Method, device and medium for screening key influencing factors of photovoltaic prediction

Abstract

The application discloses a photovoltaic prediction key influence factor screening method, device, equipment and medium, comprising: obtaining candidate influence factor data and corresponding photovoltaic power prediction accuracy index data in a historical period; based on the photovoltaic power prediction accuracy index data, determining the prediction accuracy comprehensive index of each historical sample; based on the candidate influence factor data, performing principal component analysis to obtain a loading matrix and an association strength parameter of each principal component; based on the loading matrix and the association strength parameter, determining the comprehensive influence weight score of each candidate influence factor; based on the comprehensive influence weight score of each candidate influence factor, screening the key influence factor from the candidate influence factor. According to the loading matrix, the comprehensive influence weight score of each influence factor is determined, and the screening of the influence factor is performed, which can help the power station operation and maintenance personnel to quickly identify and focus on the environmental and equipment factors that have the greatest impact on power generation prediction, thereby effectively improving the intelligent management level of the power station.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, equipment, and medium for screening key influencing factors in photovoltaic forecasting. Background Technology

[0002] The accuracy of photovoltaic power prediction is affected by a variety of meteorological and conditional factors, including but not limited to: solar irradiance, ambient temperature, photovoltaic panel temperature, humidity, wind speed, air pressure, historical power, and module degradation coefficient. In constructing the prediction model, selecting appropriate features is crucial for improving model performance and reducing computational complexity.

[0003] However, while existing prediction methods can identify factors related to prediction errors through correlation analysis or dimensionality reduction techniques, they struggle to explain from a physical perspective why these factors are critical, as well as the causal relationships and influence transmission paths between them. Summary of the Invention

[0004] This invention provides a method, apparatus, equipment, and medium for screening key influencing factors in photovoltaic forecasting. By performing dimensionality reduction and principal component extraction on candidate influencing factors through principal component analysis, the corresponding loading matrix is ​​determined. Based on the loading matrix, the comprehensive influence weight score of each influencing factor is determined and the influencing factors are screened. This can provide precise guidance for researchers to perform feature engineering and model optimization, and can also help power plant operation and maintenance personnel quickly identify and focus on the environmental and equipment factors that have the greatest impact on power generation forecasting, thereby effectively improving the intelligent management level of power plants.

[0005] According to one aspect of the present invention, a method for screening key influencing factors in photovoltaic forecasting is provided, characterized in that it includes:

[0006] Acquire candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within historical time periods;

[0007] Based on the photovoltaic power prediction accuracy index data, the comprehensive index of prediction accuracy for each historical sample is determined.

[0008] Principal component analysis was performed based on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component;

[0009] Based on the load matrix and correlation strength parameters, the comprehensive influence weight score of each candidate influencing factor is determined;

[0010] Based on the comprehensive influence weight score of each candidate influencing factor, key influencing factors are selected from the candidate influencing factors.

[0011] According to another aspect of the present invention, a screening device for key influencing factors in photovoltaic prediction is provided, characterized in that it comprises:

[0012] The data acquisition module is used to acquire candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within historical time periods;

[0013] The accuracy index determination module is used to determine the comprehensive accuracy index of prediction for each historical sample based on photovoltaic power prediction accuracy index data.

[0014] The correlation strength parameter determination module is used to perform principal component analysis based on candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component;

[0015] The comprehensive influence weight score determination module is used to determine the comprehensive influence weight score of each candidate influencing factor based on the load matrix and correlation strength parameters.

[0016] The influencing factor screening module is used to screen out key influencing factors from the candidate influencing factors based on the comprehensive influence weight score of each candidate influencing factor.

[0017] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0018] At least one processor; and

[0019] A memory that is communicatively connected to at least one processor; wherein,

[0020] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to perform the method for screening key influencing factors for photovoltaic prediction according to any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions for causing a processor to execute a method for screening key influencing factors for photovoltaic prediction according to any embodiment of the present invention.

[0022] The technical solution of this invention involves acquiring candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within a historical time period; determining the comprehensive prediction accuracy index for each historical sample based on the photovoltaic power prediction accuracy index data; performing principal component analysis based on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component; determining the comprehensive influence weight score of each candidate influencing factor based on the loading matrix and the correlation strength parameters; and selecting key influencing factors from the candidate influencing factors based on the comprehensive influence weight scores of each candidate influencing factor. Based on the above technical solution, principal component analysis is used to reduce the dimensionality of candidate influencing factors and extract principal components to determine the corresponding loading matrix. The comprehensive influence weight scores of each influencing factor are then determined based on the loading matrix, and influencing factors are selected. This provides precise guidance for researchers in feature engineering and model optimization, and also helps power plant operation and maintenance personnel quickly identify and focus on the environmental and equipment factors that have the greatest impact on power generation prediction, thereby effectively improving the intelligent management level of the power plant.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a method for screening key influencing factors in photovoltaic prediction provided by an embodiment of the present invention;

[0026] Figure 2 This is a flowchart of a method for screening key influencing factors in photovoltaic prediction provided by an embodiment of the present invention;

[0027] Figure 3 A schematic diagram of a screening device for key influencing factors in photovoltaic prediction provided in an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] Figure 1 This is a flowchart illustrating a method for screening key influencing factors in photovoltaic power prediction, provided by an embodiment of the present invention. This embodiment is applicable to situations where, from multiple candidate factors that may affect the accuracy of photovoltaic power prediction, the truly critical influencing factors are objectively, quantitatively, and automatically identified, and their degree of influence is precisely ranked. This method can be executed by a device for screening key influencing factors in photovoltaic power prediction, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method specifically includes the following steps:

[0032] S110. Obtain candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within the historical time period.

[0033] The historical time period can be understood as a continuous collection of complete time-series samples over multiple days from the Wield photovoltaic power plant. Candidate influencing factor data can be meteorological and equipment time-series data that may interfere with photovoltaic power prediction, including raw time-series values ​​across multiple dimensions such as total radiance, direct radiance, diffuse irradiance, ambient temperature, photovoltaic panel temperature, relative humidity, wind speed, air pressure, historical photovoltaic power, and module attenuation coefficient. Photovoltaic power prediction accuracy index data can be understood as a set of multi-dimensional error evaluation indicators calculated after extrapolating photovoltaic output using the basic prediction model. The basic prediction model can be any time-series prediction model among LSTM, Random Forest, XGBoost, and CNN, used to generate photovoltaic power prediction values ​​to calculate the accuracy index. Accuracy indicators include cost-based evaluation indicators such as root mean square error, mean absolute error, and mean absolute percentage error, used to quantify the magnitude of deviation in photovoltaic prediction under a single historical sample.

[0034] Specifically, the process involves acquiring candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within historical time periods. For example, this can be done by connecting to the database system of photovoltaic power plants, obtaining candidate influencing factor data corresponding to the photovoltaic power plants, and then standardizing and normalizing the collected data to obtain the corresponding influencing factor matrix. Finally, the photovoltaic power prediction accuracy index data corresponding to each influencing factor can be determined based on the collected candidate influencing factor data.

[0035] Based on the above technical solution, the acquisition of candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within a historical time period includes: acquiring candidate influencing factor data for at least two time samples within a historical time period; inputting the candidate influencing factor data for at least two time samples into a basic prediction model to predict photovoltaic power, obtaining predicted power values ​​for at least two time samples; determining at least two prediction accuracy index values ​​for at least two time samples based on the predicted power values ​​and actual power values ​​of at least two time samples; arranging the candidate influencing factor data for at least two time samples in chronological order to generate a candidate influencing factor matrix; and arranging the at least two prediction accuracy index values ​​for at least two time samples in chronological order to generate a prediction accuracy index matrix.

[0036] The candidate influencing factor data includes irradiance data, temperature data, humidity data, wind speed data, air pressure data, and historical power data. Prediction accuracy indicators include at least two of the following: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination. Irradiance data is the time-series value of solar radiation intensity received by the Earth's surface, used to determine the basic energy input for photovoltaic (PV) conversion. Temperature data includes ambient temperature and PV panel temperature, used to characterize the inhibitory effect of temperature on PV module conversion efficiency. Humidity data is relative humidity, used to reflect the degree to which water vapor and haze weaken solar radiation transmittance. Wind speed data is the real-time air velocity at the site. Air pressure data is the atmospheric pressure at the site. Historical power data is the time-series value of actual PV power output at past times. The basic prediction model is any one of the time-series prediction models: LSTM, Random Forest, XGBoost, or CNN, used to output the predicted PV power for a single sample. The actual power value is the actual power output of the site collected by the PV inverter. The predicted power value is the estimated PV power output obtained by the basic prediction model based on the candidate factors. The root mean square error (RMSE) is an error evaluation metric that amplifies significant prediction deviations. The mean absolute error (MAE) is an averaged absolute deviation metric used to visually represent the overall prediction error magnitude. The mean absolute percentage error (MASE) is a dimensionless relative error metric used to compare prediction levels across time periods and output ranges. The coefficient of determination (COD) is a goodness-of-fit metric used to quantify the overall explanatory power of candidate factors for photovoltaic output fluctuations. The candidate influencing factor matrix is ​​a standardized two-dimensional dataset constructed with time samples as rows and various candidate influencing factors as columns. The prediction accuracy metric matrix is ​​a two-dimensional dataset constructed with time samples as rows and various accuracy evaluation metrics as columns.

[0037] Specifically, a continuous historical time series spanning multiple days is defined as the data collection interval. The system connects to the meteorological monitoring terminal of the photovoltaic power plant and the inverter data acquisition system to batch read the raw time series data of irradiance, temperature, humidity, wind speed, air pressure, and historical power corresponding to each time sample. Basic cleaning operations are performed, including filling in missing values ​​and removing anomalous jump data. The cleaned single-sample candidate factor features are then fed into a pre-set basic prediction model, which outputs the photovoltaic predicted power value for the corresponding time sample. The actual photovoltaic power value stored in the inverter at the same time is retrieved, and the predicted value is paired with the actual value. Accuracy indicators such as root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination are calculated sequentially to form a complete accuracy evaluation result for each single sample. Following chronological order, the multi-dimensional candidate factors of all samples are arranged into row vectors and concatenated to construct a candidate influencing factor matrix. The four types of accuracy indicators for each sample are arranged sequentially to construct a prediction accuracy indicator matrix. It should be noted that after both matrices are constructed, Z-score standardization is uniformly performed to eliminate dimensional differences between different features and error indicators.

[0038] It should be noted that the basic prediction model can be any of LSTM, Random Forest, XGBoost, or CNN. Taking LSTM as an example, this model uses the standardized candidate influencing factor time series feature vectors composed of irradiance, temperature, humidity, wind speed, air pressure, and historical power corresponding to each historical time sample as a unified input. The model first receives the standardized multi-dimensional time series features through the input layer, and then feeds them into multiple LSTM hidden layers with forget gates, input gates, and output gates to mine the long-term and short-term time series dependencies between photovoltaic power output and various meteorological and power features. After compressing the redundant feature dimensions through a fully connected transition layer, it is connected to the linearly activated output layer, and finally outputs the photovoltaic power prediction value of the corresponding time sample.

[0039] The technical solution of this invention comprehensively collects multi-dimensional meteorological and historical power candidate factors, generates predicted power matching each time sample based on a unified basic prediction model, quantifies prediction deviation with multiple types of complementary accuracy indicators, constructs an aligned candidate factor matrix and accuracy indicator matrix according to time sequence, and completes dimensionless standardization preprocessing. This solves the defects of traditional influencing factor analysis, which only uses a single error indicator, has mismatch between the time sequence of features and evaluation samples, and suffers from dimensional interference in calculation. It ensures that the subsequent identification of key influencing factors and the ranking of importance are consistent with the physical mechanism of photovoltaic photoelectric conversion, and improves the objectivity and interpretability of feature selection.

[0040] S120. Based on the photovoltaic power prediction accuracy index data, determine the comprehensive index of prediction accuracy for each historical sample.

[0041] Historical samples can be complete time-series data units collected from photovoltaic power plants at a single moment, or they can be historical data divided into natural days, with each natural day corresponding to one historical sample. The comprehensive prediction accuracy index can be used to characterize the overall prediction effect of a single sample; the closer the value is to 1, the higher the photovoltaic power prediction accuracy for that sample.

[0042] Specifically, based on the cost-type and benefit-type indicators within the matrix, cost-type indicators such as root mean square error, mean absolute error, and mean absolute percentage error (MASE) – which indicate poorer prediction performance with larger values ​​– undergo positive transformation. Benefit-type indicators such as the coefficient of determination – which indicate better prediction performance with larger values ​​– retain their original standardized values, resulting in a unified positive evaluation matrix. The entropy weight method is then used to objectively calculate the weights for each prediction accuracy indicator, allocating weights based on the information entropy of each indicator's data. This weighted normalized evaluation matrix is ​​then generated. Based on the weighted matrix, the maximum value across all indicator dimensions is extracted as the positive ideal solution, and the minimum value as the negative ideal solution. The Euclidean distance from each historical sample to both the positive and negative ideal solutions is calculated. Finally, the comprehensive prediction accuracy index for the corresponding sample is calculated based on the ratio of these two distances. After calculation, the indexes are stored in chronological order.

[0043] The technical solution of this invention provides a unified positive processing for prediction accuracy indicators of multiple types and different attributes. It uses the entropy weight method to achieve objective weighting of indicators, integrates the multi-dimensional and scattered prediction effect evaluation standards into a unified quantitative indicator that can be directly compared, and provides a single target variable for subsequent calculation of the correlation strength between each principal component and the overall prediction accuracy. This ensures that the evaluation benchmark for ranking the importance of subsequent candidate influencing factors is unified and the results are objective and reliable.

[0044] Based on the above technical solution, and using photovoltaic power prediction accuracy index data, the comprehensive prediction accuracy index for each historical sample is determined. This includes: standardizing the prediction accuracy index data to obtain a standardized accuracy index matrix; positiveizing the standardized accuracy index matrix based on the index type of each prediction accuracy index to obtain a positiveized accuracy index matrix; determining the objective weight of each prediction accuracy index using the entropy weight method based on the positiveized accuracy index matrix; and determining the comprehensive prediction accuracy index for each historical sample using the approximation ideal solution ranking method based on the objective weights and the positiveized accuracy index matrix.

[0045] The indicators include benefit-type indicators and cost-type indicators. Standardization is achieved through Z-score standardization, which eliminates differences in dimensions and numerical ranges between indicators of different precision levels, unifying data distribution characteristics. The standardized precision indicator matrix is ​​a dimensionless two-dimensional matrix of indicators after correction for mean and standard deviation. Benefit-type indicators are those where larger values ​​represent better photovoltaic prediction results, such as the coefficient of determination. Cost-type indicators are those where larger values ​​represent larger prediction deviations and worse prediction results, such as root mean square error, mean absolute error, and mean absolute percentage error. Forwarding can be a transformation operation based on the direction of indicator performance, ensuring that the trend of all indicator values ​​is positively correlated with prediction accuracy. The forwarding precision indicator matrix is ​​a standardized matrix after all indicators have undergone a unified transformation of their performance. The entropy weighting method is an objective weighting algorithm that assigns weights to indicators based on the dispersion of data information. The objective weights can be the proportion of each precision indicator output by the entropy weighting method within the comprehensive evaluation system. The Approximation to Ideal Solution Ranking Method, also known as the TOPSIS multi-attribute decision evaluation method, is used to integrate multi-indicator weighted data to quantify the comprehensive prediction level of a single sample.

[0046] Specifically, based on the prediction accuracy index data, the Z-score standardization formula is used to perform mean-to-zero and standard-deviation-to-one operations for each index, generating a standardized accuracy index matrix. The matrix distinguishes between benefit-type and cost-type indicators. For benefit-type indicators such as the coefficient of determination, standardized values ​​are directly retained. For the three types of error-type cost indicators, a forward transformation is performed by adding a minimum constant and taking the reciprocal. The forward transformation matrix is ​​traversed to calculate the sample proportion and information entropy for each indicator. Based on the information entropy difference, objective weights are assigned to each accuracy index. The objective weights are multiplied by the forward transformation matrix to obtain a weighted normalized matrix. The maximum value of each dimension of the matrix is ​​extracted as the positive ideal solution, and the minimum value as the negative ideal solution. The Euclidean distance between each historical sample and the two types of ideal solutions is calculated, and the prediction accuracy comprehensive index for the corresponding sample is obtained by combining the distance ratio.

[0047] The technical solution of this invention integrates the scattered multi-dimensional prediction error evaluation indicators into a single comparable quantitative index, constructs a unified evaluation benchmark with the overall prediction accuracy as the core, and ensures that the identification and ranking results of key influencing factors in subsequent photovoltaic prediction are objective, stable, and in line with the actual change law of photovoltaic power output.

[0048] S130. Based on the candidate influencing factor data, perform principal component analysis to obtain the loading matrix and the correlation strength parameters of each principal component.

[0049] Principal Component Analysis (PCA) is a dimensionality reduction algorithm used to remove information redundancy among multiple factors and extract mutually independent comprehensive principal components. The loading matrix can be an intermediate core matrix output from the PCA process. The elements in the matrix are the loading coefficients of the original candidate factors on each principal component. The absolute values ​​of the loading coefficients directly characterize the correlation strength between the original factors and their corresponding principal components. The principal components are uncorrelated comprehensive variables obtained through linear transformation of the original multi-dimensional influencing factors. The correlation strength parameter is the Pearson correlation coefficient, calculated between the principal component score sequence and the comprehensive prediction accuracy index, used to quantify the impact of a single set of principal components on the overall accuracy of photovoltaic prediction.

[0050] Specifically, matrix operations are performed on the candidate influencing factor matrix to solve for the covariance matrix. Eigenvalue decomposition is then performed on the covariance matrix to obtain eigenvalues ​​sorted from largest to smallest and their corresponding orthogonal unit eigenvectors. The number of principal components to be retained can be determined based on the screening rule that the cumulative variance contribution rate is not less than 85%. Loading coefficients are calculated by combining the eigenvalues ​​and eigenvectors, generating loading matrices for all original factors and screened principal components. Subsequently, the standardized candidate factor matrix is ​​projected onto the eigenvectors corresponding to each principal component, and the score sequences of all historical samples under each principal component are calculated. Pearson correlation calculations are then performed on each set of principal component score sequences and the time-series prediction accuracy composite index to obtain the correlation coefficient of the corresponding principal component as the correlation strength parameter.

[0051] Based on the above technical solution, principal component analysis is performed on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component. This includes: standardizing the candidate influencing factor data to obtain a standardized influencing factor matrix; performing principal component analysis based on the standardized influencing factor matrix to obtain the loading matrix and eigenvectors; determining the principal component scores of each historical sample on each principal component based on the standardized influencing factor matrix and eigenvectors; and determining the correlation coefficient between each principal component and the prediction accuracy based on the principal component scores of each historical sample on each principal component and the comprehensive index of prediction accuracy of each historical sample, using the correlation coefficient as the correlation strength parameter of each principal component.

[0052] Here, the eigenvectors are orthogonal vectors obtained from the eigenvalue decomposition of the covariance matrix. The loading matrix is ​​generated by operations on the eigenvectors and their corresponding eigenvalues; the absolute values ​​of the loading coefficients within the matrix reflect the correlation strength between the original factors and the principal components. The principal component score is the comprehensive feature value projected onto the principal component basis for a single historical sample. The comprehensive prediction accuracy index is a unified evaluation value for a single sample generated by fusing multiple accuracy indices, characterizing the overall performance of photovoltaic prediction for a single sample. The correlation coefficient is the Pearson correlation coefficient, used to quantify the linear correlation between the principal component score sequence and the comprehensive prediction accuracy index. The correlation strength parameter is the Pearson correlation coefficient corresponding to the principal component, used to characterize the strength of the principal component's influence on photovoltaic prediction accuracy.

[0053] Specifically, the original candidate influencing factor data is read, and Z-score standardization is used to correct the mean and standard deviation feature by feature, generating a standardized influencing factor matrix. Based on the standardized matrix, the covariance matrix is ​​solved, and eigenvalue decomposition is performed on the covariance matrix to obtain ordered eigenvalues ​​and corresponding unit orthogonal eigenvectors. The eigenvalues ​​and eigenvectors are combined to generate a complete loading matrix, and effective principal components are selected based on the cumulative variance contribution rate. Matrix projection operations are performed on the standardized influencing factor matrix and eigenvectors to calculate the principal component scores for all historical samples on each retained principal component, forming a complete principal component score time series matrix. Based on the comprehensive index sequence of prediction accuracy for each sample, the Pearson correlation coefficient is calculated by pairing each column of principal component score sequence with the comprehensive index sequence. The calculated correlation coefficient is defined as the association strength parameter of the corresponding principal component.

[0054] The technical solution of this invention first standardizes the candidate influencing factors to eliminate dimensional interference, obtains the load matrix and eigenvector carrying the original feature association information through PCA decomposition, obtains the principal component scores of each sample based on projection operation, and then solves the association strength parameters of each principal component by combining the prediction accuracy comprehensive index. This solves the defect of traditional PCA that only completes dimensionality reduction and cannot establish a quantitative association between principal components and prediction accuracy. It retains the physical correspondence of the original meteorological and power factors, and ensures that the final photovoltaic prediction key influencing factor identification results have both data objectivity and physical interpretability.

[0055] Based on the above technical solution, principal component analysis is performed on the standardized influencing factor matrix to obtain the loading matrix and eigenvectors. This includes: calculating the covariance matrix of the standardized influencing factor matrix; performing eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and its corresponding eigenvector; determining the number of principal components based on the variance contribution rate of each eigenvalue; calculating the loading coefficients of each original influencing factor on each principal component based on the eigenvalues ​​and eigenvectors corresponding to the number of principal components; and generating the loading matrix based on the loading coefficients.

[0056] The covariance matrix can be a square matrix used to characterize the degree of linear correlation between candidate influencing factors. Eigenvalue decomposition involves linear matrix transformation operations to decompose the covariance matrix, separating eigenvalues ​​representing the information content of the data and eigenvectors representing the direction of transformation. Eigenvalues ​​are numerical values ​​that measure the amount of original data information carried by the corresponding principal component; the larger the eigenvalue, the more effective information the corresponding principal component contains. Eigenvectors are unit orthogonal vectors matching the eigenvalues. The variance contribution rate is the proportion of a single set of eigenvalues ​​to the sum of all eigenvalues, used to quantify the proportion of original data fluctuations explained by a single principal component. The number of principal components is the number of retained principal components selected based on the cumulative variance contribution rate threshold. The loading coefficient is a quantitative indicator of the correlation between the original candidate influencing factor and a single principal component; the higher the absolute value of the loading coefficient, the stronger the contribution of the original factor to that principal component.

[0057] Specifically, the covariance matrix of the standardized influencing factor matrix is ​​calculated, eigenvalue decomposition is performed on the covariance matrix, all eigenvalues ​​and their corresponding unit orthogonal eigenvectors are output, all eigenvalues ​​are sorted in descending order of value, and the variance contribution rate corresponding to each eigenvalue is accumulated. When the accumulated variance contribution rate reaches a preset threshold, the counting stops, and the number of accumulated corresponding eigenvalues ​​is determined as the final number of principal components to be retained. The eigenvalues ​​and eigenvectors corresponding to the filtered principal components are extracted, and the eigenvector components are multiplied by the square root of the corresponding eigenvalue to obtain the loading coefficient of each original influencing factor under each principal component. All loading coefficients are arranged in a regular manner according to the row and column correspondence to generate the loading matrix.

[0058] The technical solution of this invention solves the covariance matrix based on the standardized matrix and completes eigenvalue decomposition. It uses the variance contribution rate to screen effective principal components and generates a loading matrix through joint operation of eigenvalues ​​and eigenvectors. The loading matrix completely retains the contribution degree of each type of candidate influencing factor in each comprehensive principal component. This provides an objective and traceable core calculation data source for subsequent TOPSIS multi-attribute evaluation based on principal component correlation strength parameters and quantification of the influence weight of each original factor on photovoltaic prediction accuracy. It also ensures that the subsequent ranking results of key influencing factors have physical mechanism support.

[0059] S140. Based on the load matrix and correlation strength parameters, determine the comprehensive influence weight score of each candidate influencing factor.

[0060] Among them, the comprehensive influence weight score is a single quantitative value calculated by integrating the load correlation information and the principal component influence weights according to the TOPSIS model. The higher the score, the more critical the influence of the candidate factor on the accuracy of photovoltaic power prediction.

[0061] Specifically, all loading coefficients in the loading matrix are extracted and their absolute values ​​are taken. The positive and negative signs of the loadings are eliminated, and only the correlation strength between the original factors and the principal components is retained. A basic evaluation matrix is ​​constructed with candidate influencing factors as evaluation objects and each principal component as evaluation indicators. The objective evaluation weight of each principal component is calculated based on the correlation strength parameters of all principal components. The proportion of the absolute value of the correlation coefficient of a single principal component to the sum of the absolute values ​​of the correlation coefficients of all principal components is used as the indicator weight of that principal component. The basic evaluation matrix is ​​vector normalized to eliminate dimensional differences, and then multiplied with the principal component weights to obtain a weighted normalized evaluation matrix. The maximum value of each indicator dimension is extracted from the weighted matrix as the positive ideal solution and the minimum value as the negative ideal solution. The Euclidean distance from each candidate influencing factor to the two types of ideal solutions is calculated. Finally, the comprehensive influence weight score corresponding to each candidate factor is obtained by combining the ratio of the two types of distances.

[0062] Based on the above technical solution, the comprehensive influence weight score of each candidate influencing factor is determined based on the load matrix and correlation strength parameters. This includes: taking the absolute value of each load coefficient in the load matrix to obtain the absolute value load matrix; determining the evaluation weight of each principal component based on the correlation strength parameters of each principal component; constructing a multi-attribute evaluation data source for each candidate influencing factor based on the absolute value load matrix and the evaluation weight of each principal component; and evaluating the candidate influencing factor using the approximation ideal solution ranking method based on the multi-attribute evaluation data source to obtain the comprehensive influence weight score of each candidate influencing factor.

[0063] Among them, the loading coefficient is the quantitative value of the correlation between the original factor and the corresponding principal component. The larger the absolute value, the higher the contribution of the original factor to the information of the principal component. The absolute value loading matrix is ​​the matrix obtained by taking the absolute value of all elements of the loading matrix. The correlation strength parameter is the Pearson correlation coefficient between the principal component and the prediction accuracy comprehensive index, reflecting the magnitude of the influence of a single principal component on the photovoltaic prediction accuracy. The evaluation weight of the principal component is the weight value objectively calculated based on the correlation strength parameter, used to distinguish the importance of different principal components in the evaluation system. The multi-attribute evaluation data source is the weighted normalized matrix obtained by fusing the absolute value loading matrix and the principal component evaluation weights. The approximation ideal solution ranking method, i.e., the TOPSIS multi-attribute decision algorithm, is used to fuse multi-dimensional indicators to quantify the comprehensive quality level of a single object. The comprehensive influence weight score is a quantitative index ranging from 0 to 1. The higher the score, the higher the comprehensive influence of the candidate factor on the photovoltaic power prediction accuracy.

[0064] Specifically, the complete load matrix output by the preceding PCA is read, all load coefficients in the matrix are traversed and their absolute values ​​are uniformly taken to generate an absolute value load matrix; the correlation strength parameters corresponding to all principal components are extracted, and the evaluation weight corresponding to each principal component is calculated based on the proportion of the absolute value of each parameter to the sum of the absolute values ​​of all principal component parameters; the absolute value load matrix is ​​subjected to vector normalization to eliminate dimensional differences, and then the normalized matrix is ​​multiplied column by column with the principal component evaluation weights to construct a data source adapted to TOPSIS operation; based on the multi-attribute evaluation data source, the maximum value of each dimension index is extracted as the positive ideal solution and the minimum value is extracted as the negative ideal solution; the Euclidean distance between each type of candidate influencing factor and the two types of ideal solutions is calculated, and the comprehensive influence weight score of the corresponding factor is solved by combining the distance ratio.

[0065] The technical solution of this invention adopts a unified correlation strength evaluation standard by taking the absolute value of the load coefficient, objectively allocates the evaluation weight of each principal component according to the correlation strength parameter of the principal component, and then uses the TOPSIS model to solve the comprehensive influence weight score after merging and generating a standardized multi-attribute evaluation data source. This solves the shortcomings of traditional factor importance analysis, which relies only on single-dimensional correlation and cannot integrate the contributions of multiple principal components. It can simultaneously take into account the correlation degree between the original factors and each principal component, as well as the actual role of the principal components themselves in the prediction accuracy. It quantitatively outputs the comprehensive influence level of each meteorological and power feature, providing an objective quantitative basis for feature selection and feature priority ranking before photovoltaic prediction model training. It eliminates low-contribution redundant features and retains high-weight core features, reducing the computational complexity of the model and simultaneously improving the overall accuracy of photovoltaic power prediction and the physical interpretability of the model.

[0066] Based on the above technical solution, an approximation ideal solution ranking method is used to evaluate the multi-attribute evaluation data source to obtain the comprehensive influence weight score of each candidate influencing factor. This includes: performing vector normalization processing on the multi-attribute evaluation data source to obtain a normalized evaluation matrix, and determining the positive and negative ideal solutions of each evaluation index based on the normalized evaluation matrix; determining the first distance value between each candidate influencing factor and the positive ideal solution and the second distance value between each candidate influencing factor and the negative ideal solution based on the normalized evaluation matrix, the positive ideal solution, and the negative ideal solution; and determining the comprehensive influence weight score of each candidate influencing factor based on the first and second distance values.

[0067] Vector normalization is a standardization operation that eliminates differences in the numerical scales of different principal component indices. The normalized evaluation matrix is ​​the matrix obtained after vector normalization from the multi-attribute evaluation data source. The evaluation indices are the principal components retained through PCA screening, representing multi-dimensional comprehensive information dimensions. The positive ideal solution is the set of maximum values ​​corresponding to each column of principal component indices in the normalized evaluation matrix, representing the ideal factor state that contributes the most to the accuracy of photovoltaic prediction. The negative ideal solution is the set of minimum values ​​corresponding to each column of principal component indices in the normalized evaluation matrix, representing the ideal factor state that contributes the least to the accuracy of photovoltaic prediction. The first distance value is the Euclidean distance from a single candidate influencing factor sample to the positive ideal solution; a larger value indicates that the factor is further away from the optimal contribution state. The second distance value is the Euclidean distance from a single candidate influencing factor sample to the negative ideal solution; a larger value indicates that the factor is far away from the worst contribution state.

[0068] Specifically, the weighted multi-attribute evaluation data source is read, and vector normalization is performed on the principal component indicators in each column of the matrix to generate a standardized evaluation matrix. The standardized evaluation matrix is ​​then traversed column by column, extracting the maximum value combination as the positive ideal solution and the minimum value combination as the negative ideal solution from all the values ​​of each indicator, forming two fixed sets of ideal reference vectors. For each type of candidate influencing factor, such as irradiance, temperature, and humidity, the Euclidean distance between the corresponding row vector of that factor and the positive ideal solution vector is calculated as the first distance value, and the Euclidean distance between that row vector and the negative ideal solution vector is calculated as the second distance value. The second distance value is divided by the sum of the first and second distance values, and the result is the comprehensive influence weight score of the corresponding candidate factor.

[0069] The technical solution of this invention first performs vector normalization on the multi-attribute evaluation data source to eliminate scale interference, then extracts positive and negative ideal solutions as a unified evaluation benchmark, quantifies the distance between each candidate factor and the optimal and worst contribution states through Euclidean distance, and finally calculates the comprehensive influence weight score. It can comprehensively integrate the correlation contributions under multiple principal component dimensions to complete objective scoring. Based on the score, it can directly quantify and sort all candidate factors such as irradiance and temperature from key to secondary, providing an objective quantitative basis for feature simplification and customized feature selection before training photovoltaic prediction models. While reducing the dimension of model input features and reducing computational overhead, it retains the core physical features that play a dominant role in prediction accuracy, taking into account both model running efficiency and photovoltaic power prediction accuracy, and improving the physical interpretability of the feature selection process.

[0070] S150. Based on the comprehensive influence weight score of each candidate influencing factor, key influencing factors are selected from the candidate influencing factors.

[0071] Among them, the key influencing factors are the high-weight original features retained after score screening. These are the input features that are given priority in the training stage of the photovoltaic prediction model and can dominate the pattern of photovoltaic power output change and support the improvement of prediction accuracy.

[0072] Specifically, all candidate influencing factors are first sorted in descending order of their comprehensive influence weight scores to create a complete list of factor importance. Two filtering logics are pre-configured: a fixed score threshold or a fixed number of features to retain. If the threshold filtering logic is used, the sorted factor list is traversed, retaining candidate factors with comprehensive influence weight scores higher than the preset threshold as key influencing factors, and discarding redundant features with scores below the threshold that contribute little to prediction accuracy. If the fixed number filtering logic is used, a specified number of high-scoring factors at the top of the sorted list are directly selected as key influencing factors, discarding low-contribution secondary factors at the bottom. After filtering, the key influencing factors are separately aggregated to form a concise feature set, which will serve as the input sample for the photovoltaic power prediction model.

[0073] The technical solution of this invention involves acquiring candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within a historical time period; determining the comprehensive prediction accuracy index for each historical sample based on the photovoltaic power prediction accuracy index data; performing principal component analysis based on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component; determining the comprehensive influence weight score of each candidate influencing factor based on the loading matrix and the correlation strength parameters; and selecting key influencing factors from the candidate influencing factors based on the comprehensive influence weight scores of each candidate influencing factor. Based on the above technical solution, principal component analysis is used to reduce the dimensionality of candidate influencing factors and extract principal components to determine the corresponding loading matrix. The comprehensive influence weight scores of each influencing factor are then determined based on the loading matrix, and influencing factors are selected. This provides precise guidance for researchers in feature engineering and model optimization, and also helps power plant operation and maintenance personnel quickly identify and focus on the environmental and equipment factors that have the greatest impact on power generation prediction, thereby effectively improving the intelligent management level of the power plant.

[0074] In one possible implementation of the present invention Figure 2 This is a flowchart illustrating a method for screening key influencing factors in photovoltaic forecasting, provided by an embodiment of the present invention. The embodiment further describes the technical solution of this method, such as... Figure 2 As shown, the method includes:

[0075] Step 1: Obtain historical photovoltaic operation data and perform preprocessing.

[0076] Specifically, collect A historical time sample, such as Historical data for 3 days, each sample contains two data matrices:

[0077] ;

[0078] in Indicates the number of samples, such as m days; This indicates the number of candidate influencing factors, including but not limited to: total irradiance, direct irradiance, diffuse irradiance, ambient temperature, photovoltaic panel temperature, relative humidity, wind speed, air pressure, historical power, and module attenuation coefficient.

[0079] First, for each historical sample, a basic prediction model (such as LSTM, Random Forest, XGBoost, CNN, etc.) is used to predict photovoltaic power, obtaining the predicted value. Then, the power output of that sample is calculated. One evaluation indicator:

[0080] ;

[0081] in The evaluation metrics include, but are not limited to, root mean square error (RMSE), mean absolute error (MAE), absolute coefficient, and mean absolute percentage error (MAPE).

[0082] To eliminate the influence of inconsistent dimensions among the indicators, the matrix was modified. and Perform Z-score standardization:

[0083] ;

[0084] in, , . right Matrix processing is similar.

[0085] ;

[0086] After standardization, all indicators are converted into dimensionless values ​​with a mean of 0 and a standard deviation of 1, which facilitates subsequent comprehensive calculations.

[0087] Step 2: Construct a comprehensive index for prediction accuracy.

[0088] The standardized performance evaluation matrix obtained in step 1 This can be viewed as a multi-indicator decision problem, with m historical samples as the evaluation objects and p prediction accuracy indicators as the evaluation indicators. It is worth noting that the evaluation indicators may be divided into cost-based and benefit-based indicators; therefore, the indicators must be positively evaluated before assessment.

[0089] No changes are needed for benefit-oriented indicators. ;

[0090] For cost-related indicators, the reciprocal method can be used for positive transformation, where c is a sufficiently small positive number to avoid division by zero: ;

[0091] To calculate the weight of each indicator, the entropy weight method is used to objectively assign weights to each indicator. First, the indicator weight of the i-th sample under the k-th indicator is calculated: ;

[0092] The entropy value of the kth indicator is: ;

[0093] Then the entropy weight of the k-th index is... ;

[0094] Obtain the weight vector The final weighted normalized matrix is

[0095] Positive ideal solution Composed of the maximum values ​​of each indicator:

[0096] ;

[0097] Negative ideal solution It consists of the minimum values ​​of each indicator:

[0098] ;

[0099] The Euclidean distance from each sample to the positive and negative ideal solutions can be calculated:

[0100] ;

[0101] ;

[0102] Based on the above results, the overall prediction accuracy index for each sample can be calculated:

[0103] ;in A higher value indicates higher prediction accuracy for that sample.

[0104] Step 3: PCA dimensionality reduction and principal component extraction.

[0105] Now, let's analyze the standardized candidate influencing factor matrix from step 1. To perform principal component analysis, the covariance matrix R is first calculated: ;

[0106] in Standardization has been performed. Eigenvalue decomposition is then performed on the covariance matrix R:

[0107] ;

[0108] Obtain eigenvalues and its corresponding unit orthogonal eigenvectors ;

[0109] Calculate the variance contribution rate of each eigenvalue: ;

[0110] The cumulative variance contribution rate of the first k eigenvalues ​​is: Select The minimum value of k is taken as the number of principal components.

[0111] The formula for calculating the loading matrix L (of size n×k) of the first k principal components is:

[0112] ;in This represents the loading coefficient of the j-th original element on the k-th principal component. For the j-th component of the k-th eigenvector, This represents the k-th eigenvalue. The absolute value of the loading coefficient reflects the correlation strength between the j-th original factor and the k-th principal component; the closer this value is to 1, the stronger the correlation.

[0113] Step 4: Identification and ranking of key factors based on PCA-TOPSIS.

[0114] The eigenvectors have been obtained in step 3. , The standardized candidate factor matrix is ​​now presented. Projected to On each principal component, the score matrix of each historical sample is obtained. : That is, the first The sample at the th Scores on each principal component.

[0115] Then calculate the first Principal component score sequence Combined with the prediction accuracy index sequence obtained in step 2 Pearson correlation coefficient between them:

[0116] ; The absolute value reflects the first The strength of the association between each principal component and "prediction accuracy".

[0117] The load matrix obtained in step 3 As a new data source for evaluation, the evaluation object is There are 10 original candidate influencing factors, and the evaluation index is: First, construct the evaluation matrix based on the principal components. :

[0118] ;

[0119] To eliminate the influence of positive and negative factors and consider only the correlation strength, the matrix elements are selected based on the absolute values ​​of the loading coefficients. The weights of the principal components are determined by their correlation strength with the overall prediction accuracy index.

[0120] ;

[0121] That is, the first The weight of each principal component is equal to the absolute value of its correlation coefficient as a percentage of the selected components. The proportion of the sum of the absolute values ​​of the correlation coefficients of the principal components. The evaluation matrix A is obtained by vector normalization. ,

[0122] ;

[0123] Then, the weighted normalized matrix B can be constructed:

[0124] ;

[0125] Since a larger absolute value of the loading coefficient indicates a stronger correlation between the factor and the principal components, the maximum value is taken as the positive ideal solution for all indicators. The minimum value is taken as the negative ideal solution. Calculate the Euclidean distance from each original influencing factor to the positive and negative ideal solutions. This allows us to obtain the relative position of the original influencing factors, i.e., the final score of the influencing weight:

[0126] ;in , The larger the value, the more critical the factor is to the accuracy of photovoltaic forecasts. The score indicates the magnitude of the influence of the original candidate factors on photovoltaic prediction.

[0127] This invention is based entirely on historical data. Both PCA and TOPSIS calculations involve deterministic and repeatable mathematical transformations, thus avoiding the biases inherent in the subjectivity of traditional Delphi methods or expert scoring methods. Secondly, compared to simply using "black box" prediction models, this invention offers superior interpretability, making the previously opaque prediction logic clear and credible. Furthermore, this invention overcomes the limitation of traditional correlation analysis, which can only assess the relationship between a single factor and prediction error at a time. It can simultaneously incorporate dozens of candidate factors for comprehensive consideration and complete the overall ranking within a multi-dimensional evaluation index framework, making the analysis results more closely reflect the complex and ever-changing engineering realities. Finally, the ranking list of influencing factors output by this invention is clear and intuitive, providing precise guidance for researchers in feature engineering and model optimization, and helping power plant operation and maintenance personnel quickly identify and focus on the environmental and equipment factors that have the greatest impact on power generation prediction, thereby effectively improving the intelligent management level of power plants.

[0128] Figure 3 This is a schematic diagram of a device for screening key influencing factors in photovoltaic prediction, provided as an embodiment of the present invention. Figure 3 As shown, the device includes: a data acquisition module 310, an accuracy index determination module 320, a correlation strength parameter determination module 330, a comprehensive influence weight score determination module 340, and an influencing factor screening module 350.

[0129] Data acquisition module 310 is used to acquire candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within a historical time period;

[0130] The accuracy index determination module 320 is used to determine the comprehensive accuracy index of prediction for each historical sample based on the photovoltaic power prediction accuracy index data.

[0131] The correlation strength parameter determination module 330 is used to perform principal component analysis based on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component.

[0132] The comprehensive influence weight score determination module 340 is used to determine the comprehensive influence weight score of each candidate influencing factor based on the load matrix and the correlation strength parameter.

[0133] The influencing factor screening module 350 is used to screen out key influencing factors from the candidate influencing factors based on the comprehensive influence weight score of each candidate influencing factor.

[0134] Based on the above technical solution, the data acquisition module is used to acquire candidate influencing factor data for at least two time samples within a historical time period, wherein the candidate influencing factor data includes irradiance data, temperature data, humidity data, wind speed data, air pressure data, and historical power data; input the candidate influencing factor data for at least two time samples into the basic prediction model to predict photovoltaic power, thereby obtaining predicted power values ​​for at least two time samples; based on the predicted power values ​​and actual power values ​​for at least two time samples, determine at least two prediction accuracy index values ​​for at least two time samples, wherein the prediction accuracy index values ​​include at least two of the following: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination; arrange the candidate influencing factor data for at least two time samples in chronological order to generate a candidate influencing factor matrix, and arrange the at least two prediction accuracy index values ​​for at least two time samples in chronological order to generate a prediction accuracy index matrix.

[0135] Based on the above technical solution, the accuracy index determination module is used to standardize the prediction accuracy index data to obtain a standardized accuracy index matrix; based on the index type of each prediction accuracy index, the standardized accuracy index matrix is ​​positively oriented to obtain a positively oriented accuracy index matrix, wherein the index types include benefit-type indicators and cost-type indicators; based on the positively oriented accuracy index matrix, the objective weight of each prediction accuracy index is determined using the entropy weight method; based on the objective weight and the positively oriented accuracy index matrix, the comprehensive prediction accuracy index of each historical sample is determined using the approximation ideal solution ranking method.

[0136] Based on the above technical solution, the correlation strength parameter determination module is used to standardize the candidate influencing factor data to obtain a standardized influencing factor matrix; perform principal component analysis based on the standardized influencing factor matrix to obtain a loading matrix and eigenvectors; determine the principal component scores of each historical sample on each principal component based on the standardized influencing factor matrix and the eigenvectors; determine the correlation coefficient between each principal component and the prediction accuracy based on the principal component scores of each historical sample on each principal component and the comprehensive index of the prediction accuracy of each historical sample, and use the correlation coefficient as the correlation strength parameter of each principal component.

[0137] Based on the above technical solution, the correlation strength parameter determination module is used to calculate the covariance matrix of the standardized influencing factor matrix, perform eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and the corresponding eigenvector; determine the number of principal components based on the variance contribution rate of each eigenvalue; calculate the loading coefficient of each original influencing factor on each principal component based on the eigenvalue and eigenvector corresponding to the number of principal components; and generate the loading matrix according to the loading coefficient.

[0138] Based on the above technical solution, the comprehensive influence weight score determination module is used to take the absolute value of each load coefficient in the load matrix to obtain an absolute value load matrix; determine the evaluation weight of each principal component based on the correlation strength parameter of each principal component; construct a multi-attribute evaluation data source for each candidate influencing factor based on the absolute value load matrix and the evaluation weight of each principal component; and perform an approximation ideal solution ranking method evaluation based on the multi-attribute evaluation data source to obtain the comprehensive influence weight score of each candidate influencing factor.

[0139] Based on the above technical solution, the comprehensive influence weight score determination module is used to perform vector normalization processing on the multi-attribute evaluation data source to obtain a normalized evaluation matrix. Based on the normalized evaluation matrix, the positive ideal solution and the negative ideal solution of each evaluation index are determined. Based on the normalized evaluation matrix, the positive ideal solution and the negative ideal solution, the first distance value between each candidate influencing factor and the positive ideal solution and the second distance value between each candidate influencing factor and the negative ideal solution are determined. Based on the first distance value and the second distance value of each candidate influencing factor, the comprehensive influence weight score of each candidate influencing factor is determined.

[0140] The technical solution of this invention involves acquiring candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within a historical time period; determining the comprehensive prediction accuracy index for each historical sample based on the photovoltaic power prediction accuracy index data; performing principal component analysis based on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component; determining the comprehensive influence weight score of each candidate influencing factor based on the loading matrix and the correlation strength parameters; and selecting key influencing factors from the candidate influencing factors based on the comprehensive influence weight scores of each candidate influencing factor. Based on the above technical solution, principal component analysis is used to reduce the dimensionality of candidate influencing factors and extract principal components to determine the corresponding loading matrix. The comprehensive influence weight scores of each influencing factor are then determined based on the loading matrix, and influencing factors are selected. This provides precise guidance for researchers in feature engineering and model optimization, and also helps power plant operation and maintenance personnel quickly identify and focus on the environmental and equipment factors that have the greatest impact on power generation prediction, thereby effectively improving the intelligent management level of the power plant.

[0141] The photovoltaic prediction key influencing factor screening device provided in the embodiments of the present invention can execute the photovoltaic prediction key influencing factor screening method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0142] Figure 4 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0143] like Figure 4As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0144] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0145] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the screening method for key influencing factors in photovoltaic forecasting.

[0146] In some embodiments, the method for screening key influencing factors for photovoltaic forecasting can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for screening key influencing factors for photovoltaic forecasting described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the method for screening key influencing factors for photovoltaic forecasting by any other suitable means (e.g., by means of firmware).

[0147] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0148] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0149] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0150] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0151] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0152] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0153] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0154] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for screening key influencing factors in photovoltaic forecasting, characterized in that, include: Acquire candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within historical time periods; Based on the photovoltaic power prediction accuracy index data, a comprehensive index of prediction accuracy for each historical sample is determined. Principal component analysis was performed based on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component; Based on the load matrix and the correlation strength parameter, the comprehensive influence weight score of each candidate influencing factor is determined; Based on the comprehensive influence weight score of each candidate influencing factor, key influencing factors are selected from the candidate influencing factors.

2. The method according to claim 1, characterized in that, The acquisition of candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within historical time periods includes: Acquire candidate influencing factor data for at least two time samples within a historical time period, wherein the candidate influencing factor data includes irradiance data, temperature data, humidity data, wind speed data, air pressure data, and historical power data; Input candidate influencing factor data from at least two time samples into the basic prediction model to predict photovoltaic power, and obtain the predicted power values ​​for at least two time samples. Based on the predicted power value and the actual power value of at least two time samples, at least two prediction accuracy index values ​​are determined for at least two time samples, wherein the prediction accuracy index values ​​include at least two of the following: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination. Arrange the candidate influencing factor data of at least two time samples in chronological order to generate a candidate influencing factor matrix, and arrange the at least two prediction accuracy index values ​​of at least two time samples in chronological order to generate a prediction accuracy index matrix.

3. The method according to claim 1, characterized in that, The determination of the comprehensive prediction accuracy index for each historical sample based on the photovoltaic power prediction accuracy index data includes: The prediction accuracy index data is standardized to obtain a standardized accuracy index matrix. Based on the index type of each prediction accuracy index, the standardized accuracy index matrix is ​​positively processed to obtain a positively processed accuracy index matrix, wherein the index type includes benefit-type indicators and cost-type indicators. Based on the positive accuracy index matrix, the objective weights of each prediction accuracy index are determined using the entropy weight method. Based on the objective weights and the positive accuracy index matrix, the comprehensive index of prediction accuracy for each historical sample is determined using the approximation ideal solution ranking method.

4. The method according to claim 1, characterized in that, The principal component analysis based on the candidate influencing factor data yields the loading matrix and the correlation strength parameters of each principal component, including: The candidate influencing factor data are standardized to obtain a standardized influencing factor matrix; Principal component analysis was performed based on the standardized influencing factor matrix to obtain the loading matrix and eigenvectors. Based on the standardized influencing factor matrix and the eigenvector, the principal component scores of each historical sample on each principal component are determined. Based on the principal component scores of each historical sample on each principal component and the comprehensive index of prediction accuracy of each historical sample, the correlation coefficient between each principal component and the prediction accuracy is determined, and the correlation coefficient is used as the correlation strength parameter of each principal component.

5. The method according to claim 4, characterized in that, Principal component analysis is performed based on the standardized influencing factor matrix to obtain the loading matrix and eigenvectors, including: Calculate the covariance matrix of the standardized influencing factor matrix, and perform eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and the corresponding eigenvector. Based on the variance contribution rate of each eigenvalue, the number of principal components is determined. Based on the eigenvalues ​​and eigenvectors corresponding to the number of principal components, the loading coefficients of each original influencing factor on each principal component are calculated, and the loading matrix is ​​generated based on the loading coefficients.

6. The method according to claim 4, characterized in that, Based on the load matrix and the correlation strength parameter, the comprehensive influence weight score of each candidate influencing factor is determined, including: The absolute value of each load coefficient in the load matrix is ​​taken to obtain the absolute value load matrix. Based on the correlation strength parameters of each principal component, the evaluation weight of each principal component is determined. Based on the absolute value loading matrix and the evaluation weight of each principal component, a multi-attribute evaluation data source for each candidate influencing factor is constructed. Based on the multi-attribute evaluation data source, the approximation ideal solution ranking method is used for evaluation to obtain the comprehensive influence weight score of each candidate influencing factor.

7. The method according to claim 6, characterized in that, The evaluation based on the multi-attribute evaluation data source, using the approximation ideal solution ranking method, yields the comprehensive influence weight score for each candidate influencing factor, including: The multi-attribute evaluation data source is subjected to vector normalization processing to obtain a normalized evaluation matrix, and based on the normalized evaluation matrix, the positive ideal solution and negative ideal solution of each evaluation index are determined. Based on the normalized evaluation matrix, the positive ideal solution, and the negative ideal solution, determine the first distance value between each candidate influencing factor and the positive ideal solution, and the second distance value between each candidate influencing factor and the negative ideal solution; Based on the first distance value and the second distance value of each candidate influencing factor, the comprehensive influence weight score of each candidate influencing factor is determined.

8. A device for screening key influencing factors in photovoltaic forecasting, characterized in that, include: The data acquisition module is used to acquire candidate influencing factor data and corresponding photovoltaic power prediction accuracy index data within historical time periods; The accuracy index determination module is used to determine the comprehensive accuracy index of prediction for each historical sample based on the photovoltaic power prediction accuracy index data. The correlation strength parameter determination module is used to perform principal component analysis based on the candidate influencing factor data to obtain the loading matrix and the correlation strength parameters of each principal component; The comprehensive influence weight score determination module is used to determine the comprehensive influence weight score of each candidate influencing factor based on the load matrix and the correlation strength parameter. The influencing factor screening module is used to screen out key influencing factors from the candidate influencing factors based on the comprehensive influence weight score of each candidate influencing factor.

9. An electronic device, the electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the screening method for key influencing factors in photovoltaic forecasting as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the screening method for key influencing factors in photovoltaic prediction as described in any one of claims 1-7.

Images