Research On Load Forecasting Method Of High Permeability Distributed Photovoltaic Distribution Network Based On Deep Learning

The distribution network load forecasting is an important link in the safe operation and dispatching control of the power grid,and it is o f great significance to the power grid planning.With the large number of new energy sources connected to the grid,the penetration rate of distributed photovoltaics in the distribution network is gradually increasing.The distributed photovoltaic system i s generally installed after the electric meter,and its output is invisible,which has a greater impact on the net load forecast of the distribution network,and also brings challenges to the operation of the grid.Due to the coupling of the intermittent nature of photovoltaic power and the volatility of actual load,it is more difficult to predict the net load of the distribution network.In response to the above problems,this article aims to avoid the installation of expensive metering facilities and use as few known conditions as possible to decouple the photovoltaic power of unmonitored distributed photovoltaic system users from the net load to improve the accuracy of net load prediction when the distributed photovoltaic system exists.The method proposed in this paper is divided into two stages.The first stage is net load decoupling: firstly,identify whether the user has installed a distributed photovoltaic system,and then the photovoltaic power of a small number of known installed distributed photovoltaic users and the actual load of the identified non-distributed photovoltaic users are used as feature vectors to construct a "photovoltaic-load" decoupling model.The second stage is net load forecasting: the proposed prediction model based on frequency domain decomposition and deep learning is used to predict the historical photovoltaic power and actual load obtained by decoupling respectively.The prediction model first obtains the amplitude and phase of each frequency sine wave by fast Fourier decomposition of the original quantity,and then analyzes the correlation between the decomposed components and the original data to determine the frequency demarcation point of the frequency domain decomposition.The low-frequency component and high-frequency component are predicted,and the final prediction result is obtained through additive reconstruction.A case study using Sydney’s actual data set shows that the method proposed in this paper has good performance in terms of distributed photovoltaic system identification,net load decoupling and forecasting of the distribution network.