Urban Electricity Demand Forecasting with a Hybrid Machine Learning Model

作者：Ligang Hou, Xin Shen, Liqi Zhang · 年份：2023 · DOI：10.1109/icnsc58704.2023.10319027 · 被引用次数：10 · 研究领域：Energy Load and Power Forecasting、Forecasting Techniques and Applications、Image and Signal Denoising Methods

Urban electricity demand forecasting in the mid- and long-term with high accuracy is vital for the power systems’ operation and planning. Many factors of uncertainties and non-linearities, in addition to economic trends and seasonal cycles, should be taken into account during forecasting. This study presents a hybrid machine learning model, i.e., ARIMA-LSTM (ARLS), for accurate mid- and long-term urban electricity demand forecasting. The model adopts the following three-fold ideas: a) incorporating various factors that affect electricity load, such as economic trends and seasonal cycles, and then selecting important features from them; b) employing an autoregressive integrated moving average (ARIMA) model to decompose the time series into level, trend, and seasonal components, and a long short-term memory (LSTM) network to account for the relevant non-linear features; and c) fusing the outputs of each learning module by using ensemble mechanism to enhance the forecast accuracy. By doing so, it combines the advantages of both traditional and ML approaches for time-series analysis. The model is tested on the monthly electricity consumption data of two Chinese cities, and the results indicate that it surpasses the existing state-of-the-art models in terms of performance.