A residual correction power prediction method, system and readable storage medium

Abstract

The application discloses a residual correction power prediction method and system and a readable storage medium, and belongs to the technical field of wind power generation. Firstly, NWP prediction data, NWP historical data, unit operation state data and wind measurement tower data are acquired. Then, power trend prediction and power residual prediction are carried out. The power trend prediction adopts multiple similarity measurement indexes to preliminarily screen a power candidate set, and an attention unit is adopted to carry out secondary screening on the power candidate set to increase the adaptability of the model. The power residual prediction part adopts a channel-independent prediction idea to independently model parameters of different key variables, thereby reducing mutual confusion in the prediction process of different variables. Finally, the power trend prediction result and the power residual prediction result are fused, and finally, a power prediction result is obtained, and the power prediction accuracy is improved.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a residual correction power prediction method, system, and readable storage medium. Background Technology

[0002] With the continuous increase in installed wind power capacity, the volatility, randomness, and intermittency of wind power generation have a more significant impact on the power grid. Accurate power forecasting can effectively reduce the impact of wind power generation on the power grid. Most existing technologies rely on Numerical Weather Prediction (NWP) data for power forecasting, neglecting the influence of turbine operating conditions on power prediction. Some schemes consider turbine operating conditions and combine them with NWP forecast data for power prediction, but simply using both sets of data as input features for model training offers limited improvement in prediction accuracy. Some technologies employ similarity day theory, finding a corresponding power matrix based on the similarity between NWP forecast data and historical NWP data as model input for power prediction. However, the similarity measurement criterion is singular and cannot reflect data similarity from multiple dimensions, thus affecting the accuracy of power prediction. Summary of the Invention

[0003] This invention provides a residual correction power prediction method, system, and readable storage medium, which can perform power prediction more accurately.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A residual-corrected power prediction method includes the following steps:

[0006] Acquire meteorological forecast data, unit operation data, and wind tower data; the meteorological forecast data includes historical NWP data and predicted NWP data;

[0007] Power trend prediction is performed based on meteorological forecast data and unit operation data to obtain a power prediction result matrix;

[0008] Power trend prediction is performed based on historical NWP data, unit operation data and forecast weather data to obtain a power residual prediction matrix;

[0009] Wind power prediction is performed based on the power prediction result matrix and the power residual prediction matrix.

[0010] Furthermore, power trend prediction is performed using meteorological forecast data and unit operation data to obtain a power prediction result matrix, including:

[0011] The prediction period is divided into multiple time periods. The similarity between the predicted NWP data and the historical NWP data in each time period is calculated to obtain the M closest historical NWP matrices. The historical unit operating power corresponding to the M historical NWP matrices is selected as the candidate power matrix. After calculating N similarity indicators, N×M historical NWP vectors and N×M historical unit power vectors are obtained for each time period.

[0012] For each time period, the N×M historical unit power vectors are filtered, and the K most important historical unit power vectors are selected as candidate results for the power of that time period.

[0013] The candidate power results for the given time period are normalized, and the K results with the largest weights are selected to form a candidate matrix. The candidate matrix is ​​then subjected to a Hadamard inner product operation to obtain the power candidate matrix. The above operation is performed on each time period to obtain the top K power candidate vectors for the entire prediction sequence.

[0014] The top K power candidate vectors of the entire prediction sequence are used as input to perform power trend prediction and obtain the power trend prediction result.

[0015] Furthermore, an LSTM model is used for power trend prediction.

[0016] Furthermore, the filtering of the N×M historical unit power vectors for each time period includes: using an attention mechanism to filter the N×M historical unit power vectors for each time period.

[0017] Furthermore, the similarity between the predicted NWP data and the historical NWP data for each time period is calculated separately, including calculating the similarity between the predicted NWP data and the historical NWP data for each time period from two dimensions: distance similarity and morphological similarity.

[0018] Furthermore, based on historical NWP data, unit operation data, and forecast weather data, power trend prediction is performed to obtain a power residual prediction matrix, including:

[0019] Key variables were screened from historical NWP data, meteorological tower data, and unit operation data, and key variables with high correlation coefficients with power were selected.

[0020] Each key variable is predicted independently using multiple Res-MLP modules, and historical information is used to predict the variable's information for the next day.

[0021] Furthermore, the Res-MLP module includes a multi-layer DNN network. The first layer of the DNN network uses Leaky ReLU as the activation function, the second layer of the DNN network uses a linear activation function, then goes through a Dropout operation, and finally passes through a residual connection layer and normalization to obtain the output result.

[0022] Furthermore, based on the power prediction result matrix and the power residual prediction matrix, wind power prediction is performed, including the following steps:

[0023] The power prediction result matrix is ​​phase-added with the power residual prediction matrix to obtain a new prediction matrix. The new prediction matrix is ​​used as the input of the LSTM network. A time-step independent prediction method is adopted, and a separate weight is assigned to each time step for power prediction to obtain the power prediction results for each future time period, thus completing the power prediction.

[0024] A residual-corrected power prediction system includes:

[0025] The data acquisition module is used to acquire meteorological forecast data, unit operation data, and wind tower data; the meteorological forecast data includes historical NWP data and predicted NWP data.

[0026] The power trend prediction module is used to predict power trends based on meteorological forecast data and unit operation data, and obtain a power prediction result matrix.

[0027] The power residual prediction module is used to predict power trends based on historical NWP data, unit operation data and forecast weather data, and obtain a power residual prediction matrix.

[0028] The power residual correction module is used to predict wind power based on the power prediction result matrix and the power residual prediction matrix.

[0029] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power prediction method described above.

[0030] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0031] This invention makes full use of NWP forecast data, NWP historical data, unit operating status data, and anemometer data for power prediction. The overall power prediction is divided into power trend prediction and power residual prediction. After obtaining the power trend prediction and power residual prediction results, the prediction results are fused to further improve the accuracy of power prediction.

[0032] Furthermore, in order to reduce the confusion problem in the calculation of long-term similarity metrics, this invention segments the long-term data and uses a similarity calculation method to screen the power candidate matrix for each segment, which further improves the accuracy of power trend prediction.

[0033] Furthermore, power trend prediction employs multiple similarity metrics for initial screening of the power candidate set, and uses attention units for secondary screening of the power candidate set to increase the model's adaptability and improve the accuracy of power trend prediction.

[0034] Furthermore, the power residual prediction section uses independent channels to independently model the parameters of different key variables, reducing the mutual confusion in the prediction process of different variables and further improving the accuracy of power residual prediction.

[0035] Furthermore, power prediction and power residuals are fused together, and the concept of channel independence is adopted to assign independent learnable weights to each prediction time step, thereby improving the prediction effect at each time step. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of a power prediction method with residual correction provided in an embodiment of the present invention;

[0037] Figure 2 The power trend prediction process provided in the embodiments of the present invention;

[0038] Figure 3 The power residual prediction process provided in the embodiments of the present invention;

[0039] Figure 4 This is a schematic diagram of the Res-MLP module structure provided in an embodiment of the present invention;

[0040] Figure 5 This is a flowchart of the residual correction process provided in an embodiment of the present invention;

[0041] Figure 6 This is a schematic diagram of the structure of a residual correction power pre-system provided in an embodiment of the present invention;

[0042] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives and technical solutions of this invention clearer and easier to understand, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0044] Figure 1This is a schematic diagram of the overall scheme provided in the embodiment of the present invention. The present invention makes full use of NWP prediction data, NWP historical data, unit operating status data, and anemometer data to design a power prediction algorithm, thereby completing the power prediction function.

[0045] It should be noted that the power prediction method in this embodiment of the invention can be executed by the power prediction system provided in this embodiment of the invention. This wind farm power prediction system can be an electronic device, or it can be configured within an electronic device to achieve the power prediction function.

[0046] The electronic device can be any stationary or mobile computing device capable of data processing, such as a laptop, smartphone, wearable device, or other mobile computing device, or a desktop computer or other stationary computing device, or a server, or other types of computing devices. This embodiment does not impose any restrictions on this.

[0047] Reference Figure 1 A residual-corrected power prediction method includes the following steps:

[0048] Step S101: Data acquisition is performed. The acquired data includes: Numerical Weather Prediction (NWP) data, generator operation data, and anemometer data. NWP data is divided into historical NWP data and predicted future NWP data. Variables include: wind speed and direction data at different elevations, as well as temperature and humidity data. Generator operation data includes: wind speed, wind direction, power, engine speed, propeller pitch angle, ambient temperature, and ambient humidity. Anemometer data includes: wind speed, wind direction, temperature, and humidity information at different elevations. In this embodiment, different elevations can be wind speed and direction data at altitudes of 10 meters, 20 meters, 30 meters, 40 meters, 50 meters, 60 meters, 70 meters, 80 meters, 90 meters, and 100 meters, as well as ambient temperature, ambient humidity, and atmospheric pressure data.

[0049] The acquired data is used to support step S102 power trend prediction and step S103 power residual prediction. Specifically, the power trend prediction module in step S102 requires historical NWP data, future NWP data, and unit operation data; the power residual prediction module in step S103 requires historical NWP data, unit operation data, and meteorological tower data.

[0050] Step S102 performs power trend prediction, specifically as follows: Figure 2 As shown. Power trend prediction is based on the similarity day theory, and a power matrix selection is performed. Specifically, the following steps are included:

[0051] First, long-term time series are segmented. To improve prediction accuracy, the long series is divided into segments, and similarity is calculated. For example, short-term power prediction involves predicting the power trend for the next day. If the time resolution is 15 minutes, then the power trend results for the next 96 points need to be predicted. Here, one day is divided into 6 short time series, that is, each time period consists of 16 data points in 4 hours, and similarity is calculated. Of course, power trends for the next two days, three days, etc. can also be predicted. The time resolution can be set to 10 minutes, 20 minutes, 30 minutes, 60 minutes, etc., and the time periods covered by the segmented short time series can be 2 hours, 4 hours, 6 hours, etc.

[0052] Of course, the data needs to be normalized before calculation to eliminate the influence of different dimensions in the data. The normalization formula is as follows:

[0053]

[0054] in, Here, x represents the original data after normalization, mean(x) represents the mean of the original data, and std(x) represents the standard deviation of the original data.

[0055] Subsequently, the similarity between the predicted NWP data and historical NWP data is calculated for each time period. Similarity is measured from two dimensions: distance similarity and morphological similarity. Similarity calculations include Euclidean distance, Mahalanobis distance, DTW, etc., including but not limited to the aforementioned metrics. Euclidean distance D... em The formula for calculating (x,y) is:

[0056]

[0057] Mahalanobis distance D md The formula for calculating (x,y) is:

[0058]

[0059] Where, ∑ -1 Let x be the covariance matrix of two vectors. i Let be the coordinates of point i in the x-direction, and be the coordinates of point i in the y-direction.

[0060] Within each time period, by calculating different similarity metrics, the M closest historical NWP matrices are obtained. Historical NWP data and historical unit operating data are one-to-one corresponding in time; therefore, the corresponding historical unit operating power is selected as the candidate power matrix. After calculating the three similarity metrics, 3×M historical NWP vectors and 3×M historical unit power vectors are obtained for each time period.

[0061] Then, an attention mechanism is used to filter the 3×M historical generator power vectors for each time period, selecting the K most important historical generator power vectors as candidate results for that time period. Taking the candidate power selection for the first time period as an example, the secondary screening process of the power candidate set using the attention mechanism is explained in detail, and the formula is expressed as follows:

[0062]

[0063] Where, β i For the calculated phase velocity results, X nwp_pre This is the NWP forecast data for this time period. Here are several historical NWP data sets with high similarity obtained through similarity calculation. There are a total of 3×M vectors. [;] represents channel connections, v and w are learnable weight matrices, and b is the bias vector. Since the historical NWP data vectors obtained after similarity calculation are 3×M, the above calculation process yields β=[β1,β2,…β…]. 3×M After normalizing β, the calculation formula is as follows:

[0064]

[0065] The normalized weights are then expressed as: Then, the K results with the highest weights are selected as candidate matrices. The Hadamard inner product operation is then performed on these candidate matrices to obtain the power candidate matrix, which contains K vectors. Through this process, the top K power candidate vectors can be obtained from the 3×M power candidate vectors. Applying this process to each time period yields the top K power candidate vectors for the entire prediction sequence.

[0066] Finally, the top K power candidate vectors of the entire prediction sequence, i.e., a 96×K matrix, are used as input to perform power trend prediction using an LSTM model, thereby obtaining the power trend prediction result. A two-layer LSTM structure is adopted, with the hidden layer length of each LSTM layer set to 50, so the prediction result is a 96×50 power prediction result matrix.

[0067] Step S103: Perform power residual prediction. This mainly employs a channel-independent prediction approach, independently modeling parameters from different data sources to predict power residuals. This reduces mutual confusion during the prediction of different variables. The prediction process is as follows: Figure 3 As shown. The data sources used include historical NWP data, meteorological tower data, and unit operation data. Each data source is used to perform channel-independent prediction using the Res-MLP module to obtain the future prediction results, i.e., the power residual prediction matrix.

[0068] The Res-MLP module structure is as follows: Figure 4 As shown, key variables are selected from historical NWP data, meteorological tower data, and unit operation data. The variable selection process can use the correlation coefficient method, choosing key variables with high correlation coefficients to power. The calculation of the correlation coefficients is not detailed here. Each variable is predicted independently using a Res-MLP module, using historical information to predict the variable's information for the next day. The Res-MLP module consists of a multi-layer DNN network. The first layer uses Leaky ReLU as the activation function, the second layer uses a linear activation function, then undergoes Dropout, and finally passes through a residual connection layer and a normalization process (Add & normalize) to obtain the output result. To increase the accuracy of the prediction, S Res-MLP modules can be used to predict future variables.

[0069] Assuming there are A key variables for the generator data, B key variables for the meteorological tower data, and C key variables for the historical NWP data, the Res-MLP module will produce a prediction for the next day. The predicted results for the generator data will be a 96×A matrix, the meteorological tower data a 96×B matrix, and the historical NWP data a 96×C matrix. These three prediction matrices are concatenated and used as input to an LSTM model. A two-layer LSTM network is used, with each layer having the same hidden layer length of 50 as the LSTM network used in power trend prediction. The resulting power residual prediction matrix after the two-layer LSTM network is a 96×50 matrix.

[0070] Step S104, power residual correction, the process is as follows: Figure 5 As shown. The system receives the power trend prediction result matrix from step S102 and the power residual prediction result matrix from step S103, and then fuses the two parts. Specifically, the power trend prediction result matrix and the power residual prediction result matrix are added in phase to obtain a new prediction matrix with a size of 96×50. Next, the fused new prediction matrix is ​​used as the input to the LSTM network. The hidden layer output length of the LSTM network is set to 50, so the output result of the LSTM network is a 96×50 matrix. Finally, a time-step independent prediction method is used, with a separate weight layer assigned to each time step for power prediction, ultimately obtaining the prediction results for the power of the next 96 times, completing the power prediction.

[0071] After completing the network model construction, model training can be performed to obtain the optimal model. For the data to be predicted, the same data processing methods as in the training phase can be used to meet the model's input requirements. The data is then input into the trained optimal model to obtain the power prediction result. The data to be predicted and the historical training data have the same data structure; the training process is completed using historical data, and the prediction is completed using the current data.

[0072] Reference Figure 6 As shown, the present invention provides a power prediction system with residual correction, including a data acquisition module, a power trend prediction module, a power residual prediction module and a power residual correction module.

[0073] The data acquisition module is used to acquire meteorological forecast data, unit operation data, and wind tower data; the meteorological forecast data includes historical NWP data and predicted NWP data.

[0074] The power trend prediction module performs power trend prediction based on meteorological forecast data and unit operation data, and obtains a power prediction result matrix.

[0075] The power residual prediction module predicts power trends based on historical NWP data, unit operation data and forecast weather data, and obtains a power residual prediction matrix.

[0076] The power residual correction module performs wind power prediction based on the power prediction result matrix and the power residual prediction matrix.

[0077] Reference Figure 7 As shown, this invention provides an electronic device comprising: a processor, a memory, and a bus. The memory stores computer-executed instructions, and the processor is connected to the memory via the bus. When the electronic device is running, the processor executes the computer-executed instructions stored in the memory, thereby enabling the processor to run a program corresponding to the computer-executed instructions by reading the computer-executed instructions stored in the memory, for executing the power prediction method proposed in any of the above embodiments of this invention. The bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The symbol is represented by only one line, but this does not mean that there is only one bus or one type of bus.

[0078] To implement the above embodiments, the present invention also proposes a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the residual correction power prediction method proposed in any of the above embodiments of the present invention.

[0079] To implement the above embodiments, the present invention also proposes a computer program product, including a computer program that, when executed by a processor, implements the residual correction power prediction method proposed in any of the above embodiments of the present invention.

[0080] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0081] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0082] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0083] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0084] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0085] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0086] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0087] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A power prediction method with residual correction, characterized in that, Includes the following steps: Acquire meteorological forecast data, unit operation data, and wind tower data; the meteorological forecast data includes historical NWP data and predicted NWP data; Power trend prediction is performed based on meteorological forecast data and unit operation data to obtain a power prediction result matrix; Power trend prediction is performed based on historical NWP data, unit operation data and forecast weather data to obtain a power residual prediction matrix; Wind power prediction is performed based on the power prediction result matrix and the power residual prediction matrix. The power trend prediction based on meteorological forecast data and unit operation data yields a power prediction result matrix, including: The prediction period is divided into multiple time periods. The similarity between the predicted NWP data and the historical NWP data in each time period is calculated to obtain the M closest historical NWP matrices. The historical unit operating power corresponding to the M historical NWP matrices is selected as the candidate power matrix. After calculating N similarity indicators, N×M historical NWP vectors and N×M historical unit power vectors are obtained for each time period. For each time period, the N×M historical unit power vectors are filtered, and the K most important historical unit power vectors are selected as candidate results for the power of that time period. The candidate power results for the given time period are normalized, and the K results with the largest weights are selected to form a candidate matrix. The candidate matrix is ​​then subjected to a Hadamard inner product operation to obtain the power candidate matrix. The above operation is performed on each time period to obtain the top K power candidate vectors for the entire prediction sequence. The top K power candidate vectors of the entire prediction sequence are used as input to perform power trend prediction and obtain the power trend prediction result.

2. The power prediction method with residual correction according to claim 1, characterized in that, Power trend prediction using LSTM model.

3. The power prediction method with residual correction according to claim 1, characterized in that, The filtering of the N×M historical unit power vectors for each time period includes: using an attention mechanism to filter the N×M historical unit power vectors for each time period.

4. The residual-corrected power prediction method according to claim 1, characterized in that, The similarity between the predicted NWP data and the historical NWP data for each time period is calculated separately, including from two dimensions: distance similarity and morphological similarity.

5. The residual-corrected power prediction method according to claim 1, characterized in that, The power trend prediction based on historical NWP data, unit operation data, and predicted weather forecast data yields a power residual prediction matrix, including: Key variables were screened from historical NWP data, meteorological tower data, and unit operation data, and key variables with high correlation coefficients with power were selected. Each key variable is predicted independently using multiple Res-MLP modules, and historical information is used to predict the variable's information for the next day.

6. The residual-corrected power prediction method according to claim 5, characterized in that, The Res-MLP module includes a multi-layer DNN network. The first layer of the DNN network uses Leaky ReLU as the activation function, the second layer of the DNN network uses a linear activation function, and then the output result is obtained after the Dropout operation and the residual connection layer and normalization.

7. The residual-corrected power prediction method according to claim 1, characterized in that, The wind power prediction based on the power prediction result matrix and the power residual prediction matrix includes the following steps: The power prediction result matrix is ​​phase-added with the power residual prediction matrix to obtain a new prediction matrix. The new prediction matrix is ​​used as the input of the LSTM network. A time-step independent prediction method is adopted, and a separate weight is assigned to each time step for power prediction to obtain the power prediction results for each future time period, thus completing the power prediction.

8. A residual-corrected power prediction system for implementing the method of claim 1, characterized in that, include: The data acquisition module is used to acquire weather forecast data, unit operation data, and wind measurement tower data; The weather forecast data includes historical NWP data and predicted NWP data; The power trend prediction module is used to predict power trends based on meteorological forecast data and unit operation data, and obtain a power prediction result matrix. The power residual prediction module is used to predict power trends based on historical NWP data, unit operation data and forecast weather data, and obtain a power residual prediction matrix. The power residual correction module is used to predict wind power based on the power prediction result matrix and the power residual prediction matrix.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

Images