A clean energy output prediction method, device and equipment based on meteorological pattern recognition and feature bias correction and a medium

Abstract

The application discloses a clean energy output prediction method and device based on meteorological mode identification and feature deviation correction, equipment and medium, including: constructing a daily scale meteorological feature vector according to a meteorological sample set; obtaining the membership of each meteorological sample under different meteorological modes, and determining the corresponding meteorological mode; screening the meteorological variables corresponding to each meteorological mode based on mutual information, and constructing the corresponding meteorological feature subspace, and obtaining the corresponding input vector based on the meteorological feature subspace and feature projection; constructing the corresponding clean energy output prediction model, and constructing a clean energy output prediction model set through a Bagging integration strategy; projecting the corrected to-be-predicted meteorological data to the meteorological feature subspace, and calling the corresponding clean energy output prediction model set to obtain a clean energy output prediction result. The application belongs to the field of energy prediction. The application can improve the accuracy of clean energy prediction.

Description

Technical Field

[0001] This invention relates to the field of energy forecasting, and in particular to a method, apparatus, equipment, and medium for predicting clean energy output based on meteorological pattern recognition and characteristic deviation correction. Background Technology

[0002] With the rapid growth of installed capacity of clean energy sources such as wind power and photovoltaics, their share of power output in the power system continues to increase. However, their inherent randomness, volatility, and uncertainty pose severe challenges to the safe operation and dispatching decisions of the power grid. Since the output of clean energy is highly dependent on meteorological conditions, and the dominant meteorological factors and their changing patterns vary significantly under different weather conditions, the output characteristics exhibit obvious non-stationarity and nonlinearity.

[0003] Existing forecasting methods often employ a unified feature system and a single model to model all meteorological scenarios, making it difficult to characterize the differences in the power output formation mechanisms under different meteorological models. This leads to a significant decrease in forecast accuracy when facing weather shifts or extreme events. Furthermore, high-dimensional meteorological data contains substantial redundancy and correlation; directly inputting the entire dataset easily introduces noise and increases model complexity. Moreover, the numerical weather prediction data used in actual forecasts exhibits systematic deviations from historical measured data. Without correction, this will result in inconsistent input distributions between the training and forecasting phases, weakening model reliability. Therefore, there is an urgent need to construct a novel forecasting method that integrates meteorological model recognition, differentiated feature modeling, key meteorological factor screening, and NWP bias correction to achieve high-precision and robust forecasting of clean energy power output under complex and variable meteorological conditions. Summary of the Invention

[0004] This invention provides a method, apparatus, equipment, and medium for predicting clean energy output based on meteorological pattern recognition and feature deviation correction. This solves the technical problem of low accuracy in clean energy output prediction in the prior art and achieves the technical effect of improving the accuracy of clean energy prediction.

[0005] In a first aspect, the present invention provides a method for predicting clean energy output based on meteorological pattern recognition and feature bias correction, comprising:

[0006] Acquire and preprocess historical power output sequences of clean energy power plants within the target area and meteorological data for the corresponding time periods; Based on the preprocessed meteorological data, a meteorological sample set is constructed, and a daily-scale meteorological feature vector is constructed based on the meteorological sample set; Based on the fuzzy clustering method, the daily meteorological feature vectors are clustered to obtain the membership degree of each meteorological sample under different meteorological models, and the corresponding meteorological models are determined. For each meteorological model, the meteorological variables corresponding to each meteorological model are selected based on mutual information, and the corresponding meteorological feature subspace is constructed. Based on the meteorological feature subspace and feature projection, the corresponding input vector is obtained. For each meteorological model, a corresponding clean energy output prediction model is constructed based on the meteorological feature subspace corresponding to each meteorological model, and a set of clean energy output prediction models is constructed through the Bagging integration strategy. After bias correction of the meteorological data to be predicted, the corrected meteorological data to be predicted is projected onto the meteorological feature subspace, and the corresponding clean energy output prediction model set is called to obtain the clean energy output prediction result.

[0007] Furthermore, based on the preprocessed meteorological data, a meteorological sample set is constructed, and a daily-scale meteorological feature vector is constructed based on the meteorological sample set, including: Construct the first The meteorological sample set for the day and the active power time series sample set include:

[0008]

[0009] in, For the first A collection of meteorological samples from the day. For the first Daily active power time series samples, For day sequence number, t For the first Time sampling time within the day, For the first The set of time indices corresponding to each day For the first Heavenly Historical meteorological data at any given time

[0010] in, Meteorological element dimension; For the first Heavenly The active power value at any given time; Construct daily-scale meteorological feature vectors based on meteorological sample sets. ,include:

[0011] in, d Indicates the first d sky, Indicates the first Each meteorological element , The total number of meteorological elements. To indicate the first Day Time Index Set The number of elements, Meteorological elements j In the d The daily average value of the day To represent meteorological elements j In the Heavenly t The observed value at time, Meteorological elements j In the d Daily standard deviation of days Meteorological elements j The change between adjacent observation times Meteorological elements j In the Heavenly The observed value at time, For the first d Weather and meteorological elements j The average rate of change and Meteorological elements j In the Daily maximum and daily minimum values, Meteorological elements j In the d The range of changes in the sky For the first M Meteorological elements in the first d The daily average value of the day For the first M Meteorological elements in the first d Daily standard deviation of days Meteorological elements M In the d The range of changes in the sky For the first d Weather and meteorological elements M The average rate of change This is a column vector of daily meteorological features. The dimension is .

[0012] Furthermore, based on fuzzy clustering, daily meteorological feature vectors are clustered to obtain the membership degree of each meteorological sample under different meteorological models, and the corresponding meteorological models are determined, including: Determine the number of clusters in fuzzy clustering K Construct a clustering effectiveness evaluation function:

[0013] in, K The number of clusters, This is the membership matrix. As an intra-class compactness index, The intra-class separation index, As an index of inter-class overlap, It is an exponential adjustment parameter, and ; For any diurnal meteorological feature vector Define its relation to the first The membership degree of each meteorological model is ,satisfy:

[0014]

[0015] Definition of the first The central vector corresponding to each meteorological model is : The meteorological model center is jointly determined by minimizing the fuzzy clustering objective function. membership degree ,include:

[0016] in, Let be the objective function. For historical sample days, The number of clusters, For fuzzy weighted index and , Represents the Euclidean norm; Update membership degree iteratively. With cluster center ,include:

[0017] when season The membership degree of the rest is 0;

[0018] The iteration stops when the change in the membership matrix between two consecutive iterations is less than a preset threshold. After clustering is completed, the meteorological model membership vector corresponding to each meteorological sample is obtained, including:

[0019] in, This is the membership vector of the meteorological model; Determine the first based on the principle of maximum membership. The corresponding weather pattern for the day:

[0020] in, For the first Weather patterns for the day.

[0021] Furthermore, for each meteorological model, meteorological variables corresponding to each model are selected based on mutual information, and a corresponding meteorological feature subspace is constructed. Based on the meteorological feature subspace and feature projection, the corresponding input vector is obtained, including: Meteorological samples are divided according to meteorological models. The subsample set includes:

[0022] in, This indicates that it belongs to a weather model. The sample day index set; In meteorological models The following is a construction of mutual information indicators between meteorological variables and clean energy output, including:

[0023] in, Indicates weather model Next Random variables corresponding to meteorological variables This represents the random variable of clean energy output under meteorological models. For the first Meteorological variables in meteorological models The correlation strength between the output of clean energy and meteorological characteristics was analyzed and used as a screening indicator. For mutual information; The definition of mutual information includes:

[0024] in, Let be the joint probability density function. and These are the marginal probability density functions; In meteorological models Below, mutual information indicators Sort the data and select meteorological variables that are greater than the preset mutual information value to construct the first... A subset of meteorological features under a meteorological model;

[0025] in, For the first A subset of meteorological features from a meteorological model. For the first The number of effective meteorological variables selected from each meteorological model In the first Under the meteorological model, the selected [number]th [item] after mutual information sorting The original variable index number of each meteorological variable. For the first Under the weather model, the first The mutual information value between a meteorological variable and the forecast target; Meteorological variables corresponding to mutual information values ​​greater than a preset value are combined to obtain a meteorological feature subspace, including:

[0026] in, No. The first weather model Tian Shike meteorological feature subspace, For the number Meteorological variables in the first Tian Shike The observed values, A subset of meteorological features The original variable index number corresponding to each meteorological variable in the data; Definition of the first The characteristic projection operator of each meteorological model is: The characteristic projection operator is used to map the entire meteorological variable space to the first... The meteorological feature subspace corresponding to the model; In the Tian Shike The feature projection results are as follows:

[0027] in,

[0028] in, For the first Tian Shike The complete set of meteorological variables vectors, For transpose, Meteorological elements In the Heavenly The observed value at time;

[0029] in, Indicates weather model The corresponding meteorological feature subspace vector.

[0030] Furthermore, for each meteorological model, a corresponding clean energy output prediction model is constructed based on the meteorological feature subspace corresponding to each meteorological model. Then, through a Bagging ensemble strategy, a set of clean energy output prediction models is constructed, including: Based on feature projection operator It belongs to the meteorological model. The samples are projected onto the corresponding meteorological feature subspace to construct the model training input vector;

[0031] in, In the first Under the first weather pattern, the first Day, time The meteorological feature projection vector is the model input vector; No. The training sample set for each weather model includes:

[0032] in, For the first Training sample set for each weather model For the first The sky The actual output of clean energy at all times; Definition of the first The corresponding prediction models for each weather pattern are as follows:

[0033] in, For the first Under the weather model, the first Tian Shike The predicted output For the first Clean energy output prediction model function corresponding to each meteorological model; Based on the gradient boosting regression model, the prediction model is updated, including:

[0034] in, For the total number of regression trees, For the first In mode of the first A regression tree; Construct the objective function corresponding to the prediction model, including:

[0035] in, The objective function corresponding to the prediction model, Let the prediction error loss function be... For the first The complexity regularization term for trees;

[0036] in, This is the actual power. To predict power; In the In each iteration, the first and second derivatives of the loss function with respect to the predicted values ​​are introduced, including:

[0037] in, For the first-order gradient, It is a second-order gradient. The sign of the partial derivative. For the first A sample at time... The actual power generation capacity of clean energy This represents the prediction power of the current model for this sample at the end of the previous training round. Pattern training sample set Perform multiple resampling operations to construct Individual training set And train a prediction sub-model on each sub-training set. It integrates the outputs of multiple sub-models, including:

[0038] in, In the first k Under the first weather model, for the first The predicted clean energy output value at time t, output by the Bagging integrated prediction model; A set of clean energy output prediction models corresponding one-to-one with each meteorological model was obtained, including:

[0039] in, A set of models for predicting clean energy output.

[0040] Furthermore, after bias correction of the meteorological data to be predicted, the corrected meteorological data to be predicted is projected onto the meteorological feature subspace, and the corresponding clean energy output prediction model set is called to obtain the clean energy output prediction results, including: For each meteorological variable, a corresponding meteorological bias is constructed, including:

[0041] in, For the first j The meteorological variable in the first Weather deviation at time t For the first j The meteorological variable in the first Tian Shike t The true value, For the first j The meteorological variable in the first Tian Shike t The forecast value; No. The definition of the mean deviation of the meteorological model corresponding to each meteorological variable includes:

[0042] in, For the first The mean deviation of the meteorological model corresponding to each meteorological variable For the first time in history d The first day of the sample j Meteorological deviations of individual meteorological variables; Determine the set of mean deviations for each meteorological model, including:

[0043] in, This is the set of mean deviations under various meteorological models. For the first The mean deviation of the meteorological model corresponding to each meteorological variable dimension; For the predicted date The Middle At what time, the corresponding forecast meteorological variable vector is:

[0044] in, For the first The first predicted day The column vector of forecast meteorological variables at any given time. For the first The meteorological variable in the first The predicted day Forecast value for the time; Obtain the daily-scale meteorological feature vector corresponding to the prediction date. Based on the FCM meteorological model that has been trained in historical periods, the predicted date is calculated. Membership degree under various meteorological models:

[0045] in, The first known historical stage Each meteorological model center, The first known historical stage Each meteorological model center, For prediction date In the Membership degree under a weather pattern; The prediction date is determined using the maximum membership principle. Belonging meteorological model:

[0046] Determine the forecast date The meteorological model corresponding to the meteorological variable vector is used, and bias correction is performed, including:

[0047] in, For the first The meteorological variable in the first Predicted days The predicted value after bias correction at time, For the first The meteorological variable in the first Predicted days Forecast value for the time, The first one obtained from historical sample statistics The first weather model The mean deviation of each variable; The corrected forecast period's meteorological variable vectors are projected onto the meteorological feature subspace of the meteorological model, including:

[0048]

[0049] in, In the first Predicted days At that moment, the corrected meteorological variable vector, In the first Predicted days At this moment, the first The values ​​of the meteorological variables after bias correction This is the joint meteorological feature vector used as input to the prediction model; Get the first Predicted days The clean energy output forecast results for each moment include:

[0050] in, For the first Predicted days Real-time clean energy output forecast results To match the predicted date Belonging meteorological model Corresponding clean energy output prediction model; Based on the Bagging integration strategy, the clean energy output forecast results are integrated and updated, including:

[0051] in, For weather models Next Predicted days Real-time clean energy output forecast results For weather models Based on the first A prediction sub-model constructed from a resampled training set.

[0052] Furthermore, historical power output sequences of clean energy power plants within the target area and corresponding meteorological data for the corresponding time periods are acquired and preprocessed, including: The historical output sequences of clean energy power plants in the target area and the meteorological data for the corresponding time periods are further time-stamped, abnormal data are removed, and linear interpolation is performed.

[0053] Secondly, the present invention provides a clean energy output prediction device based on meteorological pattern recognition and feature deviation correction, comprising: The historical data processing module is used to acquire and preprocess the historical output sequence of clean energy power stations in the target area and the meteorological data of the corresponding time period; The meteorological sample set module is used to construct a meteorological sample set based on preprocessed meteorological data, and to construct a daily-scale meteorological feature vector based on the meteorological sample set; The clustering module is used to cluster daily meteorological feature vectors based on fuzzy clustering, obtain the membership degree of each meteorological sample under different meteorological models, and determine the corresponding meteorological model. The mutual information filtering module is used to filter the meteorological variables corresponding to each meteorological model based on mutual information, construct the corresponding meteorological feature subspace, and obtain the corresponding input vector based on the meteorological feature subspace and feature projection. The prediction model set module is used to construct corresponding clean energy output prediction models for each meteorological model based on the meteorological feature subspace corresponding to each meteorological model, and to construct a set of clean energy output prediction models through the Bagging integration strategy. The prediction results module is used to correct the deviation of the meteorological data to be predicted, project the corrected meteorological data to the meteorological feature subspace, and call the corresponding clean energy output prediction model set to obtain the clean energy output prediction results.

[0054] Thirdly, the present invention provides an electronic device, comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to execute a clean energy output prediction method based on meteorological pattern recognition and feature deviation correction, as provided in the first aspect.

[0055] Fourthly, the present invention provides a non-transitory computer-readable storage medium, wherein when the instructions in the non-transitory computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to execute a clean energy output prediction method based on meteorological pattern recognition and feature deviation correction as provided in the first aspect.

[0056] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention proposes a clean energy output prediction method based on meteorological model recognition and differentiated feature modeling. First, daily-scale meteorological features are constructed using historical meteorological data, and the meteorological samples are clustered using fuzzy C-means clustering. The meteorological model is determined through the maximum membership principle. Second, under the constraints of the meteorological model, mutual information indices are introduced from the original meteorological variable time series to screen high-value meteorological features, constructing a differentiated meteorological feature subspace that matches different meteorological models, effectively reducing meteorological feature redundancy and highlighting key driving factors. Based on this, clean energy output prediction models are trained for each meteorological model, enabling the models to fully learn the nonlinear mapping relationship between meteorological elements and output under different meteorological conditions. Simultaneously, considering the systematic bias between numerical weather prediction meteorological data and historical measured meteorological data used in the prediction stage, a model-based numerical weather prediction meteorological variable bias modeling and correction mechanism is introduced to correct the input meteorological features in the prediction stage, reducing the impact of systematic bias in the input data on the accuracy of clean energy output prediction. Finally, in the prediction stage, the corresponding prediction model is dynamically selected through meteorological model recognition to complete the clean energy output prediction.

[0057] This invention utilizes fuzzy C-means clustering to classify meteorological models, combines mutual information to screen key features and reduce redundancy; customizes prediction models for different meteorological models to enhance nonlinear mapping capabilities; and introduces a meteorological model-driven NWP bias correction mechanism to alleviate systematic bias between measured and forecast data, thereby significantly improving the robustness and accuracy of predictions while dynamically matching the optimal model. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart illustrating a clean energy output prediction method based on meteorological pattern recognition and feature deviation correction provided by the present invention. Figure 2 This is a flowchart illustrating another clean energy output prediction method based on meteorological pattern recognition and feature deviation correction provided by the present invention. Detailed Implementation

[0060] This invention provides a method for predicting clean energy output based on meteorological pattern recognition and feature deviation correction, which solves the technical problem of low accuracy in predicting clean energy output in the prior art.

[0061] The technical solution of this invention is to solve the above-mentioned technical problems, and the overall idea is as follows: A method for predicting clean energy output based on meteorological pattern recognition and feature bias correction includes: acquiring and preprocessing historical output sequences of clean energy power plants within a target area and meteorological data for corresponding time periods; constructing a meteorological sample set based on the preprocessed meteorological data, and constructing a daily-scale meteorological feature vector based on the meteorological sample set; clustering the daily-scale meteorological feature vector based on fuzzy clustering to obtain the membership degree of each meteorological sample under different meteorological patterns, and determining the corresponding meteorological pattern; for each meteorological pattern, filtering the meteorological variables corresponding to each meteorological pattern based on mutual information, constructing the corresponding meteorological feature subspace, and obtaining the corresponding input vector based on the meteorological feature subspace and feature projection; for each meteorological pattern, constructing the corresponding clean energy output prediction model based on the meteorological feature subspace corresponding to each meteorological pattern, and constructing a clean energy output prediction model set through a Bagging ensemble strategy; after bias correction of the meteorological data to be predicted, projecting the corrected meteorological data to be predicted onto the meteorological feature subspace, and calling the corresponding clean energy output prediction model set to obtain the clean energy output prediction result.

[0062] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0063] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0064] S11: Acquire and preprocess historical power output sequences of clean energy power plants within the target area and meteorological data for the corresponding time periods.

[0065] Specifically, this includes: continuing to align the historical output sequences of clean energy power plants within the target area with the corresponding meteorological data for the same period using timestamps, removing outlier data, and performing linear interpolation.

[0066] It can obtain the historical output sequence of clean energy power plants in the target area, as well as the corresponding meteorological data for the same period.

[0067] Historical power output sequences can include time series of active power from wind or solar power, denoted as... Historical meteorological data can include time series of elements such as wind speed, wind direction, irradiance, temperature, humidity, air pressure, and cloud cover, denoted as... ,in, express At any given time, M represents the dimension of the meteorological feature.

[0068] Historical power output sequences were aligned with meteorological data by timestamp, and data cleaning and normalization processes were performed, including missing value imputation, outlier removal, and normalization, to ensure that the data quality met the modeling requirements.

[0069] For example, missing data points can be estimated using linear interpolation; outliers that significantly deviate from the average level are removed using the three-standard-deviation principle. After preprocessing, a clean energy output sequence with controllable quality and corresponding meteorological element sequences are obtained, providing a reliable data foundation for subsequent analysis.

[0070]

[0071] The above formula represents the linear interpolation process: when time point When data is missing, utilize the adjacent time points before and after it. and numerical value , Linear estimation is performed according to time proportions to fill in missing values.

[0072]

[0073] in and These represent the mean and standard deviation of all historical values ​​of the variable, respectively. The preprocessing steps described above effectively improve the reliability and consistency of the data. Existing research also emphasizes the importance of high-quality data for the accuracy of power output prediction.

[0074] S12. Based on the preprocessed meteorological data, a meteorological sample set is constructed, and a daily-scale meteorological feature vector is constructed based on the meteorological sample set.

[0075] Specifically, it includes: Construct the first The meteorological sample set for the day and the active power time series sample set include:

[0076]

[0077] in, For the first A collection of meteorological samples from the day. For the first Daily active power time series samples, For day sequence number, t For the first Time sampling time within the day, For the first The set of time indices corresponding to each day For the first Heavenly Historical meteorological data at any given time

[0078] in, Meteorological element dimension; For the first Heavenly The active power value at any given time; Based on the construction of the first Weather sample set Statistical quantities and variation characteristics that reflect the overall meteorological features of a day are extracted from intraday meteorological time series to form a daily-scale meteorological feature vector. .

[0079] Construct daily-scale meteorological feature vectors based on meteorological sample sets. ,include:

[0080] in,d Indicates the first d sky, Indicates the first Each meteorological element , The total number of meteorological elements. Indicates the first Day Time Index Set The number of elements, Meteorological elements j In the d The daily average value of the day To represent meteorological elements j In the Heavenly t The observed value at time, Meteorological elements j In the d Daily standard deviation of days Meteorological elements j The change between adjacent observation times Meteorological elements j In the Heavenly The observed value at time, For the first d Weather and meteorological elements j The average rate of change and Meteorological elements j In the Daily maximum and daily minimum values, Meteorological elements j In the d The range of changes in the sky For the first M Meteorological elements in the first d The daily average value of the day For the first M Meteorological elements in the first d Daily standard deviation of days Meteorological elements M In the d The range of changes in the sky For the first d Weather and meteorological elements M The average rate of change This is a column vector of daily meteorological features. The dimension is .

[0081] For any meteorological element Its diurnal scale characteristics may include, but are not limited to: daily average. Daily standard deviation Daily extreme value , and amplitude Intraday rate of change statistics , Combining the above features constitutes the first Meteorological feature vector of the day .

[0082] The intraday characteristics of renewable energy output under different weather conditions have been well understood. Dividing historical samples by weather type and extracting output statistics for different types for modeling has greatly improved the prediction accuracy under multiple weather conditions. Identifying similar days by assessing the similarity of multiple meteorological factors to improve prediction can be regarded as another way of utilizing diurnal scale characteristics.

[0083] Therefore, the diurnal scale features extracted in step S12 can effectively characterize the pattern differences of the power output curves under different meteorological conditions, laying the foundation for pattern recognition and sub-pattern modeling.

[0084] S13. Based on the fuzzy clustering method, the daily-scale meteorological feature vectors are clustered to obtain the membership degree of each meteorological sample under different meteorological models, and the corresponding meteorological models are determined.

[0085] Because the daily-scale meteorological feature vector set obtained in step S12 The differences are significant and cannot be clearly divided into distinct clusters. Therefore, hard clustering methods such as k-means cannot be used to partition them. Instead, each feature vector should be assigned a weight to indicate its degree of belonging to a certain cluster. Therefore, this invention chooses the fuzzy clustering method FCM to construct and identify meteorological patterns from historical meteorological samples, avoiding the discontinuity of pattern boundaries caused by hard classification. Based on fuzzy clustering, daily meteorological feature vectors are clustered to obtain the membership degree of each meteorological sample under different meteorological models, and the corresponding meteorological models are determined. Specifically, this includes: When using the FCM algorithm as an unsupervised learning method, the number of clusters *c* needs to be determined before clustering, as this directly affects the clustering results. The optimal number of clusters is... Determined by the W effectiveness index.

[0086] Determine the number of clusters in fuzzy clustering K Construct a clustering effectiveness evaluation function:

[0087] in, K The number of clusters, This is the membership matrix. As an intra-class compactness index, The intra-class separation index, As an index of inter-class overlap, It is an exponential adjustment parameter, and This is used to adjust the sensitivity of intra-class compactness, inter-class separation, and inter-class overlap in the clustering effectiveness evaluation function, by making... Determine the optimal number of clusters by selecting the optimal value. ; For any diurnal meteorological feature vector Define its relation to the first The membership degree of each meteorological model is ,satisfy:

[0088]

[0089] Define the center vector corresponding to any meteorological model as: ,include: The meteorological model center is jointly determined by minimizing the fuzzy clustering objective function. membership degree ,include:

[0090] in, Let be the objective function. For historical sample days, The number of clusters is also the total number of weather models. For fuzzy weighted index and , Represents the Euclidean norm; Given a fuzzy weighting index Under the condition of iterative update and ,include:

[0091] when season The remaining membership degrees are 0.

[0092]

[0093] Repeat the above update process, and stop iterating when the change in the membership matrix between two consecutive iterations is less than a preset threshold; After clustering is completed, the meteorological model membership vector corresponding to each meteorological sample is obtained, including:

[0094] in, This is the membership vector of the meteorological model; Determine the first based on the principle of maximum membership. The corresponding weather models for the day include:

[0095] in, For the first Weather patterns for the day.

[0096] Membership vectors are used to characterize the degree of membership of samples under different meteorological patterns. Based on this, the maximum membership principle is used to determine the pattern of the samples. It is classified into the meteorological model with the highest membership degree.

[0097] S14. For each meteorological model, the meteorological variables corresponding to each meteorological model are selected based on mutual information, and the corresponding meteorological feature subspace is constructed. Based on the meteorological feature subspace and feature projection, the corresponding input vector is obtained.

[0098] For the different meteorological models identified in step S13, the differences in the degree of influence of each meteorological element on clean energy output under different meteorological models are fully considered, and differentiated meteorological feature variables that match the meteorological models are constructed to improve the relevance of input features and reduce the interference of redundant features on the prediction model.

[0099] For each meteorological model, meteorological variables corresponding to each model are selected based on mutual information, and a corresponding meteorological feature subspace is constructed. Based on the meteorological feature subspace and feature projection, the corresponding input vector is obtained, including: Meteorological samples are divided according to meteorological models. The subsample set includes:

[0100] in, This indicates that it belongs to a weather model. The sample day index set; In the subsequent feature selection and subspace construction process, for any meteorological model Only using The historical samples within the data are the (historical) meteorological data (i.e., the original meteorological variable time series) defined in step one. With active power time series and (clean energy output time series) Perform statistical analysis.

[0101] In meteorological models For each historical meteorological data point, a mutual information index is constructed between the historical meteorological data and clean energy output to quantify the effective information contained in the output change of the meteorological variable under the meteorological model.

[0102]

[0103] in, Indicates weather model Next Random variables corresponding to meteorological variables This represents the random variable of clean energy output under meteorological models. For the first Meteorological variables in meteorological models The correlation strength between the output of clean energy and meteorological characteristics was analyzed and used as a screening indicator. Mutual information; mutual information value The larger the value, the better in weather models. The stronger the statistical dependence between the original meteorological variable and clean energy output, the higher its explanatory power for changes in output.

[0104] The definition of mutual information includes:

[0105] in, Let be the joint probability density function. and These are the marginal probability density functions; In meteorological models Below, mutual information indicators Sort the data and select meteorological variables that are greater than the preset mutual information value to construct the first... A subset of meteorological features under a meteorological model;

[0106] in, For the first A subset of meteorological features from a meteorological model. For the first The number of effective meteorological variables selected from each meteorological model In the first Under the meteorological model, the selected [number]th [item] after mutual information sorting The original variable index number of each meteorological variable. For the first Under the first weather pattern, the first The mutual information value between a meteorological variable and the forecast target; Meteorological variables corresponding to mutual information values ​​greater than a preset value are combined to obtain a (differentiated) meteorological feature subspace, including:

[0107] in, No. The first weather model Tian Shike meteorological feature subspace, For the number Meteorological variables in the first Tian Shike The observed values, A subset of meteorological features The original variable index number corresponding to each meteorological variable in the data; The (differentiated) meteorological feature subspace reflects the meteorological model The original combination of meteorological variables that plays a dominant role in changes in clean energy output enables subsequent forecasting models to focus on learning the key driving factors under this meteorological pattern.

[0108] To unify the input format for subsequent model training and prediction phases, a meteorological model is defined. Feature projection operator .

[0109] Definition of the first The characteristic projection operator of each meteorological model is: , Used to spatially map the entire meteorological variable space to the first The meteorological feature subspace corresponding to the model.

[0110] This projection is a selection mapping based on the set of variable indices, at the... Tian Shike The feature projection results include:

[0111] in,

[0112] in, For the first Tian Shike The complete set of meteorological variables vectors, For transpose, Meteorological elements In the Heavenly The observed value at time;

[0113] in, Indicates weather model The corresponding meteorological feature subspace vector.

[0114] The feature projection result can be for any historical sample or any predicted sample.

[0115] S15. For each meteorological model, a corresponding clean energy output prediction model is constructed based on the meteorological feature subspace corresponding to each meteorological model. A set of clean energy output prediction models is constructed through the Bagging integration strategy.

[0116] For the different meteorological models constructed in step S13, and in conjunction with the differentiated meteorological feature subspace determined in step S14, corresponding clean energy output prediction models are constructed respectively, forming a mapping relationship between meteorological models and prediction models.

[0117] Based on feature projection operator It belongs to the meteorological model. The samples are projected onto the corresponding meteorological feature subspace to construct the model training input vector;

[0118] in, In the first Under the first weather pattern, the first Day, time The meteorological feature projection vector is the model input vector; No. The training sample set for each weather model includes:

[0119] in, For the first Training sample set for each weather model For the first The sky The actual output of clean energy at all times; For each meteorological model, a corresponding clean energy output prediction sub-model is constructed to describe the driving relationship between the original meteorological variables and the changes in clean energy output under that meteorological model.

[0120] Definition of the first The corresponding prediction models for each weather pattern are as follows:

[0121] in, For the first Under the first weather pattern, the first Tian Shike The predicted output For the first Clean energy output prediction model function corresponding to each meteorological model; Based on the gradient boosting regression model, the prediction model is updated, including:

[0122] in, For the total number of regression trees, For the first In mode of the first A regression tree; The objective function for model training can be composed of a loss function term and a model complexity regularization term.

[0123] Construct the objective function corresponding to the prediction model, including:

[0124] in, The objective function corresponding to the prediction model, The prediction error loss function is preferably the squared error. For the first The complexity regularization term for trees;

[0125] in, This is the actual power. To predict power; In the In each iteration, the first and second derivatives of the loss function with respect to the predicted values ​​are introduced, including:

[0126] in, This is the first-order gradient (residual direction). This is the second-order gradient (curvature information). The sign of the partial derivative. For the first A sample at time... The actual power generation capacity of clean energy This represents the prediction power of the current model for this sample at the end of the previous training round. By solving for the new regression tree using a second-order approximation method, the model achieves a balance between fitting accuracy and generalization ability.

[0127] To improve the stability of the prediction model under different sample perturbations and extreme weather conditions, a Bagging ensemble learning strategy is introduced in each meteorological model.

[0128] Pattern training sample set Perform multiple resampling operations to construct Individual training set And train a prediction sub-model on each sub-training set. It integrates the outputs of multiple sub-models, including:

[0129] in, In the first k Under the first weather model, for the first The predicted clean energy output value at time t, output by the Bagging integrated prediction model; A set of clean energy output prediction models corresponding one-to-one with each meteorological model was obtained, including:

[0130] in, A set of models for predicting clean energy output.

[0131] Each prediction model is trained in its corresponding meteorological feature subspace, and is used in subsequent steps S16 for prediction model invocation based on meteorological pattern recognition results and clean energy output prediction.

[0132] S16. After bias correction of the meteorological data to be predicted, the corrected meteorological data to be predicted is projected onto the meteorological feature subspace, and the corresponding clean energy output prediction model set is called to obtain the clean energy output prediction result.

[0133] Considering the systematic bias between numerical weather prediction meteorological data and historical meteorological data used in the forecasting stage, the input meteorological characteristics in the forecasting stage are corrected to reduce the impact of systematic bias in the input data on the accuracy of clean energy output forecasting.

[0134] After bias correction of the meteorological data to be predicted, the corrected meteorological data is projected onto the meteorological feature subspace, and the corresponding clean energy output prediction model set is called to obtain the clean energy output prediction results, including: Based on historical samples, obtain (measured) meteorological variable time series aligned with clean energy output time. And the corresponding time series of (numerical weather) forecast meteorological variables. .

[0135] For each meteorological variable, a corresponding meteorological bias is constructed, including:

[0136] in, For the first j The meteorological variable in the first Weather deviation at time t For the first j The meteorological variable in the first Tian Shike t The true value, For the first j The meteorological variable in the first Tian Shike t The forecast value; Based on the meteorological model classification results obtained in step S13, historical weather data are grouped according to meteorological models. Below, only using The data within the period were used to statistically model the deviations of variables in various historical weather data.

[0137] No. The definition of the mean deviation of the meteorological model corresponding to each meteorological variable includes:

[0138] in, For the first The mean deviation of the meteorological model corresponding to each meteorological variable For the first time in history d The first day of the sample j Meteorological deviations of individual meteorological variables; Determine the set of mean deviations for each meteorological model, including:

[0139] in, This is the set of mean deviations under various meteorological models. For the first The mean deviation of the meteorological model corresponding to each meteorological variable dimension; For the predicted date The Middle At what time, the corresponding forecast meteorological variable vector is:

[0140] in, For the first The first predicted day The column vector of forecast meteorological variables at any given time. For the first The meteorological variable in the first The predicted day Forecast value for the time; Obtain the daily-scale meteorological feature vector corresponding to the prediction date. Based on the FCM meteorological model that has been trained in historical periods, the predicted date is calculated. Membership degree under various meteorological models:

[0141] in, The first known historical stage Each meteorological model center, The first known historical stage Each meteorological model center, For prediction date In the Membership degree under a weather pattern; The prediction date is determined using the maximum membership principle. Belonging meteorological model:

[0142] Determine the forecast date The meteorological model corresponding to the meteorological variable vector is used, and bias correction is performed, including:

[0143] in, For the first The meteorological variable in the first Predicted days The predicted value after bias correction at time, For the first The meteorological variable in the first Predicted days Forecast value for the time, The first one obtained from historical sample statistics The first weather model The mean deviation of each variable; The corrected forecast period's meteorological variable vectors are projected onto the meteorological feature subspace of the meteorological model, including:

[0144]

[0145] in, In the first Predicted days At that moment, the corrected meteorological variable vector, In the first Predicted days At this moment, the first The values ​​of the meteorological variables after bias correction This is the joint meteorological feature vector used as input to the prediction model; Get the first Predicted days The clean energy output forecast results for each moment include:

[0146] in, For the first Predicted days Real-time clean energy output forecast results To match the predicted date Belonging meteorological model Corresponding clean energy output prediction model; Based on the Bagging integration strategy, the clean energy output forecast results are integrated and updated, including:

[0147] in, For weather models Next Predicted days Real-time clean energy output forecast results For weather models Based on the first A prediction sub-model constructed from a resampled training set.

[0148] Through steps S11-S16, a clean energy output prediction process based on meteorological model recognition was implemented, yielding the first... Predicted days Clean energy output forecast at any time The prediction method can dynamically select appropriate prediction models and input features for different meteorological models, effectively improving prediction accuracy and stability under complex meteorological conditions. The clean energy output prediction method based on meteorological model recognition and differentiated feature modeling provided in this invention can also be referenced... Figure 2 .

[0149] In summary, this invention proposes a clean energy output prediction method based on meteorological model recognition and differentiated feature modeling. First, daily-scale meteorological features are constructed using historical meteorological data, and the meteorological samples are clustered using fuzzy C-means clustering. The meteorological model is determined through the maximum membership principle. Second, under the constraints of the meteorological model, mutual information indices are introduced from the original meteorological variable time series to screen high-value meteorological features, constructing a differentiated meteorological feature subspace that matches different meteorological models, effectively reducing meteorological feature redundancy and highlighting key driving factors. Based on this, clean energy output prediction models are trained for each meteorological model, enabling the models to fully learn the nonlinear mapping relationship between meteorological elements and output under different meteorological conditions. Simultaneously, considering the systematic bias between numerical weather prediction meteorological data and historical measured meteorological data used in the prediction stage, a model-based numerical weather prediction meteorological variable bias modeling and correction mechanism is introduced to correct the input meteorological features in the prediction stage, reducing the impact of systematic bias in the input data on the accuracy of clean energy output prediction. Finally, in the prediction stage, the corresponding prediction model is dynamically selected through meteorological model recognition to complete the clean energy output prediction.

[0150] This invention utilizes fuzzy C-means clustering to classify meteorological models, combines mutual information to screen key features and reduce redundancy; customizes prediction models for different meteorological models to enhance nonlinear mapping capabilities; and introduces a meteorological model-driven NWP bias correction mechanism to alleviate systematic bias between measured and forecast data, thereby significantly improving the robustness and accuracy of predictions while dynamically matching the optimal model.

[0151] Based on the same inventive concept, this invention provides a clean energy output prediction device based on meteorological pattern recognition and feature deviation correction, comprising: The historical data processing module is used to acquire and preprocess the historical output sequence of clean energy power stations in the target area and the meteorological data of the corresponding time period; The meteorological sample set module is used to construct a meteorological sample set based on preprocessed meteorological data, and to construct a daily-scale meteorological feature vector based on the meteorological sample set; The clustering module is used to cluster daily meteorological feature vectors based on fuzzy clustering, obtain the membership degree of each meteorological sample under different meteorological models, and determine the corresponding meteorological model. The mutual information filtering module is used to filter the meteorological variables corresponding to each meteorological model based on mutual information, construct the corresponding meteorological feature subspace, and obtain the corresponding input vector based on the meteorological feature subspace and feature projection. The prediction model set module is used to construct corresponding clean energy output prediction models for each meteorological model based on the meteorological feature subspace corresponding to each meteorological model, and to construct a set of clean energy output prediction models through the Bagging integration strategy. The prediction results module is used to correct the deviation of the meteorological data to be predicted, project the corrected meteorological data to the meteorological feature subspace, and call the corresponding clean energy output prediction model set to obtain the clean energy output prediction results.

[0152] Based on the same inventive concept, the present invention also provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to perform a clean energy output prediction based on meteorological pattern recognition and feature bias correction, as described above.

[0153] Based on the same inventive concept, the present invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to perform a clean energy output prediction based on meteorological pattern recognition and feature deviation correction as described above.

[0154] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0155] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0156] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0157] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0158] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0159] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting clean energy output based on meteorological pattern recognition and feature bias correction, characterized in that, include: Acquire and preprocess historical power output sequences of clean energy power plants within the target area and meteorological data for the corresponding time periods; Based on the preprocessed meteorological data, a meteorological sample set is constructed, and a daily-scale meteorological feature vector is constructed based on the meteorological sample set; Based on the fuzzy clustering method, the daily meteorological feature vectors are clustered to obtain the membership degree of each meteorological sample under different meteorological models, and the corresponding meteorological models are determined. For each meteorological model, the meteorological variables corresponding to each meteorological model are selected based on mutual information, and the corresponding meteorological feature subspace is constructed. Based on the meteorological feature subspace and feature projection, the corresponding input vector is obtained. For each meteorological model, a corresponding clean energy output prediction model is constructed based on the meteorological feature subspace corresponding to each meteorological model, and a set of clean energy output prediction models is constructed through the Bagging integration strategy. After bias correction of the meteorological data to be predicted, the corrected meteorological data to be predicted is projected onto the meteorological feature subspace, and the corresponding clean energy output prediction model set is called to obtain the clean energy output prediction result.

2. The clean energy output prediction method based on meteorological pattern recognition and feature bias correction as described in claim 1, characterized in that, Based on the preprocessed meteorological data, a meteorological sample set is constructed, and a daily-scale meteorological feature vector is constructed based on the meteorological sample set, including: Construct the first The meteorological sample set for the day and the active power time series sample set include: in, For the first A collection of meteorological samples from the day. For the first Daily active power time series samples, For day sequence number, t For the first Time sampling time within the day, For the first The set of time indices corresponding to each day For the first Heavenly Historical meteorological data at any given time in, Meteorological element dimension; For the first Heavenly The active power value at any given time; Construct daily-scale meteorological feature vectors based on meteorological sample sets. ,include: in, d Indicates the first d sky, Indicates the first Each meteorological element , The total number of meteorological elements. To indicate the first Day Time Index Set The number of elements, Meteorological elements j In the d The daily average value of the day To represent meteorological elements j In the Heavenly t The observed value at time, Meteorological elements j In the d Daily standard deviation of days Meteorological elements j The change between adjacent observation times Meteorological elements j In the Heavenly The observed value at time, For the first d Weather and meteorological elements j The average rate of change and Meteorological elements j In the Daily maximum and daily minimum values, Meteorological elements j In the d The range of changes in the sky For the first M Meteorological elements in the first d The daily average value of the day For the first M Meteorological elements in the first d Daily standard deviation of days Meteorological elements M In the d The range of changes in the sky For the first d Weather and meteorological elements M The average rate of change This is a column vector of daily meteorological features. The dimension is .

3. The clean energy output prediction method based on meteorological pattern recognition and feature bias correction as described in claim 2, characterized in that, Based on fuzzy clustering, daily meteorological feature vectors are clustered to obtain the membership degree of each meteorological sample under different meteorological models, and the corresponding meteorological models are determined, including: Determine the number of clusters in fuzzy clustering K Construct a clustering effectiveness evaluation function: in, K The number of clusters, This is the membership matrix. As an intra-class compactness index, The intra-class separation index, As an index of inter-class overlap, It is an exponential adjustment parameter, and ; For any diurnal meteorological feature vector Define its relationship to the first The membership degree of each meteorological model is ,satisfy: Definition of the first The central vector corresponding to each meteorological model is : The meteorological model center is jointly determined by minimizing the fuzzy clustering objective function. membership degree ,include: in, Let be the objective function. For historical sample days, The number of clusters, For fuzzy weighted index and , Represents the Euclidean norm; Update membership degree iteratively. With cluster center ,include: when season The membership degree of the rest is 0; The iteration stops when the change in the membership matrix between two consecutive iterations is less than a preset threshold. After clustering is completed, the meteorological model membership vector corresponding to each meteorological sample is obtained, including: in, This is the membership vector of the meteorological model; Determine the first based on the principle of maximum membership. The corresponding weather pattern for the day: in, For the first Weather patterns for the day.

4. The clean energy output prediction method based on meteorological pattern recognition and feature bias correction as described in claim 3, characterized in that, For each meteorological model, meteorological variables corresponding to each model are selected based on mutual information, and a corresponding meteorological feature subspace is constructed. Based on the meteorological feature subspace and feature projection, the corresponding input vector is obtained, including: Meteorological samples are divided according to meteorological models. The subsample set includes: in, This indicates that it belongs to a weather model. The sample day index set; In meteorological models The following is a construction of mutual information indicators between meteorological variables and clean energy output, including: in, Indicates weather model Next Random variables corresponding to meteorological variables This represents the random variable of clean energy output under meteorological models. For the first Meteorological variables in meteorological models The correlation strength between the output of clean energy and meteorological characteristics was analyzed and used as a screening indicator. For mutual information; The definition of mutual information includes: in, Let be the joint probability density function. and These are the marginal probability density functions; In meteorological models Below, mutual information indicators Sort the data and select meteorological variables that are greater than the preset mutual information value to construct the first... A subset of meteorological features under a meteorological model; in, For the first A subset of meteorological features from a meteorological model. For the first The number of effective meteorological variables selected from each meteorological model In the first Under the meteorological model, the selected [number]th [item] after mutual information sorting The original variable index number of each meteorological variable. For the first Under the weather model, the first The mutual information value between a meteorological variable and the forecast target; Meteorological variables corresponding to mutual information values ​​greater than a preset value are combined to obtain a meteorological feature subspace, including: in, No. The first weather model Tian Shike The meteorological feature subspace, For the number Meteorological variables in the first Tian Shike The observed values, A subset of meteorological features The original variable index number corresponding to each meteorological variable in the data; Definition of the first The characteristic projection operator of each meteorological model is: The characteristic projection operator is used to map the entire meteorological variable space to the first... The meteorological feature subspace corresponding to the model; In the Tian Shike The feature projection results are as follows: in, in, For the first Tian Shike The complete set of meteorological variables vectors, For transpose, Meteorological elements In the Heavenly The observed value at time; in, Indicates weather model The corresponding meteorological feature subspace vector.

5. The clean energy output prediction method based on meteorological pattern recognition and feature bias correction as described in claim 4, characterized in that, For each meteorological model, a corresponding clean energy output prediction model is constructed based on the meteorological feature subspace corresponding to each model. Furthermore, a set of clean energy output prediction models is built using a Bagging ensemble strategy, including: Based on feature projection operator It belongs to the meteorological model. The samples are projected onto the corresponding meteorological feature subspace to construct the model training input vector; in, In the first Under the first weather pattern, the first Day, time The meteorological feature projection vector is the model input vector; No. The training sample set for each weather model includes: in, For the first Training sample set for each weather model For the first The sky The actual output of clean energy at all times; Definition of the first The corresponding prediction models for each weather pattern are as follows: in, For the first Under the weather model, the first Tian Shike The predicted output For the first Clean energy output prediction model function corresponding to each meteorological model; Based on the gradient boosting regression model, the prediction model is updated, including: in, For the total number of regression trees, For the first In mode of the first A regression tree; Construct the objective function corresponding to the prediction model, including: in, The objective function corresponding to the prediction model, Let the prediction error loss function be... For the first The complexity regularization term for trees; in, This is the actual power. To predict power; In the In each iteration, the first and second derivatives of the loss function with respect to the predicted values ​​are introduced, including: in, For the first-order gradient, It is a second-order gradient. The sign of the partial derivative. For the first Each sample at time... The actual power generation capacity of clean energy This represents the prediction power of the current model for this sample at the end of the previous training round. For pattern training sample set Perform multiple resampling operations to construct Individual Training Set And train a prediction sub-model on each sub-training set. It integrates the outputs of multiple sub-models, including: in, In the first k Under the weather model, for the first The predicted clean energy output value at time t, output by the Bagging integrated prediction model; A set of clean energy output prediction models corresponding one-to-one with each meteorological model was obtained, including: in, A set of models for predicting clean energy output.

6. The clean energy output prediction method based on meteorological pattern recognition and feature bias correction as described in claim 5, characterized in that, After bias correction of the meteorological data to be predicted, the corrected meteorological data is projected onto the meteorological feature subspace, and the corresponding clean energy output prediction model set is called to obtain the clean energy output prediction results, including: For each meteorological variable, a corresponding meteorological bias is constructed, including: in, For the first j The meteorological variable in the first Weather deviation at time t For the first j The meteorological variable in the first Tian Shike t The true value, For the first j The meteorological variable in the first Tian Shike t The forecast value; No. The definition of the mean deviation of the meteorological model corresponding to each meteorological variable includes: in, For the first The mean deviation of the meteorological model corresponding to each meteorological variable For the first time in history d The first day of the sample j Meteorological deviations of individual meteorological variables; Determine the set of mean deviations for each meteorological model, including: in, This is the set of mean deviations under various meteorological models. For the first The mean deviation of the meteorological model corresponding to each meteorological variable dimension; For the predicted date The Middle At what time, the corresponding forecast meteorological variable vector is: in, For the first The first predicted day The column vector of forecast meteorological variables at any given time. For the first The meteorological variable in the first The predicted day Forecast value for the time; Obtain the daily-scale meteorological feature vector corresponding to the prediction date. Based on the FCM meteorological model that has been trained in historical periods, the predicted date is calculated. Membership degree under various meteorological models: in, The first known historical stage Each meteorological model center, The first known historical stage Each meteorological model center, For prediction date In the Membership degree under a weather pattern; The prediction date is determined using the maximum membership principle. Belonging meteorological model: Determine the forecast date The meteorological model corresponding to the meteorological variable vector is used, and bias correction is performed, including: in, For the first The meteorological variable in the first Predicted days The predicted value after bias correction at time, For the first The meteorological variable in the first Predicted days Forecast value for the time, The first one obtained from historical sample statistics The first weather model The mean deviation of each variable; The corrected forecast period's meteorological variable vectors are projected onto the meteorological feature subspace of the meteorological model, including: in, In the first Predicted days At that moment, the corrected meteorological variable vector, In the first Predicted days At this moment, the first The values ​​of the meteorological variables after bias correction This is the joint meteorological feature vector used as input to the prediction model; Get the first Predicted days The clean energy output forecast results for each moment include: in, For the first Predicted days Real-time clean energy output forecast results To match the predicted date Belonging meteorological model Corresponding clean energy output prediction model; Based on the Bagging integration strategy, the clean energy output forecast results are integrated and updated, including: in, For weather models Next Predicted days Real-time clean energy output forecast results For weather models Based on the first A prediction sub-model constructed from a resampled training set.

7. The clean energy output prediction method based on meteorological pattern recognition and feature bias correction as described in claim 1, characterized in that, Acquire and preprocess historical power output sequences of clean energy power plants within the target area and corresponding meteorological data for the corresponding time periods, including: The historical output sequences of clean energy power plants in the target area and the meteorological data for the corresponding time periods are further time-stamped, abnormal data are removed, and linear interpolation is performed.

8. A clean energy output prediction device based on meteorological pattern recognition and feature deviation correction, characterized in that, include: The historical data processing module is used to acquire and preprocess the historical output sequence of clean energy power stations in the target area and the meteorological data of the corresponding time period; The meteorological sample set module is used to construct a meteorological sample set based on preprocessed meteorological data, and to construct a daily-scale meteorological feature vector based on the meteorological sample set; The clustering module is used to cluster daily meteorological feature vectors based on fuzzy clustering, obtain the membership degree of each meteorological sample under different meteorological models, and determine the corresponding meteorological model. The mutual information filtering module is used to filter the meteorological variables corresponding to each meteorological model based on mutual information, construct the corresponding meteorological feature subspace, and obtain the corresponding input vector based on the meteorological feature subspace and feature projection. The prediction model set module is used to construct corresponding clean energy output prediction models for each meteorological model based on the meteorological feature subspace corresponding to each meteorological model, and to construct a set of clean energy output prediction models through the Bagging integration strategy. The prediction results module is used to correct the deviation of the meteorological data to be predicted, project the corrected meteorological data to the meteorological feature subspace, and call the corresponding clean energy output prediction model set to obtain the clean energy output prediction results.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute a clean energy output prediction method based on meteorological pattern recognition and feature deviation correction as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the non-transitory computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform a clean energy output prediction method based on meteorological pattern recognition and feature deviation correction as described in any one of claims 1 to 7.

Images