Household Energy Consumption Prediction Using Evolutionary Ensemble Neural Network

Abstract

Low-voltage local electricity intelligent management is an essential portion of smart grid research. Thereinto, a precise prediction of domestic energy consumption is a pivot in establishing household/neighbourhood energy management system to achieve local smart solutions including consumption auto-balancing, micro generation & storage system, neighbourhood energy sharing, etc. Recent years, a large amount of literature has considered the use of artificial neural networks (ANNs) on electric load forecasting. Various ANN structures and configurations are employed and discussed. However, the solutions are generally developed as case by case studies. The advised network for each specific problem is commonly selected through empirical or enumerative approaches. In this article, an evolutionary ensemble approach is proposed to pool ANNs with various structures and configurations to forecast domestic energy consumption efficiently. The approach utilizes an evolutionary method to select and reproduce better performed network individuals in a network pool to optimize prediction quality. Forecast results demonstrate that the approach achieves a more accurate energy consumption prediction comparing with ANNs with commonly used configurations.