A reliable short-term water demand forecast would be a great tool to assist engineers and operators to manage daily operation and save energy usage. In this paper, the authors introduce a novel strategy to forecast short-term water demand in a real time base. This strategy first matches current water demands with either a known demand pattern or a dynamic demand pattern, then uses the pattern as a guide to direct the hourly predictions for next 24 hours. The authors demonstrate that this strategy is an optimal hybrid of pattern recognition, time series analysis and regression analysis. It not only inherits the advantages from these three approaches, but also gains the ability to recognize the impact of external factors in real time and dynamically adjusts the forecast for better accuracy. Although the strategy of forecasting demand with demand profiling has great potential as illustrated in this paper, there is much fine-tuning to be done before it is ready for industrial applications.
展开▼
