Wind power generation power prediction method and system
By extracting and fusion of on-site observed weather data and numerical weather forecast data, the problem of inaccurate wind power generation prediction caused by relying on a single input in the prior art is solved, and more accurate and reliable wind power generation prediction is achieved.
Patent Information
- Application Number
- CN202411872336.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-05-30
AI Technical Summary
In the prior art, the reliance on single input data for wind power predictions has resulted in inaccurate predictions.
By inputting on-site observed weather data and numerical weather forecast data into the power prediction model after training, a set neural network is used for feature extraction, and the feature sequences of the two data are fused through a feature adaptive fusion algorithm to obtain the predicted wind power generation power.
Through reasonable and effective feature fusion, data reflecting real weather conditions can be more comprehensive, thereby improving the accuracy and reliability of wind power forecasts.
Smart Images
Figure CN120069147A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of new energy power generation prediction, and particularly relates to a wind power generation prediction method and system. Background Art
[0002] With the increasing attention of people to energy shortage and environmental problems, wind power generation, as an alternative solution, can effectively alleviate the global warming problem caused by the large-scale use of fossil fuels, and thus has received more and more attention.
[0003] However, the large-scale integration of wind power into the power system will cause problems such as volatility, intermittency, and randomness, posing severe challenges to the safety, stability, and economy of the power system operation. Therefore, the accurate prediction of wind power generation has become a key factor in wind power grid connection governance and integration. Accurate wind power prediction can be used to reasonably arrange power generation plans and maintain grid balance, providing a reliable basis for optimizing grid operation and block management, that is, the power generation plan and operation and maintenance strategy can be optimized according to the prediction results of wind power. By reasonably arranging maintenance, shutdown, and fault handling, etc., the power generation efficiency and reliability of the wind farm can be maximized.
[0004] In recent years, with the rapid development of artificial intelligence technology, many new wind power generation prediction algorithms represented by machine learning or deep learning have emerged. The literature (Peng Jianing, Xu Heyong. Wind power generation prediction algorithm based on random forest and support vector regression [J]. Thermal Energy and Power Engineering, 2024, 39(05): 143-149.) proposed an accurate wind power generation prediction algorithm based on a random forest model and a support vector regression model. This algorithm is based on regression trees and random forest models to construct an optimal feature set, and uses the optimal feature set as input to achieve the prediction of wind power generation. The literature (Liu Xuli, Mo Yuchang, Wu Zhe, etc. Hybrid deep learning model for ultra-short-term wind power prediction [J]. Journal of Huaqiao University (Natural Science Edition), 2022, 43(05): 668-676.) proposed a hybrid deep learning model based on discrete wavelet transform, temporal convolutional network, and long short-term memory neural network for predicting ultra-short-term wind power.
[0005] Wind power generation prediction is carried out based on traditional artificial intelligence algorithms. That is, for example, in the wind power generation prediction algorithms based on random forest and support vector regression, in-situ observed weather data and original wind power data are input for prediction; in the hybrid deep learning model for ultra-short-term wind power prediction, original wind power data is input for prediction. This way of predicting wind power generation by inputting single meteorological data has limitations, that is, it overly relies on the weather conditions and wind conditions reflected by the single input meteorological data, and it is difficult to provide a multi-dimensional reflection of the real weather and wind conditions, resulting in inaccurate prediction results obtained from the wind power generation prediction based on a single input. Summary of the Invention
[0006] The purpose of the present invention is to provide a wind power generation prediction method and system, which are used to solve the problem in the prior art that the prediction results obtained by relying on data with a single input for wind power generation prediction are inaccurate.
[0007] To achieve the above purpose, the present invention provides a wind power generation prediction method. In this method, in-situ observed weather data and numerical weather prediction data are input into a trained power prediction model to obtain the predicted wind power generation; the training method of the power prediction model includes:
[0008] 1) Feature extraction is respectively carried out on a set of in-situ observed weather data and numerical weather prediction data samples in the training set through the set neural network of the power prediction model to obtain corresponding feature sequences;
[0009] 2) Feature fusion is carried out on the feature sequence corresponding to the in-situ observed weather data and the feature sequence corresponding to the numerical weather prediction data samples through the feature adaptive fusion algorithm of the power prediction model. The obtained fusion features are then passed through the set fully connected layer to obtain the predicted wind power generation; the parameters of the power prediction model are updated according to the predicted wind power generation and the actual value of the power generation corresponding to the sample; Steps 1)-2) are iterated until the stop condition is met.
[0010] Beneficial effects: The present invention provides a new wind power prediction method based on different data-driven and feature fusion. This method trains a feature sequence capable of extracting in-situ observed weather data and numerical weather prediction data, and then fuses the feature sequences corresponding to these two types of data to obtain a fusion feature of these two types of data. A power prediction model for outputting the predicted wind power is obtained based on this fusion feature, so as to achieve the effect of predicting wind power through in-situ observed weather data and numerical weather prediction data. Therefore, through this method, the data information of the relevant weather conditions provided by the input of two different dimensions of weather data can be reasonably and effectively feature-fused, making the weather conditions reflected by the two different dimensions of weather data closer to the real situation, so that the prediction results obtained by using this method for wind power prediction are more accurate and reliable.
[0011] Further, the set neural network is a bidirectional long short-term memory network.
[0012] Further, the method for fusing the feature sequence corresponding to the in-situ observed weather data and the feature sequence corresponding to the numerical weather prediction data by the feature adaptive fusion algorithm of the power prediction model includes:
[0013]
[0014] Among them, is the fusion feature; is the feature sequence obtained by processing the feature sequence corresponding to the in-situ observed weather data using a fully connected layer and a sigmoid activation function; is the feature sequence obtained by processing the feature sequence corresponding to the numerical weather prediction data using a fully connected layer; F mul is the set operator for channel multiplication.
[0015] Further, the in-situ observed weather data includes data corresponding to at least one parameter selected from each in-situ observed weather characteristic parameter according to the linear correlation degree with the power generation power; each in-situ observed weather characteristic parameter includes: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure.
[0016] Further, the numerical weather prediction data includes data corresponding to at least one parameter selected from each numerical weather prediction characteristic parameter according to the linear correlation degree with the power generation power; each numerical weather prediction characteristic parameter includes: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure.
[0017] Further, the method for selecting at least one parameter from the local observed weather characteristic parameters according to the degree of linear correlation with the power generation is as follows: calculating the degree of linear correlation between each local observed weather characteristic parameter and the power generation according to the Pearson correlation coefficient, and screening out at least one parameter with the largest correlation with the power generation; the formula for calculating the degree of linear correlation according to the Pearson correlation coefficient is:
[0018]
[0019] where r xy is the degree of linear correlation between each parameter and the power generation; is the mean value of each parameter; is the mean value of the power generation; m is the number in the feature sequence.
[0020] Further, the method for selecting at least one parameter from the numerical weather prediction characteristic parameters according to the degree of linear correlation with the power generation is as follows: calculating the degree of linear correlation between each numerical weather prediction characteristic parameter and the power generation according to the Pearson correlation coefficient, and screening out at least one parameter with the largest correlation with the power generation; the formula for calculating the degree of linear correlation according to the Pearson correlation coefficient is:
[0021]
[0022] where r xy is the degree of linear correlation between each parameter and the power generation; is the mean value of each parameter; is the mean value of the power generation; m is the number in the feature sequence.
[0023] Further, the local observed weather characteristic parameters include: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure;
[0024] The numerical weather prediction characteristic parameters include: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure.
[0025] Further, the local observed weather data and the numerical weather prediction data are respectively obtained by performing data cleaning processing, feature parameter screening, and standardization processing on the collected original local observed weather data and the collected original numerical weather prediction data.
[0026] The present invention also provides a new wind power prediction system based on different data-driven and feature fusion, including a processor for executing a computer program to implement the steps of the wind power prediction method based on different data-driven and feature fusion.
[0027] The wind power prediction system based on different data-driven and feature fusion can achieve the same beneficial effects as the above-mentioned wind power prediction method based on different data-driven and feature fusion. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 It is a flowchart example of the training method of the power prediction model in the embodiment of the wind power prediction method of the present invention;
[0029] Figure 2a It is a schematic diagram of data processing of in-situ observed weather data and numerical weather forecast data in the embodiment of the wind power prediction method of the present invention;
[0030] Figure 2b It is a schematic diagram of feature fusion of the feature sequence corresponding to the in-situ observed weather data and the feature sequence corresponding to the numerical weather forecast data sample in the embodiment of the wind power prediction method of the present invention;
[0031] Figure 2c It is a schematic diagram of obtaining the predicted wind power through the set fully connected layer by the fusion feature in the embodiment of the wind power prediction method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0032] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0033] Embodiment of the wind power prediction method
[0034] This embodiment provides a wind power prediction method, which constructs a power prediction model and inputs two different-dimensional weather data into the power prediction model to obtain the predicted wind power. The model is trained through two important steps (i.e., the feature extraction step and the feature fusion step), that is, the method extracts features from two different-dimensional weather data respectively, and then fuses the two groups of feature sequences to obtain the fusion feature to predict the wind power.
[0035] The wind power prediction method of this embodiment is specifically: input the in-situ observed weather data and the numerical weather forecast data into the trained power prediction model to obtain the predicted wind power; wherein the training method of the power prediction model includes:
[0036] 1) The set neural network of the power prediction model (in this embodiment, the set neural network is a bidirectional long short-term memory network) extracts features from a set of in-situ observed weather data and numerical weather prediction data samples in the training set respectively to obtain corresponding feature sequences;
[0037] 2) The feature adaptive fusion algorithm of the power prediction model fuses the feature sequences corresponding to the in-situ observed weather data and the feature sequences corresponding to the numerical weather prediction data samples;
[0038] In this embodiment, the method of fusing the feature sequences corresponding to the in-situ observed weather data and the feature sequences corresponding to the numerical weather prediction data by the feature adaptive fusion algorithm of the power prediction model includes:
[0039]
[0040] In the above formula, is the fused feature; is the feature sequence obtained by processing the feature sequence corresponding to the in-situ observed weather data with a fully connected layer and a sigmoid activation function; is the feature sequence obtained by processing the feature sequence corresponding to the numerical weather prediction data with a fully connected layer; F mul is the set operator of channel multiplication.
[0041] The obtained fused feature is then passed through a set fully connected layer to obtain the predicted wind power generation; the parameters of the power prediction model are updated according to the predicted wind power generation and the actual value of the power generation corresponding to the sample; Iterate steps 1)-2) until the stop condition is met.
[0042] Specifically, as Figure 1 shown, before predicting the wind power generation, the training set data obtained from the previously collected in-situ observed weather data and numerical weather prediction data is used as input to train the constructed wind power generation prediction model (i.e., the power prediction model) to obtain the trained wind power generation prediction model.
[0043] After obtaining the above trained wind power generation prediction model, the processed in-situ observed weather data and numerical weather prediction data are input into the model to obtain the predicted wind power generation.
[0044] The training method of the wind power generation prediction model is specifically to perform feature extraction first and then feature fusion, where the feature extraction part specifically includes:
[0045] As Figure 2aAs shown, in this embodiment, a bidirectional long short-term memory network (Bi-LSTM network) is used to extract the in-situ observed weather data features from the in-situ observed weather data in the training set. Extract the numerical weather prediction data features from the numerical weather prediction data in the training set. This neural network uses two independent fully connected layers to learn forward and backward simultaneously. The extraction in both directions improves the information content and stability. In other embodiments, existing unidirectional LSTM networks (long short-term memory networks), RNN networks (recursive neural networks), or Transformer structures (a neural network structure based on self-attention mechanism, suitable for processing long sequence data) can also be used for feature extraction to extract the feature sequences corresponding to the in-situ observed weather data and the numerical weather prediction data.
[0046] Specifically, the expression of the Bi-LSTM output is as follows:
[0047]
[0048] Where LSTM(·) is the traditional LSTM function. are the weights of the forward hidden sequence and the backward hidden sequence respectively, and b represents the bias term of the Bi-LSTM output layer. The input of the LSTM is the input value at the current moment. The output value h at the previous moment. t-1 And the cell state c at the previous moment. t-1 , and the output cell state c. t And the output value h. t . The key of the LSTM lies in the control of the long-term state c. Through the forget gate (f t ), input gate (i t ), and output gate (o t ), information is removed or added to the cell state. The forget gate determines how much of the previous cell state c t-1 is retained to the current moment c t ; the input gate determines how much of the current input of the network is saved to the cell state c t ; the output gate controls the influence of the long-term memory on the current output.
[0049] The expression of the LSTM output is as follows:
[0050]
[0051] h t = o t tanh(c t )
[0052] Where, f t , it , o t , c t , h t are the states of the forget gate, input gate, output gate, input node, and intermediate output, respectively; represents the intermediate input; W f , W i , W o , W c represents the matrix weights of the corresponding gate; b f , b i , b o , b c represent the bias terms of each gate, respectively; Sigmoid represents the sigmoid activation function; tanh represents the tanh activation function.
[0053] The feature fusion part specifically includes:
[0054] After feature extraction is completed, as Figure 2b shown, a feature adaptive fusion gate is used to obtain the fused feature, and the calculation formula of the fused feature is as follows:
[0055]
[0056] Among them, represents the fused feature; f joint-lmd and f joint-nwp represent the processed in-situ observed weather data features and the processed numerical weather prediction data features, respectively (that is, a fully connected layer and the sigmoid activation function are used for f lmd , and a fully connected layer is used for f nwp to obtain the corresponding f joint-lmd and f joint-nwp ); F mul represents channel multiplication (in this embodiment, it represents the multiplication of two one-dimensional features. For example, a = [a1, a2,..., an] and b = [b1, b2,..., bn], and the new feature obtained after channel multiplication F mul is c = [a1*b1, a2*b2,..., an*bn]); the calculation formulas of f joint-lmd and f joint-nwp are as follows:
[0057]
[0058] Among them, σ is the sigmoid activation function, W d1 , W d2 , W p1 , W p2 represent the fully connected layer weights, b d1 , b d2 , b p1, b p2 represents the bias term of the fully connected layer; RELU is the rectified linear unit, which is an activation function widely used in neural networks.
[0059] As Figure 2c shown, after obtaining the fused features (i.e., the joint features shown in the figure), input the fused features into the set fully connected layer (in this embodiment, the set fully connected layer is a multi-layer fully connected layer. Using a multi-layer fully connected layer can better extract the information of the fused features and improve the accuracy of the prediction result. Figure 2b As shown in
[0060] ① Set a precision threshold during model training; that is, the stopping condition is: when the prediction accuracy of the network reaches or exceeds this precision threshold, the training process can stop. This usually means that the network has fully learned the features in the training data and can generalize to unseen data to a certain extent.
[0061] ② Set a maximum number of iterations during model training; that is, the stopping condition is: when the number of iterations reaches this limit, regardless of the performance of the network, the training will stop. This helps to avoid overfitting and waste of computing resources.
[0062] ③ During training, usually a part of the data is used as a validation set to evaluate the performance of the network; that is, the stopping condition is: when the validation error reaches a certain condition (such as the validation error is less than the training error, the validation error no longer decreases significantly, etc.), the training can stop. This helps to prevent overfitting and ensure that the network has good generalization ability on unseen data. In other embodiments, other iteration stopping conditions can also be selected, which will not be elaborated here.
[0063] In this embodiment, the above-mentioned on-site observed weather data includes the data corresponding to at least one parameter selected from each on-site observed weather characteristic parameter according to the linear correlation degree with the power generation; each on-site observed weather characteristic parameter includes: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure.
[0064] The method of selecting at least one parameter from each local observed weather characteristic parameter according to the degree of linear correlation with the power generation is as follows: calculating the degree of linear correlation between each local observed weather characteristic parameter and the power generation according to the Pearson correlation coefficient, and screening out at least one parameter with the largest correlation with the power generation; the formula for calculating the degree of linear correlation according to the Pearson correlation coefficient is:
[0065]
[0066] where r xy is the degree of linear correlation between each parameter and the power generation; is the mean value of each parameter; is the mean value of the power generation; m is the number in the feature sequence.
[0067] Specifically, the local observed weather data contains each local observed weather characteristic parameter, and these characteristic parameters include: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure.
[0068] In this embodiment, the Pearson correlation coefficient is used to calculate the degree of linear correlation between the above-mentioned local observed weather characteristic parameters and the power generation, and at least one parameter with the largest correlation with the power generation is screened out. The data corresponding to the screened parameter is used as the local observed weather data participating in feature extraction. The Pearson correlation coefficient is used to calculate and analyze the degree of linear correlation between each local observed weather characteristic parameter and the power generation, and the formula is as follows:
[0069]
[0070] where r xy is the degree of linear correlation between each local observed weather characteristic parameter and the power generation; is the mean value of each local observed weather characteristic parameter; is the mean value of the power generation; m is the number in the feature sequence.
[0071] In this embodiment, the above-mentioned numerical weather prediction data includes the data corresponding to at least one parameter selected from each numerical weather prediction characteristic parameter according to the degree of linear correlation with the power generation; each numerical weather prediction characteristic parameter includes: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure.
[0072] The method of selecting at least one parameter from each numerical weather prediction characteristic parameter according to the degree of linear correlation with the power generation includes: calculating the degree of linear correlation between each numerical weather prediction characteristic parameter and the power generation according to the Pearson correlation coefficient, and screening out at least one parameter with the largest correlation with the power generation; the formula for calculating the degree of linear correlation according to the Pearson correlation coefficient is:
[0073]
[0074] where r xy is the degree of linear correlation between each parameter and the power generation; is the mean value of each parameter; is the mean value of the power generation; m is the number in the feature sequence.
[0075] Specifically, the numerical weather prediction data contains each numerical weather prediction characteristic parameter, and these characteristic parameters include: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the fan hub, wind speed at the fan hub, temperature, humidity, and air pressure.
[0076] The method of selecting at least one parameter from each numerical weather prediction characteristic parameter according to the degree of linear correlation with the power generation includes: calculating the degree of linear correlation between each numerical weather prediction characteristic parameter and the power generation according to the Pearson correlation coefficient, and screening out at least one parameter with the largest correlation with the power generation; the formula for calculating the degree of linear correlation according to the Pearson correlation coefficient is:
[0077]
[0078] where r xy is the degree of linear correlation between each numerical weather prediction characteristic parameter and the power generation; is the mean value of each numerical weather prediction characteristic parameter; is the mean value of the power generation; m is the number in the feature sequence.
[0079] In this embodiment, the above-mentioned on-site observed weather data and numerical weather prediction data are obtained by respectively performing data cleaning, feature parameter screening, and standardization processing on the collected original on-site observed weather data and the collected original numerical weather prediction data.
[0080] Specifically, after collecting the original on-site observed weather data and the original numerical weather prediction data, performing data cleaning, feature parameter screening, and standardization processing on them can obtain the above-mentioned on-site observed weather data and numerical weather prediction data.
[0081] In this embodiment, the process of data cleaning includes: deleting interfering data, filling the missing values with linear interpolation and the mean of the previous and next values, and then processing each column with fitting curve interpolation; handling outliers with breakpoint setting and mean replacement.
[0082] Standardize the in-situ observed weather data and numerical weather prediction data. Specifically, standardization performs a linear transformation on the data and maps the result to the range of 0 - 1. The transformation formula is as follows:
[0083]
[0084] where is the standardized feature data (i.e., the standardized in-situ observed weather feature parameters and the standardized numerical weather prediction feature parameters); x is the sample data before standardization (i.e., the in-situ observed weather feature parameters before standardization or the numerical weather prediction feature parameters before standardization), max(x) is the maximum value, min(x) is the minimum value, mean(x) is the data mean, and σ is the standard deviation of the sample data;
[0085] In this embodiment, the standardized data is divided into training set data, validation set, and test set data according to the ratios of 60%, 20%, and 20% respectively. The training set data is used as the input of the wind power prediction model to train the model; at the same time, the validation set data is added to preliminarily verify the performance of the model; finally, the test set data is added for testing, and the accuracy rate (C R ) is used to verify the model performance. The formula is:
[0086]
[0087] In the above formula, observed i represents the actual power generation at time i; predicted i represents the predicted power generation at time i; n represents the number of data; C i represents the installed capacity at time i.
[0088] Embodiment of Wind Power Prediction System
[0089] This embodiment provides a wind power prediction system, which includes a processor. The processor stores executable program instructions for implementing the wind power prediction method in the above-mentioned embodiment of the wind power prediction method.
[0090] Since the specific working mode and working principle of the wind power prediction system in this embodiment have been described in detail in the above-mentioned embodiment of the wind power prediction method, they will not be elaborated here.
[0091] It should be understood that the above specific embodiments of the present invention are only used for illustrative explanation or interpretation of the principles of the present invention, and do not constitute a limitation on the present invention.
Claims
1. A method for predicting wind power generation, characterized in that: The local weather observation data and numerical weather forecast data are input into the trained power prediction model to obtain the predicted wind power generation power; the training method of the power prediction model includes: 1) The power prediction model is used to extract features from a set of local weather observation data and numerical weather forecast data samples in the training set to obtain corresponding feature sequences; 2) The feature sequence corresponding to the local weather observation data and the feature sequence corresponding to the numerical weather forecast data sample are feature fused through the feature adaptive fusion algorithm of the power prediction model, and the obtained fusion features are then used to obtain the predicted wind power generation power through the set full connection layer; the parameters of the power prediction model are updated according to the predicted wind power generation power and the actual value of the power generation power corresponding to the sample; steps 1)-2) are iterated until the stop condition is met.
2. The method for predicting wind power generation according to claim 1, characterized in that: The neural network is set to be a bidirectional long short-term memory network.
3. The wind power generation power prediction method according to claim 1 or 2, characterized in that: The method of fusing the feature sequence corresponding to the local observation weather data and the feature sequence corresponding to the numerical weather forecast data by the feature adaptive fusion algorithm of the power prediction model includes: in, For fusion features; The feature sequence corresponding to the locally observed weather data is obtained by processing it with a fully connected layer and a sigmoid activation function; The feature sequence corresponding to the numerical weather forecast data is processed by a fully connected layer and obtained; F mul Set operator for channel multiplication.
4. The wind power generation prediction method according to claim 1 or 2, characterized in that: The locally observed weather data include data corresponding to at least one parameter selected from various locally observed weather characteristic parameters according to the degree of linear correlation with the power generation power; the locally observed weather characteristic parameters include: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the wind turbine hub, wind speed at the wind turbine hub, temperature, humidity and air pressure.
5. The wind power generation prediction method according to claim 1 or 2, characterized in that: The numerical weather forecast data include data corresponding to at least one parameter selected from each numerical weather forecast characteristic parameter according to the degree of linear correlation with the power generation power; each numerical weather forecast characteristic parameter includes: 10-meter wind direction, 10-meter wind speed, 30-meter wind direction, 30-meter wind speed, 50-meter wind direction, 50-meter wind speed, 70-meter wind direction, 70-meter wind speed, wind direction at the wind turbine hub, wind speed at the wind turbine hub, temperature, humidity and air pressure.
6. The method for predicting wind power generation according to claim 4, characterized in that: The method of selecting at least one parameter from each locally observed weather characteristic parameter according to the degree of linear correlation with the power generation includes: calculating the degree of linear correlation between each locally observed weather characteristic parameter and the power generation according to the Pearson correlation coefficient, and selecting at least one parameter with the greatest correlation with the power generation; the formula for calculating the degree of linear correlation according to the Pearson correlation coefficient is: Among them, r xy is the linear correlation degree between each parameter and the generated power; is the mean value of each parameter; is the mean value of the generated power; m is the number of characteristic sequences.
7. The method for predicting wind power generation according to claim 5, characterized in that: The method of selecting at least one parameter from each numerical weather forecast characteristic parameter according to the degree of linear correlation with the power generation includes: calculating the degree of linear correlation between each numerical weather forecast characteristic parameter and the power generation according to the Pearson correlation coefficient, and screening out at least one parameter with the greatest correlation with the power generation; the formula for calculating the degree of linear correlation according to the Pearson correlation coefficient is: Among them, r xy is the linear correlation degree between each parameter and the generated power; is the mean value of each parameter; is the mean value of the generated power; m is the number of characteristic sequences.
8. The wind power generation power prediction method according to claim 1 or 2, characterized in that: The on-site weather observation data and the numerical weather forecast data are obtained by respectively performing data cleaning processing, characteristic parameter screening and standardization processing on the collected original data of on-site weather observation and the collected original data of numerical weather forecast.
9. A wind power generation power prediction system, comprising a processor, characterized in that: The processor is used to execute a computer program to implement the steps of the wind power generation prediction method according to any one of claims 1 to 8.