An iterative updating heuristic search strategy for groundwater contamination source identification based on an ACPSO–ELM surrogate system

• Published in: Stochastic environmental research and risk assessment Vol. 35; no. 10; pp. 2153 - 2172
• Main Authors: Wang, Han, Lu, Wenxi, Chang, Zhenbo
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.10.2021; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Aquatic Pollution; Artificial neural networks; Chemistry and Earth Sciences; Computational Intelligence; Computer applications; Computer Science; Computer simulation; Contamination; Earth and Environmental Science; Earth Sciences; Environment; Feedback; Groundwater; Groundwater pollution; Heuristic; Iterative methods; Learning algorithms; Machine learning; Math. Appl. in Environmental Science; Mathematical models; Original Paper; Particle swarm optimization; Physics; Probability distribution; Probability Theory and Stochastic Processes; Problem solving; Search methods; Simulation; Statistics for Engineering; Waste Water Technology; Water Management; Water Pollution Control; Iterative updating heuristic search; Feedback correction step; Groundwater contamination source identification; Surrogate system; ACPSO–ELM
• ISSN: 1436-3240; 1436-3259
• DOI: 10.1007/s00477-021-01994-2

The simulation-random statistics (SRS) method is one of the effective approaches to solving groundwater contamination source identification (GCSI) challenges. In this paper, we applied a heuristic search iterative process (HSIP) based on SRS to provide interval estimation and probability distribution for an unknown variable to enhance our understanding of the unknown variable. Posterior distribution has been a subject of interest in HSIP; therefore, it is imperative to update prior distributions using final iterations obtained from posterior distributions, which has been seldom considered in previous research. This paper first proposed such updates and subsequently designed an updating HSIP (UHSIP). This paper first proposed a feedback correction step and incorporated it into UHSIP to achieve feedback improvements and simultaneous updates of an unknown variable and prior distribution. In UHSIP, a surrogate system such as extreme learning machine (ELM) is usually established for a numerical simulation model to reduce computational cost. The particle swarm optimization (PSO) algorithm has been demonstrated to optimize input weights and hidden layer biases in ELM. However, the PSO algorithm is prone to premature convergence and its calculation accuracy does not satisfy precision requirements when optimizing complex functions. To enhance accuracy, this paper first proposed a new adaptive chaotic PSO (ACPSO) algorithm to establish ACPSO–ELM surrogate system. The two newly proposed approaches were tested in two case studies based on groundwater inorganic and organic contaminations. Results reveal that UHSIP can effectively enhance identification accuracy, and ACPSO–ELM surrogate system has a higher approximation accuracy to the simulation model than PSO–ELM. Overall, the two main innovations in this paper (UHSIP and ACPSO–ELM surrogate system) present promising and effective solutions to GCSI challenge.