WO2025112382A1 - Short-term wind power forecasting method based on progressive deep learning of multi-source data - Google Patents

Abstract

Disclosed in the present invention is a short-term wind power forecasting method based on progressive deep learning of multi-source data, comprising the following steps: S1, training a forecasting model by means of a deep learning network structure; and S2, computing wind speed data of each wind turbine by means of the forecasting model, and carrying out wind power forecasting. The present invention has high forecasting accuracy; a wind speed forecasting model is trained on the basis of multi-source data, and thus, the forecasting accuracy of computing the actual wind speed can be greatly improved.

Description

A short-term wind power forecasting method based on progressive deep learning of multi-source data Technical Field

The present invention relates to the technical field of wind power prediction, and in particular to a short-term wind power prediction method based on multi-source data progressive deep learning.

Background Art

Wind power forecasting is of vital importance in the field of renewable energy. As the global demand for clean energy continues to increase, wind energy, as a green and sustainable energy source, has attracted increasing attention. However, the instability and randomness of wind energy have brought challenges to the operation of the power system. Therefore, accurate wind power forecasting, especially short-term wind power forecasting, is crucial to reduce wind abandonment, optimize the daily power generation plan and cold and hot standby of conventional power sources, and adjust the maintenance plan to more accurately reflect the utilization efficiency of wind energy resources, which is crucial to optimizing the operation of the power system and ensuring the stability and economy of power supply.

Wind power prediction, its current relevant data often comes from weather forecasts, which simplify and approximate the basic equations of atmospheric motion, and introduce meteorological physical laws and boundary condition constraints. The time resolution of wind speed forecasts in numerical weather forecasts is generally 1 hour and 3 hours, and the spatial resolution is generally 9km×9km and 12.5km×12.5km. For wind farms, the generally required time resolution of wind power prediction is 15 minutes, and the wind speed forecast at the wind turbine location is required. Therefore, the current relevant wind power prediction methods cannot fully meet the actual needs. Among the current related technologies, the main method is to predict wind power for a large area containing a wind farm. It is still difficult to play a role in predicting the wind power of each wind turbine in the wind farm, and the wind power prediction data for a large area is not applicable to each specific wind turbine therein.

Summary of the invention

The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a short-term wind power prediction method based on progressive deep learning of multi-source data to solve the problems raised in the background technology.

In order to solve the above technical problems, the technical solution adopted by the present invention is: a short-term wind power prediction method based on multi-source data progressive deep learning, which comprises the following steps:

S1: Use deep learning network structure to train prediction model;

S2: Use the prediction model to calculate the wind speed data of each wind turbine and predict the wind power;

Furthermore, the step S1 specifically includes the following process:

S1.1. Collect multi-source data required for the prediction model;

S1.2, preprocessing the multi-source data collected in step 1.1;

S1.3, dividing the processed data set;

S1.4. Train the deep learning model to obtain the prediction model function f(x).

Furthermore, the step S2 specifically includes the following process:

S2.1. Collect weather forecast data for a certain wind farm;

S2.2, using the trained prediction model f(x), calculate the wind speed of a certain wind turbine after a period of time;

S2.3. Obtain the wind speed data and wind power data of a certain wind turbine after a period of time.

Furthermore, the step S1.1 is specifically as follows: collecting numerical values and weather forecast data related to wind speed of each wind turbine to form a multi-source data set, and the data related to wind speed includes two parts: one is the weather forecast data updated for a period of time for the wind farm, including wind direction, short-term weather type, rainfall type, and temperature; the other is the data to be collected at each wind turbine location, including wind turbine height, terrain type, temperature, air pressure, humidity, sea level pressure, short-wave radiation from the ground downward, long-wave radiation from the ground downward, and total cloud cover;

The multi-source data are labeled and divided into two groups, one group is the real-time collected data _X1 , _X2 ... _Xn at each wind turbine location, and the other group is the weather forecast data _Y1 , _Y2 ... _Yn a period of time before the real-time collected data of each wind turbine.

Furthermore, in step S1.2, specific measures for data preprocessing include: correcting or deleting incomplete or erroneous data, processing duplicate data, processing outliers, unifying data formats, data normalization and data encoding.

Furthermore, the step S1.3 is specifically as follows: the preprocessed multi-source data is divided into a training set, a validation set and a test set according to a certain ratio, wherein the training set is a data set used to train the machine learning model, the validation set is a data set set aside separately during the deep learning model training process for evaluating model performance and adjusting hyperparameters, and the test set is a data set used to finally evaluate the performance of the machine learning model.

Furthermore, in step S1.4, the deep learning model includes various deep learning neural network structures, and the deep neural network includes an input layer, an output layer and at least one hidden layer; the network parameters are adjusted, and the number of network layers, the number of neurons, batch normalization, and random inactivation are adjusted to train the network until the maximum number of iterations is reached or the network learning rate converges; mean square error is selected as the loss function and mean absolute error is selected as the evaluation accuracy; the number of network layers and the number of neurons are set and adjusted according to the application in different regional scenarios until the value of the loss function converges or the maximum number of iterations is reached, and the model training is completed;

Combine the multi-source data X ₁ , X ₂ , ...X _n collected in real time at each wind turbine location in step S1.1 with the real-time data collected at each wind turbine The multi-source data Y ₁ , Y ₂ ...Y _n of the weather forecast in the previous period of time are used as the input layer, and the number of hidden layers is set according to different data amounts, at least greater than 1 layer. By adjusting the activation function, weight and other parameters, the relationship between the real-time multi-source data X ₁ , X ₂ ...X _n of each wind turbine and the multi-source data Y ₁ , Y ₂ ...Y _n provided by the weather forecast in the previous period of time is obtained through training, that is, the prediction model function f(x) is obtained.

Furthermore, the wind farm weather forecast data collected in step S2.1 includes weather forecast data Y ₁ , Y ₂ ..Y _n of a large area of a certain wind farm;

The step S2.2 specifically comprises: using the trained wind speed prediction model f(x) of each wind turbine, combined with the weather forecast data Y ₁ , Y ₂ ..Y _n collected in step S2.1, to calculate the wind speed data at the location of each wind turbine after a period of time;

The step S2.3 is specifically: obtaining the wind speed data at each wind turbine position after a period of time, and then accurately predicting the wind power data of a specific wind turbine after a period of time.

Beneficial effects of the present invention:

1. The present invention has high prediction accuracy. It uses a wind speed prediction model based on multi-source data, which can greatly improve the prediction accuracy of the actual wind speed calculation;

2. The present invention conducts progressive deep learning on multi-source data to train the prediction model. Progressive deep learning refers to deep learning on the current actual wind speed and other data and the data provided by the weather forecast in the previous period, that is, deep learning is conducted progressively over a period of time, and learning can be performed on the historical parameters related to the wind speed of each wind turbine, which can break through the limitations of the existing forecast accuracy being restricted by the existing meteorological science theory, the time resolution, spatial resolution and other limitations of numerical weather forecasts;

3. The present invention breaks through the limitation that the wind speed of each specific wind turbine cannot be accurately predicted. Through long-term progressive deep learning of the wind turbines at each specific location, the prediction model function after the progressive deep learning is obtained. Then, based on the traditional weather forecast data, the wind speed change of each wind turbine after a period of time can be accurately predicted. This breaks through the problem that the original weather forecast data cannot be fully applicable to the actual situation of a specific wind turbine for large-area forecasts containing wind farms.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a process of training a prediction model using a deep learning network structure according to the present invention;

FIG2 is a schematic diagram of a flow chart of the present invention for calculating the wind speed data of each wind turbine using a prediction model;

FIG3 is a schematic diagram of the logical structure of the prediction model trained using a deep learning network structure in the present invention.

DETAILED DESCRIPTION

The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Embodiment 1: As shown in Figures 1 to 3, a short-term wind power prediction method based on multi-source data progressive deep learning is proposed. The testing method comprises the following steps:

S1: Use deep learning network structure to train prediction model;

S1.1. Collect multi-source data required for the prediction model;

S1.2, preprocessing the multi-source data collected in step 1.1;

S1.3, dividing the processed data set;

S1.4. Train the deep learning model to obtain the prediction model function f(x).

S2: Use the prediction model to calculate the wind speed data of each wind turbine and predict the wind power;

S2.1. Collect weather forecast data for a certain wind farm;

S2.2, using the trained prediction model f(x), calculate the wind speed of a certain wind turbine after a period of time;

S2.3. Obtain the wind speed data and wind power data of a certain wind turbine after a period of time.

Furthermore, the step S1.1 is specifically as follows: collecting numerical values and weather forecast data related to wind speed of each wind turbine to form a multi-source data set, and the data related to wind speed includes two parts: one is the weather forecast data updated for a period of time for the wind farm, including wind direction, short-term weather type, rainfall type, and temperature; the other is the data to be collected at each wind turbine location, including wind turbine height, terrain type, temperature, air pressure, humidity, sea level pressure, short-wave radiation from the ground downward, long-wave radiation from the ground downward, and total cloud cover;

The multi-source data are labeled and divided into two groups, one group is the real-time collected data _X1 , _X2 ... _Xn at each wind turbine location, and the other group is the weather forecast data _Y1 , _Y2 ... _Yn a period of time before the real-time collected data of each wind turbine.

Furthermore, in step S1.2, specific measures for data preprocessing include: correcting or deleting incomplete or erroneous data, processing duplicate data, processing outliers, unifying data formats, data normalization and data encoding.

Furthermore, the step S1.3 is specifically as follows: the preprocessed multi-source data is divided into a training set, a validation set and a test set according to a certain ratio, wherein the training set is a data set used to train the machine learning model, the validation set is a data set set aside separately during the deep learning model training process for evaluating model performance and adjusting hyperparameters, and the test set is a data set used to finally evaluate the performance of the machine learning model.

Furthermore, in step S1.4, the deep learning model includes various deep learning neural network structures, and the deep neural network includes an input layer, an output layer and at least one hidden layer; the network parameters are adjusted, and the number of network layers, the number of neurons, batch normalization, and random inactivation are adjusted to train the network until the maximum number of iterations is reached or the network learning rate converges; at the same time, the mean square error is selected as the loss function and the mean absolute error is selected as the evaluation accuracy; the number of network layers and the number of neurons are set according to the application in different regional scenarios. and adjust until the loss function converges or the maximum number of iterations is reached, completing the model training;

The multi-source data _X1 , _X2 ... _Xn collected in real time at the location of each wind turbine in step S1.1 and the multi-source data _Y1 , _Y2 ... _Yn of the weather forecast for a period of time before the real-time data of each wind turbine are collected are used as the input layer, the number of hidden layers is set according to different data amounts, and is at least greater than 1 layer. After training by adjusting parameters such as activation functions and weights, the relationship between the real-time multi-source data _X1 , _X2 ... _Xn of each wind turbine and the multi-source data _Y1 , _Y2 ... _Yn provided by the weather forecast for a period of time before is obtained, that is, the prediction model function f(x) is obtained.

Furthermore, the wind farm weather forecast data collected in step S2.1 includes weather forecast data Y ₁ , Y ₂ ..Y _n of a large area of a certain wind farm;

The step S2.2 specifically comprises: using the trained wind speed prediction model f(x) of each wind turbine, combined with the weather forecast data Y ₁ , Y ₂ ..Y _n collected in step S2.1, to calculate the wind speed data at the location of each wind turbine after a period of time;

The step S2.3 is specifically: obtaining the wind speed data at each wind turbine position after a period of time, and then accurately predicting the wind power data of a specific wind turbine after a period of time.

Embodiment 2:

S1: Use deep learning network structure to train prediction model;

The specific steps are described in detail as follows:

Step 1.1: Collect multi-source data required for the prediction model. Collect the numerical values and weather forecast data related to wind speed of each wind turbine to form a multi-source data set. The data related to wind speed includes two parts: one is the weather forecast data updated every 1h for the wind farm, such as wind direction, short-term weather type, rainfall type, temperature, etc.; the other is the data that needs to be collected at each wind turbine location, such as wind turbine height, terrain type, temperature, air pressure, humidity, sea level pressure, ground downward shortwave radiation, ground downward longwave radiation, total cloud cover, etc.

Note: The "1h" mentioned here and in this patent is an example for the convenience of explanation, and does not refer to only 1h. It means the time interval for updating the current weather forecast data or the time interval set by each wind farm that is greater than or equal to this time interval, which can be 1.5h, 2h, etc., all within the scope of protection and description of this patent.

The multi-source data in step 1.1 can be labeled as shown in Table 1 and Table 2, where Table 1 is the real-time data collected at each wind turbine location, and Table 2 is the weather forecast data 1 hour before the real-time data collection of each wind turbine. Here is an example, if the whole day is divided into hours, there are 24 groups of data, such as 12 o'clock (24-hour system) Table 1 is the real-time multi-source data collected at each wind turbine location, and Table 2 is the data provided by the weather forecast for the large area of the wind farm at 11 o'clock.

Table 1 Multi-source data tags collected in real time at each wind turbine location (12 o'clock)

Table 2 Weather forecast multi-source data labels for each wind turbine 1 hour before real-time data collection (11:00)

Step 1.2: Preprocess the multi-source data collected in step 1.1. Data preprocessing is to improve the quality of data, enhance the consistency and integrity of data, and make it more able to meet the needs of analysis or mining. A series of operations of data preprocessing include processing "dirty" data, accurately extracting data, adjusting the data format, etc., so as to obtain a set of high-quality data that meets the standards of accuracy, completeness, and simplicity. The specific measures of data preprocessing include: correcting or deleting incomplete or erroneous data, processing duplicate data, processing outliers, unifying data formats, data normalization, and data encoding.

① Correcting or deleting incomplete or erroneous data, processing duplicate data, processing outliers, etc. can be divided into the following steps:

1) Identify incomplete or erroneous data: This can be done by checking for outliers, missing values, 1) Filling missing values: This can be done by using simple deletion or weighting. Simple deletion directly deletes cases with missing values. The weighting method reduces the bias by weighting the complete data. The weight of the case can be obtained by logistic or probit regression. 2) Handling outliers: Outliers (abnormal values) are the norm of data distribution. Data outside a specific distribution area or range is usually defined as abnormal or noise, which can usually be handled by deleting outliers.

② Unified data format, data normalization and data encoding, etc. can be divided into the following steps:

1) Unify data format: If the data format is not unified, you can use the corresponding method to convert it. You can use Python's pandas library to convert the data format; 2) The normalization formula can be used as follows; 3) Data encoding can use one-hot encoding and label encoding and other encoding methods.

Step 1.3: Divide the processed data set. Divide the preprocessed multi-source data into training set, validation set and test set in a ratio of 7:1.5:1.5. The training set is the data set used to train the machine learning model, the validation set is the data set set aside during the deep learning model training process for evaluating model performance and adjusting hyperparameters, and the test set is the data set used to finally evaluate the performance of the machine learning model.

Step 1.4: Train the deep learning model. The deep learning model includes various deep learning neural network structures, and the deep neural network should include an input layer, an output layer, and at least one hidden layer. The network parameters can be adjusted by adjusting the number of network layers, the number of neurons, batch normalization, and random inactivation to train the network until the maximum number of iterations is reached or the network learning rate converges. At the same time, the mean square error is selected as the loss function and the mean absolute error is selected as the evaluation accuracy. Of course, the number of network layers and the number of neurons here can be set and adjusted according to the application in different regional scenarios until the value of the loss function converges or the maximum number of iterations is reached to complete the model training.

The real-time multi-source data _X1 , _X2 ... _Xn collected at the location of each wind turbine in Table 1 in step 1.1 and the multi-source data _Y1 , _Y2 ... _Yn of the weather forecast 1 hour before the real-time data collected by each wind turbine in Table 2 are used as the input layer. The number of hidden layers can be set according to different data amounts and should be at least greater than 1 layer. By adjusting parameters such as activation functions and weights, the relationship between the real-time multi-source data of each wind turbine and the multi-source data provided by the weather forecast 1 hour before can be trained to obtain, that is, the prediction model function f(x) is obtained.

S2: Use the prediction model to calculate the wind speed data of each wind turbine and predict the wind power;

The specific steps are described in detail as follows:

Step 2.1: Collect weather forecast data Y ₁ , Y ₂ ...Y _n , etc. for a large area containing a wind farm at a certain time;

Step 2.2: Using the trained wind speed prediction model f(x) for each wind turbine, combined with the weather forecast data Y ₁ , Y ₂ ...Y _n collected in step 2.1, calculate the wind speed data at each wind turbine location 1 hour later;

Step 2.3: Calculate the wind speed data at each wind turbine location 1 hour later, and then accurately predict the specific wind power data of a specific wind turbine 1 hour later.

For Example 2, the following supplementary explanations are given:

1. The "1h" mentioned in this patent is an example for the convenience of explanation and does not refer to only 1h. It means the time interval for updating the current weather forecast data or the time interval set by each wind farm that is greater than or equal to this time interval, which can be 1.5h, 2h, etc., all within the scope of protection of this patent.

2. In the stage of training the prediction model, the multi-source data in step 1, this patent only lists some parameters here, such as wind turbine height, terrain type, temperature, air pressure, humidity, sea level air pressure, ground-down shortwave radiation, ground-down longwave radiation, total cloud cover, etc. It is not exhaustive, and some parameters can be deleted according to actual conditions, and the parameters of influencing factors related to wind speed can be supplemented, which should be within the scope of protection of this patent.

3. In the prediction model training stage, the deep learning network structure in step 4 does not specifically refer to a specific deep learning algorithm, such as a convolutional neural network, etc. All deep learning neural network structures should be within the scope of protection of this patent.

The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims (8)

A short-term wind power prediction method based on multi-source data progressive deep learning is characterized in that it includes the following steps: S1: Use deep learning network structure to train prediction model; S2: Use the prediction model to calculate the wind speed data of each wind turbine and predict the wind power; According to the method for short-term wind power prediction based on multi-source data progressive deep learning according to claim 1, it is characterized in that: step S1 specifically includes the following process: S1.1. Collect multi-source data required for the prediction model; S1.2, preprocessing the multi-source data collected in step 1.1; S1.3, dividing the processed data set; S1.4. Train the deep learning model to obtain the prediction model function f(x). According to the method for short-term wind power prediction based on multi-source data progressive deep learning according to claim 2, it is characterized in that: step S2 specifically includes the following process: S2.1. Collect weather forecast data for a certain wind farm; S2.2, using the trained prediction model f(x), calculate the wind speed of a certain wind turbine after a period of time; S2.3. Obtain the wind speed data and wind power data of a certain wind turbine after a period of time. According to the method for short-term wind power prediction based on multi-source data progressive deep learning as described in claim 2, it is characterized in that: the step S1.1 is specifically: collecting numerical values and weather forecast data related to wind speed of each wind turbine to form a multi-source data set, and the data related to wind speed includes two parts: one is the weather forecast data updated for a period of time for the wind farm, including wind direction, short-term weather type, rainfall type, and temperature; the other is the data to be collected at each wind turbine position, including wind turbine height, terrain type, temperature, air pressure, humidity, sea level pressure, downward short-wave radiation from the ground, downward long-wave radiation from the ground, and total cloud cover; The multi-source data are labeled and divided into two groups, one group is the real-time collected data _X1 , _X2 ... _Xn at each wind turbine location, and the other group is the weather forecast data _Y1 , _Y2 ... _Yn a period of time before the real-time collected data of each wind turbine. According to the method for short-term wind power prediction based on multi-source data progressive deep learning according to claim 2, it is characterized in that: in the step S1.2, the specific measures for data preprocessing include: correcting or deleting incomplete or erroneous data, processing duplicate data, processing outliers, unifying data format, data normalization processing and data encoding. According to the short-term wind power prediction method based on multi-source data progressive deep learning according to claim 2, it is characterized in that; the step S1.3 is specifically: the preprocessed multi-source data is divided into a training set, a validation set and a test set according to a certain ratio, wherein the training set is a data set for training the machine learning model, the validation set is a data set set aside separately during the deep learning model training process for evaluating the model performance and adjusting the hyperparameters, and the test set is a data set for the final evaluation of the machine learning model performance. According to the method for short-term wind power prediction based on multi-source data progressive deep learning according to claim 2, it is characterized in that: in the step S1.4, the deep learning model includes various deep learning neural network structures, and the deep neural network includes an input layer, an output layer and at least one hidden layer; the network parameters are adjusted, and the network is trained by adjusting the number of network layers, the number of neurons, batch normalization, and random inactivation until the maximum number of iterations is reached or the network learning rate converges; at the same time, the mean square error is selected as the loss function and the mean absolute error is selected as the evaluation accuracy; here, the number of network layers and the number of neurons are set and adjusted according to the application in different regional scenarios until the value of the loss function converges or the maximum number of iterations is reached, and the model training is completed; The multi-source data _X1 , _X2 ... _Xn collected in real time at the location of each wind turbine in step S1.1 and the multi-source data _Y1 , _Y2 ... _Yn of the weather forecast for a period of time before the real-time data of each wind turbine are collected are used as the input layer, the number of hidden layers is set according to different data amounts, and is at least greater than 1 layer. After training by adjusting parameters such as activation functions and weights, the relationship between the real-time multi-source data _X1 , _X2 ... _Xn of each wind turbine and the multi-source data _Y1 , _Y2 ... _Yn provided by the weather forecast for a period of time before is obtained, that is, the prediction model function f(x) is obtained. The short-term wind power prediction method based on multi-source data progressive deep learning according to claim 3 is characterized in that: the wind farm weather forecast data collected in step S2.1 includes weather forecast data Y ₁ , Y ₂ ...Y _n of a large area of a certain wind farm; The step S2.2 specifically comprises: using the trained wind speed prediction model f(x) of each wind turbine, combined with the weather forecast data Y ₁ , Y ₂ ..Y _n collected in step S2.1, to calculate the wind speed data at the location of each wind turbine after a period of time; The step S2.3 is specifically: obtaining the wind speed data at each wind turbine position after a period of time, and then accurately predicting the wind power data of a specific wind turbine after a period of time.