CN-122000877-A - Photovoltaic power prediction method and device based on historical data and weather

CN122000877ACN 122000877 ACN122000877 ACN 122000877ACN-122000877-A

Abstract

A photovoltaic power prediction method and device based on historical data and weather. The method comprises the steps of obtaining historical irradiation data, historical output data and historical weather data, generating a plurality of daily irradiation change curves through classifying the historical irradiation data, constructing a localized irradiation feature library, obtaining numerical weather forecast data of a target photovoltaic power station, retrieving the target daily irradiation change curve, correcting the numerical weather forecast data to generate target weather data, obtaining real-time power station operation data, inputting the target weather data and the real-time power station operation data into a machine learning prediction model to generate an initial power prediction curve, obtaining latest actual measurement power data, determining prediction deviation, correcting the initial power prediction curve and obtaining a final power prediction curve. By implementing the technical scheme provided by the application, the adaptability of the prediction curve to local unique climate rules is improved, so that the overall accuracy of photovoltaic power prediction is effectively improved.

Inventors

PAN YINGCHAO
DUAN XIAOHAN

Assignees

北京如实智慧电力科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260119

Claims (10)

1. A photovoltaic power prediction method based on historical data and weather, characterized in that the method is applied to a server, and comprises: acquiring historical irradiation data, historical output data and historical weather data of a target photovoltaic power station; Classifying the historical irradiation data to generate a plurality of daily irradiation change curves representing different weather patterns, and constructing a localized irradiation feature library according to the daily irradiation change curves, the historical output data and the historical weather data; Acquiring numerical weather forecast data of the target photovoltaic power station according to a preset forecast starting moment and a preset forecast duration; Searching a target daily irradiation change curve matched with the numerical weather forecast data in the localized irradiation feature library, and correcting the numerical weather forecast data by utilizing the target daily irradiation change curve to generate target meteorological data; Acquiring real-time power station operation data of the target photovoltaic power station, inputting the target meteorological data and the real-time power station operation data into a preset machine learning prediction model, and generating an initial power prediction curve; And acquiring the latest actual measurement power data of the target photovoltaic power station, determining a prediction deviation according to the latest actual measurement power data and the initial power prediction curve, and correcting the initial power prediction curve based on the prediction deviation to obtain a final power prediction curve.
2. The method of claim 1, wherein the generating a plurality of daily radiation profiles representing different weather patterns by classifying the historical radiation data and constructing a localized radiation profile based on the daily radiation profiles, the historical output data, and the historical weather data comprises: sequentially executing the processing of dividing according to natural days, aligning the time axis and normalizing the amplitude to the historical irradiation data to obtain a plurality of daily irradiation data sequences; Based on the data distribution characteristics of the plurality of daily irradiation data sequences, a preset unsupervised clustering algorithm is applied to divide the plurality of daily irradiation data sequences into a plurality of data clusters; For each data cluster, generating a corresponding daily irradiation change curve by calculating the mass centers of all daily irradiation data sequences in the data cluster; For each data cluster, extracting an output record corresponding in time to a daily irradiation data sequence associated with the data cluster from the historical output data, and calculating to obtain a historical output characteristic vector based on the output record; For each data cluster, extracting weather records corresponding in time to a daily irradiation data sequence associated with the data cluster from the historical weather data, and calculating a historical weather feature vector based on the weather records; and carrying out association storage on each daily irradiation change curve, the corresponding historical output characteristic vector and the corresponding historical weather characteristic vector to construct the localized irradiation characteristic library.
3. The method of claim 1, wherein retrieving a target daily irradiance profile matching the numerical weather forecast data in the localized irradiance signature library and correcting the numerical weather forecast data with the target daily irradiance profile, generating the target weather data, comprises: extracting weather features from the numerical weather forecast data, and constructing weather feature vectors to be matched based on the weather features; for each daily irradiation change curve in the localized irradiation feature library, calculating Euclidean distance between the weather feature vector to be matched and a historical weather feature vector associated with the daily irradiation change curve to obtain a dissimilarity degree score corresponding to the daily irradiation change curve; determining a target dissimilarity score with the smallest numerical value in all the dissimilarity scores, and determining a daily irradiation change curve corresponding to the target dissimilarity score as the target daily irradiation change curve; Carrying out fusion calculation on the irradiance prediction value in the numerical weather forecast data and the target daily irradiation change curve to obtain a corrected irradiance prediction value; And replacing the irradiance predicted value in the numerical weather forecast data with the corrected irradiance predicted value to generate the target meteorological data.
4. The method of claim 3, wherein said fusing the irradiance prediction in the numerical weather forecast data with the target daily irradiance profile to obtain a corrected irradiance prediction comprises: Obtaining a predicted value of total daily irradiation through integral operation of the predicted value of irradiance along the predicted time length; Obtaining a reference day total irradiation amount by carrying out integral operation on the target day irradiation change curve along the predicted time length; calculating an amplitude adjustment coefficient according to the predicted daily total irradiation amount value and the reference daily total irradiation amount; Multiplying the amplitude adjustment coefficient by the target daily irradiation change curve, and determining a multiplication result as the corrected irradiance prediction value.
5. The method of claim 1, wherein the obtaining the latest measured power data of the target photovoltaic power station, determining a prediction bias from the latest measured power data and the initial power prediction curve, and correcting the initial power prediction curve based on the prediction bias, to obtain a final power prediction curve, comprises: Taking the actual measurement power sequence of the target photovoltaic power station in a preset backtracking time period before the starting time of the predicted time period as the latest actual measurement power data; Extracting a historical predicted power sequence corresponding to each time point of the latest measured power data in the preset backtracking time length from the initial power prediction curve; calculating average power deviation in the preset backtracking time length according to the latest actual measured power data and the historical predicted power sequence, and determining the average power deviation as the predicted deviation; inputting the predicted deviation into a preset deviation attenuation algorithm, and calculating to obtain a correction quantity sequence attenuated along the predicted duration; and superposing the correction amount sequence and the initial power prediction curve to generate the final power prediction curve.
6. The method according to claim 1, wherein the method further comprises: Continuously acquiring an actual measurement power value corresponding to the time point of the final power prediction curve from a data acquisition system of the target photovoltaic power station in the time interval of the prediction duration to obtain subsequent actual measurement power data; Judging whether the prediction error continuously exceeds a preset performance degradation threshold value in a preset evaluation time window or not by calculating the prediction error between the follow-up measured power data and the final power prediction curve; When the prediction error is judged to continuously exceed the performance degradation threshold, triggering an online model updating instruction, and constructing an incremental training sample set according to the follow-up measured power data acquired in the preset evaluation time window and the target meteorological data corresponding to the time range of the follow-up measured power data.
7. The method of claim 6, wherein the method further comprises: In response to the model online update instruction, deconstructing the machine learning prediction model into a shared feature layer and a specific task layer; When the machine learning prediction model is updated, configuring network parameters of the shared feature layer into a gradient non-updated state; Based on the increment training sample set, only carrying out gradient calculation and weight update on the network parameters of the specific task layer through a back propagation algorithm to obtain an updated specific task layer; And combining the updated specific task layer with the shared feature layer with unchanged network parameters to generate an updated machine learning prediction model.
8. An electronic device comprising a processor, a memory, a user interface, and a network interface, the memory for storing instructions, the user interface and the network interface each for communicating with other devices, the processor for executing instructions stored in the memory to cause the electronic device to perform the method of any of claims 1-7.
9. A computer readable storage medium storing instructions which, when executed, perform the method of any one of claims 1-7.
10. A computer program product, characterized in that the computer program product, when run on an electronic device, causes the electronic device to perform the method of any of claims 1-7.

Description

Photovoltaic power prediction method and device based on historical data and weather Technical Field The application relates to the technical field of photovoltaic power generation, in particular to a photovoltaic power prediction method and device based on historical data and weather. Background Photovoltaic power prediction is a key link for ensuring safe and stable operation of a power grid. In the prior art, the photovoltaic power prediction method generally depends on two main data sources, namely, universal numerical weather forecast data issued by a weather service organization on one hand and historical output data acquired by a photovoltaic power station on the other hand. In practice, the two types of data are used as core input, and are processed through a pre-built machine learning model or a statistical model to generate a power prediction curve in a specific time period in the future, so that a decision basis is provided for power system dispatching. However, the above-mentioned prior art has a technical problem of insufficient accuracy in practical application. The root of the method is that the numerical weather forecast data adopted by the method is usually a wide-area and general model output, and the method cannot be deeply adapted to local weather features of the area where a specific photovoltaic power station is located. Therefore, when the actual weather change is influenced by specific geographical factors such as terrain, water area and the like to present a unique and periodic evolution rule, the deviation between the universal weather forecast data and the local actual weather conditions is obviously increased, and the deviation is directly transmitted to the output end of the prediction model, so that the accuracy of the final power prediction is reduced. Disclosure of Invention In order to solve the technical problems, the application provides a photovoltaic power prediction method and device based on historical data and weather. In a first aspect of the present application, a photovoltaic power prediction method based on historical data and weather is provided, and the following technical scheme is adopted: acquiring historical irradiation data, historical output data and historical weather data of a target photovoltaic power station; Classifying the historical irradiation data to generate a plurality of daily irradiation change curves representing different weather patterns, and constructing a localized irradiation feature library according to the daily irradiation change curves, the historical output data and the historical weather data; Acquiring numerical weather forecast data of the target photovoltaic power station according to a preset forecast starting moment and a preset forecast duration; Searching a target daily irradiation change curve matched with the numerical weather forecast data in the localized irradiation feature library, and correcting the numerical weather forecast data by utilizing the target daily irradiation change curve to generate target meteorological data; Acquiring real-time power station operation data of the target photovoltaic power station, inputting the target meteorological data and the real-time power station operation data into a preset machine learning prediction model, and generating an initial power prediction curve; And acquiring the latest actual measurement power data of the target photovoltaic power station, determining a prediction deviation according to the latest actual measurement power data and the initial power prediction curve, and correcting the initial power prediction curve based on the prediction deviation to obtain a final power prediction curve. By adopting the technical scheme, a localized irradiation characteristic library is constructed, and the numerical weather forecast data and the historical irradiation characteristics of the target power station are matched and corrected, so that systematic deviation between the general forecast data and local actual meteorological conditions is effectively relieved. On the basis, the real-time operation data and the machine learning model are combined to conduct preliminary prediction, and a deviation correction mechanism based on the latest actual measured power is introduced to form closed loop optimization. The method improves the adaptability of the prediction curve to the local unique climate rule, thereby effectively improving the overall accuracy of photovoltaic power prediction. Optionally, the step of classifying the historical irradiation data to generate a plurality of daily irradiation change curves representing different weather patterns, and constructing a localized irradiation feature library according to the daily irradiation change curves, the historical output data and the historical weather data includes: sequentially executing the processing of dividing according to natural days, aligning the time axis and normalizing the amplitude to the historical irradiatio