Market Oriented Trading Electricity Forecasting Method Based on Tensor ARIMA

A tensor ARIMA prediction method based on feature reconstruction is studied to address the characteristics of short length and limited sample data in market-oriented trading monthly electricity quantity sequences. Firstly, the curve correlation between market-oriented transaction electricity and macroeconomic data such as regional gross domestic product and total retail sales of consumer goods was analyzed. Then, marketoriented transaction electricity and economic data with high correlation were used as input matrices, and block Hankel tensor ARIMA technology was used to jointly predict multiple time series data with correlation in the matrix. The proposed method was validated based on real direct transaction data and economic data. The experimental results show that the proposed method can effectively screen economic data that is theoretically related to market-oriented trading electricity, reveal and quantify the inherent relationship between market-oriented electricity data and economic data, and can accurately predict market-oriented electricity with a small sample size, greatly improving the accuracy of electricity prediction.