Research On Forecasting Method Of Short-term Electricity Consumption Based On Smart Electricity Meter

Electric energy is an important symbol of modern industry and a necessity of people’s life.In recent years,due to industrial upgrading,coupled with the impact of high temperature in summer and cold in winter,the demand for electricity continues to surge,resulting in a large power gap.It is difficult to store electric energy resources,so reasonable dispatching of electric energy resources becomes an important way to make full use of electric energy resources and avoid waste of electric energy resources.Load forecasting is the key method to realize the rational dispatching of power resources.Improving the accuracy and stability of load forecasting model can provide decision making basis for power dispatching departments.For short-term local power system,it is very difficult to establish accurate load prediction model,so improving the accuracy of short-term load prediction is an important problem to be solved urgently at present.In this paper,the power data collected by smart meters ar e taken as samples,and two different ways are used to construct load forecasting models.The research work of this paper is mainly divided into the following three aspects:1、Power load data contains information of other related factors,which is complicated and interferes with each other.In order to avoid interference between information influence the effect of forecast model,this article is based on " components to predict + sequence decomposition reconstruction" thoughts,come up with using the integrated empirical mode decomposition（EEMD）algorithm for power load curve decomposition,access to power load under different frequency of intrinsic mode function（IMF）,the decomposition of the IMF,Limit gradient enhancement model（XGBoost）was constructed for prediction,and the XGBoost model of each component was reconstructed and superimposed to obtain the final EEMD-XGBoost prediction model.2、For the traditional time series data of power load,the classical time series model ARIMA is adopted.Based on the limitations of box-Jenkins,AIC and BIC grading methods of ARIMA model,a method of optimizing ARIMA model parameters based on particle swarm optimization（PSO）is proposed.The fitness function and related parameters were set by population optimization method,and the order values of ARIMA model were determined.3、Based on the idea of "sequence prediction + residual correction",the ARIMA-LSTM model based on particle swarm optimization is proposed.ARIMA model is composed of a linear autoregressive equation and a residual moving average equation.ARIMA model has a good effect on fitting linear data,and the data that cannot be fitted is noise.ARIMA model extracts linear information and part of noise information from power load data,and there is also a lot of nonlinear information.LSTM model has good fitting ability to nonlinear data,and the final prediction model is formed by adding the predicted values of ARIMA model and LSTM.Finally,PSO-ARIMA-LSTM prediction model is superior to EEMD-XGBoost prediction model in terms of stability and accuracy.