MultiCycleNet: Multiple Cycles Self-Boosted Neural Network for Short-term Electric Household Load Forecasting

Abstract

• Proposed a deep learning framework for household load forecasting. • Used correlative data in multiple cycles to explore power consumption patterns. • Used contextual information around the data to improve predicted accuracy. • The proposed framework outperforms the benchmarks with fluctuating load profiles. Household load forecasting plays an important role in future grid planning and operation. However, compared with aggregated load forecasting, household load forecasting faces the challenge of the uncertainty of prolific load profiles. This paper presents a novel multiple cycles self-boosted neural network (MultiCycleNet) framework for household load forecasting, which aims to solve the uncertainty problem of household load profiles through the correlation analysis of electricity consumption patterns in multiple cycles. The basic idea of the proposed framework is that the predictor can learn customers’ power consumption patterns better by learning the features and contextual information of relevant load profiles in multiple historical cycles. We use two real-life datasets: 1. the household load consumption dataset from Low Carbon London project led by United Kingdom (UK) Power Networks and 2. the UK Domestic Appliance-Level Electricity (UK-DALE) dataset to evaluate the effectiveness of the proposed framework. Compared with the state-of-the-art methods, experimental results show that the proposed framework is effective and outperforms the state-of-the-art methods by 19.83%, 10.46%, 11.14% and 9.02% in terms of mean squared error, root mean squared error, mean absolute error and mean absolute percent error, respectively.