Forecasting Daily Urban Water Demand Using Artificial Neural Networks Based on Evolutionary Algorithms, A Case Study of Soufiyan Urban Water

Forecasting of water demand in water supply systems is essential to water resources management and its distribution as properly. According to non-linear and oscillation process of water consumption and its affecting variables, the use of non-linear models such as neural networks have get more success in this field. On the other hand, these models have some defects such as the need to more training data and weakness in finding global optimal solutions. In this study by combining the multi-layer neural networks with PSO and ICA evolutionary algorithms, the mentioned defects eliminated firstly, and then the neural networks trained and the daily water demand of Soufiyan city is predicted based on weather parameters. The results show that the hybrid neural network with PSO and ICA algorithms had better performance compared to a network that trained by LM classical algorithm. The hybrid model of neural network with PSO algorithm has correlation coefficient equal to 0.98 which have the more accurate solutions than other models in any of the warm and cold seasons. Also water demand forecasting with proposed hybrid model in the next 10 years revealed that water demand will be increased about 40% in this city.