CN116706907B - Photovoltaic power generation prediction method and related equipment based on fuzzy reasoning - Google Patents

Abstract

The invention provides a photovoltaic power generation prediction method and related equipment based on fuzzy reasoning, and relates to the technical field of photovoltaic power generation, wherein the method comprises the following steps: acquiring training environment data and training photovoltaic data; carrying out fuzzy division on training environment data to obtain a training environment data set; carrying out fuzzy division on the training photovoltaic data to obtain a training photovoltaic data set; inputting the training environment data set into a preset neural network for training to obtain a predicted photovoltaic data set corresponding to the training environment data set; according to the predicted photovoltaic data set and the training photovoltaic data set, parameter adjustment is carried out on the neural network until the neural network converges, and a photovoltaic prediction model is obtained; when the current environment data is acquired, calculating a current photovoltaic predicted value corresponding to the current environment data based on the photovoltaic predicted model. The invention can combine multisource information, fully exert the nonlinear modeling capability of the neural network and the human intelligent characteristics of fuzzy reasoning, and improve the prediction accuracy of photovoltaic power generation.

Description

Photovoltaic power generation prediction method and related equipment based on fuzzy reasoning

Technical field

The present invention relates to the technical field of photovoltaic power generation, and in particular to a photovoltaic power generation prediction method and related equipment based on fuzzy reasoning.

Background technique

Photovoltaic prediction is a key technology that can help photovoltaic power stations better utilize solar energy resources, improve power generation efficiency, reduce operating costs, and ensure the safety and stability of the power grid. The main purpose of photovoltaic forecasting is to predict the power generation or output power of photovoltaic power stations in the future based on factors such as the location of the photovoltaic power station and component characteristics. The difficulty of photovoltaic prediction is that photovoltaic power generation is affected by many factors, such as solar radiation, cloud cover, temperature changes, wind speed and direction, etc. These factors are uncertain and nonlinear, resulting in photovoltaic power generation being random and volatile.

Current photovoltaic power generation forecasting still lacks an accurate and adaptable method.

Contents of the invention

The present invention provides a photovoltaic power generation prediction method and related equipment based on fuzzy reasoning to solve the defect of lack of accurate photovoltaic power generation prediction method in the prior art and realize accurate prediction of photovoltaic power generation.

The present invention provides a photovoltaic power generation prediction method based on fuzzy reasoning, including:

Obtain training environment data and training photovoltaic data;

Fuzzy division is performed on the training environment data to obtain a training environment data set. Fuzzy division is performed on the training photovoltaic data to obtain a training photovoltaic data set. The training environment data set includes several pairs of environmental data. The pairs of environmental data are including training environment data values and environment labels corresponding to each training environment data value; the training photovoltaic data set includes several pairs of photovoltaic data pairs; the photovoltaic data pairs include training photovoltaic data values and each of the training photovoltaic data The photovoltaic tag corresponding to the value;

The training environment data set is input into a preset neural network, and the neural network is controlled to predict the training environment data set to obtain a predicted photovoltaic data set corresponding to the training environment data set. The neural network Includes fuzzy functions;

According to the predicted photovoltaic data set and the training photovoltaic data set, adjust parameters of the neural network until the neural network converges to obtain a photovoltaic prediction model;

When current environmental data is obtained, the current photovoltaic prediction value corresponding to the current environmental data is calculated based on the photovoltaic prediction model.

According to a photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention, the fuzzy division of the training environment data to obtain the training environment data set includes:

Calculate the environment feature vector corresponding to each of the training environment data;

According to the similarity between the environmental feature vectors, perform preliminary classification on the training environment data to obtain several initial environment clusters;

Perform similarity calculation on the initial environment clustering to obtain a clustering similarity value;

According to the cluster similarity value, the initial environment clusters are merged, and the data in the merged initial environment clusters are updated until the initial environment clusters meet the preset stopping criteria, and several A target environment cluster and an environment label corresponding to the target environment cluster;

According to the target environment clustering and the environment label, an environment data pair is generated.

According to a photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention, the training environment data is initially classified according to the similarity between the environmental feature vectors, and several initial environment clusters are obtained, including:

Split the environment feature vector to obtain the first environment comparison value and the second environment comparison value corresponding to the environment feature vector;

Calculate a first similarity between the first environment comparison values;

According to the first similarity, divide the environmental feature vector to obtain a first cluster;

Calculating a second degree of similarity between the second environment comparison values in the first cluster;

According to the second similarity, the environment feature vector is divided to obtain initial environment clustering.

According to a photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention, before obtaining training environment data and training photovoltaic data, it also includes:

Obtain historical environmental data, update environmental data, historical photovoltaic data and update photovoltaic data;

According to the preset update cycle, the updated environment data and historical environment data are selected to obtain training environment data; and,

The historical photovoltaic data and the updated photovoltaic data are selected to obtain training photovoltaic data.

According to a photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention, the updated environment data and historical environment data are selected according to a preset update cycle, and the training environment data obtained includes:

Determine candidate environment data among the updated environment data and the historical environment data according to the update cycle;

Compare the candidate environment data with reference environment data corresponding to the candidate environment time, and determine outliers in the candidate environment data, where the candidate environment time is the time corresponding to the candidate environment data;

According to the abnormal value and the candidate environment time, the candidate environment data is adjusted to obtain training environment data.

According to a photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention, comparing the candidate environment data with the reference environment data corresponding to the candidate environment time, and determining the abnormal values in the candidate environment data includes:

Based on the reference environmental data, determine prediction parameters;

Generate an environment prediction model based on the reference environment data and the prediction parameters;

Based on the environment prediction model, generate predicted environment data corresponding to the candidate environment time;

The predicted environment data and the candidate environment data are compared to determine outliers in the candidate environment data.

According to a photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention, the training environment data set is input into a preset neural network, and the neural network is controlled to predict the training environment data set, Obtaining the predicted photovoltaic data set corresponding to the training environment data set includes:

Input the training environment data set to the input layer of the neural network, and control the input layer to transmit the training environment data set to the hidden layer of the neural network;

Control each computing node in the hidden layer to calculate the training environment data set to obtain the corresponding initial value;

Input the initial value into the fuzzy function, and control the fuzzy function to perform fuzzy calculation on the initial value to obtain predicted photovoltaic data;

Control the transmission of the predicted photovoltaic data to the output layer of the neural network, and control the output layer output.

The invention also provides a photovoltaic power generation prediction device based on fuzzy reasoning, including:

Acquisition module, used to obtain training environment data and training photovoltaic data;

A dividing module, used to fuzzy divide the training environment data to obtain a training environment data set, and to fuzzy divide the training photovoltaic data to obtain a training photovoltaic data set, where the training environment data set includes several pairs of environmental data Yes, the environment data pair includes a training environment data value and an environment label corresponding to each training environment data value. The training photovoltaic data set includes several pairs of photovoltaic data pairs, and the photovoltaic data pairs include training photovoltaic data values and The photovoltaic label corresponding to each of the training photovoltaic data values;

An input module for inputting the training environment data set into a preset neural network, and controlling the neural network to predict the training environment data set to obtain a predicted photovoltaic data set corresponding to the training environment data set. , the neural network includes a fuzzy function;

An adjustment module, configured to adjust parameters of the neural network according to the predicted photovoltaic data set and the training photovoltaic data set until the neural network converges to obtain a photovoltaic prediction model;

A prediction module, configured to calculate the current photovoltaic prediction value corresponding to the current environmental data based on the photovoltaic prediction model when current environmental data is obtained.

The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above fuzzy inference-based methods. Photovoltaic power generation forecasting method.

The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, any one of the above fuzzy inference-based photovoltaic power generation prediction methods is implemented.

The photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention can achieve more accurate photovoltaic data prediction and provide important support for the power scheduling and stable power supply of the photovoltaic power generation system. The present invention obtains the training environment data and the training photovoltaic data and fuzzy divides them to obtain the training environment data set and the training photovoltaic data set. This makes full use of multi-source information, further combines environmental factors, expands the input dimension, and more comprehensively reflects the operating status and influencing factors of the photovoltaic power generation system, improving the accuracy and reliability of prediction. Then the training environment data set is input into the preset neural network, and the neural network is controlled to predict it to obtain a predicted photovoltaic data set. Then, according to the predicted photovoltaic data set and the training photovoltaic data set, the parameters of the neural network are adjusted until the neural network converges, and the photovoltaic prediction model is obtained. This restores human intuition and thinking and better captures complex nonlinear relationships by giving full play to the nonlinear modeling capabilities of neural networks and the human intelligence characteristics of fuzzy reasoning. Therefore, based on the photovoltaic prediction model, the current photovoltaic prediction value corresponding to the current environmental data is calculated more accurately and has higher adaptability.

Description of the drawings

In order to explain the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are of the present invention. For some embodiments of the invention, those of ordinary skill in the art can also obtain other drawings based on these drawings without exerting creative efforts.

Figure 1 is a schematic flow chart of a photovoltaic power generation prediction method based on fuzzy reasoning provided by the present invention;

Figure 2 is a schematic structural diagram of a photovoltaic power generation prediction device based on fuzzy reasoning provided by the present invention;

Figure 3 is a schematic structural diagram of the electronic device provided by the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention more clear, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention. , not all examples. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present invention.

The following describes a photovoltaic power generation prediction method based on fuzzy reasoning of the present invention in conjunction with Figure 1, which includes:

S100. Obtain training environment data and training photovoltaic data;

S200. Perform fuzzy division on the training environment data to obtain a training environment data set. Fuzzy division on the training photovoltaic data to obtain a training photovoltaic data set. The training environment data set includes several pairs of environmental data. The environment The data pairs include training environment data values and environment labels corresponding to each of the training environment data values. The training photovoltaic data set includes several pairs of photovoltaic data pairs. The photovoltaic data pairs include training photovoltaic data values and each of the training The photovoltaic tag corresponding to the photovoltaic data value;

S300. Input the training environment data set into a preset neural network, and control the neural network to predict the training environment data set to obtain a predicted photovoltaic data set corresponding to the training environment data set. Neural networks include fuzzy functions;

S400. According to the predicted photovoltaic data set and the training photovoltaic data set, adjust parameters of the neural network until the neural network converges to obtain a photovoltaic prediction model;

S500. When obtaining current environmental data, calculate the current photovoltaic prediction value corresponding to the current environmental data based on the photovoltaic prediction model.

Specifically, enough data is collected to train the neural network to build a photovoltaic prediction model. Training environment data and training photovoltaic data can be obtained from actual photovoltaic power plants or simulated photovoltaic systems, or can be downloaded from public databases or websites. Training environment data refers to the environment data used to train subsequent neural networks. The training environment data is the temperature, radiation, humidity, wind speed, etc. corresponding to a certain moment. Training photovoltaic data refers to the photovoltaic data used to train the neural network. In order to ensure the rationality of the training results, the training photovoltaic data and the training environment data correspond to the same time. The time gap between different data can be in days or hours, and can be adjusted according to the demand for forecast accuracy. These data can be stored and represented in the form of tables or matrices, with each row corresponding to a moment and each column corresponding to an input signal. In order to facilitate training, the training environment data and training photovoltaic data can be standardized. For example, all inputs can be converted into the range of 0~1 using maximum and minimum normalization, which is helpful for the training of neural networks.

Then the continuous numerical data is converted into discrete categorical data to facilitate the input and output of the neural network. Fuzzy partitioning is a method that divides numerical data into several intervals according to certain rules and standards, and assigns a label to each interval. Fuzzy divisions can be set based on actual conditions and experience, or can be generated automatically or semi-automatically using some algorithms or tools. The training environment data set includes several pairs of environment data, and the pair of environment data includes a training environment data value and an environment label corresponding to each training environment data value. Taking temperature as an example, environmental labels can include high temperature, medium temperature, and low temperature. The training photovoltaic data set includes several pairs of photovoltaic data, and the photovoltaic data pairs include training photovoltaic data values and photovoltaic tags corresponding to each of the training photovoltaic data values. Taking photovoltaic power as an example, photovoltaic labels can include high power, medium power and low power.

By fuzzy dividing the training environment data and training photovoltaic data, the diversity of the data can be increased and avoid over-fitting or under-fitting during the model training process due to accurate values. On the other hand, fuzzy division can better divide the data into different groups, highlight its representative characteristics through environmental labels and photovoltaic labels, rather than its original characteristics, and reduce the difficulty of the prediction process. In addition, photovoltaic power generation scenarios often have interference from special environments. Fuzzy division can weaken the interference from special environments, allowing subsequent models to have better applicability to the scenario.

The neural network is then used to learn the mapping relationship from environmental variables to photovoltaic output and obtain a photovoltaic prediction model. Neural network is a nonlinear model composed of multiple nodes (neurons) and connections (weights), which can approximate any complex function by adjusting the weights. There are many types and structures of neural networks. In this embodiment, the forward propagation neural network is used, that is, the number of input layer nodes is N (corresponding to N input signals), and the hidden layer nodes can be freely adjusted according to needs. In this embodiment, it is set is 10, and the number of output layer nodes is 1 (corresponding to 1 output signal). The number of input layer nodes is related to the number of input environmental labels and photovoltaic labels. Each environmental data pair in the training environment data set is used as the input of the neural network, and through nonlinear transformation of the hidden layer, the predicted photovoltaic data value of the output layer is obtained. These predicted PV data values constitute the predicted PV data set corresponding to the training environment data set.

There is a gap between the predicted PV data set and the training PV data set, which can be represented by a loss value. Calculate the loss value between the predicted photovoltaic data set and the training photovoltaic data set, and based on the loss value, optimize the parameters of the neural network so that it can better fit the training data, thereby improving the prediction accuracy. The photovoltaic labels in the photovoltaic data set represent the relative position of a certain photovoltaic value in the entire data set, while the photovoltaic labels in the predicted photovoltaic data set represent the prediction of the data characteristics of the entire data set. In this embodiment, in addition to the traditional comparison of loss values between values, it also includes the comparison of photovoltaic labels. Therefore, during the training process, the direction of training can be determined faster, the speed of training can be accelerated, and at the same time, it can be accurately Predicting photovoltaic labels can also better improve the overall data prediction effect, thereby improving the accuracy of the entire data.

The process of reversely transmitting the loss value to the neural network and updating the weight according to certain rules and methods. This process can use the gradient descent method, stochastic gradient descent method, Newton method, etc. In this embodiment, the gradient descent method is used, that is, by calculating the mean square error between the predicted photovoltaic data and the training photovoltaic data as the loss function, and solving the partial derivative of the loss function with respect to the weight as the gradient, and then in the opposite direction of the gradient with a certain step size to update the weights. This process can be repeated multiple times until the loss function reaches the minimum value or changes very little, that is, the neural network converges. At this point, the parameters of the neural network are determined, and the photovoltaic prediction model is obtained.

Use already trained photovoltaic prediction models to predict photovoltaic output for new or future environmental data. Current environmental data refers to temperature, radiation, humidity, wind speed, etc. at a certain moment or within a certain period of time. The current environmental data also needs to be fuzzy divided to obtain the current environmental data set, including several pairs of environmental data. Each pair of environmental data in the current environmental data set is used as the input of the neural network, and through the calculation of the hidden layer and the output layer, the current photovoltaic prediction value is obtained. These current photovoltaic prediction values constitute the current photovoltaic prediction result set, which can be used to evaluate, analyze or control the performance and status of the photovoltaic system.

This invention makes full use of multi-source information, combines environmental factors, expands the input dimension, and improves the accuracy of prediction by acquiring training data, performing fuzzy division, inputting into a neural network, performing parameter adjustment, obtaining a photovoltaic prediction model, calculating current photovoltaic prediction values, etc. and reliability. It can also give full play to the nonlinear modeling capabilities of neural networks and the human intelligence characteristics of fuzzy reasoning, restore human intuition and thinking, better capture complex nonlinear relationships, and improve the universality and robustness of predictions. Ultimately, more accurate predictions of photovoltaic power or power generation can be achieved, providing important support for power dispatching and stable power supply of photovoltaic power generation systems.

In another implementation, the fuzzy division of the training environment data to obtain the training environment data set includes:

Calculate the environment feature vector corresponding to each of the training environment data;

According to the similarity between the environmental feature vectors, perform preliminary classification on the training environment data to obtain several initial environment clusters;

Perform similarity calculation on the initial environment clustering to obtain a clustering similarity value;

According to the cluster similarity value, the initial environment clusters are merged, and the data in the merged initial environment clusters are updated until the initial environment clusters meet the preset stopping criteria, and several A target environment cluster and an environment label corresponding to the target environment cluster;

According to the target environment clustering and the environment label, an environment data pair is generated.

Specifically, the training environment data is first converted into a numerical environment feature vector for subsequent similarity calculation and cluster analysis. Here, the environment feature vector can be converted to the standardized method mentioned above. The result of calculating the environment feature vector is a matrix, in which each row represents a training environment data and each column represents a feature dimension.

Then, based on the similarity between the environment feature vectors, similar training environment data are grouped into a group to form initial environment clustering. Similarity is a numerical value that measures the degree of similarity between two data, such as Euclidean distance and cosine similarity. The greater the similarity, the more likely it is that the two data are in the same class. Therefore, based on the similarity value, the training environment data can be initially classified and several initial environment clusters can be obtained. The number of initial classifications can be determined based on a preset similarity value threshold, or can be specified before classification.

Then, based on the similarity between the initial environment clusters, cluster similarity values are obtained for subsequent merging operations. The clustering similarity value is a value that measures the degree of similarity between two clusters, such as average distance and minimum distance. When calculating the clustering similarity, the cluster center, cluster boundary, cluster internal distribution, etc. can be used as the numerical values used in the calculation.

Then according to the cluster similarity value, the initial environment clusters with similarity higher than a certain threshold are merged into a larger cluster, and the data and environment feature vectors in the merged cluster are updated to improve the clustering. Class quality and accuracy. The merge operation is to merge two or more clusters into one cluster, and the update operation is to recalculate or adjust the data and environment feature vectors in the merged clusters, such as using a weighted average. A stopping criterion is set in advance to determine whether to continue the merging operation. In this embodiment, the maximum number of clusters and the maximum number of iterations can be used as the stopping criterion. The final cluster obtained after the merging operation is completed is the target environment cluster. By naming or numbering each target environment cluster, the environment label corresponding to each target environment cluster is obtained.

After obtaining the target environment clusters and environment labels, environment data pairs can be generated. An environment data pair is a data structure composed of two elements, where the first element is a training environment data, and the second element is the environment label of the target environment cluster to which the training environment data belongs.

Through similar fuzzy division, the training photovoltaic data is divided to obtain the training photovoltaic data set, which specifically includes:

Calculate the photovoltaic feature vector corresponding to each of the training photovoltaic data;

According to the similarity between the photovoltaic feature vectors, perform preliminary classification on the training photovoltaic data to obtain several initial photovoltaic clusters;

Perform similarity calculation on the initial photovoltaic clustering to obtain a clustering similarity value;

According to the cluster similarity value, the initial photovoltaic clusters are merged, and the data in the merged initial photovoltaic clusters are updated until the initial photovoltaic clusters meet the preset stopping criteria, and several A target photovoltaic cluster and a photovoltaic label corresponding to the target photovoltaic cluster;

Photovoltaic data pairs are generated based on the target photovoltaic clustering and the photovoltaic tag.

The above implementation method can dynamically adjust the number and scope of clusters according to different environmental characteristics, photovoltaic characteristics and conditions, thereby better reflecting the diversity and complexity of photovoltaic power generation. Flexibly merging or splitting clusters based on their similarity values also helps capture changes and trends in photovoltaic power generation.

In another implementation, the training environment data is initially classified according to the similarity between the environment feature vectors, and several initial environment clusters are obtained, including:

Split the environment feature vector to obtain the first environment comparison value and the second environment comparison value corresponding to the environment feature vector;

Calculate a first similarity between the first environment comparison values;

According to the first similarity, divide the environmental feature vector to obtain a first cluster;

Calculating a second degree of similarity between the second environment comparison values in the first cluster;

According to the second similarity, the environment feature vector is divided to obtain initial environment clustering.

Specifically, the environmental feature vector is decomposed into two sub-vectors, representing different environmental attributes respectively. For example, if the environmental feature vector is a four-dimensional vector representing environmental factors such as temperature, humidity, irradiance, and wind speed, it can be split into two two-dimensional vectors representing temperature, humidity, and radiance respectively. In this way, comparisons and clustering can be performed based on different environmental attributes.

Calculate the degree of similarity between each two data on the first environment comparison value. The higher the similarity, the closer the two data are in the first environment comparison value. Then calculate the degree of similarity between each two data on the second environment comparison value. Similar to the previous step of "calculating the first similarity between the first environment comparison values", different measurement methods can also be used to calculate the similarity. The difference is that here only the similarity between data within the same category needs to be calculated, without considering the similarity between different categories.

The data is further divided into smaller categories according to the second similarity, so that the data within the same category are also similar in the second environment comparison value, and the data between different categories are also similar in the second environment comparison value. Degree is low. This way, the data can be sorted by minor differences in the comparison values of the second environment. Finally, the initial environment clustering based on the environment feature vector is obtained.

Through a similar clustering method, the photovoltaic feature vectors can be initially classified to obtain the initial photovoltaic clustering, which specifically includes:

Split the photovoltaic feature vector to obtain the first photovoltaic comparison value and the second photovoltaic comparison value corresponding to the photovoltaic feature vector;

Calculating a first degree of similarity between the first photovoltaic comparison values;

According to the first similarity, divide the photovoltaic feature vector to obtain a first cluster;

calculating a second degree of similarity between the second photovoltaic comparison values in the first cluster;

According to the second similarity, the photovoltaic feature vector is divided to obtain initial photovoltaic clustering.

In another implementation, before obtaining the training environment data and the training photovoltaic data, the method further includes:

Obtain historical environmental data, update environmental data, historical photovoltaic data and update photovoltaic data;

According to the preset update cycle, the updated environment data and historical environment data are selected to obtain training environment data; and,

The historical photovoltaic data and the updated photovoltaic data are selected to obtain training photovoltaic data.

Specifically, relevant data is collected from different data sources, such as weather stations, satellites, sensors, etc. Historical environmental data refers to environmental data in the past period, including factors that affect photovoltaic power generation, such as temperature, humidity, wind speed, and radiation. Updated environmental data refers to the most recently collected environmental data. Historical photovoltaic data refers to the output power or electricity of the solar power generation system, and updated photovoltaic data refers to the most recently collected photovoltaic data.

As time goes by, historical environmental data may not necessarily be applicable to the existing environment. Therefore, it is necessary to select historical environmental data, updated environmental data, historical photovoltaic data and updated photovoltaic data to select training environment data suitable for training in the near future. and training PV data.

The update period refers to the time range of the extracted data distribution. For example weekly, monthly or quarterly. The selection method can be selected according to different objectives and conditions, such as random sampling, stratified sampling or group sampling, etc. According to the update cycle, training environment data and training photovoltaic data can be selected to increase the adaptability of the photovoltaic prediction model obtained by subsequent training.

In another implementation, selecting the updated environment data and historical environment data according to a preset update cycle to obtain training environment data includes:

According to the update cycle, determine the candidate environment data among the updated environment data and the historical environment data;

Compare the candidate environment data with reference environment data corresponding to the candidate environment time, and determine outliers in the candidate environment data, where the candidate environment time is the time corresponding to the candidate environment data;

According to the abnormal value and the candidate environment time, the candidate environment data is adjusted to obtain training environment data.

Specifically, both the updated environment data and the historical environment data exist at a time corresponding to each environment data. Therefore, the candidate environment data in the updated environment data and the historical environment data can be determined according to the time range in the update cycle. For example, if the updated environment data corresponds to one week and the surrounding area is updated to two weeks, then the most recent week of data is extracted from the historical environment data as candidate environment data.

The time period corresponding to the candidate environment data is the candidate environment time. By comparing the historical environment data corresponding to the candidate environment time, the fluctuation value can be obtained. For example, if the candidate environmental data corresponds to the period from July 1 to July 31, the historical environmental data of last year and the year before that can be used as the corresponding reference environmental data, or the environment corresponding to June 1 to June 30 can be used. data as reference environmental data. Based on the candidate environment data and the reference environment data, outliers in the candidate environment data can be determined. Outliers refer to candidate environmental data that deviate from the normal range compared with reference environmental data. For example, if the temperature in the candidate environment data is higher or lower than the temperature in the reference environment data by a certain threshold, then this temperature is an outlier.

A reasonable threshold is set in advance to determine whether the difference between the candidate environment data and the reference environment data exceeds the normal range. The selection of threshold can be determined based on different environmental factors and data characteristics. For example, statistical methods such as mean, standard deviation, quantile, etc. are used to calculate the central trend and degree of dispersion of reference environmental data, and then the threshold is determined based on a certain multiple or proportion. Compare the candidate environment data with the reference environment data, find those candidate environment data that exceed the threshold, and mark them as outliers.

After excluding outliers, the number of candidate environmental data will be reduced. Therefore, based on the update period, some environmental data are selected before the selected historical environmental data, so that the number of finally selected environmental data is the same as the number of environmental data corresponding to the update period. . In the process of adding historical environment data, outlier detection will also be performed to ensure that the data in the training environment data is accurate.

The method of obtaining training photovoltaic data is similar, including:

Determine candidate photovoltaic data in the updated photovoltaic data and the historical photovoltaic data according to the update cycle;

Compare the candidate photovoltaic data with the reference photovoltaic data corresponding to the candidate photovoltaic time, and determine the abnormal values in the candidate photovoltaic data, wherein the candidate photovoltaic time is the time corresponding to the candidate photovoltaic data;

According to the abnormal value and the candidate photovoltaic time, the candidate photovoltaic data is adjusted to obtain training photovoltaic data.

In another implementation, comparing the candidate environment data with the reference environment data corresponding to the candidate environment time, and determining the abnormal values in the candidate environment data includes:

Based on the reference environmental data, determine prediction parameters;

Generate an environment prediction model based on the reference environment data and the prediction parameters;

Based on the environment prediction model, calculate the predicted environment data corresponding to the candidate environment time;

The predicted environment data and the candidate environment data are compared to determine outliers in the candidate environment data.

Specifically, differential processing is performed on the reference environmental data to eliminate trends and seasonal factors in the data and make it a stationary sequence. The number and order of differential processing can be selected according to the characteristics and needs of the data. Then, calculate the prediction parameters corresponding to the reference environmental data. Taking the ARIMA model as an example, the prediction parameters include the autoregressive term (p), the moving average term (q) and the number of differences (d). Then an environmental prediction model is generated based on the reference environmental data and prediction parameters. Still taking the ARIMA model as an example, since the basic structure of the ARIMA model is known, the environmental prediction model can be obtained by substituting the reference environmental data and prediction parameters into the basic structure of the model, and testing the fitting effect and residuals of the environmental prediction model. distributed. If the environmental prediction model does not fit the characteristics of the data or there are significant residual correlations, the parameters need to be re-tuned. Based on the environment prediction model, the predicted environment data corresponding to the candidate environment time can be calculated. For example, the candidate environment time is input into a fitted environment prediction model to obtain corresponding predicted environment data.

The predicted environment data and the candidate environment data are compared to determine outliers in the candidate environment data. There can be many methods of comparison, such as calculating the difference, ratio, correlation, etc. between the two.

In another implementation, the neural network includes an input layer, a hidden layer and an output layer. The hidden layer includes several computing nodes and fuzzy functions. The training environment data set is input to a preset neural network. network, and control the neural network to predict the training environment data set, and obtain the predicted photovoltaic data set corresponding to the training environment data set including:

Input the training environment data set to the input layer, and control the input layer to transmit the training environment data set to the computing node;

Control each of the computing nodes to calculate the training environment data set and obtain the corresponding initial value;

Input the initial value into the fuzzy function, and control the fuzzy function to perform fuzzy calculation on the initial value to obtain predicted photovoltaic data;

Control the transmission of the predicted photovoltaic data to the output layer, and control the output layer to output.

Specifically, the training environment data set is first converted into a format suitable for neural network processing and distributed to different computing nodes. A computing node is a unit that performs specific operations in a neural network and is usually composed of multiple neurons. Define the structure of the input layer in advance, such as how many neurons the input layer has, how many feature values each neuron receives, and how the input layer is connected to the computing nodes.

Then, the environmental data set needs to be preprocessed, such as normalization, normalization, missing value handling, etc., so that the input layer can correctly receive and parse the data and transmit it to the computing node.

Then let the computing node perform corresponding operations based on the data from the input layer to obtain the initial value. The initial value refers to the value output by the calculation node that has not been processed by the activation function. Define the structure of the computing node in advance, including how many neurons the computing node has, how many weights and biases each neuron has, and how the computing nodes are connected. In addition, it is also necessary to define the operation method of the calculation node, such as using weighted sum, matrix multiplication, convolution and other methods. The calculation method of the computing node in this embodiment is preferably weighted summation. The initial value is then converted into predicted photovoltaic data through a fuzzy function. Fuzzy functions refer to mathematical functions that can handle uncertainty and ambiguity. Preset the type of fuzzy function, including the use of fuzzy logic, fuzzy sets, fuzzy rules, etc., as well as the parameters of the fuzzy function, such as the use of membership functions, fuzzy operators, and inference mechanisms.

Finally, the predicted PV data are transferred from the fuzzy function to the output layer and displayed or saved by the output layer. The output layer is the layer in the neural network responsible for outputting the final result. Setting the output layer includes the number of neurons, the connection method between the output layer and the fuzzy function, etc.

Finally, define the output method of the output layer, such as denormalization and denormalization. Control the output layer output data.

This method uses fuzzy functions to deal with uncertainty and ambiguity, thereby improving the accuracy and robustness of predictions. Through the combination of fuzzy theory and neural network technology, the accuracy and adaptability of photovoltaic prediction are improved.

Referring to FIG. 2 , the photovoltaic power generation prediction device based on fuzzy reasoning provided by the present invention is described below. The photovoltaic power generation prediction device based on fuzzy reasoning described below and the photovoltaic power generation prediction method based on fuzzy reasoning described above can correspond to each other. The apparatus includes an acquisition module 210, a partitioning module 220, an input module 230, an adjustment module 240 and a prediction module 250.

The acquisition module 210 is used to acquire training environment data and training photovoltaic data;

The division module 220 is used to perform fuzzy division on the training environment data to obtain a training environment data set, and perform fuzzy division on the training photovoltaic data to obtain a training photovoltaic data set. The training environment data set includes several pairs of environmental data pairs, The environmental data pairs include training environment data values and environmental labels corresponding to each training environment data value. The training photovoltaic data set includes several pairs of photovoltaic data pairs. The photovoltaic data pairs include training photovoltaic data values and each The photovoltaic tag corresponding to the training photovoltaic data value;

The input module 230 is used to input the training environment data set into a preset neural network, and control the neural network to predict the training environment data set to obtain a predicted photovoltaic data set corresponding to the training environment data set. , the neural network includes a fuzzy function;

The adjustment module 240 is configured to adjust parameters of the neural network according to the predicted photovoltaic data set and the training photovoltaic data set until the neural network converges to obtain a photovoltaic prediction model;

The prediction module 250 is configured to calculate the current photovoltaic prediction value corresponding to the current environmental data based on the photovoltaic prediction model when current environmental data is obtained.

Figure 3 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 3, the electronic device may include: a processor (processor) 310, a communications interface (Communications Interface) 320, a memory (memory) 330 and a communication bus 340. Among them, the processor 310, the communication interface 320, and the memory 330 complete communication with each other through the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a photovoltaic power generation prediction method based on fuzzy reasoning, which method includes:

Obtain training environment data and training photovoltaic data;

Fuzzy division is performed on the training environment data to obtain a training environment data set. Fuzzy division is performed on the training photovoltaic data to obtain a training photovoltaic data set. The training environment data set includes several pairs of environmental data. The pairs of environmental data are including training environment data values and environment labels corresponding to each training environment data value; the training photovoltaic data set includes several pairs of photovoltaic data pairs; the photovoltaic data pairs include training photovoltaic data values and each of the training photovoltaic data The photovoltaic tag corresponding to the value;

The training environment data set is input into a preset neural network, and the neural network is controlled to predict the training environment data set to obtain a predicted photovoltaic data set corresponding to the training environment data set. The neural network Includes fuzzy functions;

According to the predicted photovoltaic data set and the training photovoltaic data set, adjust parameters of the neural network until the neural network converges to obtain a photovoltaic prediction model;

When current environmental data is obtained, the current photovoltaic prediction value corresponding to the current environmental data is calculated based on the photovoltaic prediction model.

In addition, the above-mentioned logical instructions in the memory 330 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. .

In another aspect, the present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored, which is implemented when executed by a processor to execute the fuzzy inference-based photovoltaic power generation prediction method provided by the above methods. , the method includes:

Obtain training environment data and training photovoltaic data;

Fuzzy division is performed on the training environment data to obtain a training environment data set. Fuzzy division is performed on the training photovoltaic data to obtain a training photovoltaic data set. The training environment data set includes several pairs of environmental data. The pairs of environmental data are including training environment data values and environment labels corresponding to each training environment data value; the training photovoltaic data set includes several pairs of photovoltaic data pairs; the photovoltaic data pairs include training photovoltaic data values and each of the training photovoltaic data The photovoltaic tag corresponding to the value;

The training environment data set is input into a preset neural network, and the neural network is controlled to predict the training environment data set to obtain a predicted photovoltaic data set corresponding to the training environment data set. The neural network Includes fuzzy functions;

According to the predicted photovoltaic data set and the training photovoltaic data set, adjust parameters of the neural network until the neural network converges to obtain a photovoltaic prediction model;

When current environmental data is obtained, the current photovoltaic prediction value corresponding to the current environmental data is calculated based on the photovoltaic prediction model.

The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in One location, or it can be distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Persons of ordinary skill in the art can understand and implement the method without any creative effort.

Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the part of the above technical solution that essentially contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disc, optical disk, etc., including a number of instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be used Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent substitutions are made to some of the technical features; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. A photovoltaic power generation prediction method based on fuzzy reasoning is characterized by comprising the following steps:

acquiring training environment data and training photovoltaic data;

performing fuzzy division on the training environment data to obtain a training environment data set, performing fuzzy division on the training photovoltaic data to obtain a training photovoltaic data set, wherein the training environment data set comprises a plurality of pairs of environment data pairs, each pair of environment data comprises a training environment data value and an environment label corresponding to each training environment data value, the training photovoltaic data set comprises a plurality of pairs of photovoltaic data, and each pair of photovoltaic data comprises a training photovoltaic data value and a photovoltaic label corresponding to each training photovoltaic data value;

Inputting the training environment data set into a preset neural network, and controlling the neural network to predict the training environment data set to obtain a predicted photovoltaic data set corresponding to the training environment data set, wherein the neural network comprises a fuzzy function;

according to the predicted photovoltaic data set and the training photovoltaic data set, parameter adjustment is carried out on the neural network until the neural network converges, and a photovoltaic prediction model is obtained;

when current environment data are acquired, calculating a current photovoltaic predicted value corresponding to the current environment data based on the photovoltaic predicted model;

the performing fuzzy division on the training environment data to obtain a training environment data set comprises:

calculating an environment characteristic vector corresponding to each training environment data;

according to the similarity between the environment feature vectors, primarily classifying the training environment data to obtain a plurality of initial environment clusters;

performing similarity calculation on the initial environment clusters to obtain cluster similarity values;

combining the initial environment clusters according to the cluster similarity value, and updating the data in the combined initial environment clusters until the initial environment clusters meet a preset stopping standard to obtain a plurality of target environment clusters and environment labels corresponding to the target environment clusters;

Generating an environment data pair according to the target environment cluster and the environment label;

the step of performing fuzzy division on the training photovoltaic data to obtain a training photovoltaic data set comprises the following steps: calculating a photovoltaic sign vector corresponding to each piece of training photovoltaic data;

according to the similarity between the photovoltaic sign vectors, primarily classifying the training photovoltaic data to obtain a plurality of initial lights Fu Julei;

performing similarity calculation on the initial photovoltaic clusters to obtain cluster similarity values;

combining the initial photovoltaic clusters according to the cluster similarity value, and updating the data in the combined initial photovoltaic clusters until the initial photovoltaic clusters meet a preset stopping standard to obtain a plurality of target photovoltaic clusters and photovoltaic labels corresponding to the target photovoltaic clusters;

generating a photovoltaic data pair according to the target photovoltaic cluster and the photovoltaic label;

the initial classification of the training environment data according to the similarity between the environment feature vectors, and the obtaining of a plurality of initial environment clusters comprises:

splitting the environment feature vector to obtain a first environment comparison value and a second environment comparison value corresponding to the environment feature vector;

Calculating a first similarity between the first environmental comparison values;

dividing the environmental feature vectors according to the first similarity to obtain a first cluster;

calculating a second similarity between the second environmental comparison values in the first cluster;

dividing the environment feature vector according to the second similarity to obtain an initial environment cluster;

the training photovoltaic data is initially classified according to the similarity between the photovoltaic sign vectors to obtain a plurality of initial photovoltaic clusters, including:

splitting the photovoltaic sign vector to obtain a first photovoltaic comparison value and a second photovoltaic comparison value corresponding to the photovoltaic sign vector;

calculating a third similarity between the first photovoltaic comparison values;

dividing the photovoltaic sign vector according to the third similarity to obtain a second cluster;

calculating a fourth similarity between the second photovoltaic comparison values in the second clusters;

and dividing the photovoltaic sign vector according to the fourth similarity to obtain an initial photovoltaic cluster.

2. The fuzzy inference based photovoltaic power generation prediction method of claim 1, further comprising, prior to the acquiring the training environment data and the training photovoltaic data:

Acquiring historical environment data, updated environment data, historical photovoltaic data and updated photovoltaic data;

according to a preset updating period, the updating environment data and the historical environment data are decimated to obtain training environment data; the method comprises the steps of,

and decimating the historical photovoltaic data and the updated photovoltaic data to obtain training photovoltaic data.

3. The fuzzy inference-based photovoltaic power generation prediction method according to claim 2, wherein the decimating the updated environmental data and the historical environmental data according to a preset update period to obtain training environmental data includes:

determining candidate environmental data in the updated environmental data and the historical environmental data according to the updating period;

comparing the candidate environment data with reference environment data corresponding to candidate environment time, and determining an abnormal value in the candidate environment data, wherein the candidate environment time is the time corresponding to the candidate environment data;

and adjusting the candidate environment data according to the abnormal value and the candidate environment time to obtain training environment data.

4. The fuzzy inference-based photovoltaic power generation prediction method of claim 3, wherein the comparing the candidate environmental data with reference environmental data corresponding to a candidate environmental time, determining an outlier in the candidate environmental data comprises:

Determining a prediction parameter based on the reference environmental data;

generating an environment prediction model according to the reference environment data and the prediction parameters;

generating predicted environment data corresponding to the candidate environment time based on the environment prediction model;

comparing the predicted environment data with the candidate environment data, and determining an abnormal value in the candidate environment data.

5. The photovoltaic power generation prediction method based on fuzzy inference according to any one of claims 1 to 4, wherein the inputting the training environment data set into a preset neural network, and controlling the neural network to predict the training environment data set, to obtain a predicted photovoltaic data set corresponding to the training environment data set includes:

inputting the training environment data set to an input layer of the neural network, and controlling the input layer to transmit the training environment data set to a hidden layer of the neural network;

controlling each computing node in the hidden layer to compute the training environment data set to obtain a corresponding initial value;

inputting the initial value into the fuzzy function, and controlling the fuzzy function to perform fuzzy calculation on the initial value to obtain predicted photovoltaic data;

And controlling the predicted photovoltaic data to be transmitted to an output layer of the neural network, and controlling the output layer to output.

6. A photovoltaic power generation prediction device based on fuzzy reasoning is characterized by comprising:

the acquisition module is used for acquiring training environment data and training photovoltaic data;

the system comprises a dividing module, a fuzzy dividing module and a fuzzy dividing module, wherein the dividing module is used for carrying out fuzzy dividing on training environment data to obtain a training environment data set and carrying out fuzzy dividing on the training photovoltaic data to obtain a training photovoltaic data set, the training environment data set comprises a plurality of pairs of environment data pairs, the environment data pairs comprise training environment data values and environment labels corresponding to each training environment data value, the training photovoltaic data set comprises a plurality of pairs of photovoltaic data, and the photovoltaic data pairs comprise training photovoltaic data values and photovoltaic labels corresponding to each training photovoltaic data value;

the input module is used for inputting the training environment data set into a preset neural network, controlling the neural network to predict the training environment data set, and obtaining a predicted photovoltaic data set corresponding to the training environment data set, wherein the neural network comprises a fuzzy function;

The adjustment module is used for carrying out parameter adjustment on the neural network according to the predicted photovoltaic data set and the training photovoltaic data set until the neural network converges to obtain a photovoltaic prediction model;

the prediction module is used for calculating a current photovoltaic predicted value corresponding to the current environment data based on the photovoltaic predicted model when the current environment data are acquired;

the performing fuzzy division on the training environment data to obtain a training environment data set comprises:

calculating an environment characteristic vector corresponding to each training environment data;

according to the similarity between the environment feature vectors, primarily classifying the training environment data to obtain a plurality of initial environment clusters;

performing similarity calculation on the initial environment clusters to obtain cluster similarity values;

combining the initial environment clusters according to the cluster similarity value, and updating the data in the combined initial environment clusters until the initial environment clusters meet a preset stopping standard to obtain a plurality of target environment clusters and environment labels corresponding to the target environment clusters;

generating an environment data pair according to the target environment cluster and the environment label;

The step of performing fuzzy division on the training photovoltaic data to obtain a training photovoltaic data set comprises the following steps: calculating a photovoltaic sign vector corresponding to each piece of training photovoltaic data;

according to the similarity between the photovoltaic sign vectors, primarily classifying the training photovoltaic data to obtain a plurality of initial lights Fu Julei;

performing similarity calculation on the initial photovoltaic clusters to obtain cluster similarity values;

combining the initial photovoltaic clusters according to the cluster similarity value, and updating the data in the combined initial photovoltaic clusters until the initial photovoltaic clusters meet a preset stopping standard to obtain a plurality of target photovoltaic clusters and photovoltaic labels corresponding to the target photovoltaic clusters;

generating a photovoltaic data pair according to the target photovoltaic cluster and the photovoltaic label;

the initial classification of the training environment data according to the similarity between the environment feature vectors, and the obtaining of a plurality of initial environment clusters comprises:

splitting the environment feature vector to obtain a first environment comparison value and a second environment comparison value corresponding to the environment feature vector;

calculating a first similarity between the first environmental comparison values;

Dividing the environmental feature vectors according to the first similarity to obtain a first cluster;

calculating a second similarity between the second environmental comparison values in the first cluster;

dividing the environment feature vector according to the second similarity to obtain an initial environment cluster;

the training photovoltaic data is initially classified according to the similarity between the photovoltaic sign vectors to obtain a plurality of initial photovoltaic clusters, including:

splitting the photovoltaic sign vector to obtain a first photovoltaic comparison value and a second photovoltaic comparison value corresponding to the photovoltaic sign vector;

calculating a third similarity between the first photovoltaic comparison values;

dividing the photovoltaic sign vector according to the third similarity to obtain a second cluster;

calculating a fourth similarity between the second photovoltaic comparison values in the second clusters;

and dividing the photovoltaic sign vector according to the fourth similarity to obtain an initial photovoltaic cluster.

7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the fuzzy inference based photovoltaic power generation prediction method of any of claims 1 to 5 when the computer program is executed by the processor.

8. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the fuzzy inference based photovoltaic power generation prediction method of any of claims 1 to 5.