Photovoltaic generating capacity short-term prediction system

Abstract

The invention relates to the technical field of photovoltaic power generation capacity prediction, in particular to a photovoltaic power generation capacity short-term prediction system based on multi-source data fusion, which comprises a data acquisition module, a feature processing module, a model construction module and a prediction optimization module. According to the method, historical power generation data, meteorological data and satellite remote sensing data are fused, wavelet neural analysis and an I nform prediction model are utilized, and power generation change rate classification and concentrated time period modeling are combined, so that the photovoltaic power output space-time correlation is accurately depicted. An interference function dynamically adjusts a prediction result under extreme weather, the adaptability is enhanced, and meanwhile the prediction precision and robustness are improved through feature selection and model integration. According to the invention, high-precision, real-time and high-adaptability photovoltaic generating capacity prediction support can be provided for power grid dispatching.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy and power system, and specifically to short-term prediction of photovoltaic power generation. Background Art

[0002] In recent years, with the continuous optimization of artificial intelligence algorithms, photovoltaic power generation prediction methods have gradually integrated multi-source data and complex models, and can improve prediction accuracy by combining photovoltaic power station data, meteorological data, satellite remote sensing data and other information. At the same time, the application of spatiotemporal correlation modeling technology has enhanced the ability to describe the correlation between photovoltaic power output in different regions, and ultra-short-term predictions and predictions under extreme weather conditions have also been further developed to meet the real-time and adaptability requirements of power grid dispatching.

[0003] At present, in the field of photovoltaic power generation prediction, statistical model method and machine learning method are the main research directions. The statistical model method uses historical data to establish a prediction model through methods such as time series analysis, regression analysis and non-parametric models. It has the advantage of simple calculation, but there are certain limitations in accuracy. Machine learning rules use algorithms such as support vector machines, neural networks, decision trees and random forests to explore the complex relationship between photovoltaic power and influencing factors, showing high prediction accuracy, but its training process is complex and has high requirements for data volume and quality. In addition, although the physical model method has high accuracy, it is complex in calculation and highly dependent on system parameters.

[0004] Although existing technologies have made some progress in photovoltaic power generation forecasting, there is still room for improvement in the deep application of multi-source data fusion, the refinement of feature engineering, and the comprehensiveness of model optimization. For example, the prediction ability of existing methods under extreme weather conditions needs to be further improved, and the uncertainty quantification and visualization of prediction results also need to be strengthened. Summary of the invention

[0005] The technical problem solved by the present invention is that the existing technology fails to fully integrate multi-source data to improve the prediction accuracy in photovoltaic power generation prediction, lacks in-depth modeling of the spatiotemporal correlation of photovoltaic power output, and has insufficient prediction capabilities under extreme weather conditions. In addition, the existing methods have limitations in the quantification and visualization of the uncertainty of the prediction results, and cannot meet the real-time and adaptability requirements of power grid dispatching.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a photovoltaic power generation short-term prediction system based on multi-source data fusion, including a data acquisition module, a feature processing module, a model building module and a prediction optimization module;

[0007] The data acquisition module is used to obtain historical power generation data, meteorological data and satellite remote sensing data of the target photovoltaic power station within a historical time period;

[0008] The feature processing module is used to clean, standardize and reduce the dimension of the collected data, and extract feature variables that have a significant impact on the photovoltaic power generation;

[0009] The model construction module trains the prediction model using historical data through wavelet neural analysis combined with the Informer prediction model, determines the model parameters, and evaluates the prediction accuracy of the model;

[0010] The prediction optimization module adjusts the hyperparameters according to the model prediction results, screens the most predictive feature variables, and integrates multiple prediction models to form a comprehensive prediction model, and finally outputs the prediction results.

[0011] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: the historical power generation data includes historical time points, historical power generation power and the operating state of photovoltaic modules; the meteorological data includes irradiance, temperature, wind speed and humidity; the satellite remote sensing data includes cloud cover and surface reflectivity; the historical time points are represented as each time point continuously monitored within the first historical time period, and the first historical time period represents a historical time period with a period of one year.

[0012] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: the feature extraction logic of the feature processing module includes:

[0013] Obtain adjacent historical time points and the corresponding historical power generation power, calculate the difference between adjacent historical time points, denoted as the first difference, calculate the difference between the corresponding historical power generation powers, denoted as the second difference, calculate the ratio of the second difference to the first difference, and take the absolute value of the ratio, and set the absolute value of the ratio as the power generation change rate; when the latter historical power generation power is greater than the former historical power generation power, mark the absolute value of the ratio as a positive change, and when the latter historical power generation power is less than the former historical power generation power, mark the absolute value of the ratio as a negative change.

[0014] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: the feature variable category is represented as the category of the power generation change degree of the target photovoltaic power station at the current moment relative to the previous moment, and the feature variable category includes the first category, the second category, the third category and the fourth category; the first category and the second category represent the power generation change rate of negative change, wherein, the absolute value of the negative change of the first category is greater than the absolute value of the negative change of the second category; the third category represents the power generation change rate of positive change and the power generation change amount with a value of 0; the fourth category represents the power generation change rate of positive change, and the positive change of the fourth category is greater than the positive change of the third category.

[0015] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: the classification logic of the centralized time corresponding to each characteristic variable category includes:

[0016] Clean the power generation change rate, convert the corresponding historical time points into hourly time periods, set the number of clusters to 4, classify the power generation change rate into each characteristic variable category, set labels for each power generation change rate with the characteristic variable category, and input the labeled power generation change rate into the K-means algorithm to obtain the time periods corresponding to each characteristic variable category, where the time period is expressed as A hours to B hours, where the value of A is less than B, and both A and B are distributed in the range of 0 to 24; set the time period corresponding to the fourth category as the centralized time, and the centralized time is expressed as the centralized time of the historical day where the historical time point is located.

[0017] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: obtain the irradiance and temperature of each day in the first historical time period, and obtain the corresponding centralized time period of each day in the first historical time period;

[0018] Set the first value as the ideal irradiance, set the second value as the ideal temperature, obtain the centralized time period corresponding to the ideal irradiance and ideal temperature, denoted as the ideal centralized time period, calculate the differences between the irradiance and temperature of each day and the first value and the second value, denoted as the irradiance difference and the temperature difference, and calculate the difference between the corresponding centralized time period and the ideal centralized time period, denoted as the duration difference;

[0019] Calculate the weights of the meteorological data, and the calculation logic of the weights includes:

[0020] Perform multiple regression analysis with the irradiance difference and the temperature difference as independent variables and the duration difference as the dependent variable to obtain the coefficient of the irradiance difference and the coefficient of the temperature difference, and set the coefficient of the irradiance difference and the coefficient of the temperature difference as the weight of the irradiance and the weight of the temperature.

[0021] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: construct an interference function according to the weights, and the calculation expression of the interference function is:

[0022]

[0023] wherein, T is the duration difference, x i represents the i-th independent variable, and w i is the weight corresponding to the i-th independent variable;

[0024] Input the current meteorological data into the interference model to obtain the current time difference, obtain the current day, obtain the corresponding concentrated time period for the current day, perform a summation calculation on the current time difference and the concentrated time period to obtain a first sum value, and set the first sum value as the new concentrated time period.

[0025] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: the modeling logic of the prediction model includes:

[0026] Obtain the historical time points, historical power generation, meteorological data, and satellite remote sensing data of the target photovoltaic power station, perform missing value processing and outlier processing on the historical time points, historical power generation, meteorological data, and satellite remote sensing data, divide the target photovoltaic power station into unit areas with a first length, the area of the unit area is larger than the area of a single photovoltaic module, and only one set of photovoltaic modules can be accommodated in the unit area, obtain the historical positions of the photovoltaic modules after unit area division, and use the historical positions of the photovoltaic modules, historical power generation, and meteorological data as variables to respectively extract a position matrix, a power matrix, and a meteorological matrix;

[0027] Set the sampling time step. When there are photovoltaic modules within the sampling time step, set the elements corresponding to the position matrix to 1, otherwise, set the elements corresponding to the position matrix to 0. Normalize the historical power generation so that the historical power generation is distributed between 0 and 1, and assign the normalized historical power generation to the elements in the power matrix. When the meteorological data is sunny, set the elements corresponding to all unit areas on the target photovoltaic power station in the meteorological matrix to 1, otherwise, set the elements corresponding to all unit areas on the target photovoltaic power station in the meteorological matrix to 0;

[0028] Encode the position matrix, power matrix, and meteorological matrix through wavelet transform encoding, select a deep learning model, use the encoded matrices as the input quantities of the model, and use the corresponding concentrated time period and the corresponding power generation change rate as the output quantities to construct a prediction model.

[0029] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: input the corresponding meteorological data and the corresponding satellite remote sensing data into the prediction model to obtain the concentrated time period and the power generation change rate, obtain the corresponding current meteorological data, and obtain a new concentrated time period and a new power generation change rate through an interference function;

[0030] Set the third value as the power generation change rate threshold, compare the absolute value of the new power generation change rate with the power generation change amount threshold, and perform a first operation according to the comparison result. The first operation includes adjusting the parameters of the prediction model and sending a completion signal;

[0031] When the absolute value of the new power generation change rate is greater than or equal to the power generation change rate threshold, set the first operation to adjust the parameters of the prediction model. The adjustment includes continuously decreasing or continuously increasing until the absolute value of the new power generation change rate is less than the power generation change rate threshold, then stop adjusting the parameters of the prediction model and send a completion signal. When the absolute value of the new power generation change rate is less than the power generation change rate threshold, set the first operation to send a completion signal.

[0032] As a preferred solution of the short-term photovoltaic power generation prediction system based on multi-source data fusion according to the present invention, wherein: establish a first mapping relationship, which is expressed as the mapping relationship between meteorological data, any new concentrated time period of a day and the parameters of the prediction model. By inputting meteorological data and the concentrated time period of any day, the corresponding parameters of the prediction model can be obtained.

[0033] Technical effects of the present invention: Through the application of multi-source data fusion and deep learning models, the system can comprehensively analyze the historical power generation data, meteorological data and satellite remote sensing data of the photovoltaic power station, and improve the prediction accuracy. By classifying the power generation change rate and modeling the concentrated time period, the system can accurately depict the spatio-temporal correlation of photovoltaic power output. The introduction of the interference function enables the system to dynamically adjust the prediction results under extreme weather conditions and enhance the adaptability of the prediction. In addition, through feature selection and model integration, the system further improves the robustness and accuracy of the prediction, providing reliable support for power grid scheduling. Description of the Drawings

[0034] Figure 1 It is the overall structural block diagram of the short-term photovoltaic power generation prediction system of the present invention, showing the connection relationship between the data acquisition module, the feature processing module, the model construction module and the prediction optimization module.

[0035] Figure 2 It is the flow schematic diagram of the feature processing module in the present invention, describing in detail the calculation logic of the power generation change rate and its classification process.

[0036] Figure 3 It is the modeling logic diagram of the prediction model of the present invention, showing the generation of the position matrix, the power matrix and the meteorological matrix and the input-output relationship after wavelet transform coding.

[0037] Figure 4 It is the application flow chart of the interference function of the present invention, explaining the specific steps of adjusting the concentrated time period and the power generation change rate through the current meteorological data.

[0038] The reference numerals are as follows:

[0039] 1. Data acquisition module; 2. Feature processing module; 3. Model construction module; 4. Prediction optimization module; 5. Power generation change rate; 6. Location matrix; 7. Power matrix; 8. Meteorological matrix; 9. Interference function; 10. Concentrated time period. Detailed implementation manner

[0040] The present invention provides a short-term photovoltaic power generation prediction system based on multi-source data fusion. The overall structure of the system is as Figure 1 shown, including a data acquisition module 1, a feature processing module 2, a model construction module 3, and a prediction optimization module 4. These modules work together through data flow and logical connection relationships to achieve accurate prediction of photovoltaic power generation. In practical applications, the data acquisition module 1 is responsible for obtaining raw data from the external environment and transmitting it to the feature processing module 2. The feature processing module 2 cleans and extracts features from the data and then transfers it to the model construction module 3. The model construction module 3 uses deep learning algorithms to complete modeling and inputs the results into the prediction optimization module 4. Finally, the prediction optimization module 4 outputs the optimized prediction results.

[0041] The specific operation principle of the data acquisition module 1 is as follows: This module obtains historical power generation data, meteorological data, and satellite remote sensing data of the target photovoltaic power station through a sensor network and a data interface. The historical power generation data includes historical time points, historical power generation power, and the operating status of photovoltaic modules. Among them, the historical time points are represented as each time point continuously monitored within the first historical time period, and the first historical time period is usually set to the time span of a complete year. The meteorological data includes variables such as irradiance, temperature, wind speed, and humidity, while the satellite remote sensing data covers information such as cloud cover and surface reflectivity. These data are collected in real time through a standardized data transmission protocol and stored in a database, and then transmitted to the feature processing module 2 for further processing.

[0042] The operation process of the feature processing module 2 is as Figure 2As shown, its core task is to clean, standardize, and reduce the dimensionality of the collected data, and extract the feature variables that have a significant impact on the photovoltaic power generation. First, the module calculates the first difference between adjacent historical time points and the second difference of the corresponding historical power generation, and then obtains the power generation change rate 5. The calculation logic of the power generation change rate 5 is to take the absolute value of the ratio of the second difference to the first difference, and mark it as a positive change or a negative change according to the magnitude relationship of the historical power generation before and after. For the convenience of subsequent modeling, the feature processing module 2 divides the power generation change rate 5 into four categories: the first category and the second category represent negative changes, and the negative change amplitude of the first category is greater than that of the second category; the third category represents positive changes and zero changes; the fourth category represents positive changes and the amplitude is greater than that of the third category. In addition, the module also classifies the power generation change rate 5 into each category through the K-means clustering algorithm, and maps the labeled power generation change rate 5 to the hourly time period, so as to determine the concentrated time period 10.

[0043] The modeling logic of the model construction module 3 is as Figure 3 As shown, its main function is to complete the prediction of photovoltaic power generation through wavelet neural analysis combined with the Informer prediction model. First, the module processes the missing values and outliers of the historical time points, historical power generation, meteorological data, and satellite remote sensing data, and then divides the target photovoltaic power station into multiple unit areas of the first length, and each unit area only accommodates a group of photovoltaic modules. On this basis, the module generates a position matrix 6, a power matrix 7, and a meteorological matrix 8. The position matrix 6 is used to identify the presence or absence of photovoltaic modules within the sampling time step, the power matrix 7 records the normalized historical power generation, and the meteorological matrix 8 reflects the changes in meteorological conditions. These matrices are encoded through wavelet transform and used as the input of the deep learning model, and the output of the model is the concentrated time period 10 and the power generation change rate 5. Through training and parameter adjustment, the model can capture the spatio-temporal correlation of photovoltaic power output.

[0044] The operation process of the prediction optimization module 4 is as Figure 4 As shown, its core lies in dynamically adjusting the prediction result through the interference function 9. First, the module inputs the current meteorological data and satellite remote sensing data into the prediction model to obtain the preliminary concentrated time period 10 and the power generation change rate 5. Subsequently, the module uses the interference function 9 to calculate the current time difference, and adds it to the concentrated time period 10 to obtain a new concentrated time period. The mathematical expression of the interference function 9 is that T is equal to w i x iThe cumulative sum, where xi is the independent variable and wi is the weight. The weights are calculated based on multiple regression analysis, with the irradiance difference and temperature difference as independent variables and the duration difference as the dependent variable. The module also compares the new power generation change rate 5 with a preset power generation change rate threshold. If the absolute value of the new power generation change rate 5 is greater than or equal to the threshold, the parameters of the prediction model are adjusted until the absolute value of the new power generation change rate 5 is less than the threshold.

[0045] The collaboration relationship between each module is as follows: The data acquisition module 1 obtains the raw data through the sensor network and stores it in the database. The feature processing module 2 reads the data from the database, cleans and extracts features, and then transfers the processed data to the model construction module 3. The model construction module 3 uses deep learning algorithms to complete the modeling and inputs the prediction results into the prediction optimization module 4. The prediction optimization module 4 dynamically adjusts the prediction results through the interference function 9 and outputs the finally optimized prediction results to the user interface or the power grid dispatching system. Throughout the process, each module cooperates with each other through data flow and logical control to ensure the efficient operation of the system.

[0046] In practical applications, the present system can be deployed in the monitoring center of large-scale photovoltaic power stations to support power grid dispatching decisions. For example, at 10 am on a certain day, the system obtains the current meteorological data and satellite remote sensing data through the data acquisition module 1. The feature processing module 2 calculates the power generation change rate 5 and determines the concentrated time period 10. The model construction module 3 generates a prediction model and outputs the preliminary prediction results. The prediction optimization module 4 adjusts the prediction results through the interference function 9 and outputs the finally optimized predicted value of photovoltaic power generation. This predicted value can be directly used to guide power grid dispatching, improving the stability and economy of power grid operation.

[0047] The implementation of the present system requires high-performance computing devices and a stable network environment to ensure the real-time nature of data acquisition, processing, and modeling. In addition, the system needs to regularly update historical data and model parameters to adapt to changes in the operating conditions of photovoltaic power stations. It can be seen from the above detailed description that the technical solution of the present invention has high operability and practicality, and can meet the needs of short-term prediction of photovoltaic power generation.

[0048] To enable relevant personnel in the technical field to better understand and implement the present invention, the following further supplements the specific implementation principles of the present invention in combination with a specific application scenario.

[0049] In the monitoring center of a large-scale photovoltaic power station, the short-term photovoltaic power generation prediction system provided by the present invention is deployed. The system realizes the accurate prediction of photovoltaic power generation through the coordinated operation of the data acquisition module 1, the feature processing module 2, the model construction module 3, and the prediction optimization module 4. The following will be elaborated in detail in combination with the reference numerals and specific operation steps in the attached drawings.

[0050] First, at 9:00 am, the data acquisition module 1 starts running and obtains the historical power generation data, meteorological data, and satellite remote sensing data of the target photovoltaic power station through the sensor network and data interface. The historical power generation data includes the historical time points, historical power generation power, and operating status of photovoltaic modules per hour in the past year; the meteorological data covers variables such as current irradiance, temperature, wind speed, and humidity; the satellite remote sensing data contains cloud cover and surface reflectivity information. These data are collected in real time through a standardized data transmission protocol, stored in the database, and then transmitted to the feature processing module 2 for further processing.

[0051] Next, the feature processing module 2 performs cleaning, standardization, and dimensionality reduction processing on the collected data according to the Figure 2 process shown. The module first calculates the first difference between adjacent historical time points and the second difference corresponding to the historical power generation power, and then obtains the power generation change rate 5. The calculation logic of the power generation change rate 5 is to take the absolute value of the ratio of the second difference to the first difference and mark it as a positive change or a negative change according to the magnitude relationship between the historical power generation powers before and after. Subsequently, the module divides the power generation change rate 5 into four categories: the first category and the second category represent negative changes, and the negative change amplitude of the first category is greater than that of the second category; the third category represents positive changes and zero changes; the fourth category represents positive changes and the amplitude is greater than that of the third category. Through the K-means clustering algorithm, the module classifies the labeled power generation change rate 5 into each category and maps it to the hour-level time period to determine the concentrated time period 10. For example, if the power generation change rate on a certain day is mainly concentrated from 2 pm to 4 pm, this time period is marked as the concentrated time period 10.

[0052] Subsequently, the model construction module 3 completes the construction of the prediction model according to the Figure 3 modeling logic shown. First, the module processes the missing values and outliers of the historical time points, historical power generation power, meteorological data, and satellite remote sensing data, and then divides the target photovoltaic power station into multiple unit areas of the first length, and each unit area only accommodates a group of photovoltaic modules. On this basis, the module generates a position matrix 6, a power matrix 7, and a meteorological matrix 8. The position matrix 6 is used to identify the presence or absence of photovoltaic modules within the sampling time step, the power matrix 7 records the normalized historical power generation power, and the meteorological matrix 8 reflects the changes in meteorological conditions. These matrices are encoded through wavelet transform and used as the input of the deep learning model, and the output of the model is the concentrated time period 10 and the power generation change rate 5. Through training and parameter adjustment, the model can capture the spatio-temporal correlation of photovoltaic power output. For example, when the meteorological conditions on a certain day show sunny, the elements corresponding to all unit areas in the meteorological matrix 8 are set to 1, while they are set to 0 on cloudy or rainy days.

[0053] Finally, the prediction optimization module 4 follows theFigure 4 The process shown dynamically adjusts the prediction results. The module first inputs the current meteorological data and satellite remote sensing data into the prediction model to obtain the initial concentrated time period 10 and the power generation change rate 5. Subsequently, the module uses the interference function 9 to calculate the current time difference and sums it with the concentrated time period 10 to obtain a new concentrated time period. The mathematical expression of the interference function 9 is that T is equal to the cumulative sum of w i x i , where x i is the independent variable and w i is the weight. The calculation of the weight is based on multiple regression analysis, with the irradiance difference and temperature difference as independent variables and the time difference as the dependent variable. For example, if the current irradiance is lower than the ideal irradiance and the temperature is higher than the ideal temperature, the module will adjust the time difference according to the weight, thereby correcting the concentrated time period 10. In addition, the module also compares the new power generation change rate 5 with a preset power generation change rate threshold. If the absolute value of the new power generation change rate 5 is greater than or equal to the threshold, the parameters of the prediction model are adjusted until the absolute value of the new power generation change rate 5 is less than the threshold.

[0054] The cooperation relationship between the modules is as follows: The data acquisition module 1 obtains the raw data through the sensor network and stores it in the database. The feature processing module 2 reads the data from the database, cleans and extracts features, and then transfers the processed data to the model construction module 3. The model construction module 3 uses deep learning algorithms to complete the modeling and inputs the prediction results into the prediction optimization module 4. The prediction optimization module 4 dynamically adjusts the prediction results through the interference function 9 and outputs the finally optimized prediction results to the user interface or the power grid dispatching system. Throughout the process, the modules cooperate with each other through data flow and logical control to ensure the efficient operation of the system.

[0055] Through the above steps, this system can significantly improve the accuracy and adaptability of photovoltaic power generation prediction in practical applications. For example, at 10 am on a certain day, the system obtains the current meteorological data and satellite remote sensing data through the data acquisition module 1. The feature processing module 2 calculates the power generation change rate 5 and determines the concentrated time period 10. The model construction module 3 generates a prediction model and outputs the preliminary prediction results. The prediction optimization module 4 adjusts the prediction results through the interference function 9 and outputs the finally optimized photovoltaic power generation prediction value. This prediction value can be directly used to guide the power grid dispatching, improving the stability and economy of the power grid operation.

[0056] The content not described in detail in the specification belongs to the prior art well-known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited, and conventional equipment can be used. In this technical solution, the electrical control components not mentioned are not shown in the figure because they belong to the prior art, and will not be described here either.

Claims

1. A short-term photovoltaic power generation prediction system, characterized in that, It includes a data acquisition module (1), a feature processing module (2), a model construction module (3), and a prediction optimization module (4); The data acquisition module (1) is used to obtain the historical power generation data, meteorological data, and satellite remote sensing data of the target photovoltaic power station within the first historical time period. The historical power generation data includes historical time points, historical power generation power, and the operating status of photovoltaic modules. The meteorological data includes irradiance, temperature, wind speed, and humidity. The satellite remote sensing data includes cloud cover and surface reflectivity; The feature processing module (2) is used to clean, standardize, and reduce the dimension of the collected data, and extract feature variables that have a significant impact on the power generation of photovoltaic power; The model construction module (3) combines wavelet neural analysis with the Informer prediction model, and uses historical data to train the prediction model to determine model parameters; The prediction optimization module (4) adjusts hyperparameters according to the model prediction results, screens the most predictive feature variables, and integrates multiple prediction models to form a comprehensive prediction model, and finally outputs the prediction results.

2. The short-term photovoltaic power generation prediction system according to claim 1, characterized in that: The feature extraction logic of the feature processing module (2) includes: Obtain adjacent historical time points and corresponding historical power generation power, calculate the difference between adjacent historical time points, denoted as the first difference, calculate the difference between the corresponding historical power generation power, denoted as the second difference, calculate the ratio of the second difference to the first difference, and take the absolute value of the ratio. Set the absolute value of the ratio as the power generation change rate (5); when the subsequent historical power generation power is greater than the previous historical power generation power, mark the absolute value of the ratio as a positive change, and when the subsequent historical power generation power is less than the previous historical power generation power, mark the absolute value of the ratio as a negative change.

3. The short-term photovoltaic power generation prediction system according to claim 1, wherein: The feature variable category represents the category of the power generation change degree of the target photovoltaic power station at the current moment relative to the previous moment. The feature variable category includes the first category, the second category, the third category, and the fourth category; the first category and the second category represent the power generation change rate (5) of negative change, where the absolute value of the negative change in the first category is greater than the absolute value of the negative change in the second category; the third category represents the power generation change rate (5) of positive change and the power generation change amount with a value of 0; the fourth category represents the power generation change rate (5) of positive change, and the positive change in the fourth category is greater than the positive change in the third category.

4. The short-term photovoltaic power generation prediction system according to claim 1, wherein: The modeling logic of the model construction module (3) includes: Obtain the historical time points, historical power generation power, meteorological data, and satellite remote sensing data of the target photovoltaic power station, perform missing value processing and outlier processing, divide the target photovoltaic power station into unit areas with a first length. The area of the unit area is larger than the area of a single photovoltaic module, and only one set of photovoltaic modules can be accommodated in the unit area. Obtain the historical positions of photovoltaic modules after unit area division, and use the historical positions of photovoltaic modules, historical power generation power, and meteorological data as variables to extract a position matrix (6), a power matrix (7), and a meteorological matrix (8) respectively; Set the sampling time step. When there is a photovoltaic module within the sampling time step, set the element corresponding to the position matrix (6) to 1; otherwise, set the element corresponding to the position matrix (6) to 0. Normalize the historical power generation to make the historical power generation distributed between 0 and 1, and assign the normalized historical power generation to the elements corresponding to the power matrix (7). When the meteorological data is sunny, set the elements corresponding to all unit areas on the target photovoltaic power station in the meteorological matrix (8) to 1; otherwise, set the elements corresponding to all unit areas on the target photovoltaic power station in the meteorological matrix (8) to 0. Encode the position matrix (6), power matrix (7) and meteorological matrix (8) through wavelet transform coding. Select a deep learning model, use the encoded matrices as the input variables of the model, and use the corresponding concentrated time period (10) and the corresponding power generation change rate (5) as the output variables to construct a prediction model.

5. The short-term photovoltaic power generation prediction system according to claim 1, characterized in that: The prediction optimization module (4) dynamically adjusts the prediction result through the interference function (9), and the calculation expression of the interference function (9) is: Among them, T is the duration difference, and x i represents the i-th independent variable, and w i is the weight corresponding to the i-th independent variable; Input the current meteorological data into the interference model to obtain the current time difference. Obtain the current day and the corresponding concentrated time period (10) of the current day. Perform a summation calculation on the current time difference and the concentrated time period (10) to obtain the first sum value, and set the first sum value as the new concentrated time period (10).

Images