Wind power prediction method based on feature reconstruction

Abstract

The invention provides a wind power prediction method based on feature reconstruction. Original meteorological data and original unit data are subjected to expansion processing to generate first input characteristics with time delay, nonlinearity or average characteristics, so that meteorological information and unit operation information can be indicated from multiple aspects, after the first input characteristics are obtained, dimensionality reduction and recombination are carried out on the first input characteristics, dimensionality with large information amount is reserved, and the system performance is improved. In addition, optimization can be carried out in real time in the feature reconstruction process to obtain the optimal reconstruction parameter, the feature reconstruction process is executed according to the optimal reconstruction parameter to obtain the second input feature with the optimal performance, and finally the second input feature is input into the target model to execute the corresponding prediction task. According to the method, the comprehensiveness of the input information is improved through feature expansion, and the data dimension is reduced and redundant and noise information is removed through feature reconstruction, so that the data volume is reduced, the data quality is improved, and the effectiveness and efficiency of wind power prediction can be improved.

Description

Technical Field

[0001] The present application relates to the field of renewable energy power generation, and in particular to a wind power prediction method based on feature reconstruction. Background Art

[0002] In wind power forecasting, the nonlinearity and non-stationarity of data are the main challenges. Wind power is affected by many complex factors, such as wind speed, wind direction, temperature, humidity, etc. Traditional forecasting methods are difficult to fully capture complex multidimensional factors, which will cause large forecasting errors.

[0003] As a data analysis and processing technology, feature reconstruction can mine and expand the potential information in the original data, screen out the core features that have a greater impact on wind power, better reflect the fluctuation law of wind power, reduce the interference of noise and redundant information, and thus improve the generalization ability and prediction accuracy of the prediction model. However, the current prediction method based on wind power features has limitations in feature selection and parameter optimization, and it is difficult to achieve automatic and efficient feature reconstruction under complex practical application scenarios and data input, resulting in poor prediction results. Therefore, a wind power prediction method based on feature reconstruction is needed to improve the effectiveness of wind power feature extraction, thereby improving the effectiveness of wind power prediction. Summary of the invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical deficiencies, especially the technical defect of insufficient effectiveness of wind power feature extraction in the prior art.

[0005] In a first aspect, the present application provides a wind power prediction method based on feature reconstruction, the method comprising:

[0006] Obtain the original meteorological data and original unit data of the target wind farm to establish a data set;

[0007] Generate a first input feature according to the original meteorological data and the original unit data;

[0008] Reconstructing the first input feature by using a factor analysis algorithm to obtain a second input feature;

[0009] Wherein, the dimension of the second input feature is smaller than the dimension of the first input feature;

[0010] The second input feature is input into the target model to obtain a wind power prediction result.

[0011] As an optional implementation manner, generating the first input feature according to the original meteorological data and the original unit data includes:

[0012] One or more of a time series lag feature, a nonlinear feature and a moving average feature are generated according to the original meteorological data and the original unit data.

[0013] As an optional implementation manner, the original meteorological data includes wind speed data and wind direction data, and the original unit data includes power data;

[0014] And, the generation process of the timing hysteresis feature includes:

[0015] Determine a lag step length according to the time delay characteristics of the target wind farm, and process the original meteorological data and the original unit data according to the lag step length to generate a preset number of time series lag features;

[0016] The generation process of the nonlinear feature includes:

[0017] Generate one or more of wind speed cube feature, wind speed square feature, wind direction sine feature, wind direction cosine feature, and wind speed and wind direction combined feature according to the wind speed data and wind direction data;

[0018] The generation process of the moving average feature includes:

[0019] Determine the sliding window length;

[0020] A moving average feature of the power data is generated within the sliding window length.

[0021] As an optional implementation manner, reconstructing the first input feature by using a factor analysis algorithm to obtain the second input feature includes:

[0022] Determine a sample covariance matrix of each dimension in the first input feature;

[0023] Establishing a factor analysis model, determining a factor loading matrix by principal component estimation method, and extracting a preset number of common factors from the first input features;

[0024] Performing a rotation operation on the common factors, calculating a factor axis rotation matrix that maximizes the variance of the rotated factors, and generating a preset number of reconstruction factors;

[0025] The first input feature is processed through the sample covariance matrix, the factor analysis model, the factor loading matrix, and the factor axis rotation matrix to obtain the second input feature.

[0026] As an optional implementation, the method further includes:

[0027] By means of optimal search, the number of preset common factors in the factor analysis model is optimized to obtain a second input feature whose target performance meets preset conditions.

[0028] As an optional implementation manner, the optimal search includes:

[0029] Setting a search step size and a learning rate that are negatively correlated with the number of iterations and / or the rate of improvement of the target performance;

[0030] Iteratively obtaining the second input feature according to the search step size and the learning rate until the target performance corresponding to the second input feature meets a preset condition;

[0031] And, in the optimal search process, the method further includes:

[0032] If the change rate of the target performance is less than a preset threshold, the early stopping mechanism is triggered to stop the optimal search, and during the search process, the feature whose target performance exceeds the performance threshold is selected as the second input feature.

[0033] As an optional implementation, the target model includes a LightGBM model;

[0034] The step of inputting the second input feature into the target model to obtain a wind power prediction result includes:

[0035] Dividing the data in the data set into training data and test data;

[0036] Inputting the second input feature corresponding to the training data into the LightGBM model, and training the LightGBM model according to the histogram approximation algorithm and the leaf node according to the depth growth strategy;

[0037] The second input feature corresponding to the test data is input into the trained LightGBM model to obtain a wind power prediction result.

[0038] In a second aspect, the present application provides a wind power prediction device based on feature reconstruction, the device comprising:

[0039] An acquisition module is used to acquire the original meteorological data and original unit data of the target wind farm to establish a data set;

[0040] A processing module, configured to generate a first input feature according to the original meteorological data and the original unit data;

[0041] The processing module is further used to perform feature reconstruction on the first input feature through a factor analysis algorithm to obtain a second input feature;

[0042] Wherein, the dimension of the second input feature is smaller than the dimension of the first input feature;

[0043] The processing module is further used to input the second input feature into the target model to obtain a wind power prediction result.

[0044] In a third aspect, the present application provides a computer device comprising one or more processors and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the method described in the first aspect are performed.

[0045] In a fourth aspect, the present application provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, causes the one or more processors to perform the steps of the method described in the first aspect.

[0046] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0047] Based on any of the above embodiments, the present application generates a first input feature with time delay, nonlinear or average characteristics by expanding and processing the original meteorological data and the original unit data, so as to indicate meteorological information and unit operation information from multiple aspects, provide a data basis for subsequent predictions, and improve the effectiveness of the predictions. After obtaining the first input feature, the first input feature is reduced in dimension and reorganized by means of feature reconstruction, retaining the dimension with a large amount of information. It can also be optimized in real time during the feature reconstruction process to obtain the optimal reconstruction parameters, and execute the feature reconstruction process according to the optimal reconstruction parameters to obtain the second input feature with the best performance. Finally, the second input feature is input into the target model to perform the corresponding prediction task. In summary, the present application improves the comprehensiveness of the input information through feature expansion, reduces the data dimension and removes redundant and noise information through feature reconstruction, thereby reducing the data volume and improving the data quality, thereby improving the effectiveness and efficiency of wind power prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0049] Figure 1 A schematic diagram of a flow chart of a wind power prediction method based on feature reconstruction provided in one embodiment of the present application;

[0050] Figure 2A schematic diagram of the effect of a wind power prediction method based on feature reconstruction provided by an embodiment of the present application;

[0051] Figure 3 A schematic diagram of the effect of a wind power prediction method based on feature reconstruction provided by an embodiment of the present application;

[0052] Figure 4 This is a diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0054] With the proposal of the dual carbon goals and the increasing demand for renewable energy, wind energy has gradually become one of the most promising forms of energy in the world due to its advantages of cleanliness, low cost and abundant resources. However, the intermittent and unstable nature of wind power generation has brought great challenges to the operation and dispatch of power grids. Therefore, accurate wind power forecasting is of great significance to improving wind power absorption capacity, optimizing power grid dispatching and ensuring the safe and stable operation of power systems.

[0055] In wind power forecasting, the nonlinearity and non-stationarity of data are the main challenges. Wind power is affected by many complex factors, such as wind speed, wind direction, temperature, humidity, etc. Traditional forecasting methods are difficult to fully capture complex multidimensional factors, which will cause large forecasting errors. Feature reconstruction, as a data analysis and processing technology, can mine and expand the potential information in the original data, screen out the core features that have a greater impact on wind power, better reflect the fluctuation law of wind power, reduce the interference of noise and redundant information, and thus improve the generalization ability and prediction accuracy of the forecasting model. However, the existing technology still has limitations in feature selection and parameter optimization, and it is difficult to achieve automatic and efficient feature reconstruction under complex practical application scenarios and data inputs, which affects the efficiency and accuracy of the forecasting model. Therefore, developing a forecasting method that realizes adaptive feature reconstruction has important application value for improving the accuracy of wind power forecasting.

[0056] The technical concept of the present application is to generate a first input feature with time delay, nonlinear or average characteristics by expanding and processing the original meteorological data and the original unit data, so as to indicate meteorological information and unit operation information from multiple aspects, provide a data basis for subsequent predictions, and improve the effectiveness of predictions. After obtaining the first input feature, the first input feature is reduced in dimension and reorganized by means of feature reconstruction, retaining the dimension with a large amount of information. It can also be optimized in real time during the feature reconstruction process to obtain the optimal reconstruction parameters, and the feature reconstruction process is executed according to the optimal reconstruction parameters to obtain the second input feature with the best performance. Finally, the second input feature is input into the target model to perform the corresponding prediction task. In summary, the present application improves the comprehensiveness of the input information through feature expansion, reduces the data dimension and removes redundant and noise information through feature reconstruction, thereby reducing the amount of data and improving the data quality, thereby improving the effectiveness and efficiency of wind power prediction.

[0057] The technical solution provided by the embodiment of the present application may include the following beneficial effects: 1) The feature expansion model established in the present application generates multi-dimensional input features through time lag and nonlinear mapping, which can enrich the input information of the model and help the model capture more key factors affecting wind power changes; 2) The feature reconstruction model established in the present application uses factor analysis technology to extract hidden variables behind observable variables, which can reduce data dimensions, remove redundant and noise information, and help improve the efficiency and accuracy of the model when processing complex data; 3) The automatic parameter optimization model established in the present application automatically selects the optimal parameters through variable step search and early stopping technology, which can improve the model training efficiency and ensure that the prediction model can quickly converge to the optimal state, thereby improving the accuracy and stability of wind power prediction; 4) The wind power prediction model based on LightGBM established in the present application can reduce the computational complexity and improve the model training efficiency through the histogram approximation algorithm and the leaf node depth growth strategy, thereby ensuring the prediction accuracy and efficiency under the wind power prediction task with large data sets and high-dimensional features. The present invention can overcome the problem that other existing wind power prediction models fail to effectively reconstruct the input and automatically extract the optimal features.

[0058] See also Figure 1 , Figure 1 A schematic diagram of a flow chart of a wind power prediction method based on feature reconstruction provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0059] S101, obtaining original meteorological data and original unit data of the target wind farm to establish a data set;

[0060] The data set contains multi-dimensional features associated with the original meteorological data and the original unit data. The specific form of the features can be determined by the feature expansion and feature reconstruction methods in the corresponding implementation methods.

[0061] S102, generating a first input feature according to the original meteorological data and the original unit data;

[0062] The first input feature is expanded based on the original meteorological data and the original unit data so that the feature can more fully reflect the relationship between meteorological conditions and wind power conditions. Specifically, it may include time lag features, nonlinear features, and moving average features, etc. Please refer to the description of the corresponding implementation method.

[0063] S103, reconstructing the first input feature by using a factor analysis algorithm to obtain a second input feature;

[0064] Wherein, the dimension of the second input feature is smaller than the dimension of the first input feature;

[0065] Based on the factor analysis algorithm, the hidden variables behind the observable variables can be calculated, and feature reconstruction can be performed on the first input feature, thereby obtaining a second input feature with reduced dimension and higher information validity. In the process of obtaining the second input feature, parameter optimization can be performed in real time. The present application also provides a parameter optimization method for variable step size search to achieve optimization of the second input feature.

[0066] S104: Input the second input feature into the target model to obtain a wind power prediction result.

[0067] The target model can adopt any prediction model. In this application, the LightGBM model will be used as the target model for explanation based on actual application scenarios.

[0068] As an optional implementation manner, generating the first input feature according to the original meteorological data and the original unit data includes:

[0069] One or more of a time series lag feature, a nonlinear feature and a moving average feature are generated according to the original meteorological data and the original unit data.

[0070] Specifically, in actual application scenarios, the original data can be recorded as: express Wind speed at the moment, including wind speed at 10m above the ground, wind speed at 50m above the ground, wind speed at 90m above the ground, etc.; express Wind direction at the moment, including wind direction at 10m above the ground, wind direction at 50m above the ground, wind direction at 90m above the ground, etc.; express The temperature of the moment; express Humidity at all times; express The wind power at the time instant. More input features can also be included in practical applications.

[0071] This embodiment performs feature expansion processing on the original meteorological data and the original unit data so that the input features used for wind power prediction can more comprehensively reflect the unit conditions, improve the effectiveness of the features, and thus improve the effectiveness and efficiency of power prediction.

[0072] As an optional implementation manner, the original meteorological data includes wind speed data and wind direction data, and the original unit data includes power data;

[0073] And, the generation process of the timing hysteresis feature includes:

[0074] Determine a lag step length according to the time delay characteristics of the target wind farm, and process the original meteorological data and the original unit data according to the lag step length to generate a preset number of time series lag features;

[0075] The generation process of the nonlinear feature includes:

[0076] Generate one or more of wind speed cube feature, wind speed square feature, wind direction sine feature, wind direction cosine feature, and wind speed and wind direction combined feature according to the wind speed data and wind direction data;

[0077] The generation process of the moving average feature includes:

[0078] Determine the sliding window length;

[0079] A moving average feature of the power data is generated within the sliding window length.

[0080] This implementation can be combined with the previous implementation, and the details can be referred to as follows:

[0081] Generate hysteresis features: For the hysteresis step , the lagged value of the feature is expressed as , , etc. Lag step length It can be set according to the specific time delay characteristics of the wind farm, e.g. Represents a lag of one time step (15min or 1h).

[0082] Generate nonlinear features: Generate cubic features of wind speed , square characteristics of wind speed , the sinusoidal characteristics of wind direction , the cosine characteristic of wind direction And the wind speed and wind direction combination characteristics etc., are used to capture the nonlinear relationship between power and wind speed.

[0083] Generate moving average feature: Generate the moving average feature of power, which is specifically implemented by the following formula.

[0084]

[0085] in, Indicates the length of the sliding window. The specific reference value can be set to 3 or 5 to smooth power data and reduce the impact of short-term fluctuations.

[0086] This implementation method can enrich the input information of the model by mathematically processing the wind speed data, wind direction data and power data, and generating multi-dimensional input features according to time lag, nonlinear mapping and sliding average, which is conducive to the model capturing more key factors affecting the change of wind power, improving the effectiveness of the input features and the universality of the model, thereby improving the effectiveness and efficiency of power prediction.

[0087] As an optional implementation manner, reconstructing the first input feature by using a factor analysis algorithm to obtain the second input feature includes:

[0088] Determine a sample covariance matrix of each dimension in the first input feature;

[0089] Establishing a factor analysis model, determining a factor loading matrix by principal component estimation method, and extracting a preset number of common factors from the first input features;

[0090] Performing a rotation operation on the common factors, calculating a factor axis rotation matrix that maximizes the variance of the rotated factors, and generating a preset number of reconstruction factors;

[0091] The first input feature is processed through the sample covariance matrix, the factor analysis model, the factor loading matrix, and the factor axis rotation matrix to obtain the second input feature.

[0092] In a specific application scenario, suppose that after feature expansion and standardization, the first input feature has Features , the factor analysis model can be established as follows:

[0093]

[0094] Where: Appearing in the expression of each variable, it is called a common factor and is an unobservable latent variable; Only with features Related, called special factors, are The part explained by the common factors is also an unobservable latent variable; the coefficient is called factor loading, It is called the load matrix. The above formula can be expressed in the following matrix form:

[0095]

[0096] Where: is the feature matrix, is the common factor matrix, is a special factor matrix, is the load matrix. And it satisfies:

[0097]

[0098]

[0099] The factor loading matrix was estimated using the principal component estimation method. and the special variance matrix :

[0100]

[0101]

[0102]

[0103] Where: is the sample covariance matrix Before The largest eigenvalue; is the corresponding orthogonal unit eigenvector; For the Special factor variance estimates; is the sample covariance matrix No. diagonal elements; The load matrix Line The estimated value of the column.

[0104] Calculate the common factor estimation vector corresponding to the sample and obtain the preset number of common factors:

[0105]

[0106] A factor axis rotation operation is performed on the preset number of common factors, a factor axis rotation matrix that maximizes the variance of the rotated factors is calculated, and a preset number of second input features are generated.

[0107] As can be seen from the above analysis, in order to obtain the second input feature, the actual problem is to calculate the sample covariance matrix, and then estimate the factor loading matrix, so as to calculate a preset number of common factors, and finally calculate a preset number of second input features based on the varimax rotation method.

[0108] This implementation method extracts latent variables behind observable variables through a factor analysis model, which can reduce data dimensions, remove redundant and noise information, and extract the main factors that have the greatest impact on the power prediction task. Then, the physical interpretability of the main factors is improved through the factor axis rotation operation, thereby realizing the reconstruction of the first input feature and obtaining the second input feature, thereby improving the efficiency and accuracy of data processing and ultimately improving the effect of wind power prediction.

[0109] As an optional implementation, the method further includes:

[0110] By means of optimal search, the number of preset common factors in the factor analysis model is optimized to obtain a second input feature whose target performance meets preset conditions.

[0111] During the feature reconstruction process of the present application, the optimal search can also be performed in real time to optimize the number of common factors, thereby optimizing the performance of the second input feature from the calculation source, and then determining the second input feature that meets the model performance standards, improving the model training efficiency and ensuring that the prediction model can quickly converge to the optimal state, thereby improving the accuracy and stability of wind power prediction.

[0112] As an optional implementation manner, the optimal search includes:

[0113] Setting a search step size and a learning rate that are negatively correlated with the number of iterations and / or the rate of improvement of the target performance;

[0114] Iteratively obtaining the second input feature according to the search step size and the learning rate until the target performance corresponding to the second input feature meets a preset condition;

[0115] And, in the optimal search process, the method further includes:

[0116] If the change rate of the target performance is less than a preset threshold, the early stopping mechanism is triggered to stop the optimal search, and during the search process, the feature whose target performance exceeds the performance threshold is selected as the second input feature.

[0117] In actual application scenarios, the second input feature may be the feature obtained in the last iteration, or a set of features with the best target performance in the search process may be set as the second input feature.

[0118] Specifically, the variable step length search strategy is expressed as:

[0119]

[0120] in, represents the number of iterations, Indicates The learning rate at the iteration, represents the initial learning rate.

[0121] Use early stopping technology to control the parameter optimization process. Set the model performance evaluation metric, such as root mean square error or prediction accuracy; set the maximum tolerance for consecutive non-improvements; stop training when the monitoring indicator does not improve significantly within several consecutive iterations (root mean square error does not decrease or prediction accuracy does not increase).

[0122] The algorithm flow is shown in Table 1, which describes the execution process of the algorithm in the form of pseudocode combined with annotations, where the initial reconstruction parameters and the optimal reconstruction parameters both refer to the data used to process the first input feature to calculate the second input feature.

[0123] Table 1: Determining the optimal reconstruction parameters

[0124]

[0125] This implementation method sets a variable search step size and learning rate. In the early stage of the search when the performance is significantly improved, a larger step size and a higher learning rate are set to quickly approach the optimal solution region. In the later stage of the search when convergence is gradually approached, a smaller step size and a lower learning rate are set to achieve fine adjustment of the parameters. During the optimal search process, changes in key performance indicators of the model are continuously monitored. When it is detected that the model performance has not been significantly improved within several consecutive steps, the early stopping mechanism is triggered to automatically stop the search, thereby avoiding continuing invalid searches when the performance has not been significantly improved, thereby optimizing training efficiency and reducing the waste of storage resources.

[0126] As an optional implementation, the target model includes a LightGBM model;

[0127] The step of inputting the second input feature into the target model to obtain a wind power prediction result includes:

[0128] Dividing the data in the data set into training data and test data;

[0129] Inputting the second input feature corresponding to the training data into the LightGBM model, and training the LightGBM model according to the histogram approximation algorithm and the leaf node according to the depth growth strategy;

[0130] The second input feature corresponding to the test data is input into the trained LightGBM model to obtain a wind power prediction result.

[0131] This implementation method divides the data set and inputs the optimized second input feature of the corresponding data in the data set into the wind power prediction model based on LightGBM. During the prediction process, the histogram approximation algorithm and the leaf node depth growth strategy can be used to reduce the computational complexity and improve the model training efficiency, thereby ensuring the prediction accuracy and efficiency in wind power prediction tasks with large data sets and high-dimensional features.

[0132] like Figure 2 As shown, Figure 2 A schematic diagram of the effect of a wind power prediction method based on feature reconstruction provided in an embodiment of the present application is used to illustrate the prediction performance of the original features, the first input features and the second input features under the same data set, which is more intuitive in combination with the data in Table 2.

[0133] Table 2 Prediction performance using different input features

[0134]

[0135] like Figure 3 As shown, Figure 3 A schematic diagram of the effect of a wind power prediction method based on feature reconstruction provided in one embodiment of the present application is used to demonstrate the optimization process of the second input feature.

[0136] Depend on Figure 2 It can be seen that the wind power prediction result obtained by the method provided by this application, that is, using the second input feature prediction, is closer to the actual value than the prediction using the original feature prediction and the prediction using the first input feature prediction. It can be seen from Table 2 that the prediction result after feature expansion is more accurate than the prediction using the original feature; after further adaptive optimal reconstruction of the expanded feature to obtain the second input feature, the prediction performance of the model is further improved. Figure 3 It can be seen that through the method provided in this application, when searching for the optimal reconstruction parameters, if the performance index has not been significantly improved within a period of time, the search is stopped in advance, and the optimal performance obtained is close to the global optimal performance, and the feature dimension after reconstruction is smaller, and the model efficiency is higher.

[0137] The embodiment of the present application further provides a wind power prediction device based on feature reconstruction, characterized in that the device comprises:

[0138] An acquisition module is used to acquire the original meteorological data and original unit data of the target wind farm to establish a data set;

[0139] A processing module, configured to generate a first input feature according to the original meteorological data and the original unit data;

[0140] The processing module is further used to perform feature reconstruction on the first input feature through a factor analysis algorithm to obtain a second input feature;

[0141] Wherein, the dimension of the second input feature is smaller than the dimension of the first input feature;

[0142] The processing module is further used to input the second input feature into the target model to obtain a wind power prediction result.

[0143] By expanding and processing the original meteorological data and the original unit data, a first input feature with time delay, nonlinearity or average characteristics is generated, which can indicate meteorological information and unit operation information from multiple aspects, provide a data basis for subsequent predictions, and improve the effectiveness of predictions. After obtaining the first input feature, the first input feature is reduced in dimension and reorganized by means of feature reconstruction, retaining the dimension with a large amount of information. It can also be optimized in real time during the feature reconstruction process to obtain the optimal reconstruction parameters. The feature reconstruction process is executed according to the optimal reconstruction parameters to obtain the second input feature with the best performance. Finally, the second input feature is input into the target model to perform the corresponding prediction task. In summary, the present application improves the comprehensiveness of the input information through feature expansion, reduces the data dimension and removes redundant and noise information through feature reconstruction, thereby reducing the amount of data and improving the data quality, thereby improving the effectiveness and efficiency of wind power prediction.

[0144] As an optional implementation, the specific manner in which the processing module generates the first input feature according to the original meteorological data and the original unit data includes:

[0145] One or more of a time series lag feature, a nonlinear feature and a moving average feature are generated according to the original meteorological data and the original unit data.

[0146] This embodiment performs feature expansion processing on the original meteorological data and the original unit data so that the input features used for wind power prediction can more comprehensively reflect the unit conditions, improve the effectiveness of the features, and thus improve the effectiveness and efficiency of power prediction.

[0147] As an optional implementation, the original meteorological data includes wind speed data and wind direction data, and the original unit data includes power data;

[0148] And, the specific manner in which the processing module executes the generation process of the timing lag feature includes:

[0149] Determine a lag step length according to the time delay characteristics of the target wind farm, and process the original meteorological data and the original unit data according to the lag step length to generate a preset number of time series lag features;

[0150] The specific manner in which the processing module executes the generation process of the nonlinear feature includes:

[0151] Generate one or more of wind speed cube feature, wind speed square feature, wind direction sine feature, wind direction cosine feature, and wind speed and wind direction combined feature according to the wind speed data and wind direction data;

[0152] The specific manner in which the processing module performs the generation process of the moving average feature includes:

[0153] Determine the sliding window length;

[0154] A moving average feature of the power data is generated within the sliding window length.

[0155] This implementation method can enrich the input information of the model by mathematically processing the wind speed data, wind direction data and power data, and generating multi-dimensional input features according to time lag, nonlinear mapping and sliding average, which is conducive to the model capturing more key factors affecting the change of wind power, improving the effectiveness of the input features and the universality of the model, thereby improving the effectiveness and efficiency of power prediction.

[0156] As an optional implementation manner, the processing module performs feature reconstruction on the first input feature through a factor analysis algorithm to obtain a specific manner of the second input feature, including:

[0157] Determine a sample covariance matrix of each dimension in the first input feature;

[0158] Establishing a factor analysis model, determining a factor loading matrix by principal component estimation method, and extracting a preset number of common factors from the first input features;

[0159] Performing a rotation operation on the common factors, calculating a factor axis rotation matrix that maximizes the variance of the rotated factors, and generating a preset number of reconstruction factors;

[0160] The first input feature is processed through the sample covariance matrix, the factor analysis model, the factor loading matrix, and the factor axis rotation matrix to obtain the second input feature.

[0161] This implementation method extracts latent variables behind observable variables through a factor analysis model, which can reduce data dimensions, remove redundant and noise information, and extract the main factors that have the greatest impact on the power prediction task. Then, the physical interpretability of the main factors is improved through the factor axis rotation operation, thereby realizing the reconstruction of the first input feature and obtaining the second input feature, thereby improving the efficiency and accuracy of data processing and ultimately improving the effect of wind power prediction.

[0162] As an optional implementation, the processing module is further configured to:

[0163] By means of optimal search, the number of preset common factors in the factor analysis model is optimized to obtain a second input feature whose target performance meets preset conditions.

[0164] During the feature reconstruction process of the present application, the optimal search can also be performed in real time to optimize the number of common factors, thereby optimizing the performance of the second input feature from the calculation source, and then determining the second input feature that meets the model performance standards, improving the model training efficiency and ensuring that the prediction model can quickly converge to the optimal state, thereby improving the accuracy and stability of wind power prediction.

[0165] As an optional implementation manner, the specific manner in which the processing module performs the optimal search includes:

[0166] Setting a search step size and a learning rate that are negatively correlated with the number of iterations and / or the rate of improvement of the target performance;

[0167] Iteratively obtaining the second input feature according to the search step size and the learning rate until the target performance corresponding to the second input feature meets a preset condition;

[0168] And, in the optimal search process, the processing module is also used for:

[0169] If the change rate of the target performance is less than a preset threshold, the early stopping mechanism is triggered to stop the optimal search, and during the search process, the feature whose target performance exceeds the performance threshold is selected as the second input feature.

[0170] This implementation method sets a variable search step size and learning rate. In the early stage of the search when the performance is significantly improved, a larger step size and a higher learning rate are set to quickly approach the optimal solution region. In the later stage of the search when convergence is gradually approached, a smaller step size and a lower learning rate are set to achieve fine adjustment of the parameters. During the optimal search process, changes in key performance indicators of the model are continuously monitored. When it is detected that the model performance has not been significantly improved within several consecutive steps, the early stopping mechanism is triggered to automatically stop the search, thereby avoiding continuing invalid searches when the performance has not been significantly improved, thereby optimizing training efficiency and reducing the waste of storage resources.

[0171] As an optional implementation, the target model includes a LightGBM model;

[0172] The specific manner in which the processing module inputs the second input feature into the target model to obtain the wind power prediction result includes:

[0173] Dividing the data in the data set into training data and test data;

[0174] Inputting the second input feature corresponding to the training data into the LightGBM model, and training the LightGBM model according to the histogram approximation algorithm and the leaf node according to the depth growth strategy;

[0175] The second input feature corresponding to the test data is input into the trained LightGBM model to obtain a wind power prediction result.

[0176] This implementation method divides the data set and inputs the optimized second input feature of the corresponding data in the data set into the wind power prediction model based on LightGBM. During the prediction process, the histogram approximation algorithm and the leaf node depth growth strategy can be used to reduce the computational complexity and improve the model training efficiency, thereby ensuring the prediction accuracy and efficiency in wind power prediction tasks with large data sets and high-dimensional features.

[0177] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the processing module can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above-determined module. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0178] Indicatively, Figure 4 As shown, Figure 4 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 300 may be provided as a server. Figure 4The computer device 300 includes a processing component 302, which further includes one or more processors, and a memory resource represented by a memory 301 for storing instructions executable by the processing component 302, such as an application. The application stored in the memory 301 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 302 is configured to execute the instructions to perform the text recognition method of any of the above embodiments.

[0179] The computer device 300 may further include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate based on an operating system stored in the memory 301, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.

[0180] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0181] An embodiment of the present application provides a storage medium, in which computer-readable instructions are stored. When the computer-readable instructions are executed by one or more processors, the one or more processors execute a method provided in any embodiment.

[0182] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0183] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

[0184] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A wind power prediction method based on feature reconstruction, characterized in that: The method comprises: Obtain the original meteorological data and original unit data of the target wind farm to establish a data set; Generate a first input feature according to the original meteorological data and the original unit data; Reconstructing the first input feature by using a factor analysis algorithm to obtain a second input feature; Wherein, the dimension of the second input feature is smaller than the dimension of the first input feature; The second input feature is input into the target model to obtain a wind power prediction result.

2. The method according to claim 1, characterized in that The step of generating a first input feature according to the original meteorological data and the original unit data includes: One or more of a time series lag feature, a nonlinear feature and a moving average feature are generated according to the original meteorological data and the original unit data.

3. The method according to claim 2, characterized in that The original meteorological data includes wind speed data and wind direction data, and the original unit data includes power data; And, the generation process of the timing hysteresis feature includes: Determine a lag step length according to the time delay characteristics of the target wind farm, and process the original meteorological data and the original unit data according to the lag step length to generate a preset number of time series lag features; The generation process of the nonlinear feature includes: Generate one or more of wind speed cube feature, wind speed square feature, wind direction sine feature, wind direction cosine feature, and wind speed and wind direction combined feature according to the wind speed data and wind direction data; The generation process of the moving average feature includes: Determine the sliding window length; A moving average feature of the power data is generated within the sliding window length.

4. The method according to claim 1, characterized in that: The step of reconstructing the first input feature by using a factor analysis algorithm to obtain a second input feature includes: Determine a sample covariance matrix of each dimension in the first input feature; Establishing a factor analysis model, determining a factor loading matrix by principal component estimation method, and extracting a preset number of common factors from the first input features; Performing a rotation operation on the common factors, calculating a factor axis rotation matrix that maximizes the variance of the rotated factors, and generating a preset number of reconstruction factors; The first input feature is processed through the sample covariance matrix, the factor analysis model, the factor loading matrix, and the factor axis rotation matrix to obtain the second input feature.

5. The method according to claim 4, characterized in that The method further comprises: By means of optimal search, the number of preset common factors in the factor analysis model is optimized to obtain a second input feature whose target performance meets preset conditions.

6. The method according to claim 5, characterized in that The optimal search includes: Setting a search step size and a learning rate that are negatively correlated with the number of iterations and / or the rate of improvement of the target performance; Iteratively obtaining the second input feature according to the search step size and the learning rate until the target performance corresponding to the second input feature meets a preset condition; And, in the optimal search process, the method further includes: If the change rate of the target performance is less than a preset threshold, the early stopping mechanism is triggered to stop the optimal search, and during the search process, the feature whose target performance exceeds the performance threshold is selected as the second input feature.

7. The method according to any one of claims 1 to 6, characterized in that: The target model includes a LightGBM model; The step of inputting the second input feature into the target model to obtain a wind power prediction result includes: Dividing the data in the data set into training data and test data; Inputting the second input feature corresponding to the training data into the LightGBM model, and training the LightGBM model according to the histogram approximation algorithm and the leaf node depth growth strategy; The second input feature corresponding to the test data is input into the trained LightGBM model to obtain a wind power prediction result.

8. A wind power prediction device based on feature reconstruction, characterized in that: The device comprises: An acquisition module is used to acquire the original meteorological data and original unit data of the target wind farm to establish a data set; A processing module, configured to generate a first input feature according to the original meteorological data and the original unit data; The processing module is further used to perform feature reconstruction on the first input feature through a factor analysis algorithm to obtain a second input feature; Wherein, the dimension of the second input feature is smaller than the dimension of the first input feature; The processing module is further used to input the second input feature into the target model to obtain a wind power prediction result.

9. A computer device, characterized in that: The method comprises one or more processors and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the method according to any one of claims 1 to 7 are executed.

10. A storage medium, characterized in that: The storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the method according to any one of claims 1 to 7.

Images