In the water treatment process, a main objective is to improve the
water quality and also minimize the production costs. To achieve these, an
integrated monitoring and control system has been established through flow
prediction and pump scheduling. This paper proposes a new integrated solution
for predictions and optimal pump control by learning algorithms. Flow
prediction has usually been studied for daily or monthly estimation, which is
insufficient for real-time control of a water treatment plant (hereafter WTP). An
hourly based estimator is proposed to track the steady change of flow demand.
Unlike electricity, water can be stored in huge tanks for more than a dozen
hours, which can be used for saving energy and increasing water quality. Pump
on/off minimization is considered to improve the water quality. If influent water
to a water treatment plant varies, then output turbidity and particles are
increasing, which could possibly be supplied to citizens. The proposed on/off
minimization algorithm is expected to prevent those particles from leaking and
to secure public health.