ELM evaluation model of regional groundwater quality based on the crow search algorithm

According to the multi-parameter evaluation of groundwater quality, an evaluation model of groundwater quality based on the improved Extreme Learning Machine (ELM) was proposed to resolve fuzziness of the water quality evaluation and incompatibility of water parameters. A training sample set and testing sample set were randomly generated according to the classification standards of groundwater quality, then Crow Search Algorithm (CSA) was used to optimize the input weights and thresholds of hidden-layer neurons of the ELM; thus, the CSA-ELM evaluation model of groundwater quality was constructed based on optimization of the ELM by the CSA. Base on the training sample set and testing sample set, the CSA-ELM model was tested. The test results indicate that the evaluating precision and generalization ability of the CSA-ELM model reach a high level and can be used for comprehensive evaluations of groundwater quality. The Jiansanjiang Administration in Heilongjiang Province, China, was used as an example; the groundwater quality of 15 farms in this region was evaluated based on the CSA-ELM model. The groundwater quality in this region was generally good, and the groundwater quality appeared to have spatial distribution characteristics. Compared with the Nemerow Index Method (NIM), the CSA-ELM evaluation model of groundwater quality is more reasonable and can be used for the comprehensive evaluation of groundwater quality. The stability of the NIM, ELM model, back propagation (BP) model and CSA-ELM model was analyzed using the theory of serial number summation and Spearman's correlation coefficient. The stability of the NIM and BP model in groundwater quality evaluation was poor, while the stability of the ELM model and CSA-ELM model was relatively superior. The ranked results of stability are CSA-ELM model > ELM model > NIM > BP model. The reliability of the NIM, ELM model, BP model and CSA-ELM model was analyzed using the theory of distinction degree. The reliability of the NIM was not good, although its distinction degree was large; the distinction degrees of the ELM model, BP model and CSA-ELM model were close to each other. The ranked results of reliability are CSA-ELM model > ELM model > BP model. The CSA-ELM model can provide a stable and reliable evaluation method for the evaluation of related fields and thus has important practical applicability.