City Water Demand Forecasting Based on Improved BP Neural Network

Abstract

City water demand forecasting is of great significance in reducing the cost of electricity consumption and municipal planning. Back-propagation (BP) neural network has been widely adopted in water demand forecasting in recent years. But BP performs unsatisfactorily in terms of training time and global searching ability, so in this paper we improve BP by two heuristic algorithms, namely, genetic algorithm (ga) and particle swarm optimization (PSO), respectively. The testing and verification of the three algorithms (BP, ga+BP, PSo+BP) have been conducted on real-life water demand forecasting of Beijing city. The obtained results demonstrate that, in spite of the execution time consumed, both ga+BP and PSo+BP performed with higher accuracy and less errors than BP. The obtained results also illustrate that PSo+BP slightly outperformed ga+BP in terms of forecasting accuracy.