Regional photovoltaic power generation prediction method based on federated learning and deep neural network

Abstract

The invention relates to a photovoltaic power generation prediction method, and aims to provide a regional photovoltaic power generation prediction method based on federated learning and a deep neural network. Comprising the following steps: respectively establishing neural network models at each photovoltaic prediction point and a cloud server based on a federated learning framework, performing classification according to a Pearson's correlation coefficient between a meteorological variable of each photovoltaic prediction point and photovoltaic generating capacity, and performing neural network initialization corresponding to all classifications at a cloud; local data are utilized to train each photovoltaic prediction point model and aggregate at the cloud, and only model parameters instead of training data are transmitted in the process. According to the method, large-scale data transmission in the communication process can be avoided, the communication time consumption and the bandwidth required by communication are remarkably reduced, and the model training efficiency and economy are improved; the problems of insufficient computing power, insufficient data and the like of a photovoltaic prediction point edge computing device in a traditional model training method are avoided; the classification algorithm enables the photovoltaic prediction points to be better matched with the prediction model, and improves the prediction precision of the generating capacity.

Description

technical field [0001] The invention relates to a photovoltaic power generation prediction method of a power system, which belongs to a regional classification photovoltaic power generation collaborative prediction method based on federated learning. Background technique [0002] In order to achieve carbon neutrality goals, the utilization of renewable energy has received increasing attention. Among them, solar energy has proven to be one of the cleanest and most abundant energy sources. Using solar energy for photovoltaic power generation can reduce the use of fossil fuels and reduce carbon emissions. However, the volatility of solar energy brings great challenges to the formulation of photovoltaic power generation plans, and accurate prediction of photovoltaic power generation output is of great significance to ensure the stable operation of the power grid. [0003] Traditional photovoltaic power generation power prediction methods can be divided into physical methods an...

AI Technical Summary

Problems solved by technology

However, the photovoltaic power of a single point is often greatly affected by fluctuating meteorological factors, and the result is quite different from the actual power generation.

At the same time, the neural network prediction method that has attracted much attention in recent years has improved the accuracy of photovoltaic prediction, but training the neural network requires a large computing power, and the cost of building a computing device for each point that needs to be predicted is too high

At the same time, there is a certain "data island" problem, and the meteorological data and photovoltaic power generation data in different regions cannot communicate with each other, resulting in the data not being fully utilized

[0006] Some researchers have also proposed that merging the data of each point to the central server for unified model training will lead to data leakage and privacy leakage in the process of data transmission in various places. Therefore, direct remote transmission and merging of original data cannot effectively solve the problem

Images