Deriving flood-control release rules on a reservoir operation system

Abstract

To provide the optimal real-time releases for operators on a multi-reservoir system during typhoon, the study proposes the methodology to extract a set of optimal multi-reservoir operation rules for flood control. The two classification technologies, namely decision-tree algorithm (C5.0) and neural-based decision-tree algorithm (NDT) are employed in the extracting rules. The tasks involves four parts: (1) collection of data, (2) development of the model of optimal flood control operation and solution of optimal release patterns, (3) building of the model of the decision-tree algorithm and neural-based decision-tree algorithm and extraction of the decision trees, and (4) comparison of the developed rules and the current rules. The developed methodology is applied to Shihmen reservoir located in the Tahan river basin in northern Taiwan. The optimal flood-control operation rules involve rules with respect to peak-flow-preceding stage and peak-flow-proceeding stage. In order to generate the optimal data, the reservoir operation optimization model for flood control is applied to the Shihmen reservoir system. This optimization model can identify the best amount of water released at each flood period. Data of 36 typhoons (1987-2004) are available. The generated optimal results of the total 1438 hourly data (36 typhoons), including 335 records of the peak-flow- preceding stage and 1103 records of the peak-flow- proceeding stage, can then serve as the training and testing datasets of the NDT model. The application of the Shihmen reservoir operation verifies the superior performance of the NDT model in contrast to the traditional decision-tree algorithm (C5.0) and current operating rules. Consequently, the developed methodology demonstrates its feasibility.