Photovoltaic power generation power prediction method and system

Abstract

The present application relates to photovoltaic power generation prediction technical field, specifically disclose photovoltaic power generation power prediction method and system, including first abnormal value is carried out abnormal reason judgment, abnormal reason includes snow removal causes and other causes, based on the first abnormal value caused by snow removal and constructs compensation sequence and obtains the first prediction model of the to-be-corrected sequence according to different snowfall grades, the to-be-corrected sequence is corrected through compensation sequence, the present application is further analyzed by the first abnormal value caused by the snow removal equipment starting under the snow weather, so as to optimize the prediction process under the snow weather, so as to optimize the photovoltaic power generation prediction method of removing abnormal value at present mainstream, the robustness of prediction result under the snow weather and the operation of snow removal equipment is improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation prediction technology, specifically to a method and system for predicting photovoltaic power generation. Background Technology

[0002] Predicting photovoltaic (PV) power generation is beneficial for electricity planning, thus it is of great significance. Weather factors are a crucial influence on PV power generation, and various weather-based prediction technologies exist. For example, Chinese invention patent CN115529002A combines power data with weather data to strengthen the correlation between PV power and various environmental factors. It also considers the varying degrees of impact of different weather conditions on PV equipment, classifying basic indicators and calculating their weights, thereby increasing the causal relationship between environmental factors and power generation and improving the accuracy of subsequent PV power generation predictions. While weather classification methods can improve prediction accuracy, the accuracy remains low due to the unique characteristics of rain and snow. This is especially true now that some PV power generation systems are equipped with automatic snow removal devices. For a period after the snow removal devices are activated, the PV power generation system can still operate. The interference from the snow removal devices reduces the robustness of the clustered power generation prediction method, making it more prone to errors and deviations in the output results when the snow removal devices are activated.

[0003] In view of this, the present invention proposes a method and system for predicting photovoltaic power generation. By analyzing the data generated during the operation of snow removal equipment, the power generation efficiency of photovoltaic power plants under snowfall weather can be predicted more accurately, while providing planning and management for the operation of snow removal equipment. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for predicting photovoltaic power generation, and to solve the following technical problems:

[0005] How can we analyze the data generated during the operation of snow removal equipment to more accurately predict the power generation efficiency of photovoltaic power plants under snowfall, and at the same time provide planning and management for the operation of snow removal equipment?

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] The photovoltaic power generation prediction method includes the following steps:

[0008] Obtain forecast weather data for the predicted snowfall date of the target photovoltaic power station and obtain the snowfall coefficient based on the forecast weather data;

[0009] Snowfall days are classified into several snowfall levels based on the snowfall coefficient and the start-up time of snow removal equipment;

[0010] The predicted weather data is substituted into the first prediction model to predict the power generation of the target photovoltaic power station on the predicted snowfall day and obtain the first prediction sequence.

[0011] Acquire and store the actual power generation data of each power generation unit of the target photovoltaic power station on the predicted snowfall day of different snowfall levels, compare the first prediction sequence with the actual power generation data, obtain multiple first outliers and record the position of the first outlier, the position of the outlier is the time of occurrence of the outlier;

[0012] The cause of the first outlier is determined, including causes caused by snow removal and other causes. A compensation sequence is constructed based on the first outlier caused by snow removal.

[0013] The first prediction model's sequence to be corrected is obtained based on different snowfall levels, and the sequence to be corrected is corrected using the compensation sequence.

[0014] The second prediction sequence is output using the modified first prediction model.

[0015] The above technical solution provides a method for predicting the power generation of photovoltaic power plants under snowfall. This invention further analyzes and utilizes the first outlier caused by the activation of snow removal equipment under snowfall, thereby optimizing the prediction process under snowfall and improving the robustness of the prediction results under snowfall and snow removal equipment operation.

[0016] As a further technical solution of the present invention: the process of obtaining the snowfall coefficient includes:

[0017] Standardize the predicted snowfall duration and the predicted average snowfall amount;

[0018] The snowfall coefficient is calculated using the standardized predicted snowfall duration as the base and the standardized predicted average snowfall amount as the exponent.

[0019] As a further technical solution of the present invention: the process of classifying snowfall levels includes:

[0020] Divide the range of snowfall coefficient values ​​into at least three equally spaced initial intervals;

[0021] The activation coefficient of each photovoltaic power station falling within the initial range is calculated based on historical data of snowfall days;

[0022] The average start-up coefficient of photovoltaic power stations on several snowy days corresponding to the snowfall coefficient at the midpoint of the initial interval is selected as the benchmark value, and continuous identification and search are performed on both sides of the initial interval at preset intervals.

[0023] Based on the identified search results, several second intervals are constructed, and different snowfall levels are set for different second intervals according to their numerical values.

[0024] The above technical solution provides a process for constructing different snowfall levels. The snowfall level of the present invention is set based on the snowfall process and the activation status of snow removal equipment during the snowfall process. The greater the snowfall or the more times the snow removal equipment is activated during the snowfall process, the higher the snowfall level. After the snowfall level of the present invention is divided, the snowfall time and the number of times the snow removal equipment is activated during the snowfall process are similar within each snowfall level, which can provide a basis for subsequent prediction result correction.

[0025] As a further technical solution of the present invention: the identification and search process includes:

[0026] Set the average start-up coefficient of photovoltaic power stations corresponding to several snowfall days at the midpoint of the initial interval as the benchmark value, and calculate the difference between the benchmark value and the average start-up coefficient of photovoltaic power stations corresponding to the next snowfall coefficient during the identification and search process;

[0027] If the difference is not greater than the critical value, the identification search is considered successful, and the average starting coefficient of the photovoltaic power station corresponding to the next snowfall coefficient is used to replace the original benchmark value for the next round of identification search.

[0028] If the difference is less than the critical value, the identification search is judged to have failed. The next snowfall coefficient during the identification search process is recorded as one endpoint of the second interval. Continuous identification search is performed on both sides of the initial interval to obtain one endpoint each. The value range between the two endpoints is the second interval.

[0029] As a further technical solution of the present invention: the process of obtaining the activation coefficient of the snow removal equipment of a target photovoltaic power station on a snowfall day includes:

[0030] Through the formula:

[0031]

[0032] in This refers to the activation coefficient of the corresponding device. , and These are the preset fitting parameters. It is a preset startup state function and , This refers to the number of times the device has been started. It is a functional relationship between the number of times the device is started and the start time;

[0033] Anomaly detection was performed on the activation coefficients of all snow removal equipment in the target photovoltaic power station. The activation coefficient of the target photovoltaic power station is the average value of the remaining activation coefficients of the snow removal equipment after removing outliers.

[0034] As a further technical solution of the present invention: the process of determining the cause of the anomaly includes:

[0035] Obtain the actual power generation curve of each photovoltaic device in the photovoltaic power station;

[0036] Obtain the number of times the power generation rises vertically in the actual power generation curve;

[0037] If the absolute value of the difference between the number of times the power generation increases vertically and the number of times the corresponding snow removal equipment is started is not greater than 2, and the position of the first abnormal value coincides with the start time of any snow removal equipment, then the current first abnormal value is determined to be caused by snow removal.

[0038] Otherwise, it is judged to be caused by other reasons.

[0039] As a further technical solution of the present invention: the process of constructing the compensation sequence includes:

[0040] For any first outlier, retrieve its preceding and following non-outlier data;

[0041] Through the formula:

[0042]

[0043] Get the compensation value corresponding to the first outlier, where As compensation value, To predict the duration of snowfall, The standard snowfall duration set for the current snowfall coefficient. and These are the non-abnormal data points preceding and following the first outlier. This is the first outlier currently;

[0044] Multiple compensation values ​​are used to construct a compensation sequence.

[0045] As a further technical solution of the present invention: the process of correcting the sequence to be corrected by the compensation sequence includes:

[0046] A second predictive model for the start-up time of snow removal equipment was constructed for different snowfall levels;

[0047] The predicted sequence of snow removal equipment start-up time output by the second prediction model is used as the correction sequence to be used for the first prediction sequence;

[0048] If the number of sequences to be corrected is not greater than the number of compensation sequences for the corresponding snowfall level, then the values ​​of the compensation sequences will replace the corresponding sequences to be corrected by interpolation according to the sequence order.

[0049] If the number of sequences in the sequence to be corrected is greater than the number of compensation sequences for the corresponding snowfall level, the number of compensation sequences is extended using the least squares method until the number of compensation sequences is the same as the number of sequences in the sequence to be corrected. Then, according to the sequence order, the values ​​of the compensation sequences are used to replace the corresponding sequences in the sequence to be corrected by interpolation. The least squares method finds the best function match of the data by minimizing the sum of squares of the errors. The least squares method can be used to obtain unknown data and minimize the sum of squares of the errors between the obtained data and the actual data.

[0050] The above technical solution provides a method for correcting a first prediction sequence to obtain a second prediction sequence. In the correction process, the first prediction sequence of the present invention is corrected based on the value of the first outlier. Compared with direct rejection, it can more accurately predict the total power generation. In addition, the present invention takes the start-up of snow removal equipment into consideration in the prediction process rather than directly rejecting the data anomalies caused by the start-up of snow removal equipment, which makes the prediction results more robust.

[0051] Another object of the present invention is to provide a method for predicting photovoltaic power generation, comprising:

[0052] The weather monitoring module obtains the predicted weather data for the predicted snowfall day of the target photovoltaic power station and obtains the snowfall coefficient based on the predicted weather data;

[0053] A snowfall classification module, which divides snowfall days into several snowfall levels based on snowfall coefficient and snow removal equipment start-up time;

[0054] The first prediction model takes predicted weather data as input, outputs the predicted power generation of the target photovoltaic power station on the day of snowfall, and obtains the first prediction sequence.

[0055] An anomaly analysis module acquires and stores the actual power generation data of each power generation unit of the target photovoltaic power station on the predicted snowfall day with different snowfall levels, compares the first prediction sequence with the actual power generation data, acquires multiple first anomalies and records the position of the first anomalies, and judges the cause of the anomalies, including those caused by snow removal and other causes, and constructs a compensation sequence based on the first anomalies caused by snow removal.

[0056] The second prediction model outputs a sequence of predicted snow removal equipment start-up times as the correction sequence to be used for the first prediction sequence.

[0057] The correction module corrects the sequence to be corrected using a compensation sequence.

[0058] The beneficial effects of this invention are:

[0059] (1) This invention further analyzes and uses the first outlier caused by the start of snow removal equipment in snowy weather, thereby optimizing the prediction process in snowy weather, thereby optimizing the current mainstream photovoltaic power generation prediction method for removing outliers, and improving the robustness of the prediction results under snowy weather and snow removal equipment operation.

[0060] (2) The snowfall level of the present invention is set based on the snowfall process and the start-up status of the snow removal equipment during the snowfall process. The greater the snowfall or the more times the snow removal equipment is started during the snowfall process, the higher the snowfall level. After the snowfall level of the present invention is divided, the snowfall time in each snowfall level is similar and the number of times the snow removal equipment is started during the snowfall process is similar, which can provide a basis for the subsequent prediction result correction.

[0061] (3) The first prediction sequence of the present invention is corrected based on the value of the first outlier during the correction process. Compared with direct rejection, it can more accurately predict the total power generation. In addition, the present invention takes the start-up of snow removal equipment into consideration during the prediction process instead of directly rejecting the data anomalies caused by the start-up of snow removal equipment, which makes the prediction results more robust. Attached Figure Description

[0062] The invention will now be further described with reference to the accompanying drawings.

[0063] Figure 1 This is a schematic diagram of the prediction method steps of the present invention;

[0064] Figure 2 This is a schematic diagram of the composition relationship of the prediction system of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Please see Figure 1 As shown, in one embodiment, a photovoltaic power generation prediction method is provided, including the following steps:

[0067] S100. Obtain the predicted weather data for the predicted snowfall day of the target photovoltaic power station and obtain the snowfall coefficient based on the predicted weather data. The predicted weather data includes the predicted snowfall duration and the predicted average snowfall amount. The snowfall coefficient is positively proportional to the predicted snowfall duration and the predicted average snowfall amount.

[0068] S200: Based on the snowfall coefficient and the start-up time of snow removal equipment, snowfall days are divided into several snowfall levels;

[0069] S300. Substitute the predicted weather data into the first prediction model to predict the power generation of the target photovoltaic power station on the predicted snowfall day and obtain the first prediction sequence. In this embodiment, the prediction sequence is a numerical sequence of power generation for each hour. The first prediction model is a combination of multiple prediction methods that remove outliers. The prediction result with the smallest mean square error is selected for output. The prediction methods that remove outliers include existing technologies such as isolated forest.

[0070] S400: Acquire and store the actual power generation data of each power generation unit of the target photovoltaic power station on the predicted snowfall day of different snowfall levels, compare the first prediction sequence with the actual power generation data, obtain multiple first outliers and record the position of the first outlier, the position of the outlier is the time of occurrence of the outlier;

[0071] S500: Determine the cause of the first outlier. The cause of the outlier includes snow removal and other causes. Construct a compensation sequence based on the first outlier caused by snow removal.

[0072] S600: Obtain the sequence to be corrected for the first prediction model according to different snowfall levels, and correct the sequence to be corrected by the compensation sequence;

[0073] S700, output the second prediction sequence through the modified first prediction model.

[0074] This embodiment provides a method for predicting the power generation of a photovoltaic power station under snowfall. The present invention optimizes the prediction process under snowfall by further analyzing and using the first outlier caused by the start of snow removal equipment. This optimizes the current mainstream photovoltaic power generation prediction method that removes outliers and improves the robustness of the prediction results under snowfall and snow removal equipment operation.

[0075] The process of obtaining the snowfall coefficient includes:

[0076] S110. Standardize the predicted snowfall duration and the predicted average snowfall amount;

[0077] S120. Using the standardized predicted snowfall duration as the base and the standardized predicted average snowfall amount as the exponent, the snowfall coefficient is calculated. As an example, the formula is used... To obtain the snowfall coefficient, where This is the snowfall coefficient. For correction factor, As a compensation constant, For the standardized predicted snowfall duration, This represents the standardized predicted average snowfall.

[0078] The process of classifying snowfall levels includes:

[0079] S210. Divide the range of snowfall coefficient values ​​into at least three equally spaced initial intervals. The impact of snowfall on snow removal mechanisms can be divided into at least three types: one is no impact when snowfall is very low and snow removal mechanisms do not need to operate; another is a moderate impact when snowfall is high and snow removal mechanisms need to be started intermittently; and the third is no impact when snowfall is extremely high and photovoltaic power stations stop operating, and snow removal mechanisms also stop operating.

[0080] S220. Calculate the activation coefficient of each photovoltaic power station that falls within the initial range based on historical data of snowfall days;

[0081] S230. Select the average start-up coefficient of the photovoltaic power station on several snowy days corresponding to the snowfall coefficient at the midpoint of the initial interval as the benchmark value, and at the same time, continuously identify and search to both sides of the initial interval at a preset interval.

[0082] S240. Construct several second intervals based on the identified search results and set different snowfall levels for different second intervals according to the numerical values.

[0083] This embodiment provides a process for constructing different snowfall levels. The snowfall levels of this invention are set based on the snowfall process and the activation status of snow removal equipment during the snowfall process. The greater the snowfall or the more times the snow removal equipment is activated during the snowfall process, the higher the snowfall level. After the snowfall levels of this invention are divided, the snowfall time and the number of times the snow removal equipment is activated during the snowfall process are similar within each snowfall level, which can provide a basis for subsequent prediction result correction.

[0084] The process of identification and search includes:

[0085] Set the average start-up coefficient of photovoltaic power stations corresponding to several snowfall days at the midpoint of the initial interval as the benchmark value, and calculate the difference between the benchmark value and the average start-up coefficient of photovoltaic power stations corresponding to the next snowfall coefficient during the identification and search process;

[0086] If the difference is not greater than the critical value, the identification search is considered successful, and the average starting coefficient of the photovoltaic power station corresponding to the next snowfall coefficient is used to replace the original benchmark value for the next round of identification search.

[0087] If the difference is less than the critical value, the identification search is judged to have failed. The next snowfall coefficient during the identification search process is recorded as one endpoint of the second interval. Continuous identification search is performed on both sides of the initial interval to obtain one endpoint each. The value range between the two endpoints is the second interval.

[0088] The process of obtaining the activation coefficient of the snow removal equipment of a target photovoltaic power station on a snowfall day includes:

[0089] Through the formula:

[0090]

[0091] in This refers to the activation coefficient of the corresponding device. , and These are the preset fitting parameters. It is a preset startup state function and , This refers to the number of times the device has been started. It is a functional relationship between the number of times the device is started and the start time;

[0092] Anomaly detection was performed on the activation coefficients of all snow removal equipment in the target photovoltaic power station. The activation coefficient of the target photovoltaic power station is the average value of the remaining activation coefficients of the snow removal equipment after removing outliers.

[0093] As one embodiment, the combination of fitting parameters can be... , as well as The purpose of binomial fitting is to make the final result present as a binomial curve and to make it smoother. Obviously, the fitting parameters can also be other combinations, which will not be elaborated here. The start-up state function is a preset function in exponential form. Its purpose is to ensure that the start-up coefficient values ​​of devices with the same number of starts in each snowfall process do not differ too much, while increasing the start-up coefficient values ​​of devices with different number of starts.

[0094] The process of determining the cause of an anomaly includes:

[0095] Obtain the actual power generation curve of each photovoltaic device in the photovoltaic power station;

[0096] Obtain the number of times the power generation rises vertically in the actual power generation curve;

[0097] If the absolute value of the difference between the number of times the power generation increases vertically and the number of times the corresponding snow removal equipment is started is not greater than 2, and the position of the first abnormal value coincides with the start time of any snow removal equipment, then the current first abnormal value is determined to be caused by snow removal.

[0098] Otherwise, it is judged to be caused by other reasons.

[0099] It should be noted that the numerical anomalies caused by snow removal are regular in snowy weather. Whenever the snow removal mechanism is started, the power generation of the photovoltaic equipment will increase sharply under constant sunlight conditions. Based on this pattern, the cause of the anomaly can be accurately determined.

[0100] The process of constructing the compensation sequence includes:

[0101] For any first outlier, retrieve its preceding and following non-outlier data;

[0102] Through the formula:

[0103]

[0104] Get the compensation value corresponding to the first outlier, where As compensation value, To predict the duration of snowfall, The standard snowfall duration set for the current snowfall coefficient. and These are the non-abnormal data points preceding and following the first outlier. This is the first outlier currently;

[0105] Multiple compensation values ​​are used to construct a compensation sequence.

[0106] The process of correcting the sequence to be corrected using the compensation sequence includes:

[0107] A second prediction model for the start-up time of snow removal equipment is constructed for different snowfall levels. The second prediction model adopts either a time series model or a neural network model.

[0108] The predicted sequence of snow removal equipment start-up time output by the second prediction model is used as the correction sequence to be used for the first prediction sequence;

[0109] If the number of sequences to be corrected is not greater than the number of compensation sequences for the corresponding snowfall level, then the values ​​of the compensation sequences will replace the corresponding sequences to be corrected by interpolation according to the sequence order.

[0110] If the number of sequences in the sequence to be corrected is greater than the number of compensation sequences for the corresponding snowfall level, the number of compensation sequences is extended using the least squares method until the number of compensation sequences is the same as the number of sequences in the sequence to be corrected. Then, according to the sequence order, the values ​​of the compensation sequences are used to replace the corresponding sequences in the sequence to be corrected by interpolation. The least squares method finds the best function match of the data by minimizing the sum of squares of the errors. The least squares method can be used to obtain unknown data and minimize the sum of squares of the errors between the obtained data and the actual data.

[0111] This embodiment provides a method for correcting a first prediction sequence to obtain a second prediction sequence. The first prediction sequence of the present invention is corrected based on the value of a first outlier during the correction process. Compared with direct rejection, it can more accurately predict the total power generation. In addition, the present invention takes the start-up of snow removal equipment into consideration during the prediction process instead of directly rejecting the data anomalies caused by the start-up of snow removal equipment, making the prediction results more robust.

[0112] refer to Figure 2 Another object of the present invention is to provide a photovoltaic power generation prediction system, comprising:

[0113] The weather monitoring module obtains the predicted weather data for the predicted snowfall day of the target photovoltaic power station and obtains the snowfall coefficient based on the predicted weather data;

[0114] A snowfall classification module, which divides snowfall days into several snowfall levels based on snowfall coefficient and snow removal equipment start-up time;

[0115] The first prediction model takes predicted weather data as input, outputs the predicted power generation of the target photovoltaic power station on the day of snowfall, and obtains the first prediction sequence.

[0116] An anomaly analysis module acquires and stores the actual power generation data of each power generation unit of the target photovoltaic power station on the predicted snowfall day with different snowfall levels, compares the first prediction sequence with the actual power generation data, acquires multiple first anomalies and records the position of the first anomalies, and judges the cause of the anomalies, including those caused by snow removal and other causes, and constructs a compensation sequence based on the first anomalies caused by snow removal.

[0117] The second prediction model outputs a sequence of predicted snow removal equipment start-up times as the correction sequence to be used for the first prediction sequence.

[0118] The correction module corrects the sequence to be corrected using a compensation sequence.

[0119] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for predicting photovoltaic power generation, characterized in that, Includes the following steps: Obtain forecast weather data for the predicted snowfall date of the target photovoltaic power station and obtain the snowfall coefficient based on the forecast weather data; Snowfall days are classified into several snowfall levels based on the snowfall coefficient and the start-up time of snow removal equipment; The predicted weather data is substituted into the first prediction model to predict the power generation of the target photovoltaic power station on the predicted snowfall day and obtain the first prediction sequence. Acquire and store the actual power generation data of each power generation unit of the target photovoltaic power station on the predicted snowfall day with different snowfall levels, and compare the first prediction sequence with the actual power generation data to obtain multiple first outliers and record the position of the first outliers; The cause of the first outlier is determined, including snow removal. A compensation sequence is constructed based on the first outlier caused by snow removal. The process of constructing the compensation sequence includes: For any first outlier, retrieve its preceding and following non-outlier data; Through the formula: Get the compensation value corresponding to the first outlier, where As compensation value, To predict the duration of snowfall, The standard snowfall duration set for the current snowfall coefficient. and These are the non-abnormal data points preceding and following the first outlier. This is the first outlier currently; Multiple compensation values ​​are constructed into a compensation sequence; The process of correcting the sequence to be corrected using the compensation sequence includes: A second predictive model for the start-up time of snow removal equipment was constructed for different snowfall levels; The predicted sequence of snow removal equipment start-up time output by the second prediction model is used as the correction sequence to be used for the first prediction sequence; If the number of sequences to be corrected is not greater than the number of compensation sequences for the corresponding snowfall level, then the values ​​of the compensation sequences will replace the corresponding sequences to be corrected by interpolation according to the sequence order. If the number of sequences in the sequence to be corrected is greater than the number of compensation sequences for the corresponding snowfall level, the number of compensation sequences is extended by least squares until the number of compensation sequences is the same as the number of sequences in the sequence to be corrected. Then, the value of the compensation sequence replaces the corresponding sequence of the sequence to be corrected by interpolation according to the sequence order. The first prediction model's sequence to be corrected is obtained based on different snowfall levels, and the sequence to be corrected is corrected using the compensation sequence. The second prediction sequence is output using the modified first prediction model.

2. The photovoltaic power generation prediction method according to claim 1, characterized in that, The process of obtaining the snowfall coefficient includes: Standardize the predicted snowfall duration and the predicted average snowfall amount; The snowfall coefficient is calculated using the standardized predicted snowfall duration as the base and the standardized predicted average snowfall amount as the exponent.

3. The photovoltaic power generation prediction method according to claim 1, characterized in that, The process of classifying snowfall levels includes: Divide the range of snowfall coefficient values ​​into at least three equally spaced initial intervals; The activation coefficient of each photovoltaic power station falling within the initial range is calculated based on historical data of snowfall days; The average start-up coefficient of photovoltaic power stations on several snowy days corresponding to the snowfall coefficient at the midpoint of the initial interval is selected as the benchmark value, and continuous identification and search are performed on both sides of the initial interval at preset intervals. Based on the identified search results, several second intervals are constructed, and different snowfall levels are set for different second intervals according to their numerical values.

4. The photovoltaic power generation prediction method according to claim 3, characterized in that, The identification and search process includes: Set the average start-up coefficient of photovoltaic power stations corresponding to several snowfall days at the midpoint of the initial interval as the benchmark value, and calculate the difference between the benchmark value and the average start-up coefficient of photovoltaic power stations corresponding to the next snowfall coefficient during the identification and search process; If the difference is not greater than the critical value, the identification search is considered successful, and the average starting coefficient of the photovoltaic power station corresponding to the next snowfall coefficient is used to replace the original benchmark value for the next round of identification search. If the difference is greater than the critical value, the identification search is judged to have failed. The next snowfall coefficient during the identification search process is recorded as one of the endpoints of the second interval.

5. The photovoltaic power generation prediction method according to claim 1, characterized in that, The process of obtaining the activation coefficient of the snow removal equipment of a target photovoltaic power station on a snowfall day includes: Through the formula: in This refers to the activation coefficient of the corresponding device. , and These are the preset fitting parameters. It is a preset startup state function and , This refers to the number of times the device has been started. It is a functional relationship between the number of times the device is started and the start time; Anomaly detection was performed on the activation coefficients of all snow removal equipment in the target photovoltaic power station. The activation coefficient of the target photovoltaic power station is the average value of the remaining activation coefficients of the snow removal equipment after removing outliers.

6. The photovoltaic power generation prediction method according to claim 1, characterized in that, The process of determining the cause of an anomaly includes: Obtain the actual power generation curve of each photovoltaic device in the photovoltaic power station; Obtain the number of times the power generation rises vertically in the actual power generation curve; If the absolute value of the difference between the number of times the power generation increases vertically and the number of times the corresponding snow removal equipment is started is not greater than 2, and the position of the first abnormal value coincides with the start time of any snow removal equipment, then the current first abnormal value is determined to be caused by snow removal. Otherwise, it is judged to be caused by other reasons.

7. A photovoltaic power generation prediction system, characterized in that, The method for predicting photovoltaic power generation as described in any one of claims 1-6 includes: The weather monitoring module obtains the predicted weather data for the predicted snowfall day of the target photovoltaic power station and obtains the snowfall coefficient based on the predicted weather data; A snowfall classification module, which divides snowfall days into several snowfall levels based on snowfall coefficient and snow removal equipment start-up time; The first prediction model takes predicted weather data as input, outputs the predicted power generation of the target photovoltaic power station on the day of snowfall, and obtains the first prediction sequence. An anomaly analysis module acquires and stores the actual power generation data of each power generation unit of the target photovoltaic power station on the predicted snowfall day with different snowfall levels, compares the first prediction sequence with the actual power generation data, acquires multiple first anomalies and records the position of the first anomalies, and judges the cause of the anomalies, including those caused by snow removal, and constructs a compensation sequence based on the first anomalies caused by snow removal. The second prediction model outputs a sequence of predicted snow removal equipment start-up times as the correction sequence to be used for the first prediction sequence. The correction module corrects the sequence to be corrected using a compensation sequence.

Images