Regional distributed photovoltaic power prediction method under extreme data loss

Abstract

The invention discloses a regional distributed photovoltaic power prediction method under extreme data loss, and the method comprises the steps: carrying out the equal-resolution arrangement of the obtainable meteorological data and generated power data of a distributed photovoltaic power station and a centralized photovoltaic power station in a region; carrying out abnormal value identification and cubic spline interpolation processing on data missing and abnormal power generation points; for a centralized monitored photovoltaic power station with complete data sampling in a region, a lightweight gradient elevator method is adopted to perform modeling on a meteorological data-photovoltaic power generation relationship, and a frost optimization algorithm is adopted to optimize hyper-parameters of a model so as to improve prediction precision; and based on the prediction result of the monitored photovoltaic power station in the region, combining an up-sampling method and a transfer learning method to improve the estimation precision of the non-monitoring photovoltaic power station. According to the method, the output of the non-monitoring photovoltaic power station lacking data acquisition can be estimated in the scene of large-scale data missing, the output data of the photovoltaic power station can be complemented, and the method has better adaptive capacity and robustness.

Description

Technical Field

[0001] The present invention belongs to the field of new energy, and in particular relates to a method for predicting regional distributed photovoltaic power under extreme data loss. Background Art

[0002] At present, the proportion of renewable energy power generation is growing rapidly. Among them, photovoltaic power generation, as one of the most mainstream forms of renewable energy power generation, is gradually changing from a large-capacity centralized power generation mode to a mode where large-capacity centralized power generation and small-capacity distributed power generation coexist. There has been a substantial increase in small-scale distributed photovoltaic systems (BTM), especially rooftop photovoltaics and small photovoltaic systems without data monitoring equipment. Although the capacity of such photovoltaic systems is generally small, their number is large and their distribution range is wide, so the impact of their total power generation on the overall power system cannot be ignored. According to the International Energy Agency (IEA), half of the global solar capacity growth will be small distributed generators, so accurate prediction of BTM photovoltaic power generation systems is crucial to the safe and stable operation of the power system.

[0003] At present, for the prediction of distributed photovoltaic cluster power generation, domestic and foreign research mainly focuses on power generation areas with complete data sampling, but rarely involves distributed photovoltaic power generation areas with large-scale data missing. For distributed photovoltaic power generation prediction with complete data sampling, deep learning methods occupy a dominant position. For example, Chinese patent application CN202311472223.3 "Distributed photovoltaic power generation power prediction system and method based on deep learning framework" proposes a distributed photovoltaic power generation power prediction system and method based on a deep learning framework, which pre-processes abnormal data in data samples through the RNN network module, and trains the data and time series features of the training set based on the attention mechanism through the transformer network module; Chinese patent application CN202410506841.3 "A short-term prediction method for distributed photovoltaic power generation based on big data" A graph neural network method is proposed to obtain relevant data of photovoltaic power stations, construct an undirected graph based on the relevant data of photovoltaic power stations, and obtain the original node features of all nodes in the graph for prediction; Chinese patent application CN202311736745.X "Distributed photovoltaic power generation WaveNet-BiLstm prediction method and device" uses WaveNet network to construct a distributed photovoltaic power generation prediction model time feature extraction module, and then uses BiLSTM network for power generation prediction model prediction module; Chinese patent application CN202311790426.7 "A distributed photovoltaic power generation prediction system and method" screens similar meteorological scenes and converts historical The damage data of photovoltaic power station operation corresponding to meteorological data is imported into the damage judgment strategy, and a meteorological damage judgment model is constructed. Then these similar scenarios and their historical power data are input into the power generation prediction strategy for preliminary prediction; the above photovoltaic prediction methods are all based on deep learning methods, but the implementation and effect of deep learning methods rely on complete data. In general, the distributed photovoltaic power generation predictions of these methods rely on complete meteorological data and photovoltaic power generation prediction data. With the development of decentralized distributed photovoltaics, many power stations lack data monitoring equipment due to construction reasons and data monitoring cost issues. For some areas, more than half are small-capacity distributed power stations that lack data power generation information. Therefore, the current deep learning distributed photovoltaic power generation prediction method based on big data is no longer applicable.

[0004] In response to the problem of missing data of unmonitored power stations in regional distributed photovoltaic power generation, related studies have also proposed and explored the problem of completing power generation data of unmonitored distributed photovoltaic power stations. For example, Chinese patent application CN202310503584.3 "A method for estimating the output power of unmonitored distributed photovoltaic power stations" proposes to determine the low-voltage distributed power source partition or aggregate according to its geographical location or the feeder to which it belongs to estimate the photovoltaic power generation output and make corrections based on the power. Chinese patent application CN202011474288.8 "A method and system for estimating distributed photovoltaic total radiation based on satellite remote sensing" uses a satellite solar radiation correction model and linear interpolation method to estimate complete total radiation data to solve the problem of unmonitored total radiation of distributed photovoltaic power stations. However, these methods all estimate the output of unmonitored photovoltaic power stations that lack data collection to a certain extent, and overall do not improve the prediction accuracy of the entire distributed photovoltaic power generation cluster from the perspective of large-scale data missing. Therefore, it is urgent to propose a more accurate regional distributed photovoltaic power prediction method in large-scale data missing scenarios. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention proposes a regional distributed photovoltaic power prediction method under extreme data loss, which estimates the output of distributed photovoltaic power stations in the region that lack data monitoring by combining transfer learning and weight optimization. This method can not only estimate the output of a single unmonitored photovoltaic power station, promote the stable operation and regulation of the distribution network, but also further improve the prediction accuracy of the entire regional distributed photovoltaic power generation, effectively solve the impact of large-scale data loss on the prediction accuracy of regional distributed photovoltaic power generation, and improve the accuracy and stability of the prediction.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: A method for predicting regional distributed photovoltaic power under extreme data loss includes the following steps: Step 1: sort out the available meteorological data and power generation data of the distributed photovoltaic power stations and centralized photovoltaic power stations in the region at different resolutions, and perform outlier identification and cubic spline interpolation processing on data missing and abnormal power generation points; Step 2: For centralized monitored power stations with complete data sampling in the region, the LightGBM method is used to model the relationship between meteorological data and photovoltaic power generation, and the frost optimization algorithm is used to optimize the hyperparameters of the LightGBM model to improve the prediction accuracy of centralized photovoltaic power generation in the region, providing a benchmark for subsequent predictions of unmonitored power stations; Step 3: Based on the prediction results of the monitored photovoltaic power stations in the region, the combined weight optimization is used to estimate the output of the unmonitored photovoltaic power stations in the region. At the same time, the power generation model of the monitored photovoltaic power stations is migrated to the unmonitored power stations to improve the estimation accuracy of the unmonitored photovoltaic power stations. Step 4: Summarize the prediction results of monitored PV power stations with complete data collection and the estimated results of unmonitored PV power stations with insufficient data sampling to obtain the distributed PV power prediction results in areas with extreme data loss.

[0007] Furthermore, the step 1 comprises: Step 1-1: Determine all available power generation-related data of regional distributed photovoltaics, and organize all input variable data and output variable data at equal resolution, where the input data is the meteorological data in the region, including available weather forecast data of light radiation, temperature, humidity, and air pressure, and the output variables include the power generation data of monitored photovoltaic power stations and the power generation data of unmonitored photovoltaic power stations; Step 1-2: Collect historical operation data of each regional distributed photovoltaic power station, standardize the data, and divide it into training set and test set; (1) (2) (3) in, A historical set representing historical meteorological data and power generation components; Represents standardized historical data; represents the averaging function, It represents the standard deviation function; represents the training set taken from the historical data after standardization, represents the test set taken from the historical data after standardization, Indicates the proportion of the training set to the total data set; Step 1-3: Identify outliers in the sampled data and use the cubic spline method for interpolation and completion; (4) In each interval The lower limit of the interval And the regional limit Inside[ ]Construct a As a smooth curve function of the variable, establish a fitting formula ,in , , , is the coefficient to be determined, and the missing points of regional data are fitted and completed through the fitting formula of each interval.

[0008] Furthermore, the step 2 comprises the following steps: Step 2-1: For all monitored photovoltaic power stations with complete data collection in the region, a power generation data prediction model is built. The LightGBM method is used to model the relationship between meteorological input and power generation output. The training target formula is shown in Equation (5)-Equation (6): (5) (6) in, For the The characteristics of sampled data, there are M data in total, is the corresponding true label, is the corresponding predicted label, To calculate the mean square error function, t represents the number of iterations, represents the target at the tth iteration, represents the newly generated CART fitting function at the tth iteration, represents the regularization term; T represents the number of leaves, and is the hyperparameter of the penalty term, Represents the weight value assigned to leaf node j; Step 2-2: Use the Frost Optimization algorithm to optimize the hyperparameters of the LightGBM model to further improve the accuracy and robustness of the power prediction of the monitored PV power station: set up =[n_estimators, num_leaves, max_depth, learning_rate] as optimization variables, n_estimators, num_leaves, max_depth, learning_rate are the number of iterations of the LightGBM model algorithm, the number of leaves, the maximum depth, and the learning rate, respectively. is the lower limit of the optimization variable, is the upper limit of the optimization variable; set up For the The optimal hyperparameters at the iteration time are updated using the soft time search strategy during the optimization process as follows: (7) In the formula, is the position of the updated hyperparameter, n represents the particle number, and t is the iteration number; is a random number in the range (-1,1). is a random number between (0,1), h is the degree of adhesion, which is a random number in the range of (0,1) and is used to control the distance between the centers of two gray particles. is the maximum number of iterations.

[0009] Furthermore, the step 3 comprises the following steps: Step 3-1: Determine the monitored PV power stations in the area and the set of power stations trained by the RiME-LightGBM algorithm , Power generation model collection And the collection of power stations without monitoring ; Weather forecast collection for monitored and unmonitored PV power stations , ; The capacity set of monitored photovoltaic power stations and unmonitored photovoltaic power stations, where there are M monitored photovoltaic power stations and N unmonitored photovoltaic power stations in the region; (8) in, Representing the region There are photovoltaic power stations monitored, Representing the region A power generation model with monitoring photovoltaic power station, Represents the nth unmonitored PV power station in the region, Representing the region There are meteorological datasets for monitoring photovoltaic power plants. Representing the region A meteorological dataset of an unmonitored PV power station. Representing the region The capacity of the photovoltaic power station is monitored. Representing the region The capacity of an unmonitored PV plant; Step 3-2: Determine the predicted value of the monitored photovoltaic power station in the area, bring the weather forecast into the LightGBM model trained based on historical data, and obtain the predicted value of the monitored photovoltaic power station in the area: (9) In the formula, represents the power generation forecast value of the mth monitored PV power station, is the power generation model of the mth monitored photovoltaic power station, is the weather forecast value of the mth photovoltaic power station; Step 3-3: For unmonitored PV plants with missing data, output estimates can be obtained by optimizing the weighted combination of power generation modes of monitored PV plants. : (10) in, is the weight, is the capacity of the nth unmonitored PV power station, is the capacity of the mth monitored PV power station; Represents the power generation of the unmonitored photovoltaic power station estimated by weighted combination optimization; The power generation model of the monitored PV power station is directly migrated to the unmonitored PV power station to obtain the output estimate: (11) in, represents the power generation of the unmonitored photovoltaic power station estimated by model migration optimization, is the combined weight of the power generation models of the unmonitored PV power station and different monitored PV power stations, This is the meteorological data set of the unmonitored photovoltaic power station; Step 3-4: Combine the output estimation result of transfer learning in formula (11) with the output estimation result of weight combination in formula (10) to improve the output estimation accuracy of the unmonitored PV power station: (12) in, The power generation of the unmonitored photovoltaic power station is estimated by combining weighted combination optimization and model migration optimization. , , , is the weight coefficient, which is solved by the optimization method. The photovoltaic power prediction result of the whole area is: (13) in, Represents the predicted photovoltaic power value of the fully distributed photovoltaic power generation area.

[0010] The beneficial effects of the present invention compared with the prior art are: (1) The present invention can estimate the output of unmonitored photovoltaic power stations that lack data collection in scenarios where large-scale data is missing, thereby completing the distributed photovoltaic output data; (2) Compared with traditional data-driven methods, the present invention can adapt to the situation of large-scale data missing and has better adaptability and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of regional distributed photovoltaic power generation; Figure 2 The present invention is a flow chart of a method for predicting regional distributed photovoltaic power under extreme data loss. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0013] The following is a method for predicting regional distributed photovoltaic power under extreme data loss proposed by the present invention in conjunction with the accompanying drawings. The specific implementation methods in this specification are only for explaining the invention in detail, and are not intended to limit the present invention.

[0014] Figure 1 A schematic diagram of regional distributed photovoltaic power generation is given. Regional distributed photovoltaic power generation includes monitored photovoltaic power stations with complete data collection in the region and unmonitored photovoltaic power stations with missing data sampling. For the 1st, 2nd, and mth photovoltaic power generation units with complete data sampling in the region, These are the 1st, 2nd, 3rd, 4th, and nth unmonitored photovoltaic power generation units in the area that lack data sampling.

[0015] like Figure 2 As shown, the regional distributed photovoltaic power prediction method under extreme data loss of the present invention includes the following steps: Step 1: sort out the available meteorological data and power generation data of the distributed photovoltaic power stations and centralized photovoltaic power stations in the region at different resolutions, and perform outlier identification and cubic spline interpolation processing on data missing and abnormal power generation points; Step 2: For centralized monitored photovoltaic power stations with complete data sampling in the region, the LightGBM (Lightweight Gradient Boosting Machine) method is used to model the relationship between meteorological data and photovoltaic power generation, and the frost optimization algorithm is used to optimize the hyperparameters of the LightGBM model to improve the prediction accuracy of centralized photovoltaic power generation in the region, providing a benchmark for subsequent predictions of unmonitored photovoltaic power stations; Step 3: Based on the prediction results of the monitored photovoltaic power stations in the region, a combined weight optimization method is used to estimate the output of the unmonitored power stations in the region. At the same time, the power generation model of the monitored photovoltaic power stations is migrated to the unmonitored power stations to improve the output estimation accuracy of the unmonitored photovoltaic power stations. Step 4: Summarize the prediction results of monitored power stations with complete data collection and the estimated results of unmonitored power stations with insufficient data sampling to obtain the distributed photovoltaic power prediction results in the area with extreme data missing.

[0016] Specifically, the step 1 specifically includes the following steps: Step 1-1: Determine all available power generation-related data of regional distributed photovoltaics, and organize all input variable data and output variable data at equal resolution. The input data is mainly meteorological data in the region, including available meteorological forecast data such as light radiation, temperature, humidity, and air pressure. The output variables include the power generation data of monitored photovoltaic power stations and the power generation data of unmonitored photovoltaic power stations. Step 1-2: Collect historical operation data of each regional distributed photovoltaic power station, standardize the data, and divide it into training set and test set; (1) (2) (3) in, Represents a historical set of historical meteorological data and power generation. The resolution is usually 15 minutes or 1 hour, which means that data is sampled every 15 minutes or 1 hour. Represents standardized historical data; represents the averaging function, It represents the standard deviation function; represents the training set taken from the historical data after standardization, represents the test set taken from the historical data after standardization, Indicates the proportion of the training set to the total data set; Step 1-3: Identify outliers in the sampled data and use the cubic spline method for interpolation and completion; (4) In each interval The lower limit of the interval And the regional limit Inside[ ]Construct a As a smooth curve function of the variable, establish a fitting formula ,in , , , is the coefficient to be determined, and the missing points of regional data are fitted and completed through the fitting formula of each interval.

[0017] Specifically, the step 2 specifically includes the following steps: Step 2-1: A power generation data prediction model is built for all monitored photovoltaic power stations in the region with complete data collection. The LightGBM (Lightweight Gradient Boosting Machine) method is used to model the relationship between meteorological input and power generation output. The training target formula is shown in Equation (5)-Equation (6).

[0018] (5)

[0019] (6) in, For the The characteristics of sampled data, there are M data in total, is the corresponding true label, is the corresponding predicted label, To calculate the mean square error function, t represents the number of iterations, represents the target at the tth iteration, represents the newly generated CART fitting function at the tth iteration, represents the regularization term; T represents the number of leaves, and is the hyperparameter of the penalty term, Represents the weight value assigned to leaf node j.

[0020] Step 2-2: Use the Frost Optimization algorithm to optimize the hyperparameters of the LightGBM model to further improve the accuracy and robustness of the power prediction of the monitored PV power station: set up =[n_estimators, num_leaves, max_depth, learning_rate] as optimization variables, n_estimators, num_leaves, max_depth, learning_rate are the number of iterations of the LightGBM model algorithm, the number of leaves, the maximum depth, and the learning rate, respectively. is the lower limit of the optimization variable, is the upper limit of the optimization variable.

[0021] set up For the The optimal hyperparameters at the iteration time are updated using the soft time search strategy during the optimization process as follows: (7) In the formula, is the position of the updated hyperparameter, n represents the particle number, and t is the iteration number. is a random number in the range (-1,1). is a random number between (0,1), h is the degree of adhesion, which is a random number in the range of (0,1) and is used to control the distance between the centers of two gray particles. is the maximum number of iterations.

[0022] Specifically, the step 3 comprises the following steps: Step 3-1: Determine the monitored PV power stations in the area and the set of power stations trained by the RiME-LightGBM (Frost Optimization Algorithm Optimized Lightweight Gradient Boosting Machine) algorithm , Power generation model collection And the collection of power stations without monitoring ; Weather forecast collection for monitored and unmonitored PV power stations , ; The capacity set of monitored photovoltaic power stations and unmonitored photovoltaic power stations, where there are M monitored photovoltaic power stations and N unmonitored photovoltaic power stations in the region.

[0023] (8) in, Representing the region There are photovoltaic power stations with monitoring, Representing the region A power generation model with monitoring photovoltaic power station, Represents the nth unmonitored PV power station in the region, Representing the region There are meteorological datasets for monitoring photovoltaic power plants. Representing the region A meteorological dataset of an unmonitored PV power station. Representing the region The capacity of the photovoltaic power station is monitored. Representing the region The capacity of an unmonitored PV power station.

[0024] Step 3-2: Determine the predicted value of the monitored photovoltaic power station in the area, bring the weather forecast into the LightGBM model trained based on historical data, and obtain the predicted value of the monitored photovoltaic power station in the area: (9) In the formula, represents the power generation forecast value of the mth monitored PV power station, is the power generation model of the mth monitored photovoltaic power station, is the weather forecast value of the mth photovoltaic power station.

[0025] Step 3-3: For unmonitored PV plants with missing data, output estimates can be obtained by optimizing the weighted combination of power generation modes of monitored PV plants. : (10) in, is the weight, is the capacity of the nth unmonitored PV power station, is the capacity of the mth monitored PV power station; Represents the power generation of the unmonitored PV power station estimated using weighted combination optimization.

[0026] On the other hand, the power generation model of the monitored PV power station can be directly transferred to the unmonitored PV power station to obtain the output estimate: (11) in, represents the power generation of the unmonitored photovoltaic power station estimated by model migration optimization, is the combined weight of the power generation models of the unmonitored PV power station and different monitored PV power stations, This is the meteorological data set of the unmonitored photovoltaic power station; Step 3-4: Combine the output estimation result of transfer learning in formula (11) with the output estimation result of weight combination in formula (10) to improve the output estimation accuracy of the unmonitored PV power station; (12) in, The power generation of the unmonitored photovoltaic power station is estimated by combining weighted combination optimization and model migration optimization. , , , is the weight coefficient, which is solved by the optimization method. The photovoltaic power prediction result of the whole area is: (13) in, Represents the predicted photovoltaic power value of the fully distributed photovoltaic power generation area.

[0027] The above implementation modes are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for predicting regional distributed photovoltaic power under extreme data loss, characterized in that: The steps include: Step 1: sort out the available meteorological data and power generation data of the distributed photovoltaic power stations and centralized photovoltaic power stations in the region at different resolutions, and perform outlier identification and cubic spline interpolation processing on data missing and abnormal power generation points; Step 2: For centralized monitored power stations with complete data sampling in the region, the LightGBM method is used to model the relationship between meteorological data and photovoltaic power generation, and the frost optimization algorithm is used to optimize the hyperparameters of the LightGBM model to improve the prediction accuracy of centralized photovoltaic power generation in the region, providing a benchmark for subsequent predictions of unmonitored power stations; Step 3: Based on the prediction results of the monitored photovoltaic power stations in the region, the combined weight optimization is used to estimate the output of the unmonitored photovoltaic power stations in the region. At the same time, the power generation model of the monitored photovoltaic power stations is migrated to the unmonitored power stations to improve the estimation accuracy of the unmonitored photovoltaic power stations. Step 4: Summarize the prediction results of monitored PV power stations with complete data collection and the estimated results of unmonitored PV power stations with insufficient data sampling to obtain the distributed PV power prediction results in areas with extreme data loss.

2. According to the method of claim 1, the regional distributed photovoltaic power prediction method under extreme data loss is characterized in that: The step 1 comprises: Step 1-1: Determine all available power generation-related data of regional distributed photovoltaics, and organize all input variable data and output variable data at equal resolution. The input data is the meteorological data in the region, including available meteorological forecast data of light radiation, temperature, humidity, and air pressure. The output variables include the power generation data of monitored photovoltaic power stations and the power generation data of unmonitored photovoltaic power stations.

3. The method for predicting regional distributed photovoltaic power under extreme data loss according to claim 2, characterized in that: The step 1 also includes: Step 1-2: Collect historical operation data of each regional distributed photovoltaic power station, standardize the data, and divide it into training set and test set; （1） （2） （3） in, A historical set representing historical meteorological data and power generation components; Represents standardized historical data; represents the averaging function, It represents the standard deviation function; represents the training set taken from the historical data after standardization, represents the test set taken from the historical data after standardization, Indicates the ratio of the training set to the total data set.

4. The method for predicting regional distributed photovoltaic power under extreme data loss according to claim 3, characterized in that: The step 1 also includes: Step 1-3: Identify outliers in the sampled data and use the cubic spline method for interpolation and completion; （4） In each interval The lower limit of the interval And the regional limit Inside[ ]Construct a As a smooth curve function of the variable, establish a fitting formula ,in , , , is the coefficient to be determined, and the missing points of regional data are fitted and completed through the fitting formula of each interval.

5. The method for predicting regional distributed photovoltaic power under extreme data loss according to claim 1, characterized in that: The step 2 comprises: Step 2-1: For all monitored photovoltaic power stations with complete data collection in the region, a power generation data prediction model is built. The LightGBM method is used to model the relationship between meteorological input and power generation output. The training target formula is shown in Equation (5)-Equation (6): （5） （6） in, For the The characteristics of the sampled data, there are M data in total, is the corresponding true label, is the corresponding predicted label, To calculate the mean square error function, t represents the number of iterations, represents the target at the tth iteration, represents the newly generated CART fitting function at the tth iteration, represents the regularization term; T represents the number of leaves, and is the hyperparameter of the penalty term, Represents the weight value assigned to leaf node j.

6. The method for predicting regional distributed photovoltaic power under extreme data loss according to claim 5, characterized in that: The step 2 also includes: Step 2-2: Use the Frost Optimization algorithm to optimize the hyperparameters of the LightGBM model to further improve the accuracy and robustness of the power prediction of the monitored PV power station: set up =[n_estimators, num_leaves, max_depth, learning_rate] as optimization variables, n_estimators, num_leaves, max_depth, learning_rate are the number of iterations of the LightGBM model algorithm, the number of leaves, the maximum depth, and the learning rate, respectively. is the lower limit of the optimization variable, is the upper limit of the optimization variable; set up For the The optimal hyperparameters at the iteration time are updated using the soft time search strategy during the optimization process as follows: （7） In the formula, is the position of the updated hyperparameter, n represents the particle number, and t is the iteration number; is a random number in the range (-1,1). is a random number between (0,1), h is the degree of adhesion, which is a random number in the range of (0,1) and is used to control the distance between the centers of two gray particles. is the maximum number of iterations.

7. The method for predicting regional distributed photovoltaic power under extreme data loss according to claim 1, characterized in that: The step 3 comprises: Step 3-1: Determine the monitored PV power stations in the area and the set of power stations trained by the RiME-LightGBM algorithm , Power generation model collection And the collection of power stations without monitoring ; Weather forecast collection for monitored and unmonitored PV power stations , ; The capacity set of monitored photovoltaic power stations and unmonitored photovoltaic power stations, where there are M monitored photovoltaic power stations and N unmonitored photovoltaic power stations in the region; （8） in, Representing the region There are photovoltaic power stations with monitoring, Representing the region A power generation model with monitoring photovoltaic power station, Represents the nth unmonitored PV power station in the region, Representing the region There are meteorological datasets for monitoring photovoltaic power plants. Representing the region A meteorological dataset of an unmonitored PV power station. Representing the region The capacity of the photovoltaic power station is monitored. Representing the region The capacity of an unmonitored PV power station.

8. The method for predicting regional distributed photovoltaic power under extreme data loss according to claim 7, characterized in that: The step 3 also includes: Step 3-2: Determine the predicted value of the monitored photovoltaic power station in the area, bring the weather forecast into the LightGBM model trained based on historical data, and obtain the predicted value of the monitored photovoltaic power station in the area: （9） In the formula, represents the power generation forecast value of the mth monitored PV power station, is the power generation model of the mth monitored photovoltaic power station, is the weather forecast value of the mth photovoltaic power station.

9. A method for predicting regional distributed photovoltaic power under extreme data loss according to claim 8, characterized in that: The step 3 also includes: Step 3-3: For unmonitored PV plants with missing data, output estimates can be obtained by optimizing the weighted combination of power generation modes of monitored PV plants. : （10） in, is the weight, is the capacity of the nth unmonitored PV power station, is the capacity of the mth monitored PV power station; Represents the power generation of the unmonitored photovoltaic power station estimated by weighted combination optimization; The power generation model of the monitored PV power station is directly migrated to the unmonitored PV power station to obtain the output estimate: （11） in, represents the power generation of the unmonitored photovoltaic power station estimated by model migration optimization, is the combined weight of the power generation models of the unmonitored PV power station and different monitored PV power stations, This is the meteorological dataset of the unmonitored photovoltaic power station.

10. A method for predicting regional distributed photovoltaic power under extreme data loss according to claim 9, characterized in that: The step 3 also includes: Step 3-4: Combine the output estimation result of transfer learning in formula (11) with the output estimation result of weight combination in formula (10) to improve the output estimation accuracy of the unmonitored PV power station: （12） in, The power generation of the unmonitored photovoltaic power station is estimated by combining weighted combination optimization and model migration optimization. , , , is the weight coefficient, which is solved by the optimization method. The photovoltaic power prediction result of the whole area is: （13） in, Represents the predicted photovoltaic power value of the fully distributed photovoltaic power generation area.

Images