Search

• CEEMD was applied to the joint identification of simulation model parameters and GCSs information. • CEEMD had a significant noise reduction effect when used to denoise the observed concentrations. • The accuracy of the identification results improved when the noise-reduced observed concentrations were applied. Identification of groundwater contaminant sources (GCSs) relies on actual observed data, and the observed data directly affects the accuracy of the identification results. However, the observed data inevitably contains noise due to accidental and systematic errors, and the identification results of GCSs based on the observed data containing noise are usually not reliable. This seriously restricts the rational design of contamination remediation plans and contamination risk assessment. To solve this problem, complementary ensemble empirical mode decomposition (CEEMD) was applied to the joint identification of simulation model parameters and GCSs information in this study for the first time. The wavelet analysis and CEEMD methods were used to reduce the noise in observed concentrations respectively, and the noise reduction effect of the two methods was compared and analyzed. Then the simulated concentrations, observed concentrations and noise-reduced observed concentrations with the best noise reduction effect were then applied to identify simulation model parameters and GCSs information, respectively, after which the corresponding identification results were compared and analyzed. The results showed that CEEMD could reduce the noise contained in the observed concentrations more effectively and make it closer to the actual values when compared with the wavelet analysis method. The accuracy of identification results based on the noise-reduced observed concentrations by was improved. As the noise contained in the observed concentrations increased, the noise reduction effect of the CEEMD decreased. Nevertheless, the accuracy of the identification results based on the noise-reduced observed concentrations was still higher than that of those based on the observed concentrations with the same intensity noise without denoising. [ABSTRACT FROM AUTHOR]

• LSTM is a deep learning method which can applied to build a surrogate model for GCSI. • The accuracy of LSTM is highest in a group of methods including RBF, Kriging and KELM. • The use of LSTM-based surrogate modeling significantly accelerate the GCSI process. When simulation–optimization (S/O) is used for groundwater contamination source identification (GCSI), to reduce the calculation load and time generated by calling the simulation model, a surrogate model is often used instead of the simulation model. However, when the conversion relationship between the simulation model inputs and outputs is complex and the model is highly nonlinear, the commonly used approach of building a surrogate model may lose its advantage. This study use and check a deep learning method with the long-short term memory (LSTM) network which has great potential for characterizing the input–output conversion relationship of complex nonlinear numerical simulations, to a surrogate model of the simulation model. The accuracy of the surrogate model developed by the LSTM method was compared with that of commonly used surrogate models developed by the Kriging method, the radial basis function network method, and the kernel extreme learning machine method. The surrogate model with the highest accuracy was linked to the optimization model, and the optimization model was solved to identify the contamination source information. Results show that compared with the other three methods, the surrogate model constructed by the LSTM method had the best accuracy and generalization performance. LSTM is therefore an effective method of building a surrogate model. Linking the LSTM surrogate model to an optimization model and then solving the optimization model can save approximately 99% of the computing load and time otherwise required. [ABSTRACT FROM AUTHOR]

• Improve the extreme learning machine by particle swarm optimization. • Applying PSO-ELM to establish surrogate model for numerical simulation model. • The surrogate model based on the PSO-ELM was embedded in the optimization model. • Surrogate model could avoid computational disasters during the optimization process. • Uncertainty of parameters is considered in GCSI based on PSO-ELM predictive model. When the simulation-optimization method is applied to groundwater contamination source identification (GCSI), the numerical simulation model is usually embedded in the optimization model as a constraint condition. Hundreds and thousands of simulation model calls, while solving the optimization model, often lead to computational disaster. Establishing a surrogate model for the simulation model can effectively alleviate this drawback. When using extreme learning machine (ELM) to establish a surrogate model for a simulation model, the output weights are calculated using randomly given input weights and hidden-layer deviations, which may lead to inadequate accuracy of the surrogate model and insufficient generalization ability to samples not appearing in the training set. However, by combining the ELM with particle swarm optimization (PSO), the PSO can be used to optimize the selection of the input weights and hidden-layer deviations of the ELM; thus, calculating the output weights and establishing a surrogate model based on the PSO-ELM instead of the simulation model embedded in the optimization model to complete the GCSI. The uncertainty of the parameters is rarely considered when the simulation-optimization method is applied to GCSI. Thus, the uncertainty of the parameters is considered in GCSI based on the PSO-ELM. The results show that compared with the ELM, the PSO-ELM can establish the surrogate model with higher accuracy. The surrogate model based on PSO-ELM can be embedded in the optimization model to effectively solve GCSI problems. The predictive model based on PSO-ELM can predict the release histories of contamination sources corresponding to different random parameters. [ABSTRACT FROM AUTHOR]