Method for predicting generating capacity of distributed photovoltaic power station by combining grid point meteorological data

Abstract

The invention discloses a distributed photovoltaic power station generating capacity prediction method based on qualified point meteorological data, and relates to the technical field of photovoltaic power generation, and the method comprises the steps: obtaining the geographic information and historical power generation data of a photovoltaic power station, carrying out the space matching through combining the historical grid point meteorological data, and building the grid point-position correlation; extracting a plurality of groups of historical data based on the association, performing weather scene classification, training a scene-based generating capacity prediction model, and constructing a prediction network; and inputting real-time grid point meteorological data into the prediction network, and performing power generation prediction through a dynamic feature selection mechanism to obtain a high-precision power generation prediction result. According to the method, the technical problems of low prediction precision and poor adaptability of a traditional photovoltaic power generation capacity prediction method in the prior art when coping with complex meteorological changes are solved, and the purpose of predicting the power generation capacity by introducing high-precision grid point meteorological data and a special scenario prediction model is achieved. And the prediction precision and adaptability of the photovoltaic generating capacity under the complex meteorological condition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and in particular to a method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data. Background Art

[0002] With the continuous growth of global energy demand and the emphasis on environmental protection, the use of clean energy has become particularly important. Photovoltaic power generation, as a green and renewable energy source, has received widespread attention due to its pollution-free and sustainable characteristics. In particular, distributed photovoltaic power stations have the advantages of on-site power generation and on-site consumption, which can effectively alleviate the problem of power loss during power transmission. However, since photovoltaic power generation is highly dependent on meteorological conditions such as solar radiation, it has significant volatility and uncertainty. Therefore, how to accurately predict the power generation of photovoltaic power stations has become the key to its large-scale application in power grids.

[0003] Traditional photovoltaic power generation prediction methods usually rely on historical power generation data and simple weather forecast models, which makes it difficult to fully consider the differences in different meteorological conditions within a spatial range, and the prediction accuracy is often not high when meteorological conditions change rapidly. Summary of the invention

[0004] The present application provides a distributed photovoltaic power station power generation prediction method combined with grid meteorological data, which is used to solve the technical problems of low prediction accuracy and poor adaptability of traditional photovoltaic power generation prediction methods in the prior art when dealing with complex meteorological changes.

[0005] The present application provides a method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data, the method comprising: obtaining geographic information data and historical power generation data of a target photovoltaic power station; collecting historical grid meteorological data, spatially matching it with the geographic information data, and establishing a grid-position association; based on the grid-position association, correlating and extracting the historical grid meteorological data and historical power generation data to obtain multiple groups of historical power generation data; using the multiple groups of historical power generation data to extract weather scenarios to obtain multiple historical weather scenarios, and training corresponding power generation prediction models based on the multiple historical weather scenarios to construct a power generation prediction network; based on the grid-position association, collecting real-time grid meteorological data, inputting it into the power generation prediction network, and introducing a dynamic feature selection mechanism to predict power generation to obtain a power generation prediction result of the target photovoltaic power station.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The distributed photovoltaic power station power generation prediction method combined with grid meteorological data provided in the present application relates to the field of photovoltaic power generation technology. By combining grid meteorological data with photovoltaic power station geographic information, a grid-location association is established, historical weather scenarios are extracted, a multi-scenario power generation prediction model is constructed, and real-time meteorological data is used for dynamic feature selection and prediction optimization. The technical problems of low prediction accuracy and poor adaptability of traditional photovoltaic power generation prediction methods in the prior art when dealing with complex meteorological changes are solved, and the technical effect of improving the prediction accuracy and adaptability of photovoltaic power generation under complex meteorological conditions is achieved by introducing high-precision grid meteorological data and a special scenario-based prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic diagram of a flow chart of a method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data provided in an embodiment of the present application; Figure 2 A schematic diagram of a process for constructing a power generation prediction network in a distributed photovoltaic power station power generation prediction method combined with grid meteorological data provided in an embodiment of the present application; Figure 3 A schematic diagram of a process for introducing a dynamic feature selection mechanism to predict power generation in a distributed photovoltaic power station power generation prediction method combined with grid meteorological data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] The present application provides a distributed photovoltaic power station power generation prediction method combined with grid meteorological data, which is used to solve the technical problems of low prediction accuracy and poor adaptability of traditional photovoltaic power generation prediction methods in the prior art when dealing with complex meteorological changes.

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0011] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices.

[0012] Embodiment 1, as Figure 1 As shown, the present application provides a method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data, the method comprising: P10: Obtain geographic information data and historical power generation data of the target PV power station.

[0013] Optionally, the first step before starting the distributed photovoltaic power station power generation prediction process is to fully obtain the geographic information data and historical power generation data of the target photovoltaic power station. Among them, the geographic information data covers the specific geographical location of the photovoltaic power station (such as longitude and latitude coordinates), topographic features (such as descriptions of natural environments such as mountains, plains or waters), altitude (an important factor affecting light intensity and temperature), and the shading of the surrounding environment (such as the degree of shading of sunlight by buildings, trees, etc.). These data are collected through high-precision maps, geographic information systems (GIS) and field surveys. This information is crucial for accurately evaluating the amount of solar radiation received by the photovoltaic power station and the impact of environmental factors on power generation efficiency.

[0014] At the same time, it is also necessary to collect the historical power generation data of the target photovoltaic power station, that is, the actual power generation records of the power station in different time periods. These data are collected and sorted through the monitoring system or energy management system (EMS) of the power station, reflecting the actual power generation performance of the power station in different seasons and weather conditions, and are important sample data for training power generation prediction models. Through in-depth analysis of these historical data, the periodic changes in power generation of the power station and the impact of specific weather conditions on power generation can be identified, thereby providing strong support for subsequent power generation predictions.

[0015] In order to ensure the reliability of the data, it is necessary to clean the historical power generation data. This process includes removing abnormal data points caused by equipment failure or maintenance, and filling in missing values ​​to ensure the integrity and accuracy of the data. Through the detailed acquisition and processing of geographic information data and historical power generation data, the model can more accurately understand the power generation characteristics of photovoltaic power stations under different meteorological conditions, laying a solid data foundation for subsequent power generation forecasts. The high-quality acquisition of these data directly affects the accuracy and adaptability of the prediction model, enabling the model to better reflect the actual operation of photovoltaic power stations.

[0016] P20: Collect historical grid meteorological data, perform spatial matching with the geographic information data, and establish grid-location association.

[0017] Furthermore, step P20 of the embodiment of the present application further includes: P21: Based on a preset long-term period, the interactive meteorological platform collects historical grid meteorological data; P22: Based on the geographic information data of the target photovoltaic power station, key location information is extracted, and spatial position matching is performed with the historical grid meteorological data, and based on the spatial matching result, the grid-position association is established based on neighboring grid points or interpolation calculations.

[0018] It should be understood that in the distributed photovoltaic power station power generation prediction process, collecting historical grid meteorological data and spatially matching it with geographic information data is an important part of establishing a prediction model. First, according to the preset long-term period, the historical grid meteorological data is systematically collected through an interactive meteorological platform (such as a meteorological station or a numerical weather forecast center). Among them, the long-term period refers to the time period setting for data collection within a longer time span (such as a time range of 1 year or more). This periodic collection can ensure that the model covers a variety of seasonal meteorological changes, such as high radiation in summer and low radiation in winter, and comprehensively captures the long-term impact of meteorology on power generation. The historical grid meteorological data is distributed in a gridded form, including temperature, humidity, wind speed, wind direction, solar radiation intensity and other meteorological elements. It has a certain resolution in space and time and can accurately reflect the atmospheric conditions.

[0019] Subsequently, according to the geographic information data of the target photovoltaic power station, key location information is extracted, such as the latitude and longitude coordinates, altitude, etc. of the power station. This information is the basis for spatial matching. The key location information is spatially matched with the historical grid meteorological data, and the meteorological data that accurately corresponds to the power station location is obtained by finding the nearest grid data or using spatial interpolation methods (such as inverse distance weighted interpolation, Kriging interpolation, etc.). For example, if the location of the power station is not completely located on a certain meteorological grid point, but between multiple grid points, two methods can be used to establish a grid-location association: first, directly select the data of one or more grid points closest to the location of the photovoltaic power station as the meteorological data input of the power station. This method is simple and easy to use and is suitable for situations where the meteorological data resolution is high. Second, when the power station location is between multiple grid points, an interpolation algorithm (such as bilinear interpolation, inverse distance weighted interpolation, etc.) can be used to calculate the meteorological data value of the power station location. The interpolation calculation calculates the meteorological element value at the target location by combining the data of the surrounding grid points and taking the distance as the weight. For example, the bilinear interpolation algorithm can calculate the solar radiation intensity or temperature value at the location based on the distance from the power station to the four surrounding grid points. The interpolation method can smoothly reflect spatial changes and is more suitable for situations where the meteorological data resolution is low or the power station is located in an area with complex terrain.

[0020] Through the above spatial matching and interpolation calculations, the association between the photovoltaic power station and the meteorological grid points is finally established, so that in the subsequent model training and real-time prediction, the meteorological conditions corresponding to the power station location can be quickly matched. This association ensures that the meteorological data used accurately reflects the weather conditions at the location of the photovoltaic power station, avoiding prediction deviations caused by data errors.

[0021] P30: Based on the grid-position association, the historical grid meteorological data and the historical power generation data are associated and extracted to obtain multiple groups of historical power generation data.

[0022] Furthermore, step P30 of the embodiment of the present application further includes: P31: Based on the grid-position association, the historical grid meteorological data and the historical power generation data are aligned in time dimension and unified in data frequency to obtain a standard historical sample data sequence; P32: For the standard historical sample data sequence, each time point in the historical time period is traversed to extract associated data to form multiple groups of sample data, wherein each group of sample data contains multiple meteorological elements and corresponding power generation sample data; P33: According to the historical operation and maintenance records of the target photovoltaic power station, the multiple groups of sample data are screened for effective time periods to obtain multiple groups of historical power generation data.

[0023] Specifically, based on the established grid-location association, the historical grid meteorological data and historical power generation data are deeply integrated and extracted. First, the historical grid meteorological data are aligned with the historical power generation data of the photovoltaic power station in the time dimension, that is, to ensure that the meteorological data and power generation data at the same time are accurately matched. For example, if the recording frequency of meteorological data and power generation data is different (such as meteorological data is recorded every hour, while power generation data is recorded every 15 minutes), alignment processing needs to be performed through interpolation, averaging or aggregation.

[0024] Next, adjust the meteorological data and power generation data to a consistent time frequency to generate standardized time series data. For example, the power generation data can be aggregated hourly, or the meteorological data can be linearly interpolated to make the time frequencies of the two consistent, ensuring that the model input data is synchronized in time, thereby avoiding data deviations caused by frequency differences. This step can use time series analysis techniques and data preprocessing software, such as Python's Pandas library, which can efficiently process large-scale time series data and achieve data alignment and frequency unification.

[0025] Next, the standardized data sequence is traversed, and the data samples at each time point are extracted one by one. The traversed time period can be all historical periods from the start of the operation of the photovoltaic power station to the current time, or a specific season or year can be selected according to actual needs. The core of this process is that for each time point, the values ​​of various meteorological elements (such as temperature, humidity, wind speed, solar radiation intensity, etc.) are paired with the corresponding power generation sample data to form multiple groups of sample data. Each group of sample data is a complete record containing multiple meteorological elements and power generation data, which together constitute the training set of the power generation prediction model. The implementation of this step depends on data processing and mining techniques, such as data frame operations, loop traversal, and conditional screening.

[0026] Finally, according to the historical operation and maintenance records of the target PV power station, the above multiple sets of sample data are screened for effective time periods to remove abnormal or invalid power generation data caused by equipment failure, maintenance downtime or other non-weather factors, ensuring that the sample data used for training can truly reflect the relationship between weather conditions and power generation. Effective time period screening usually involves a detailed analysis of operation and maintenance records, as well as statistical tests and outlier detection of sample data. Through this step, multiple sets of high-quality historical power generation data are obtained, providing reliable training samples for subsequent power generation prediction model training, so that the power generation prediction model can better understand the impact of meteorological changes on power generation performance, thereby improving the reliability and accuracy of the prediction.

[0027] P40: Use the multiple sets of historical power generation data to extract weather scenarios, obtain multiple historical weather scenarios, and train corresponding power generation prediction models based on the multiple historical weather scenarios to construct a power generation prediction network.

[0028] Further, such as Figure 2 As shown, step P40 of the embodiment of the present application also includes: P41: Based on historical power generation records, multiple key meteorological elements are extracted, and the key meteorological elements include but are not limited to solar radiation intensity, cloud cover, temperature, and sunshine duration; P42: Element thresholds are set for the multiple key meteorological elements respectively, and based on the element thresholds, the multiple groups of historical power generation data are weather clustered to obtain multiple historical weather scenarios; P43: According to the multiple historical weather scenarios, the multiple groups of historical power generation data are divided into multiple data subsets, and scenario labels are generated for each data subset; P44: Using the multiple data subsets, power generation prediction models are trained under different weather scenarios, and the trained multiple models are combined to generate a power generation prediction network, and the power generation prediction network also has a built-in weather scenario classifier.

[0029] Optionally, by extracting multiple key meteorological elements based on historical power generation records and analyzing weather scenarios, prediction models are trained and combined to improve the prediction accuracy of the model under different meteorological conditions.

[0030] First, key meteorological factors that have a significant impact on power generation are extracted from historical power generation records. These factors are not limited to solar radiation intensity (which directly affects the ability of photovoltaic panels to convert light energy received into electrical energy), but also include cloud cover (affecting the penetration of solar radiation), temperature (affecting photovoltaic panel efficiency and battery chemical performance), and sunshine duration (reflecting the duration of effective light in a day). The extraction of these key meteorological factors relies on data analysis and feature selection techniques, such as correlation analysis and principal component analysis (PCA), which can identify variables with the most predictive value for power generation.

[0031] Next, element thresholds are set for the multiple key meteorological elements. These thresholds are set based on the statistical characteristics of historical data and the specific environment of the photovoltaic power station. They are numerical limits set for each key meteorological element to distinguish different weather conditions. For example, thresholds for high radiation (greater than 800W / ㎡), medium radiation (400-800W / ㎡) and low radiation (less than 400W / ㎡) can be set for solar radiation intensity. Based on these element thresholds, weather clustering analysis is performed on multiple groups of historical power generation data. By classifying meteorological data at different time points, similar meteorological conditions are divided into the same category. For example, data with sunny days, high temperatures, and long sunshine duration can be clustered into one scenario, while cloudy and low radiation data can be clustered into another scenario. The implementation of this step relies on clustering algorithms, such as K-means or hierarchical clustering, which can automatically discover weather patterns based on the intrinsic structure of the data.

[0032] Subsequently, based on the identified historical weather scenarios, multiple groups of historical power generation data were divided into multiple data subsets, and each subset was assigned a scenario label, such as "high radiation on sunny days" and "moderate radiation on cloudy days". The scenario labels are used to identify the meteorological category to which the subset data belongs. These labels not only simplify the organization and management of data, but also provide a clear classification basis for subsequent model training.

[0033] Finally, these scenario-labeled data subsets are used to train power generation prediction models under different weather scenarios. These models can be based on machine learning regression models, such as linear regression, decision trees, random forests, or support vector machines (SVMs), or they can be deep learning models, such as neural networks. Each model is optimized for a specific weather scenario to improve the accuracy and generalization of the prediction. After training, these models are combined into a power generation prediction network, which also has a built-in weather scenario classifier to automatically identify the current weather scenario during real-time prediction and select the most appropriate model for power generation prediction. The implementation of this step relies on ensemble learning technology and model fusion algorithms, which can fully utilize the prediction advantages of multiple models and improve the overall prediction performance.

[0034] This multi-model collaborative working mode can automatically switch prediction models under different weather scenarios, thereby improving prediction accuracy and adaptability, making the power generation prediction of photovoltaic power stations more accurate and efficient.

[0035] P50: Based on the grid-position association, real-time grid meteorological data is collected and input into the power generation prediction network, and a dynamic feature selection mechanism is introduced to perform power generation prediction to obtain a power generation prediction result of a target photovoltaic power station.

[0036] Further, such as Figure 3 As shown, step P50 of the embodiment of the present application also includes: P51: According to the preset collection frequency, the interactive meteorological platform collects real-time grid meteorological data and inputs it into the power generation prediction network; P52: The weather scenario classifier in the power generation prediction network is used to perform scenario classification, obtain the target weather scenario, and match it to obtain the target power generation prediction model; P53: A dynamic feature selection mechanism is introduced to perform feature importance analysis on the real-time grid meteorological data and configure feature importance identification; P54: According to the feature importance identification, after adjusting the feature weight, the target power generation prediction model is used to perform power generation prediction to obtain the power generation prediction result of the target photovoltaic power station.

[0037] Preferably, based on the grid-location association, real-time meteorological data is collected and combined with a power generation prediction network for prediction, aiming to provide accurate power generation prediction for photovoltaic power stations based on real-time meteorological data.

[0038] First, according to the preset collection frequency (such as every hour, every half hour, etc., the specific frequency is set according to the prediction accuracy requirements), through real-time interaction with the meteorological platform, the real-time grid meteorological data of the target photovoltaic power station is collected. These data include a variety of meteorological elements, such as solar radiation intensity, cloud cover, temperature, wind speed, etc., which are crucial for evaluating the power generation potential at present and in the future. The collected real-time data is then input into the power generation prediction network to provide a basis for subsequent prediction analysis.

[0039] Next, the weather scenario classifier embedded in the power generation forecast network is used to classify the real-time meteorological data. By comparing the similarity between real-time data and historical weather scenarios, the current weather scenario (such as sunny, cloudy, rainy, etc.) can be quickly identified, and the most appropriate power generation forecast model can be matched accordingly. The implementation of the weather scenario classifier relies on machine learning algorithms, such as decision trees, support vector machines, or deep learning models, which can accurately distinguish different weather patterns and provide a reliable basis for model selection.

[0040] Then, a dynamic feature selection mechanism is introduced to perform feature importance analysis on real-time grid meteorological data. The core of this step is to evaluate the contribution of each meteorological element to power generation forecasting, that is, their weight in the forecasting model. For example, on sunny days, solar radiation may be more important, while on cloudy days, cloud cover may be a more important feature. By applying feature selection algorithms such as recursive feature elimination, random forest importance assessment, or Lasso regression, each feature can be assigned an importance label that reflects their relative importance in the forecasting process.

[0041] Finally, the feature weights are adjusted according to the feature importance identifiers to optimize the performance of the prediction model. Specifically, for features with higher importance, their weights may be increased to play a greater role in the prediction process; for features with lower importance, their weights may be reduced or even ignored. The adjusted feature weights are applied to the target power generation prediction model, which predicts power generation based on real-time meteorological data and the adjusted feature weights, and finally outputs the power generation prediction results of the target PV power station. The implementation of this step relies on model training and optimization techniques, such as gradient descent, stochastic gradient descent, or Adam optimizer, which can ensure that the model maintains high performance and accuracy in real-time prediction.

[0042] Through the above steps, real-time and efficient prediction of photovoltaic power station power generation can be achieved. This method can not only adjust the model input according to the latest meteorological data, but also improve the adaptability of the model under complex weather conditions through a dynamic feature selection mechanism, thereby providing accurate power generation prediction support for power grid dispatching and photovoltaic power station management. This solution ensures the efficient operation of photovoltaic power stations under various weather changes and helps to increase the proportion of renewable energy in the power system.

[0043] Furthermore, step P53 of the embodiment of the present application also includes: P53-1: Based on multiple historical weather scenarios, with meteorological characteristics as input and actual power generation as output, multiple corresponding feature contribution evaluation models are trained respectively; P53-2: Match and obtain the feature contribution evaluation model corresponding to the target weather scenario, perform feature importance analysis on the real-time grid meteorological data, and obtain the feature importance ranking under the target weather scenario; P53-3: Based on the feature importance ranking, dynamically select meteorological features in combination with the importance threshold, and identify the feature importance.

[0044] Specifically, in the in-depth processing stage of distributed photovoltaic power station power generation prediction, by introducing a dynamic feature selection mechanism, the model's attention to various meteorological features can be automatically adjusted according to different weather scenarios, and dynamic feature selection of real-time grid meteorological data can be realized, thereby further improving the accuracy and efficiency of power generation prediction.

[0045] First, based on a variety of historical weather scenarios, with meteorological features (such as solar radiation intensity, cloud cover, temperature, wind speed, etc.) as input variables and actual power generation as output variables, multiple corresponding feature contribution evaluation models are trained. These models may be regression models based on machine learning, such as linear regression, random forest, gradient boosting tree (GBM) or support vector regression (SVR), which can learn and quantify the contribution of each meteorological feature to the actual power generation forecast. During the training process, the model will continuously adjust its internal parameters to minimize the prediction error, so as to accurately capture the complex relationship between features and power generation.

[0046] Next, in the real-time prediction process, the weather scenario classifier is used to identify the current weather scenario type (such as whether it is sunny, cloudy, etc.). According to the identified target weather scenario, the previously trained corresponding feature contribution evaluation model is selected. For example, if the classifier determines that the current scenario is "cloudy", the feature contribution evaluation model trained under the cloudy scenario is selected. Subsequently, the model is used to perform feature importance analysis on the real-time grid meteorological data, and the feature importance ranking under the target weather scenario is obtained by calculating the contribution score of each feature.

[0047] Finally, based on the feature importance ranking, the meteorological features are dynamically selected in combination with the preset importance threshold. Specifically, the importance threshold is a pre-set value used to filter out the meteorological features that need to be considered in the forecast. Only features with an importance score higher than the threshold will be retained and given a higher weight in subsequent forecasts. According to the feature importance ranking and the importance threshold, the features with the most predictive value in the current weather scenario are screened out. For example, if the solar radiation score is much higher than the cloud cover in a sunny scenario, the solar radiation will be preferred. The screened features are marked so that they are given higher weights in the forecast model. The marking process can ensure that the model pays more attention to the impact of these features on power generation when forecasting by weighting the importance scores of the features. For example, the values ​​of these high-weight features can be amplified when inputting the model, so that the model can capture the changes of these features more sensitively.

[0048] Through the above steps, the system can dynamically adjust the meteorological feature weights of the model input according to the current weather scenario during the real-time prediction process. This method ensures that the model can flexibly select the most predictive features according to different weather changes, thereby providing more accurate power generation forecast results under complex and changing meteorological conditions.

[0049] Furthermore, the embodiment of the present application further includes step P60, and step P60 further includes: P61: Based on the demand for high-precision short-term forecasts, preset the short-term data update frequency; P62: Based on historical power generation records, extract instantaneous meteorological change data and power generation data, and build an ultra-short-term prediction model; P63: According to the short-term data update frequency, real-time grid meteorological data is collected and input into the ultra-short-term prediction model to quickly predict power generation and obtain short-term high-precision prediction results for short-term power load regulation.

[0050] In a possible embodiment of the present application, in order to meet the needs of high-precision short-term forecasting, rapid prediction of power generation and power load regulation of photovoltaic power stations within a short time range are achieved through the setting of data update frequency and real-time application of models.

[0051] First, according to the specific needs of high-precision short-term forecasts, a short-term data update frequency is preset. This frequency is usually much higher than the daily meteorological data update cycle, and may be updated every few minutes or even every second to ensure that the forecast model can capture the rapid changes in meteorological conditions in a timely manner. Presetting a higher data update frequency can collect the latest meteorological data more frequently, reflect subtle changes in the weather in a timely manner, and ensure that the forecast model is always calculated based on the latest input.

[0052] Next, instantaneous meteorological change data and power generation data are extracted from historical power generation records for building an ultra-short-term prediction model. The instantaneous meteorological change data include changes in solar radiation intensity, rapid cloud movement, temperature fluctuations, etc. in a short period of time (such as within a few minutes), which are essential for capturing rapidly changing meteorological conditions. At the same time, the corresponding power generation data also records the actual power generation under these instantaneous meteorological changes. Through machine learning or deep learning algorithms, such as long short-term memory networks (LSTM), gated recurrent units (GRU) or convolutional neural networks (CNN), an ultra-short-term prediction model that can learn and predict the relationship between instantaneous meteorological changes and power generation is constructed.

[0053] Finally, according to the preset short-term data update frequency, real-time grid meteorological data is collected and input into the ultra-short-term prediction model. Based on the latest meteorological data, the model quickly calculates the power generation forecast results in the short future time (such as a few minutes to a few hours). These short-term high-precision forecast results are of great significance for the real-time regulation of power load. Using the forecast results, the power grid dispatching system can quickly adjust the power load distribution to match the output changes of the photovoltaic power station. For example, when the forecast shows that the power generation will decrease in the next few minutes, the power gap can be made up by increasing the discharge of energy storage equipment or dispatching other power sources. Conversely, when the power generation is predicted to increase, the output of other peak-shaving power sources can be reduced or the charging load can be increased, thereby achieving dynamic balance of the power grid.

[0054] Furthermore, the embodiment of the present application further includes step P70, and step P70 further includes: P71: Collect long-term historical power generation data, conduct seasonal trend analysis, and identify long-term meteorological characteristics; P72: Based on the long-term meteorological characteristics, train a medium- and long-term prediction model, which can perform periodic corrections to the power generation prediction results based on the seasonal changes in historical data to obtain medium- and long-term trend analysis results; P73: Integrate the short-term high-precision prediction results and the medium- and long-term trend analysis results to generate a power generation prediction curve.

[0055] Optionally, in order to improve the prediction accuracy of the model at different time scales, the embodiment of the present application further combines short-term and medium- to long-term prediction results to ensure that the model can accurately reflect the power generation of the photovoltaic power station when dealing with instantaneous weather changes and long-term seasonal fluctuations.

[0056] First, collect long-term historical power generation data, which covers the power generation records of photovoltaic power stations in multiple seasons and years and contains rich seasonal change information. By performing seasonal trend analysis on these data, the key factors affecting the long-term changes in power generation, namely long-term meteorological characteristics, can be identified. These characteristics may include seasonal changes in solar radiation intensity, temperature fluctuations, precipitation patterns, etc., which have a significant impact on the power generation potential of photovoltaic power stations. The implementation of seasonal trend analysis relies on time series analysis techniques, such as seasonal decomposition (STL), autoregressive integrated moving average model (ARIMA) or seasonal neural network model, which can reveal the periodic laws in the data.

[0057] Next, the medium- and long-term forecasting model is trained based on the identified long-term meteorological characteristics. The model can use time series analysis methods such as SARIMA (Seasonal Autoregressive Integrated Moving Average Model) or LSTM (Long Short-Term Memory Network), which can effectively capture seasonal and cyclical changes in the data. During the forecasting process, the model will periodically correct the forecast results based on the seasonal patterns of historical data to ensure the accuracy and reliability of the forecast results. For example, when entering winter, the model will automatically adjust the forecast results to reflect the impact of shortened sunshine hours and weakened solar radiation, thereby providing a more accurate long-term power generation forecast.

[0058] Finally, the short-term high-precision prediction results and the medium- and long-term trend analysis results are fused. When the time ranges of the short-term and medium- and long-term predictions overlap, the results are fused by weighted average or segmented combination. The weighted average method assigns different weights to the short-term and long-term prediction results according to the reliability of the two predictions within the time range. Generally, the short-term results have a higher weight in a short time (such as 1 hour), while the medium- and long-term results have a higher weight in a longer time range (such as 1 day or weeks). The fused prediction results generate a complete power generation forecast curve, covering the power generation changes from minutes to weeks in the future. This prediction curve not only reflects the immediate impact of real-time weather changes on power generation, but also reflects the trend impact of seasonal meteorological changes on long-term power generation. This prediction curve can be used to guide the daily operation of photovoltaic power plants, the formulation of equipment maintenance plans, and the dispatch optimization of power supply by the power grid system. By combining short-term and medium- and long-term prediction results at different time scales, photovoltaic power plant managers can have a more comprehensive understanding of future power generation changes and respond to the challenges brought by power generation fluctuations in advance.

[0059] In summary, the embodiments of the present application have at least the following technical effects: This application combines grid meteorological data with geographic information of photovoltaic power stations, establishes grid-location association, extracts historical meteorological and power generation data, and classifies weather scenarios. It trains a multi-scenario prediction model to build a power generation prediction network, and uses real-time meteorological data and a dynamic feature selection mechanism to input the prediction network for power generation prediction. The application achieves the technical effect of improving the prediction accuracy and adaptability of photovoltaic power generation under complex meteorological conditions by introducing high-precision grid meteorological data and a specialized scenario-based prediction model.

[0060] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0062] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for predicting power generation of distributed photovoltaic power stations in combination with grid meteorological data, characterized in that: The method comprises: Obtain geographic information data and historical power generation data of the target photovoltaic power station; Collect historical grid meteorological data, perform spatial matching with the geographic information data, and establish grid-location associations; Based on the grid point-position association, the historical grid point meteorological data and the historical power generation data are associated and extracted to obtain multiple groups of historical power generation data; Using the multiple sets of historical power generation data, extracting weather scenarios to obtain multiple historical weather scenarios, and training corresponding power generation prediction models based on the multiple historical weather scenarios to construct a power generation prediction network; Based on the grid-position association, real-time grid meteorological data is collected and input into the power generation prediction network, and a dynamic feature selection mechanism is introduced to perform power generation prediction to obtain a power generation prediction result of a target photovoltaic power station.

2. The method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data according to claim 1, characterized in that: Collect historical grid meteorological data, perform spatial matching with the geographic information data, and establish grid-location associations, including: Based on a preset long-term period, the interactive meteorological platform collects historical grid meteorological data; According to the geographic information data of the target photovoltaic power station, key location information is extracted, and spatial location matching is performed with the historical grid meteorological data, and according to the spatial matching result, the grid point-location association is established based on neighboring grid points or interpolation calculation.

3. The method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data according to claim 1, characterized in that: Based on the grid point-position association, the historical grid point meteorological data and the historical power generation data are associated and extracted to obtain multiple groups of historical power generation data, including: Based on the grid-position association, the historical grid meteorological data and the historical power generation data are aligned in time dimension and unified in data frequency to obtain a standard historical sample data sequence; For the standard historical sample data sequence, traverse each time point in the historical time period, extract related data, and form multiple groups of sample data, wherein each group of sample data contains multiple meteorological elements and corresponding power generation sample data; According to the historical operation and maintenance records of the target photovoltaic power station, the multiple groups of sample data are screened for effective time periods to obtain multiple groups of historical power generation data.

4. The method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data according to claim 3, characterized in that: Using the multiple sets of historical power generation data, extracting weather scenarios to obtain multiple historical weather scenarios, and training corresponding power generation prediction models based on the multiple historical weather scenarios to construct a power generation prediction network, including: Based on historical power generation records, multiple key meteorological elements are extracted, including but not limited to solar radiation intensity, cloud cover, temperature, and sunshine duration; Setting element thresholds for the multiple key meteorological elements respectively, and performing weather clustering on the multiple groups of historical power generation data based on the element thresholds to obtain multiple historical weather scenarios; According to the multiple historical weather scenarios, the multiple groups of historical power generation data are divided into multiple data subsets, and a scenario label is generated for each data subset; The multiple data subsets are used to train power generation prediction models under different weather scenarios, and the trained multiple models are combined to generate a power generation prediction network, which also has a weather scenario classifier embedded in it.

5. The method for predicting power generation of distributed photovoltaic power stations in combination with grid meteorological data according to claim 4, characterized in that: Real-time grid meteorological data is collected and input into the power generation prediction network, and a dynamic feature selection mechanism is introduced to perform power generation prediction, including: According to the preset collection frequency, the interactive meteorological platform collects real-time grid meteorological data and inputs it into the power generation prediction network; Performing scenario classification through the weather scenario classifier in the power generation prediction network to obtain a target weather scenario, and matching to obtain a target power generation prediction model; Introducing a dynamic feature selection mechanism, performing feature importance analysis on the real-time grid meteorological data, and configuring feature importance identifiers; After adjusting the feature weights according to the feature importance identifiers, the target power generation prediction model is used to perform power generation prediction to obtain a power generation prediction result of the target photovoltaic power station.

6. The method for predicting power generation of a distributed photovoltaic power station in combination with grid meteorological data according to claim 5, characterized in that: A dynamic feature selection mechanism is introduced to perform feature importance analysis on the real-time grid meteorological data and configure feature importance identifiers, including: Based on various historical weather scenarios, with meteorological characteristics as input and actual power generation as output, multiple corresponding feature contribution evaluation models are trained respectively; Matching and obtaining a feature contribution evaluation model corresponding to the target weather scenario, performing feature importance analysis on the real-time grid meteorological data, and obtaining a feature importance ranking under the target weather scenario; Based on the feature importance ranking, meteorological features are dynamically selected in combination with importance thresholds, and feature importance identification is performed.

7. The method for predicting power generation of distributed photovoltaic power stations in combination with grid meteorological data according to claim 1, characterized in that: The method further comprises: Based on the demand for high-precision short-term forecasts, the short-term data update frequency is preset; Based on historical power generation records, instantaneous meteorological change data and power generation data are extracted to build an ultra-short-term prediction model; According to the short-term data update frequency, real-time grid meteorological data is collected and input into the ultra-short-term prediction model to quickly predict power generation, and obtain short-term high-precision prediction results for short-term power load regulation.

8. The method for predicting power generation of distributed photovoltaic power stations in combination with grid meteorological data according to claim 7, characterized in that: The method further comprises: Collect long-term historical power generation data, conduct seasonal trend analysis, and identify long-term meteorological characteristics; Based on the long-term meteorological characteristics, a medium- and long-term prediction model is trained, and the medium- and long-term prediction model can perform periodic corrections on power generation prediction results based on seasonal changes in historical data to obtain medium- and long-term trend analysis results; The short-term high-precision prediction results and the medium- and long-term trend analysis results are integrated to generate a power generation prediction curve.

Images