HYDRODYNAMIC AND WATER QUALITY SURROGATE MODELLING FOR RESERVOIR OPERATION

摘要

In order to include the water quality state in the reservoir release policy for long and short term optimization, a forecasting surrogate model was developed using M5 model trees algorithm. This data-driven model is able to recreate the state of the system while reducing the number of input variables from the original numerical model. After testing the methodology it was concluded that the model can represent the discharges in the system accurately based on a single year routing calculation. In terms of the main stream water quality it can also represent the seasonal trends. However it is also concluded that more extreme pollution events need to be included inside the training data set in order to achieve the correct representation of high pollution inflows in the system.