Method and device for screening numerical weather forecast data of wind power prediction model

Abstract

The invention relates to the technical field of wind power, and discloses a numerical weather forecast data screening method and device for a wind power prediction model, and the method comprises the steps: obtaining the accumulated wind power and numerical weather forecast data of a wind power plant group in a target region at historical time, and imitating the wind speed characteristics; inputting the imitated wind speed features and the original wind speed features into a neural network model, and iteratively updating a first weight and a second weight of a filtering layer of the neural network model and a third weight of a multi-layer perceptron until accumulated wind power is output, so as to obtain a latest first weight, a latest second weight and a latest third weight; based on the latest first weight, the second weight and the third weight, generating counterfeit statistics; determining a counterfeit threshold value based on a preset error rate threshold value and the counterfeit statistical magnitude; and selecting the original wind speed characteristics of which the mapped counterfeit statistical magnitude is greater than a counterfeit threshold value from the original wind speed characteristics, and taking the corresponding numerical weather forecast data as the numerical weather forecast data of the wind power prediction model.

Description

Technical Field

[0001] The present invention relates to the field of wind power technology, and in particular to a method and device for screening numerical weather forecast data of a wind power prediction model. Background Art

[0002] At present, the goal of short-term wind power forecasting for regional wind farms is to introduce numerical weather forecast data (NWP) to build a joint forecasting model that can capture the spatiotemporal dependencies of wind farms in the region, thereby improving the accuracy of overall wind power forecasting. Compared with the traditional "one farm, one measurement" model, regional centralized forecasting can better reflect the complex spatiotemporal characteristics of wind farms, adapt to the centralized operation and management needs of wind farms, and provide more competitive forecasting results in power market bidding.

[0003] Traditional methods often use linear or nonlinear relationships to measure the correlation between different numerical weather forecast data and wind farm power based on the principle of similarity to select numerical weather forecast data with high correlation. However, traditional methods usually rely on manual experience for screening, which makes it difficult to quantify the error rate and may be limited when facing wind farms with complex spatiotemporal dependencies. Summary of the invention

[0004] In view of this, the present invention provides a method and device for screening numerical weather forecast data of a wind power prediction model to solve the problem of how to screen numerical weather forecast data of a wind power prediction model.

[0005] In a first aspect, the present invention provides a method for screening numerical weather forecast data of a wind power prediction model, the method comprising: Obtaining the accumulated wind power of at least one wind farm group in a target area in historical time and N numerical weather forecast data, as well as N simulated wind speed characteristics mapped one-to-one with the N original wind speed characteristics, wherein the N numerical weather forecast data include the N original wind speed characteristics, and N is a positive integer; Input the N simulated wind speed features and the N original wind speed features into a neural network model, and iteratively update the N first weights and the N second weights of the filter layer of the neural network model, and the N third weights of the multilayer perceptron of the neural network model, until the neural network model outputs the accumulated wind power, and obtains the latest N first weights, the latest N second weights and the latest N third weights, wherein the N first weights are used to weight the N original wind speed features, and the N second weights are used to weight the N simulated wind speed features; Based on the latest N first weights, the latest N second weights and the latest N third weights, N simulated statistics are generated, wherein each of the simulated statistics indicates the difference between one of the simulated wind speed features and the mapped original wind speed feature, and each of the simulated statistics maps one of the original wind speed features; Determine a counterfeit threshold based on a preset error rate threshold and the N counterfeit statistics; The original wind speed feature whose mapped imitation statistic is greater than the imitation threshold is selected from the N original wind speed features to obtain the selected original wind speed feature, and the numerical weather forecast data corresponding to the selected original wind speed feature is used as the numerical weather forecast data of the wind power prediction model.

[0006] In an optional implementation, the N third weights are the product of a pre-weight matrix and a post-weight matrix; The above-mentioned inputting the above-mentioned N simulated wind speed characteristics and the above-mentioned N original wind speed characteristics into the neural network model comprises: Inputting the N simulated wind speed features and the N original wind speed features into the filtering layer, and determining the weighted sum of the N simulated wind speed features and the N original wind speed features using the N first weights and the N second weights to obtain N fused wind speed features; Using the pre-weight matrix, the N fused wind speed features are mapped to the hidden layer of the multi-layer perceptron to obtain the output of the hidden layer; The output of the hidden layer is mapped to the output layer of the multilayer perceptron using the post-weight matrix to obtain the output of the output layer.

[0007] In an optional implementation, selecting the original wind speed feature whose mapped simulated statistic is greater than the simulated threshold from the N original wind speed features as numerical weather forecast data for the wind power prediction model includes: From the N counterfeit statistics, find out whether there is a counterfeit statistic greater than the counterfeit threshold; If so, then all counterfeit statistics greater than the counterfeit threshold are counted to obtain the target counterfeit statistics; The original wind speed characteristics mapped by the above target simulated statistics are used as numerical weather forecast data for the above wind power prediction model.

[0008] In an optional implementation, the N simulated wind speed features and the N original wind speed features are input into the filtering layer, and the weighted sum of the N simulated wind speed features and the N original wind speed features is determined using the N first weights and the N second weights to obtain N fused wind speed features, including: Generate the N fused wind speed features according to the N simulated wind speed features, the N original wind speed features, the N first weights, the N second weights, and the first preset relationship; The first preset relationship is:

[0009] in, is the jth fused wind speed feature among the above N fused wind speed features, is the jth original wind speed feature among the above N original wind speed features, is the jth simulated wind speed feature among the above N simulated wind speed features, is the jth first weight among the above N first weights, is the j-th second weight among the above N second weights.

[0010] In an optional implementation, the N simulation statistics are generated based on the latest N first weights, the latest N second weights and the latest N third weights, including: Generate the N simulated statistics according to the latest N first weights, the latest N second weights, the latest N third weights, and a second preset relationship; The second preset relationship is:

[0011] in, , , is the jth simulation statistic among the above N simulation statistics, is the jth latest first weight among the above N latest first weights, is the jth latest second weight among the above N latest second weights, is the jth latest third weight among the above-mentioned N latest third weights.

[0012] In an optional implementation manner, the forgery threshold is determined based on the preset error rate threshold and the N forgery statistics, including: Determining the counterfeit threshold according to the preset error rate threshold, the N counterfeit statistics and the third preset relationship; The third preset relationship is: ,

[0013] in, is the above counterfeit threshold, is the error rate threshold preset above, is the jth simulation statistic among the above N simulation statistics, is the set of N non-zero absolute values ​​of the simulated statistics.

[0014] In a second aspect, the present invention provides a numerical weather forecast data screening device for a wind power prediction model, the device comprising: An acquisition module, used to acquire the accumulated wind power of at least one wind farm group in a target area in historical time and N numerical weather forecast data, and N simulated wind speed characteristics mapped one-to-one with the N original wind speed characteristics, wherein the N numerical weather forecast data include the N original wind speed characteristics, and N is a positive integer; An updating module, used for inputting the N simulated wind speed features and the N original wind speed features into a neural network model, and iteratively updating the N first weights and the N second weights of the filter layer of the neural network model, and the N third weights of the multilayer perceptron of the neural network model, until the neural network model outputs the accumulated wind power, and obtaining the latest N first weights, the latest N second weights and the latest N third weights, wherein the N first weights are used to weight the N original wind speed features, and the N second weights are used to weight the N simulated wind speed features; A generating module, for generating N simulated statistics based on the latest N first weights, the latest N second weights and the latest N third weights, wherein each of the simulated statistics indicates a difference between one of the simulated wind speed features and the mapped original wind speed feature, and each of the simulated statistics maps one of the original wind speed features; A determination module, used for determining a counterfeit threshold based on a preset error rate threshold and the N counterfeit statistics; A selection module is used to select the original wind speed feature whose mapped imitation statistic is greater than the imitation threshold from the above-mentioned N original wind speed features, obtain the selected original wind speed feature, and use the numerical weather forecast data corresponding to the selected original wind speed feature as the numerical weather forecast data of the wind power prediction model.

[0015] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the numerical weather forecast data screening method for the wind power prediction model of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the numerical weather forecast data screening method for a wind power prediction model of the first aspect or any corresponding embodiment thereof.

[0017] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the numerical weather forecast data screening method for a wind power prediction model of the first aspect or any corresponding embodiment thereof.

[0018] The embodiment provided by the present invention has the following effects: by introducing a neural network model, using the weighted processing of wind speed characteristics and a multi-layer perceptron structure, the processing of original wind speed characteristics and simulated wind speed characteristics is made more accurate, thereby improving the accuracy of wind power prediction. By generating simulated statistics and comparing them with a preset error rate threshold, the most relevant wind speed characteristics can be automatically screened out, avoiding the interference of redundant or irrelevant data, making the input data of the model more effective and representative. The iterative update and weighting mechanism in the neural network enable the model to adaptively adjust the weights in multiple rounds of training, gradually adapt to the conditions of wind farm groups in different regions, thereby enhancing the model's adaptability to different environmental conditions. The method is not only applicable to the original wind speed characteristic data, but also can process simulated wind speed characteristics, so that the numerical weather forecast data can be better utilized and the prediction ability of complex weather changes can be improved. By setting the imitation threshold, the method can automatically select data with strong relevance to the prediction task, avoid manual intervention, reduce the workload of manual data screening, and improve the automation of data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 is a flow chart of a method for screening numerical weather forecast data of a wind power prediction model according to an embodiment of the present invention; Figure 2 is a schematic diagram of a numerical weather forecast data screening device for a wind power prediction model according to an embodiment of the present invention; Figure 3 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0022] According to an embodiment of the present invention, an embodiment of a method for screening numerical weather forecast data for a wind power prediction model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0023] The Transformer model (a deep learning architecture) can be used to build a short-term wind power prediction model for wind farms. Training the Transformer model requires appropriate numerical weather forecast data and corresponding measured power values.

[0024] In this embodiment, a method for screening numerical weather forecast data of a wind power prediction model is provided, which can be used for a computer, Figure 1 FIG. 1 is a flow chart of a method for screening numerical weather forecast data according to a wind power prediction model according to an embodiment of the present invention. Figure 1 The method comprises: Step S101, obtaining the accumulated wind power of at least one wind farm group in a target area in historical time and N numerical weather forecast data, as well as N simulated wind speed characteristics mapped one-to-one with the N original wind speed characteristics.

[0025] Among them, N numerical weather forecast data include N original wind speed features, N is a positive integer. First, it is necessary to collect the accumulated wind power in historical time (indicating the total power generation of the target wind farm group in a certain period of history) and numerical weather forecast data. The key components of these data are wind speed features (original wind speed and simulated wind speed features). N simulated wind speed features are constructed using the autoencoder method.

[0026] One way to obtain it is that the historical cumulative wind power data of the wind farm group may be obtained through the on-site monitoring system, which reflects the actual power generation capacity of the wind farm. Numerical weather forecast data is usually the wind speed, wind direction, air pressure and other data calculated by the meteorological service provider through the atmospheric model. The simulated wind speed characteristics can be generated through the known historical data and prediction model to assist in the prediction of future wind power.

[0027] Step S102, input N simulated wind speed features and N original wind speed features into the neural network model, and iteratively update N first weights and N second weights of the filter layer of the neural network model, and N third weights of the multilayer perceptron of the neural network model, until the neural network model outputs the accumulated wind power, and obtains the latest N first weights, the latest N second weights, and the latest N third weights.

[0028] Among them, N first weights are used to weight N original wind speed features, and N second weights are used to weight N simulated wind speed features. The purpose is to input N simulated wind speed features and N original wind speed features into the neural network and train the model to obtain weights. N first weights are used to weight the original wind speed features, indicating the importance of the original wind speed features to predicting the accumulated wind power. N second weights are used to weight the simulated wind speed features, indicating the importance of the simulated wind speed features to predicting the accumulated wind power. N third weights are related to the multilayer perceptron part of the neural network, which helps the model combine the weighted input features into the final wind power output.

[0029] One updating method is: through the input simulated wind speed features and original wind speed features, the neural network model iteratively updates the weights of its filter layer and multi-layer perceptron, and learns how to map these input features to the predicted output of accumulated wind power.

[0030] Step S103, generating N simulated statistics based on the latest N first weights, the latest N second weights and the latest N third weights.

[0031] Each simulated statistic indicates the difference between a simulated wind speed feature and a mapped original wind speed feature, and each simulated statistic maps an original wind speed feature. The purpose is to generate a simulated statistic based on the latest weights obtained by the neural network to measure the difference between the simulated wind speed feature and the corresponding original wind speed feature.

[0032] One way to generate is that each simulated statistic indicates the difference between a simulated wind speed feature and its corresponding original wind speed feature. The difference may be calculated based on the change of weights or directly based on the difference of input features. The statistic reflects the accuracy of the simulated wind speed feature and its importance in the model prediction.

[0033] Step S104, determining a counterfeit threshold based on a preset error rate threshold and N counterfeit statistics.

[0034] The purpose is to determine a counterfeit threshold based on a preset error rate threshold and N counterfeit statistics.

[0035] One way to determine this is: the error rate threshold is used to measure the tolerance for forecast errors and is set based on historical data analysis or model evaluation indicators. The relationship between the simulated statistic and the error rate helps determine which raw wind speed features will be considered unreliable and thus screened out or replaced.

[0036] Step S105, selecting the original wind speed feature whose mapped imitation statistic is greater than the imitation threshold from the N original wind speed features, obtaining the selected original wind speed feature, and using the numerical weather forecast data corresponding to the selected original wind speed feature as the numerical weather forecast data of the wind power prediction model.

[0037] The purpose is to select those features that are greatly different from the simulated statistics, that is, the difference is greater than the simulated threshold, from the N original wind speed features, and input them into the wind power prediction model as effective numerical weather forecast data.

[0038] One selection method is to compare the difference of the original wind speed features (derived from the imitation statistic) with the imitation threshold, and select features with larger differences, indicating that these features contribute more to the prediction model and can better correlate with the wind power learned by the model. These selected features will be used as input data to improve the prediction accuracy of the wind power prediction model.

[0039] The numerical weather forecast data screening method of the wind power prediction model provided by the present invention, by introducing a neural network model, using the weighted processing of wind speed characteristics and a multi-layer perceptron structure, makes the processing of the original wind speed characteristics and the simulated wind speed characteristics more accurate, thereby improving the accuracy of wind power prediction. By generating a simulated statistic and comparing it with a preset error rate threshold, the most relevant wind speed characteristics can be automatically screened out, avoiding the interference of redundant or irrelevant data, making the input data of the model more effective and representative. The iterative update and weighting mechanism in the neural network enable the model to adaptively adjust the weights in multiple rounds of training, gradually adapt to the conditions of wind farm groups in different regions, thereby enhancing the adaptability of the model to different environmental conditions. The method is not only applicable to the original wind speed characteristic data, but also can process simulated wind speed characteristics, so that the numerical weather forecast data can be better utilized and the prediction ability of complex weather changes can be improved. By setting the imitation threshold, the method can automatically select data with strong relevance to the prediction task, avoid manual intervention, reduce the workload of manual data screening, and improve the automation of data processing.

[0040] Furthermore, a centralized short-term wind power prediction model for a wind farm group can be constructed by using the Transformer model. In this model, the input features are the NWP data corresponding to the filtered original wind speed features, including the predicted values ​​of wind speed and wind direction for 24 hours on the same day, and the output is the actual measured value of regional wind power for 24 hours on the same day. By training these data with the Transformer model, an efficient model that can predict short-term wind power is finally obtained.

[0041] In an optional embodiment, in order to further use the neural network model, the N third weights are the product of the front weight matrix and the rear weight matrix; the N simulated wind speed features and the N original wind speed features are input into the neural network model, including: N simulated wind speed features and N original wind speed features are input into the filtering layer, and the weighted sum of the N simulated wind speed features and the N original wind speed features is determined by using N first weights and N second weights to obtain N fused wind speed features; the N fused wind speed features are mapped to the hidden layer of the multilayer perceptron by using the front weight matrix to obtain the output of the hidden layer; the output of the hidden layer is mapped to the output layer of the multilayer perceptron by using the back weight matrix to obtain the output of the output layer.

[0042] In this embodiment, the simulated wind speed feature can be a feature generated by the original wind speed feature through some preprocessing means, such as wind speed prediction values ​​and simulated values ​​calculated by some meteorological models. The simulated feature helps to capture potential information from different sources in the data. The original wind speed feature can be the actual observed wind speed data. The original wind speed feature often contains real information directly related to wind power. The function of the filter layer is to perform weighted processing on the input N simulated wind speed features and the original wind speed features. Here, each feature is weighted and summed using the first weight and the second weight to obtain the weighted sum as the fused wind speed feature. In this way, the model can more effectively process and fuse wind speed data from different sources. Through the pre-weight matrix, the fused wind speed feature is mapped to the hidden layer of the multilayer perceptron. The multilayer perceptron is a feedforward neural network that can learn complex nonlinear relationships from the input through operations such as activation functions, weights and biases. In this process, the role of the pre-weight matrix is ​​to convert the fused wind speed feature into the hidden layer input required by the neural network for more complex feature learning. The post-weight matrix maps the output of the hidden layer to the output layer. The result of the output layer is usually the predicted value of wind power or other related quantities. Through the operation of the post-weight matrix, the neural network can transform the features learned by the hidden layer into the final prediction result.

[0043] The method provided in this embodiment, by combining the original wind speed features and the simulated wind speed features, can obtain wind speed information from multiple angles, enhance the adaptability of the model to wind speed changes, and reduce the noise or deviation that may be introduced by a single feature. By using the first weight and the second weight to weight different features, the neural network can be dynamically adjusted according to the importance of the features to ensure that the model focuses on the most valuable features for predicting wind power. Through multi-layer perceptrons and pre- and post-weight matrices, the network can effectively learn the complex nonlinear relationship between wind speed and wind power, and improve the accuracy of the prediction. The combination of multi-layer perceptrons and weighted features enables the model to have stronger generalization capabilities and adaptability when facing different wind farms, different time periods and different environmental conditions.

[0044] In an optional implementation, in order to further screen the original wind speed characteristics, the original wind speed characteristics whose mapped simulated statistics are greater than the simulated threshold are selected from N original wind speed characteristics as numerical weather forecast data for the wind power prediction model, including: from the N simulated statistics, searching whether there is a simulated statistic greater than the simulated threshold; if so, counting all simulated statistics greater than the simulated threshold to obtain a target simulated statistic; and using the original wind speed characteristics mapped to the target simulated statistic as the numerical weather forecast data for the wind power prediction model.

[0045] In this embodiment, the simulated statistic is an indicator used to describe wind speed data generated by some methods (such as simulation, external calculation, etc.), which may be a statistical value related to wind speed characteristics, such as mean, variance, standard deviation, etc. The simulated threshold is a set value used to determine whether certain simulated statistics are important enough to be used as effective wind speed characteristics. That is, if a simulated statistic is greater than this threshold, it is considered to have a strong prediction ability. First, the simulated statistics related to the original wind speed characteristics are calculated, and then compared with the simulated threshold to determine which simulated statistics exceed the threshold. Collect all simulated statistics that meet the conditions for subsequent processing. The collection of all simulated statistics that are greater than the simulated threshold is called the target simulated statistic. For the target simulated statistic (the statistic that meets the threshold), the corresponding original wind speed characteristics will be selected as the input data of the wind power prediction model. The connection between the selected wind speed characteristics and the simulated statistics is established through a mapping relationship. The original wind speed characteristics finally selected will be used as numerical weather forecast data and input into the wind power prediction model to help improve the prediction ability of the model.

[0046] The method provided in this embodiment can more accurately select wind speed features that are highly relevant to wind power prediction by screening the counterfeit statistic and the counterfeit threshold. This makes the feature input more reasonable and reduces the interference of irrelevant features or noise on the model. Through the high-quality original wind speed features screened out, the model can obtain more accurate numerical weather forecast data, thereby enhancing the accuracy and robustness of wind power prediction. For a large number of original wind speed features, through this screening and mapping method, the amount of data to be processed can be effectively reduced, redundancy can be reduced, and the efficiency of model training and prediction can be improved. By selecting the original wind speed features corresponding to the counterfeit statistic greater than the counterfeit threshold, it is possible to avoid using noise or irrelevant wind speed features as model input, thereby reducing the risk of overfitting. Numerical weather forecast data is a key input for wind power prediction. The screened original wind speed features can provide higher quality weather data for the wind power prediction model and improve the prediction accuracy.

[0047] In an optional implementation, in order to further determine the fused wind speed features, N simulated wind speed features and N original wind speed features are input into the filtering layer, and the weighted sum of the N simulated wind speed features and the N original wind speed features is determined using N first weights and N second weights to obtain N fused wind speed features, including: Generate the N fused wind speed features according to the N simulated wind speed features, the N original wind speed features, the N first weights, the N second weights, and the first preset relationship; The first preset relationship is:

[0048] in, is the jth fused wind speed feature among the above N fused wind speed features, is the jth original wind speed feature among the above N original wind speed features, is the jth simulated wind speed feature among the above N simulated wind speed features, is the jth first weight among the above N first weights, is the j-th second weight among the above N second weights.

[0049] In this embodiment, the fusion process is achieved by weighting the weighted sum of the original wind speed features and the simulated wind speed features. Each original wind speed feature and the simulated wind speed feature are weighted by corresponding weights, and the impact of each feature on the final result can be determined according to the relevance or importance of each feature. A filter layer can be used to further optimize the selection and weighting of features. In the filter layer, the original wind speed features and the simulated wind speed features will adjust the weighting coefficients according to their feature importance, so as to select the optimal weighted combination, which is very important for ensuring that the fused features have high quality.

[0050] The method provided in this embodiment optimizes the accuracy and reliability of wind speed features by weighted fusion of multiple wind speed features. The method dynamically adjusts the weights so that the method can adapt to the features of different data sources and has strong robustness and low computational complexity.

[0051] In an optional implementation, in order to further determine the counterfeit statistic, the N counterfeit statistics are generated based on the latest N first weights, the latest N second weights and the latest N third weights, including: Generate the N simulated statistics according to the latest N first weights, the latest N second weights, the latest N third weights, and a second preset relationship; The second preset relationship is:

[0052] in, , , is the jth simulation statistic among the above N simulation statistics, is the jth latest first weight among the above N latest first weights, is the jth latest second weight among the above N latest second weights, is the jth latest third weight among the above-mentioned N latest third weights.

[0053] In this embodiment, these input data are weighted through the filtering layer. Specifically, each wind speed feature (whether original or simulated) is associated with a corresponding first weight or second weight, and the role of the weight is to determine the influence of each feature according to the importance of the wind speed feature to the wind power prediction model. Then, using the weighted sum method, the weighted values ​​of each original wind speed feature and the simulated wind speed feature are added to generate a fused wind speed feature.

[0054] The method provided in this embodiment can adjust the weight according to the contribution of each wind speed feature in wind power prediction by obtaining the first weight and the second weight, thereby generating a more comprehensive and predictive fused wind speed feature. The weighted sum fusion method can balance the contribution of information between the original wind speed feature and the simulated wind speed feature, so that the influence of each feature can be flexibly adjusted according to actual needs, avoiding the excessive influence of certain features on the prediction results. The fusion method can effectively suppress the influence of redundant features, reduce the interference of noise on wind power prediction, and improve the reliability of the prediction. By weighted fusion of multiple features, the complexity of the original data can be reduced, the calculation process can be optimized, and the efficiency of model training and prediction can be improved.

[0055] In an optional implementation, in order to further determine the counterfeit threshold, the counterfeit threshold is determined based on the preset error rate threshold and the N counterfeit statistics, including: The absolute values ​​of the N simulated statistics are sorted from large to small.

[0056] Determining the counterfeit threshold according to the preset error rate threshold, the N counterfeit statistics and the third preset relationship; The third preset relationship is: ,

[0057] in, is the above counterfeit threshold, is the error rate threshold preset above, is the jth simulation statistic among the above N simulation statistics, is the set of N non-zero absolute values ​​of the simulated statistics.

[0058] In this embodiment, by first sorting the absolute values ​​of N counterfeit statistics from large to small, t can be selected starting from the absolute value of the largest counterfeit statistic without traversing all counterfeit statistics. After sorting, it is possible to directly determine which feature statistic values ​​have a greater impact on the screening results, simplifying the calculation process. Through the error rate threshold (q), the counterfeit threshold (T) can be accurately controlled to ensure that the error of the counterfeit statistic does not exceed the set tolerance range, thereby improving the robustness of the system. Based on the distribution characteristics of the counterfeit statistic, the threshold (t) is dynamically adjusted so that the counterfeit threshold (T) is more adapted to the actual data situation. In this way, different counterfeit features can be flexibly responded to according to the actual performance of the data. Counterfeit statistics may affect the reliability of the prediction results. By setting a suitable counterfeit threshold, the influence of counterfeit statistics that do not meet expectations can be effectively suppressed, making the model more accurate.

[0059] The method provided in this embodiment can effectively control the error in the counterfeit statistic by determining the counterfeit threshold, so that the system's tolerance for error is consistent with the set error rate threshold. Accurately controlling the impact of the counterfeit statistic can effectively avoid the impact of abnormal data or counterfeit statistics with excessive errors on the stability of the model, thereby improving the overall prediction accuracy. The threshold is dynamically adjusted based on the distribution characteristics of the counterfeit statistic, so that the method can cope with the diversity of different data sets and has strong adaptability.

[0060] The present invention provides a numerical weather forecast data screening device for a wind power prediction model. Figure 2 1 is a schematic diagram of a numerical weather forecast data screening device for a wind power prediction model according to an embodiment of the present invention, the device comprising: The acquisition module 201 is used to acquire the accumulated wind power of at least one wind farm group in the target area in the historical time and N numerical weather forecast data, and N simulated wind speed characteristics mapped one-to-one with the N original wind speed characteristics, wherein the N numerical weather forecast data include the N original wind speed characteristics, and N is a positive integer; An updating module 202 is used to input the N simulated wind speed features and the N original wind speed features into a neural network model, and iteratively update the N first weights and the N second weights of the filter layer of the neural network model, and the N third weights of the multilayer perceptron of the neural network model, until the neural network model outputs the accumulated wind power, and obtains the latest N first weights, the latest N second weights and the latest N third weights, wherein the N first weights are used to weight the N original wind speed features, and the N second weights are used to weight the N simulated wind speed features; A generating module 203, configured to generate N simulated statistics based on the latest N first weights, the latest N second weights and the latest N third weights, wherein each of the simulated statistics indicates a difference between one of the simulated wind speed features and the mapped original wind speed feature, and each of the simulated statistics maps one of the original wind speed features; A determination module 204, configured to determine a counterfeit threshold based on a preset error rate threshold and the N counterfeit statistics; The selection module 205 is used to select the original wind speed feature whose mapped imitation statistic is greater than the imitation threshold from the above-mentioned N original wind speed features, obtain the selected original wind speed feature, and use the numerical weather forecast data corresponding to the selected original wind speed feature as the numerical weather forecast data of the wind power prediction model.

[0061] The numerical weather forecast data screening device of the wind power prediction model provided by the present invention, by introducing a neural network model, using the weighted processing of wind speed characteristics and a multi-layer perceptron structure, makes the processing of the original wind speed characteristics and the simulated wind speed characteristics more accurate, thereby improving the accuracy of wind power prediction. By generating a simulated statistic and comparing it with a preset error rate threshold, the most relevant wind speed characteristics can be automatically screened out, avoiding the interference of redundant or irrelevant data, making the input data of the model more effective and representative. The iterative update and weighting mechanism in the neural network enable the model to adaptively adjust the weights in multiple rounds of training, gradually adapt to the conditions of wind farm groups in different regions, thereby enhancing the adaptability of the model to different environmental conditions. The method is not only applicable to the original wind speed characteristic data, but also can process simulated wind speed characteristics, so that the numerical weather forecast data can be better utilized and the prediction ability of complex weather changes can be improved. By setting the imitation threshold, the method can automatically select data with strong relevance to the prediction task, avoid manual intervention, reduce the workload of manual data screening, and improve the automation of data processing.

[0062] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0063] The numerical weather forecast data screening device of the wind power prediction model in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0064] The embodiment of the present invention also provides a computer device having the above Figure 2 The numerical weather forecast data screening device of the wind power prediction model shown.

[0065] See also Figure 3 , Figure 3 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 3As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 3 A processor 10 is taken as an example.

[0066] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include an integrated circuit. The integrated circuit may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0067] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0068] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0069] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0070] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0071] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0072] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0073] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for screening numerical weather forecast data for a wind power prediction model, characterized in that: The method comprises: Obtaining the accumulated wind power of at least one wind farm group in a target area in historical time and N numerical weather forecast data, as well as N simulated wind speed characteristics mapped one-to-one with the N original wind speed characteristics, wherein the N numerical weather forecast data include the N original wind speed characteristics, and N is a positive integer; Inputting the N simulated wind speed features and the N original wind speed features into a neural network model, and iteratively updating N first weights and N second weights of a filter layer of the neural network model, and N third weights of a multilayer perceptron of the neural network model, until the neural network model outputs the accumulated wind power, to obtain the latest N first weights, the latest N second weights, and the latest N third weights, wherein the N first weights are used to weight the N original wind speed features, and the N second weights are used to weight the N simulated wind speed features; Based on the latest N first weights, the latest N second weights and the latest N third weights, N simulated statistics are generated, wherein each of the simulated statistics indicates the difference between one simulated wind speed feature and the mapped original wind speed feature, and each of the simulated statistics maps one original wind speed feature; Determining a counterfeit threshold based on a preset error rate threshold and the N counterfeit statistics; An original wind speed feature whose mapped imitation statistic is greater than the imitation threshold is selected from the N original wind speed features to obtain a selected original wind speed feature, and numerical weather forecast data corresponding to the selected original wind speed feature is used as numerical weather forecast data for a wind power prediction model.

2. The method according to claim 1, characterized in that The N third weights are the product of the front weight matrix and the rear weight matrix; The step of inputting the N simulated wind speed characteristics and the N original wind speed characteristics into a neural network model comprises: Inputting the N simulated wind speed features and the N original wind speed features into the filtering layer, and determining a weighted sum of the N simulated wind speed features and the N original wind speed features using the N first weights and the N second weights to obtain N fused wind speed features; Mapping the N fused wind speed features to a hidden layer of the multilayer perceptron using the pre-weight matrix to obtain an output of the hidden layer; The output of the hidden layer is mapped to the output layer of the multilayer perceptron using the post-weight matrix to obtain the output of the output layer.

3. The method according to claim 1, characterized in that The method further comprises selecting an original wind speed feature whose mapped imitation statistic is greater than the imitation threshold from the N original wind speed features as numerical weather forecast data of the wind power prediction model, including: From the N counterfeit statistics, searching whether there is a counterfeit statistic greater than the counterfeit threshold; If yes, then all counterfeit statistics greater than the counterfeit threshold are counted to obtain a target counterfeit statistic; The original wind speed characteristics mapped by the target simulated statistics are used as numerical weather forecast data of the wind power prediction model.

4. The method according to claim 2, characterized in that: The step of inputting the N simulated wind speed features and the N original wind speed features into the filtering layer, and using the N first weights and the N second weights to determine the weighted sum of the N simulated wind speed features and the N original wind speed features to obtain N fused wind speed features comprises: Generate the N fused wind speed features according to the N simulated wind speed features, the N original wind speed features, the N first weights, the N second weights, and a first preset relationship; The first preset relationship is: in, is the jth fused wind speed feature among the N fused wind speed features, is the jth original wind speed feature among the N original wind speed features, is the jth simulated wind speed feature among the N simulated wind speed features, is the jth first weight among the N first weights, is the j-th second weight among the N second weights.

5. The method according to any one of claims 1 to 4, characterized in that: The generating N simulated statistics based on the latest N first weights, the latest N second weights and the latest N third weights comprises: Generate the N simulated statistics according to the latest N first weights, the latest N second weights, the latest N third weights, and a second preset relationship; The second preset relationship is: in, , , is the jth replica statistic among the N replica statistics, is the jth latest first weight among the latest N first weights, is the jth latest second weight among the latest N second weights, It is the jth latest third weight among the N latest third weights.

6. The method according to any one of claims 1 to 4, characterized in that: The step of determining the counterfeit threshold based on the preset error rate threshold and the N counterfeit statistics includes: Sort the absolute values ​​of the N simulated statistics from large to small; Determining the counterfeit threshold according to the preset error rate threshold, the N counterfeit statistics and a third preset relationship; The third preset relationship is: ， in, is the counterfeit threshold, is the preset error rate threshold, is the jth replica statistic among the N replica statistics, is the set of N non-zero absolute values ​​of the simulated statistics.

7. A numerical weather forecast data screening device for a wind power prediction model, characterized in that: The device comprises: An acquisition module, used to acquire the accumulated wind power of at least one wind farm group in a target area in historical time and N numerical weather forecast data, and N simulated wind speed characteristics mapped one-to-one with the N original wind speed characteristics, wherein the N numerical weather forecast data include the N original wind speed characteristics, and N is a positive integer; An updating module, used for inputting the N simulated wind speed features and the N original wind speed features into a neural network model, and iteratively updating N first weights and N second weights of a filter layer of the neural network model, and N third weights of a multilayer perceptron of the neural network model, until the neural network model outputs the accumulated wind power, and obtaining the latest N first weights, the latest N second weights, and the latest N third weights, wherein the N first weights are used to weight the N original wind speed features, and the N second weights are used to weight the N simulated wind speed features; A generating module, configured to generate N simulated statistics based on the latest N first weights, the latest N second weights and the latest N third weights, wherein each of the simulated statistics indicates a difference between one of the simulated wind speed features and the mapped original wind speed feature, and each of the simulated statistics maps one of the original wind speed features; A determination module, configured to determine a counterfeit threshold based on a preset error rate threshold and the N counterfeit statistics; A selection module is used to select the original wind speed feature whose mapped imitation statistic is greater than the imitation threshold from the N original wind speed features, obtain the selected original wind speed feature, and use the numerical weather forecast data corresponding to the selected original wind speed feature as the numerical weather forecast data of the wind power prediction model.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the numerical weather forecast data screening method for the wind power prediction model according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the numerical weather forecast data screening method for a wind power prediction model according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to enable a computer to execute the numerical weather forecast data screening method for the wind power prediction model according to any one of claims 1 to 6.

Images