Radial basis collocation method with parameters optimized for estimating pollutant release history

Abstract

The identification of pollutant source release history in rivers is important for emergency response of pollution accidents and formulating remediation strategies. Space-time radial basis collocation method (RBCM), as a meshless method with strong applicability, can directly estimate the release history from the concentration data measured at downstream observation sites. However, the uncertainty of specific parameters in space-time RBCM is the main factor affecting the accuracy of estimation. Therefore, a way to solve the parameters efficiently and accurately is essential. For this purpose, a new model which combines space-time RBCM and differential evolution algorithm (DEA) is established to identify the source release history. First of all, efficient parameter optimizer DEA is introduced to search the parameters that affect the estimation accuracy of space-time RBCM. Then, a new loss function considering the imbalance configuration of RBCM nodes is designed to ensure the rationality of the parameters obtained by DEA. The results of numerical cases and real field case show that the proposed method can accurately estimate the real release history with low time consumption. It is also demonstrated that DEA is more efficient than k-fold cross-validation in searching the optimal parameters for space-time RBCM, and the parameters obtained from the new loss function can make the estimated release history more precise.

Keywords: Differential evolution algorithm; Loss function; Meshless method; Pollutant release history identification; Radial basis collocation method; Surface water.