A fast and robust simulation-optimization methodology for stormwater quality management

Highlights

• A new framework was proposed for the simulation and optimization of urban LIDs practices.

• The MOSFLA-based method showed advantages in robustness and provides more cost-effective LIDs.

• Markov-based routing within a pipe network is more efficient with good performance.

Abstract

The simulation and optimization of low impact development (LID) practices has been a key research topic in stormwater management. In this study, a fast and robust framework was proposed for providing the optimal design of LID practices by coupling a physically based model, the Markov chain, with the multi-objective shuffled frog leaping algorithm (MOSFLA). The proposed framework was then tested for chemical oxygen demand (COD) reduction in a typical urban catchment in China. The storm water management model (SWMM) was used to provide the flow/COD data during the baseline scenario, and the Markov chain method was then incorporated as a subset of the physically based model. Based on the results, the computational efficiency was improved by 500 times when the new framework was used, and the robustness of the optimal results increased over 50% compared to commonly used algorithms. The relative error between the SWMM and the Markov chain method was less than 5%, indicating that a satisfactory performance could be obtained using the proposed framework. This new method provides a useful tool for optimizing LID practices and green infrastructure, especially for complex urban catchments.

Introduction

The extent of urban areas has been increasing globally, which has resulted in dramatic environmental change, such as increased temperatures, altered hydrology, and elevated water pollution (Newcomer et al., 2014, Yang et al., 2016, Hamed et al., 2018). Specifically, urbanization seals native soils with impervious roofs, roads, and pavement, resulting in an altered hydrological cycle (Londo et al., 2013) and serious nonpoint source (NPS) pollution (Chen et al., 2018, Dai et al., 2019, Hou et al., 2018). The mitigation of stormwater impacts on flow and water quality requires a range of low impact development (LID) practices (Dietz, 2007), such as rainfall gardens, ponds, infiltration systems, vegetated roofs and other control measures. Currently, LIDs have been suggested to be an efficient measure for restoring flow regimes (Zhu and Chen, 2017) and mitigating NPS pollution (Hong et al., 2018), as well as for green infrastructure or sponge city construction (Li et al., 2019).

Urban LIDs vary greatly in their efficiency, cost and design parameters; therefore, the most important question is how to reach the optimal design of LID practices at the catchment scale (Booth et al., 2002, Alfredo et al., 2010). The philosophy behind this question is to replicate the natural hydrological process by using a comprehensive analysis of the type, number, location and combination of many LIDs, particularly at a large scale. Two tools can be utilized for the implementation of LID design. One powerful tool is the physically based urban hydrological model, such as the storm water management model (SWMM), the hydrological simulation program-FORTAN (HSPF) model, the source loading and management model (SLAMM), and the storm water drainage system design and analysis program (DRAINS). These models can simulate the entire rainfall-runoff and NPS processes from land surfaces through a channel or pipe network and can quantify their impacts on the receiving water bodies. As one of the most commonly used models, the SWMM has incorporated the LID module for quantifying the detailed hydrological and hydraulic response to the LID design (Haris et al., 2016, Qin et al., 2013). Furthermore, evolutionary algorithms have also been used for the optimal design of LID in recent years due to the large number, complex configuration, and distributed location of LIDs, especially for complex catchments. For this purpose, famous algorithms, such as the non-dominated sorting genetic algorithm II (NSGA-II) (Deb et al., 2002), the annealing-based multi-objective optimization algorithm (Bandyopadhyay et al., 2008), and the multi-objective particle swarm optimization (Abido, 2009), have been commonly used. To date, the NSGA-II represents one of the most commonly used algorithms for the optimal design of LID practices due to its iterative and parallel subpopulation features.

Based on the above development, coupling a multi-objective optimization algorithm with an urban model has been explored more recently for the optimal design of LIDs. To date, a few decision support systems (DSSs) have been developed. For example, the best management practice decision support system (BMPDSS) has been developed for stormwater management at both the plot scale and the catchment scale (Cheng et al., 2009). This tool is based on the integration of the SWMM and an evolutionary algorithm for a detailed understanding of LIDs (Jia et al., 2012). The benefit-cost (B/C) analysis has also been considered for the mitigation of megacity floods (Huang et al., 2018). However, it should be noted that although the advances in urban models have benefited stakeholders with the possibility for comparing the efficiency of different LIDs, these comprehensive models are often physically based and have many distributed parameters.

Thus, the execution of models such as the SWMM is computationally expensive and technically complex, if not impossible, especially for urban catchments during the optimization process (Chen et al., 2015). If an effective simulator can be proposed, which decreases the computational complexity while maintaining the full use of the watershed model’s strengths, a satisfactory performance could be obtained at large scale simulation. To solve this problem, the Markov chain has been used as a substitute for traditional models for improving the computational efficiency of LID practices (Grimvall and Stalnacke, 1996). However, there are also several problems in these commonly used algorithms, such as poor robustness or low computational efficiency. Instead, the multi-objective shuffled frog leaping algorithm (MOSFLA), as a state-of-the-art method in multiple objective optimization, has been developed. MOSFLA is a novel multi-objective optimization algorithm, which is easy and convenient to code with less control parameters, and it has been testified compatible with handling comprehensive optimization problems including the non-linear and high dimensional discrete systems. It is mostly used in reservoir operation and water resources system optimization in the field of hydrology and environment. However, as far as we know, few scientists have used the advanced MOSFLA for optimizing LID practices, especially from the perspective of NPS reduction.

The objectives of this study were as follows: (1) to develop a simulation-optimization framework to generate cost-benefit information for the optimal design of LID practices; (2) to improve the computation efficiency by introducing the Markov simulator; and (3) to solve the robustness problem by incorporating the MOSFLA. The new framework was then tested in a typical urban catchment in China, and the reliability of results was validated by comparing them to the most-commonly used SWMM and NSGA-II.