利用粒子群算法求解管网污染源反向追踪模型

An optimal simulation method for identifying pollution source in water distribution networks based on the particle swarm optimization

针对供水管网中突发污染事件的污染源识别问题,构建污染源反向追踪的模拟-优化数学模型,利用粒子群优化算法求解污染物的侵入位置、时间及侵入速度信息。分别以监测点处污染物模拟质量浓度与实际质量浓度为数据源,应用模型对试验供水管网进行污染源识别研究,分析管网拓扑结构、粒子群算法中种群规模及惯性权重参数设置对结果的影响。结果表明,当参数设置合理时,基于粒子群优化算法的污染源模拟-优化反向追踪模型具有较高的准确率和计算效率。

The paper is inclined to propose an optimal simulation method based on the particle swarm optimization（PSO） to be applied for identifying the pollution source in the water distribution networks. In the paper, we have also chosen PSO as an optimized tool and the EPANET water distribution system model as a simulator. For the research purpose, we have first of all analyzed the changing situation of the pollutant Cr（VI） concentrations with time and space in the pipe network in accordance with the experiments of Cr（VI） solution intru- sion into the laboratory water distribution network, on which basis we have calibrated and established the hydraulic and water quality mod- els, which should be coincided with the actual experimental scenario in the pipe network. And then, we have managed to set up a simulation-optimization model for identifying the pollution source in the water distribution networks as a calculation program compiled in the Matlab, with the sum of the square difference between the simulated pollutant and the actually measured concentrations at monitoring spots as the objective function and the EPANET toolbox as the embedded simulation engine. Besides, we have also adopted the particle swarm optimization algorithm to collect the information and data of the pollution source spots, the starting time and speed of the pollutant intrusion. Thirdly, based on the Cr （VI） intrusion experiment and the simulation-optimization model, we have been trying to solve the pollution source identification problem by using the simulated pollution concentrations and measured ones respectively as the data resource. The method we have proposed has thus been proved viable. And, finally, by setting the parameters influencing the model and comparing the output results, we have also made a thorough analysis of the influential factors, such as the network topologic structure, the population size and the inertiaweight in the particle swarm optimization program. What is more, the computational accuracy and efficiency of the suggested simulation-optimization method turn out to be high enough if the following conditions can be satisfied： the network structure should be made similar enough to the actual one, few joints are similar downstream to the real contamination sources, and the population size and inertiaweight in the particle swarm optimization should be organized reasonably.