Parallel Optimisation of Pump Schedules with a Thread-Safe Variant of EPANET Toolkit

The optimisation of pump operations in water distribution networks is an ongoing research topic motivated by the great energy savings that a careful scheduling of pumps may achieve. State-of-the-art approaches often combine an optimisation algorithm, such as Evolutionary Algorithms, and full hydraulic simulation. Despite the advances in optimisation techniques and hardware performance, the time required to obtain a near-optimal schedule is still measured in hours. Multiple CPUs are increasingly used to parallelise tasks that require a high computation time. In fact, parallel optimisation algorithms are widely present in the literature. On the other hand, hydraulic simulators that support parallel computation are scarce. The most popular research simulator, EPANET, was not designed with concurrency in mind. In this paper, a new thread-safe variant of the EPANET Toolkit is proposed. As an application example, we propose a parallel variant of an Ant Colony Optimisation (ACO) algorithm for optimal pump scheduling in water distribution networks. The thread-safe EPANET library is combined with the parallel ACO algorithm to achieve reduced computation time in a multi-core computer. Experimental results show that an initial computation time close to 2 hours may be reduced to less than half an hour without sacrificing the quality of the results. The number of ants is also identified as a parameter that influences execution time in the parallel ACO approach proposed in this paper.