Distributed photovoltaic power station power prediction method and system based on meteorological similar day

Abstract

The application discloses a distributed photovoltaic power station power prediction method and system based on meteorological similar days, and comprises the following steps: obtaining historical data and meteorological data of a day to be predicted; dividing the meteorological data of the day to be predicted into time periods to obtain meteorological data of the day to be predicted in multiple time periods, and sequentially selecting historical data of the same weather type as the meteorological data of the day to be predicted in the time period to perform meteorological similarity weighted training to correspondingly obtain power prediction models in the multiple time periods; sequentially taking the meteorological data of the day to be predicted in different time periods as the input of the power prediction models in the corresponding time periods to sequentially output power prediction in different time periods; and performing smoothing processing on the power prediction in different time periods according to a filter function to obtain the final power prediction of the day to be predicted. The application can accurately perform power prediction on meteorological data in different time periods respectively, and can improve the accuracy and reliability of the power prediction of the distributed photovoltaic power station.

Description

Technical Field

[0001] This invention relates to the field of distributed photovoltaic power plants, and in particular to a method and system for predicting the power of distributed photovoltaic power plants based on meteorological similarity days. Background Technology

[0002] Currently, the medium- and low-voltage distribution networks in distributed photovoltaic (PV) systems have long been in a "blind spot" for new energy regulation, posing a significant challenge to grid operation. Therefore, distributed PV power prediction is a fundamental supporting technology for solving these regulation problems, and improving the accuracy of distributed PV predictions is urgently needed.

[0003] Photovoltaic power forecasting can be divided into two categories based on the forecast time: short-term power forecasting and medium- to long-term power forecasting. Short-term photovoltaic power forecasting is based on historical power output data, numerical weather prediction (NWP), and measured meteorological data to establish a power forecasting model to predict the photovoltaic output power for the next 24 hours. Furthermore, meteorological data has a significant impact on the forecast results, and the characteristics of meteorological data, such as long-term regularity and local randomness, also bring certain difficulties to photovoltaic power forecasting.

[0004] However, existing centralized photovoltaic (PV) forecasting methods all require modeling based on local data with the same spatiotemporal resolution, including measured and predicted data, as well as weather and electrical data. Distributed PV systems, however, are numerous, widely distributed, and exhibit significant individual variations. While centralized PV forecasting methods are often used for forecasting individual small PV systems, their accuracy is insufficient. Therefore, distributed PV systems mainly face two problems: first, poor power data quality, often containing long periods of null values ​​or outliers, while electricity generation data is relatively complete; second, the spatially dispersed nature of distributed PV systems leads to relatively high randomness in power generation. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by proposing a method and system for predicting the power of distributed photovoltaic power plants based on meteorological similarity days, which can improve the accuracy of power prediction for distributed photovoltaic power plants.

[0006] In a first aspect, the present invention provides a method for predicting the power output of a distributed photovoltaic power station based on meteorologically similar days, the method comprising: Acquire historical data and meteorological data for the day to be predicted; The meteorological data for the day to be predicted is divided into time periods, resulting in meteorological data for multiple time periods. Historical data with the same weather type as the meteorological data for the day to be predicted within each time period are then selected for weighted training based on meteorological similarity, resulting in power prediction models for multiple time periods. Specifically, the selection of historical data with the same weather type as the meteorological data for the day to be predicted within each time period for weighted training based on meteorological similarity involves: Obtain the meteorological factors corresponding to the meteorological data of the day to be predicted under the current period, and calculate the similarity between the meteorological factors of the day to be predicted under the current period and the meteorological factors of several historical meteorological data with the same weather type. Use the similarity as the weight of the loss function of the power prediction model, and perform multiple training sessions to obtain the power prediction model with the same weather type as the day to be predicted under the corresponding period. Furthermore, the loss function is calculated as follows: , in, The output of the power prediction model is... For the input of the first k Historical meteorological data with the same weather type, For the corresponding historical actual power data, The similarity between the meteorological data for the corresponding forecast date and the k-th historical meteorological data with the same weather type is calculated. N The number of historical data entries; Meteorological data for the day to be predicted at different time periods are used as inputs to the power prediction model for the corresponding time period, and power predictions for different time periods are output sequentially. The power predictions for different time periods are smoothed using a filtering function to obtain the final power prediction for the day to be predicted.

[0007] This invention uses meteorological data from different time periods as input to a power prediction model for the corresponding time period. By processing the meteorological data differently for each time period, it can accurately predict power for each period, improving the accuracy and reliability of power prediction for distributed photovoltaic power plants. Furthermore, by using the similarity between the meteorological data for the predicted day and meteorological data of the same weather type as weights to calculate a loss function, and after training, it can provide accurate power predictions for predicted days similar to historical meteorological data, further improving the accuracy of power prediction for distributed photovoltaic power plants. This invention also uses the similarity between meteorological factors of the same weather type as weights in the loss function, which can simulate the implicit power correlation between similar weather types, improving the accuracy of power prediction for weather types similar to historical meteorological data.

[0008] Furthermore, after acquiring historical power data, the historical theoretical power data to be repaired is then restored. Specifically, based on the historical meteorological data and historical power generation data, the historical power data of distributed photovoltaic power is restored; wherein, A complete historical power data segment is selected from the historical data, and the power generation curve of the complete historical power data is numerically integrated to obtain the estimated daily historical power. Based on the estimated daily historical power and the corresponding actual daily historical power, the daily historical output correction coefficient is calculated. Based on the daily historical output correction coefficient, the historical theoretical power to be corrected in the historical power data is corrected.

[0009] This invention corrects the theoretical power data to be repaired, which can correct incomplete or outlier data and expand the historical power data that can be used for training, thereby improving the quality of historical power data and thus improving the accuracy of power prediction for distributed photovoltaic power plants.

[0010] Furthermore, the step of repairing the historical theoretical power to be repaired in the historical power data according to the daily historical power output correction coefficient specifically involves: Based on historical meteorological data and the solar-to-electricity model, the historical theoretical power to be repaired is obtained, and the historical theoretical power to be repaired is repaired according to the daily historical power output correction coefficient.

[0011] Furthermore, the acquisition of the historical meteorological data specifically involves: Meteorological stations are gridded, and the meteorological data from the gridded stations are interpolated to obtain historical meteorological data on the latitude and longitude coordinates of the distributed photovoltaic power station.

[0012] Furthermore, the similarity is calculated by the Euclidean distance between the meteorological factors corresponding to the day to be predicted and the meteorological factors of several historical meteorological data with the same weather type.

[0013] Furthermore, after acquiring historical data and meteorological data for the day to be predicted, the process also includes: dividing the historical data into time periods based on radiation intensity, and classifying the historical data after dividing the time periods according to weather type to obtain historical data; wherein, Based on radiation intensity, the historical data for a day is divided into morning and afternoon periods; and the historical data for the morning and afternoon periods are further divided into weather types, resulting in 8 types of historical data, including: sunny morning, sunny afternoon, rainy morning, rainy afternoon, cloudy morning, cloudy afternoon, foggy morning, and foggy afternoon.

[0014] Furthermore, the calculation of the similarity can be expressed as: , in, and These are vectors representing historical meteorological data with the same weather type as the day to be predicted, and the corresponding meteorological data for the day to be predicted. Let be the Euclidean distance between the two.

[0015] Secondly, the present invention also provides a distributed photovoltaic power station power prediction system based on meteorological similarity days, comprising: The data acquisition module is used to acquire historical data and meteorological data for the day to be predicted; The power prediction model acquisition module is used to divide the meteorological data of the day to be predicted into time periods, obtain meteorological data of the day to be predicted in multiple time periods, and sequentially select historical data with the same weather type as the meteorological data of the day to be predicted in the same time period for meteorological similarity weighted training, thereby obtaining power prediction models for multiple time periods. The process involves sequentially selecting historical data of the same weather type as the meteorological data for the day to be predicted during that time period for weighted meteorological similarity training. Obtain the meteorological factors corresponding to the meteorological data of the day to be predicted under the current period, and calculate the similarity between the meteorological factors of the day to be predicted under the current period and the meteorological factors of several historical meteorological data with the same weather type. Use the similarity as the weight of the loss function of the power prediction model, and perform multiple training sessions to obtain the power prediction model with the same weather type as the day to be predicted under the corresponding period. Furthermore, the loss function is calculated as follows: , in, The output of the power prediction model is... For the input of the first k Historical meteorological data with the same weather type, For the corresponding historical actual power data, The similarity between the meteorological data for the corresponding forecast date and the k-th historical meteorological data with the same weather type is calculated. N The number of historical data entries; The power prediction module is used to take meteorological data of the day to be predicted at different time periods as input to the power prediction model at the corresponding time period, and output the power prediction at different time periods in sequence. The filtering module is used to smooth the power forecast for different time periods according to the filtering function, so as to obtain the final power forecast for the day to be predicted. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the process for predicting the power of a distributed photovoltaic power station based on meteorological similar days, provided in an embodiment of the present invention. Figure 2This is a schematic diagram comparing the power prediction accuracy of a photovoltaic power station in August 2019, provided by an embodiment of the present invention. Figure 3 This is a comparative schematic diagram of power prediction provided by an embodiment of the present invention; Figure 4 This is a schematic diagram of the process for predicting the power of a distributed photovoltaic power station based on meteorological similar days, provided in an embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of a distributed photovoltaic power station power prediction system based on meteorological similar days provided in an embodiment of the present invention. Detailed Implementation

[0017] 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.

[0018] See Figure 1 This is a flowchart illustrating the power prediction process for distributed photovoltaic power plants based on meteorologically similar days, provided in an embodiment of the present invention. It includes steps S11 to S14, specifically: Step S11: Obtain historical data and meteorological data for the day to be predicted; wherein, historical data includes: historical meteorological data, historical electricity data and historical power data.

[0019] Specifically, after acquiring historical power data, the historical theoretical power to be repaired is repaired, including: repairing the historical power data of distributed photovoltaics based on the historical meteorological data and historical power generation data; wherein, a complete historical power data segment is selected from the historical data, and the power generation curve of the complete historical power data is numerically integrated to obtain the daily historical estimated power; based on the daily historical estimated power and the corresponding daily historical actual power, the daily historical output correction coefficient is calculated; and based on the daily historical output correction coefficient, the historical theoretical power to be repaired in the historical power data is repaired.

[0020] Specifically, the step of repairing the historical theoretical power to be repaired in the historical power data according to the daily historical power output correction coefficient includes: obtaining the historical theoretical power to be repaired according to historical meteorological data and the photoelectric conversion model, and repairing the historical theoretical power to be repaired according to the daily historical power output correction coefficient.

[0021] It is worth noting that the historical data acquired for distributed photovoltaic (PV) power stations includes: historical meteorological data, historical power data, and historical electricity data. Historical meteorological data includes daily historical maximum, minimum, and average temperatures, relative humidity, rainfall, and wind speed. Furthermore, for periods with missing or outlier values ​​in the historical data, corrections are needed to provide sufficient historical data for training the power prediction model. It is noteworthy that the historical theoretical power data requiring correction includes the missing or outlier values. Since the daily historical power generation data for distributed PV power stations is relatively complete and highly reliable, historical meteorological data can be interpolated from historical meteorological data from gridded meteorological stations to obtain the meteorological conditions at the latitude and longitude coordinates of the distributed PV power station. This historical meteorological data, along with historical electricity data, can then be used to correct the historical theoretical power data.

[0022] Therefore, the period of historical power data without missing or outlier values ​​is selected as the complete historical power data, and the power generation curve of the complete historical power data for each day is numerically integrated to obtain the estimated daily historical power. Based on the corresponding actual daily historical power collected, the corresponding daily historical correction coefficient is calculated.

[0023] Preferably, the daily historical correction factor can be expressed as: , in, and These are the estimated daily historical electricity consumption and the actual daily historical electricity consumption, respectively.

[0024] Preferably, the theoretical historical power data can be expressed as: , in, For this type of power station collection j exist i Predicted irradiance at time of day The total effective area of ​​this type of power station cluster. This is the factor affecting the power output of this type of power plant's tilt angle. For this type of power station to be grouped in i Predict the conversion efficiency at constant temperature. This is the energy conversion loss factor based on the duration of use of this type of power plant.

[0025] Preferably, the restoration of the theoretical historical power data to be restored can be expressed as: , in, For this type of power station collectionj exist i The theoretical historical power data to be repaired at any given moment. This is the corresponding average correction factor.

[0026] The acquisition of historical meteorological data specifically involves: gridding the meteorological stations and interpolating the meteorological data from the gridded stations to obtain the historical meteorological data of the latitude and longitude coordinates of the distributed photovoltaic power station.

[0027] In addition, after acquiring historical data and meteorological data for the day to be predicted, the process also includes: dividing the historical data into time periods based on radiation intensity, and classifying the historical data after dividing the time periods into weather types based on weather type to obtain historical data; wherein, based on radiation intensity, the historical data of a day is divided into morning and afternoon periods; and the historical data of the morning and afternoon periods are classified into weather types respectively to obtain 8 types of historical data, including: sunny morning, sunny afternoon, rainy morning, rainy afternoon, cloudy morning, cloudy afternoon, foggy morning, and foggy afternoon.

[0028] It is worth noting that, due to the implicit correspondence between radiation intensity and power values ​​under similar weather conditions, classifying the weather types in historical meteorological data and clustering them yields four weather clustering results: sunny, rainy, cloudy, and foggy. Since weather conditions vary throughout the day, based on irradiance intensity, time periods are sequentially divided at the peak values ​​of historical meteorological data, resulting in morning and afternoon periods. Correspondingly, power prediction models for these periods are also divided into morning and afternoon power prediction models. Furthermore, the historical meteorological data for both morning and afternoon periods is further refined to achieve a more detailed weather classification, ultimately resulting in eight weather types in the historical meteorological data.

[0029] Preferably, the algorithm for clustering weather types is the K-means clustering algorithm; where K is 4.

[0030] Step S12: Divide the meteorological data of the day to be predicted into time periods to obtain meteorological data of the day to be predicted in multiple time periods. Then, select historical data with the same weather type as the meteorological data of the day to be predicted in the same time period for meteorological similarity weighted training to obtain power prediction models for multiple time periods.

[0031] Specifically, the step of sequentially selecting historical data with the same weather type as the meteorological data of the day to be predicted under the given time period for meteorological similarity weighted training involves: obtaining the meteorological factors corresponding to the meteorological data of the day to be predicted under the given time period, calculating the similarity between the meteorological factors of the day to be predicted under the given time period and the meteorological factors of several historical meteorological data with the same weather type, using the similarity as the weight of the loss function of the power prediction model, and performing multiple training sessions to obtain a power prediction model with the same weather type as the day to be predicted under the corresponding time period.

[0032] Specifically, the meteorological factors of the meteorological data for the day to be predicted are used to form a feature vector, where the meteorological factor is the corresponding irradiance. The feature vector of the meteorological data for the day to be predicted can be expressed as: , in, Let be the irradiance at time i during the day, and .

[0033] The feature vector of historical meteorological data can be represented as: , in, Let be the radiation intensity at time i on day d, and . .

[0034] Calculate the similarity between the meteorological factors of the day to be predicted and the meteorological factors of several historical meteorological data with the same weather type.

[0035] Preferably, the similarity is calculated by the Euclidean distance between the meteorological factors corresponding to the day to be predicted and the meteorological factors of several historical meteorological data with the same weather type.

[0036] Preferably, the similarity calculation can be expressed as: , in, and These are vectors representing historical meteorological data with the same weather type as the day to be predicted, and the corresponding meteorological data for the day to be predicted. Let be the Euclidean distance between the two.

[0037] Preferably, the loss function can be calculated as follows: , in, The output of the power prediction model is... For the input of the first k Historical meteorological data with the same weather type, For the corresponding historical actual power data, For the meteorological data of the corresponding forecast date and the weather type, the first... k Similarity between historical meteorological data N The number of historical data entries.

[0038] Preferably, the power prediction model uses a multilayer perceptron for power prediction.

[0039] To evaluate the accuracy of the power prediction model, the accuracy of the power prediction model is calculated based on the capacity of the distributed photovoltaic power station, the predicted power, and the corresponding historical power data.

[0040] Preferably, the accuracy evaluation uses capacity precision for assessment, which can be expressed as: , Where N is the number of power output points collected in a day. This represents the historical actual power value at the i-th time point within a day. Let Cap be the predicted power value at the i-th time point within a day, and Cap be the capacity of distributed photovoltaic power.

[0041] See Figure 2 This is a schematic diagram comparing the power prediction accuracy of a photovoltaic power station in August 2019, provided by an embodiment of the present invention. In the figure, the photovoltaic model based on meteorological similarity day (MSD-PV) proposed in this application has a higher accuracy than the traditional MLP model. The accuracy of the traditional MLP model is about 85%, while the average accuracy of the MSD-PV model proposed in this application is 91.53%. It can be seen that the MSD-PV model of this application has a higher accuracy.

[0042] Step S13: Take the meteorological data of the day to be predicted under different time periods as the input of the power prediction model under the corresponding time period, and output the power prediction under different time periods in sequence.

[0043] It is worth noting that by inputting the meteorological data for the morning and afternoon of the same day into the corresponding trained power prediction models for the morning and afternoon weather types, respectively, the power prediction results for the morning and afternoon periods are obtained. These power prediction results can be expressed as follows: ,in, The power prediction results are for the morning and afternoon periods, respectively.

[0044] It is worth noting that the power prediction model for the corresponding time period takes the meteorological data of the day to be predicted under the corresponding time period as input. Specifically, the weather types of the historical meteorological data are clustered, and the time periods are divided according to the irradiance intensity to obtain the historical meteorological data of the weather types corresponding to different time periods. The historical meteorological data corresponding to different time periods are used as the input of the power prediction model for the corresponding time period.

[0045] Step S14: Smooth the power predictions for different time periods according to the filtering function to obtain the final power prediction for the day to be predicted.

[0046] Preferably, the filtering function may employ a db4 filtering algorithm for filtering.

[0047] See Figure 3 This is a comparative schematic diagram of power prediction provided by an embodiment of the present invention. In the figure, the MSD-PV model proposed in this invention still achieves good power prediction in periods that are difficult to predict, and also achieves better power prediction results in periods with large historical actual power fluctuations. This indicates that the MSD-PV model proposed in this invention has higher power prediction accuracy. Furthermore, when actual irradiance fluctuates, the MSD-PV model of this invention has better prediction tracking ability and stronger prediction model generalization ability. It is especially suitable for areas with less short-term weather changes and more frequent long-term changes, and has significant advantages over the traditional MLP model.

[0048] This invention also provides a complete flowchart illustrating the power prediction process for distributed photovoltaic power plants based on meteorologically similar days, see [link / reference]. Figure 4 This is a schematic diagram of the process for predicting the power of a distributed photovoltaic power station based on meteorological similar days, provided in an embodiment of the present invention.

[0049] This invention also provides a schematic diagram of a distributed photovoltaic power station power prediction system based on meteorological similarity days, see [link / reference]. Figure 5 This is a schematic diagram of the structure of a distributed photovoltaic power station power prediction system based on meteorological similar days provided in an embodiment of the present invention. The system includes: a data acquisition module 51, a power prediction model acquisition module 52, a power prediction module 53, and a filtering module 54.

[0050] The data acquisition module 51 is used to acquire historical data and meteorological data for the day to be predicted; The power prediction model acquisition module 52 is used to divide the meteorological data of the day to be predicted into time periods, obtain meteorological data of the day to be predicted in multiple time periods, and sequentially select historical data with the same weather type as the meteorological data of the day to be predicted in the time period for meteorological similarity weighted training, thereby obtaining power prediction models in multiple time periods. The power prediction module 53 is used to take the meteorological data of the day to be predicted under different time periods as the input of the power prediction model under the corresponding time period, and output the power prediction under different time periods in sequence. The filtering module 54 is used to smooth the power prediction for different time periods according to the filtering function, so as to obtain the final power prediction for the day to be predicted.

[0051] This invention uses meteorological data from different time periods as input to the power prediction model for the corresponding time period, and performs differentiated processing on the meteorological data from different time periods. This enables accurate power prediction for meteorological data from different time periods, improving the accuracy and reliability of power prediction for distributed photovoltaic power plants. Furthermore, by using the similarity between the meteorological data of the day to be predicted and meteorological data of the same weather type as weights to calculate the loss function, after training, it can provide accurate power predictions for the day to be predicted that is similar to historical meteorological data, further improving the accuracy of power prediction for distributed photovoltaic power plants.

[0052] Those skilled in the art will understand that embodiments of this application may also include computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0053] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the power output of a distributed photovoltaic power station based on meteorologically similar days, characterized in that, The method includes: Acquire historical data and meteorological data for the day to be predicted; The meteorological data of the day to be predicted is divided into time periods to obtain meteorological data of the day to be predicted in multiple time periods. Then, historical data with the same weather type as the meteorological data of the day to be predicted in the same time period are selected for meteorological similarity weighted training to obtain power prediction models for multiple time periods. The process involves sequentially selecting historical data of the same weather type as the meteorological data for the day to be predicted during that time period for weighted meteorological similarity training. Obtain the meteorological factors corresponding to the meteorological data of the day to be predicted under the current period, and calculate the similarity between the meteorological factors of the day to be predicted under the current period and the meteorological factors of several historical meteorological data with the same weather type. Use the similarity as the weight of the loss function of the power prediction model, and perform multiple training sessions to obtain the power prediction model with the same weather type as the day to be predicted under the corresponding period. The similarity is calculated as follows: ,in, For the meteorological data of the corresponding forecast date and the weather type, the first... k Similarity between historical meteorological data and These are vectors of historical meteorological data with the same weather type as the day to be predicted, and the corresponding meteorological data for the day to be predicted. Let be the Euclidean distance between the two. The loss function is calculated as follows: ,in, The output of the power prediction model is... For the input of the first k Historical meteorological data with the same weather type, For the corresponding historical actual power data, N The number of historical data entries; Meteorological data for the day to be predicted at different time periods are used as inputs to the power prediction model for the corresponding time period, and power predictions for different time periods are output sequentially. The power predictions for different time periods are smoothed using a filtering function to obtain the final power prediction for the day to be predicted.

2. The method for predicting the power output of a distributed photovoltaic power station based on meteorological similarity days as described in claim 1, characterized in that, After acquiring historical power data, the historical theoretical power data to be repaired is then restored. Specifically, this involves restoring the historical power data of distributed photovoltaic systems based on the historical meteorological data and historical power generation data. A complete historical power data segment is selected from the historical data, and the power generation curve of the complete historical power data is numerically integrated to obtain the estimated daily historical power. Based on the estimated daily historical power and the corresponding actual daily historical power, the daily historical output correction coefficient is calculated. Based on the daily historical output correction coefficient, the historical theoretical power to be corrected in the historical power data is corrected.

3. The distributed photovoltaic power station power prediction method based on meteorological similarity days as described in claim 2, Its features are, The step of repairing the historical theoretical power in the historical power data according to the daily historical power output correction coefficient is as follows: Based on historical meteorological data and the solar-to-electricity model, the historical theoretical power to be repaired is obtained, and the historical theoretical power to be repaired is repaired according to the daily historical power output correction coefficient.

4. The method for predicting the power output of a distributed photovoltaic power station based on meteorological similarity days as described in claim 1, characterized in that, The acquisition of the historical meteorological data specifically involves: Meteorological stations are gridded, and the meteorological data from the gridded stations are interpolated to obtain historical meteorological data on the latitude and longitude coordinates of the distributed photovoltaic power station.

5. The method for predicting the power output of a distributed photovoltaic power station based on meteorological similarity days as described in claim 1, characterized in that, The similarity is calculated by the Euclidean distance between the meteorological factors corresponding to the day to be predicted and the meteorological factors of several historical meteorological data with the same weather type.

6. The method for predicting the power output of a distributed photovoltaic power station based on meteorological similarity days as described in claim 1, characterized in that, After acquiring historical data and meteorological data for the day to be predicted, the process further includes: dividing the historical data into time periods based on radiation intensity, and classifying the historical data into weather types based on weather type to obtain historical data; wherein, Based on radiation intensity, the historical data for a day is divided into morning and afternoon periods; and the historical data for the morning and afternoon periods are further divided into weather types, resulting in 8 types of historical data, including: sunny morning, sunny afternoon, rainy morning, rainy afternoon, cloudy morning, cloudy afternoon, foggy morning, and foggy afternoon.

7. A distributed photovoltaic power station power prediction system based on meteorological similarity days, characterized in that, include: The data acquisition module is used to acquire historical data and meteorological data for the day to be predicted; The power prediction model acquisition module is used to divide the meteorological data of the day to be predicted into time periods, obtain meteorological data of the day to be predicted in multiple time periods, and sequentially select historical data with the same weather type as the meteorological data of the day to be predicted in the same time period for meteorological similarity weighted training, thereby obtaining power prediction models for multiple time periods. The process involves sequentially selecting historical data of the same weather type as the meteorological data for the day to be predicted during that time period for weighted meteorological similarity training. Obtain the meteorological factors corresponding to the meteorological data of the day to be predicted under the current period, and calculate the similarity between the meteorological factors of the day to be predicted under the current period and the meteorological factors of several historical meteorological data with the same weather type. Use the similarity as the weight of the loss function of the power prediction model, and perform multiple training sessions to obtain the power prediction model with the same weather type as the day to be predicted under the corresponding period. The similarity is calculated as follows: ,in, For the meteorological data of the corresponding forecast date and the weather type, the first... k Similarity between historical meteorological data and These are vectors of historical meteorological data with the same weather type as the day to be predicted, and the corresponding meteorological data for the day to be predicted. Let be the Euclidean distance between the two. The loss function is calculated as follows: ， in, The output of the power prediction model is... For the input of the first k Historical meteorological data with the same weather type, For the corresponding historical actual power data, N The number of historical data entries; The power prediction module is used to take meteorological data of the day to be predicted at different time periods as input to the power prediction model at the corresponding time period, and output the power prediction at different time periods in sequence. The filtering module is used to smooth the power forecast for different time periods according to the filtering function, so as to obtain the final power forecast for the day to be predicted.

Images