CN120105216A - Air quality prediction model training method, air quality prediction method and system - Google Patents
Air quality prediction model training method, air quality prediction method and system Download PDFInfo
- Publication number
- CN120105216A CN120105216A CN202510601073.4A CN202510601073A CN120105216A CN 120105216 A CN120105216 A CN 120105216A CN 202510601073 A CN202510601073 A CN 202510601073A CN 120105216 A CN120105216 A CN 120105216A
- Authority
- CN
- China
- Prior art keywords
- air quality
- data
- prediction model
- quality prediction
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Primary Health Care (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本发明提供了一种空气质量预测模型训练方法、空气质量预测方法及系统，涉及计算机技术领域，该空气质量预测模型训练方法首先获取历史空气质量数据及其对应的空气质量真实值；而后通过特征注意力选择模型对历史空气质量数据进行特征提取和加权处理，得到空气质量时序数据；再通过空气质量预测模型学习空气质量时序数据的目标隐藏状态向量，并基于目标隐藏状态向量确定空气质量预测值；其中，空气质量预测模型为基于MLSTM构建的模型；最后根据空气质量真实值和空气质量预测值，对空气质量预测模型进行调整。本发明能够提高空气质量预测模型预测空气质量的准确性。
The present invention provides an air quality prediction model training method, an air quality prediction method and a system, and relates to the field of computer technology. The air quality prediction model training method first obtains historical air quality data and its corresponding true air quality value; then extracts and weights the historical air quality data through a feature attention selection model to obtain air quality time series data; then learns the target hidden state vector of the air quality time series data through an air quality prediction model, and determines the air quality prediction value based on the target hidden state vector; wherein the air quality prediction model is a model constructed based on MLSTM; finally, the air quality prediction model is adjusted according to the true air quality value and the air quality prediction value. The present invention can improve the accuracy of the air quality prediction model in predicting air quality.
Description
Technical Field
The invention relates to the technical field of computers, in particular to an air quality prediction model training method, an air quality prediction method and an air quality prediction system.
Background
By selecting a proper deep neural network, the complex nonlinear relation is fitted by utilizing the historical data reported by the environmental air monitoring, so that the aim of environmental quality prediction is fulfilled. There are many methods for predicting environmental quality by using deep neural network, such as dividing air quality data into different granularities, training a model by contrast learning, obtaining an air quality prediction value by obtaining a multi-granularity contrast learning air quality prediction model, or combining multi-channel data input with an LSTM (Long Short-Term Memory) model to predict air quality.
However, in practical applications, it is found that the above-mentioned related method for predicting environmental quality achieves a low accuracy of the prediction effect.
Disclosure of Invention
Accordingly, the present invention is directed to an air quality prediction model training method, an air quality prediction method and an air quality prediction system, which can improve the accuracy of air quality prediction model prediction of air quality.
In a first aspect, an embodiment of the present invention provides an air quality prediction model training method, including the steps of:
acquiring historical air quality data and a corresponding air quality true value thereof;
performing feature extraction and weighting processing on the historical air quality data through a feature attention selection model to obtain air quality time sequence data;
learning a target hidden state vector of the air quality time sequence data through an air quality prediction model, and determining an air quality predicted value based on the target hidden state vector, wherein the air quality prediction model is a model constructed based on MLSTM;
and adjusting the air quality prediction model according to the air quality true value and the air quality prediction value.
In some embodiments, the feature attention selection model comprises a convolution layer, a self-attention mechanism module and a normalization module, wherein the normalization module comprises a full connection layer and an activation layer;
The method for obtaining the air quality time sequence data comprises the following steps of:
Extracting features of the input historical air quality data through the convolution layer to obtain a plurality of feature vector matrixes;
Calculating each eigenvector matrix based on the initialized weight matrix through the self-attention mechanism module to obtain the attention weight of each eigenvector matrix, wherein the attention weight is used for representing the importance degree of the eigenvector matrix;
And normalizing the attention weight of each eigenvector matrix by the normalization module through a full connection layer and an activation function to obtain air quality time sequence data.
In some embodiments, the calculating the feature vector matrices based on the initialized weight matrix to obtain the attention weight of each feature vector matrix includes:
The first weight matrix comprises a Key matrix, a Value matrix and a Query matrix;
Calculating each eigenvector matrix according to the first weight matrix to obtain a second weight matrix of each eigenvector matrix;
And calculating the attention weight of each feature vector matrix according to a preset self-attention formula and the second weight matrix.
In some embodiments, the learning the target hidden state vector of the air quality time series data by the air quality prediction model, and determining the air quality prediction value based on the target hidden state vector, includes:
Calculating the air quality time sequence data at each moment by using an air quality prediction model to obtain an original hidden state vector corresponding to each moment;
Performing multi-wheel gating interaction and linear interpolation operation on the air quality time sequence data at the current moment and the original hidden state vector at the last moment to obtain target hidden state vectors corresponding to all moments;
and calculating the target hidden state vector based on a preset prediction algorithm, and determining an air quality predicted value.
In some embodiments, the performing multiple-round gating interaction and linear interpolation on the air quality time sequence data at the current time and the original hidden state vector at the previous time to obtain a target hidden state vector corresponding to each time includes:
performing multi-round gating interaction on the air quality time sequence data at the current moment and the original hidden state vector at the last moment according to the designated number of interaction rounds to obtain an air quality iteration vector corresponding to the air quality time sequence data and a hidden state iteration vector corresponding to the original hidden state vector;
and respectively carrying out linear difference operation on the air quality iteration vector and the hidden state iteration vector to correspondingly obtain a target air quality time sequence vector and a target hidden state vector.
In some embodiments, said adjusting the air quality prediction model based on the air quality actual value and the air quality predicted value comprises:
Determining a loss value between the air quality true value and the air quality predicted value according to a preset Huber loss function;
and adjusting the air quality prediction model according to the loss value.
In some embodiments, the Huber loss function comprises:
when the error between the air quality true value and the air quality predicted value is smaller than a preset error threshold value, adopting a mean square error MSE function as the Huber loss function;
And when the error between the air quality true value and the air quality predicted value is not smaller than the error threshold value, adopting an average absolute error MAE function as the Huber loss function.
In a second aspect, an embodiment of the present invention provides an air quality prediction method, including the steps of:
acquiring original air quality data;
Performing feature extraction and weighting processing on the original air quality data through a feature attention selection model to obtain air quality input data;
Inputting the air quality input data into an air quality prediction model, wherein the air quality prediction model is any one of the air quality prediction models;
And predicting an air quality value according to the air quality input data by adopting the air quality prediction model.
In a third aspect, an embodiment of the present invention provides an air quality prediction model training system, the system including:
The first data acquisition module is used for acquiring historical air quality data and corresponding air quality true values thereof;
the first feature processing module is used for carrying out feature extraction and weighting processing on the historical air quality data through a feature attention selection model to obtain air quality time sequence data;
The first air quality prediction module is used for learning a target hiding state vector of the air quality time sequence data through an air quality prediction model and determining an air quality prediction value based on the target hiding state vector, wherein the air quality prediction model is a model constructed based on MLSTM;
and the model adjustment module is used for adjusting the air quality prediction model according to the air quality true value and the air quality predicted value.
In a fourth aspect, embodiments of the present invention provide an air quality prediction system, the system comprising:
the second data acquisition module is used for acquiring original air quality data;
The second feature processing module is used for carrying out feature extraction and weighting processing on the original air quality data through a feature attention selection model to obtain air quality input data;
the data input module is used for inputting the air quality input data into an air quality prediction model, wherein the air quality prediction model is any one of the air quality prediction models;
And the second air quality prediction module is used for predicting an air quality value according to the air quality input data by adopting the air quality prediction model.
In a fifth aspect, an embodiment of the present invention further provides an electronic device, including a memory, and a processor, where the memory stores a computer program that can be executed on the processor, where the processor executes the computer program to implement the steps of the air quality prediction model training method mentioned in the first aspect or the steps of the air quality prediction method mentioned in the second aspect.
The embodiment of the invention has at least the following beneficial effects:
The invention provides an air quality prediction model training method, an air quality prediction method and an air quality prediction system, wherein the air quality prediction model training method firstly acquires historical air quality data and corresponding air quality true values thereof; the method comprises the steps of obtaining historical air quality data through feature extraction and weighting processing, obtaining air quality time sequence data through a feature attention selection model, learning a target hiding state vector of the air quality time sequence data through an air quality prediction model, and determining an air quality prediction value based on the target hiding state vector, wherein the air quality prediction model is constructed based on MLSTM, and finally adjusting the air quality prediction model according to an air quality true value and the air quality prediction value.
In the scheme, the characteristic extraction and weighting treatment are carried out on the historical air quality data through the characteristic attention selection model, so that the characteristics with strong importance and high contribution degree can be amplified, the noise characteristics can be reduced, the high-quality air quality time sequence data can be obtained, and the subsequent beneficial influence on the air quality prediction model can be increased. By determining the air quality predicted value based on the SLSET constructed air quality predicted model, the long-term dependency relationship in the air quality time sequence data can be extracted, so that the air quality predicted model learns more context information, and the air quality predicting effect of the air quality predicted model is improved. And finally, according to the air quality true value and the air quality predicted value, the air quality predicted model is adjusted, so that the model can be conveniently and rapidly converged, and the expected training purpose is achieved. Therefore, in the model training process, by amplifying the characteristics with high contribution degree and reducing noise characteristics and by utilizing SLSET learning context information, the air quality prediction model in the training process can optimize training efficiency, improve model performance, enhance generalization capability and reduce debugging cost, and the trained air quality prediction model can obtain better air quality prediction effect.
Additional features and advantages of the invention will be set forth in the description which follows, or in part will be obvious from the description, or may be learned by practice of the invention.
In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of an air quality prediction model training method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an air quality prediction model training process according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a feature attention selection model according to an embodiment of the present invention;
Fig. 4 is a schematic structural diagram of MLSTM according to an embodiment of the present invention;
FIG. 5 is a flowchart of an air quality prediction method according to an embodiment of the present invention;
FIG. 6 is a schematic structural diagram of an air quality prediction model training system according to an embodiment of the present invention;
FIG. 7 is a schematic diagram of an air quality prediction system according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Icon:
601-a first data acquisition module, 602-a first feature processing module, 603-a first air quality prediction module, 604-a model adjustment module;
701-a second data acquisition module, 702-a second feature processing module, 703-a data input module, 704-a second air quality prediction module;
101-processor, 102-memory, 103-bus, 104-communication interface.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
At present, the accuracy of the prediction effect obtained by the method for predicting the environmental quality by using the deep neural network is low. In view of this problem, the embodiment of the present invention considers that the related art mostly ignores that noise data existing in air quality data may interfere with a final prediction result, and that there is a problem of losing context information of a time sequence model, resulting in poor prediction effect of the model.
Based on the method, the air quality prediction model training method and the air quality prediction system, the method performs feature extraction and weighting processing on historical air quality data through a feature attention selection model to screen noise data, calculates air quality time sequence data based on the air quality prediction model MLSTM, can effectively alleviate the problem of context information loss, and further can improve accuracy of weather quality prediction of the model.
In order to facilitate understanding of the embodiment, firstly, the air quality prediction model training method disclosed by the embodiment of the invention is described in detail, and is applied to a scene of model training, wherein the trained model is a model for air quality prediction, and particularly an air quality prediction model constructed based on MLSTM, for example, and the method relates to the artificial intelligence fields of deep learning, data mining, big data analysis and the like. As shown in fig. 1, an air quality prediction model training method may include the steps of:
step S101, historical air quality data and corresponding air quality true values thereof are obtained.
Step S102, performing feature extraction and weighting processing on the historical air quality data through a feature attention selection model to obtain air quality time sequence data.
And step S103, learning a target hidden state vector of the air quality time sequence data through an air quality prediction model, and determining an air quality predicted value based on the target hidden state vector, wherein the air quality prediction model is a model constructed based on MLSTM.
And step S104, adjusting the air quality prediction model according to the air quality true value and the air quality prediction value.
For a better understanding of the solution, the following detailed description will be given for this embodiment in conjunction with the schematic diagram of the air quality prediction model training process shown in fig. 2.
For step S101, reference may be made to the following for an embodiment of acquiring historical air quality data and its corresponding air quality truth value.
In one possible embodiment, a campus is taken as an example of a scenario of air quality to be measured. One or more air quality monitoring devices are installed in a central monitoring quality area of the campus or in a critical area of the campus where air quality is to be measured. Air quality data, such as index data including but not limited to PM2.5, PM10, noise, air temperature, air humidity, wind speed, total particle suspension, nitric oxide, sulfur dioxide carbon monoxide and the like, is acquired by the air quality monitoring device according to preset acquisition time (such as once per hour).
A dataset is constructed using a large amount of historical air quality data acquired over a period of time. Specifically, the historical air quality data is segmented into a training set, a validation set and a test set. The training set is used for training the air quality prediction model based on the characteristic attention mechanism, the verification set is used for verifying the training effect of the air quality prediction model, and the training can be stopped when the air quality prediction model converges. The test set is used to evaluate the final performance of the air quality prediction model.
The time window selection is then performed, i.e. in the air quality prediction phase using an air quality prediction model, reasoning is required from the historical air quality data. In one example, historical air quality data at a time granularity of 12 hours, 24 hours, 36 hours, or 48 hours, respectively, may be referenced to predict air quality for a future preset time (e.g., 6 hours in the future).
For step S102, in the present embodiment, referring to FIG. 3, the feature attention selection model may include a convolution layer, a self-attention mechanism module, and a normalization module, where the normalization module includes a full connection layer and an activation layer.
In one embodiment, the feature Attention selection model is built on the basis of the existing ECANet model, ECANet is an Efficient Channel Attention (ECA) Module for depth CNN. Since ECANet is a model based on the computer vision field, and the historical air quality data to be processed by the present embodiment is a sequence level feature extraction task, the attention mechanism of raw ECANet cannot be fully applied to the sequence level feature extraction task of the present embodiment.
Aiming at the problem, the embodiment improves the traditional attention mechanism into a self-attention mechanism on the basis of ECANet, and constructs and obtains a characteristic attention selection model, so that the relation among the characteristics can be better obtained, the importance degree of each characteristic is clear, and a better characteristic selection effect is achieved.
The feature attention selection model in the embodiment can be divided into a three-layer structure, wherein the first layer structure comprises a convolution layer for extracting feature information of historical air quality data, the second layer structure comprises a self-attention mechanism module for calculating importance degrees of each feature, and the third layer structure comprises a normalization module, wherein the normalization module comprises a full-connection layer and an activation layer and is used for acquiring weights of all the features to distinguish importance degrees among the features.
Based on the above-mentioned feature attention selection model, the present embodiment performs feature extraction and weighting processing on the historical air quality data through the feature attention selection model to obtain air quality time series data, which may include the following.
Step S201, feature extraction is carried out on input historical air quality data through a convolution layer, and a plurality of feature vector matrixes are obtained.
In this embodiment, the historical air quality data of T times may be selected, where the historical air quality data of each time includes F pollution factors, and a basic data matrix of t×f is formed as input data of the feature attention selection model. The input data of T is input to a convolution layer for feature extraction, and a feature vector matrix X is obtained.
Step S202, calculating each feature vector matrix based on the initialized weight matrix through a self-attention mechanism module to obtain attention weights of each feature vector matrix, wherein the attention weights are used for representing the importance degree of the feature vector matrix.
The embodiment comprises an initialized first weight matrix, wherein the first weight matrix comprises a Key matrixValue matrixAnd Query matrixCalculating each eigenvector matrix according to the first weight matrix to obtain a second weight matrix of each eigenvector matrix; and calculating the attention weight of each feature vector matrix according to a preset self-attention formula and a second weight matrix.
Specifically, three first weight matrixes are initialized randomly and are respectively a Query matrixKey matrixAnd Value matrixReferring to the following formula (1), each first weight matrix is calculated with the eigenvector matrix X to obtain a second weight matrix, and the second weight matrix is Q, K, V matrices respectively:
(1)
then, according to a preset self-attention formula (2), calculating to obtain the attention weight of each feature vector matrix:
(2)
wherein, Representing the attention weight.
Step S203, the attention weight of each feature vector matrix is normalized by a normalization module through a full connection layer and an activation function, and air quality time sequence data are obtained.
After the attention weight of the feature vector matrix is obtained, referring to formula (3), the embodiment adopts the full connection layer and utilizes the activation function of the activation layer to convert the attention weight into a value between 0 and 1, thereby obtaining the final attention moment matrix:
(3)
The attention matrix of each moment in the time period T is obtained through the stepsThereby forming time series data, namely air quality time series data which can be expressed asWherein t represents the time instant t,Air quality time series data representing the t-th time.
Attention moment array obtained according to the present embodimentAnd the air quality time sequence data comprises weighted characteristic data obtained by attention mechanism calculation, so that the characteristics with strong importance and high contribution degree are highlighted and amplified, the influence of noise characteristics on model prediction is reduced, and support is provided for later study.
In the embodiment, the self-attention mechanism of the feature attention selection model directly models the global relation, so that information loss is reduced, and the feature extraction effect is improved. Through stacking a plurality of layers of self-attention mechanisms, the characteristic attention selection model can extract multi-level characteristics, and the expression capacity is enhanced, so that the characteristics with high contribution degree are improved, the influence of low contribution degree or noise on a prediction result is reduced, the characteristic relation of air quality time sequence data is extracted more effectively, a better characteristic extraction effect is achieved, more proper weight is given, and the learning effect of the model is improved.
For the air quality prediction model constructed based on MLSTM in step S103, MLSTM refers to modified Mogrifer LSTM (Long Short-Term Memory network).
In some embodiments, the air quality time series data obtained in the previous embodiments may be usedAs input to the LSTM model. Wherein, the input of the LSTM model at the moment t is,Indicating the hidden state of the last memory cell,Long term memory representing the input of the previous time.Indicating the hidden state of the current moment,Representing the long-term memory output at the current time.、、Respectively representing a forgetting door, an input door and an output door.
Forgetting doorFor determining the state of the cell at the previous timeHow much information needs to be discarded and the remaining information is passed to the state of the current cell. Forgetting doorThe calculation mode of (2) is as follows:
(4)
wherein, A weight matrix representing the forgetting gate,A bias variable representing a forgetting gate.
Input doorTime-series data for determining air quality input at present momentAnd the hidden state of the last momentThe information of that part of the memory can be reserved to the current state of the unitIs a kind of medium. Input doorThe calculation mode of (2) is as follows:
(5)
wherein, Representing a weight matrix of the input gates,Representing the bias variable of the input gate.
Candidate vectorRepresenting the current stateState information of (2) from the last moment to conceal the stateAnd current time inputAnd (5) calculating to obtain the product. The calculation method is as follows:
(6)
Memory state of current cell The calculation method is as follows:
(7)
for the output gate of the current unit, controlling the state of the current unit The useful information in (a) is saved to the next time. Output doorThe calculation formula of (2) is as follows:
(8)
wherein, Representing the weight matrix of the output gate,Representing the bias variable of the output gate.
Finally calculating hidden state output of the current unitThe formula is as follows:
(9)
In the above formula, W represents a weight matrix, and b represents a bias variable. In the LSTM model, the input of the current state in the process of the related information flow of the air quality time sequence data can be found Output from last stateThe interaction between the gates occurs so that the relationship between the two is one-dimensional. The interaction of the input and contextual information is inadequate, which can result in some loss of contextual information.
Thus, to enhance the dependency between contexts, the present embodiment utilizes MLSTM to learn the relationship between features. Referring to FIG. 4, a diagram of MLSTM is shown, MLSTM is implemented by alternating air quality timing data based on the original LSTMAnd the hidden state of the last moment(Also denoted as h prev) to overcome the problem of insufficient one-dimensional relationship of context.
Based on this, the present embodiment can construct an air quality prediction model capable of learning context information based on MLSTM, learn a target hidden state vector of air quality time series data through the air quality prediction model, and determine an air quality prediction value based on the target hidden state vector. The present embodiment may include the following steps S301 to S303.
In step S301, air quality time sequence data at each moment is calculated by an air quality prediction model, so as to obtain an original hidden state vector corresponding to each moment.
Step S302, carrying out multi-wheel gating interaction and linear interpolation operation on the air quality time sequence data at the current moment and the original hidden state vector at the last moment to obtain target hidden state vectors corresponding to all moments.
The multi-round gating interaction in the embodiment comprises the steps of carrying out multi-round gating interaction on air quality time sequence data at the current moment and an original hidden state vector at the last moment according to the designated number of interaction rounds to obtain an air quality iteration vector corresponding to the air quality time sequence data and a hidden state iteration vector corresponding to the original hidden state vector.
Specifically, the air quality characteristic at the current momentAnd the hidden state of the last momentInter-modulating according to the designated number of interactive rounds to obtain air quality iterative vector corresponding to air quality time sequence data, wherein, for example, the air quality iterative vector is adoptedAn air quality iteration vector representing the r-th round iteration at time t, and a hidden state iteration vector corresponding to the original hidden state vector is obtained, for example, byAnd a hidden state iteration vector of the r-th round of iteration representing the hidden layer state of the previous unit. Each wheelAndThe calculation formula of (2) is as follows:
(10)
(11)
Wherein r represents the number of interactive rounds, AndRepresenting a randomly initialized matrix. R represents air quality characteristicsHidden stateThe two parameters are updated alternately. After r-round iterationAnd (3) withMore long-term dependence information is contained, so that the air quality prediction model can achieve better effect in the training stage.
Although the air quality iteration vectorHidden state iteration vectorAfter multiple rounds of iterative updating, a long-term context dependency relationship can be obtained to improve the performance of the model, but the operation can lead the air quality prediction model to pay attention to the context information in a transitional manner, so that the characteristic information at the current moment is ignored. Thus, the method is applicable to a variety of applications. In order to ensure that the context dependent information is obtained while still focusing on the original features, the present embodiment may perform a linear interpolation operation, i.e., iterate the vector for the air quality, with reference to the following formulas (12) and (13)Hidden state iteration vectorRespectively performing linear difference operation to obtain target air quality time sequence vectorAnd a target hidden state vectorThereby balancing the original characteristic information and the context.
(12)
(13)
Then, as shown in FIG. 4, the target air quality timing vector after balancing the original characteristic information and the context dependent informationAnd a target hidden state vectorThe prediction capability of the air quality prediction model can be further improved by inputting the air quality prediction model into the LSTM model.
Step S303, calculating a target hidden state vector based on a preset prediction algorithm, and determining an air quality predicted value.
In this embodiment, the final output module of the air quality prediction model is composed of a full connection layer, which is generally used as the last layer of the neural network for outputting the final prediction result. The function of the fully connected layer is to connect all nodes of the previous layer to all nodes of the current layer, so that linear combination of the features is realized, and a higher-level feature representation is obtained. The calculation mode of the full connection layer is simple and easy to realize and understand. Meanwhile, as the output of the full-connection layer is the result of the combined action of all nodes of the previous layer, the global information of the input data can be captured to a certain extent, and the learning of the air quality prediction model is facilitated.
Thus, in MLSTM-based air quality prediction models, air quality timing dataLearning through MLSTM layers to generate a target hidden state vector representing an air quality prediction resultThe object hiding state vectorAn output module of the air quality prediction model is also required to finally generate an air quality prediction value for a preset future time (e.g., 24 hours).
The output module of the air quality prediction model learns through the full connection layer and hides the state vector of the targetAnd (5) performing operation, and finally integrating dimension reduction into an air quality predicted value. The specific operation process formula is as follows (14):
(14)
wherein, For the final air quality prediction value, for MLSTM learned timing information,Representing the weight of the object to be weighed,Is a bias variable.
The above embodiment uses MLSET as the main structure of the air quality prediction model, and air quality time series dataTraining is performed. In this embodiment, MLSET can extract air quality time sequence data by applying the air quality prediction model to the time sequence prediction fieldThe long-term dependency relationship in the air quality prediction model enables the air quality prediction model to learn more context information, so that the air quality prediction effect of the air quality prediction model is improved.
For step S104, the present embodiment may determine a loss value between the air quality actual value and the air quality predicted value according to a preset Huber loss function, and adjust the air quality prediction model according to the loss value.
In this embodiment, a smoothed average absolute error Huber loss function is selected when selecting the model-trained loss function. Huber loss function pass parameterThe selection of the MSE (Mean Squared Error, mean square error) function and the MAE (Mean Absolute Error ) function is controlled as thresholds.
Wherein the MSE function is shown in equation (15) below:
(15)
wherein, Representing a true value of the air quality,Represents the air quality predicted value, n represents the air quality actual value and the number of air quality predicted values, n=2 in this example.
The MSE function is led everywhere, and when the error is reduced, the gradient is also reduced, so that the convergence of the air quality prediction model is facilitated. However, when the error between the air quality actual value and the air quality predicted value is relatively large, that is, when the outlier is the maximum, the derivative value is large, so that the error is amplified, and the training is unstable.
The MAE function is shown in equation (16) below:
(16)
The MAE function calculates the absolute value of the error between the air quality predicted value and the air quality true value, which is a measure of the average error range of the air quality predicted value and the air quality true value, regardless of the error direction. MAE functions are less sensitive to abrupt values than MSE functions, but are not conductive at 0, and the gradient caused by smaller losses is also large, which is detrimental to convergence.
In order to balance the advantages and disadvantages of both the MSE function and the MAE function, the present embodiment uses a Huber loss function, referring to equation (17), which includes:
(17)
when the error between the air quality actual value and the air quality predicted value is smaller than a preset error threshold value When the MSE function is adopted as the Huber loss function, namely, when the error between the air quality true value and the air quality predicted value is taken as the valueThe Huber loss function is equivalent to the MSE function;
when the error between the air quality actual value and the air quality predicted value is not less than the error threshold value When the MAE function is used as Huber loss function, that is, when the error between the air quality true value and the air quality predicted value is taken as the valueAndIn the range of (2), the Huber loss function is equivalent to the MAE function.
The advantages of the MSE function and the MAE function can be effectively fused through the Huber loss function, the MSE function and the MAE function are insensitive to abrupt values, and the derivative around 0 is smoother.
The influence of the extreme noise data on the prediction result of the air quality prediction model can be effectively relieved by using the Huber loss function, and the robustness and generalization of the air quality prediction are enhanced.
Through the embodiment, the model based on MLSTM is constructed and trained, and the air quality prediction model based on the characteristic attention mechanism can be obtained.
After the trained air quality prediction model is obtained, the air quality data acquired by the air quality acquisition equipment can be input into the air quality prediction model, and future air quality values are predicted by the air quality prediction model.
Based on this, as shown in fig. 5, the present embodiment provides an air quality prediction method, which includes the steps of:
s501, acquiring original air quality data. The embodiment can collect the original air quality data in a period of time according to the preset collection time through the air quality collection device.
S502, performing feature extraction and weighting processing on the original air quality data through a feature attention selection model to obtain air quality input data. According to the embodiment, the original air quality data is subjected to feature extraction and weighting processing through the feature attention selection model, so that the air quality data with high contribution degree can be improved, the air quality data or noise data with low contribution degree can be reduced, and the obtained air quality input data has higher quality, so that the accuracy of the air quality prediction model on a prediction result can be improved subsequently.
S503, inputting air quality input data into an air quality prediction model, wherein the air quality prediction model is obtained according to the air quality prediction model training method provided by the embodiment.
S504, an air quality prediction model is adopted, and an air quality value is predicted according to air quality input data. The air quality prediction model can be directly applied to the prediction of the air quality value and obtain a good prediction effect after training, and can extract long-term dependency relationship in the air quality input data based on MLSTM as a main structure and learn more context information, so that the prediction accuracy of the air quality value can be effectively improved.
In summary, the air quality prediction model training method provided by the embodiment of the invention comprises the steps of firstly obtaining historical air quality data and corresponding air quality true values thereof, then carrying out feature extraction and weighting treatment on the historical air quality data through a feature attention selection model to obtain air quality time sequence data, then learning a target hidden state vector of the air quality time sequence data through the air quality prediction model, and determining an air quality prediction value based on the target hidden state vector, wherein the air quality prediction model is a model constructed based on MLSTM, and finally adjusting the air quality prediction model according to the air quality true values and the air quality prediction value.
In the scheme, the characteristic extraction and weighting treatment are carried out on the historical air quality data through the characteristic attention selection model, so that the characteristics with strong importance and high contribution degree can be amplified, the noise characteristics can be reduced, the high-quality air quality time sequence data can be obtained, and the subsequent beneficial influence on the air quality prediction model can be increased. By determining the air quality predicted value based on the SLSET constructed air quality predicted model, the long-term dependency relationship in the air quality time sequence data can be extracted, so that the air quality predicted model learns more context information, and the air quality predicting effect of the air quality predicted model is improved. And finally, according to the air quality true value and the air quality predicted value, the air quality predicted model is adjusted, so that the model can be conveniently and rapidly converged, and the expected training purpose is achieved. Therefore, in the model training process, by amplifying the characteristics with high contribution degree and reducing noise characteristics and by utilizing SLSET learning context information, the air quality prediction model in the training process can optimize training efficiency, improve model performance, enhance generalization capability and reduce debugging cost, and the trained air quality prediction model can obtain better air quality prediction effect.
Corresponding to the above method embodiment, the embodiment of the present invention provides an air quality prediction model training system, as shown in fig. 6, which includes the following modules:
a first data acquisition module 601, configured to acquire historical air quality data and corresponding air quality real values thereof;
The first feature processing module 602 is configured to perform feature extraction and weighting processing on the historical air quality data through a feature attention selection model to obtain air quality time sequence data;
The first air quality prediction module 603 is configured to learn a target hidden state vector of the air quality time sequence data through an air quality prediction model, and determine an air quality predicted value based on the target hidden state vector, where the air quality prediction model is a model constructed based on MLSTM;
The model adjustment module 604 is configured to adjust the air quality prediction model according to the air quality actual value and the air quality predicted value.
In some implementations, the feature attention selection model includes a convolution layer, a self-attention mechanism module, and a normalization module including a full connection layer and an activation layer, the first feature processing module 602 includes:
the convolution unit is used for extracting the characteristics of the input historical air quality data through the convolution layer to obtain a plurality of characteristic vector matrixes;
The self-attention mechanism unit is used for calculating each eigenvector matrix based on the initialized weight matrix through the self-attention mechanism module to obtain the attention weight of each eigenvector matrix, wherein the attention weight is used for representing the importance degree of the eigenvector matrix;
The normalization unit is used for normalizing the attention weight of each feature vector matrix by the normalization module through the full connection layer and the activation function to obtain air quality time sequence data.
In some embodiments, the self-attention mechanism unit is further to:
The first weight matrix comprises a Key matrix, a Value matrix and a Query matrix;
Calculating each eigenvector matrix according to the first weight matrix to obtain a second weight matrix of each eigenvector matrix;
And calculating the attention weight of each feature vector matrix according to a preset self-attention formula and the second weight matrix.
In some embodiments, the first air quality prediction module 603 includes:
The data calculation unit is used for calculating the air quality time sequence data at each moment through an air quality prediction model to obtain an original hidden state vector corresponding to each moment;
the gating interaction and linear interpolation operation unit is used for carrying out multi-wheel gating interaction and linear interpolation operation on the air quality time sequence data at the current moment and the original hidden state vector at the last moment to obtain target hidden state vectors corresponding to all moments;
And the prediction unit is used for calculating the target hiding state vector based on a preset prediction algorithm and determining an air quality predicted value.
In some embodiments, the gating interaction and linear interpolation operation unit is further configured to:
performing multi-round gating interaction on the air quality time sequence data at the current moment and the original hidden state vector at the last moment according to the designated number of interaction rounds to obtain an air quality iteration vector corresponding to the air quality time sequence data and a hidden state iteration vector corresponding to the original hidden state vector;
and respectively carrying out linear difference operation on the air quality iteration vector and the hidden state iteration vector to correspondingly obtain a target air quality time sequence vector and a target hidden state vector.
In some implementations, the model adjustment module 604 includes:
A loss value determining unit, configured to determine a loss value between the air quality actual value and the air quality predicted value according to a preset Huber loss function;
And the adjusting unit is used for adjusting the air quality prediction model according to the loss value.
In some embodiments, the Huber loss function comprises:
when the error between the air quality true value and the air quality predicted value is smaller than a preset error threshold value, adopting a mean square error MSE function as the Huber loss function;
And when the error between the air quality true value and the air quality predicted value is not smaller than the error threshold value, adopting an average absolute error MAE function as the Huber loss function.
The air quality prediction model training system provided by the embodiment has the same technical characteristics as the air quality prediction model training method provided by the embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved. For a brief description, reference may be made to the corresponding content of the foregoing air quality prediction model training method embodiments where the embodiments are not mentioned.
Corresponding to the above method embodiments, the embodiment of the present invention provides an air quality prediction system, as shown in fig. 7, which includes the following modules:
a second data acquisition module 701, configured to acquire raw air quality data;
a second feature processing module 702, configured to perform feature extraction and weighting processing on the raw air quality data through a feature attention selection model, so as to obtain air quality input data;
A data input module 703, configured to input the air quality input data into an air quality prediction model, where the air quality prediction model is the air quality prediction model described in the foregoing embodiment;
a second air quality prediction module 704, configured to predict an air quality value according to the air quality input data using the air quality prediction model.
The air quality prediction system provided in this embodiment has the same technical characteristics as the air quality prediction method provided in the foregoing embodiment, so that the same technical problems can be solved, and the same technical effects can be achieved. For a brief description, reference may be made to the corresponding content of the air quality prediction method embodiments described above, where the embodiments are not mentioned.
The embodiment also provides an electronic device, the structural schematic diagram of which is shown in fig. 8, where the electronic device includes a processor 101 and a memory 102, where the memory 102 is configured to store one or more computer instructions, and the one or more computer instructions are executed by the processor to implement the air quality prediction model training method or the air quality prediction method.
The electronic device shown in fig. 8 further comprises a bus 103 and a communication interface 104, the processor 101, the communication interface 104 and the memory 102 being connected by the bus 103.
The memory 102 may include a high-speed random access memory (RAM, random Access Memory), and may further include a non-volatile memory (non-volatile memory), such as at least one disk memory. Bus 103 may be an ISA bus, a PCI bus, an EISA bus, or the like. The buses may be classified as address buses, data buses, control buses, etc. For ease of illustration, only one bi-directional arrow is shown in FIG. 8, but not only one bus or type of bus.
The communication interface 104 is configured to connect with at least one user terminal and other network units through a network interface, and send the encapsulated IPv4 message or the IPv4 message to the user terminal through the network interface.
The processor 101 may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in the processor 101 or instructions in the form of software. The Processor 101 may be a general-purpose Processor, including a central processing unit (Central Processing Unit, CPU), a network Processor (Network Processor, NP), a digital signal Processor (DIGITAL SIGNAL Processor, DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable gate array (FPGA) or other Programmable logic device, discrete gate or transistor logic device, or discrete hardware components. The various methods, steps and logic blocks of the disclosure in the embodiments of the disclosure may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the embodiments of the present disclosure may be embodied directly in hardware, in a decoded processor, or in a combination of hardware and software modules in a decoded processor. The software modules may be located in a random access memory, flash memory, read only memory, programmable read only memory, or electrically erasable programmable memory, registers, etc. as well known in the art. The storage medium is located in the memory 102, and the processor 101 reads information in the memory 102, and in combination with its hardware, performs the steps of the method of the previous embodiment.
The embodiment of the invention also provides a readable storage medium, and a computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the steps of the air quality prediction model training method or the steps of the air quality prediction method in the previous embodiment are executed.
In the several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. The above-described apparatus embodiments are merely illustrative, for example, the division of the units is merely a logical function division, and there may be other manners of division in actual implementation, and for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some communication interface, indirect coupling or communication connection of devices or units, electrical, mechanical, or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a non-volatile computer readable storage medium executable by a processor. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. The storage medium includes a U disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, an optical disk, or other various media capable of storing program codes.
It should be noted that the foregoing embodiments are merely illustrative embodiments of the present invention, and not restrictive, and the scope of the invention is not limited to the embodiments, and although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that any modification, variation or substitution of some of the technical features of the embodiments described in the foregoing embodiments may be easily contemplated within the scope of the present invention, and the spirit and scope of the technical solutions of the embodiments do not depart from the spirit and scope of the embodiments of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A method of training an air quality prediction model, the method comprising the steps of:
acquiring historical air quality data and a corresponding air quality true value thereof;
performing feature extraction and weighting processing on the historical air quality data through a feature attention selection model to obtain air quality time sequence data;
learning a target hidden state vector of the air quality time sequence data through an air quality prediction model, and determining an air quality predicted value based on the target hidden state vector, wherein the air quality prediction model is a model constructed based on MLSTM;
and adjusting the air quality prediction model according to the air quality true value and the air quality prediction value.
2. The method of claim 1, wherein the feature attention selection model comprises a convolution layer, a self-attention mechanism module, and a normalization module comprising a full connection layer and an activation layer;
The method for obtaining the air quality time sequence data comprises the following steps of:
Extracting features of the input historical air quality data through the convolution layer to obtain a plurality of feature vector matrixes;
Calculating each eigenvector matrix based on the initialized weight matrix through the self-attention mechanism module to obtain the attention weight of each eigenvector matrix, wherein the attention weight is used for representing the importance degree of the eigenvector matrix;
And normalizing the attention weight of each eigenvector matrix by the normalization module through a full connection layer and an activation function to obtain air quality time sequence data.
3. The method according to claim 2, wherein the calculating each of the feature vector matrices based on the initialized weight matrix to obtain the attention weight of each of the feature vector matrices includes:
The first weight matrix comprises a Key matrix, a Value matrix and a Query matrix;
Calculating each eigenvector matrix according to the first weight matrix to obtain a second weight matrix of each eigenvector matrix;
And calculating the attention weight of each feature vector matrix according to a preset self-attention formula and the second weight matrix.
4. The method of claim 1, wherein the learning the target hidden state vector of the air quality time series data by the air quality prediction model and determining the air quality prediction value based on the target hidden state vector comprises:
Calculating the air quality time sequence data at each moment by using an air quality prediction model to obtain an original hidden state vector corresponding to each moment;
Performing multi-wheel gating interaction and linear interpolation operation on the air quality time sequence data at the current moment and the original hidden state vector at the last moment to obtain target hidden state vectors corresponding to all moments;
and calculating the target hidden state vector based on a preset prediction algorithm, and determining an air quality predicted value.
5. The method according to claim 4, wherein performing multi-round gating interaction and linear interpolation on the air quality time sequence data at the current time and the original hidden state vector at the previous time to obtain target hidden state vectors corresponding to respective times comprises:
performing multi-round gating interaction on the air quality time sequence data at the current moment and the original hidden state vector at the last moment according to the designated number of interaction rounds to obtain an air quality iteration vector corresponding to the air quality time sequence data and a hidden state iteration vector corresponding to the original hidden state vector;
and respectively carrying out linear difference operation on the air quality iteration vector and the hidden state iteration vector to correspondingly obtain a target air quality time sequence vector and a target hidden state vector.
6. The method of claim 1, wherein said adjusting the air quality prediction model based on the air quality truth value and the air quality prediction value comprises:
Determining a loss value between the air quality true value and the air quality predicted value according to a preset Huber loss function;
and adjusting the air quality prediction model according to the loss value.
7. The method of claim 6, wherein the Huber loss function comprises:
when the error between the air quality true value and the air quality predicted value is smaller than a preset error threshold value, adopting a mean square error MSE function as the Huber loss function;
And when the error between the air quality true value and the air quality predicted value is not smaller than the error threshold value, adopting an average absolute error MAE function as the Huber loss function.
8. A method of air quality prediction, the method comprising the steps of:
acquiring original air quality data;
Performing feature extraction and weighting processing on the original air quality data through a feature attention selection model to obtain air quality input data;
Inputting the air quality input data into an air quality prediction model, wherein the air quality prediction model is an air quality prediction model according to any one of claims 1-7;
And predicting an air quality value according to the air quality input data by adopting the air quality prediction model.
9. An air quality prediction model training system, the system comprising the following modules:
The first data acquisition module is used for acquiring historical air quality data and corresponding air quality true values thereof;
the first feature processing module is used for carrying out feature extraction and weighting processing on the historical air quality data through a feature attention selection model to obtain air quality time sequence data;
The first air quality prediction module is used for learning a target hiding state vector of the air quality time sequence data through an air quality prediction model and determining an air quality prediction value based on the target hiding state vector, wherein the air quality prediction model is a model constructed based on MLSTM;
and the model adjustment module is used for adjusting the air quality prediction model according to the air quality true value and the air quality predicted value.
10. An air quality prediction system, the system comprising the following modules:
the second data acquisition module is used for acquiring original air quality data;
The second feature processing module is used for carrying out feature extraction and weighting processing on the original air quality data through a feature attention selection model to obtain air quality input data;
a data input module for inputting the air quality input data into an air quality prediction model, the air quality prediction model being the air quality prediction model of any one of claims 1-7;
And the second air quality prediction module is used for predicting an air quality value according to the air quality input data by adopting the air quality prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510601073.4A CN120105216A (en) | 2025-05-12 | 2025-05-12 | Air quality prediction model training method, air quality prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510601073.4A CN120105216A (en) | 2025-05-12 | 2025-05-12 | Air quality prediction model training method, air quality prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN120105216A true CN120105216A (en) | 2025-06-06 |
Family
ID=95881426
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510601073.4A Pending CN120105216A (en) | 2025-05-12 | 2025-05-12 | Air quality prediction model training method, air quality prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN120105216A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120382570A (en) * | 2025-06-27 | 2025-07-29 | 甘肃展通管业有限责任公司 | A method and system for automatic batching control of PVC pipe production |
| CN120744328A (en) * | 2025-09-05 | 2025-10-03 | 歌尔股份有限公司 | State prediction method and head display device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114626567A (en) * | 2020-12-10 | 2022-06-14 | 中国科学院深圳先进技术研究院 | A Prediction Method of Air Quality Index |
| CN116629623A (en) * | 2023-04-19 | 2023-08-22 | 中广核(北京)新能源科技有限公司 | Short-term wind power prediction method based on improved Mogrifier gating interaction mechanism |
| CN117909384A (en) * | 2024-01-16 | 2024-04-19 | 沈阳建筑大学 | Multivariate time series prediction method and system based on wavelet denoising and multi-scale feature extraction |
| CN118154878A (en) * | 2024-04-01 | 2024-06-07 | 中国矿业大学 | Training method of thyroid nodule segmentation model, nodule segmentation method and device |
| CN118196293A (en) * | 2024-03-26 | 2024-06-14 | 北京航空航天大学 | A Conditional Generation Method for 3D Indoor Scenes Based on Graph Representation |
-
2025
- 2025-05-12 CN CN202510601073.4A patent/CN120105216A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114626567A (en) * | 2020-12-10 | 2022-06-14 | 中国科学院深圳先进技术研究院 | A Prediction Method of Air Quality Index |
| CN116629623A (en) * | 2023-04-19 | 2023-08-22 | 中广核(北京)新能源科技有限公司 | Short-term wind power prediction method based on improved Mogrifier gating interaction mechanism |
| CN117909384A (en) * | 2024-01-16 | 2024-04-19 | 沈阳建筑大学 | Multivariate time series prediction method and system based on wavelet denoising and multi-scale feature extraction |
| CN118196293A (en) * | 2024-03-26 | 2024-06-14 | 北京航空航天大学 | A Conditional Generation Method for 3D Indoor Scenes Based on Graph Representation |
| CN118154878A (en) * | 2024-04-01 | 2024-06-07 | 中国矿业大学 | Training method of thyroid nodule segmentation model, nodule segmentation method and device |
Non-Patent Citations (2)
| Title |
|---|
| JING LIU等: ""PM2.5 Pollution Prediction Service based on Mogrifier LSTM Model"", 《2023 IEEE 29TH INTERNATIONAL CONFERENCE ON PARALLEL AND DISTRIBUTED SYSTEMS (ICPADS)》, 26 March 2024 (2024-03-26), pages 511 - 516 * |
| 刘峤等: "《基于深度学习的自然语言处理和语音识别》", 31 January 2024, pages: 103 - 104 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120382570A (en) * | 2025-06-27 | 2025-07-29 | 甘肃展通管业有限责任公司 | A method and system for automatic batching control of PVC pipe production |
| CN120744328A (en) * | 2025-09-05 | 2025-10-03 | 歌尔股份有限公司 | State prediction method and head display device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3772023B1 (en) | Method for determining quantization parameters in neural network and related products | |
| CN120105216A (en) | Air quality prediction model training method, air quality prediction method and system | |
| CN111797122B (en) | Method and device for predicting change trend of high-dimensional reappearance concept drift stream data | |
| CN111783540B (en) | A method and system for human action recognition in video | |
| CN116760742A (en) | Network traffic anomaly detection method and system based on multi-stage hybrid spatio-temporal fusion | |
| CN114077913B (en) | Method and system for multi-step prediction of future wind speed based on automatic reservoir neural network | |
| CN112232486B (en) | An optimization method for YOLO spiking neural network | |
| CN113852432A (en) | RCS-GRU model-based spectrum prediction sensing method | |
| CN111860783A (en) | Graph node low-dimensional representation learning method, device, terminal device and storage medium | |
| WO2021042857A1 (en) | Processing method and processing apparatus for image segmentation model | |
| CN118277770A (en) | Obstacle sensing method, device, electronic device and storage medium | |
| CN119484065B (en) | Network security situation awareness method based on deep neural network | |
| CN116431597A (en) | Method, electronic device and computer program product for training a data classification model | |
| CN113328908A (en) | Abnormal data detection method and device, computer equipment and storage medium | |
| CN114861671B (en) | Model training method, device, computer equipment and storage medium | |
| Meng et al. | An effective weighted vector median filter for impulse noise reduction based on minimizing the degree of aggregation | |
| CN118536548A (en) | Training method, encoding method, decoding method and device of codec model | |
| US20250156532A1 (en) | Anomaly detection | |
| CN116664833A (en) | Method for improving target re-identification model capacity and target re-identification method | |
| CN114611695A (en) | Method, device, terminal and storage medium for air quality prediction | |
| CN119810459A (en) | Method, device, equipment and storage medium for determining abnormal IoT number card | |
| CN118797426A (en) | Model training method, device, equipment and computer storage medium | |
| CN117150326A (en) | New energy node output power prediction method, device, equipment and storage medium | |
| CN117371482A (en) | A cloud platform operation and maintenance time series knowledge graph reasoning method | |
| CN117131384A (en) | Retraining method and device for time sequence data model and computer equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |