Research On Residential Power Load Optimization Based On Hybrid Long-short Memory Neural Network

With the advancement of the energy Internet,the proportion of electric energy in the final energy consumption is increasing,and the peak load of the power grid is increasing year by year.In 2020,the peak load of the national power grid will reach 875 million kilowatts.The time-period demand pattern of residential electricity consumption makes the peak-valley difference of electricity consumption widened,and the peak load is high and short-lived.Especially the large-scale centralized activation of high-power home appliances such as air conditioners and water heaters on the user side makes the electricity load increase instantaneously and the load curve peaks.The phenomenon of seasonal and regional peak power supply shortages,which is dominated by residential electricity consumption,has become a common problem throughout the country.Residents lateral load accurate prediction for the operation of the electric power demand response and the response of incentives set up to provide strong data support,and large power appliances such as air conditioner,electric water heater load is one of the important objects of demand response control,accurate short-term load forecasting results to evaluate residents adjustable load potential of demand side response,aggregation,residential electricity optimization control strategy.Firstly,this thesis explains the current business status of residential demand response and the theoretical research related to load forecasting at home and abroad.The characteristics of high operating power of adjustable load and relatively stable over a period of time in demand response are analyzed.We proposed a Non-intrusive Load Monitoring architecture based on event detection,which classifies and aggregates diversified home appliance load resources,and decomposes the load characteristics of high power consumption appliances.Secondly,this paper proposed an SSA-LSTM model for load forecasting of high-power home appliances based on deep learning.The model considers residential environmental factors and holiday factors,and uses the long-term learning ability of the LSTM network to predict the high-power consumption load of time series data.On this basis,due to the excellent population optimization capability of sparrow search algorithm,the window size,batch size and the number of hidden layer units in the LSTM network were optimized by taking the load prediction value and the MAE of sample data as the objective function,and the optimal load prediction network model was obtained.Finally,this paper uses the air-conditioning operating data of a family in the United States for four months in 2016 to conduct a large number of simulation experiments to verify that the SSA-LSTM model can predict the trend of the load curve very well.Compared with the traditional deep learning model,the network structure optimized by the SSA algorithm has superior effectiveness and stability.