Offshore wind power prediction method and system based on neural network

Abstract

The invention relates to the technical field of data analysis, in particular to an offshore wind power prediction method and system based on a neural network. The method comprises the following steps: firstly, acquiring current meteorological data, historical power data and historical meteorological data of a wind power plant; then, a historical meteorological data set meeting a threshold value judgment condition is obtained through window sliding and meteorological data similarity calculation, and mean value calculation is carried out on the historical power data corresponding to the historical meteorological data set to obtain first predicted power data; then, determining a training strategy of a wind power prediction model based on a hybrid neural network, and inputting the current meteorological data into the wind power prediction model to obtain second prediction power data; evaluating the first predicted power data and the second predicted power data to obtain a wind power predicted value; and finally, correcting the wind power predicted value by using an offshore wind power correction formula to obtain a final wind power predicted value.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method and system for predicting offshore wind power based on a neural network. Background Art

[0002] As an important low-carbon clean energy, wind energy will play an important role in the future energy structure transformation due to its sustainable utilization characteristics. However, the power output of offshore wind power is relatively unstable, which is mainly due to the randomness of wind energy itself. Therefore, accurate prediction of offshore wind power in advance has become an indispensable technical requirement.

[0003] The power generation of offshore wind power is affected by many factors, especially wind speed and wind direction, which are the key factors in determining the power generation. Since meteorological factors often show periodic fluctuations related to various factors such as seasons, atmospheric circulation and geographical location, the fluctuation of offshore wind power has significant trend and periodic characteristics. Through in-depth analysis of these meteorological variables, the accuracy of offshore wind power prediction can be effectively improved, thereby providing a scientific basis for power dispatch and reserve.

[0004] At present, the existing technology still has shortcomings in offshore wind power forecasting; the existing technology does not take into account factors such as sudden changes in wind speed, temperature and wind direction in the offshore wind power forecasting process, resulting in large forecasting errors. In addition, the existing technology ignores the operating status of the wind farm itself, the impact of equipment aging and changes in meteorological conditions, which limits the forecasting accuracy and reduces the effectiveness and accuracy of offshore wind power forecasting.

[0005] Therefore, a neural network-based offshore wind power prediction method and system are proposed. Summary of the invention

[0006] The purpose of the present invention is to provide an offshore wind power prediction method and system based on a neural network. First, the current meteorological data, historical power data and historical meteorological data of the wind farm are obtained; then, a historical meteorological data set that meets the threshold judgment condition is obtained by using window sliding and meteorological data similarity calculation, and the historical power data corresponding to the historical meteorological data set is averaged to obtain the first predicted power data; then, a training strategy for a wind power prediction model based on a hybrid neural network is determined, and the current meteorological data is input into the wind power prediction model to obtain the second predicted power data; the first predicted power data and the second predicted power data are evaluated to obtain a wind power prediction value; finally, the wind power prediction value is corrected using an offshore wind power correction formula to obtain a final wind power prediction value.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for predicting offshore wind power based on a neural network, comprising:

[0009] Obtain current meteorological data, historical power data and historical meteorological data of the wind farm;

[0010] Constructing a current meteorological data matrix and a historical meteorological data matrix respectively according to the current meteorological data and the historical meteorological data; wherein the current meteorological data is a set of continuous meteorological data;

[0011] The historical meteorological data matrix is ​​extracted by using a window sliding method to obtain a historical meteorological data set, and meteorological data similarity is calculated with the current meteorological data matrix, and the meteorological data similarity and the historical power data are combined to obtain the first predicted power data;

[0012] Constructing a wind power prediction model based on a hybrid neural network, training the wind power prediction model according to the first predicted power data, the historical meteorological data and the historical power data to obtain a final wind power prediction model; inputting the current meteorological data into the final wind power prediction model to obtain second predicted power data; performing weighted summation on the first predicted power data and the second predicted power data to obtain a wind power prediction value;

[0013] The wind power forecast value is corrected using an offshore wind power correction formula to obtain a final wind power forecast value.

[0014] Furthermore, the current meteorological data includes: wind speed, wind direction, air pressure, temperature and humidity.

[0015] Further, the historical meteorological data matrix is ​​extracted by using a window sliding method to obtain a historical meteorological data set, and meteorological data similarity is calculated with the current meteorological data matrix, and the specific implementation process of combining the meteorological data similarity and the historical power data to obtain the first predicted power data includes:

[0016] Get the current meteorological data matrix and the historical meteorological data matrix;

[0017] In the historical meteorological data matrix, a window of size N is slid, and each time a set of sliding data is extracted as a historical meteorological data set, and meteorological data similarity calculation is performed with the current meteorological data matrix to obtain the meteorological data similarity of the historical meteorological data set, and the window slides one time step each time until the historical meteorological data matrix is ​​completely traversed;

[0018] If there is a historical meteorological data set whose meteorological data similarity is greater than the similarity threshold, the meteorological data set that meets the judgment conditions is sorted according to time, the proportion of the meteorological data set that meets the judgment conditions is recorded, and the historical power data corresponding to the first N meteorological data sets are averaged to obtain the first predicted power data; otherwise, the first predicted power data is 0.

[0019] Further, a wind power prediction model based on a hybrid neural network is constructed, the wind power prediction model is trained according to the first predicted power data, the historical meteorological data and the historical power data to obtain a final wind power prediction model; the current meteorological data is input into the final wind power prediction model to obtain second predicted power data; the first predicted power data and the second predicted power data are weighted and summed to obtain a specific implementation process of a wind power prediction value, including:

[0020] Construct a wind power prediction model based on hybrid neural network;

[0021] Acquiring historical meteorological data, historical power data and first predicted power data;

[0022] Perform window sliding and meteorological data similarity calculation on the historical meteorological data matrix to obtain a historical meteorological data set that meets the similarity judgment conditions;

[0023] Determine a model training strategy according to the proportion of the first predicted power data and the historical meteorological data set; wherein the model training strategy includes: a first model training strategy, a second model training strategy and a third model training strategy;

[0024] The first model training strategy includes: if the first predicted power data is 0, inputting the pre-processed historical meteorological data into the wind power prediction model for training to obtain a final wind power prediction model;

[0025] The second model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is higher than a preset threshold, inputting the preprocessed historical meteorological data set into the wind power prediction model for training to obtain the final wind power prediction model;

[0026] The third model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is not higher than a preset threshold, the pre-processed historical meteorological data is input into the wind power prediction model for training to obtain a pre-trained wind power prediction model; then, the pre-processed historical meteorological data set is input into the pre-trained wind power prediction model for training to obtain the final wind power prediction model;

[0027] Inputting current meteorological data into the final wind power prediction model to obtain second predicted power data;

[0028] Comprehensively evaluate the first predicted power data and the second predicted power data to obtain a wind power prediction value.

[0029] Furthermore, the wind power forecast value is corrected using the offshore wind power correction formula, and the calculation formula for the final wind power forecast value is obtained as follows:

[0030]

[0031] Among them, P o is the final wind power prediction value, P r is the predicted value of wind power, R sl is the aging coefficient of wind farm equipment, Δfx is the deviation between the actual wind direction and the reference wind direction, α is the wind speed coefficient, and fs ref is the reference wind speed, fs is the actual wind speed, β is the temperature coefficient, T is the actual temperature, T ref is the reference temperature.

[0032] An offshore wind power prediction system based on a neural network comprises: a system control module, a data acquisition module, a data processing module, a power prediction module, a data correction module and a data output module;

[0033] Wherein, the system control module is used to control the start, pause and stop of the system;

[0034] The data acquisition module is used to obtain current meteorological data, historical power data and historical meteorological data of the wind farm;

[0035] The data processing module is used to pre-process the data acquired by the data acquisition module;

[0036] The power prediction module is used to process and evaluate the data and output a wind power prediction value; wherein the power prediction module includes: a first power prediction unit and a second power prediction unit;

[0037] The data correction module is used to correct the wind power prediction value according to the offshore wind power correction formula;

[0038] The data output module is used to output and display the corrected final wind power forecast value.

[0039] Furthermore, the first power prediction unit calculates the meteorological data similarity between the current meteorological data matrix and the historical meteorological data matrix, and obtains the first predicted power data by combining the meteorological data similarity and the historical power data. The specific implementation process includes:

[0040] Get the current meteorological data matrix and the historical meteorological data matrix;

[0041] In the historical meteorological data matrix, a window of size N is slid, and each time a set of sliding data is extracted as a historical meteorological data set, and meteorological data similarity calculation is performed with the current meteorological data matrix to obtain the meteorological data similarity of the historical meteorological data set, and the window slides one time step each time until the historical meteorological data matrix is ​​completely traversed;

[0042] If there is a historical meteorological data set whose meteorological data similarity is greater than the similarity threshold, the meteorological data set that meets the judgment conditions is sorted according to time, the proportion of the meteorological data set that meets the judgment conditions is recorded, and the historical power data corresponding to the first N meteorological data sets are averaged to obtain the first predicted power data; otherwise, the first predicted power data is 0.

[0043] Furthermore, the calculation formula of the meteorological data similarity is:

[0044]

[0045] Wherein, MDS represents the similarity of the meteorological data; dde() represents the error function of the meteorological data; M P Represented as the current meteorological data matrix; M h It represents the historical meteorological data set; N represents the window size; K represents the number of meteorological data types; Represented as the jth data in the i-th group of data of the current meteorological data matrix; Represented as the jth data in the i-th group of data in the historical meteorological data set.

[0046] Furthermore, the second power prediction unit inputs the current meteorological data into the final wind power prediction model to obtain the second predicted power data, and the specific implementation process includes:

[0047] Construct a wind power prediction model based on hybrid neural network;

[0048] Acquiring historical meteorological data, historical power data and first predicted power data;

[0049] Perform window sliding and meteorological data similarity calculation on the historical meteorological data matrix to obtain a historical meteorological data set that meets the similarity judgment conditions;

[0050] Determine a model training strategy according to the proportion of the first predicted power data and the historical meteorological data set; wherein the model training strategy includes: a first model training strategy, a second model training strategy and a third model training strategy;

[0051] The first model training strategy includes: if the first predicted power data is 0, inputting the pre-processed historical meteorological data into the wind power prediction model for training to obtain a final wind power prediction model;

[0052] The second model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is higher than a preset threshold, inputting the preprocessed historical meteorological data set into the wind power prediction model for training to obtain the final wind power prediction model;

[0053] The third model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is not higher than a preset threshold, the pre-processed historical meteorological data is input into the wind power prediction model for training to obtain a pre-trained wind power prediction model; then, the pre-processed historical meteorological data set is input into the pre-trained wind power prediction model for training to obtain the final wind power prediction model;

[0054] The current meteorological data is input into the final wind power prediction model to obtain second predicted power data.

[0055] Furthermore, the data correction module uses the offshore wind power correction formula to correct the wind power prediction value, and obtains the calculation formula of the final wind power prediction value as follows:

[0056]

[0057] Among them, P o is the final wind power prediction value, P r is the predicted value of wind power, R sl is the aging coefficient of wind farm equipment, Δfx is the deviation between the actual wind direction and the reference wind direction, α is the wind speed coefficient, and fs ref is the reference wind speed, fs is the actual wind speed, β is the temperature coefficient, T is the actual temperature, T ref is the reference temperature.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. The present invention proposes a power prediction function based on a sliding window and data similarity for obtaining the first predicted power data; the function first performs a window sliding on the historical meteorological data matrix, and calculates the meteorological data similarity between each obtained historical meteorological data set and the current meteorological data matrix; then, the similarity threshold is used to determine whether the historical meteorological data similar to the current meteorological data is contained, and the corresponding historical power data is averaged to obtain the first predicted power data; the function uses the similarity between historical data and current data to make an initial prediction of offshore wind power, which can effectively improve the effectiveness and accuracy of offshore wind power prediction.

[0060] 2. The present invention proposes a wind power prediction function based on a hybrid neural network model for obtaining second predicted power data; this function inputs current meteorological data into the hybrid neural network model for processing, and outputs the second predicted power data; wherein, the training process of the hybrid neural network model is determined by a training strategy based on the proportion of the first predicted power data and a historical meteorological data set that meets the meteorological data similarity judgment condition; finally, the first predicted power data and the second predicted power data are comprehensively evaluated to obtain a wind power prediction value; this function combined with similarity mean prediction and model prediction can effectively improve the effectiveness and accuracy of offshore wind power prediction.

[0061] 3. The present invention proposes a wind power correction function for correcting the wind power prediction value according to the actual situation; the function uses the offshore wind power correction formula to modify the wind power prediction value to obtain the final wind power prediction value; wherein, the offshore wind power correction formula is set according to the aging degree of the wind farm equipment, the wind direction deviation value, the temperature deviation value and the wind speed deviation value; the function uses the offshore wind power correction formula to ensure that the final prediction value is more in line with the actual situation, which can effectively improve the effectiveness and accuracy of the offshore wind power prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic flow chart of a method for predicting offshore wind power based on a neural network according to the present invention;

[0063] Figure 2 A schematic diagram of a process for obtaining first predicted power data according to the present invention;

[0064] Figure 3 It is a structural schematic diagram of a wind power prediction model based on a hybrid neural network of the present invention;

[0065] Figure 4 It is a structural schematic diagram of a neural network-based offshore wind power prediction system of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] As an important low-carbon clean energy, wind energy will play an important role in the future energy structure transformation due to its sustainable utilization characteristics. However, the power output of offshore wind power is relatively unstable, which is mainly due to the randomness of wind energy itself. Therefore, accurate prediction of offshore wind power in advance has become an indispensable technical requirement.

[0068] The power generation of offshore wind power is affected by many factors, especially wind speed and wind direction, which are the key factors in determining the power generation. Since the fluctuation of offshore wind power has significant trend and periodic characteristics, meteorological factors often show periodic fluctuations related to various factors such as seasons, atmospheric circulation and geographical location. Through in-depth analysis of these meteorological variables, the accuracy of offshore wind power prediction can be effectively improved, thereby providing a scientific basis for power dispatch and reserve.

[0069] At present, the existing technology still has shortcomings in offshore wind power forecasting; the existing technology does not take into account factors such as sudden changes in wind speed, temperature and wind direction in the offshore wind power forecasting process, resulting in large forecasting errors. In addition, the existing technology ignores the operating status of the wind farm itself, the impact of equipment aging and changes in meteorological conditions, which limits the forecasting accuracy and reduces the effectiveness and accuracy of offshore wind power forecasting.

[0070] Embodiment 1

[0071] The specific implementation process in the embodiment of the present application is implemented by a neural network-based offshore wind power prediction method of the present invention, see Figure 1 The process of the method proposed in the present invention is described in the following; the offshore wind power prediction method based on a neural network comprises:

[0072] S10. Obtain current meteorological data, historical power data and historical meteorological data of the wind farm;

[0073] S20. Calculate the meteorological data similarity between the current meteorological data matrix and the historical meteorological data matrix to obtain the first predicted power data;

[0074] S30. Training a wind power prediction model based on a hybrid neural network according to the first predicted power data, the historical meteorological data and the historical power data, and inputting the current meteorological data into the trained wind power prediction model to obtain second predicted power data;

[0075] S40. Comprehensively evaluate the first predicted power data and the second predicted power data to obtain a wind power prediction value;

[0076] S50. Correct the wind power forecast value using the offshore wind power correction formula to obtain a final wind power forecast value.

[0077] Furthermore, the specific implementation process of a neural network-based offshore wind power prediction method is as follows:

[0078] Obtain current meteorological data, historical power data and historical meteorological data of the wind farm;

[0079] Constructing a current meteorological data matrix and a historical meteorological data matrix respectively according to the current meteorological data and the historical meteorological data; wherein the current meteorological data is a set of continuous meteorological data;

[0080] The historical meteorological data matrix is ​​extracted by using a window sliding method to obtain a historical meteorological data set, and meteorological data similarity is calculated with the current meteorological data matrix, and the meteorological data similarity and the historical power data are combined to obtain the first predicted power data;

[0081] Constructing a wind power prediction model based on a hybrid neural network, training the wind power prediction model according to the first predicted power data, the historical meteorological data and the historical power data to obtain a final wind power prediction model; inputting the current meteorological data into the final wind power prediction model to obtain second predicted power data; performing weighted summation on the first predicted power data and the second predicted power data to obtain a wind power prediction value;

[0082] The wind power forecast value is corrected using an offshore wind power correction formula to obtain a final wind power forecast value.

[0083] In this embodiment, a method for offshore wind power prediction based on a neural network is proposed; first, the current meteorological data, historical power data and historical meteorological data of the wind farm are obtained; then, a historical meteorological data set that meets the threshold judgment condition is obtained by using window sliding and meteorological data similarity calculation, and the historical power data corresponding to the historical meteorological data set is averaged to obtain first predicted power data; then, a training strategy for a wind power prediction model based on a hybrid neural network is determined, and the current meteorological data is input into the wind power prediction model to obtain second predicted power data; the first predicted power data and the second predicted power data are evaluated to obtain a wind power prediction value; finally, the wind power prediction value is corrected using an offshore wind power correction formula to obtain a final wind power prediction value; this method can effectively improve the effectiveness and accuracy of offshore wind power prediction.

[0084] For the purpose of specific explanation, the following embodiments are described as follows:

[0085] Obtain current meteorological data, historical power data and historical meteorological data of the wind farm;

[0086] The historical meteorological data is time series data, which corresponds one to one with the historical power data, and the two are connected through a time relationship;

[0087] Furthermore, the current meteorological data includes: wind speed, wind direction, air pressure, temperature and humidity.

[0088] There are five types of meteorological data used in this embodiment, including: wind speed, wind direction, air pressure, temperature and humidity; of course, the meteorological data that affect offshore wind power are not limited to these, and can be supplemented according to actual conditions; among them, meteorological data other than wind direction are presented in the form of numerical values. In order to facilitate the subsequent numerical comparison of wind direction, the actual wind direction angle is recorded, and the angle deviation is used to express the wind direction difference; the collection of these meteorological data provides a data basis for subsequent power prediction and power correction, further improving the effectiveness and accuracy of offshore wind power prediction.

[0089] Further, a current meteorological data matrix and a historical meteorological data matrix are constructed respectively according to the current meteorological data and the historical meteorological data; wherein the current meteorological data is a set of continuous meteorological data;

[0090] Furthermore, the window sliding method is used to extract the historical meteorological data matrix to obtain a historical meteorological data set, and the meteorological data similarity is calculated with the current meteorological data matrix. The process diagram of obtaining the first predicted power data by combining the meteorological data similarity and the historical power data can be referred to. Figure 2 ,include:

[0091] S110. Obtain the current meteorological data matrix and the historical meteorological data matrix;

[0092] S120. In the historical meteorological data matrix, a window of size N is slid, and a set of sliding data is extracted each time as a historical meteorological data set, and the window slides one time step each time until the historical meteorological data matrix is ​​completely traversed to obtain a historical meteorological data set group;

[0093] S130. Calculate the meteorological data similarity between each of the historical meteorological data sets in the historical meteorological data set group and the current meteorological data matrix to obtain the meteorological data similarity of each of the historical meteorological data sets;

[0094] S140. Perform a threshold judgment on the meteorological data similarity of each historical meteorological data set. If there is a historical meteorological data set whose meteorological data similarity is greater than the similarity threshold, sort the meteorological data sets that meet the judgment conditions by time, and calculate the mean of the historical power data corresponding to the first N meteorological data sets to obtain the first predicted power data; otherwise, set the first predicted power data to 0.

[0095] In this embodiment, the number N of meteorological data sets selected is set to 5; of course, this value can be adjusted according to actual conditions.

[0096] Furthermore, the calculation formula of the meteorological data similarity is:

[0097]

[0098] Wherein, MDS represents the similarity of the meteorological data; dde() represents the error function of the meteorological data; M P Represented as the current meteorological data matrix; M h It represents the historical meteorological data set; N represents the window size; K represents the number of meteorological data types; Represented as the jth data in the i-th group of data of the current meteorological data matrix; Represented as the jth data in the i-th group of data in the historical meteorological data set.

[0099] In this embodiment, a power prediction function based on a sliding window and data similarity is proposed to obtain the first predicted power data; this function first slides the window of the historical meteorological data matrix, and calculates the meteorological data similarity between each obtained historical meteorological data set and the current meteorological data matrix; then, the similarity threshold is used to determine whether there is historical meteorological data similar to the current meteorological data, and the corresponding historical power data is averaged to obtain the first predicted power data; this function uses the similarity between historical data and current data to make an initial prediction of offshore wind power, which can effectively improve the effectiveness and accuracy of offshore wind power prediction.

[0100] Further, a wind power prediction model based on a hybrid neural network is constructed, the wind power prediction model is trained according to the first predicted power data, the historical meteorological data and the historical power data to obtain a final wind power prediction model; the current meteorological data is input into the final wind power prediction model to obtain second predicted power data; the first predicted power data and the second predicted power data are weighted and summed to obtain a specific implementation process of a wind power prediction value, including:

[0101] Construct a wind power prediction model based on hybrid neural network;

[0102] The wind power prediction model based on the hybrid neural network in this embodiment is a hybrid model combining residual convolution and self-attention structure. The structure of the hybrid model is as follows: Figure 3 As shown, it includes: an input layer, a feature extraction layer, 3 residual convolution layers, 3 Transformer layers, a feature fusion layer and an output layer; wherein the input layer is used to convert the current meteorological data from the data space to the feature space for further processing; the feature extraction layer is used to extract deep features to enhance the feature expression capability; the residual convolution layer and the Transformer layer are respectively in different branches, each for obtaining spatial local features and global features; the feature fusion layer is used to fuse the spatial local features and the global features; the output layer is used to output the second predicted power data.

[0103] Further, obtaining historical meteorological data, historical power data and first predicted power data;

[0104] Furthermore, the historical meteorological data matrix is ​​subjected to window sliding and meteorological data similarity calculation to obtain a historical meteorological data set that meets the similarity judgment conditions;

[0105] Further, according to the proportion of the first predicted power data and the historical meteorological data set, a model training strategy is determined; wherein the model training strategy includes: a first model training strategy, a second model training strategy and a third model training strategy;

[0106] Furthermore, the first model training strategy includes: if the first predicted power data is 0, inputting the pre-processed historical meteorological data into the wind power prediction model for training to obtain a final wind power prediction model;

[0107] Furthermore, the second model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is higher than a preset threshold, inputting the preprocessed historical meteorological data set into the wind power prediction model for training to obtain the final wind power prediction model;

[0108] In this embodiment, the preset threshold is set to 0.5. The preset threshold will vary with the amount of historical meteorological data and can be flexibly adjusted according to the actual amount of data.

[0109] Further, the third model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is not higher than a preset threshold, the pre-processed historical meteorological data is input into the wind power prediction model for training to obtain a pre-trained wind power prediction model; then, the pre-processed historical meteorological data set is input into the pre-trained wind power prediction model for training to obtain the final wind power prediction model;

[0110] In this embodiment, the preprocessing process includes data cleaning, deduplication and normalization; data cleaning and data deduplication are used to eliminate missing data, erroneous data and duplicate data; data normalization is used to increase the training speed of the prediction model; the preprocessed data will be divided into a training set and a test set in a ratio of 8:2.

[0111] Further, the current meteorological data is input into the final wind power prediction model to obtain second predicted power data;

[0112] Furthermore, the first predicted power data and the second predicted power data are comprehensively evaluated to obtain a wind power prediction value; the calculation formula of the wind power prediction value is:

[0113] P r =ω*P fi +(1-ω)*P se ;

[0114]

[0115] Among them, P r is represented by the wind power prediction value; ω is represented by the adaptive weight factor; λ is represented by the first prediction weight factor, which is set to 0.5; P fi Represented as the first predicted power data; P se Denoted as the second predicted power data.

[0116] In this embodiment, a wind power prediction function based on a hybrid neural network model is proposed to obtain second predicted power data; this function inputs current meteorological data into the hybrid neural network model for processing, and outputs the second predicted power data; wherein, the training process of the hybrid neural network model is determined by a training strategy based on the proportion of the first predicted power data and a historical meteorological data set that meets the meteorological data similarity judgment condition; finally, the first predicted power data and the second predicted power data are comprehensively evaluated to obtain a wind power prediction value; this function, combined with similarity mean prediction and model prediction, can effectively improve the effectiveness and accuracy of offshore wind power prediction.

[0117] Furthermore, the wind power forecast value is corrected using the offshore wind power correction formula, and the calculation formula for the final wind power forecast value is obtained as follows:

[0118]

[0119] α+β=1.0;

[0120] Among them, P o is the final wind power prediction value, P r is the predicted value of wind power, R sl is the aging coefficient of wind farm equipment, Δfx is the deviation between the actual wind direction and the reference wind direction, α is the wind speed coefficient, and fs ref is the reference wind speed, fs is the actual wind speed, β is the temperature coefficient, T is the actual temperature, T ref is the reference temperature.

[0121] In this embodiment, a wind power correction function is proposed for correcting the wind power prediction value according to the actual situation; this function uses the offshore wind power correction formula to modify the wind power prediction value to obtain the final wind power prediction value; wherein, the offshore wind power correction formula is set according to the aging degree of wind farm equipment, wind direction deviation value, temperature deviation value and wind speed deviation value; this function uses the offshore wind power correction formula to ensure that the final prediction value is more in line with the actual situation, which can effectively improve the effectiveness and accuracy of offshore wind power prediction.

[0122] The present invention collects current meteorological data of wind farm A and historical meteorological data and historical power data of the past year; wherein the current meteorological data are multiple sets of data collected under good meteorological conditions; pre-processed data are obtained by performing data cleaning, deduplication and normalization on the historical meteorological data and the historical power data; the pre-processed data are divided into a training set and a test set in a ratio of 8:2.

[0123] In order to verify the actual effect of the offshore wind power prediction method based on neural network proposed in the present invention, multiple groups of comparative experiments were designed; among them, method one applied the offshore wind power prediction method based on neural network proposed in the present invention, the method included a first power prediction based on sliding window and data similarity and a second power prediction based on a hybrid neural network model; method two only used the second power prediction method based on the hybrid neural network model to predict offshore wind power, omitting the first power prediction based on sliding window and data similarity; method three used convolutional neural network to predict wind power.

[0124] The three methods first use the training set to train their respective prediction models, and then input the current meteorological data into the respectively trained prediction models to obtain the wind power prediction power and calculate the ratio of the wind power prediction power within a reasonable range.

[0125] Table 1. Comparison of the effects of different prediction schemes

[0126]

[0127] As shown in Table 1, the comprehensive method (method one) combining the first power prediction based on sliding window and data similarity and the second power prediction based on hybrid neural network model performed best, indicating that the method proposed in the present invention is the most effective in offshore wind power prediction.

[0128] Embodiment 2

[0129] As an embodiment of the present invention, refer to Figure 4 , an offshore wind power prediction system based on neural network, comprising: a system control module, a data acquisition module, a data processing module, a power prediction module, a data correction module and a data output module;

[0130] Wherein, the system control module is used to control the start, pause and stop of the system;

[0131] The data acquisition module is used to obtain current meteorological data, historical power data and historical meteorological data of the wind farm;

[0132] The data processing module is used to pre-process the data acquired by the data acquisition module;

[0133] The power prediction module is used to process and evaluate the data and output a wind power prediction value; wherein the power prediction module includes: a first power prediction unit and a second power prediction unit;

[0134] Furthermore, the first power prediction unit calculates the meteorological data similarity between the current meteorological data matrix and the historical meteorological data matrix, and obtains the first predicted power data by combining the meteorological data similarity and the historical power data. The specific implementation process includes:

[0135] Get the current meteorological data matrix and the historical meteorological data matrix;

[0136] In the historical meteorological data matrix, a window of size N is slid, and each time a set of sliding data is extracted as a historical meteorological data set, and meteorological data similarity calculation is performed with the current meteorological data matrix to obtain the meteorological data similarity of the historical meteorological data set, and the window slides one time step each time until the historical meteorological data matrix is ​​completely traversed;

[0137] If there is a historical meteorological data set whose meteorological data similarity is greater than the similarity threshold, the meteorological data set that meets the judgment conditions is sorted according to time, the proportion of the meteorological data set that meets the judgment conditions is recorded, and the historical power data corresponding to the first N meteorological data sets are averaged to obtain the first predicted power data; otherwise, the first predicted power data is 0.

[0138] Furthermore, the calculation formula of the meteorological data similarity is:

[0139]

[0140] Wherein, MDS represents the similarity of the meteorological data; dde() represents the error function of the meteorological data; M P Represented as the current meteorological data matrix; M h It represents the historical meteorological data set; N represents the window size; K represents the number of meteorological data types; Represented as the jth data in the i-th group of data of the current meteorological data matrix; Represented as the jth data in the i-th group of data in the historical meteorological data set.

[0141] Further, the second power prediction unit trains a wind power prediction model according to the first predicted power data, the historical meteorological data and the historical power data to obtain a final wind power prediction model; inputs the current meteorological data into the final wind power prediction model to obtain second predicted power data; and performs weighted summation on the first predicted power data and the second predicted power data to obtain a specific implementation process of a wind power prediction value, including:

[0142] Construct a wind power prediction model based on hybrid neural network;

[0143] Acquiring historical meteorological data, historical power data and first predicted power data;

[0144] Perform window sliding and meteorological data similarity calculation on the historical meteorological data matrix to obtain a historical meteorological data set that meets the similarity judgment conditions;

[0145] Determine a model training strategy according to the proportion of the first predicted power data and the historical meteorological data set; wherein the model training strategy includes: a first model training strategy, a second model training strategy and a third model training strategy;

[0146] The first model training strategy includes: if the first predicted power data is 0, inputting the pre-processed historical meteorological data into the wind power prediction model for training to obtain a final wind power prediction model;

[0147] The second model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is higher than a preset threshold, inputting the preprocessed historical meteorological data set into the wind power prediction model for training to obtain the final wind power prediction model;

[0148] The third model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is not higher than a preset threshold, the pre-processed historical meteorological data is input into the wind power prediction model for training to obtain a pre-trained wind power prediction model; then, the pre-processed historical meteorological data set is input into the pre-trained wind power prediction model for training to obtain the final wind power prediction model;

[0149] Inputting current meteorological data into the final wind power prediction model to obtain second predicted power data;

[0150] Comprehensively evaluate the first predicted power data and the second predicted power data to obtain a wind power prediction value; wherein the calculation formula of the wind power prediction value is:

[0151] P r =ω*P fi +(1-ω)*P se ;

[0152]

[0153] Among them, P r is represented by the wind power prediction value; ω is represented by the adaptive weight factor; λ is represented by the first prediction weight factor, which is set to 0.5; P fi Represented as the first predicted power data; P se Denoted as the second predicted power data.

[0154] In order to further verify the accuracy of the power prediction of the present invention under different data sets, multiple groups of comparative experiments were designed; the present invention collects historical meteorological data, historical power data and current meteorological data of wind farm B; wherein, the current meteorological data is collected under good meteorological conditions; by screening the historical meteorological data to meet the condition requirements of three model training strategies, three historical meteorological data sets are obtained, which are recorded as the first data set, the second data set and the third data set; then, the wind power prediction model proposed by the present invention and the traditional convolutional neural network model are respectively trained using the three historical meteorological data sets, and then the current meteorological data are input into the respective trained models to obtain the second predicted power data; wherein, the present invention adopts different training strategies for different data sets, while the traditional convolutional neural network model is directly trained; finally, the second predicted power data obtained by each output are compared.

[0155] Table 2. Comparison of prediction effects of different data sets

[0156]

[0157] As shown in Table 2, compared with the traditional convolutional neural network model, the neural network model proposed in the present invention has the highest accuracy in predicting wind power under different training sets, and even in the first data set containing the most sudden abnormal meteorological data, it still has a high prediction accuracy, indicating that the neural network and training strategy proposed in the present invention can effectively improve the accuracy and reliability of offshore wind power prediction.

[0158] The data correction module is used to correct the wind power prediction value according to the offshore wind power correction formula;

[0159] Furthermore, the data correction module uses the offshore wind power correction formula to correct the wind power prediction value, and obtains the calculation formula of the final wind power prediction value as follows:

[0160]

[0161] Among them, P o is the final wind power prediction value, P r is the predicted value of wind power, R sl is the aging coefficient of wind farm equipment, Δfx is the deviation between the actual wind direction and the reference wind direction, α is the wind speed coefficient, and fs ref is the reference wind speed, fs is the actual wind speed, β is the temperature coefficient, T is the actual temperature, T ref is the reference temperature.

[0162] The data output module is used to output and display the corrected final wind power forecast value.

[0163] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting offshore wind power based on neural network, characterized in that: include: Obtain current meteorological data, historical power data and historical meteorological data of the wind farm; Constructing a current meteorological data matrix and a historical meteorological data matrix respectively according to the current meteorological data and the historical meteorological data; wherein the current meteorological data is a set of continuous meteorological data; The historical meteorological data matrix is ​​extracted by using a window sliding method to obtain a historical meteorological data set, and meteorological data similarity is calculated with the current meteorological data matrix, and the meteorological data similarity and the historical power data are combined to obtain the first predicted power data; Constructing a wind power prediction model based on a hybrid neural network, training the wind power prediction model according to the first predicted power data, the historical meteorological data and the historical power data to obtain a final wind power prediction model; inputting the current meteorological data into the final wind power prediction model to obtain second predicted power data; performing weighted summation on the first predicted power data and the second predicted power data to obtain a wind power prediction value; The wind power forecast value is corrected using an offshore wind power correction formula to obtain a final wind power forecast value.

2. The offshore wind power prediction method based on neural network according to claim 1 is characterized in that: The current meteorological data includes: wind speed, wind direction, air pressure, temperature and humidity.

3. The offshore wind power prediction method based on neural network according to claim 1 is characterized in that: The specific implementation process of extracting the historical meteorological data matrix by using the window sliding method to obtain a historical meteorological data set, calculating the meteorological data similarity with the current meteorological data matrix, and combining the meteorological data similarity with the historical power data to obtain the first predicted power data includes: Get the current meteorological data matrix and the historical meteorological data matrix; In the historical meteorological data matrix, a window of size N is slid, and each time a set of sliding data is extracted as a historical meteorological data set, and meteorological data similarity calculation is performed with the current meteorological data matrix to obtain the meteorological data similarity of the historical meteorological data set, and the window slides one time step each time until the historical meteorological data matrix is ​​completely traversed; If there is a historical meteorological data set whose meteorological data similarity is greater than the similarity threshold, the meteorological data set that meets the judgment conditions is sorted according to time, the proportion of the meteorological data set that meets the judgment conditions is recorded, and the historical power data corresponding to the first N meteorological data sets are averaged to obtain the first predicted power data; otherwise, the first predicted power data is 0.

4. The offshore wind power prediction method based on neural network according to claim 1 is characterized in that: Constructing a wind power prediction model based on a hybrid neural network, training the wind power prediction model according to the first predicted power data, the historical meteorological data and the historical power data to obtain a final wind power prediction model; inputting the current meteorological data into the final wind power prediction model to obtain second predicted power data; The specific implementation process of performing weighted summation on the first predicted power data and the second predicted power data to obtain the wind power prediction value includes: Construct a wind power prediction model based on hybrid neural network; Acquiring historical meteorological data, historical power data and first predicted power data; Perform window sliding and meteorological data similarity calculation on the historical meteorological data matrix to obtain a historical meteorological data set that meets the similarity judgment conditions; Determine a model training strategy according to the proportion of the first predicted power data and the historical meteorological data set; wherein the model training strategy includes: a first model training strategy, a second model training strategy and a third model training strategy; The first model training strategy includes: if the first predicted power data is 0, inputting the pre-processed historical meteorological data into the wind power prediction model for training to obtain a final wind power prediction model; The second model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is higher than a preset threshold, inputting the preprocessed historical meteorological data set into the wind power prediction model for training to obtain the final wind power prediction model; The third model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is not higher than a preset threshold, the pre-processed historical meteorological data is input into the wind power prediction model for training to obtain a pre-trained wind power prediction model; then, the pre-processed historical meteorological data set is input into the pre-trained wind power prediction model for training to obtain the final wind power prediction model; Inputting current meteorological data into the final wind power prediction model to obtain second predicted power data; Comprehensively evaluate the first predicted power data and the second predicted power data to obtain a wind power prediction value.

5. The offshore wind power prediction method based on neural network according to claim 1 is characterized in that: The wind power forecast value is corrected using the offshore wind power correction formula to obtain the final wind power forecast value: Among them, P o is the final wind power prediction value, P r is the predicted value of wind power, R sl is the aging coefficient of wind farm equipment, Δfx is the deviation between the actual wind direction and the reference wind direction, α is the wind speed coefficient, and fs ref is the reference wind speed, fs is the actual wind speed, β is the temperature coefficient, T is the actual temperature, T ref is the reference temperature.

6. An offshore wind power prediction system based on neural network, characterized in that: include: System control module, data acquisition module, data processing module, power prediction module, data correction module and data output module; Wherein, the system control module is used to control the start, pause and stop of the system; The data acquisition module is used to obtain current meteorological data, historical power data and historical meteorological data of the wind farm; The data processing module is used to pre-process the data acquired by the data acquisition module; The power prediction module is used to process and evaluate the data and output a wind power prediction value; wherein the power prediction module includes: a first power prediction unit and a second power prediction unit; The data correction module is used to correct the wind power prediction value according to the offshore wind power correction formula; The data output module is used to output and display the corrected final wind power forecast value.

7. The offshore wind power prediction system based on neural network according to claim 6, characterized in that: The first power prediction unit calculates the meteorological data similarity between the current meteorological data matrix and the historical meteorological data matrix, and obtains the first predicted power data by combining the meteorological data similarity and the historical power data. The specific implementation process includes: Get the current meteorological data matrix and the historical meteorological data matrix; In the historical meteorological data matrix, a window of size N is slid, and each time a set of sliding data is extracted as a historical meteorological data set, and meteorological data similarity calculation is performed with the current meteorological data matrix to obtain the meteorological data similarity of the historical meteorological data set, and the window slides one time step each time until the historical meteorological data matrix is ​​completely traversed; If there is a historical meteorological data set whose meteorological data similarity is greater than the similarity threshold, the meteorological data set that meets the judgment conditions is sorted according to time, the proportion of the meteorological data set that meets the judgment conditions is recorded, and the historical power data corresponding to the first N meteorological data sets are averaged to obtain the first predicted power data; otherwise, the first predicted power data is 0.

8. The offshore wind power prediction system based on neural network according to claim 7, characterized in that: The calculation formula of the meteorological data similarity is: Wherein, MDS represents the similarity of the meteorological data; dde() represents the error function of the meteorological data; M P Represented as the current meteorological data matrix; M h It is represented as the historical meteorological data set; N is represented as the window size; K is represented as the number of types of meteorological data; Represented as the jth data in the i-th group of data of the current meteorological data matrix; Represented as the j-th data in the i-th group of data in the historical meteorological data set.

9. The offshore wind power prediction system based on neural network according to claim 6, characterized in that: The specific implementation process of the second power prediction unit inputting the current meteorological data into the final wind power prediction model to obtain the second predicted power data includes: Construct a wind power prediction model based on hybrid neural network; Acquiring historical meteorological data, historical power data and first predicted power data; Perform window sliding and meteorological data similarity calculation on the historical meteorological data matrix to obtain a historical meteorological data set that meets the similarity judgment conditions; Determine a model training strategy according to the proportion of the first predicted power data and the historical meteorological data set; wherein the model training strategy includes: a first model training strategy, a second model training strategy and a third model training strategy; The first model training strategy includes: if the first predicted power data is 0, inputting the pre-processed historical meteorological data into the wind power prediction model for training to obtain a final wind power prediction model; The second model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is higher than a preset threshold, inputting the preprocessed historical meteorological data set into the wind power prediction model for training to obtain the final wind power prediction model; The third model training strategy includes: if the first predicted power data is not 0 and the proportion of the historical meteorological data set is not higher than a preset threshold, the pre-processed historical meteorological data is input into the wind power prediction model for training to obtain a pre-trained wind power prediction model; then, the pre-processed historical meteorological data set is input into the pre-trained wind power prediction model for training to obtain the final wind power prediction model; The current meteorological data is input into the final wind power prediction model to obtain second predicted power data.

10. The offshore wind power prediction system based on neural network according to claim 6, characterized in that: The data correction module uses the offshore wind power correction formula to correct the wind power prediction value, and the calculation formula for obtaining the final wind power prediction value is: Among them, P o is the final wind power prediction value, P r is the predicted value of wind power, R sl is the aging coefficient of wind farm equipment, Δfx is the deviation between the actual wind direction and the reference wind direction, α is the wind speed coefficient, and fs ref is the reference wind speed, fs is the actual wind speed, β is the temperature coefficient, T is the actual temperature, T ref is the reference temperature.

Images