CN-121983950-A - Mountain area distribution network distributed photovoltaic prediction method, system, equipment and medium considering prediction error

CN121983950ACN 121983950 ACN121983950 ACN 121983950ACN-121983950-A

Abstract

The invention relates to the technical field of power prediction of new energy sources of electric power systems, and discloses a mountain area distribution network distributed photovoltaic prediction method, a system, equipment and a medium for taking prediction errors into account, wherein the method comprises the steps of firstly collecting meteorological data, historical output data and topographic data of a distributed photovoltaic field station with meteorological observation capability; based on multi-source data, a dynamic time warping and dynamic self-organizing mapping algorithm is adopted to cluster stations, regions similar to mountainous climate are divided, a photovoltaic output prediction model integrating a long-period memory network and a time convolution network is built for each region, a preliminary prediction value is generated, a conditional error probability distribution model is built by using a prediction error sequence on the basis, and finally correction prediction is carried out on all stations by combining the prediction model and the error model, so that a cluster output result is output. The method effectively improves the prediction precision and reliability of the mountain area under the high-proportion distributed photovoltaic access scene, and is particularly suitable for stations lacking local meteorological data.

Inventors

ZHENG YOUZHUO
ZUO HONGYU
YAN JIANGFENG
ZHANG WANCHENG
ZHANG HENGRONG
CHEN KAILEI
LUO CHAOYI
LIU ANJIANG
LI YUE
HAO SHUQING
HUA LONG
MIAO YU
WENG DI
XU YUTAO
DOU CHEN

Assignees

贵州电网有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251218

Claims (10)

1. The mountain area distribution network distributed photovoltaic prediction method considering the prediction error is characterized by comprising the following steps of: Acquiring multi-source data of a plurality of distributed photovoltaic field stations in a mountain area distribution network, wherein the multi-source data comprise meteorological data, photovoltaic historical output data and topographic data; Carrying out regional clustering on the distributed photovoltaic field stations based on multi-source data, and dividing the distributed photovoltaic field stations to obtain a plurality of mountain climate similar regions; building a photovoltaic output prediction model aiming at each mountain climate similar area, and generating preliminary prediction values of all stations by using the corresponding model; constructing a conditional error model based on the deviation between the preliminary predicted value and the actual output value; And combining the photovoltaic output prediction model and the conditional error model, predicting and correcting the output of each distributed photovoltaic field station, and outputting the prediction result of the distributed photovoltaic clusters of the mountain area distribution network.
2. The method for predicting distributed photovoltaics of a mountain area distribution network, which takes prediction errors into account, according to claim 1, wherein the regional clustering of the distributed photovoltaics stations based on the multi-source data comprises: calculating meteorological time sequence distances of any two distributed photovoltaic field stations by adopting a dynamic time warping algorithm; The meteorological data and the topographic data are fused and then input into a dynamic self-organizing map network, and station clustering is achieved through competition learning; And dividing the distributed photovoltaic field station into mountain climate similar areas according to the meteorological time sequence distance and the self-organizing map clustering result.
3. The method for predicting distributed photovoltaic in a mountain area distribution network, wherein the predicting model of photovoltaic output is constructed according to claim 2, and the method comprises the following steps: the long-term memory neural network and the time convolution neural network are deployed in parallel, and feature extraction is carried out on the historical multi-source data sequences respectively; the time sequence characteristics output by the long-term memory neural network and the multi-scale characteristics output by the time convolution neural network are weighted and fused through an attention mechanism to form a comprehensive characteristic representation; a preliminary predicted value of the photovoltaic output for a future period is generated based on the composite characteristic representation.
4. A mountain area distribution network distributed photovoltaic prediction method taking account of prediction errors as claimed in claim 3, wherein said constructing a conditional error model comprises: Calculating a prediction error sequence between a historical preliminary predicted value and an actual output value of each distributed photovoltaic station in each mountain climate similar area; And taking meteorological data, topographic data and historical output data at corresponding moments as conditional variables, and establishing a conditional probability distribution model of the prediction error.
5. The method for distributed photovoltaic prediction of a mountain area distribution network, which accounts for prediction errors, as set forth in claim 4, wherein the conditional probability distribution model is a mixed gaussian model, and the number of components, weights, means and variances of the mixed gaussian model are determined by training data.
6. The method for predicting distributed photovoltaic in a mountain area distribution network with prediction error taken into account as claimed in claim 5, wherein predicting and error correcting the output of each distributed photovoltaic field station comprises: generating a single-station preliminary predicted value by using a photovoltaic output prediction model; extracting an error correction amount corresponding to a preset quantile from the conditional error model based on the currently input multi-source data; The error correction is added to the preliminary predicted value to obtain a corrected single-station predicted value; and aggregating all corrected single-station predicted values in the same mountain climate similar region to obtain a region cluster predicted result.
7. A method of distributed photovoltaic prediction of a mountain area distribution network taking into account prediction errors as defined in claim 6, further comprising: For a distributed photovoltaic field station without meteorological observation equipment, attributing the distributed photovoltaic field station to a mountain area climate similar area with meteorological features closest to the topographic features according to topographic data and geographic positions of the distributed photovoltaic field station; and using the photovoltaic output prediction model and the conditional error model corresponding to the mountain climate similar region to complete prediction and correction.
8. A mountain power distribution network distributed photovoltaic prediction system taking prediction errors into account, applying the method as claimed in any one of claims 1 to 7, comprising: The data acquisition module is used for acquiring multi-source data of a plurality of distributed photovoltaic stations in the mountain area distribution network, wherein the multi-source data comprises meteorological data, photovoltaic historical output data and topographic data; The division module is used for carrying out region clustering on the distributed photovoltaic field stations based on the multi-source data, and dividing the distributed photovoltaic field stations into a plurality of mountain climate similar regions; The preliminary prediction module is used for constructing a photovoltaic output prediction model aiming at each mountain climate similar area and generating preliminary prediction values of all stations by utilizing the corresponding model; the error acquisition module is used for constructing a conditional error model based on the deviation between the preliminary predicted value and the actual output value; And the correction prediction module is used for predicting the output of each distributed photovoltaic field station and correcting errors by combining the photovoltaic output prediction model and the conditional error model, and outputting the prediction result of the distributed photovoltaic clusters of the power distribution network in the mountain area.
9. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of a mountain area distribution network distributed photovoltaic prediction method according to any one of claims 1-7, wherein the prediction error is accounted for.
10. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the steps of a mountain area distribution network distributed photovoltaic prediction method according to any one of claims 1 to 7, taking into account prediction errors.

Description

Mountain area distribution network distributed photovoltaic prediction method, system, equipment and medium considering prediction error Technical Field The invention relates to the technical field of new energy power prediction of electric power systems, in particular to a mountain area distribution network distributed photovoltaic prediction method, system, equipment and medium considering prediction errors. Background Under the background of a novel power system, a large number of new energy types such as distributed photovoltaic and the like are connected into a power distribution side, so that the power distribution network is gradually changed into electric energy bidirectional flow from the traditional electric energy downstream. Because the distributed photovoltaic output has the characteristics of high volatility, strong randomness, space unbalance and the like, the distributed photovoltaic can influence the quality of the power distribution network voltage and the safety boundary of operation and maintenance scheduling when being counted into a power distribution side. Particularly, for provinces mainly formed on a plateau and a mountain area, after the traditional mountain area distribution network used by the provinces is connected with high-proportion distributed photovoltaic, typical conditions such as reverse power flow overload and power quality reduction can occur, so that the problem to be solved urgently is predicted how to perform accurate and reliable distributed photovoltaic output. At the current stage, the method for forecasting and researching the photovoltaic output mainly comprises a physical modeling method and a data driving method, wherein the physical modeling method has higher dependence on irradiance, temperature and component parameters, and the depth characteristics of meteorological data and historical photovoltaic power generation can be accurately captured for the data driving method represented by a LSTM, TCN, GRU-based neural network, so that the effect of improving the photovoltaic output forecasting precision is achieved. However, in the case of a distributed photovoltaic scene facing a mountain power distribution network, due to the lack of meteorological data of part of the distributed photovoltaic stations, it is difficult to directly apply a neural network model considering the meteorological data. On the other hand, most of existing prediction models focus on single stations or small-range scenes, lack of distributed photovoltaic prediction conditions oriented to large-scale large-range scenes, and are not enough in processing and correcting capabilities of prediction errors. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a mountain area distribution network distributed photovoltaic prediction method, a system, equipment and a medium for taking prediction errors into account, which can solve the problems that in the prior art, the mountain area distribution network distributed photovoltaic prediction is difficult to apply a neural network model due to lack of meteorological data, a large-scale scene prediction method is lacked, and cluster result errors are large due to lack of prediction errors are not considered. In order to solve the technical problems, the invention provides the following technical scheme: In a first aspect, the present invention provides a method for predicting distributed photovoltaic in a mountain area distribution network, which includes: Acquiring multi-source data of a plurality of distributed photovoltaic field stations in a mountain area distribution network, wherein the multi-source data comprise meteorological data, photovoltaic historical output data and topographic data; Carrying out regional clustering on the distributed photovoltaic field stations based on multi-source data, and dividing the distributed photovoltaic field stations to obtain a plurality of mountain climate similar regions; building a photovoltaic output prediction model aiming at each mountain climate similar area, and generating preliminary prediction values of all stations by using the corresponding model; constructing a conditional error model based on the deviation between the preliminary predicted value and the actual output value; And combining the photovoltaic output prediction model and the conditional error model, predicting and correcting the output of each distributed photovoltaic field station, and outputting the prediction result of the distributed photovoltaic clusters of the mountain area distribution network. The mountain area distribution network distributed photovoltaic prediction method considering the prediction error is used as a preferable scheme, wherein the regional clustering of the distributed photovoltaic field stations based on the multi-source data comprises the following steps: calculating meteorological time sequence