Construct Intelligent Optimal Operation Rules for Pumping Stations during Typhoon and Storm Periods

The rapid urbanization in metropolitan areas causes less water infiltration, flashy floods and shorter rainfall concentration time. To effectively manage urban inundation problems, pumping stations play an important role for flood mitigation in urban areas. The operation rules for pumping stations have been designed to determine the number of duty pumps based only on the water levels of the front storage pool (FSP). Nevertheless, current pump operation depends mainly on experienced operators rather than on pump operation rules. In practice, pump operation needs to not only consider the discharge amount to rivers, the number of duty pumps, FSP water levels, as well as the water head difference of the surrounding river and the FSP but also avoid switching a pump on or off too frequently within a short time. Therefore, this study aims to propose an approach to deriving multi-objective optimal pump operation rules through the Non-dominated Sorting Genetic Algorithm-II (NSGA-II) in consideration of the water head difference of the surrounding river and the FSP as an objective during typhoon periods for urban flood control. The Yu-Cheng pumping station of Taipei City in Taiwan is the study area. The optimization search of the multi-objective pump operation rules is conducted by the non-dominated sorting and crowding-distance calculation of the NSGA-II, in which the objective functions identify good chromosomes in order to select the optimal pump operation rules from the Pareto-front. Two NSGA-II models with different objective functions are established to investigate the impacts of various objective functions on pump operation rules in this study. The objective functions of Model A include: (A1) minimize the standard deviation of FSP water levels at t+i and t+i+1; (A2) minimize the accumulated peak FSP water levels; and (A3) minimize the accumulated absolute differences on the numbers of duty pumps at t+i and t+i+1. Model B has the same objective functions as Model A except for the first objective function, which is modified to minimize the absolute differences of FSP water levels at t+i and t+i+1 in Model B. This study first adopts 14 typhoon and storm events to search the optimal pump operation rules for the two models and further applies 3 additional events to simulating the optimal pump operations of the two constructed models, for which the two optimization models are compared with current pump operation rules and historical operation. Results indicate that Model A performs better than current pump operation rules, except for the first objective function (A1). Model B performs better than current pump operation rules in terms of all three objective functions, for which the improvement rates can achieve 43.14%, 2.79% and 71.27% for objective functions B1, B2 and B3, respectively. In addition, Model A produces more fluctuations than Model B, which means pumps are switched on and off more frequently by Model A. However Model A and Model B perform more stable than current pump operation rules and thus can effectively reduce the number of duty pumps for the whole operational period. As a consequence, the derived multi-objective optimal pump operation rules can avoid switching a pump on or off too frequently within a short time, and it makes little difference in pump operations based on the optimal pump operation rules suggested by the proposed model and the experiences of operators. The multi-objective optimal pump operation rules of the propose models are considered superior to current pump operation rules, and therefore can provide useful information to operators at pumping stations for real-time urban flood control.