A distributed photovoltaic power station power prediction updating method and system
By constructing a distributed photovoltaic power plant cluster, using common information for unified prediction and combining individual differences for correction, the problems of high cost of independent deployment of distributed photovoltaic power plants and inconsistent prediction results are solved, realizing low-cost and high-precision power prediction at the plant level.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING GUODIAN NANZI POWER GRID AUTOMATION CO LTD
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies struggle to generate stable prediction results applicable to each distributed photovoltaic power station without deploying a complete prediction system for each station individually. This results in problems such as high costs for independent deployment at each station, lack of independent prediction conditions, difficulty in unifying commonalities and individual characteristics, and difficulty in uniformly optimizing prediction results.
By constructing a power station cluster, common information on weather changes, irradiance trends, and time cycles from multiple power stations within the same cluster is used to perform unified power prediction, forming a cluster-level power prediction result. This result is then corrected based on individual differences among the power stations to form the final power prediction value for each power station.
It enables the generation of power prediction results that can be applied to various sites at a lower deployment cost, improves the availability and adaptability of site-level prediction, reduces the duplication of computing power and operation and maintenance resources, and improves the accuracy and applicability of prediction results.
Smart Images
Figure CN122437488A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of new energy photovoltaic power generation prediction technology, and in particular to a method and system for updating power prediction of distributed photovoltaic power plants. Background Technology
[0002] With the rapid development of distributed photovoltaic (PV) systems, a large number of small-scale, geographically dispersed PV power plants are often located within the grid dispatch area. Compared with centralized large-scale PV power plants, distributed PV power plants are more numerous, have smaller individual capacities, and exhibit greater differences in operating conditions, leading to a significant increase in the difficulty of constructing and maintaining their forecasting systems.
[0003] From the perspective of current technological development, research on photovoltaic power prediction started earlier abroad. Related technologies mainly revolve around centralized large-scale photovoltaic power plants, regional renewable energy power prediction, and electricity market dispatch needs. They typically rely on relatively complete meteorological observation systems, comprehensive historical operating data, and stable single-site modeling conditions. Under this technological approach, the prediction targets are mostly single large-scale power plants, centrally connected photovoltaic power sources, or regional total power. The technological focus is usually on improving the accuracy of single-site predictions, optimizing regional total power prediction results, or improving the accuracy of market transactions and dispatch arrangements.
[0004] China's distributed photovoltaic (PV) industry is developing rapidly, with large-scale grid integration and diverse site types. Particularly in county-level areas, industrial parks, and commercial / industrial rooftops, distributed PV sites are characterized by their large number, small capacity, dispersed layout, and uneven operation and maintenance levels. Compared to centralized large-scale power plants, many distributed PV sites suffer from insufficient data accumulation, incomplete meteorological information, limited communication conditions, and inadequate local forecasting system configurations. This makes it difficult to directly adapt foreign technologies designed for large-scale single-site or regional total volume forecasting to China's large-scale distributed PV scenarios.
[0005] In existing technologies, power prediction for distributed photovoltaic (PV) power plants typically suffers from the following problems:
[0006] (1) High cost of independent deployment for each distributed photovoltaic power station. If a complete power prediction system is deployed for each distributed photovoltaic power station, it is necessary to configure independent data access, feature processing, model training, operation and maintenance monitoring and result publishing capabilities, which results in high construction costs and complex maintenance, making it difficult to promote in large-scale distributed scenarios.
[0007] (2) Some sites lack independent forecasting conditions. Many distributed photovoltaic sites lack historical data, complete meteorological information, or even the data conditions required to independently deploy forecasting models, resulting in the inability to stably output site-level forecasting results.
[0008] (3) There is a problem of commonality and individuality among power stations. Multiple photovoltaic power stations in the same area are usually affected by similar weather systems, irradiance variation trends and seasonal cycles, and have regional commonalities that can be used for unified modeling; however, different power stations have differences in terms of installed capacity, component orientation, shading conditions, equipment performance and local operating environment. If only the total cluster is predicted, it is difficult to directly form a result that meets the application needs of the power stations; if each power station is modeled completely, it will bring high construction and maintenance costs.
[0009] (4) It is difficult to uniformly distribute and continuously optimize the prediction results. Under the existing model, the sources of prediction results at the field station level are scattered and of varying quality, making it difficult to form a unified prediction generation mechanism, a unified result output mechanism, and a unified error feedback closed-loop mechanism.
[0010] Therefore, neither the foreign technical approach of large-scale single-site or regional total prediction, nor the domestic prediction system construction that is gradually moving towards distributed scenarios, has effectively solved the problem of "how to stably generate prediction results that can be used by each site without deploying a complete prediction system for each of the many distributed photovoltaic sites". Summary of the Invention
[0011] The purpose of this invention is to provide a method and system for updating power prediction for distributed photovoltaic (PV) power plants. By constructing a power plant cluster, common information from multiple power plants within the same cluster regarding weather changes, irradiance trends, and time periods is used to perform unified power prediction. First, a cluster-level power prediction result is generated. Then, this cluster-level power prediction result is mapped to the basic prediction value for each power plant, and further site-specific corrections are made based on individual plant differences, thereby generating the final power prediction value for each power plant. This invention enables the generation of power prediction results applicable to various power plants at a relatively low deployment cost, and improves the availability and adaptability of site-level predictions. This invention is achieved through the following technical solutions.
[0012] In a first aspect, the present invention provides a method for updating the power prediction of distributed photovoltaic power plants, comprising:
[0013] Acquire basic information about multiple distributed photovoltaic power stations, and construct a cluster based on the basic information and calculate the capacity share of each distributed photovoltaic power station in the cluster;
[0014] Based on the capacity ratio, the meteorological characteristics pre-acquired by each distributed photovoltaic power station are weighted and aggregated to obtain cluster-level meteorological characteristics;
[0015] Extract the temporal features of the predicted time from the pre-acquired meteorological feature data, and combine the cluster-level meteorological features, temporal features, and cluster-level historical power features into cluster-level input features;
[0016] The cluster-level input features are input into the prediction model to perform unified power prediction for distributed photovoltaic power stations, and the cluster-level power prediction results are obtained.
[0017] The cluster-level power prediction results are mapped according to the capacity ratio and historical contribution coefficient to obtain the basic prediction values for each distributed photovoltaic power station.
[0018] The final power prediction value for each distributed photovoltaic power station is obtained by correcting the basic prediction value of each distributed photovoltaic power station.
[0019] The actual power value of each distributed photovoltaic power station is obtained. The difference between the actual power value and the basic prediction value and the final power prediction value is calculated to obtain the cluster-level prediction error and the power station-level prediction error. The final power prediction value is updated by using the cluster-level prediction error and the power station-level prediction error.
[0020] Optionally, the basic information includes the site identifier, installed capacity, geographical location, operating status, and historical power generation data; the meteorological characteristics include irradiance, temperature, humidity, and wind speed; and the prediction model includes one or more of machine learning models, deep learning models, and ensemble prediction models.
[0021] In practical applications, one or more of the above prediction models can be selected for use. When multiple prediction models are used, the corresponding weights are determined based on the prediction accuracy of each model within the historical time window, and the prediction results of multiple prediction models are weighted and fused.
[0022] Optionally, the capacity percentage is calculated using the following formula:
[0023] ,
[0024] in, and All are serial numbers of distributed photovoltaic power stations, with values ranging from 1 to... , This represents the total number of distributed photovoltaic power stations in the cluster. and The first The and the first The installed capacity of a distributed photovoltaic power station For the first The capacity percentage of each distributed photovoltaic power station.
[0025] Optionally, the cluster-level meteorological characteristics are calculated using the following formula:
[0026] ,
[0027] ,
[0028] In the formula, It is a cluster-level meteorological feature. This is a dimension of meteorological characteristics, with values ranging from 1 to... , The total number of dimensions of meteorological features. For the first Cluster meteorological characteristics, For the first The first distributed optical field station Pre-acquired meteorological characteristics.
[0029] Optionally, cluster-level input features are calculated using the following formula:
[0030] ,
[0031] In the formula, For cluster-level prediction, the input feature vector is... For time feature vectors, This is a cluster-level historical power feature vector. This indicates a vertical concatenation operation of vectors.
[0032] Optionally, the cluster-level power prediction result is calculated using the following formula:
[0033] ,
[0034] In the formula, For cluster-level power prediction results, This refers to the index of the prediction sub-model in the prediction model, with a value ranging from 1 to... , This represents the total number of prediction sub-models in the prediction model. For the first The fusion weights of the prediction sub-models For the first Power prediction results of each prediction sub-model;
[0035] The first Power prediction results of each prediction sub-model It is obtained by calculation using the following formula:
[0036] ,
[0037] In the formula, For the first The weight vector of the output layer of each predictive sub-model has the same dimension as the hidden layer. for transpose, For the first The bias scalar of the output layer of each prediction sub-model For the first The hidden feature vectors of each prediction sub-model are calculated using the following formula:
[0038] ,
[0039] In the formula, For the first Nonlinear activation functions for each predictive sub-model. For the first The weight matrix of the feature transformation layer of each prediction sub-model For the first The bias vector of the feature transformation layer of each predictive sub-model has a dimension equal to the hidden layer dimension.
[0040] Optionally, the basic predicted value is calculated using the following formula:
[0041] ,
[0042] In the formula, and The first The and the first The historical contribution coefficient of each distributed photovoltaic power station For cluster-level power prediction results, For the first Basic predicted values for a distributed photovoltaic power station.
[0043] Optionally, the final power prediction value is calculated using the following formula:
[0044] ,
[0045] In the formula, For the first The site correction weights corresponding to each distributed photovoltaic power station, and , For the first Auxiliary forecast values for individual distributed photovoltaic power stations. For the first Final power prediction values for each distributed photovoltaic power station.
[0046] Optionally, updating the final power prediction value using cluster-level prediction errors and site-level prediction errors includes: updating the first... The and the first Historical contribution coefficient of distributed photovoltaic power stations and Through the updated historical contribution coefficient and Update # Basic forecast value for a distributed photovoltaic power station Update the prediction error at the station level. Site correction weights for individual distributed photovoltaic power stations ; through the updated version of the updated version Basic forecast value for a distributed photovoltaic power station and the updated version Site correction weights for individual distributed photovoltaic power stations Jointly update the Final power forecast values for each distributed photovoltaic power station .
[0047] Secondly, the present invention provides a distributed photovoltaic power prediction and update system, comprising:
[0048] The cluster building module is used to obtain basic information of multiple distributed photovoltaic power stations, and to build a cluster based on the basic information and calculate the capacity ratio of each distributed photovoltaic power station in the cluster.
[0049] The meteorological feature generation module is used to weight and aggregate the meteorological feature data pre-acquired by each distributed photovoltaic power station according to the capacity ratio to obtain cluster-level meteorological features;
[0050] The cluster-level input feature generation module is used to extract the time features of the predicted time from the pre-acquired meteorological feature data, and combine the cluster-level meteorological features, time features and cluster-level historical power features into cluster-level input features;
[0051] The power prediction module is used to input cluster-level input features into the prediction model to perform unified power prediction for distributed photovoltaic power stations and obtain cluster-level power prediction results.
[0052] The site mapping module is used to map the cluster-level power prediction results according to the capacity ratio and historical contribution coefficient to obtain the basic prediction values of each distributed photovoltaic site.
[0053] The site correction module is used to correct the basic forecast values of each distributed photovoltaic power station to obtain the final power forecast value of each distributed photovoltaic power station.
[0054] The update module is used to obtain the actual power value of each distributed photovoltaic power station, and to calculate the difference between the actual power value and the basic prediction value and the final power prediction value to obtain the cluster-level prediction error and the power station-level prediction error. The final power prediction value is updated by using the cluster-level prediction error and the power station-level prediction error.
[0055] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0056] This invention introduces a distributed photovoltaic (PV) power prediction and update method. It combines common information from multiple PV sites within the same cluster regarding weather changes, irradiance trends, and time periods. First, it uses an existing prediction model to generate a unified cluster-level power prediction result, fully utilizing the common information from multiple sites within the same region. This cluster-level power prediction result is then mapped to the base prediction value for each site, and further site-specific corrections are made based on individual site differences, resulting in the final power prediction value for each site. This fully utilizes site-level mapping and site-specific corrections to compensate for individual differences between different sites, balancing regional commonalities with site-specific characteristics. Compared to traditional site-by-site deployment, this invention uses unified cluster modeling and prediction capabilities for multiple sites within a region. By replacing independent site-by-site deployment with unified cluster prediction, it generates power prediction results for each site at a lower cost. Furthermore, it improves the site adaptability of the prediction results by combining commonalities extraction with individual corrections.
[0057] By replacing independent deployment at each distributed photovoltaic (PV) site with unified forecasting based on PV clusters, redundant construction of computing power and O&M resources is reduced, lowering deployment costs. For sites that previously lacked independent forecasting capabilities, site-level forecasting results can be obtained through unified forecasting by PV clusters, site-level mapping, and site-specific corrections, improving the availability of site-level forecasts. Site-specific corrections support the introduction of auxiliary corrections for some sites with existing auxiliary forecasting information. For sites without existing forecasting capabilities, the final result is obtained based on cluster forecasting and site mapping, without relying on all sites having complete local forecasting systems, thus improving project applicability. Through a unified result output and actual power feedback mechanism, the historical contribution coefficient and site-specific correction weights are continuously optimized, resulting in more accurate final power forecast values for each distributed PV site. Attached Figure Description
[0058] Figure 1 The diagram shown is a schematic flowchart of a distributed photovoltaic power prediction and update method in one embodiment of the present invention. Detailed Implementation
[0059] The following description, in conjunction with the accompanying drawings and specific embodiments, provides further details. In this description, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature.
[0060] Example 1 This embodiment introduces a method for power prediction and updating of distributed photovoltaic power plants, such as... Figure 1 As shown, it includes the following steps:
[0061] Acquire basic information about multiple distributed photovoltaic power stations, and construct a cluster based on the basic information and calculate the capacity share of each distributed photovoltaic power station in the cluster;
[0062] Based on the capacity ratio, the meteorological characteristics pre-acquired by each distributed photovoltaic power station are weighted and aggregated to obtain cluster-level meteorological characteristics;
[0063] Extract the temporal features of the predicted time from the pre-acquired meteorological feature data, and combine the cluster-level meteorological features, temporal features, and cluster-level historical power features into cluster-level input features;
[0064] The cluster-level input features are input into the prediction model to perform unified power prediction for distributed photovoltaic power stations, and the cluster-level power prediction results are obtained.
[0065] The cluster-level power prediction results are mapped according to the capacity ratio and historical contribution coefficient to obtain the basic prediction values for each distributed photovoltaic power station.
[0066] The final power prediction value for each distributed photovoltaic power station is obtained by correcting the basic prediction value of each distributed photovoltaic power station.
[0067] The actual power value of each distributed photovoltaic power station is obtained. The difference between the actual power value and the basic prediction value and the final power prediction value is calculated to obtain the cluster-level prediction error and the power station-level prediction error. The final power prediction value is updated by using the cluster-level prediction error and the power station-level prediction error.
[0068] Example 2 Based on Example 1, this example introduces a specific implementation process of a distributed photovoltaic power prediction and update method, including the following:
[0069] I. Building a Cluster
[0070] In one specific embodiment of the present invention, basic information of multiple distributed photovoltaic power stations within a target geographical area is first obtained. This basic information is then used to construct a distributed photovoltaic power station cluster, referred to as a power station cluster. Distributed photovoltaic power stations are simply referred to as power stations.
[0071] In one specific embodiment of the present invention, the basic information of a distributed photovoltaic power station includes the station identifier, installed capacity, geographical location, operating status, and historical power generation data.
[0072] In one specific embodiment of the present invention, the target geographical area is determined according to the power grid dispatch jurisdiction, administrative division or similar meteorological conditions, and multiple distributed photovoltaic power stations located in the target geographical area are selected as candidate power stations for building a power station cluster.
[0073] II. Calculation
[0074] 2.1 Computing capacity percentage
[0075] In one specific embodiment of the present invention, the capacity ratio is calculated using the following formula:
[0076] ,
[0077] in, and All are serial numbers of distributed photovoltaic power stations, with values ranging from 1 to... , This represents the total number of distributed photovoltaic power stations in the cluster. and The first The and the first The installed capacity of a distributed photovoltaic power station For the first The capacity percentage of each distributed photovoltaic power station.
[0078] 2.2 Calculate cluster-level meteorological characteristics
[0079] In one specific embodiment of the present invention, cluster-level meteorological characteristics are calculated using the following formula:
[0080] ,
[0081] ,
[0082] In the formula, It is a cluster-level meteorological feature. This is a dimension of meteorological characteristics, with values ranging from 1 to... , The total number of dimensions of meteorological features. For the first Cluster meteorological characteristics, For the first The first distributed optical field station Pre-acquired meteorological characteristics.
[0083] The above formula shows that each dimension of cluster-level meteorological feature is the result of a weighted sum of the meteorological features of all stations in the cluster corresponding to the same dimension according to their capacity proportions. The larger the capacity proportion of a station, the greater its contribution to the cluster-level meteorological feature of that dimension.
[0084] In one specific embodiment of the present invention, meteorological characteristics include irradiance, temperature, humidity, and wind speed.
[0085] 2.3 Calculate cluster-level input features
[0086] In one specific embodiment of the present invention, the expression for the cluster-level input feature is as follows:
[0087] ,
[0088] In the formula, For cluster-level prediction, the input feature vector is... For time feature vectors, This is a cluster-level historical power feature vector. This indicates a vertical concatenation operation of vectors.
[0089] The dimension is 3 represents the number of time feature components. The sequence length of the cluster-level historical power feature vector; the hour, month, and season of the prediction time are encoded to form the time feature vector. The expression is as follows:
[0090] ,
[0091] In the formula, For time feature vectors, The hourly feature encoding value for the predicted time. The month feature encoding value for the prediction time. The seasonal feature encoding value for the predicted time. This indicates that the vector is transposed into a column vector.
[0092] The first The length of each station before the predicted time is The historical power sequence is denoted as:
[0093]
[0094] In the formula, For the first Historical power feature vectors of each power station For the first The site is located before the predicted time. The actual power generation at each time step The time step number. , The preset historical time window length.
[0095] The historical power sequences above are weighted and summed according to the capacity proportion of each power station to obtain the cluster-level historical power vector:
[0096]
[0097] In the formula, This is a cluster-level historical power vector. For the first The site is located before the predicted time. The actual power generation at each time step. The Each component For all stations, the first time before the prediction time The power generation at each time step is weighted and summed according to capacity ratio, thereby integrating the historical power output information of multiple power stations into a cluster-level historical power sequence.
[0098] In one specific embodiment of the present invention, the time feature includes one or more of the hourly feature, monthly feature, and seasonal feature of the predicted time.
[0099] 2.4 Calculate cluster-level power prediction results
[0100] In one specific embodiment of the present invention, the cluster-level power prediction result is calculated using the following formula:
[0101] ,
[0102] In the formula, For cluster-level power prediction results, This refers to the index of the prediction sub-model in the prediction model, with a value ranging from 1 to... , The total number of prediction sub-models in the prediction model, when The time-dependent formula degenerates into a single prediction model. For the first The fusion weights of the prediction sub-models For the first Power prediction results of each prediction sub-model;
[0103] The fusion weights are determined by the average prediction error of each sub-model within a historical time window, and the calculation formula is as follows:
[0104] ,
[0105] In the formula, For the first The average prediction error of each prediction sub-model within a historical time window. It is a natural exponential function. The summation index takes values ranging from 1 to 1. ;
[0106] The above equation satisfies and The smaller the historical prediction error, the greater the fusion weight the sub-model receives.
[0107] The first The power prediction results of each prediction sub-model are calculated using the following formula:
[0108] ,
[0109] In the formula, For the first The weight vector of the output layer of each predictive sub-model has the same dimension as the hidden layer. for transpose, For the first The bias scalar of the output layer of each prediction sub-model For the first The hidden feature vectors of each prediction sub-model are calculated using the following formula:
[0110] ,
[0111] In the formula, For the first Nonlinear activation functions for each predictive sub-model. For the first The weight matrix of the feature transformation layer of each prediction sub-model has rows equal to the hidden layer dimension and columns equal to... Dimensions , For the first The bias vector of the feature transformation layer of each predictive sub-model has a dimension equal to the hidden layer dimension.
[0112] In one specific embodiment of the present invention, the prediction model includes a machine learning model, a deep learning model, and an ensemble prediction model. In practical applications, one or more of the above prediction models can be selected for use. When multiple prediction models are used, the corresponding weights are determined based on the prediction accuracy of each prediction model within a historical time window, and the prediction results of the multiple prediction models are weighted and fused.
[0113] 2.5 Calculate the basic predicted values for each distributed photovoltaic power station
[0114] In one specific embodiment of the present invention, the basic predicted value is calculated using the following formula:
[0115] ,
[0116] In the formula, and The first The and the first The historical contribution coefficient of each distributed photovoltaic power station For cluster-level power prediction results, For the first Basic predicted values for a distributed photovoltaic power station.
[0117] 2.6 Calculate the final power prediction value for each distributed photovoltaic power station.
[0118] In one specific embodiment of the present invention, the final power prediction value is calculated using the following formula:
[0119] ,
[0120] In the formula, For the first The site correction weights corresponding to each distributed photovoltaic power station, and , For the first Auxiliary forecast values for individual distributed photovoltaic power stations. For the first Final power prediction values for each distributed photovoltaic power station.
[0121] when When the value is larger, it indicates that the final power prediction value of the current distributed photovoltaic power station has a greater weighting than the basic prediction value; when... When the weight is smaller, it indicates that the final power prediction value of the current distributed photovoltaic power station has a greater emphasis on the auxiliary prediction value; when When the final power prediction is determined entirely by the baseline prediction, when... At that time, the final power prediction value is entirely determined by the auxiliary prediction value. This is achieved by utilizing the power station correction weights as described above. Correcting the final power prediction This approach allows the correction process for distributed photovoltaic (PV) power plants to be adjusted based on the completeness of data, the quality of existing auxiliary prediction information, and historical error performance of different distributed PV power plants, thereby improving the stability of the final power prediction value and the adaptability of each distributed PV power plant.
[0122] III. Update
[0123] In a specific embodiment of the present invention, updating the final power prediction value using cluster-level prediction error and site-level prediction error includes: updating the first... The and the first Historical contribution coefficient of distributed photovoltaic power stations and Through the updated historical contribution coefficient and Update # Basic forecast value for a distributed photovoltaic power station Update the prediction error at the station level. Site correction weights for individual distributed photovoltaic power stations ; through the updated version of the updated version Basic forecast value for a distributed photovoltaic power station and the updated version Site correction weights for individual distributed photovoltaic power stations Jointly update the Final power forecast values for each distributed photovoltaic power station .
[0124] IV. Examples
[0125] In one specific embodiment of the present invention, taking several distributed photovoltaic power stations within a prefecture-level city as an example, there are 50 distributed photovoltaic power stations in the prefecture-level city, with a single station installed capacity ranging from 2MW to 15MW. Most of the power stations have not independently deployed a complete power prediction system.
[0126] With this invention, a regional-level forecasting platform uniformly acquires basic information, meteorological information, and historical power data from each power station, forming a power station cluster. The platform first performs unified modeling based on the power station cluster and outputs cluster-level power prediction results. Then, based on the capacity ratio of each power station, historical contribution relationships, and local feature information, it generates the final power prediction value for each power station and distributes the final power prediction value to the corresponding power station.
[0127] In this embodiment, it is not necessary to deploy 50 complete prediction systems for each of the 50 sites. Only one clustered prediction platform is needed to output prediction results for each site, thereby reducing construction and operation costs. For most sites that do not have independently deployed prediction systems, the site-level prediction results output by this platform can be used; for sites with existing auxiliary prediction information, auxiliary corrections can be introduced. Due to the introduction of site-level mapping and site-specific corrections, the prediction results for each site have better site adaptability compared to the simple allocation method based on capacity ratio.
[0128] Example 3 This embodiment introduces a distributed photovoltaic power prediction and update system, including:
[0129] The cluster building module is used to obtain basic information of multiple distributed photovoltaic power stations, and to build a cluster based on the basic information and calculate the capacity ratio of each distributed photovoltaic power station in the cluster.
[0130] The meteorological feature generation module is used to weight and aggregate the meteorological feature data pre-acquired by each distributed photovoltaic power station according to the capacity ratio to obtain cluster-level meteorological features;
[0131] The cluster-level input feature generation module is used to extract the time features of the predicted time from the pre-acquired meteorological feature data, and combine the cluster-level meteorological features, time features and cluster-level historical power features into cluster-level input features;
[0132] The power prediction module is used to input cluster-level input features into the prediction model to perform unified power prediction for distributed photovoltaic power stations and obtain cluster-level power prediction results.
[0133] The site mapping module is used to map the cluster-level power prediction results according to the capacity ratio and historical contribution coefficient to obtain the basic prediction values of each distributed photovoltaic site.
[0134] The site correction module is used to correct the basic forecast values of each distributed photovoltaic power station to obtain the final power forecast value of each distributed photovoltaic power station.
[0135] The update module is used to obtain the actual power value of each distributed photovoltaic power station, and to calculate the difference between the actual power value and the basic prediction value and the final power prediction value to obtain the cluster-level prediction error and the power station-level prediction error. The final power prediction value is updated by using the cluster-level prediction error and the power station-level prediction error.
[0136] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.
Claims
1. A method for predicting and updating the power of distributed photovoltaic power plants, characterized in that, include: Acquire basic information about multiple distributed photovoltaic power stations, and construct a cluster based on the basic information and calculate the capacity share of each distributed photovoltaic power station in the cluster; Based on the capacity ratio, the meteorological characteristics pre-acquired by each distributed photovoltaic power station are weighted and aggregated to obtain cluster-level meteorological characteristics; Extract the temporal features of the predicted time from the pre-acquired meteorological feature data, and combine the cluster-level meteorological features, temporal features, and cluster-level historical power features into cluster-level input features; The cluster-level input features are input into the prediction model to perform unified power prediction for distributed photovoltaic power stations, and the cluster-level power prediction results are obtained. The cluster-level power prediction results are mapped according to the capacity ratio and historical contribution coefficient to obtain the basic prediction values for each distributed photovoltaic power station. The final power prediction value for each distributed photovoltaic power station is obtained by correcting the basic prediction value of each distributed photovoltaic power station. The actual power value of each distributed photovoltaic power station is obtained. The difference between the actual power value and the basic prediction value and the final power prediction value is calculated to obtain the cluster-level prediction error and the power station-level prediction error. The final power prediction value is updated by using the cluster-level prediction error and the power station-level prediction error.
2. The distributed photovoltaic power prediction and update method according to claim 1, characterized in that, The basic information includes the site identifier, installed capacity, geographical location, operating status, and historical power generation data; the meteorological characteristics include irradiance, temperature, humidity, and wind speed; and the prediction model includes one or more of the following: machine learning model, deep learning model, and ensemble prediction model.
3. The distributed photovoltaic power prediction and update method according to claim 1, characterized in that, The capacity percentage is calculated using the following formula: ， in, and All are serial numbers of distributed photovoltaic power stations, with values ranging from 1 to... , This represents the total number of distributed photovoltaic power stations in the cluster. and The first The and the first The installed capacity of a distributed photovoltaic power station For the first The capacity percentage of each distributed photovoltaic power station.
4. The distributed photovoltaic power prediction and update method according to claim 3, characterized in that, The cluster-level meteorological characteristics are calculated using the following formula: ， ， In the formula, It is a cluster-level meteorological feature. This is a dimension of meteorological characteristics, with values ranging from 1 to... , The total number of dimensions of meteorological features. For the first Cluster meteorological characteristics, For the first The first distributed optical field station Pre-acquired meteorological characteristics.
5. The distributed photovoltaic power prediction and update method according to claim 4, characterized in that, The expression for the cluster-level input features is as follows: ， In the formula, For cluster-level prediction, the input feature vector is... For time feature vectors, This is a cluster-level historical power feature vector. This indicates a vertical concatenation operation of vectors.
6. The distributed photovoltaic power prediction and update method according to claim 5, characterized in that, The cluster-level power prediction result is calculated using the following formula: ， In the formula, For cluster-level power prediction results, This refers to the index of the prediction sub-model in the prediction model, with a value ranging from 1 to... , This represents the total number of prediction sub-models in the prediction model. For the first The fusion weights of the prediction sub-models For the first Power prediction results of each prediction sub-model; The first Power prediction results of each prediction sub-model It is obtained by calculation using the following formula: ， In the formula, For the first The weight vector of the output layer of each prediction sub-model for transpose, For the first The bias scalar of the output layer of each prediction sub-model For the first The hidden feature vectors of each prediction sub-model are calculated using the following formula: ， In the formula, For the first Nonlinear activation functions for each predictive sub-model. For the first The weight matrix of the feature transformation layer of the prediction sub-model For the first The bias vector of the feature transformation layer of the prediction sub-model.
7. The distributed photovoltaic power prediction and update method according to claim 6, characterized in that, The basic predicted value is calculated using the following formula: ， In the formula, and The first The and the first The historical contribution coefficient of each distributed photovoltaic power station For cluster-level power prediction results, For the first Basic predicted values for a distributed photovoltaic power station.
8. The distributed photovoltaic power prediction and update method according to claim 7, characterized in that, The final power prediction value is calculated using the following formula: ， In the formula, For the first The site correction weights corresponding to each distributed photovoltaic power station, and , For the first Auxiliary forecast values for individual distributed photovoltaic power stations. For the first Final power prediction values for each distributed photovoltaic power station.
9. The distributed photovoltaic power prediction and update method according to claim 8, characterized in that, Updating the final power prediction value using cluster-level prediction errors and site-level prediction errors includes: updating the first... The and the first Historical contribution coefficient of distributed photovoltaic power stations and Through the updated historical contribution coefficient and Update # Basic forecast value for a distributed photovoltaic power station Update the prediction error at the station level. Site correction weights for individual distributed photovoltaic power stations ; through the updated version of the updated version Basic forecast value for a distributed photovoltaic power station and the updated version Site correction weights for individual distributed photovoltaic power stations Jointly update the Final power forecast values for each distributed photovoltaic power station .
10. A distributed photovoltaic power prediction and update system, characterized in that, include: The cluster building module is used to obtain basic information of multiple distributed photovoltaic power stations, and to build a cluster based on the basic information and calculate the capacity ratio of each distributed photovoltaic power station in the cluster. The meteorological feature generation module is used to weight and aggregate the meteorological feature data pre-acquired by each distributed photovoltaic power station according to the capacity ratio to obtain cluster-level meteorological features; The cluster-level input feature generation module is used to extract the time features of the predicted time from the pre-acquired meteorological feature data, and combine the cluster-level meteorological features, time features and cluster-level historical power features into cluster-level input features; The power prediction module is used to input cluster-level input features into the prediction model to perform unified power prediction for distributed photovoltaic power stations and obtain cluster-level power prediction results. The site mapping module is used to map the cluster-level power prediction results according to the capacity ratio and historical contribution coefficient to obtain the basic prediction values of each distributed photovoltaic site. The site correction module is used to correct the basic forecast values of each distributed photovoltaic power station to obtain the final power forecast value of each distributed photovoltaic power station. The update module is used to obtain the actual power value of each distributed photovoltaic power station, and to calculate the difference between the actual power value and the basic prediction value and the final power prediction value to obtain the cluster-level prediction error and the power station-level prediction error. The final power prediction value is updated by using the cluster-level prediction error and the power station-level prediction error.