A power distribution network photovoltaic power prediction method and system

Abstract

This invention discloses a method and system for predicting photovoltaic (PV) power in a distribution network, relating to the field of PV output prediction technology. The method includes: preprocessing collected sample data from within a county to construct an original dataset; proposing PV output prediction evaluation indicators and initializing model parameters; dividing the region into sub-regions using empirical orthogonal decomposition and agglomerative hierarchical clustering algorithms; establishing an irradiance correction model based on micrometeorological data; establishing centralized and distributed non-residential PV power station power prediction models using a multi-output neural network based on the corrected irradiance data; establishing a PV output prediction model for small-sample scenarios using model transfer; estimating the predicted output value of distributed 220V residential PV power stations by calculating power generation relationship coefficients; and summing the predicted output values ​​of all PV power stations to obtain the county-level PV power prediction result. This invention significantly improves the prediction accuracy of PV output in county-level scenarios.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power output prediction technology, specifically to a method and system for predicting photovoltaic power in a distribution network. Background Technology

[0002] With the rapid advancement of new power system construction, the photovoltaic (PV) penetration rate within the system is increasing rapidly. In areas with high PV penetration, the local consumption of distributed power sources, the transmission of power clusters, and the reliability issues of traditional power grid supply are intertwined. To meet the high utilization rate of PV power generation and the flexible dispatching to satisfy grid demand, power system operators need PV output forecasting. Accurate PV output forecasting has become a crucial support for the reliable formulation of grid dispatching plans and the safe and stable operation of the power system.

[0003] Currently, most domestic and international research on photovoltaic power prediction focuses on the effectiveness and reliability of models, neglecting sufficient consideration of real-world scenarios. Firstly, most studies only consider a single power station or power stations within a very small area, ignoring the impact of spatial uncertainties on photovoltaic prediction in large-scale scenarios such as county-level areas. Long geographical distances inevitably lead to significant differences in meteorological conditions, potentially resulting in large errors between numerical weather prediction data and actual meteorological data. Secondly, for distributed 220V residential photovoltaic systems, photovoltaic output data is often unavailable, making it difficult to establish prediction models. Thirdly, for newly built photovoltaic power stations or power stations with missing data, insufficient data in small sample scenarios severely restricts the effectiveness and accuracy of prediction models. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for predicting photovoltaic power in distribution networks, which solves the problems of spatial uncertainty, lack of power data for household photovoltaic systems, and insufficient data volume in small sample scenarios at the county level mentioned above.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] Firstly, a method for predicting photovoltaic power in a distribution network is provided, including:

[0009] After preprocessing the collected sample data from the county, an original dataset was constructed, photovoltaic output prediction and evaluation indicators were proposed, and model parameters were initialized.

[0010] The characteristics of centralized photovoltaic power output within the county are analyzed, spatial feature vectors and cosine distances between feature vectors are extracted, and the county is divided into sub-regions.

[0011] Irradiance correction models for centralized and distributed non-residential photovoltaic power stations were established using sample data collected within the county. Based on the corrected irradiance data and the original dataset, a power prediction model for photovoltaic power stations was constructed.

[0012] Using existing photovoltaic power prediction models established for photovoltaic power plants that have been in operation for more than one year, we can assist in establishing a power prediction model for photovoltaic power plants in a small sample scenario.

[0013] The power generation relationship coefficient is calculated based on the power generation of centralized photovoltaic power stations and distributed 220V household photovoltaic power stations within the sub-region, and the power of distributed 220V household photovoltaic power stations is deduced from the power generation relationship coefficient.

[0014] The final photovoltaic power prediction result for the county is obtained by summing up all the predicted output values ​​of centralized and distributed photovoltaic power.

[0015] Preferably, the step of preprocessing the collected sample data within the county to construct the original dataset specifically involves:

[0016] The collected sample data within the county are sequentially processed by missing value imputation, feature extraction, and normalization to generate the original dataset for training the prediction model.

[0017] The collected sample data within the county includes centralized photovoltaic power stations, distributed non-residential photovoltaic power stations, and distributed 220V residential photovoltaic power stations. The types of sample data collected within the county include numerical weather forecast data, micro-meteorological measurement data of each power station, and historical power and electricity generation data.

[0018] Preferably, the photovoltaic power output prediction and evaluation index includes the mean absolute error E. mae Accuracy C R Among them, mean absolute error N is the number of samples in the dataset, y t Indicates the actual output of photovoltaic power. The predicted photovoltaic output value is represented by the accuracy rate. C t This represents the operating capacity at time t.

[0019] Preferably, the step of performing feature analysis on the output of each centralized photovoltaic power station within the county, extracting key data features, and dividing the county into sub-regions specifically includes:

[0020] The structural characteristics of the output data of centralized photovoltaic power stations in the region were analyzed by using the empirical orthogonal function decomposition method, and spatial feature vectors and cosine distances between feature vectors were extracted to explore spatial commonalities. The county was spatially divided into several sub-regions by using an agglomerative hierarchical clustering algorithm.

[0021] Preferably, the method of analyzing the structural characteristics of the output data of each centralized photovoltaic power station in the region using the empirical orthogonal function decomposition method specifically includes:

[0022] The output data of all centralized photovoltaic power plants are constructed into an original data matrix Y = [y1, y2, ..., y]. m ] T y m This represents a column vector consisting of all the outputs of the m-th photovoltaic power station during this data period;

[0023] The original data matrix is ​​processed into anisotropy form to obtain a new data matrix X. m×n Let n represent the length of the time series, and calculate its covariance matrix. S3. Calculate the eigenvalues ​​and eigenvectors of the square matrix C to obtain the spatial eigenvectors of the original data matrix.

[0024] Preferably, the step of spatially dividing the county into several sub-regions using an agglomerative hierarchical clustering algorithm specifically includes:

[0025] The spatial eigenvectors of the original data matrix Y are used as input, and the number of photovoltaic power stations m is taken as the number of initial classes. Each class contains a column vector of the eigenvector matrix.

[0026] Calculate the cosine distance between each pair of column vectors, calculate the inter-class distance using the minimum variance method, merge the two closest classes into a new class, and continue until the number of classes reaches the set number k, then draw the clustering graph.

[0027] Preferably, the step of establishing irradiance correction models for centralized and distributed non-residential photovoltaic power stations using collected sample data within the county, and constructing a photovoltaic power station power prediction model based on the corrected irradiance data and the original dataset, specifically includes:

[0028] The measured values ​​of irradiance micrometeorology and numerical weather forecasts are extracted from the original datasets of each power station. The measured values ​​of irradiance at the same time a day before the correction time, the measured values ​​of irradiance at time b before the correction time, and the numerical weather forecast values ​​of irradiance at the correction time are used as input features, and the measured values ​​of irradiance at the correction time are used as output targets. Based on this, the training dataset of the irradiance correction model is constructed.

[0029] An irradiance correction prediction model was constructed using the XGBoost method based on the training dataset of the correction model.

[0030] The irradiance numerical weather forecast value and other input features at the time to be predicted are input into the irradiance correction prediction model to obtain the irradiance correction value, which is then used to update the data column of the irradiance forecast value in the initial dataset of the photovoltaic power prediction model.

[0031] Based on the corrected initial dataset, a multi-output fully connected neural network is used as a predictor to train power prediction models for centralized photovoltaic and distributed non-residential photovoltaic systems.

[0032] Preferably, the step of using established photovoltaic power prediction models for photovoltaic power plants that have been in operation for one year or more to assist in establishing power prediction models for photovoltaic power plants in small sample scenarios includes the following steps:

[0033] For power plants with data shortages, select a power plant with the same voltage level and sufficient data in the same sub-region and extract the parameters of its established photovoltaic power prediction model.

[0034] The extracted model parameters are used as the initialization parameters of the target photovoltaic power station prediction model. Based on the small sample dataset of the target photovoltaic power station, a fully connected neural network is used as the predictor to train the target photovoltaic power station prediction model.

[0035] Repeat the above steps until all eligible power plants in the sub-region are traversed, thereby obtaining several power prediction models for the target photovoltaic power plant. The average output of all power prediction models is taken as the final prediction result.

[0036] Preferably, the calculation of the power generation relationship coefficient based on the power generation of centralized photovoltaic power stations and distributed 220V residential photovoltaic power stations within a sub-region, and the deduction of the power of distributed 220V residential photovoltaic power stations using the power generation relationship coefficient, specifically includes:

[0037] Consider any distributed 220V residential photovoltaic power station in any sub-region. Divide the historical power generation of any centralized photovoltaic power station by the power generation of the distributed 220V residential photovoltaic power station, and take the average over the time span to calculate the power generation relationship coefficient of the centralized photovoltaic power station. Continue to obtain the power generation relationship coefficients of all power generation in the sub-region, and take the average to obtain the final power generation relationship coefficient of the 220V residential photovoltaic power station.

[0038] The power generation relationship coefficient is used as the ratio coefficient between the photovoltaic power of centralized photovoltaic power plants and the photovoltaic power of distributed 220V household photovoltaic power plants. Based on the power prediction average of all centralized photovoltaic power plants in the sub-region, the output prediction value of the distributed 220V household photovoltaic power plant is derived by using the ratio coefficient.

[0039] Preferably, the step of summing up all the predicted output values ​​of centralized and distributed photovoltaic power to obtain the final predicted photovoltaic power result for the county includes:

[0040] For centralized photovoltaic and distributed non-residential photovoltaic power stations, if there is sufficient data, a power prediction model can be directly established; if there is insufficient data, a model transfer method can be used to assist in building a power prediction model. For distributed 220V residential photovoltaic power stations, the power output value of 220V residential photovoltaic power stations is derived by using the power generation relationship coefficient as the prediction result.

[0041] Secondly, a photovoltaic power prediction system for a distribution network is provided, comprising:

[0042] The preprocessing module is used to construct the original dataset from the collected sample data within the county, propose photovoltaic output prediction and evaluation indicators, and initialize model parameters;

[0043] The analysis module is used to perform feature analysis on the output of each centralized photovoltaic power station in the county, extract spatial feature vectors and cosine distance between feature vectors, and divide the county into sub-regions.

[0044] The model building module is used to establish irradiance correction models for centralized and distributed non-residential photovoltaic power stations by collecting sample data within the county. Based on the corrected irradiance data and the original dataset, it constructs a photovoltaic power station power prediction model. It is also used to utilize the photovoltaic power station power prediction models established by photovoltaic power stations that have been in operation for more than one year to assist in establishing a photovoltaic power prediction model for small sample scenarios.

[0045] The calculation module calculates the power generation relationship coefficient based on the power generation of centralized photovoltaic power stations and distributed 220V household photovoltaic power stations within the sub-region, and uses the power generation relationship coefficient to deduce the power of distributed 220V household photovoltaic power stations.

[0046] The processing module is used to sum all the predicted output values ​​of centralized and distributed photovoltaic power to obtain the final photovoltaic power prediction result for the county.

[0047] Furthermore, the objects of the collected sample data within the county include centralized photovoltaic power stations, distributed non-residential photovoltaic power stations, and distributed 220V residential photovoltaic power stations. The types of the collected sample data within the county include numerical weather forecast data, micro-meteorological measurement data of each power station, and historical power and electricity generation data.

[0048] (III) Beneficial Effects

[0049] (1) The present invention provides a photovoltaic power prediction method and system for distribution networks. In view of the problem that wide-area weather forecasts cannot effectively describe local meteorological information in the context of large-area scenarios at the county level, the XGBoost method is used to establish a mapping relationship between wide-area numerical weather forecast data and local micro-meteorological measured data, an irradiance correction model is established, and the forecast data column in the original dataset is corrected based on the model, which effectively improves the accuracy of irradiance forecast information and thus improves the quality of photovoltaic power prediction.

[0050] (2) This invention provides a method and system for predicting photovoltaic power in a distribution network. It utilizes the empirical orthogonal function decomposition method to extract features from the output values ​​of centralized photovoltaic power plants, and then uses an agglomerative hierarchical clustering algorithm to divide the space into sub-regions. The large-scale prediction task at the county level is decomposed into prediction tasks within each sub-region, mitigating the adverse effects of spatial uncertainty on photovoltaic power output prediction in county-level scenarios. Furthermore, to address the problem of insufficient sample size within each sub-region, a photovoltaic power prediction model for small-sample scenarios is constructed based on model transfer theory and the established photovoltaic prediction model. This method achieves accurate prediction of both centralized and distributed non-residential photovoltaic power at the county level, overcoming the adverse effects of insufficient sample size.

[0051] (3) The present invention provides a method and system for predicting photovoltaic power in a distribution network. The ratio of the power generation of a centralized photovoltaic power station to the power generation of a distributed 220V household photovoltaic power station is used as an approximate ratio coefficient of the power generation. Based on the predicted output value of the centralized photovoltaic power station, the power generation of the distributed 220V household photovoltaic power station is estimated as its predicted output value, which solves the problem that the photovoltaic power generation of the distributed 220V household photovoltaic power station cannot be collected. Attached Figure Description

[0052] Figure 1 This is a flowchart of the method of the present invention;

[0053] Figure 2 This is a diagram showing the results of correcting irradiance forecast values ​​based on meteorological information, as presented in this invention.

[0054] Figure 3 This is a graph showing the prediction results of the present invention 24 hours before the spring season;

[0055] Figure 4 This is a graph showing the prediction results of the present invention 24 hours before the summer solstice;

[0056] Figure 5 This is a graph showing the prediction results of the present invention 24 hours before the autumn equinox;

[0057] Figure 6 This is a graph showing the prediction results of the present invention for the 24 hours before winter. Detailed Implementation

[0058] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0059] Example

[0060] like Figure 1 As shown, one embodiment of the present invention provides a method for predicting photovoltaic power in a distribution network, comprising:

[0061] After preprocessing the collected sample data from the county, an original dataset was constructed, photovoltaic output prediction and evaluation indicators were proposed, and model parameters were initialized.

[0062] The characteristics of centralized photovoltaic power output within the county are analyzed, spatial feature vectors and cosine distances between feature vectors are extracted, and the county is divided into sub-regions.

[0063] Irradiance correction models for centralized and distributed non-residential photovoltaic power stations were established using sample data collected within the county. Based on the corrected irradiance data and the original dataset, a power prediction model for photovoltaic power stations was constructed.

[0064] Using existing photovoltaic power prediction models established for photovoltaic power plants that have been in operation for more than one year, we can assist in establishing a power prediction model for photovoltaic power plants in a small sample scenario.

[0065] The power generation relationship coefficient is calculated based on the power generation of centralized photovoltaic power stations and distributed 220V household photovoltaic power stations within the sub-region, and the power of distributed 220V household photovoltaic power stations is deduced from the power generation relationship coefficient.

[0066] The final photovoltaic power prediction result for the county is obtained by summing up all the predicted output values ​​of centralized and distributed photovoltaic power.

[0067] Among them, photovoltaic power plants that have been in operation for one year or more have sufficient and representative data.

[0068] The specific implementation is as follows:

[0069] S1. Collect data, preprocess data, and propose photovoltaic output prediction and evaluation indicators.

[0070] The collected sample data includes centralized photovoltaic (PV) power plants, distributed non-residential PV power plants, and distributed 220V residential PV power plants. The data types include numerical weather prediction data, micrometeorological data from each power plant, and historical power and generation data. Historical power data could not be collected for 220V residential PV power plants. In the data preprocessing stage, outlier removal and missing value imputation were performed first. Then, correlation analysis was used for feature extraction to determine the feature space. Next, data from 10 PM to 5 AM was removed because PV output was zero during this period. Finally, the input data and the prediction target were normalized using linear normalization to remove dimensions, thus forming the sample dataset needed for training. The evaluation index for PV output prediction includes the mean absolute error E. mae Accuracy C R Among them, the mean absolute error N is the number of samples in the dataset, y t Indicates the actual output of photovoltaic power. The predicted photovoltaic output value is represented by the accuracy rate. C t This represents the operating capacity at time t.

[0071] S2. Use clustering methods to divide spatial sub-regions.

[0072] The structural characteristics of the output data of centralized photovoltaic power stations within the region were analyzed using the empirical orthogonal function decomposition method, key data features were extracted, and spatial commonalities were identified. Subsequently, the county-level region was spatially divided into several sub-regions using an agglomerative hierarchical clustering algorithm. The specific steps are as follows:

[0073] (a) Obtaining the spatial eigenvectors of the output matrix of centralized photovoltaic power stations within the county using the empirical orthogonal decomposition method: First, construct the original data matrix Y = [y1, y2, ..., y] using the output data of all centralized photovoltaic power stations. m ] T y m This represents the column vector consisting of all the outputs of the m-th photovoltaic power station during this data period. Next, the original data matrix is ​​processed into anomaly form to obtain a new data matrix X. m×n Let n represent the length of the time series, and calculate its covariance matrix. Finally, the eigenvalues ​​and eigenvectors of the square matrix C are calculated to obtain the spatial eigenvectors of the original data matrix.

[0074] (b) The steps for dividing the spatial sub-regions using the agglomerative hierarchical clustering algorithm are as follows: First, the spatial feature vectors of the original data matrix Y are used as input, and the number of photovoltaic power stations m is taken as the initial number of classes. Each class contains one column vector of the feature vector matrix. Second, the cosine distance between each pair of column vectors is calculated, and the inter-class distance is calculated using the minimum variance method. The two classes with the closest distance are merged into a new class. Finally, the previous step is repeated until the number of classes is the set number k, and the clustering graph is drawn.

[0075] S3. Construct a correction and prediction model for irradiance forecast values.

[0076] Irradiance correction models for centralized and distributed non-residential photovoltaic (PV) power plants are established using regional meteorological forecast data and micrometeorological data. This method can obtain output prediction models for centralized PV and distributed non-residential PV, but does not include distributed 220V residential PV. The steps are as follows: First, extract the measured micrometeorological irradiance values ​​and numerical weather forecast values ​​from the original datasets of each power plant. Second, use the measured irradiance values ​​at the same time a days before the correction time, the measured irradiance values ​​at time b before the correction time, and the numerical weather forecast irradiance values ​​at the correction time as input features, and the measured irradiance value at the correction time as the output target value, thus constructing the training dataset for the irradiance correction model. Finally, based on the training dataset of the correction model, the XGBoost method is used to construct the irradiance correction prediction model.

[0077] S4. Establish power prediction models for centralized and distributed non-residential photovoltaic power plants.

[0078] Based on the corrected irradiance data and the original dataset, a photovoltaic power plant power prediction model is constructed. First, the irradiance numerical weather forecast value for the time to be predicted, along with other input features, are input into the irradiance correction prediction model to obtain the corrected irradiance value. Second, this corrected value is used to update the irradiance forecast data column in the initial dataset of the photovoltaic power prediction model. Finally, based on the corrected initial dataset, a multi-output fully connected neural network is used as the predictor to train power prediction models for centralized photovoltaic and distributed non-residential photovoltaic systems.

[0079] S5. Construct a power prediction model for photovoltaic power plants in a small sample scenario.

[0080] For centralized and distributed non-residential photovoltaic (PV) power plants facing data shortages, a model transfer approach is used. This approach leverages existing power prediction models from PV power plants with ample data to assist in building power prediction models for small-sample scenarios. The steps are as follows: First, for power plants with data shortages, select a power plant within the same sub-region with the same voltage level and sufficient data, and extract its existing PV power prediction model parameters. Second, use these extracted model parameters as the initialization parameters for the target PV power plant's prediction model. Based on the target PV power plant's small-sample dataset, a fully connected neural network is used as the predictor to train the target PV power plant's prediction model. Finally, repeat the first two steps until all eligible power plants within the sub-region are traversed. This yields several power prediction models for the target PV power plant, and the average output of all power prediction models is used as the final prediction result.

[0081] S6. Calculate the predicted power of distributed 220V residential photovoltaic systems.

[0082] The method for calculating the power output of a distributed 220V residential photovoltaic (PV) power station by reverse engineering the power generation coefficient is as follows: First, consider any distributed 220V residential PV power station in any sub-region. Divide the historical power generation of any centralized PV power station by the power generation of the distributed 220V residential PV power station, and take the average over the time span to calculate the power generation relationship coefficient for that centralized PV power station. Second, repeat step S1 until all power generation relationship coefficients in the sub-region are obtained, and take the average to obtain the final power generation relationship coefficient for the 220V residential PV power station. Finally, the power generation relationship coefficient is approximated as the ratio coefficient between the photovoltaic power of the centralized PV power station and the photovoltaic power of the distributed 220V residential PV power station. Based on the average power prediction of all centralized PV power stations in the sub-region, the predicted output value of the distributed 220V residential PV power station is approximately derived from this coefficient.

[0083] S7. Establish a photovoltaic power prediction model for county-level power distribution networks.

[0084] In summary, the centralized and distributed photovoltaic (PV) zones were scaled up to form a county-level PV prediction model. For centralized and distributed non-residential PV power stations, if sufficient data was available, a power prediction model was directly established; otherwise, a model transfer method was used to assist in building the model. For distributed 220V residential PV power stations, the power output value was derived by using the power generation relationship coefficient as the prediction result. The final county-level PV power prediction result was obtained by summing all the predicted output values ​​for centralized and distributed PV.

[0085] This study uses photovoltaic (PV) output data, micrometeorological data, and numerical weather prediction data from centralized and distributed PV power stations in a county in Anhui Province, China, from December 2019 to November 2021 as the data sources for the implementation case. The time resolution of the data is 15 minutes per point. In the data preprocessing stage, data from 10 PM to 5 AM was first removed because PV output was zero during this period. Secondly, correlation analysis was used to extract features and determine the feature space. Finally, the input data and the prediction target were normalized using linear normalization to remove dimensions, thus forming the sample dataset needed for training. Since the characteristics of PV output vary in different seasons, predictions were made seasonally.

[0086] The schematic diagram of the result obtained by the irradiance forecast correction method based on micrometeorological data proposed in this invention is shown below. Figure 2As shown in the figure, there is a significant deviation between the irradiance forecast values ​​in the wide-area numerical weather prediction data and the locally measured irradiance values ​​obtained from the power plant's micrometeorological equipment. Since the irradiance forecast value is a crucial characteristic for photovoltaic power prediction, this deviation inevitably affects the accuracy and effectiveness of the prediction model. The irradiance data corrected by the proposed method can reasonably fit the measured irradiance values. Therefore, the correction method based on micrometeorological information can effectively reduce the adverse effects of wide-area numerical weather prediction deviations on the prediction model.

[0087] The point prediction results obtained using the method disclosed in this invention are shown in Table 1.

[0088] Table 1. Point prediction and evaluation results for the 24 hours prior to the invention date under different seasons.

[0089]

[0090] Since the prediction model in this invention can output prediction results from 1 to 96 points after the prediction time, i.e., from 15 minutes to 24 hours, the prediction results for the 24 hours prior to the prediction time with the greatest uncertainty are shown in Table 1. The evaluation results show that this invention can achieve superior prediction results in different seasons. The mean absolute error index (Emae) shows that the prediction deviation is higher in summer and autumn than in spring and autumn, indicating that the photovoltaic power output prediction effect is better in the first two seasons. However, the prediction accuracy (CR) can be maintained above 90%, indicating that the method proposed in this invention can guarantee the accuracy and effectiveness of county-level photovoltaic power prediction. Figure 3-6 The forecast curves for different periods of each season are shown. As can be seen from the graphs, photovoltaic power output is relatively stable in spring and autumn, resulting in better forecast performance. In contrast, photovoltaic power generation exhibits greater fluctuations in summer and winter, leading to poorer forecast performance. Overall, the forecast curves effectively describe the fluctuation trends and characteristics of the actual power output, thus demonstrating the feasibility and effectiveness of the forecasting method proposed in this invention.

[0091] Another embodiment of the present invention provides a photovoltaic power prediction system for a distribution network, comprising:

[0092] The preprocessing module is used to construct the original dataset from the collected sample data within the county, propose photovoltaic output prediction and evaluation indicators, and initialize model parameters;

[0093] The analysis module is used to perform feature analysis on the output of each centralized photovoltaic power station in the county, extract spatial feature vectors and cosine distance between feature vectors, and divide the county into sub-regions.

[0094] The model building module is used to establish irradiance correction models for centralized and distributed non-residential photovoltaic power stations by collecting sample data within the county. Based on the corrected irradiance data and the original dataset, it constructs a photovoltaic power station power prediction model. It is also used to utilize the photovoltaic power station power prediction models established by photovoltaic power stations that have been in operation for more than one year to assist in establishing a photovoltaic power prediction model for small sample scenarios.

[0095] The calculation module calculates the power generation relationship coefficient based on the power generation of centralized photovoltaic power stations and distributed 220V household photovoltaic power stations within the sub-region, and uses the power generation relationship coefficient to deduce the power of distributed 220V household photovoltaic power stations.

[0096] The processing module is used to sum all the predicted output values ​​of centralized and distributed photovoltaic power to obtain the final photovoltaic power prediction result for the county.

[0097] The collected sample data within the county includes centralized photovoltaic power stations, distributed non-residential photovoltaic power stations, and distributed 220V residential photovoltaic power stations. The types of sample data collected within the county include numerical weather forecast data, micro-meteorological measurement data of each power station, and historical power and electricity generation data.

[0098] 10. Embodiments of the present invention may be provided as methods or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0099] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for predicting photovoltaic power in a distribution network, characterized in that, include: The original dataset was constructed from the collected sample data within the county, and photovoltaic output prediction and evaluation indicators were proposed and model parameters were initialized. The characteristics of centralized photovoltaic power output within the county are analyzed, spatial feature vectors and cosine distances between feature vectors are extracted, and the county is divided into sub-regions. Irradiance correction models for centralized and distributed non-residential photovoltaic power stations were established using sample data collected within the county. Based on the corrected irradiance data and the original dataset, a power prediction model for photovoltaic power stations was constructed. Using existing photovoltaic power prediction models established for photovoltaic power plants that have been in operation for more than one year, we can assist in establishing a power prediction model for photovoltaic power plants in small sample scenarios. The power generation relationship coefficient is calculated based on the power generation of centralized photovoltaic power stations and distributed 220V household photovoltaic power stations within the sub-region. The power of the distributed 220V household photovoltaic power stations is then deduced from the power generation relationship coefficient. The final photovoltaic power prediction result for the county is obtained by summing up all the predicted output values ​​of centralized and distributed photovoltaic power. The method of using established photovoltaic power prediction models for photovoltaic power plants that have been in operation for more than one year to assist in establishing a power prediction model for photovoltaic power plants under small sample scenarios includes the following steps: For power plants with data shortages, select a power plant with the same voltage level and sufficient data in the same sub-region and extract the parameters of its established photovoltaic power prediction model. The extracted model parameters are used as the initialization parameters of the target photovoltaic power station prediction model. Based on the small sample dataset of the target photovoltaic power station, a fully connected neural network is used as the predictor to train the target photovoltaic power station prediction model. Repeat the above steps until all eligible power plants in the sub-region are traversed, thereby obtaining several power prediction models for the target photovoltaic power plant. The average output of all power prediction models is taken as the final prediction result.

2. The method for predicting photovoltaic power in a distribution network according to claim 1, characterized in that: The process of constructing the original dataset from the collected sample data within the county specifically includes: The collected sample data within the county are sequentially processed by missing value imputation, feature extraction, and normalization to generate the original dataset for training the prediction model; The collected sample data within the county includes centralized photovoltaic power stations, distributed non-residential photovoltaic power stations, and distributed 220V residential photovoltaic power stations. The types of sample data collected within the county include numerical weather forecast data, micro-meteorological measurement data of each power station, and historical power and electricity generation data.

3. The method for predicting photovoltaic power in a distribution network according to claim 2, characterized in that: The photovoltaic power output prediction and evaluation indicators include the mean absolute error E. mae Accuracy C R ; Among them, mean absolute error N is the number of samples in the dataset. Indicates the actual output of photovoltaic power. The predicted photovoltaic output value is represented by the accuracy rate. C t This represents the operating capacity at time t.

4. The method for predicting photovoltaic power in a distribution network according to claim 1, characterized in that: The process of performing feature analysis on the centralized photovoltaic power output within the county, extracting key data features, and dividing the county into sub-regions specifically includes: The structural characteristics of the output data of centralized photovoltaic power stations in the region were analyzed by using the empirical orthogonal function decomposition method, and spatial feature vectors and cosine distances between feature vectors were extracted to explore spatial commonalities. The county was spatially divided into several sub-regions by using an agglomerative hierarchical clustering algorithm.

5. The method for predicting photovoltaic power in a distribution network according to claim 4, characterized in that: The structural characteristics of the power output data of each centralized photovoltaic power station in the region, analyzed using the empirical orthogonal function decomposition method, specifically include: The output data of all centralized photovoltaic power plants are constructed into an original data matrix Y = [y1, y2, ..., y]. m ] T y m This represents a column vector consisting of all the output power of the m-th photovoltaic power station during this data period; The original data matrix is ​​processed into anisotropy form to obtain a new data matrix X. m×n Let n represent the length of the time series, and calculate its covariance matrix. Finally, the eigenvalues ​​and eigenvectors of the square matrix C are calculated to obtain the spatial eigenvectors of the original data matrix.

6. The photovoltaic power prediction method for a distribution network according to claim 4, characterized in that: The method of spatially dividing the county into several sub-regions using an agglomerative hierarchical clustering algorithm specifically includes: The spatial eigenvectors of the original data matrix Y are used as input, and the number of photovoltaic power stations m is taken as the number of initial classes. Each class contains a column vector of the eigenvector matrix. Calculate the cosine distance between each pair of column vectors, calculate the inter-class distance using the minimum variance method, merge the two closest classes into a new class, and continue until the number of classes reaches the set number k, then draw the clustering graph.

7. The method for predicting photovoltaic power in a distribution network according to claim 6, characterized in that: The process involves establishing irradiance correction models for centralized and distributed non-residential photovoltaic power stations using collected sample data from within the county. Based on the corrected irradiance data and the original dataset, a power prediction model for the photovoltaic power station is constructed, specifically including: The measured values ​​of irradiance micrometeorology and numerical weather forecasts are extracted from the original datasets of each power station. The measured values ​​of irradiance at the same time a day before the correction time, the measured values ​​of irradiance at time b before the correction time, and the numerical weather forecast values ​​of irradiance at the correction time are used as input features, and the measured values ​​of irradiance at the correction time are used as output targets. Based on this, the training dataset of the irradiance correction model is constructed. An irradiance correction prediction model was constructed using the XGBoost method based on the training dataset of the correction model. The irradiance numerical weather forecast value and other input features at the time to be predicted are input into the irradiance correction prediction model to obtain the irradiance correction value, which is then used to update the data column of the irradiance forecast value in the initial dataset of the photovoltaic power prediction model. Based on the corrected initial dataset, a multi-output fully connected neural network is used as a predictor to train power prediction models for centralized photovoltaic and distributed non-residential photovoltaic systems.

8. A photovoltaic power prediction system for a power distribution network, characterized in that, include: The preprocessing module is used to construct the original dataset from the collected sample data within the county, propose photovoltaic output prediction and evaluation indicators, and initialize model parameters; The analysis module is used to perform feature analysis on the output of each centralized photovoltaic power station in the county, extract spatial feature vectors and cosine distance between feature vectors, and divide the county into sub-regions. The model building module is used to establish irradiance correction models for centralized and distributed non-residential photovoltaic power stations by collecting sample data within the county. Based on the corrected irradiance data and the original dataset, it constructs a photovoltaic power station power prediction model. It is also used to utilize the photovoltaic power station power prediction models established by photovoltaic power stations that have been in operation for more than one year to assist in establishing a photovoltaic power prediction model for small sample scenarios. The calculation module calculates the power generation relationship coefficient based on the power generation of centralized photovoltaic power stations and distributed 220V household photovoltaic power stations within the sub-region, and uses the power generation relationship coefficient to deduce the power of distributed 220V household photovoltaic power stations. The processing module is used to sum all the predicted output values ​​of centralized and distributed photovoltaic power to obtain the final photovoltaic power prediction result for the county. The step of using established photovoltaic power prediction models for photovoltaic power plants that have been in operation for one year or more to assist in establishing a power prediction model for photovoltaic power plants in a small sample scenario includes the following steps: For power plants with data shortages, select a power plant with the same voltage level and sufficient data in the same sub-region and extract the parameters of its established photovoltaic power prediction model. The extracted model parameters are used as the initialization parameters of the target photovoltaic power station prediction model. Based on the small sample dataset of the target photovoltaic power station, a fully connected neural network is used as the predictor to train the target photovoltaic power station prediction model. Repeat the above steps until all eligible power plants in the sub-region are traversed, thereby obtaining several power prediction models for the target photovoltaic power plant. The average output of all power prediction models is taken as the final prediction result.

9. A photovoltaic power prediction system for a distribution network according to claim 8, characterized in that: The collected sample data within the county includes centralized photovoltaic power stations, distributed non-residential photovoltaic power stations, and distributed 220V residential photovoltaic power stations. The types of sample data collected within the county include numerical weather forecast data, micro-meteorological measurement data of each power station, and historical power and electricity generation data.

Images