No abstract

The forecast of electricity consumption is of great significance to adjusting the power supply dispatching scheme and optimizing economic structure. Electricity consumption prediction is a time series prediction problem in essence. Most relevant work considers establishing electricity consumption prediction models in terms of economy, temperature, region, etc. However, few studies consider the three data characteristics of trend, seasonality, and periodicity contained in time series. Therefore, this paper proposes a Long-Short Term Memory model based on time series feature fusion. This model considers the influence of three data features of time series on time prediction results and can effectively integrate time-series features into Long Short-Term Memory Model with strong self-learning ability. Experimental results show that the proposed method has higher accuracy than the basic LSTM model and the Autoregressive Integrated Moving Average Model (ARIMA).