Monitoring data mining and prediction analysis method for distributed photovoltaic power generation system

Abstract

The invention provides a monitoring data mining and prediction analysis method for a distributed photovoltaic power generation system, and the method comprises the steps: obtaining the conversion efficiency of a distributed photovoltaic module, an array layout mode and a photovoltaic support inclination angle setting parameter, carrying out the correlation analysis of the power generation amount of a photovoltaic power station, predicting the power generation amount, and obtaining a theoretical power generation amount prediction value; acquiring actual generating capacity data of the distributed photovoltaic power station, comparing the actual generating capacity data with a theoretical generating capacity predicted value, and if the actual generating capacity is significantly higher than the theoretical generating capacity predicted value, determining that the photovoltaic power station has suspected deceiving and complementing behaviors, and generating a suspected deceiving and complementing distributed photovoltaic power station list; and for the power station which is confirmed to have the cheating and supplementing behavior in the list, acquiring the grid-connected time, the actual generating capacity and the approved subsidy electric quantity, predicting the accumulated generating capacity and the accumulated approved subsidy electric quantity of the power station in the whole life cycle, and calculating the difference between the accumulated generating capacity and the accumulated approved subsidy electric quantity to obtain the estimated value of the accumulated cheating and supplementing electric quantity of the power station.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a monitoring data mining and prediction analysis method for a distributed photovoltaic power generation system. Background Art

[0002] The core technical problem faced by the monitoring of electricity fraud in distributed photovoltaic power generation systems is how to accurately identify and quantify fraudulent subsidy behaviors. Consider factors such as component efficiency, meteorological conditions, and system operation efficiency to obtain reliable theoretical power generation forecasts. Compare the actual power generation with the theoretical forecast to identify power stations with abnormally high power generation. However, this comparison may be affected by factors such as the voltage level of the grid connection point and the actual installed capacity. Therefore, it is necessary to compare the actual voltage level and installed capacity of the grid connection point with the registered data to identify over-capacity grid connection behavior. This in turn raises the question of how to determine whether the power station meets the electrical conditions for over-capacity grid connection, which requires analysis of parameters such as transformer capacity and line current carrying capacity. Finally, for power stations confirmed to have fraudulent subsidy behaviors, on-site verification is conducted to determine whether the capacity has been increased privately, and the over-issued subsidies are compensated in accordance with the regulations. These issues are interrelated and together constitute a complex technical challenge. In short, how to accurately identify the fraudulent subsidy behavior of distributed photovoltaic power generation systems and effectively supervise and deal with it while considering multiple influencing factors is a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a monitoring data mining and prediction analysis method for a distributed photovoltaic power generation system, which mainly includes:

[0004] Obtain the conversion efficiency of distributed photovoltaic modules, array layout mode and photovoltaic support tilt setting parameters, conduct correlation analysis on the power generation of photovoltaic power stations and predict the power generation to obtain the theoretical power generation prediction value;

[0005] Obtain the actual power generation data of distributed photovoltaic power stations and compare it with the theoretical power generation forecast value. If the actual power generation is significantly higher than the theoretical power generation forecast value, it is determined that the photovoltaic power station is suspected of fraudulent subsidy, and a list of distributed photovoltaic power stations suspected of fraudulent subsidy is generated;

[0006] For each power station in the list of suspected fraudulent distributed photovoltaic power stations, obtain the actual voltage level and actual installed capacity data of its grid connection point, and compare them with the registered grid connection point voltage level and registered installed capacity data of the power station. If the actual value is greater than the registered value, it is determined that the power station has exceeded the capacity grid connection behavior, and the list of suspected fraudulent distributed photovoltaic power stations is updated;

[0007] For power stations confirmed to have engaged in subsidy fraud in the list, their grid connection time, actual power generation, and approved subsidy power are obtained. The cumulative power generation and cumulative approved subsidy power of the power station over its entire life cycle are predicted, and the difference between the two is calculated to obtain an estimated value of the cumulative subsidy fraud of the power station.

[0008] Based on the estimated value of the cumulative fraudulent electricity volume of the distributed photovoltaic power station, the subsidized electricity volume of the power station over its entire life cycle is calculated and compared with the actual cumulative power generation of the power station. If the actual cumulative power generation exceeds the subsidized electricity volume over its entire life cycle, an early warning message is generated and an on-site verification order is issued. For distributed photovoltaic power stations that are found to have exceeded the power generation volume during on-site verification, the estimated value of the cumulative fraudulent electricity volume is multiplied by the subsidized electricity price to obtain the amount of fraudulent subsidies, and the subsequent subsidy funds are stopped at the same time.

[0009] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0010] The present invention discloses a monitoring data mining and predictive analysis method for distributed photovoltaic power generation systems. By comparing theoretical and actual power generation, suspected subsidy fraud power stations are identified; over-capacity grid connection is verified in combination with grid-connected data; intelligent devices are deployed to monitor anomalies in real time; machine learning models are established to determine subsidy fraud in real time; and the degree of subsidy fraud is quantitatively assessed and made public. The present invention realizes comprehensive monitoring, timely warning and accurate identification of subsidy fraud in distributed photovoltaic power stations, effectively curbs subsidy fraud, ensures the rational use of subsidy funds, and promotes the healthy development of the photovoltaic industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 The present invention is a flow chart of a monitoring data mining and predictive analysis method for a distributed photovoltaic power generation system. DETAILED DESCRIPTION

[0012] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0013] like Figure 1 In this embodiment, a monitoring data mining and prediction analysis method for a distributed photovoltaic power generation system may specifically include:

[0014] Step S101, obtaining the conversion efficiency of distributed photovoltaic components, array layout mode and photovoltaic support tilt angle setting parameters, performing correlation analysis on the power generation of the photovoltaic power station and predicting the power generation, and obtaining a theoretical power generation prediction value.

[0015] According to the technical parameters and geographical location information of photovoltaic modules, the irradiance in different time periods is calculated, and temperature correction is performed to obtain the actual irradiance and temperature data of photovoltaic modules under various working conditions; parameters are set according to the layout of photovoltaic modules and the inclination of the bracket, the shadow obstruction of photovoltaic modules is calculated, and the actual output power of photovoltaic strings at different times is determined; a power generation prediction model is established to predict the theoretical power generation in each time period; the losses in each link of the photovoltaic system are quantitatively analyzed, and multiple groups of random samples of system losses are generated. The confidence interval of the comprehensive system efficiency is calculated, and the theoretical power generation prediction value is corrected to obtain the final prediction result.

[0016] For example, according to the technical parameters and geographic location information of distributed photovoltaic components, the PVGIS radiation model is used to calculate the irradiance in different time periods, and the Sandia temperature model is used for temperature correction to obtain the actual irradiance and temperature data of the photovoltaic components under various working conditions, and the component output power characteristic curve is obtained by the polynomial regression fitting method. According to the photovoltaic array layout mode and bracket inclination setting parameters, the Ray-tracing algorithm is used to calculate the shadow obstruction of the component, and the inverter efficiency table lookup method and linear interpolation are combined to determine the actual output power of each photovoltaic string at different times, and the photovoltaic array connection method is optimized by genetic algorithm. Based on the historical operation data of photovoltaic power stations, the STL decomposition method is used to extract seasonal, periodic and random characteristics, and the environmental factor influence model is constructed in combination with meteorological forecast data. The power generation prediction model is established through the XGBoost regression tree to obtain the theoretical power generation prediction value for each time period. Using the rolling prediction method, the actual power generation is regularly compared with the predicted value, and the model parameters are updated by the exponential smoothing method to improve the prediction accuracy. The losses in each link of the photovoltaic system are quantitatively analyzed, including cable loss, inverter conversion loss and grid connection loss. The probability distribution model of various losses is established. The Latin Hypercube Sampling method is used to generate multiple groups of random samples of system losses. The confidence interval of the comprehensive system efficiency is calculated by the Bootstrap method, and the theoretical power generation prediction value is corrected to obtain the final prediction result. In practical applications, the PVGIS radiation model is first used to calculate the irradiance of a specific geographical location. Taking city a as an example, the model inputs longitude and latitude a°N, b°E, and the annual average daily irradiation is 4.2kWh / m2. Then the Sandia temperature model is used for temperature correction, and the operating temperature of the component is calculated considering the ambient temperature and wind speed factors. Assuming the ambient temperature is 25℃ and the wind speed is 3m / s, the component temperature is about 42℃. Based on these data, the component power curve is fitted by polynomial regression to obtain a characteristic equation such as P = 250*(G / 1000)*(1-0.004*(T-25)), where P is the output power, G is the irradiation intensity, and T is the component temperature. Next, the Ray-tracing algorithm is used to analyze the shadow effect. Taking a 10m×10m square array as an example, the annual shadow loss rate is calculated to be 3.5% by simulating the sun's trajectory. Combined with the inverter efficiency lookup table method, assuming that the SMA brand inverter is selected, the efficiency is 98% at 50% load, and the efficiency value under each load is obtained by linear interpolation. The genetic algorithm is used to optimize the component connection method, setting the population size to 100 and the number of iterations to 1000, and finally obtaining the optimal connection solution, which increases the power generation by 1.2%. Then, the STL decomposition method is used to process the historical power generation data to extract seasonal, trend and random components. Taking a 50kW photovoltaic power station as an example, the seasonal characteristics of summer power generation being 40% higher than winter power generation are decomposed.Combining these features with weather forecast data, an XGBoost regression tree model was constructed, with the maximum tree depth set to 6, the learning rate set to 0.1, and 1000 rounds of training to obtain the theoretical power generation forecast for each period. The rolling forecast method was used to update the forecast model every day, and the model parameters were adjusted by exponential smoothing (smoothing coefficient α = 0.3). Finally, the system loss was analyzed, assuming that the cable loss obeyed the normal distribution N (2%, 0.5%) and the inverter loss obeyed the uniform distribution U (2%, 3%). The Latin Hypercube Sampling method was used to generate 1000 sets of loss samples, and the 95% confidence interval of the system comprehensive efficiency was calculated by the Bootstrap method to be [85%, 88%]. This efficiency interval was applied to the theoretical power generation forecast to obtain the final power generation forecast result.

[0017] Step S102, obtaining the actual power generation data of the distributed photovoltaic power station, and comparing it with the theoretical power generation forecast value. If the actual power generation is significantly higher than the theoretical power generation forecast value, it is determined that the photovoltaic power station is suspected of fraudulent subsidy, and a list of distributed photovoltaic power stations suspected of fraudulent subsidy is generated.

[0018] The actual power generation data in the distributed photovoltaic power station management platform database is obtained; the actual power generation data is preprocessed by time series analysis tools to obtain a smoothed actual power generation time series; the theoretical power generation is calculated according to the geographical location, equipment parameters and meteorological data of the photovoltaic power station to obtain a theoretical power generation forecast value; the smoothed actual power generation time series is compared with the theoretical power generation forecast value, and the similarity between the two is calculated by a dynamic time warping algorithm; if the similarity is lower than a preset threshold, it is determined to be an abnormal data point; the abnormal data points are grouped to identify persistent abnormal patterns; for persistent abnormal patterns, keywords are extracted as input features in combination with the power station operation log and equipment status information; a decision tree algorithm is used to construct a subsidy fraud discrimination model to classify abnormal data; if it is determined to be a suspected subsidy fraud, the photovoltaic power station information is added to the list of suspected distributed photovoltaic power stations.

[0019] For example, the actual power generation data is obtained from the database of the distributed photovoltaic power station management platform, and the data is preprocessed by the time series analysis tool. The 3σ criterion is used to remove outliers, and the linear interpolation method is used to fill the missing values. The moving average and standard deviation of the power generation are calculated using a 3-day sliding window to obtain the smoothed actual power generation time series. According to the geographical location, equipment parameters and meteorological data of the photovoltaic power station, the theoretical power generation is calculated using the PVsyst software. Combined with the historical power generation data, the support vector regression algorithm (RBF kernel function, the penalty parameter C is set to 1.0) is used to train the power generation prediction model, and the calculated theoretical power generation is corrected to obtain a more accurate theoretical power generation prediction value. The actual power generation is compared with the theoretical power generation prediction value, and the dynamic time warping algorithm is used to calculate the similarity between the two. The similarity threshold is set to 0.8. If the similarity is lower than the threshold, it is marked as an abnormal data point. The abnormal data points are grouped by the DBSCAN clustering algorithm (the neighborhood radius ε is 0.1 and the minimum number of samples MinPts is 5) to identify the persistent abnormal pattern. For the identified abnormal patterns, combined with the power station operation log and equipment status information, text mining technology is used to extract keywords as input features, and the C4.5 decision tree algorithm (maximum tree depth is 5, minimum sample split number is 20) is used to build a fraudulent subsidy behavior discrimination model to classify abnormal data. If it is determined to be suspected fraudulent subsidy behavior, the photovoltaic power station information is added to the list of suspected fraudulent distributed photovoltaic power stations, and the time period and degree of abnormality are recorded. In practical applications, the daily power generation data of a 100kW photovoltaic power station in the past year are first extracted from the distributed photovoltaic power station management platform database. The 3σ criterion is applied to these data, and data that exceeds the mean ±3 times the standard deviation range is marked as anomalies, such as anomalies with daily power generation exceeding 800kWh. Linear interpolation is used to fill missing data. For example, if the data on May 15 is missing, it is estimated by the average value of May 14 and 16. A 3-day sliding window is used to calculate the moving average and standard deviation to obtain the smoothed actual power generation time series. Then, using PVsyst software, input the geographical coordinates of the power station, such as latitude a° north, longitude b° east, component inclination, such as 30°, azimuth, such as 0° south, and other parameters, combined with local meteorological data, calculate the theoretical daily power generation. Combine the calculation results with historical data, and use the support vector regression algorithm (RBF kernel function, C=1.0) to train the prediction model, such as input sunshine time, temperature and other features, and output the corrected theoretical power generation. Next, the dynamic time warping algorithm is used to calculate the similarity between the actual power generation and the theoretical power generation, and the threshold is set to 0.8. If the similarity on a certain day is 0.75, which is lower than the threshold, it is marked as an outlier. Use the DBSCAN algorithm (ε=0.1, MinPts=5) to cluster the outliers. If it is found that there are outliers for 7 consecutive days, it is identified as a persistent abnormal pattern.Finally, keywords such as "equipment failure" and "maintenance" in the power station operation log are extracted and combined with equipment status information such as inverter efficiency and component temperature as input features of the C4.5 decision tree. The decision tree depth is set to 5, and the minimum sample split number is 20 to build a subsidy fraud discrimination model. If the model determines that an abnormal pattern is suspected of subsidy fraud, such as the actual power generation exceeds the theoretical value by 20% for 7 consecutive days, the power station information, the abnormal time period from June 1 to June 7, and the abnormal degree exceeding 20% ​​are recorded in the suspected subsidy fraud list.

[0020] Step S103, for each power station in the list of distributed photovoltaic power stations suspected of fraudulent subsidies, obtain the actual voltage level and actual installed capacity data of its grid connection point, and compare them with the registered grid connection point voltage level and registered installed capacity data of the power station. If the actual value is greater than the registered value, it is determined that the power station has exceeded the capacity grid connection behavior, and the list of distributed photovoltaic power stations suspected of fraudulent subsidies is updated.

[0021] The actual grid-connected point voltage level and actual installed capacity data of the list of suspected subsidy fraud distributed photovoltaic power stations are obtained; according to the actual value and the registered value data, the actual value is matched with the registered value one by one to obtain a data set containing all the records of suspected subsidy fraud power stations; for each power station in the data set, the IF-THEN rule is used for judgment: if the actual voltage level is greater than the registered voltage level, the voltage level of the power station is marked as abnormal; if the actual installed capacity is greater than the preset percentage of the registered installed capacity, the installed capacity of the power station is marked as abnormal; according to the voltage level abnormality mark and the installed capacity abnormality mark, it is determined whether the power station has over-capacity grid-connected behavior: if at least one of the two abnormal marks is true, it is determined that the power station has over-capacity grid-connected behavior; the list of suspected subsidy fraud distributed photovoltaic power stations is updated using database transactions; abnormal voltage level values ​​and abnormal installed capacity values ​​are recorded; according to the updated list of suspected subsidy fraud distributed photovoltaic power stations, a map of abnormal power station distribution is generated, and the degree of abnormality is displayed through a heat map; a statistical report containing the number of power stations at each risk level and the distribution of abnormal types is generated.

[0022] Exemplarily, the actual grid connection point voltage level and actual installed capacity data of each power station in the list of suspected fraudulent distributed photovoltaic power stations are obtained from the power regulatory department database, and the registered grid connection point voltage level and registered installed capacity data of the corresponding power station are extracted from the power station filing system. The actual value is matched with the registered value one by one through the SQL LEFT JOIN operation, and all suspected fraudulent power station records are retained to handle incomplete data. For each power station, the IF-THEN rule is used for judgment, and the voltage level judgment standard is defined as the actual value shall not exceed the registered value, and the installed capacity judgment standard is that the actual value shall not exceed 110% of the registered value. If the actual voltage level is greater than the registered voltage level, the voltage level of the power station is marked as abnormal; if the actual installed capacity is greater than 110% of the registered installed capacity, the installed capacity of the power station is marked as abnormal. Combined with the voltage level abnormality mark and the installed capacity abnormality mark, it is determined whether the power station has over-capacity grid connection behavior. If at least one of the two abnormal marks is true, it is determined that the power station has over-capacity grid connection behavior, and the judgment result is added to the power station information. According to the results of the over-capacity grid-connected behavior, the database transaction is used to ensure the atomicity of the update operation, and the list of suspected fraudulent distributed photovoltaic power stations is updated. The risk level is defined: "extremely high risk" exists when both the voltage level and the installed capacity are abnormal, "high risk" exists when only one abnormality exists, and the original risk level is maintained when there is no abnormality. At the same time, the abnormal voltage level value and the abnormal installed capacity value are recorded to generate an updated list of suspected fraudulent distributed photovoltaic power stations. The geographic information visualization tool is used to generate a distribution map of abnormal power stations, and the degree of abnormality is displayed through a heat map. A statistical report is generated, which contains information such as the number of power stations at each risk level and the distribution of abnormal types, to provide data support for subsequent supervision. In actual implementation, a list of 100 suspected fraudulent distributed photovoltaic power stations is first extracted from the database of the power regulatory department. Each power station contains the actual grid-connected point voltage level, such as 10kV, 35kV, and the actual installed capacity, such as 500kW, 2MW. At the same time, the filing information of these 100 power stations is obtained from the power station filing system. Use SQL LEFTJOIN operation to match two sets of data, such as SELECT a., b. FROM actual data table a LEFTJOIN record data table b ON a. power station ID = b. power station ID. For incomplete data, such as 5 power stations lacking record information, LEFTJOIN ensures that these 5 power stations are still retained in the result set and marked as "missing record information". Next, apply the IF-THEN rule to make a judgment: IF actual voltage level > record voltage level THEN mark "voltage level abnormal"; IF actual installed capacity > record installed capacity * 1.1 THEN mark "installed capacity abnormal". For example, the actual voltage level of a power station is 35kV, the record is 10kV, marked as "voltage level abnormal"; the actual installed capacity is 2.2MW, the record is 2MW, marked as "installed capacity abnormal".According to the abnormal marking, the overcapacity grid-connected behavior is determined: if the power station has both "voltage level abnormality" and "installed capacity abnormality", it is judged as "extremely high risk"; only one abnormality is judged as "high risk". Use database transactions to update the list, and use geographic information visualization tools such as ArcGIS to generate a distribution map of abnormal power stations. Import the geographic coordinates and risk levels of 100 power stations and use heat maps to display them, such as "extremely high risk" in red and "high risk" in orange. At the same time, statistical reports are generated, including 15 extremely high-risk power stations and 30 high-risk power stations, with voltage level abnormalities accounting for 20% and installed capacity abnormalities accounting for 35%, providing intuitive data support for regulatory authorities.

[0023] Obtain the grid-connected point transformer capacity, grid-connected line current carrying capacity and booster station transformer capacity data of the PV power stations in the list, and determine whether they meet the electrical conditions for over-capacity grid connection based on the preset over-capacity grid-connected electrical condition threshold to further determine the list of distributed PV power stations suspected of fraudulent subsidies.

[0024] The grid-connected point transformer capacity, grid-connected line current carrying capacity and booster station transformer capacity data of each power station in the list of distributed photovoltaic power stations with subsidy fraud are obtained; the preset overcapacity grid-connected electrical condition threshold is used to establish a judgment rule base, which includes the transformer capacity utilization rate threshold, the line current carrying capacity ratio threshold and the booster station transformer capacity matching threshold; the data in the temporary data table are evaluated one by one to obtain the electrical condition satisfaction of each power station; a logistic regression algorithm is used to construct an accurate discrimination model for subsidy fraud, which takes the electrical condition satisfaction and the overcapacity grid-connected judgment result as input and outputs a subsidy fraud possibility score; if the subsidy fraud possibility score is higher than the preset judgment threshold, the corresponding power station is marked as confirmed subsidy fraud; if the subsidy fraud possibility score is lower than the judgment threshold, the corresponding power station is retained in the suspected list; if the judgment result is inconsistent with the original list, the corresponding power station is marked as pending review.

[0025] Exemplarily, the grid-connected transformer capacity, grid-connected line current carrying capacity and booster station transformer capacity data of each power station in the list of distributed photovoltaic power stations that are fraudulently subsidized are obtained from the power monitoring system database, and the current values ​​of these parameters are updated daily through scheduled tasks, and the collected data is stored in a temporary data table. At the same time, data quality checks are performed to mark and supplement missing or abnormal data. According to the preset threshold of overcapacity grid-connected electrical conditions, a judgment rule base is established using the Drools rule engine, including specific evaluation criteria such as transformer capacity utilization not exceeding 80%, line current carrying capacity not exceeding 70%, and booster station transformer capacity matching. The data in the temporary data table are evaluated one by one to obtain the electrical condition satisfaction of each power station. The electrical condition satisfaction is associated with the previous overcapacity grid-connected judgment results, and a logistic regression algorithm is used to construct an accurate discrimination model for fraudulent subsidy behavior. The electrical condition satisfaction and overcapacity grid-connected judgment results are input, and the fraudulent subsidy possibility score is output. The optimal fraudulent subsidy behavior judgment threshold is determined through ROC curve analysis. According to the output results of the accurate discrimination model for subsidy fraud, the list of distributed photovoltaic power stations that commit subsidy fraud is updated. Power stations with a possibility score higher than the judgment threshold are marked as confirmed subsidy fraud, and those below the threshold are retained in the suspected list. If the judgment result is inconsistent with the original list, it is marked as "

[0026] The consistency and integrity of the list update are ensured through database transactions. At the same time, the change log is recorded to save the list differences before and after the update. In actual application, first connect to the power monitoring database and extract a list of 200 suspected photovoltaic power stations that are suspected of fraudulent subsidies. Set a scheduled task at 2 a.m. every day to update the data of each power station. For example, the grid-connected transformer capacity of power station A is 1000kVA, the grid-connected line current is 300A, and the booster transformer capacity is 5000kVA. The data quality check found that the data of 15 power stations were missing, and the data of the previous day was used to fill it; the data of 5 power stations were abnormal, such as negative values, and the 30-day moving average was used instead. Drool The judgment criteria set by the s rule engine include a transformer capacity utilization threshold of 80%, a line current carrying capacity ratio threshold of 70%, and a booster station transformer capacity matching threshold of 95%. The evaluation results of power station A are that the transformer utilization rate is 85% (exceeding the standard), the line current carrying capacity ratio is 65% (normal), and the booster station matching degree is 98% (normal). The input variables of the logistic regression algorithm include the satisfaction of three electrical conditions and the previous overcapacity grid connection judgment results. The training samples are 100 historically confirmed fraud cases and 100 normal cases. The model outputs a fraud possibility score of 0.75 for power station A. Through ROC curve analysis, the optimal judgment threshold is determined to be 0.7. When updating the list, power station A is changed from "suspected" to "confirmed subsidy fraud". Among the 200 power stations, 30 were determined to be confirmed subsidy fraud, 160 remained suspected, and 10 were inconsistent, such as the original "confirmed" changed to "suspected", marked as "pending review". Use database transactions to update the list and record the change log: such as "The status of power station A changed from suspected to confirmed subsidy fraud, change time 2023-09-2502:15:30", save the complete list snapshot before and after the update, for subsequent review and tracing.

[0027] Step S104, for the power stations confirmed in the list to have committed subsidy fraud, obtain their grid connection time, actual power generation, and approved subsidy power, predict the cumulative power generation and cumulative approved subsidy power over the entire life cycle of the power station, calculate the difference between the two, and obtain an estimated value of the cumulative fraudulent subsidy power of the power station.

[0028] The power station data confirmed to have engaged in subsidy fraud in the power trading center database were obtained, including the grid-connected time, historical actual power generation and approved subsidy power; the Z-score method was used to detect outliers based on the power station data, and the missing values ​​were filled by the moving average method to obtain a cleaned historical data set; for the cleaned historical data set, the time series decomposition method was used to split the power generation data into trend, seasonal and random components; an approved subsidy power forecast model was established, and the approved subsidy power forecast model used a polynomial regression method to fit the relationship between historical approved subsidy power and time; the output results of the power generation forecast model and the approved subsidy power forecast model were accumulated to obtain the cumulative power generation forecast value and the cumulative approved subsidy power forecast value for the entire life cycle of the power station; the estimated value of the cumulative fraudulent subsidy power of the power station and its confidence interval were obtained by calculating the difference between the cumulative power generation forecast value and the cumulative approved subsidy power forecast value.

[0029] For example, the grid connection time, historical actual power generation and approved subsidy power data of power stations confirmed to have fraudulent subsidy behaviors in the list are obtained from the power trading center database, the Z-score method is used to detect outliers, and the moving average method is used to fill missing values ​​to obtain the cleaned historical data set. Based on the cleaned historical data set, the power generation data is split into trend, seasonal and random components using the time series decomposition method, the trend component is modeled using the ARI MA model, the optimal parameters are selected using the AIC criterion, the main cycle of the Fourier series is determined by periodogram analysis and the seasonal component is fitted, and the power generation prediction model for the power station's entire life cycle is constructed in combination with the statistical characteristics of the random component, while considering equipment aging factors, such as an annual decrease in power generation efficiency of 0.5%. According to the design life of the power station and the historical approved subsidy power data, an approved subsidy power prediction model is established, and the relationship between the historical approved subsidy power and time is fitted using the polynomial regression method. The optimal polynomial order is determined through cross-validation, and a regression model of subsidy intensity and time is established in combination with the quantitative analysis of historical subsidy policy changes to predict the approved subsidy power in future years. The output results of the power generation prediction model and the approved subsidy power prediction model are accumulated to obtain the cumulative power generation prediction value and the cumulative approved subsidy power prediction value of the power station's entire life cycle, respectively. The Monte Carlo simulation method is used to generate the prediction interval. Considering the uncertainty of the prediction results, the difference between the two and its confidence interval are obtained by difference calculation. The difference is used as the estimated value of the cumulative fraudulent subsidy power of the power station, and the accuracy of the overall prediction model is evaluated by cross-validation method. In practical applications, the system first extracts data of a 100kW photovoltaic power station from the power trading center database, including the grid connection time on June 1, 2018, 3 years of historical daily power generation and monthly approved subsidy power. The Z-score method is used to detect outliers, and the threshold is set to 3. Five abnormal daily power generation data are found; these outliers and 3 missing values ​​are filled by the 7-day moving average method. The cleaned power generation data is decomposed into time series to obtain trend, seasonality and random components. The ARIMA model parameters were selected using the AIC criterion, and the ARIMA (2,1,1) model was determined to fit the trend. The periodogram method was used to analyze that the main period was 365 days, and the third-order Fourier series was used to fit the seasonality. Considering the aging of the equipment, assuming that the annual power generation efficiency decreased by 0.5%, the power generation over the entire life cycle of 25 years was predicted. For the approved subsidized electricity, a third-order polynomial regression was used to fit the historical data, and the optimal order was determined by 5-fold cross validation. The historical subsidy policy was analyzed, and a logarithmic regression model of subsidy intensity and time was established to predict future subsidy changes. The cumulative prediction results were added up to obtain a cumulative power generation prediction value of 5,475,000 kWh and a cumulative approved subsidized electricity prediction value of 5,110,000 kWh. The Monte Carlo method was used to perform 10,000 simulations and generate a 95% confidence interval.The final estimate is that the cumulative amount of electricity fraudulently subsidized by the power station is 365,000±18,250kWh. The model is verified by the last 6 months of data through the holdout method, and the mean absolute percentage error (MAPE) is 3.2%, confirming the reliability of the prediction model.

[0030] The power generation, grid-connected power and irradiance data of the power station are collected in real time, and the power generation and grid-connected power of the power station are modeled using the time series analysis method. Through the model residual analysis, it is determined whether there are data anomalies, and the suspected fraudulent subsidy behavior is preliminarily identified. Among them, the correlation between power generation and irradiance, grid-connected power and grid-connected electricity price is analyzed, and the time window with abnormal correlation is identified as a high-risk period for fraudulent subsidy behavior, and fraud warning information is sent to management personnel at the same time.

[0031] Collect data from smart meters and irradiance sensors in real time to obtain power generation, grid-connected power and irradiance data; pre-process the collected data to obtain a processed time series data set; use the ARIMA model to model the processed power generation and grid-connected power time series data, and select the optimal model parameters through the AIC criterion. If the model residual does not obey the normal distribution, it is determined that there is data anomaly; for the data anomaly period, calculate the absolute value of the Pearson correlation coefficient between power generation and irradiance, grid-connected power and on-grid electricity price. If the absolute value of the correlation coefficient is lower than the preset threshold, the period is determined to be a high-risk period; if the period is determined to be a high-risk period, send a fraudulent subsidy warning information containing the abnormal period, abnormal type and abnormal degree to the preset power station management personnel.

[0032] For example, smart meters and irradiance sensors are deployed at key nodes of photovoltaic power stations. The power generation, grid-connected power and irradiance data of the power station are collected in real time through the data collection gateway. The sampling frequency is set to once every 5 minutes, and the collected data is transmitted to the data processing center in real time. At the same time, local cache and breakpoint resumption mechanism are set to handle data transmission interruptions, and the grid-connected electricity price data for the corresponding period is obtained from the power trading system. The collected historical data is preprocessed, including data cleaning, missing value filling and outlier detection. The Z-score method is used for outlier detection, and the threshold is set to 3 standard deviations. The sliding window method is used to smooth the data, which is implemented using the rolling function of pandas, and the window size is set to 1 hour to obtain the processed time series data set. The ARIMA model is used to model the processed power generation and grid-connected power time series data. The parameter selection range is set to p(0-5), d(0-2), q(0-5). The optimal model parameters are selected by the AIC criterion. The model residual is calculated and the Shapiro-Wilk test is used to determine whether it obeys the normal distribution. The p-value threshold is set to 0.05. If it does not obey the normal distribution, it is determined that there is data anomaly and the period is marked as suspected subsidy fraud. At the same time, the ARIMA model parameters are updated in real time using the online learning algorithm. For the period of suspected subsidy fraud, the absolute value of the Pearson correlation coefficient between power generation and irradiance, and grid-connected power and grid-connected electricity price is calculated. The optimal correlation coefficient threshold is determined by analyzing historical data through the ROC curve. If the absolute value of the correlation coefficient is lower than the threshold, the period is determined as a high-risk period, and the early warning mechanism is triggered to send subsidy fraud warning information containing abnormal period, abnormal type and abnormal degree to the preset power station management personnel. In the implementation of a 100kW photovoltaic power station, smart meters are installed at the grid connection point and irradiance sensors are installed on the roof, and data is collected every 5 minutes. The data acquisition gateway is equipped with 1GB local storage. When a network interruption is detected, the local cache is automatically started, and the data is resumed after the connection is restored. 288 sets of data points are collected in one day, including power generation kWh, grid-connected power kWh, irradiance W / m2 and the corresponding grid-connected electricity price (yuan / kWh). The Z-score method is used to detect outliers, and the threshold is set to 3, and 5 outliers are found; a 1-hour sliding window and 12 data points are used for smoothing. The processed power generation and grid-connected power data are modeled using the ARIMA model, and the optimal parameters are determined to be ARIMA (2,1,1) through grid search. The Shapiro-Wilk test for residual normality is used, and the p value is 0.03, which is less than the threshold of 0.05, and it is determined that there is an abnormality. The recursive least squares method is used to update the ARIMA model parameters in real time. The correlation coefficient is calculated for the abnormal period, such as the correlation coefficient between power generation and irradiance is 0.75, and the correlation coefficient between grid-connected power and grid-connected power price is -0.6. The optimal correlation coefficient threshold is determined to be 0.78 through ROC curve analysis.The time periods with an absolute value of the correlation coefficient lower than 0.78 (such as 14:00-15:00) are identified as high-risk time periods. An early warning text message is sent to the power station administrator: "Suspected fraudulent subsidy behavior was detected during the 14:00-15:00 period. The correlation between power generation and irradiance is abnormal. Please check."

[0033] A subsidy fraud identification model is established, combining multiple features such as power generation, irradiance, grid-connected power and electricity price to make real-time judgments on subsidy fraud. Based on the judgment results of the subsidy fraud identification model, the identification threshold and early warning strategy are adaptively adjusted; if subsidy fraud is identified, power restrictions and power outages are implemented, and relevant data is transmitted to the grid-connected dispatching system to reduce the subsidy for the power station.

[0034] Acquire multi-dimensional data including power generation, irradiance, grid-connected power and electricity price; perform feature engineering processing based on the multi-dimensional data, and the feature engineering processing includes data standardization, feature combination and principal component analysis dimensionality reduction; use support vector machine algorithm to build a subsidy fraud behavior identification model, and the support vector machine algorithm uses RBF kernel function, and the kernel function parameters and penalty factors are determined by grid search method; for the output results of the subsidy fraud behavior identification model, use the sliding time window method to calculate the change trend of the subsidy fraud probability score; according to the change trend, dynamically adjust the recognition threshold and warning level in combination with the historical data distribution to obtain an adaptive warning strategy; if the subsidy fraud probability score is greater than the preset threshold, the corresponding power restriction or power outage instruction is triggered, and the preset threshold includes the power restriction threshold and the power outage threshold; according to the subsidy fraud probability score, the power station subsidy is reduced in real time, and the reduction adopts the linear proportional method.

[0035] For example, multi-dimensional data such as power generation, irradiance, grid-connected power and electricity price are obtained every minute from the real-time monitoring of the power station. The original data is preprocessed through feature engineering, including data standardization and feature combination, and the principal component analysis method is used for dimensionality reduction, retaining 95% of the variance information, and constructing the input feature vector of the support vector machine model. The support vector machine algorithm is used to train the subsidy fraud behavior identification model, using the RBF kernel function, using the SMOTE algorithm to deal with the data imbalance problem, and the kernel function parameters and penalty factors are optimized by the grid search method. The model parameters are updated every hour, and the real-time data is used to determine the subsidy fraud behavior, and the subsidy fraud probability score is output. According to the judgment results of the support vector machine model, the 24-hour sliding time window method is used with a step size of 1 hour to calculate the change trend of the subsidy fraud probability score. The Z-score method is used in combination with the historical data distribution to dynamically adjust the recognition threshold and warning level to achieve adaptive warning strategy optimization. The subsidy fraud identification and early warning system is interfaced with the power station management system and the grid-connected dispatching system. When subsidy fraud is identified, power restriction or power outage instructions are automatically triggered according to the early warning level. When the probability of subsidy fraud is greater than 0.8, power restriction is triggered, and when it is greater than 0.95, power outage is triggered. The instructions are executed in real time. At the same time, the relevant data is transmitted to the grid-connected dispatching system, and the subsidy of the power station is reduced in real time according to the preset rules. The linear proportional method is used. For every increase of 0.1 in the probability of subsidy fraud, 10% of the subsidy is reduced. In the implementation of a 100kW photovoltaic power station, the system collects data from the monitoring equipment every minute, including power generation of 70kWh, irradiance of 800W / m2, grid-connected power of 68kWh and electricity price of 0.8 yuan / kWh. These data are standardized, the values ​​are mapped to the interval of 0-1, and then the features such as power generation efficiency, power generation / irradiance = 0.0875 are combined. The principal component analysis method is used to reduce the dimension, retaining the first 3 principal components, covering 95.5% of the variance information. The support vector machine with RBF kernel function was used for modeling, and the SMOTE algorithm was used to generate 100 synthetic subsidy fraud samples to balance the ratio of positive and negative samples. The optimal parameters were determined by grid search: C = 10, gamma = 0.01. The model was updated every hour, and the latest judgment result showed that the probability of subsidy fraud was 0.75. Using a 24-hour sliding window, a total of 1440 data points were used to analyze the trend, and it was found that the probability of subsidy fraud continued to rise in the past 6 hours. Using the Z-score method, the current probability of 0.75 was standardized to 2.1 standard deviations, which was higher than the historical average. The system dynamically adjusted the warning threshold to 0.7. Since 0.75 exceeded 0.7 but did not reach 0.8, a yellow warning was triggered and an alarm was sent to the management platform. If the probability continues to rise to 0.85, a 30% power restriction order will be immediately executed. At the same time, the subsidy reduction amount is calculated as the original subsidy of 100 yuan / day, the current probability of subsidy fraud is 0.75, and it is calculated by subtracting 10% for every 0.1, a reduction of 50%, and an actual payment of 50 yuan / day.

[0036] Step S105: Calculate the life cycle subsidy power of the distributed photovoltaic power station based on the estimated value of the cumulative fraudulent electricity of the distributed photovoltaic power station, and compare it with the actual cumulative power generation of the distributed photovoltaic power station. If the actual cumulative power generation exceeds the life cycle subsidy power, generate an early warning message and issue an on-site verification instruction. For distributed photovoltaic power stations that have exceeded the power generation capacity during on-site verification, multiply the estimated value of the cumulative fraudulent electricity by the subsidy price to obtain the amount of fraudulent funds, and stop disbursing subsequent subsidy funds.

[0037] Obtain the estimated value of the cumulative fraudulent electricity volume and the actual cumulative power generation data of the distributed photovoltaic power station; calculate the subsidy volume of the distributed photovoltaic power station throughout its life cycle, and the subsidy volume of the entire life cycle is obtained based on the design life of the power station, historical subsidy policy and annual attenuation rate; compare the actual cumulative power generation with the subsidy volume of the entire life cycle, and determine whether the warning mechanism is triggered based on the preset warning level standard; if the warning mechanism is triggered, generate warning information containing basic information of the power station, over-generation data and warning level; push the warning information to the on-site verification workflow, generate a verification task order and assign it to relevant personnel; determine whether there is over-generation based on the on-site verification results, and if there is over-generation, calculate the amount of fraudulent subsidy funds, which is obtained by multiplying the estimated value of the cumulative fraudulent electricity volume with the weighted average subsidy electricity price; update the renewable energy subsidy project database, and delete the information of the power station involved from the subsidy project database list; send a stop payment instruction to the subsidy issuance system, requiring the return of the excess subsidy that has been issued.

[0038] For example, the estimated value of the cumulative fraudulent electricity and the actual cumulative power generation data of the distributed photovoltaic power station are obtained from the power trading center database. According to the 25-year design life of the power station and the historical subsidy policy, the ARIMA model is used to calculate the subsidized electricity of the power station throughout its life cycle, taking into account the annual attenuation rate of 0.5% and possible policy changes of the power station. The actual cumulative power generation of the power station is compared with the calculated subsidy power of the whole life cycle using the data comparison tool, and the warning level standard is set. If it exceeds 5%, it is a yellow warning, and 10% is a red warning. If the actual cumulative power generation exceeds the subsidy power of the whole life cycle, the warning mechanism is triggered, and the warning information containing the basic information of the power station, the over-power generation data and the warning level is automatically generated. Through the power station management system interface, the warning information is pushed to the on-site verification workflow, and the verification task list is automatically generated and assigned to the relevant personnel. At the same time, the remote monitoring module is started to collect data such as power generation and irradiance every 5 minutes to provide data support for on-site verification and ensure the objectivity and accuracy of the verification. According to the on-site verification results, if it is confirmed that there is over-generation, the financial system interface is called, the weighted average method is used to calculate the historical subsidy electricity price, the estimated value of the power station's cumulative fraudulent electricity is multiplied by the weighted average subsidy electricity price, the amount of fraudulent funds is calculated, and the renewable energy subsidy project database is updated at the same time, the power station information involved is deleted from the list, and a stop payment instruction is sent to the subsidy issuance system, requiring the return of the excess subsidy that has been issued. In the actual application of a 100kW distributed photovoltaic power station, the system extracts from the database the estimated value of the power station's 5-year cumulative fraudulent electricity of 50,000kWh, and the actual cumulative power generation is 950,000kWh. Considering the 25-year design life and 0.5% annual decay rate, the ARI MA (2,1,1) model is used to predict that the life cycle subsidy electricity is 4,500,000kWh. The data comparison tool calculates the ratio of actual power generation to predicted value and obtains 21.11%, triggering a red warning. The system automatically generates warning information, including basic information such as power station ID, installed capacity, grid-connected time, over-generation of 50,000kWh and red warning level. The early warning information is pushed to the verification system through the API, and the task number NC20230925001 is generated. The remote monitoring module is started, and data is collected every 5 minutes, including power generation of 75kWh, irradiance of 850W / m2, component temperature of 42°C, etc. After the on-site verification confirms the excess power generation, the financial system calls the historical subsidized electricity price data: 0.42 yuan / kWh in the first 3 years, 0.37 yuan / kWh in the next 2 years, and the weighted average method is used to calculate 0.40 yuan / kWh. Multiply 50,000 kWh by 0.40 yuan / kWh to get 20,000 yuan in fraudulent subsidies. The system automatically deletes the power station record from the subsidy project database, sends a stop payment instruction to the subsidy issuance system, and generates a recovery notice, requiring the power station to return the 20,000 yuan excess subsidy within 30 days.

[0039] Conduct a quantitative assessment of subsidy fraud by photovoltaic power stations, obtain a quantitative scoring result for subsidy fraud, determine the severity of subsidy fraud by the power station based on the scoring grade classification rules, include power stations whose subsidy fraud severity exceeds the preset threshold in the subsidy fraud list, and publicize the subsidy fraud list through designated information disclosure channels.

[0040] The basic data of the power station to be evaluated is obtained from the photovoltaic power station management database, and the basic data includes power generation, grid-connected power and irradiance; the basic data is cleaned and standardized to obtain a data set that meets the evaluation requirements; a multi-level fuzzy comprehensive evaluation method is used to quantitatively evaluate the data set to obtain a quantitative scoring result of fraudulent subsidies; the quantitative scoring result of fraudulent subsidies is judged according to the preset scoring level classification rules to determine the severity of the power station fraudulent subsidies; if the severity of the fraudulent subsidies exceeds the preset threshold, the power station information is added to the fraudulent subsidy list database; the fraudulent subsidy list is pushed to the designated information disclosure platform through the data interface; a disclosure cycle and an automatic update mechanism are set to realize regular disclosure and dynamic management of the fraudulent subsidy list; the complaint information fed back by the information disclosure platform is received, and the complaint information is processed according to the preset review process; the fraudulent subsidy list is desensitized to obtain a desensitized fraudulent subsidy list; the desensitized fraudulent subsidy list is updated to the information disclosure platform.

[0041] Exemplarily, the basic data of the power station to be evaluated, including power generation, grid-connected power, irradiance, etc., are obtained from the photovoltaic power station management database. The raw data is cleaned and standardized through the data preprocessing module, the Z-score method is used to detect outliers, and the mean is used to fill in missing data to obtain a data set that meets the evaluation requirements. According to the preset fraud subsidy quantitative indicator system, the multi-level fuzzy comprehensive evaluation method is used to quantitatively evaluate the processed data, specifically including establishing an evaluation indicator system, such as power generation abnormality index, grid-connected power fluctuation rate, irradiance utilization efficiency, etc., using the hierarchical analysis method to determine the indicator weight, constructing a fuzzy relationship matrix, calculating the membership of each indicator, and combining the indicator weight to obtain the quantitative scoring result of the power station's fraud subsidy. The quantitative scoring result of the fraud subsidy is compared with the scoring grade division rule, and the score interval and the corresponding severity level are set. 0-60 points are mild, 61-80 points are moderate, and 81-100 points are severe. The interval mapping algorithm is used to determine the severity of the power station's fraud subsidy. If the severity exceeds the preset threshold of 80 points, the power station information is added to the fraud subsidy list database. The list of fraudulent subsidies is pushed to the designated information disclosure platform through the data interface, and digital signature technology is used to ensure the integrity and non-tamperability of the data. At the same time, the disclosure cycle and automatic update mechanism are set to realize the regular disclosure and dynamic management of the list of fraudulent subsidies. Set up complaint channels and review processes to handle feedback after the disclosure. Use data desensitization technology, such as hiding specific power generation data, only displaying the scoring results and severity levels, to protect sensitive information. In the actual application of a 100kW photovoltaic power station, the system extracts the daily power generation, grid-connected power and irradiance data of the power station for the past 30 days from the database. The Z-score method is used to detect outliers, and the threshold is set to 3. 2 days of data anomalies are found; the 30-day average value is used to fill these anomalies and 3 days of missing data. After preprocessing, a quantitative indicator system for fraudulent subsidies is constructed, including power generation anomaly index, grid-connected power fluctuation rate and irradiance utilization efficiency, with weights of 0.4, 0.3 and 0.3 respectively. Using the multi-level fuzzy comprehensive evaluation method, each indicator is standardized to the range of 0-1, such as the power generation abnormality index of 0.75, the grid-connected power fluctuation rate of 0.6, and the irradiance utilization efficiency of 0.8. The final score of 85 points is calculated by the fuzzy relationship matrix. The score level is set to 0-60 points for mild, 61-80 points for moderate, and 81-100 points for severe. 85 points is mapped to severe subsidy fraud. If the threshold exceeds 80 points, the system automatically adds the power station to the subsidy fraud list. The list is pushed to the public disclosure platform through an encrypted API, and the digital signature is generated using the SHA-256 algorithm. In the public information, the actual power generation data of the power station is replaced by asterisks, and only the "severe subsidy fraud" level is displayed. The system sets a 7-day public disclosure period, during which one complaint is received, triggering the automatic review process.

[0042] The description of the above embodiments is only used to help understand the technical solutions and core ideas of the present application. Ordinary technical personnel in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A monitoring data mining and prediction analysis method for a distributed photovoltaic power generation system, characterized in that: The method comprises: Obtain the conversion efficiency of distributed photovoltaic modules, array layout mode and photovoltaic support tilt setting parameters, conduct correlation analysis on the power generation of photovoltaic power stations and predict the power generation to obtain the theoretical power generation prediction value; Obtain the actual power generation data of distributed photovoltaic power stations and compare it with the theoretical power generation forecast value. If the actual power generation is significantly higher than the theoretical power generation forecast value, it is determined that the photovoltaic power station is suspected of fraudulent subsidy, and a list of distributed photovoltaic power stations suspected of fraudulent subsidy is generated; For each power station in the list of suspected fraudulent distributed photovoltaic power stations, obtain the actual voltage level and actual installed capacity data of its grid connection point, and compare them with the registered grid connection point voltage level and registered installed capacity data of the power station. If the actual value is greater than the registered value, it is determined that the power station has exceeded the capacity grid connection behavior, and the list of suspected fraudulent distributed photovoltaic power stations is updated; For power stations confirmed to have engaged in subsidy fraud in the list, their grid connection time, actual power generation, and approved subsidy power are obtained. The cumulative power generation and cumulative approved subsidy power of the power station over its entire life cycle are predicted, and the difference between the two is calculated to obtain an estimated value of the cumulative subsidy fraud of the power station. Based on the estimated value of the cumulative fraudulent electricity volume of the distributed photovoltaic power station, the subsidized electricity volume of the power station over its entire life cycle is calculated and compared with the actual cumulative power generation of the power station. If the actual cumulative power generation exceeds the subsidized electricity volume over its entire life cycle, an early warning message is generated and an on-site verification order is issued. For distributed photovoltaic power stations that are found to have exceeded the power generation volume during on-site verification, the estimated value of the cumulative fraudulent electricity volume is multiplied by the subsidized electricity price to obtain the amount of fraudulent subsidies, and the subsequent subsidy funds are stopped at the same time.

2. The method according to claim 1, characterized in that The method of obtaining the conversion efficiency of distributed photovoltaic modules, array layout mode and photovoltaic support inclination angle setting parameters, performing correlation analysis on the power generation of the photovoltaic power station and predicting the power generation to obtain a theoretical power generation prediction value includes: According to the technical parameters and geographical location information of photovoltaic modules, the irradiance at different time periods is calculated, and temperature correction is performed to obtain the actual irradiance and temperature data of photovoltaic modules under various working conditions; According to the layout of photovoltaic modules and the setting parameters of the bracket inclination angle, the shadow blocking of photovoltaic modules is calculated to determine the actual output power of photovoltaic strings at different times; Establish a power generation prediction model to predict the theoretical power generation in each period; The losses in each link of the photovoltaic system are quantitatively analyzed, multiple groups of random samples of system losses are generated, the confidence interval of the comprehensive system efficiency is calculated, and the theoretical power generation prediction value is corrected to obtain the final prediction result.

3. The method according to claim 1, characterized in that: The actual power generation data of the distributed photovoltaic power station is obtained and compared with the theoretical power generation forecast value. If the actual power generation is significantly higher than the theoretical power generation forecast value, it is determined that the photovoltaic power station is suspected of fraudulent subsidy, and a list of distributed photovoltaic power stations suspected of fraudulent subsidy is generated, including: Obtain actual power generation data from the distributed photovoltaic power station management platform database; The actual power generation data is preprocessed by using a time series analysis tool to obtain a smoothed actual power generation time series; According to the geographical location, equipment parameters and meteorological data of the photovoltaic power station, the theoretical power generation is calculated to obtain the theoretical power generation forecast value; The smoothed actual power generation time series is compared with the theoretical power generation forecast value, and the similarity between the two is calculated using the dynamic time warping algorithm; If the similarity is lower than the preset threshold, it is determined to be an abnormal data point; Group anomalous data points to identify persistent anomalous patterns; For persistent abnormal patterns, keywords are extracted as input features by combining power plant operation logs and equipment status information; A decision tree algorithm is used to build a fraudulent subsidy discrimination model to classify abnormal data; If it is determined to be suspected subsidy fraud, the photovoltaic power station information will be added to the list of distributed photovoltaic power stations suspected of subsidy fraud.

4. The method according to claim 1, characterized in that: For each power station in the list of suspected fraudulent distributed photovoltaic power stations, the actual voltage level and actual installed capacity data of the grid connection point are obtained, and compared with the registered grid connection point voltage level and registered installed capacity data of the power station. If the actual value is greater than the registered value, it is determined that the power station has over-capacity grid connection behavior, and the list of suspected fraudulent distributed photovoltaic power stations is updated, including: Obtain the actual grid connection point voltage level and actual installed capacity data of the list of distributed photovoltaic power stations suspected of fraudulent subsidies; According to the actual value and the recorded value data, the actual value and the recorded value are matched one by one to obtain a data set containing records of all power stations suspected of fraudulent subsidies; For each power station in the data set, the IF-THEN rule is used for judgment: if the actual voltage level is greater than the registered voltage level, the voltage level of the power station is marked as abnormal; If the actual installed capacity is greater than the preset percentage of the registered installed capacity, the installed capacity of the power station is marked as abnormal; According to the voltage level abnormality mark and the installed capacity abnormality mark, determine whether the power station has overcapacity grid-connected behavior: if at least one of the two abnormality marks is true, it is determined that the power station has overcapacity grid-connected behavior; Use database transactions to update the list of distributed photovoltaic power stations suspected of fraudulent subsidies; Record abnormal voltage level values ​​and abnormal installed capacity values; Based on the updated list of distributed photovoltaic power stations suspected of fraudulent subsidies, a distribution map of abnormal power stations is generated, and the degree of abnormality is displayed through a heat map; Generate statistical reports including the number of power plants at each risk level and the distribution of abnormal types; It also includes: obtaining the grid-connected point transformer capacity, grid-connected line current carrying capacity and booster station transformer capacity data of the photovoltaic power stations in the list, and judging whether the electrical conditions for over-capacity grid connection are met based on the preset over-capacity grid-connected electrical condition threshold to further determine the list of distributed photovoltaic power stations suspected of fraudulent subsidies.

5. The method according to claim 4, characterized in that The data of the grid-connected point transformer capacity, grid-connected line current carrying capacity and booster station transformer capacity of the photovoltaic power station in the list are obtained, and according to the preset overcapacity grid-connected electrical condition threshold, whether the overcapacity grid-connected electrical condition is met to further determine the list of suspected fraudulent distributed photovoltaic power stations, including: Obtain the grid-connected point transformer capacity, grid-connected line current carrying capacity and booster station transformer capacity data for each power station in the list of distributed photovoltaic power stations that have been defrauded of subsidies; The preset over-capacity grid-connected electrical condition threshold is adopted to establish a judgment rule base, which includes the transformer capacity utilization rate threshold, the line current carrying capacity ratio threshold and the booster station transformer capacity matching threshold; Evaluate the data in the temporary data table one by one to determine whether the electrical conditions of each power station are met; A logistic regression algorithm is used to build an accurate discrimination model for subsidy fraud. The accurate discrimination model for subsidy fraud takes the electrical condition satisfaction and the overcapacity grid connection judgment results as inputs, and outputs a subsidy fraud possibility score. If the subsidy fraud possibility score is higher than the preset judgment threshold, the corresponding power station will be marked as confirmed subsidy fraud; If the subsidy fraud possibility score is lower than the judgment threshold, the corresponding power station will be retained in the suspected list; If the determination result is inconsistent with the original list, the corresponding power station will be marked as pending review.

6. The method according to claim 1, characterized in that For the power stations confirmed in the list to have committed fraudulent subsidy, their grid connection time, actual power generation, and approved subsidy power are obtained, the cumulative power generation and cumulative approved subsidy power of the power station over its entire life cycle are predicted, and the difference between the two is calculated to obtain an estimated value of the cumulative fraudulent subsidy power of the power station, including: Obtain data on power plants confirmed to have engaged in subsidy fraud in the power trading center database. The power plant data includes the time of grid connection, historical actual power generation, and approved subsidy power; The Z-score method is used to detect outliers based on the power station data, and the missing values ​​are filled by the moving average method to obtain the cleaned historical data set; For the cleaned historical data set, the time series decomposition method is used to split the power generation data into trend, seasonal and random components; Establish a subsidy-approved electricity forecasting model, which uses a polynomial regression method to fit the relationship between historical subsidy-approved electricity and time; The output results of the power generation prediction model and the approved subsidy power prediction model are accumulated to obtain the cumulative power generation prediction value and the cumulative approved subsidy power prediction value for the entire life cycle of the power station; By calculating the difference between the cumulative power generation forecast and the cumulative approved subsidy power forecast, the estimated value of the cumulative fraudulent power generation of the power station and its confidence interval are obtained; It also includes: real-time collection of power plant power generation, grid-connected power and irradiance data, using time series analysis methods to model the power plant power generation and grid-connected power, and through model residual analysis, determine whether there are data anomalies, and preliminarily identify suspected subsidy fraud. Among them, analyze the correlation between power generation and irradiance, and the correlation between grid-connected power and on-grid electricity price, identify the time window with abnormal correlation as a high-risk period for subsidy fraud, and send subsidy fraud warning information to management personnel at the same time; establish a subsidy fraud identification model, combine multiple characteristics of power generation, irradiance, grid-connected power and electricity price, and make real-time judgments on subsidy fraud, and adaptively adjust the identification threshold and warning strategy according to the judgment results of the subsidy fraud identification model; if subsidy fraud is identified, power rationing and power outages will be adopted, and relevant data will be transmitted to the grid-connected dispatching system to reduce the subsidy for the power plant.

7. The method according to claim 6, characterized in that The power generation, grid-connected power and irradiance data of the power station are collected in real time, and the power generation and grid-connected power of the power station are modeled by time series analysis method. Through model residual analysis, it is judged whether there are data anomalies, and suspected fraudulent subsidy behaviors are preliminarily identified. Among them, the correlation between power generation and irradiance, grid-connected power and grid-connected electricity price is analyzed, and the time window with abnormal correlation is identified as a high-risk period for fraudulent subsidy behaviors. At the same time, fraud warning information is sent to management personnel, including: Collect data from smart meters and irradiance sensors in real time to obtain power generation, grid-connected power and irradiance data; Preprocess the collected data to obtain a processed time series data set; The ARIMA model is used to model the processed power generation and grid-connected power time series data, and the optimal model parameters are selected by the AIC criterion. If the model residual does not obey the normal distribution, it is determined that there is data anomaly; For the data abnormal period, the absolute value of the Pearson correlation coefficient between power generation and irradiance, grid-connected power and grid-connected electricity price is calculated. If the absolute value of the correlation coefficient is lower than the preset threshold, the period is determined to be a high-risk period; If the time period is determined to be a high-risk period, a subsidy fraud warning message including the abnormal time period, abnormal type and abnormal degree will be sent to the preset power station management personnel.

8. The method according to claim 6, Among them, it is characterized in that the subsidy fraud behavior identification model is established, and multiple characteristics such as power generation, irradiance, grid-connected power and electricity price are combined to make real-time judgments on subsidy fraud behaviors, and the identification threshold and early warning strategy are adaptively adjusted according to the judgment results of the subsidy fraud behavior identification model; if subsidy fraud behaviors are identified, power restrictions and power outages are adopted, and relevant data are transmitted to the grid-connected dispatching system to reduce the subsidies for power stations, including: Obtain multi-dimensional data including power generation, irradiance, grid-connected power and electricity price; Perform feature engineering based on multi-dimensional data, including data standardization, feature combination, and principal component analysis dimensionality reduction; The support vector machine algorithm is used to build a subsidy fraud identification model. The support vector machine algorithm uses the RBF kernel function, and the kernel function parameters and penalty factors are determined by the grid search method. Based on the output results of the subsidy fraud identification model, the sliding time window method is used to calculate the change trend of the subsidy fraud probability score; According to the changing trend, the recognition threshold and warning level are dynamically adjusted in combination with the historical data distribution to obtain an adaptive warning strategy; If the probability score of subsidy fraud is greater than a preset threshold, a corresponding power restriction or power outage instruction is triggered. The preset threshold includes a power restriction threshold and a power outage threshold; The power station subsidies are reduced in real time based on the probability score of subsidy fraud, using the linear proportional method.

9. The method according to claim 1, characterized in that: The cumulative estimated value of fraudulent electricity of distributed photovoltaic power stations is used to calculate the subsidy electricity of the power station throughout its life cycle, and the value is compared with the actual cumulative power generation of the power station. If the actual cumulative power generation exceeds the subsidy electricity of the whole life cycle, an early warning message is generated, and an on-site verification instruction is issued. For distributed photovoltaic power stations that have exceeded the power generation capacity during on-site verification, the estimated value of fraudulent electricity is multiplied by the subsidy electricity price to obtain the amount of fraudulent funds, and the subsequent subsidy funds are stopped at the same time, including: Obtain the estimated value of the cumulative fraudulent electricity volume and the actual cumulative power generation data of distributed photovoltaic power stations; Calculate the subsidy amount of distributed photovoltaic power stations over their entire life cycle, which is based on the design life of the power station, historical subsidy policies, and annual attenuation rate; Compare the actual cumulative power generation with the life cycle subsidy power, and determine whether to trigger the early warning mechanism based on the preset early warning level standards; If the early warning mechanism is triggered, early warning information including basic information of the power plant, over-generation data and early warning level will be generated; Push the early warning information to the on-site verification workflow, generate verification task orders and assign them to relevant personnel; Determine whether there is over-generation based on the on-site verification results. If there is over-generation, calculate the amount of fraudulent subsidies, which is calculated by multiplying the estimated cumulative fraudulent electricity generation by the weighted average subsidized electricity price; Update the renewable energy subsidy project database and delete the power plant information involved from the subsidy project database list; Send a stop payment instruction to the subsidy disbursement system, requiring the return of excess subsidies already disbursed; It also includes: conducting a quantitative assessment of the subsidy fraud behavior of photovoltaic power stations, obtaining a quantitative scoring result for subsidy fraud, determining the severity of the subsidy fraud of the power station based on the scoring grade classification rules, including power stations whose subsidy fraud severity exceeds the preset threshold in the subsidy fraud list, and publicizing the subsidy fraud list through designated information disclosure channels.

10. The method according to claim 9, characterized in that The subsidy fraud behavior of the photovoltaic power station is quantitatively evaluated to obtain the subsidy fraud quantitative scoring result, and the severity of the subsidy fraud of the power station is determined in combination with the scoring level classification rules, and the power station whose subsidy fraud severity exceeds the preset threshold is included in the subsidy fraud list, and the subsidy fraud list is publicized through the designated information disclosure channel, including: Obtain basic data of the power plant to be evaluated from the photovoltaic power plant management database, including power generation, grid-connected power and irradiance; Clean and standardize the basic data to obtain a data set that meets the evaluation requirements; A multi-level fuzzy comprehensive evaluation method was used to quantitatively evaluate the data set and obtain the quantitative scoring results of fraudulent subsidies; Determine the quantitative scoring results of subsidy fraud according to the preset scoring grading rules and determine the severity of subsidy fraud of power stations; If the severity of subsidy fraud exceeds the preset threshold, the power plant information will be added to the subsidy fraud list database; Push the subsidy fraud list to the designated information disclosure platform through the data interface; Set up a publicity cycle and automatic update mechanism to achieve regular publicity and dynamic management of the subsidy fraud list; Receive complaint information from the information disclosure platform and handle the complaint information according to the preset review process; Perform data desensitization on the fraud subsidy list to obtain a desensitized fraud subsidy list; The de-sensitized list of subsidy frauds will be updated to the information disclosure platform.

Images