Application of Artificial Neural Network Model In Forecasting Water Demand: Case of Kimilili Water Supply Scheme, Kenya.

Abstract

Potable water treatment and supply systems are designed and constructed to deliver adequate water to meet consumer demand requirements. Consequently, water demand forecasting is essential for the design and operations management of treated water supply systems. Correct prediction of time-varying water demand trends and the critical water demand values determines the extent to which a network can satisfy critical demand and maintain economic efficiency. This study aimed to forecast Kimilili water supply scheme water demand up to 2030. Kimilili water supply scheme being operated by Nzoia Water Services Company Limited is characterized by rapidly increasing water demand leading to persistent water supply shortages hence unplanned fluctuations in the system water production hours. The artificial Neural Network (ANN) model was utilized to forecast Kimilili water supply scheme water demand. The trained model had good performance with a coefficient of determination (R2) of 0.999972988. The results indicated that Water demand for the Kimilili water supply increased with time, and the general relationship between time and water demand was defined by a sixth-order polynomial function given by y = 9e-0x6-1e-05x5+0.0005x4-0.0115x3+0.1178x2+0.1384x+100.48. The study confirmed that ANN could simulate the water demand characteristics of the water supply very well.