Ultra-Short- Term Load Forecasting Based on Convolutional-LSTM Hybrid Networks

Abstract

The big data era in energy systems is coming with the popularity of advanced metering architectures such as smart meters, providing much more data sources for ultra-short-term load forecasting. In this context, data-drive techniques, e.g., long short-term memory (LSTM) networks and convolutional neural networks (CNNs) have gradually become essential methodology to improve load forecasting accuracy. In this paper, we identify an ultra-short-term load forecasting model based on LSTM net-works and CNNs. LSTM networks reflect the nonlinear relation between influential features and future loads, and the influential features are extracted implicitly from special CNNs. Case studies have verified the effectiveness of the proposed model in terms of both accuracy and efficiency, comparing with shallow neural networks, separate CNNs, separate LSTM networks, and sepa-rate gated recurrent unit (GRU) networks.