Photovoltaic power generation power prediction method and related equipment

Abstract

The invention provides a photovoltaic power generation power prediction method and related equipment. The method comprises the following steps: acquiring power output information of a photovoltaic power station; acquiring meteorological key data based on the meteorological large model; fusing the meteorological key data and the power output information to obtain a data feature set; the meteorological key data and the power output information in the data feature set have a mapping relationship; training the first model based on the data feature set to obtain a second model; based on real-time push data of the photovoltaic power station, correcting the second model to obtain a third model; performing prediction error analysis on the third model, and adjusting parameters in the third model in response to the situation that the prediction error is greater than a preset threshold value to obtain a fourth model; and predicting the photovoltaic power generation power by using the fourth model. The precision and reliability of photovoltaic power generation power prediction are improved through model optimization and correction, the system adaptability is improved, and stable and efficient operation of power generation power prediction is ensured.

Description

Technical Field

[0001] This application relates to the technical field of photovoltaic power generation, and particularly to a photovoltaic power generation prediction method and related equipment. Background Art

[0002] Current photovoltaic power generation prediction technologies still have significant problems in dealing with complex meteorological conditions. Relying on meteorological data provided by numerical weather prediction, its spatial resolution and temporal accuracy are limited, especially under extreme weather changes, resulting in large errors in prediction data. At the same time, traditional machine learning models are difficult to deeply mine the spatio-temporal correlation of meteorological data, have insufficient ability to capture non-linear features, and cannot effectively respond to sudden meteorological changes (such as cloud cover, severe convective weather), making the prediction results volatile and lacking stability. In addition, centralized computing platforms are difficult to meet the real-time prediction requirements, limited by network load pressure and transmission delay problems, further affecting the response speed and real-time performance of short-term power prediction.

[0003] At the operation level of photovoltaic power plants, existing technologies have not fully incorporated non-meteorological conditions such as equipment operation status and shading factors, which have a significant impact on power output in these actual environments. Traditional models mostly rely on simple feature engineering to model the relationship between meteorology and power, but lack consideration of equipment differential features and non-meteorological factors, and the generalization ability of the models is insufficient, making it difficult to adapt to the individual needs of different power plants. Especially in scenarios with shading or low equipment efficiency, the accuracy of the prediction model drops significantly. In addition, with the increase in the large-scale distribution of photovoltaic power plants, traditional equipment management methods are difficult to match the operation status of rapidly deployed equipment, and combined with the topological identification problem of low-voltage distribution network access, it further increases the complexity of equipment communication and data real-time transmission.

[0004] Generally speaking, the limitations of existing technologies are mainly reflected in the insufficient accuracy and timeliness of meteorological data, the weakening of the model's ability to express complex features, and the difficulties in communication and equipment management. Based on these problems, the stability, real-time response ability, and adaptability to equipment individual characteristics of photovoltaic power prediction systems in complex meteorological conditions urgently need technological breakthroughs and in-depth optimizations to achieve precise support and wide application for new energy power generation dispatching. Summary of the Invention

[0005] In view of this, the purpose of this application is to propose a photovoltaic power generation prediction method and related equipment.

[0006] Based on the above purpose, this application provides a photovoltaic power generation prediction method, including: Obtain the power output information of the photovoltaic power plant; obtain key meteorological data based on the meteorological large model; Fuse the meteorological key data and the power output information to obtain a data feature set; there is a mapping relationship between the meteorological key data and the power output information in the data feature set; Train a first model based on the data feature set to obtain a second model; Based on the real-time push data of the photovoltaic power station, correct the second model to obtain a third model; Conduct a prediction error analysis on the third model. In response to the prediction error being greater than a preset threshold, adjust the parameters in the third model to obtain a fourth model; Use the fourth model to predict the photovoltaic power generation power.

[0007] In a possible implementation manner, the fusing the meteorological key data and the power output information to obtain a data feature set includes: Utilize the sliding window technique to dynamically capture the time correlation relationship between the meteorological key data and the power output information through time series analysis methods; Utilize a data reconstruction algorithm to obtain the spatial correlation relationship between the meteorological key data and the power output information; Utilize a piecewise fitting algorithm to extract the power decline sequence data corresponding to the power change caused by the cloud density parameter in the meteorological key data; Taking the geographical location of the photovoltaic power station as a node, convert the spatiotemporally correlated data in the meteorological key data into a graph structure, and use a graph neural network to adjust the time correlation relationship, the spatial correlation relationship, and the power decline sequence data; Based on the adjusted time correlation relationship, spatial correlation relationship, and power decline sequence data, obtain the data feature set.

[0008] In a possible implementation manner, the training the first model based on the data feature set to obtain a second model includes: Process the data feature set using a cross-validation algorithm, input the processed data feature set into the first model, and use an adaptive learning rate optimization algorithm to dynamically adjust the learning rate of the first model to obtain a trained learning rate; Adjust the hyperparameters of the first model using a grid search algorithm or a random search algorithm to obtain trained hyperparameters; Based on the trained learning rate and the trained hyperparameters, train to obtain the second model.

[0009] In a possible implementation manner, the correcting the second model based on the real-time push data of the photovoltaic power station to obtain a third model includes: Use the second model to predict the photovoltaic power generation power on a short time scale to obtain first prediction data; Use the second model to predict the photovoltaic power generation power on a long time scale to obtain second prediction data; Calculate a first error sequence based on at least two of the first prediction data and the corresponding at least two real-time push data; Calculate a second error sequence based on at least two of the second prediction data and the corresponding at least two real-time push data; Analyze the error sources of each error data in the first error sequence and the second error sequence to obtain a first result; Perform calibration processing on the second model according to the first result to obtain the third model.

[0010] In a possible implementation manner, the performing calibration processing on the second model according to the first result includes: In response to the first result being that at least one of the meteorological key data suddenly changes, dynamically adjust the weight of the corresponding meteorological key data to perform calibration processing on the second model; In response to the first result being that the device state of the photovoltaic power station is abnormal, dynamically modify or add parameters corresponding to the abnormal device to perform calibration processing on the second model.

[0011] In a possible implementation manner, for the prediction error analysis of the third model, in response to the prediction error being greater than a preset threshold, adjust the parameters in the third model to obtain a fourth model, including: Use the third model to predict the photovoltaic power generation power on a short time scale to obtain third prediction data; Use the third model to predict the photovoltaic power generation power on a long time scale to obtain fourth prediction data; Calculate a third error sequence based on at least two of the third prediction data and the corresponding at least two historical push data; Calculate a fourth error sequence based on at least two of the fourth prediction data and the corresponding at least two historical push data; Analyze the error sources of each error data in the third error sequence and the fourth error sequence to obtain a second result; Adjust the parameters in the third model based on the second result to obtain the fourth model.

[0012] Based on the same inventive concept, an embodiment of the present application further provides a photovoltaic power generation power prediction device, including: An acquisition module, configured to acquire power output information of a photovoltaic power station; and acquire key meteorological data based on a meteorological large model; A fusion module, configured to perform fusion processing on the key meteorological data and the power output information to obtain a data feature set; there is a mapping relationship between the key meteorological data and the power output information in the data feature set; A training module, configured to train a first model based on the data feature set to obtain a second model; A calibration module, configured to perform calibration processing on the second model based on real-time push data of the photovoltaic power station to obtain a third model; An adjustment module, configured to perform prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjust parameters in the third model to obtain a fourth model; A prediction module, configured to predict photovoltaic power generation power using the fourth model.

[0013] Based on the same inventive concept, an embodiment of the present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the program, it implements the photovoltaic power generation power prediction method as described in any one of the above.

[0014] Based on the same inventive concept, an embodiment of the present application further provides a non-transitory computer-readable storage medium, where the non-transitory computer-readable storage medium stores computer instructions for causing a computer to execute the photovoltaic power generation power prediction method as described in any one of the above.

[0015] Based on the same inventive concept, an embodiment of the present application further provides a computer program product, which includes computer program instructions for causing the computer program product to execute the photovoltaic power generation power prediction method as described in any one of the above.

[0016] As can be seen from the above, the photovoltaic power prediction method and related devices provided in this application obtain the power output information of a photovoltaic power station; obtain key meteorological data based on a meteorological large model; perform fusion processing on the key meteorological data and the power output information to obtain a data feature set; there is a mapping relationship between the key meteorological data and the power output information in the data feature set; train a first model based on the data feature set to obtain a second model; perform calibration processing on the second model based on the real-time push data of the photovoltaic power station to obtain a third model; perform prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjust the parameters in the third model to obtain a fourth model; use the fourth model to predict the photovoltaic power generation power. The embodiments of this application demonstrate superiority in multiple aspects such as improving prediction accuracy, enhancing adaptability, and optimizing real-time performance. It collects power output information from a photovoltaic power station and obtains key meteorological data based on a meteorological large model, and generates a high-quality feature set after fusing the two, fully exploring the spatio-temporal correlation between power output and meteorological features. Through layer-by-layer optimization of model training, a dynamic adjustment mechanism from the first model to the fourth model is formed, realizing systematic optimization from data preparation to prediction analysis, providing strong support for solving the problems of large errors, poor adaptability, and difficulty in accurate prediction under complex weather conditions in traditional photovoltaic power prediction. It can also significantly improve the real-time performance and accuracy of photovoltaic power prediction. In short-term prediction, a sliding time window is used to dynamically capture the time correlation between meteorology and power, enabling the prediction to have a second-level dynamic response ability. In medium- and long-term prediction, by fusing key influencing factors such as cloud density, the failure of the prediction model under changing meteorological conditions is avoided, providing a reliable basis for power grid dispatching and power generation planning. At the same time, through error analysis and parameter adjustment, the models in each stage can adapt to the influence of complex weather conditions and abnormal equipment states. For example, dynamic calibration of parameters such as dust index, cleanliness, and equipment operation efficiency further reduces the error of power prediction. Further through the design of dynamic optimization, this application also enhances the adaptability and robustness of photovoltaic power prediction. In error analysis, classification processing is performed for different sources, and the model weights and parameters are adjusted as needed, gradually optimizing from the initial data feature set to the final prediction result, greatly improving the performance ability of the model under extreme climate conditions. In summary, this application effectively improves the prediction accuracy, stability, and applicability through scientific data fusion, model optimization, and multi-layer calibration mechanisms, providing technical support for the efficient operation of photovoltaic power generation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the technical solutions in the present application or related technologies, the following will briefly introduce the accompanying drawings required for use in the embodiments or related technology descriptions. Obviously, the accompanying drawings in the following descriptions are only embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0018] Figure 1 Schematic flow diagram of the photovoltaic power prediction method according to an embodiment of the present application; Figure 2 Schematic structural diagram of the photovoltaic power prediction device according to an embodiment of the present application; Figure 3 Schematic structural diagram of the electronic device according to an embodiment of the present application. Detailed implementation manners

[0019] To make the objectives, technical solutions, and advantages of the present application clearer and more understandable, the following further details the present application in combination with specific embodiments and with reference to the accompanying drawings.

[0020] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should have the ordinary meanings understood by those with ordinary skills in the field to which the present application belongs. The "first", "second", and similar terms used in the embodiments of the present application do not indicate any order, quantity, or importance, but are only used to distinguish different components. "Including" or "comprising" and similar terms mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. "Connecting" or "being connected" and similar terms are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0021] It can be understood that before using the technical solutions of the various embodiments in the present disclosure, the types, usage scopes, usage scenarios, etc. of the personal information involved will be informed to the user in an appropriate manner, and the user's authorization will be obtained.

[0022] For example, when responding to receiving an active request from the user, a prompt message is sent to the user to clearly prompt the user that the operation requested by the user will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to the electronic device, application program, server, storage medium, or other software or hardware that performs the operations of the technical solutions of the present disclosure according to the prompt message.

[0023] As an optional but non-limiting implementation, in response to receiving an active request from the user, the way to send a prompt message to the user can be, for example, in the form of a pop-up window, and the prompt message can be presented in text in the pop-up window. In addition, the pop-up window can also carry selection controls for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0024] It can be understood that the above notification and the process of obtaining user authorization are only illustrative and do not limit the implementation of the present disclosure. Other methods that meet relevant laws and regulations can also be applied to the implementation of the present disclosure.

[0025] As described in the background art section, there are still significant problems in the current photovoltaic power prediction technology when dealing with complex meteorological conditions. Relying on the meteorological data provided by numerical weather prediction, its spatial resolution and temporal accuracy are limited, especially under extreme weather changes, resulting in large errors in the prediction data. At the same time, traditional machine learning models are difficult to deeply mine the spatio-temporal correlation of meteorological data, have insufficient ability to capture non-linear features, and cannot effectively respond to sudden meteorological changes (such as cloud cover, severe convective weather), making the prediction results fluctuate greatly and lack stability. In addition, the centralized computing platform is difficult to meet the real-time prediction requirements, limited by network load pressure and transmission delay problems, further affecting the response speed and real-time performance of short-term power prediction.

[0026] At the operation level of the photovoltaic power station, the existing technology has not fully incorporated non-meteorological conditions such as equipment operation status and shading factors, and these actual environments have a significant impact on power output. Traditional models mostly rely on simple feature engineering to model the relationship between meteorology and power, but lack consideration of equipment differential features and non-meteorological factors, and the generalization ability of the models is insufficient, making it difficult to adapt to the individual needs of different power stations. Especially in scenarios with shading or low equipment efficiency, the accuracy of the prediction model drops significantly. In addition, with the increase in the large-scale distribution of photovoltaic power stations, traditional equipment management methods are difficult to match the operation status of rapidly deployed equipment, and combined with the problem of topological identification of low-voltage distribution network access, it further increases the complexity of equipment communication and data real-time transmission.

[0027] Generally speaking, the limitations of the existing technology are mainly reflected in the insufficient accuracy and timeliness of meteorological data, the weakening of the model's ability to express complex features, and the difficulties in communication and equipment management. Based on these problems, the stability, real-time response ability, and adaptability to equipment individual characteristics of the photovoltaic power prediction system under complex meteorological conditions urgently need technological breakthroughs and in-depth optimizations to achieve precise support and wide application for new energy power generation scheduling.

[0028] In view of the above considerations, an embodiment of the present application proposes a photovoltaic power prediction method and related devices, which obtain the power output information of a photovoltaic power station; obtain key meteorological data based on a meteorological large model; perform fusion processing on the key meteorological data and the power output information to obtain a data feature set; there is a mapping relationship between the key meteorological data and the power output information in the data feature set; train a first model based on the data feature set to obtain a second model; perform calibration processing on the second model based on the real-time push data of the photovoltaic power station to obtain a third model; perform prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjust the parameters in the third model to obtain a fourth model; use the fourth model to predict the photovoltaic power. The embodiment of the present application demonstrates superiority in multiple aspects such as improving prediction accuracy, enhancing adaptability, and optimizing real-time performance. Through the synergistic effect of multiple claims, the embodiment of the present application breaks through the deficiencies of traditional prediction methods in terms of real-time performance, accuracy, and adaptability to complex environments, providing more efficient, accurate, and stable technical support for photovoltaic power prediction. First, a systematic optimization process is formed from data collection to model prediction, changing the current situation that traditional prediction methods rely heavily on data quality. By introducing a meteorological large model to obtain key meteorological data, combining with the real-time power output information of the photovoltaic power station, and performing in-depth fusion processing on the two, a high-quality data feature set is generated, mining the change laws of photovoltaic power from multiple dimensions such as time correlation, space correlation, and the impact of complex weather, effectively solving the problems of single data input and insufficient accuracy. At the same time, methods such as sliding window technology, piecewise fitting algorithm, and graph neural network are used to process the data feature set, improving the spatio-temporal feature expression ability and solving the problem of insufficient extraction of correlation information between data in complex environments. Secondly, through a series of optimization algorithms, the model is trained and dynamically adjusted layer by layer, enabling the present application to have excellent performance in both short-term and medium- to long-term power prediction. During the training process, cross-validation, adaptive learning rate optimization algorithm, and hyperparameter search technology are used to dynamically adjust the first model to the second model, enabling the model to quickly adapt to the power change characteristics of different scenarios. Through the real-time calibration of the third model, the model parameters are analyzed based on two sequences of short-term prediction error and long-term prediction error, and dynamic weight adjustment and device parameter adaptation are performed in combination with real-time push data, providing an accurate calibration basis for generating the fourth model. This multi-stage calibration mechanism effectively solves the problems of insufficient prediction accuracy of the model under sudden meteorological conditions (such as sand and dust, cloud density changes) and the difficulty in compensating for the impact of device state changes. In addition, by continuously optimizing the adaptability and robustness of the prediction model, the present application can accurately predict the power output in complex weather and multi-region device scenarios. In practical applications, by dynamically optimizing the weights of device-related factors such as sand and dust index and cleanliness, the model is enabled to handle complex environments, reducing the error range caused by environmental changes or device abnormalities.The dynamic error correction mechanism enables the model to have high stability in long-term operation, especially in high wind speed and frequent weather conditions, where the error is significantly reduced, thus enhancing the reliability and adaptability of the prediction results.

[0029] In summary, this application can comprehensively improve the accuracy, real-time performance and environmental adaptability of photovoltaic power prediction, provide important technical support for grid dispatching, energy storage optimization and intelligent management of photovoltaic stations, and effectively promote the application value of photovoltaic power generation systems in smart grids.

[0030] The technical solutions of the embodiments of the present application are described in detail below through specific examples.

[0031] refer to Figure 1 The photovoltaic power generation power prediction method of the embodiment of the present application comprises the following steps: Step S101, obtaining power output information of a photovoltaic power station; obtaining key meteorological data based on a large meteorological model; Step S102, fusing the key meteorological data and the power output information to obtain a data feature set; the key meteorological data and the power output information in the data feature set have a mapping relationship; Step S103, training the first model based on the data feature set to obtain a second model; Step S104, based on the real-time push data of the photovoltaic power station, calibrate the second model to obtain a third model; Step S105, performing prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjusting parameters in the third model to obtain a fourth model; Step S106: predicting photovoltaic power generation power using the fourth model.

[0032] With respect to step S101, this application uses a large meteorological model and edge computing technology to push key meteorological data that affect photovoltaic power generation in real time, including solar irradiance, temperature, humidity, wind speed, etc. At the same time, power output information is extracted from the historical operation data of the photovoltaic power station, and the mathematical morphology denoising method is used to clean the data, the image denoising method is used to remove outliers, and the nonlinear normalization method is used to convert the data scale. The three methods are used in combination to improve the availability and accuracy of historical data.

[0033] In this embodiment, the photovoltaic power station is located in the northwest plateau area of ​​my country, and the environmental characteristics are as follows: Meteorological conditions: large temperature difference between day and night, large fluctuations in wind speed, frequent impacts of sandstorms and sudden cloud cover, and dramatic changes in irradiance.

[0034] Equipment features: The photovoltaic power generation components and sensors are exposed to a dusty environment. During long-term operation, significant changes in the equipment cleanliness will affect the power performance.

[0035] To ensure the high quality and reliability of the input data for subsequent models, the initial input data needs to be comprehensively collected and strictly preprocessed.

[0036] Specifically as follows: This photovoltaic power station needs to collect two types of core data: meteorological data and photovoltaic equipment operation data.

[0037] Meteorological data collection: This photovoltaic power station has installed edge meteorological monitoring equipment (such as solar irradiance sensors, anemometers), and data is collected in real time through Internet of Things devices.

[0038] Access the national meteorological large model application programming interface (Application Programming Interface, API) to obtain regional weather data, ensuring comprehensive coverage of the data in terms of time and space.

[0039] The collected meteorological characteristics are as follows: Solar irradiance: Updated every minute, directly from on-site sensors, and at the same time, the regional average irradiance is compared through the large model.

[0040] Temperature: Collect data on the near-surface temperature change (real-time change frequency is 1 minute).

[0041] Humidity: Used to analyze the irradiance penetration rate and the state of air clarity.

[0042] Wind speed and direction: Obtain the wind speed and direction every minute from the anemometer, which is used to evaluate the influence of sand and dust on the shading and heat dissipation of photovoltaic equipment.

[0043] Cloud density and dust index: The cloud density is provided by combining satellite observation images and regional meteorological models; the dust index is generated by combining local detection stations and the national real-time dust warning platform.

[0044] Furthermore, edge computing technology is used to compress and optimize the data preliminarily processed by meteorological sensors and send it to the central data server.

[0045] Transmission frequency: Updated every 1 minute, and the meteorological data is saved to the database in real time to support short-term prediction requirements.

[0046] Collection of photovoltaic power station operation data: Historical power data: The power output curve data for the past 1 year is extracted from the power station inverters and the power station management system at a minute granularity.

[0047] The electrical indicators such as power, current, and voltage recorded every minute are stored in the historical operation database of the power station.

[0048] Device status data: Photovoltaic module cleaning status: Use a cleanliness monitoring device to record the pollution and dust accumulation of each photovoltaic panel.

[0049] Photovoltaic module operating efficiency: Real-time collect the operating temperature rise data and efficiency parameters of the module to evaluate the device status.

[0050] The purpose of collection is to ensure that real-time power data and historical performance data can be fully combined with meteorological data and input into the model, providing practical and effective data support for predicting feature extraction.

[0051] After the data collection is completed, the collected raw data may have problems such as noise, missing values, and inconsistencies. Therefore, it is necessary to perform comprehensive preprocessing on it to ensure data quality.

[0052] An abnormal spike was found in the wind speed data during a certain period. The record shows that the normal wind speed is 57 m / s, and the wind speed jumped abnormally to 25 m / s within a certain minute and then returned to normal. This may be noise caused by a sensor failure. Subsequently, mathematical morphological filtering technology was used to denoise the data, removing such extreme spike values and smoothing the wind speed curve at the same time.

[0053] In satellite image data, there are random noise points (such as aggregated dot-shaped shadows) in the cloud-covered images, which are likely to incorporate interference information. Use the denoising algorithm of Convolutional Neural Network (CNN) to process the images, screening and removing the noise points that do not belong to natural clouds.

[0054] If the photovoltaic power data is missing the record for 10 minutes due to a sensor device failure on a certain day, resulting in the inability to match the meteorological data for this period. The interpolation method is used to complete the missing power values, and the data continuity is ensured by correcting through the context data trend.

[0055] The standardization processing of the data can normalize the irradiance data (sampling range from 200 W / m² to 1300 W / m²) to the interval of 0 - 1, ensuring the unity of the input data magnitude and avoiding the large-value features from dominating the model parameter training. For features such as temperature and humidity, the standardization method (Z-score standardization) is used to convert them into a standard distribution.

[0056] After that, align the power data and meteorological data at a minute-level time granularity to ensure that the data moments of the model input after alignment are consistent.

[0057] Environmental and regional characteristics may cause inconsistencies in data scales (such as time intervals and geographical deviations). This application ensures the unity of data from different sources through the processing of time and space scales.

[0058] For example, some historical power record data is once per hour, while real-time meteorological data is once per minute: The interpolation method is used to interpolate the hourly power data to the minute level, and at the same time, the weighted smoothing method is used to improve the fine-grained adaptability of short-term prediction.

[0059] The spatial scale also needs to be normalized: Since the geographical location of the desert plateau photovoltaic power station is different from that of the plain area and is affected by the environmental albedo and the sunlight incidence angle: According to the altitude and latitude of the power station, the irradiance data value is corrected in combination with the ground measurement results. For areas with strong cloud cover, the regional meteorological model is used for distribution correction.

[0060] The cleaned data is stored in a distributed database, which is divided into the following sub-modules: Historical data table: Stores past power and meteorological data. Real-time data table: Updates and stores real-time transmission data every minute. And it is necessary to regularly check the cleaned data. For example, compare the solar irradiance data with the monitoring values of the national meteorological center to ensure the accuracy of local sensor data.

[0061] This application significantly improves the spatio-temporal resolution of meteorological data by introducing a large meteorological model, and ensures the timeliness of data through real-time high-frequency updates (data collection at the minute level). Combining edge computing technology, through near computing and real-time data push, it significantly shortens the delay in the data transmission process and improves the data response ability. It increases the comprehensive collection range of key meteorological parameters affecting photovoltaic power (such as wind speed, wind direction, cloud density, etc.), and real-time access to monitoring devices further makes up for the blind spots of numerical weather prediction in the actual complex geographical environment. Comprehensive use of mathematical morphology denoising and image denoising techniques to eliminate spike values, jitter and abnormal deviations in the collected data. The non-linear data normalization method is added to standardize the data collected from different geographical regions or sensing devices to ensure the unity and comparability of the input data between photovoltaic power stations. For the real-time power data of photovoltaic power stations, through efficient noise elimination and data calibration methods, such as smoothing data fluctuations and eliminating extreme data points, etc., to ensure the accuracy and consistency of the input data. Image denoising and cloud image analysis techniques are specifically designed to clearly capture the influence sequence of cloud cover on power, making up for the problem of data distortion under complex weather conditions. By dynamically comparing the predicted data of the meteorological model with the actual sensing data, this technology can automatically adjust the data quality and feedback data anomalies under complex weather conditions, thereby enhancing the adaptability of the model to complex climate conditions.

[0062] Regarding step S102, in some embodiments, the fusion processing of the meteorological key data and the power output information to obtain a data feature set includes: using the sliding window technique to dynamically capture the temporal correlation relationship between the meteorological key data and the power output information through time series analysis; using a data reconstruction algorithm to obtain the spatial correlation relationship between the meteorological key data and the power output information; using a piecewise fitting algorithm to extract the power decline sequence data corresponding to the power change caused by the cloud density parameter in the meteorological key data; taking the geographical location of the photovoltaic power station as a node, converting the data with spatio-temporal correlation in the meteorological key data into a graph structure, and using a graph neural network to adjust the temporal correlation relationship, the spatial correlation relationship, and the power decline sequence data; based on the adjusted temporal correlation relationship, spatial correlation relationship, and power decline sequence data, obtaining the data feature set.

[0063] In the above steps, the collection and preprocessing of the meteorological data and photovoltaic operation data of the desert plateau photovoltaic power station have been completed, and accurate basic data input has been obtained. The goal of this step is to deeply integrate real-time meteorological data with historical power generation data, mine the non-linear spatio-temporal correlation between meteorological factors and power output, and provide high-quality input for subsequent model training.

[0064] In this embodiment, the meteorological large model is fused with photovoltaic power generation data. Based on the collected meteorological data, the meteorological large model is used to make short-term and medium- to long-term predictions of future meteorological conditions, and combined with the historical power generation data of the photovoltaic power station, the mapping relationship between meteorological factors and photovoltaic power generation is analyzed. A power data reconstruction algorithm for photovoltaic power stations considering wind direction and spatio-temporal correlation is proposed. The relationship between wind speed and the delay time of the power station is established, the power decline sequence caused by cloud occlusion in the photovoltaic curve is extracted, and a fitting model is built using a graph neural network to effectively reconstruct the power curve of the photovoltaic power station and improve the accuracy of the data in the prediction input feature set.

[0065] This photovoltaic power station has completed the collection and preprocessing of the following data: Meteorological data: complex meteorological characteristics such as solar irradiance (updated every 1 minute), temperature, humidity, wind speed, wind direction, cloud density, and dust index.

[0066] Photovoltaic power generation data: historical power data (recorded at 1-minute granularity in the past year), real-time power output, and the status of photovoltaic power station equipment (such as cleanliness and operating efficiency).

[0067] These data have all been normalized according to time and space scales and stored in a distributed database.

[0068] By analyzing the meteorological data and power data, explore the influence law of weather changes on power output, and establish the global mapping relationship of time and space.

[0069] Use historical data to analyze the time series relationship between meteorology and power changes.

[0070] Use the sliding time window technique: Design a sliding window centered on the current moment data with a time span of 3 hours to dynamically capture the correlation between historical meteorological data and power output.

[0071] In this embodiment, it is found through analysis that solar irradiance has a 30-minute lag characteristic on power output, and the influence of temperature changes on power efficiency is more significant on the hourly scale. Therefore, set data extraction rules in the sliding window: Irradiance characteristics: Use data from the past 1 hour and focus on capturing changes in the past 30 minutes.

[0072] Temperature and humidity: Set the window length to the past 3 hours and extract the gradually changing trend characteristics.

[0073] Furthermore, dynamically construct a time series feature matrix containing multi-period characteristics for subsequent model training.

[0074] For the special environment of desert photovoltaic power plants, analyze the spatio-temporal impact of geographical features on the operation of photovoltaic power plants. Measured data shows that high wind speed periods will form a self-cleaning effect on the surface of photovoltaic panels, thus improving equipment efficiency; while when low wind speed and sand and dust weather occur together, it will significantly affect power output.

[0075] In this embodiment, it is found that at 2 pm on a certain day, the wind speed drops to 2 m / s, and at the same time the sand and dust index rises, resulting in the power output being significantly lower than the expected value. The model extracts the "wind speed-sand and dust coupling effect" feature for describing the occlusion state of sand and dust on power.

[0076] In some embodiments, the power data is missing from 10:00 to 10:10 am on a certain day, which affects the analysis of lag features. Complement the missing power sequence through a fitting algorithm (such as Spline interpolation method): Piecewise fit the corresponding relationship between irradiance and power within the past 10 minutes and reconstruct the power change curve. The complemented sequence effectively fills the data gap and provides connectivity for model input.

[0077] Furthermore, construct a regional feature reconstruction model based on the Graph Neural Network (GNN) to further improve data quality through high-dimensional feature coupling.

[0078] Each node is a photovoltaic power plant collection point (divided according to geographical location). Node features include: time series meteorological features (irradiance, temperature, humidity, cloud density, etc.). Power history curve features (10-minute average value and change rate).

[0079] The edges between nodes are established based on geographical adjacency. The edge weights between two nodes are calculated through the correlation characteristics of meteorological data. For example, for the characteristic of highly similar cloud layers: edge weight > 0.8. When a high-wind area is connected to a low-wind area: the edge weight decreases.

[0080] The constructed graph neural network is used to complete the photovoltaic power curve and reconstruct its features. Specifically, on a certain day, the power decreases due to sand and dust occlusion. By extracting the irradiance-power relationship features between nodes, the occlusion influence characteristic curve is drawn. Finally, the optimized reconstructed power curve can be closer to the actual fluctuations.

[0081] After the above processing, an optimized data input feature set is generated: Meteorological features: dynamically changing solar irradiance, cloud density, temperature sequence (including sliding window features).

[0082] Power features: historical power data, completed power decline sequence.

[0083] Spatio-temporal features: feature reconstruction data based on the graph neural network.

[0084] The above high-quality input features will be directly used in the next model training stage.

[0085] In the embodiments of the present application, by introducing a sliding time window and a time series feature analysis method, the dynamic response ability of the model to complex meteorological changes is improved by dynamically extracting the time correlation of meteorological features (such as the time period when solar irradiance lags and affects power changes). Through spatial feature analysis, a reconstruction algorithm based on wind speed, wind direction, cloud density, etc. is designed to deeply explore regional specific problems (such as the relationship between wind speed and irradiance scattering in mountainous areas), and solve the limitation of the traditional model's insufficient processing ability for special meteorological conditions. Using a graph neural network to establish the spatio-temporal correlation relationship of meteorological data, and modeling the regional dynamics of multiple photovoltaic power stations as a graph structure, not only can analyze the connections between nodes, but also can handle the non-linear feature changes in specific spatial situations. A non-linear fitting algorithm for power data is constructed, and by reconstructing the power curve of the photovoltaic power station in segments, the power drop sequence caused by cloud occlusion is completed, increasing the robustness of the data. Using a graph neural network to model the complex connections between meteorological elements, such as the combined influence of wind direction and cloud density on power changes, is incorporated into the network architecture, thus realizing non-linear and in-depth mining of meteorological factors. The meteorological characteristics of different regions are calibrated according to the geographical distribution differences to eliminate the interference of heterogeneous meteorological data on the prediction results and optimize the data input accuracy. Using a graph neural network to transform meteorological data into a high-dimensional graph structure, and associating the dynamic changes of wind speed, irradiance and cloud occlusion sequence through the relationship between nodes and edges, upgrading the traditional "single-index feature extraction" to "spatio-temporal comprehensive feature mining", and improving the quality of the power prediction feature set. The data completion and correction technology is enhanced to handle the missing points in the collected data, and the power drop sequence is used as part of the simulation input to solve the problem of incomplete data under complex meteorological conditions. The power drop data caused by cloud occlusion is extracted as a segmented fitting sequence, and by designing a regional dynamic data completion algorithm, the power output curve is effectively reconstructed, making the prediction results more stable under complex conditions. In the meteorological factor analysis, the dynamic reconstruction of the roles of wind speed and wind direction is increased, providing more complete input data for power features when predicting sudden weather changes, thereby improving the response ability of the model to predictions under unstable meteorological conditions.

[0086] Regarding step S103, training the first model based on the data feature set to obtain a second model includes: processing the data feature set using a cross-validation algorithm, inputting the processed data feature set into the first model, dynamically adjusting the learning rate of the first model using an adaptive learning rate optimization algorithm to obtain a trained learning rate; adjusting the hyperparameters of the first model using a grid search algorithm or a random search algorithm to obtain trained hyperparameters; and training to obtain the second model based on the trained learning rate and the trained hyperparameters.

[0087] In this embodiment, deep learning methods are first used to model meteorological data and photovoltaic power generation historical data to build a photovoltaic power generation prediction model. Secondly, time-series neural networks such as long short-term memory networks or gated recurrent units are used to model and test, and the time-series correlation characteristics between meteorological variables and power generation are explored. The network structure parameters are adjusted synchronously during the training process to optimize the model performance. Cross-validation and grid search methods are used again to systematically optimize the hyperparameters to improve the prediction accuracy and robustness.

[0088] In the above steps, a high-quality feature dataset is generated. This data is ready for deep learning model training, with the goal of building a high-precision photovoltaic power prediction model.

[0089] In this embodiment, the optimized meteorological feature set is input, including irradiance, humidity, temperature, etc. per minute, and time series data combined with sliding windows. The historical power curve is used as an auxiliary input to capture the power generation law.

[0090] The LSTM network is divided into two layers: The first layer contains 128 hidden units, which are used to extract the relationship between short-term weather characteristics and power. The second layer contains 64 hidden units, which are used to model the relationship between power and long-term meteorological changes.

[0091] Data segmentation rules: From the optimized feature set generated in the second step, it is divided into the following according to the time series: Training set: 70%, used for model learning parameters. Includes power data and corresponding meteorological characteristics in the past year.

[0092] Validation set: 20%, used for model verification. It includes data from the past two months and is used to evaluate whether the model can effectively avoid overfitting.

[0093] Test set: 10%, used for performance evaluation. Includes data from the most recent week for the final test of predictive ability.

[0094] A 5-fold cross-validation scheme is used: the training set is divided into five groups, four of which are used for training each time, and the remaining one is used for validation to ensure that the model is universal for different data subsets.

[0095] Each training inputs a feature sequence across a time window. For example, input the feature matrix of the meteorological time series for the past 3 hours (12 time steps, 15 minutes between each step). Input the power change sequence as an auxiliary feature. Further align the power and meteorological features according to the time period, for example, pair the 30-minute lag effect of the irradiance feature with the power change. Through the model training process, the model gradually captures the complex nonlinear relationship between the time series features and the power output.

[0096] Use the Adaptive Learning Rate (Adam) optimizer during the optimization process: Adaptive learning rate adjustment accelerates convergence and improves the model training efficiency. Adopt the learning rate decay and cosine annealing strategies: The learning rate is 0.001 in the initial training stage; maintain a relatively high learning rate before the 20th training epoch to achieve fast convergence. Gradually reduce the learning rate after the 20th epoch to avoid overfitting in the later stage of training.

[0097] Set the main hyperparameters of the model as the search range and test the optimal combination one by one: The number of hidden layer units: 64, 128, 256. The time window length: 2 hours, 3 hours, 4 hours. The batch size: 16, 32, 64. Through grid search, it is found that the best configuration of the model is: 128 hidden units in the first layer and 64 hidden units in the second layer. The time window length is set to 3 hours and the batch size is 32. Supplementary optimization by random search: Randomly select hyperparameter combinations to further verify the results of grid search.

[0098] For step S104, in some embodiments, the correcting the second model based on the real-time push data of the photovoltaic power station to obtain a third model includes: using the second model to predict the photovoltaic power generation power on a short time scale to obtain first prediction data; using the second model to predict the photovoltaic power generation power on a long time scale to obtain second prediction data; calculating a first error sequence based on at least two of the first prediction data and the corresponding at least two real-time push data; calculating a second error sequence based on at least two of the second prediction data and the corresponding at least two real-time push data; analyzing the error sources of each error data in the first error sequence and the second error sequence to obtain a first result; correcting the second model according to the first result to obtain the third model.

[0099] In some embodiments, the correcting the second model according to the first result includes: in response to the first result being that at least one of the meteorological key data suddenly changes, dynamically adjusting the weight of the corresponding meteorological key data to correct the second model; in response to the first result being that the device state of the photovoltaic power station is abnormal, dynamically modifying or adding parameters corresponding to the abnormal device to correct the second model.

[0100] In this embodiment, design a daily rolling update meteorological forecast data classification algorithm, screen and complement the input model, calculate the photovoltaic power generation power at different time scales (15 minutes - 10 days in the future), then analyze the prediction error based on the actual power data, and adopt an adaptive optimization algorithm for real-time error correction to improve the prediction accuracy.

[0101] Specifically, for short - term power prediction, the meteorological feature sequence and historical power data of the past 3 hours are input, and the real - time changes in the current time period are captured through a sliding window. The features include: the sequence of solar irradiance changes (updated every 1 minute), the wind speed and temperature sequences, and the historical power change curve of the most recent 3 hours. The second model is used for short - term prediction (15 minutes to 2 hours). In some feasible embodiments, under the condition of enhanced sand - dust weather, the second model predicts that the power in the next 15 minutes is 16.8 MW and the power in the next 2 hours is 17.5 MW.

[0102] For long - term power prediction, the power curve and meteorological prediction data of the most recent 2 days are input, and the change trends of the sand - dust index and cloud density are combined. The input features cover the meteorological large - model forecast results for the next 10 days, including: the daily change curve of solar irradiance, and the dynamic trends of temperature and humidity. The second model is used to predict the hourly power changes in the next 3 days. The second model predicts that the power in the next 3 days is as follows: the morning power on the first day is 12 MW, the noon power is 18 MW, and the total power for the whole day is 180 MW. The power on the second day gradually recovers due to the weakening of sand - dust, and the total power is 200 MW.

[0103] Receive real - time power data from the photovoltaic power station for comparing with short - term prediction values. Calculate the error sequence between the short - term prediction data and the real - time power according to time periods. For example, the predicted power on the first day is 16.8 MW, and the actually measured power is 16.6 MW, with an error of 1.2%.

[0104] Compare the predicted power results for the next 3 days with the actual power generation to generate a long - term error sequence. Generate an error sequence by comparing the hourly predicted values with the actual values. Calculate the total power difference by day to evaluate the long - term prediction performance of the model. The second model predicts that the power on the second day is 200 MW, and the actually measured power generation is 195 MW, with an error rate of 2.56%.

[0105] Analyze the impact of sudden changes in the sand - dust index on power prediction. The sand - dust index rapidly rises to 90 in the morning of the day, but only the index of 70 is considered in the input features, resulting in an expansion of the power prediction deviation.

[0106] Check whether the difference between the cleanliness data of the photovoltaic modules and the actual state causes deviation in power prediction. The cleanliness monitoring shows that the surface pollution degree of the modules reaches 40%, and the model does not fully compensate for the pollution effect, resulting in a higher power.

[0107] Analyze the prediction deviation caused by regional weather differences. The actual shaded area of the northern photovoltaic panels is insufficiently calculated, while the model defaults to a higher shading rate.

[0108] Dynamically optimize the dust index weight according to the meteorological error results. Adjust the dust index weight from 0.7 to 0.9 to more accurately reflect the occlusion effect. Compensate for the power characteristics and add the cleanliness factor as a new input feature. If the error results significantly exceed the preset threshold (short-term error > 3%, long-term error > 5%), trigger the retraining of the model. For the error sequence in the all-day power prediction, add the high-dust environment data of the past 2 days as training samples and update the model structure.

[0109] For the mutation of key meteorological data, analyze the prediction deviation generated by the real-time analysis of key meteorological data, such as a sudden drop in solar irradiance, a significant increase in the dust index, a mutation in cloud density, etc. At 10:00 am one day, the dust index suddenly increased from 60 to 95 within a short period (15 minutes), resulting in the power prediction error expanding from 3% to 6%.

[0110] According to the characteristic correlation corresponding to the sudden change of key meteorological data, increase or decrease the weights of relevant features: increase the dust index weight, adjust it from the initial value of 0.7 to 0.9 (enhance the influence of the occlusion effect). Decrease the temperature weight, as the influence of temperature change on power output decreases when the dust concentration exceeds 90. After correcting the dust index weight, the model readjusts the prediction of the occlusion range, and the power prediction error decreases from 6% to 3.5% after adjustment.

[0111] For abnormal equipment status, analyze the real-time status of photovoltaic modules and compare it with historical data and records of the monitoring system. Common abnormalities include: an increase in component contamination (dust accumulation). Equipment maintenance status failure (decrease in the connection efficiency of photovoltaic panels). The photovoltaic panel cleaning monitoring shows that the surface coverage rate of the component reaches 40% (dust accumulation), the power output is low, and this change in cleanliness is not fully compensated in the model.

[0112] Add or dynamically adjust equipment-related input parameters in the second model, such as: component cleanliness index. Temperature rise efficiency compensation factor. Dynamically set the initial value of the cleanliness index to 0.8, and at the same time introduce a maintenance status compensation parameter to simulate the power reduction amplitude caused by dust accumulation.

[0113] Modify the abnormal parameters of the equipment: According to the real-time detection data record, adjust the old parameter values or add new features: assign a reduced weight to the equipment with low cleanliness (reflecting low efficiency). Enhance the power prediction correlation for well-maintained equipment. After setting additional parameters for a large number of dust-accumulated components in the second model, the power reduction trend is accurately corrected, and the prediction error is optimized from 5% to 3%.

[0114] In the embodiments of the present application, the prediction errors are classified (such as meteorological data errors, device sensing errors, etc.), and targeted correction strategies are designed in combination with the sources of deviation. For example, for the power fluctuations caused by complex meteorological conditions, the model feature weights are adjusted in real time (such as increasing the dynamic influence weight of irradiance changes). A lightweight error correction model is developed, and combined with the sliding window technology, the errors between the predicted values and the true values are dynamically compensated and adjusted in real time. For sudden cloud occlusion or sudden wind speed changes, instant correction based on segmented optimization is adopted to improve the adaptability of the model when there are drastic power fluctuations. For the situation where the power drops rapidly due to sudden cloud occlusion, by extracting the historical patterns of irradiance decline in real time, the future power trend is compensated and corrected. Multiple correction factors are introduced (such as the coupling effects of wind speed, temperature, and humidity on power output), and by dynamically comparing the actual power change characteristics, the prediction deviation under complex weather conditions is corrected. The meteorological prediction data is updated in real time (such as the monitoring of cloud dynamics), and by complementing incomplete or time-insufficient meteorological data, the input data of the model under complex conditions is ensured to be more accurate. Error feedback optimization is designed to record and track the deviations occurring in each prediction correction, and the overall prediction ability of the system is improved through a weighted optimization and parameter dynamic adjustment mechanism. The prediction results and the actual power generation data are dynamically compared daily, and the key feature weights in the model are automatically corrected (such as the influence coefficients of features such as cloud density and irradiance are updated in real time). The sliding window range is adjusted according to the error sequence to refine the time span of the input features. The operating status, maintenance records, and current efficiency of the photovoltaic device are collected (such as the cleaning status of the photovoltaic panel, etc.), and a prediction correction link is introduced to compensate for the deviation caused by the device state difference. For different types of photovoltaic power stations, the model input is calibrated according to their historical power generation laws, and the generalization ability of the model is improved by using adaptive weight settings suitable for specific regions.

[0115] Regarding step S105, in some embodiments, the analysis of the prediction error of the third model and, in response to the prediction error being greater than a preset threshold, the adjustment of the parameters in the third model to obtain a fourth model includes: using the third model to predict the photovoltaic power generation power on a short time scale to obtain third prediction data; using the third model to predict the photovoltaic power generation power on a long time scale to obtain fourth prediction data; calculating a third error sequence based on at least two of the third prediction data and the corresponding at least two historical push data; calculating a fourth error sequence based on at least two of the fourth prediction data and the corresponding at least two of the historical push data; analyzing the error sources of each error data in the third error sequence and the fourth error sequence to obtain a second result; adjusting the parameters in the third model based on the second result to obtain the fourth model.

[0116] In this embodiment, a dynamic deviation-driven optimization method is designed to compare and analyze the operation data with the historical grid-connected power generation data on the grid side to determine the prediction deviation of the model. If the prediction error exceeds the set threshold, the model optimization mechanism is triggered to dynamically adjust the selection of meteorological features, data weights, etc., to ensure the continuous optimization of the prediction system.

[0117] In this embodiment, for the short time scale, the meteorological feature data of the past 2 hours and the recent power output sequence are input, and short-term prediction is carried out in combination with the real-time environmental change parameters. It includes meteorological data (such as solar irradiance, dust index) and power historical data. The third model is used to predict the power changes in the next 15 minutes, 30 minutes, 1 hour, and 2 hours. The current dust index is 90 and the irradiance is 800 W / m². The predicted power in the next 15 minutes is 16.6 MW, and the predicted power in the next 1 hour is 17.2 MW. Real-time push data is obtained from the photovoltaic power station and compared with the prediction results of the third model. An error sequence for the short time scale is generated, and the error is recorded for each time period (such as a 15-minute interval). The predicted power on the first day is 16.8 MW, the actually measured power is 16.5 MW, and the error is 1.82%. Similarly, the errors in multiple time periods are recorded, and the error sequence is: time period 1: 1.82%, time period 2: 1.5%, time period 3: 2.8%.

[0118] For the long time scale, the power change data of the last 3 days and the meteorological prediction sequence are input, and the daily dust index and the change trend of cloud density are combined. The third model is used to predict the power trend change in the next 3 days (distributed hourly). In the case of the gradually weakening dust, the third model predicts that the power throughout the first day is 180 MW. The power rebounds on the second day, and the total power is 195 MW. The power further recovers to 210 MW on the third day. The predicted data for the next 3 days is compared with the cumulative value of the actual power to generate an error sequence for the long time scale. The predicted power on the first day is 180 MW, the actual power is 175 MW, and the error rate is 2.86%. The error rate on the second day is 1.56%, and the error rate on the third day is 2.1%.

[0119] Analyze the error sequence to determine the main deviation sources: sudden meteorological changes (such as abnormal increase in dust concentration). Decrease in component operation efficiency (low cleanliness). Uneven predicted power due to regional weather differences. The impact of the dust index rising to 95 on power shading is underestimated, and the model prediction error is concentrated at 10:00 - 11:00 in the morning. 80% of the components in the southern area have clean coverage, and the actual power reduction trend is lower than the model prediction. Trace the prediction deviation for the phenomenon of high power in the northern region and find that the irradiance factor is overestimated.

[0120] According to the error source analysis, adjust the parameters of the model: enhance the weight of the sudden weather parameter dust index from 0.8 to 0.9. Increase the compensation factor for the decrease in irradiance in the northern region to correct the impact on the occlusion environment. Use the error sequence feedback for model retraining: extract the dust increase and regional occlusion data in the past 3 days and add them to the new training samples. Adjust the number of hidden layer neurons in the LSTM model (increase from 128 to 256). After correcting the parameters and retraining the model, the power prediction error on the third day is optimized from 5.5% to 3.2%.

[0121] The embodiments of the present application significantly reduce the errors in short-term and long-term scale predictions, solve the problems of meteorological mutations, abnormal equipment states, and regional differences, enable the model to have stronger adaptability in complex environments, and provide guarantees for the accuracy and robustness of photovoltaic power generation predictions.

[0122] For step S107, use the fourth model to predict the photovoltaic power generation.

[0123] In this embodiment, the photovoltaic power generation prediction results are provided to the power grid dispatching center and the energy management system to assist in decision-making such as power grid load dispatching and energy storage system optimization. At the same time, record the error change trend during the prediction process, and continuously optimize the prediction model through machine learning methods to continuously improve its adaptability under complex weather conditions and ensure the stability and efficiency of the photovoltaic power generation system.

[0124] As can be seen from the above embodiments, the photovoltaic power prediction method described in the embodiments of the present application obtains the power output information of a photovoltaic power station; obtains meteorological key data based on a meteorological large model; performs fusion processing on the meteorological key data and the power output information to obtain a data feature set; there is a mapping relationship between the meteorological key data and the power output information in the data feature set; trains a first model based on the data feature set to obtain a second model; performs calibration processing on the second model based on the real-time push data of the photovoltaic power station to obtain a third model; performs prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjusts the parameters in the third model to obtain a fourth model; uses the fourth model to predict the photovoltaic power generation power. The embodiments of the present application provide a comprehensive solution for photovoltaic power prediction that combines accuracy, real-time performance, and complex environment adaptability. Aiming at the problem of insufficient prediction accuracy caused by the existing technology's reliance on traditional meteorological data, the present application introduces a meteorological large model to realize the real-time collection and prediction of high-resolution meteorological data, and combines it with the real-time power generation power and equipment operation status data of the photovoltaic power station to construct a deeply integrated feature set. This feature set can explore the change laws of photovoltaic power from multiple dimensions such as time correlation, spatial correlation, and complex meteorological dynamic changes, significantly improving the problem of low data input quality and difficulty in adapting to complex scenarios with a single feature. At the same time, through dynamic feature selection and interactive processing, the problem of insufficient ability of traditional models to express data spatio-temporal correlation and non-linear features is solved, providing an accurate analysis and prediction basis for power fluctuations caused by sudden weather changes (such as sandstorms, severe convective weather, etc.).

[0125] In terms of model construction and optimization, the present application adopts a series of innovative algorithms and dynamic adjustment mechanisms, effectively improving the adaptability and accuracy of the power prediction model. In multi-stage layer-by-layer training, through techniques such as cross-validation, adaptive learning rate optimization, and dynamic weight adjustment, the model can quickly adapt to the dynamic characteristics of photovoltaic power with changes in equipment status and complex weather. The model combines short-term to medium- and long-term predictions with a real-time calibration mechanism, and gradually optimizes the short-term prediction error and long-term power change laws, providing an efficient real-time correction ability for power fluctuations caused by sudden meteorological conditions. In terms of non-meteorological factor processing, the model fully considers equipment characteristics such as the operation status, cleanliness, and occlusion impact of photovoltaic equipment, and realizes accurate compensation for equipment local differences and environmental impacts through dynamic weight adjustment, thus solving the problem that traditional models fail to fully consider equipment status changes resulting in prediction deviations.

[0126] In addition, in view of the deficiencies in the data acquisition and real-time transmission links of photovoltaic power stations, this application realizes the rapid acquisition and synchronous processing of real-time data from multiple sites through embedded data acquisition and edge computing technologies, effectively alleviating the transmission delay problem caused by network load in traditional centralized computing platforms and improving the short-term prediction response speed. In a distributed photovoltaic scenario, this method can quickly adapt to the requirements of real-time observation and data management in the context of frequent device access and withdrawal. By continuously optimizing the device ledger and communication system, the stability and real-time performance of photovoltaic power station data transmission are significantly enhanced. In addition, by synchronously matching devices and data, the contradiction between the traditional ledger update speed and the rapid device access demand is resolved, providing high-quality support at the data level for power grid dispatching.

[0127] Based on the above innovative technologies, this application demonstrates excellent environmental adaptability and prediction performance in both complex meteorological conditions and multi-region scenarios. Whether in strong wind speed, cloudy and shaded weather, or in the context of large-scale access of distributed photovoltaics, the prediction results can meet the requirements of power grid load dispatching and energy storage optimization in terms of accuracy and stability. At the same time, combined with a dynamic error correction mechanism and continuous model optimization capabilities, this application has strong robustness for long-term application scenarios of photovoltaic power prediction, making device state changes and complex meteorological impacts no longer important factors hindering prediction accuracy. Overall, this application comprehensively improves the functionality and adaptability of the photovoltaic power prediction system, truly solving multiple problems in the aspects of data quality, model capabilities, and actual application scenarios in the existing technology, and providing reliable support for the intelligent management of photovoltaic power generation systems, the optimization of power grid dispatching, and the improvement of new energy consumption efficiency.

[0128] It should be noted that the method of the embodiment of this application can be executed by a single device, such as a computer or a server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In such a distributed scenario, one of the multiple devices can only execute one or more steps of the method of the embodiment of this application, and these multiple devices will interact with each other to complete the described method.

[0129] It should be noted that some embodiments of this application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be executed in a different order than in the above embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0130] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present application further provides a photovoltaic power prediction device.

[0131] Referring to Figure 2 , the photovoltaic power prediction device includes: An acquisition module 21, configured to acquire the power output information of a photovoltaic power station; and acquire meteorological key data based on a meteorological large model; A fusion module 22, configured to perform fusion processing on the meteorological key data and the power output information to obtain a data feature set; there is a mapping relationship between the meteorological key data and the power output information in the data feature set; A training module 23, configured to train a first model based on the data feature set to obtain a second model; A calibration module 24, configured to perform calibration processing on the second model based on the real-time push data of the photovoltaic power station to obtain a third model; An adjustment module 25, configured to perform prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjust the parameters in the third model to obtain a fourth model; A prediction module 26, configured to predict the photovoltaic power generation using the fourth model.

[0132] For the convenience of description, when describing the above device, it is divided into various modules according to functions for separate description. Of course, when implementing the present application, the functions of each module can be implemented in one or more software and / or hardware.

[0133] The device in the above embodiment is used to implement the corresponding photovoltaic power prediction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0134] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the photovoltaic power prediction method described in any of the above embodiments when executing the program.

[0135] Figure 3 FIG. shows a more specific schematic diagram of the hardware structure of the electronic device provided in this embodiment. The device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. Among them, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are communicatively connected to each other inside the device through the bus 1050.

[0136] The processor 1010 can be implemented in the form of a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0137] The memory 1020 can be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory 1020 can store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.

[0138] The input / output interface 1030 is used to connect to the input / output module to implement information input and output. The input / output module can be configured as a component in the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Among them, the input device can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device can include a display, a speaker, a vibrator, an indicator light, etc.

[0139] The communication interface 1040 is used to connect to a communication module (not shown in the figure) to implement communication and interaction between this device and other devices. Among them, the communication module can implement communication through a wired method (such as USB, network cable, etc.) or can implement communication through a wireless method (such as a mobile network, WIFI, Bluetooth, etc.).

[0140] The bus 1050 includes a path for transmitting information between various components of the device (such as the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0141] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in the specific implementation process, this device may also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device may also only include the components necessary to implement the solution of the embodiments of this specification and does not necessarily include all the components shown in the figure.

[0142] The electronic device of the above embodiment is used to implement the corresponding photovoltaic power generation prediction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0143] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the photovoltaic power generation prediction method described in any of the foregoing embodiments.

[0144] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0145] The computer instructions stored in the storage medium of the above embodiment are used to cause the computer to execute the photovoltaic power generation prediction method described in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0146] Based on the same inventive concept, corresponding to the photovoltaic power generation prediction method described in any of the above embodiments, the present disclosure also provides a computer program product including computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of the computer to cause the computer and / or the processor to execute the photovoltaic power generation prediction method. Corresponding to the execution subject corresponding to each step in each embodiment of the photovoltaic power generation prediction method, the processor executing the corresponding step can belong to the corresponding execution subject.

[0147] The computer program product of the above embodiment is used to cause the computer and / or the processor to execute the photovoltaic power generation prediction method described in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0148] Those of ordinary skill in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the present application (including the claims) is limited to these examples; under the concept of the present application, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations in different aspects of the embodiments of the present application as described above, and for the sake of brevity, they are not provided in detail.

[0149] In addition, for simplicity of explanation and discussion, and in order not to make the embodiments of the present application difficult to understand, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Further, the devices may be shown in block diagram form in order to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present application are to be implemented (i.e., these details should be fully within the understanding of those skilled in the art). In cases where specific details (such as circuits) are set forth to describe exemplary embodiments of the present application, it will be apparent to those skilled in the art that the embodiments of the present application may be implemented without these specific details or with variations of these specific details. Accordingly, these descriptions should be considered illustrative rather than restrictive.

[0150] Although the present application has been described in connection with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art based on the foregoing description. For example, other memory architectures (such as dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0151] The embodiments of the present application are intended to cover all such alternatives, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application shall be included within the protection scope of the present application.

Claims

1. A photovoltaic power prediction method, characterized in that, Including: Obtaining the power output information of a photovoltaic power station; obtaining key meteorological data based on a meteorological large model; Fusing the key meteorological data and the power output information to obtain a data feature set; there is a mapping relationship between the key meteorological data and the power output information in the data feature set; Training a first model based on the data feature set to obtain a second model; Based on the real-time push data of the photovoltaic power station, performing calibration processing on the second model to obtain a third model; Performing prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjusting the parameters in the third model to obtain a fourth model; Using the fourth model to predict the photovoltaic power generation power.

2. The method according to claim 1, wherein The fusing the key meteorological data and the power output information to obtain a data feature set includes: Using a sliding window technique to dynamically capture the time correlation relationship between the key meteorological data and the power output information through a time series analysis method; Using a data reconstruction algorithm to obtain the spatial correlation relationship between the key meteorological data and the power output information; Using a piecewise fitting algorithm to extract the power drop sequence data corresponding to the power change caused by the cloud density parameter in the key meteorological data; Taking the geographical location of the photovoltaic power station as a node, converting the spatiotemporally correlated data in the key meteorological data into a graph structure, and using a graph neural network to adjust the time correlation relationship, the spatial correlation relationship, and the power drop sequence data; Based on the adjusted time correlation relationship, spatial correlation relationship, and power drop sequence data, obtaining the data feature set.

3. The method according to claim 1, characterized in that The training the first model based on the data feature set to obtain a second model includes: Using a cross-validation algorithm to process the data feature set, inputting the processed data feature set into the first model, and using an adaptive learning rate optimization algorithm to dynamically adjust the learning rate of the first model to obtain a trained learning rate; Using a grid search algorithm or a random search algorithm to adjust the hyperparameters of the first model to obtain trained hyperparameters; Based on the trained learning rate and the trained hyperparameters, training to obtain the second model.

4. The method according to claim 1, wherein The performing calibration processing on the second model based on the real-time push data of the photovoltaic power station to obtain a third model includes: Using the second model to predict the photovoltaic power generation power on a short time scale to obtain first prediction data; Using the second model to predict the photovoltaic power generation power on a long time scale to obtain second prediction data; Calculating a first error sequence based on at least two of the first prediction data and the corresponding at least two real-time push data; Calculating a second error sequence based on at least two of the second prediction data and the corresponding at least two real-time push data; Analyzing the error sources of each error data in the first error sequence and the second error sequence to obtain a first result; Performing calibration processing on the second model according to the first result to obtain the third model.

5. The method according to claim 4, characterized in that Performing calibration processing on the second model according to the first result includes: In response to the first result being that at least one of the meteorological key data has a sudden change, dynamically adjusting the weight of the corresponding meteorological key data to perform calibration processing on the second model; In response to the first result being that the device status of the photovoltaic power station is abnormal, dynamically modifying or adding parameters corresponding to the abnormal device to perform calibration processing on the second model.

6. The method according to claim 1, wherein Performing prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjusting the parameters in the third model to obtain a fourth model, including: Using the third model to predict the photovoltaic power generation power on a short time scale to obtain third prediction data; Using the third model to predict the photovoltaic power generation power on a long time scale to obtain fourth prediction data; Calculating a third error sequence based on at least two of the third prediction data and corresponding at least two historical push data; Calculating a fourth error sequence based on at least two of the fourth prediction data and corresponding at least two of the historical push data; Analyzing the error sources of each error data in the third error sequence and the fourth error sequence to obtain a second result; Adjusting the parameters in the third model based on the second result to obtain the fourth model.

7. A photovoltaic power generation power prediction device, characterized in that Including: An acquisition module configured to acquire the power output information of the photovoltaic power station; Acquiring meteorological key data based on the meteorological large model; A fusion module configured to perform fusion processing on the meteorological key data and the power output information to obtain a data feature set; there is a mapping relationship between the meteorological key data and the power output information in the data feature set; A training module configured to train a first model based on the data feature set to obtain a second model; A calibration module configured to perform calibration processing on the second model based on the real-time push data of the photovoltaic power station to obtain a third model; An adjustment module configured to perform prediction error analysis on the third model, and in response to the prediction error being greater than a preset threshold, adjust the parameters in the third model to obtain a fourth model; A prediction module configured to predict the photovoltaic power generation power using the fourth model.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, the method described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the method described in any one of claims 1 to 6.

10. A computer program product, including computer program instructions, when the computer program instructions run on a computer, cause the computer to execute the method described in any one of claims 1 to 6.

Images