Photovoltaic output short-term prediction method, system and device considering spatial correlation and seasonal characteristics and medium

Abstract

The invention relates to the technical field of photovoltaic output prediction, and discloses a photovoltaic output short-term prediction method, system and device considering spatial correlation and seasonal characteristics, and a medium, and the method comprises the steps: obtaining meteorological data and output sequence data of a target photovoltaic power station; performing correlation analysis to form a main meteorological factor set; collecting output sequence data of adjacent photovoltaic power stations, calculating correlation coefficients of output sequences of the target power station and the adjacent photovoltaic power stations, and screening out a reference photovoltaic power station; jointly constructing a multi-dimensional data set based on the output data of the reference photovoltaic power station and the output related data of the target power station; the method comprises the steps of obtaining a plurality of seasonal sub-sample sets, classifying the seasonal sub-sample sets to obtain seasonal sub-sample sets, constructing neural network prediction models for the seasonal sub-sample sets, obtaining an optimal prediction model corresponding to each season through parameter optimization, and outputting a photovoltaic output prediction result. Through coupling the spatial correlation and seasonal characteristics of multiple stations, the prediction constraint driven by historical data of a single station is broken through, and the prediction precision under complex meteorological conditions is improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power output prediction technology, and in particular to a method, system, device and medium for short-term photovoltaic power output prediction that takes into account spatial correlation and seasonal characteristics. Background Technology

[0002] With the continuous growth of energy consumption and the increasing depletion of fossil fuels, my country's energy structure is gradually being adjusted, and the development of renewable and clean energy sources such as photovoltaic and wind power is gaining popularity. However, due to the significant intermittency and randomness of photovoltaic power generation, and the dispersed and multi-point grid connection of large-scale photovoltaic power plants, the power grid faces a series of problems such as harmonics, stable operation, and power quality. Current short-term photovoltaic output forecasts are still not accurate enough, thus urgently requiring more accurate short-term photovoltaic output forecasting methods to ensure the safe and economical operation of the power system and the coordinated utilization of power resources.

[0003] Current short-term photovoltaic (PV) output forecasting technologies mainly revolve around two major systems: indirect forecasting and direct forecasting. Indirect forecasting methods construct a physical mapping model between meteorological elements and PV output, combining numerical weather prediction with historical meteorological data to achieve forecasting. Its core advantage lies in the interpretability of the physical mechanism, but it suffers from inherent drawbacks such as high modeling complexity and high computational resource requirements. Direct forecasting relies on data-driven technologies, using historical output data and meteorological parameters to train statistical models, and improving forecast stability through techniques such as data dimensionality reduction and feature reconstruction. Common direct forecasting methods include time series analysis (represented by the Autoregressive Moving Average (ARIMA) model), grey theory, multiple linear regression, and various artificial intelligence forecasting methods, among which artificial neural networks are widely used.

[0004] In the research and practice of this method, some limitations of existing technologies in photovoltaic power prediction were found, which can be summarized in the following two aspects: First, insufficient utilization of spatial correlation. Traditional methods mostly focus on the historical data of the target power plant itself and NWP meteorological parameters, ignoring the output correlation characteristics of neighboring plants under similar meteorological conditions. Considering only the time dimension, it is difficult to further improve the prediction effect. Second, lack of seasonal adaptability. Photovoltaic output is significantly affected by the seasonal changes in solar altitude angle and irradiance, but existing prediction models mostly use globally uniform parameters and do not build differentiated prediction models for different seasons. These defects lead to insufficient prediction accuracy of traditional methods in complex meteorological conditions and cross-seasonal scenarios, restricting the level of precision in the coordinated dispatch of photovoltaic power plants and the grid. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a method and system for short-term photovoltaic output prediction that considers spatial correlation and seasonal characteristics to address the current problems of insufficient utilization of spatial correlation, difficulty in further improving prediction accuracy when considering only the time dimension, and insufficient prediction accuracy in complex meteorological conditions and cross-seasonal scenarios.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a short-term photovoltaic power output prediction method that considers spatial correlation and seasonal characteristics, comprising:

[0009] Acquire meteorological data and power output sequence data of the target photovoltaic power station;

[0010] The correlation between meteorological data and the output power of photovoltaic power plants is analyzed, and meteorological factors with absolute values ​​of correlation coefficients greater than preset thresholds are selected to form the main meteorological factor set.

[0011] Collect power output sequence data from nearby photovoltaic power plants, calculate the correlation coefficient between the power output sequence of the target power plant and the nearby power plants, and select reference photovoltaic power plants based on the correlation coefficient values;

[0012] A multidimensional dataset is constructed based on the power output data of a reference photovoltaic power plant and the relevant power output data of the target power plant.

[0013] The multidimensional dataset is classified according to seasonal characteristics to obtain seasonal sub-sample sets. A neural network prediction model is constructed for each seasonal sub-sample set, and the optimal prediction model for each season is obtained through parameter optimization.

[0014] Based on the season to which the expected prediction period belongs, the corresponding seasonal prediction model is invoked to output the photovoltaic power output prediction results.

[0015] As a preferred embodiment of the short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics described in this invention, the following steps are taken: The correlation between meteorological data and photovoltaic power plant output power is analyzed, and meteorological factors with correlation coefficients exceeding a preset threshold are selected to form a set of main meteorological factors, including:

[0016] The correlation coefficient between each meteorological factor and the output sequence was calculated one by one using the Pearson correlation coefficient.

[0017] Set a first threshold to filter major meteorological factors;

[0018] Multicollinearity tests were performed on the selected set of major meteorological factors to generate the final set of major meteorological factors.

[0019] As a preferred embodiment of the short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics described in this invention, the method includes: collecting power output sequence data from neighboring photovoltaic power plants and calculating the correlation coefficient between the power output sequences of the target power plant and neighboring power plants, including:

[0020] Based on the geographical coordinates of the target photovoltaic power station, define the proximity analysis range;

[0021] Obtain metadata from nearby power plants and generate a list of nearby power plants;

[0022] Collect historical power output sequence data from neighboring power plants and align the timestamps of the data from neighboring and target power plants to generate a time-aligned power output sequence dataset;

[0023] The correlation coefficients between the output sequence of the target photovoltaic power plant and the output sequence of each neighboring power plant are calculated.

[0024] As a preferred embodiment of the short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics described in this invention, the method includes: selecting reference photovoltaic power plants based on correlation coefficient values, including:

[0025] A second threshold is set to filter out neighboring power plants that are highly correlated with the output of the target power plant as reference power plants;

[0026] If multiple power stations have similar correlation coefficients, the power station closer to the target power station should be selected as the reference power station.

[0027] The beneficial effects of this preferred technical solution are as follows: Due to their geographical proximity, the output sequences of neighboring photovoltaic power plants are correlated with those of the target power plant. By incorporating the output data of the reference power plant, the deficiencies in the local data of the target power plant can be supplemented, regional photovoltaic output variation patterns can be captured, and a more comprehensive input feature set can be formed. This improves the modeling ability of the neural network model for complex nonlinear relationships and the robustness of predictions.

[0028] As a preferred embodiment of the short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics described in this invention, a multidimensional dataset is constructed based on reference photovoltaic power plant output data and target power plant output-related data, including:

[0029] Align all data timestamps;

[0030] Construct a feature matrix, including: meteorological characteristics of the target power station, historical power output of the target power station, and power output sequence of the reference power station;

[0031] Add time features to capture intraday and weekly periodic patterns;

[0032] A sliding window process is applied to the dataset to generate a multidimensional dataset suitable for prediction.

[0033] As a preferred embodiment of the short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics described in this invention, the method includes: constructing a neural network prediction model for each seasonal sub-sample set, and obtaining the optimal prediction model for each season through parameter optimization, including:

[0034] Data corresponding to the season is extracted from the timestamps of the cube to obtain four subsets;

[0035] A BP neural network prediction model is constructed. The BP neural network adopts a three-layer topology. The number of nodes in the input layer is determined according to the number of feature parameters, the number of nodes in the hidden layer is optimized by trial and error, and the output layer is the photovoltaic power prediction value.

[0036] As a preferred embodiment of the short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics described in this invention, the method includes: calling the corresponding seasonal prediction model according to the season to which the expected prediction period belongs, and outputting the photovoltaic power output prediction result, including:

[0037] Acquire real-time power output data of the target power plant, meteorological forecast data for the predicted period, and real-time power output data of the reference photovoltaic power plant, and construct a multidimensional dataset;

[0038] Input the multidimensional dataset into the optimal prediction model for the corresponding season, and output the photovoltaic power output prediction results.

[0039] Secondly, the present invention provides a short-term photovoltaic power output prediction system that considers spatial correlation and seasonal characteristics, comprising:

[0040] The acquisition module is used to acquire meteorological data and power output sequence data of the target photovoltaic power station;

[0041] The first correlation analysis module is used to analyze the correlation between meteorological data and the output power of photovoltaic power plants, and to select meteorological factors with an absolute value of correlation coefficient greater than a preset threshold to form a set of main meteorological factors.

[0042] The second correlation analysis module is used to collect the output sequence data of neighboring photovoltaic power plants, calculate the correlation coefficient between the output sequence of the target power plant and the neighboring power plants, and select reference photovoltaic power plants based on the correlation coefficient value.

[0043] The dataset construction module is used to jointly construct a multidimensional dataset based on the power output data of a reference photovoltaic power plant and the relevant power output data of the target power plant.

[0044] The model building module is used to classify the multidimensional dataset according to seasonal characteristics to obtain seasonal subsample sets. For each seasonal subsample set, a neural network prediction model is built, and the optimal prediction model for each season is obtained through parameter optimization.

[0045] The prediction output module is used to call the corresponding seasonal prediction model based on the season to which the expected prediction period belongs, and output the photovoltaic power output prediction results.

[0046] Thirdly, the present invention provides a computer device, comprising:

[0047] Memory and processor;

[0048] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of a short-term photovoltaic power output prediction method that takes into account spatial correlation and seasonal characteristics.

[0049] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method for short-term photovoltaic power output prediction considering spatial correlation and seasonal characteristics.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention analyzes the spatial correlation of the output sequence of neighboring photovoltaic power plants by calculating the correlation coefficient, and selects strongly correlated power plant data as model input to expand the data dimension; by synergistically coupling the spatial correlation of multiple power plants with the periodic seasonal characteristics of meteorological parameters, it can break through the constraints of the prediction framework driven by the historical data of a single power plant, thereby improving the prediction accuracy under complex meteorological conditions and different seasonal scenarios, and effectively improving the quality of photovoltaic output prediction. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0052] Figure 1 This is a schematic diagram of the overall process of a short-term photovoltaic power output prediction method that considers spatial correlation and seasonal characteristics, according to an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram comparing the results of the model of the present invention with those of the annual model in a short-term photovoltaic power output prediction method that considers spatial correlation and seasonal characteristics, as described in one embodiment of the present invention. Detailed Implementation

[0054] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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 protection scope of the present invention.

[0055] Example 1, referring to Figure 1 As an embodiment of the present invention, a short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics is provided, comprising:

[0056] S100: Acquire meteorological data and power output sequence data of the target photovoltaic power station;

[0057] S200: Analyze the correlation between meteorological data and the output power of photovoltaic power plants, and select meteorological factors with an absolute value of correlation coefficient greater than a preset threshold to form a set of main meteorological factors;

[0058] S300: Collect power output sequence data of nearby photovoltaic power plants, calculate the correlation coefficient between the power output sequence of the target power plant and the nearby power plants, and select reference photovoltaic power plants based on the correlation coefficient value;

[0059] S400: A multidimensional dataset is constructed based on the output data of a reference photovoltaic power plant and the relevant output data of the target power plant.

[0060] S500: Classify the multidimensional dataset according to seasonal characteristics to obtain seasonal sub-sample sets. Construct a neural network prediction model for each seasonal sub-sample set and obtain the optimal prediction model for each season through parameter optimization.

[0061] S600: Based on the season to which the expected prediction period belongs, call the corresponding seasonal prediction model and output the photovoltaic power output prediction results.

[0062] It should be noted that traditional prediction models are mostly based on data from a single power plant, often ignoring the output correlation of neighboring power plants and failing to utilize regional weather and output patterns, which limits the model's generalization ability. They are also prone to including redundant or low-correlation features, increasing model complexity and reducing prediction efficiency. At the same time, they do not optimize the model for seasonal characteristics, resulting in an inability to capture the fluctuation patterns of photovoltaic output in different seasons and a large prediction error.

[0063] To address the aforementioned main issues, steps S100-S600 incorporate the output data of a reference power station to supplement the deficiencies in the local data of the target power station, capture regional photovoltaic output variation patterns, increase the robustness of the model, and combine seasonality for optimized prediction to improve the accuracy of the prediction data.

[0064] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, a method for short-term prediction of photovoltaic power output considering spatial correlation and seasonal characteristics is provided.

[0065] In this embodiment of the application, step S100 involves acquiring meteorological data and power output sequence data of the target photovoltaic power station;

[0066] In one alternative implementation, the meteorological data of the target photovoltaic power station can be historical and real-time data, including: temperature, irradiance, wind speed, humidity, etc.; and corresponding photovoltaic power output sequence data.

[0067] In another optional implementation, based on the data obtained in the above embodiments, all data needs to be preprocessed, including missing value imputation, outlier removal, and data standardization, such as Z-score normalization, to ensure data quality and consistency. Specifically, the normalization formula is:

[0068]

[0069] Where x represents the original data, x min x max These are the historical minimum and maximum values ​​of the feature parameter, respectively.

[0070] In this embodiment of the application, step S200 analyzes the correlation between meteorological data and the output power of the photovoltaic power station, and selects meteorological factors with an absolute value of correlation coefficient greater than a preset threshold to form a set of main meteorological factors, including the following steps A1-A3:

[0071] A1: The correlation coefficient between each meteorological factor and the output sequence was calculated one by one using the Pearson correlation coefficient;

[0072] For example, the correlation between various meteorological factors and the output power of a photovoltaic power plant can be quantified using the Pearson correlation coefficient, as follows:

[0073] ρ x,y =f(X,Y)

[0074] In the formula, ρ x,y The Pearson linear correlation coefficient is used to illustrate the correlation between numerical meteorological data and photovoltaic output. Data set X consists of weather data, including irradiance, temperature, wind speed, humidity, and cloud cover. Data set Y consists of photovoltaic output series.

[0075] The formula for calculating the Pearson correlation coefficient is as follows:

[0076]

[0077] A2: Set a first threshold to filter major meteorological factors;

[0078] For example, the Pearson correlation coefficient between the photovoltaic power output sequences of various meteorological data can be calculated, and the first threshold can be set to 0.3. Meteorological data with a correlation coefficient greater than 0.3 are input into the main meteorological factor set.

[0079] A3: Perform multicollinearity tests on the selected set of major meteorological factors to generate the final set of major meteorological factors.

[0080] Specifically, multicollinearity tests can be used to ensure the independence between features through VIF analysis, generating the final set of major meteorological factors, such as irradiance, temperature, and wind speed.

[0081] In this embodiment of the application, step S300 involves collecting power output sequence data from neighboring photovoltaic power plants and calculating the correlation coefficient between the power output sequences of the target power plant and the neighboring power plants, including the following steps B1-B4:

[0082] B1: Set the proximity analysis range based on the geographical coordinates of the target photovoltaic power station;

[0083] Specifically, the analysis range is determined based on the geographical coordinates of the target photovoltaic power station, such as latitude and longitude, which is usually all photovoltaic power stations within 50-100 kilometers of the target power station.

[0084] B2: Obtain metadata from nearby power plants and generate a list of nearby power plants;

[0085] Specifically, metadata can include: power plant location, installed capacity, and power output data availability.

[0086] B3: Collect historical power output sequence data from neighboring power plants and align the timestamps of the data from neighboring and target power plants to generate a time-aligned power output sequence dataset;

[0087] B4: The relationship between the output sequence of the target photovoltaic power station and the output sequence of each neighboring power station is calculated.

[0088] In an alternative implementation, the correlation coefficient in B4 can also be analyzed using the Pearson correlation coefficient:

[0089]

[0090] In the formula: r is the Pearson correlation coefficient; x i The y represents the wind speed data for the i-th adjacent meteorological station; i Let be the i-th power data of the target wind farm; x and y are the average values ​​of the data in the sequence x and y, respectively; n is the total number of samples in the sequence.

[0091] It should be noted that Pearson correlation coefficients are generally classified as follows: 0.8-1.0 extremely strong correlation, 0.6-0.8 strong correlation, 0.4-0.6 moderate correlation, 0.2-0.4 weak correlation, and 0.0-0.2 no correlation.

[0092] Meteorological factors such as cloud disturbances, temperature, and weather changes among photovoltaic power plants in geographically close proximity have a certain time delay and similarity. The photovoltaic output sequence not only has a certain autocorrelation in time, but also has a high degree of similarity among multiple photovoltaic output sequences of photovoltaic power plants.

[0093] In another alternative implementation, the correlation coefficient can also be calculated using the Spearman correlation coefficient. For example, the output sequence of the target photovoltaic power station and the output sequence of the neighboring power station are obtained; the values ​​of each sequence are sorted in ascending order and assigned a rank; for each time point t, the difference between the rank of the target power station and the rank of the neighboring power station is calculated, and the correlation is calculated using the Spearman correlation coefficient formula.

[0094] In this embodiment of the application, step S300, which involves selecting reference photovoltaic power plants based on correlation coefficient values, includes the following steps B5-B6:

[0095] B5: Set a second threshold to filter out neighboring power stations that are highly correlated with the output of the target power station as reference power stations;

[0096] B6: If multiple power stations have similar correlation coefficients, the power station closer to the target power station should be selected as the reference power station.

[0097] For example, the screening criteria for strongly correlated power plants in the second threshold of B5 can be photovoltaic power plants with a Pearson correlation coefficient greater than 0.6 and a spatial distance of no more than 50 kilometers. The strongly correlated power plants selected based on the Pearson correlation coefficient are used as reference photovoltaic power plants, and their photovoltaic power output data are input into the photovoltaic power output prediction model.

[0098] In another alternative implementation, the feasibility of the selected reference power stations is verified based on the application, specifically whether they are driven by similar meteorological conditions. For example, the correlation between meteorological data of the target power station and the reference power station, such as irradiance and temperature, is examined to ensure that the spatial correlation is not a coincidence.

[0099] In this embodiment of the application, step S400 involves constructing a multidimensional dataset based on the output data of a reference photovoltaic power plant and the output-related data of the target power plant, including the following steps C1-C4:

[0100] C1: Align all data timestamps;

[0101] For example, ensure that all data is aligned at the same time resolution, such as 15 minutes.

[0102] C2: Construct a feature matrix, including: meteorological characteristics of the target power station, historical power output of the target power station, and power output sequence of the reference power station;

[0103] C3: Add time features to capture intraday and weekly periodic patterns;

[0104] Examples include hours and days of the week to capture intraday and weekly cyclical patterns.

[0105] C4: Perform sliding window processing on the dataset to generate a multidimensional dataset suitable for prediction.

[0106] For example, sliding window processing can predict the next hour based on data from the past 24 hours.

[0107] In this embodiment of the application, step S500 involves constructing a neural network prediction model for each seasonal subset of samples, and optimizing the parameters to obtain the optimal prediction model for each season, including the following steps D1-D2:

[0108] D1: Extract the data corresponding to the season from the timestamps of the cube to obtain four subsets;

[0109] Specifically, four seasonal sub-sample sets can be established: spring, summer, autumn, and winter.

[0110] In another alternative implementation, the cube can be further subdivided by month, for example, into 12 subsample sets.

[0111] D2: Construct a BP neural network prediction model. The BP neural network adopts a three-layer topology. The number of nodes in the input layer is determined according to the number of feature parameters. The number of nodes in the hidden layer is optimized by trial and error. The output layer is the photovoltaic power prediction value.

[0112] For example, the number of input features is equal to the feature dimension of the multidimensional dataset. For instance, 5 meteorological factors (irradiance, temperature, etc.) + historical output of the target power plant (assuming the past 4 hours, 4 time points) + output of 2 reference power plants (4 hours × 2), for a total of 5 + 4 + 8 = 17 input nodes.

[0113] Specifically, the empirical formula for determining the initial number of hidden layer nodes is:

[0114]

[0115] n1 is the number of hidden layer nodes; n is the number of input layer nodes; m is the number of output layer nodes; a is an adjustment constant between 1 and 10.

[0116] Based on historical operational data of the target photovoltaic power station and reference photovoltaic power station data obtained through spatial correlation analysis, modeling training samples were extracted after data preprocessing. The Sigmoid neuron activation function was selected, and the initial weights and thresholds were set to the minimum random number. The mathematical expression of the Sigmoid function is:

[0117]

[0118] The prediction error is measured by a loss function. The gradient of each parameter (partial derivative of the error with respect to the weights) is calculated layer by layer from the output layer in reverse. The weights and thresholds are then adjusted according to the gradient descent method.

[0119]

[0120] Where t i For the actual value, y i is the predicted value, and m is the number of neurons in the output layer.

[0121] Evaluate the model performance for each season on the test set and save the optimal BP neural network model for each season, including its weights and structure. The final output saves the optimal model parameters and structure for each season, for example, {Spring model: 2 layers, 30+15 nodes, learning rate 0.001}.

[0122] In another alternative implementation, the model can also use LSTM or GRU models other than BP; divide the training set, validation set and test set, and optimize the hyperparameters based on the validation set. The learning rate, number of hidden layer nodes and time step of the above two models can be adjusted by grid search; train the model and evaluate the performance using root mean square error (RMSE) or mean absolute error (MAE), and save the optimal model parameters for each season.

[0123] In this embodiment of the application, step S600 involves calling the corresponding seasonal prediction model based on the season to which the expected prediction period belongs, and outputting the photovoltaic power output prediction result, including:

[0124] Acquire real-time power output data of the target power plant, meteorological forecast data for the predicted period, and real-time power output data of the reference photovoltaic power plant, and construct a multidimensional dataset;

[0125] Input the multidimensional dataset into the optimal prediction model for the corresponding season, and output the photovoltaic power output prediction results.

[0126] In summary, this invention analyzes the spatial correlation of output sequences of neighboring photovoltaic power plants based on correlation coefficients, filters data from strongly correlated plants, and uses this data, along with historical data and meteorological data from the target photovoltaic power plant, as model inputs. It also identifies and removes outlier data points from the collected data; secondly, it normalizes the acquired data; and finally, it acquires a massive amount of effective data samples to improve the effectiveness of the data samples. Combining the meteorological method of dividing the year into four seasons, BP neural network models for different seasons are established according to the aforementioned implementation scheme. The performance of the BP neural network prediction model considering spatial correlation and seasonal characteristics is evaluated to improve the model's prediction accuracy.

[0127] Example 3, referring to Figure 2 Based on the previous embodiment and Table 1, this embodiment provides a comparative experiment of a short-term photovoltaic power output prediction method that considers spatial correlation and seasonal characteristics, in order to verify the beneficial effects of the method of the present invention.

[0128] Historical operating data from a photovoltaic (PV) power station in Hainan Province from June 2023 to May 2024 were selected as the data sample. Pearson correlation coefficient analysis was used to identify a strongly correlated neighboring PV power station (station 1) as a reference PV power station. Using historical data from the reference PV power station, the target PV power station, and NWP data categorized by season, a BP neural network model for different seasons was trained and optimized to predict PV output in different seasons. One day was randomly selected from each month from June 2023 to May 2024, and the actual PV output data, the prediction results of the BP neural network model considering spatial correlation and seasonal characteristics, and the prediction results of the annual model without considering spatial correlation were compared and analyzed. The results are as follows: Figure 2 As shown

[0129] To quantitatively verify the prediction results, the accuracy of different methods was evaluated using mean absolute error (MAE), root mean square error (RMSE), and root mean square error (RMSE) metrics.

[0130] MAE can be calculated using the following formula:

[0131]

[0132] RMSE can be calculated using the following formula:

[0133]

[0134] In the formula: denoted as , where x is the predicted photovoltaic power; is the actual photovoltaic power; and n is the number of samples.

[0135] Table 1. Error analysis of multi-model and year-round model predictions considering spatial correlation and seasonal characteristics. (Unit: %)

[0136]

[0137]

[0138] By analyzing the results in the table, the following conclusions can be drawn:

[0139] 1) Both prediction models have good prediction results, with prediction accuracy of over 90%.

[0140] 2) The root mean square error and mean absolute error of the multi-model photovoltaic power prediction method based on seasonal characteristics are smaller than those of the annual model. The prediction error is improved, and the stability of the prediction results is also significantly improved.

[0141] 3) The short-term photovoltaic power output prediction method that considers spatial correlation and seasonal characteristics has higher accuracy and better performance than the annual model, and can be used to design more accurate photovoltaic power prediction algorithms.

[0142] In summary, the short-term photovoltaic power output forecasting method that considers spatial correlation and seasonal characteristics expands the data dimensions and provides a more targeted seasonal classification method for photovoltaic power output forecasting, thereby effectively improving the accuracy of photovoltaic power output forecasting.

[0143] Example 4 illustrates a method for short-term photovoltaic power output prediction that considers spatial correlation and seasonal characteristics. It should be noted that the technical solution of this system for short-term photovoltaic power output prediction that considers spatial correlation and seasonal characteristics is based on the same concept as the method for short-term photovoltaic power output prediction that considers spatial correlation and seasonal characteristics described above. Details not described in detail in this example can be found in the description of the method for short-term photovoltaic power output prediction that considers spatial correlation and seasonal characteristics described above.

[0144] This embodiment also provides another short-term photovoltaic output forecasting system that considers spatial correlation and seasonal characteristics, including:

[0145] The acquisition module is used to acquire meteorological data and power output sequence data of the target photovoltaic power station;

[0146] The first correlation analysis module is used to analyze the correlation between meteorological data and the output power of photovoltaic power plants, and to select meteorological factors with an absolute value of correlation coefficient greater than a preset threshold to form a set of main meteorological factors.

[0147] The second correlation analysis module is used to collect the output sequence data of neighboring photovoltaic power plants, calculate the correlation coefficient between the output sequence of the target power plant and the neighboring power plants, and select reference photovoltaic power plants based on the correlation coefficient value.

[0148] The dataset construction module is used to jointly construct a multidimensional dataset based on the power output data of a reference photovoltaic power plant and the relevant power output data of the target power plant.

[0149] The model building module is used to classify the multidimensional dataset according to seasonal characteristics to obtain seasonal subsample sets. For each seasonal subsample set, a neural network prediction model is built, and the optimal prediction model for each season is obtained through parameter optimization.

[0150] The prediction output module is used to call the corresponding seasonal prediction model based on the season to which the expected prediction period belongs, and output the photovoltaic power output prediction results.

[0151] This embodiment also provides a computer device applicable to a short-term photovoltaic power output forecasting method considering spatial correlation and seasonal characteristics, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the short-term photovoltaic power output forecasting method considering spatial correlation and seasonal characteristics as proposed in the above embodiment.

[0152] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a short-term photovoltaic output prediction method considering spatial correlation and seasonal characteristics as proposed in the above embodiments.

[0153] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for short-term photovoltaic output prediction that considers spatial correlation and seasonal characteristics proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0154] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0155] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics, characterized in that, include: Acquire meteorological data and power output sequence data of the target photovoltaic power station; The correlation between meteorological data and the output power of photovoltaic power plants is analyzed, and meteorological factors with absolute values ​​of correlation coefficients greater than preset thresholds are selected to form the main meteorological factor set. Collect power output sequence data from nearby photovoltaic power plants, calculate the correlation coefficient between the power output sequence of the target power plant and the nearby power plants, and select reference photovoltaic power plants based on the correlation coefficient values; A multidimensional dataset is constructed based on the power output data of a reference photovoltaic power plant and the relevant power output data of the target power plant. The multidimensional dataset is classified according to seasonal characteristics to obtain seasonal sub-sample sets. A neural network prediction model is constructed for each seasonal sub-sample set, and the optimal prediction model for each season is obtained through parameter optimization. Based on the season to which the expected prediction period belongs, the corresponding seasonal prediction model is invoked to output the photovoltaic power output prediction results.

2. The method for short-term photovoltaic power output prediction considering spatial correlation and seasonal characteristics as described in claim 1, characterized in that, The correlation between meteorological data and the output power of photovoltaic power plants was analyzed, and meteorological factors with absolute correlation coefficients greater than a preset threshold were selected to form a set of main meteorological factors, including: The correlation coefficient between each meteorological factor and the output sequence was calculated one by one using the Pearson correlation coefficient. Set a first threshold to filter major meteorological factors; Multicollinearity tests were performed on the selected set of major meteorological factors to generate the final set of major meteorological factors.

3. The method for short-term photovoltaic power output prediction considering spatial correlation and seasonal characteristics as described in claim 2, characterized in that, Collect power output sequence data from nearby photovoltaic power plants and calculate the correlation coefficient between the power output sequences of the target power plant and those of nearby power plants, including: Based on the geographical coordinates of the target photovoltaic power station, define the proximity analysis range; Obtain metadata from nearby power plants and generate a list of nearby power plants; Collect historical power output sequence data from neighboring power plants and align the timestamps of the data from neighboring and target power plants to generate a time-aligned power output sequence dataset; The correlation coefficients between the output sequence of the target photovoltaic power plant and the output sequence of each neighboring power plant are calculated.

4. The method for short-term photovoltaic power output prediction considering spatial correlation and seasonal characteristics as described in claim 3, characterized in that, Reference photovoltaic power plants were selected based on correlation coefficient values, including: A second threshold is set to filter out neighboring power plants that are highly correlated with the output of the target power plant as reference power plants; If multiple power stations have similar correlation coefficients, the power station closer to the target power station should be selected as the reference power station.

5. The method for short-term photovoltaic power output prediction considering spatial correlation and seasonal characteristics as described in claim 4, characterized in that, A multidimensional dataset is constructed based on the output data of a reference photovoltaic power plant and the relevant output data of the target power plant, including: Align all data timestamps; Construct a feature matrix, including: meteorological characteristics of the target power station, historical power output of the target power station, and power output sequence of the reference power station; Add time features to capture intraday and weekly periodic patterns; A sliding window process is applied to the dataset to generate a multidimensional dataset suitable for prediction.

6. The method for short-term photovoltaic power output prediction considering spatial correlation and seasonal characteristics as described in claim 5, characterized in that, A neural network prediction model is constructed for each seasonal subset of samples, and the optimal prediction model for each season is obtained through parameter optimization, including: Data corresponding to the season is extracted from the timestamps of the cube to obtain four subsets; A BP neural network prediction model is constructed. The BP neural network adopts a three-layer topology. The number of nodes in the input layer is determined according to the number of feature parameters, the number of nodes in the hidden layer is optimized by trial and error, and the output layer is the photovoltaic power prediction value.

7. A short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics as described in claim 6, characterized in that, Based on the season to which the expected forecast period belongs, the corresponding seasonal forecast model is invoked, and the photovoltaic power output forecast results are output, including: Acquire real-time power output data of the target power plant, meteorological forecast data for the predicted period, and real-time power output data of the reference photovoltaic power plant, and construct a multidimensional dataset; Input the multidimensional dataset into the optimal prediction model for the corresponding season, and output the photovoltaic power output prediction results.

8. A short-term photovoltaic power output forecasting system considering spatial correlation and seasonal characteristics, employing the method described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire meteorological data and power output sequence data of the target photovoltaic power station; The first correlation analysis module is used to analyze the correlation between meteorological data and the output power of photovoltaic power plants, and to select meteorological factors with an absolute value of correlation coefficient greater than a preset threshold to form a set of main meteorological factors. The second correlation analysis module is used to collect the output sequence data of neighboring photovoltaic power plants, calculate the correlation coefficient between the output sequence of the target power plant and the neighboring power plants, and select reference photovoltaic power plants based on the correlation coefficient value. The dataset construction module is used to jointly construct a multidimensional dataset based on the power output data of a reference photovoltaic power plant and the relevant power output data of the target power plant. The model building module is used to classify the multidimensional dataset according to seasonal characteristics to obtain seasonal subsample sets. For each seasonal subsample set, a neural network prediction model is built, and the optimal prediction model for each season is obtained through parameter optimization. The prediction output module is used to call the corresponding seasonal prediction model based on the season to which the expected prediction period belongs, and output the photovoltaic power output prediction results.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the short-term photovoltaic power output prediction method considering spatial correlation and seasonal characteristics as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of a short-term photovoltaic power output forecasting method considering spatial correlation and seasonal characteristics as described in any one of claims 1 to 7.

Images