Week-ahead Daily Peak Load Forecasting Using Hybrid Convolutional Neural Network

Abstract

Peak load forecasting (PLF) is essential for the operation of large-scale power systems, such as security-constrained economic dispatch and rolling unit commitment. PLF is critical for peak load shaving and demand side management in distribution systems, too. Results obtained by PLF can assist market participants to develop their bidding schemes in power markets. This work presents a novel method, based on a hybrid convolutional neural network (CNN) that is cascaded with a fully-connected network, to explore week-ahead daily PLF. The proposed method uses a systematic grid search along with Adaptive Moment Estimation (Adam) optimizer to design the hybrid model. The grid search tunes the network structure and hyperparameters (such as kernel size) of the hybrid CNN while Adam optimizer adjusts the synaptic weights and parameters (e.g., values of a kernel). Realistic daily peak load data and meteorological (temperature) data in Taiwan are investigated. Simulation results obtained by the proposed hybrid deep learning model are better than those obtained by the traditional multi-layer neural network, recurrent neural network, vector auto-regressive moving average model and support vector machine.