Short-mid term electricity consumption prediction using non-intrusive attention-augmented deep learning model

Abstract

Estimates of electricity consumption (EC) can provide effective guidance for energy allocation and energy-saving measures. For improving the accuracy of short-mid term EC forecasting, a novel nonintrusive attention-augmented deep learning model is proposed. This model, which is referred to as NAP-BiLSTM, consists of a nonintrusive attention-augmented-based preprocessing (NAP) block and a regular bidirectional long short-term memory (BiLSTM). The NAP block is a kind of plug-and-play attention mechanism for time series, which can be combined with a deep neural model that processes sequential data without modifying its structure. The effectiveness of the proposed model is verified by two experiments, including univariate prediction using EC data from the U.S. and multivariate prediction using weather and energy data from Valencia. The root mean square error, mean absolute error, and mean absolute percentage error are used as metrics. The results demonstrate that the proposed model is highly promising for both univariate and multivariate analyses and outperforms state-of-the-art deep learning models.

Keywords

• Non-intrusive attention mechanism
• Preprocessing block
• Electric consumption
• BiLSTM
• Short-mid term prediction