Short-Term Net Load Forecasting Of Bus Including Distributed Photovoltaic Based On Deep Learning

Power load forecasting is an indispensable and important task in the power system.With the large-scale distributed photovoltaics into the distribution network,the net load measured on the bus includes the electricity load and distributed photovoltaic output within the jurisdiction.To fully coordinate the flexible consumption of distributed photovoltaics and ensure the balance of power between sources and loads,it is necessary to accurately forecasting the net load of the bus containing distributed photovoltaics.To reasonably and accurately forecasting the short-term net load of the bus,this paper structures an indirect net load forecasting model for bus based on the different characteristics of load demand and distributed photovoltaic power.This model consists of the bus load demand forecasting model and a distributed photovoltaic power forecasting model in the jurisdiction.Firstly,a load forecasting model based on variational mode decomposition and multi branch gated convolutional neural network is proposed to address the obvious periodicity and nonlinearity of bus load.To reduce the interference of noise on forecasting accuracy,the variational modal decomposition method is used to decompose the original load;To fully utilize the long and short time dependence of time series,extract the weekly,daily,and nearest neighbor cycle features of load data;In order to reduce the dimension of subsequences and avoid overfitting in prediction,the periodic features extracted from each subsequence are reconstructed into a two-dimensional feature dataset;Finally,input the dataset into the multi branch gated convolutional neural network model to forecasting the load demand of the bus.Secondly,in response to the strong volatility of distributed photovoltaic power caused by meteorological factors,a dual attention and gated convolutional neural network distributed photovoltaic power prediction model considering meteorological factors is constructed.To select strong correlation vectors as inputs,Pearson principle is used to analyze the degree of correlation between photovoltaic output and meteorological factors;Using feature attention to calculate meteorological feature weights and explore the relationship between meteorological factors and photovoltaic power changes;Introducing temporal attention to further extract critical moment information from predicted value information,ultimately obtaining distributed photovoltaic power prediction results in the busbar area.Finally,the bus load demand forecasting model and a distributed photovoltaic power forecasting model are combined to build an indirect bus net load forecasting model.This model fully considers the load change characteristics and distributed photovoltaic output characteristics for net load prediction.Finally,through numerical simulation and comparative experiments,the effectiveness of the proposed short-term net load forecasting model for bus with distributed photovoltaics is demonstrated.