Soil Pollution Prediction Based on KPCA and Improved Deep Sparse Extreme Learning Machine

Abstract

A soil prediction model combining kernel principal component analysis (KPCA) features extraction and optimized deep extreme learning machine(DELM) is proposed to solve the problems of traditional methods for predicting heavy metal content in soil are of low accuracy and soil testing is expensive. First of all, KPCA method is used to extract the effective parameters of soil heavy metals to realize data dimension reduction. Secondly, the input weights and biases of deep extreme learning machine (DELM) with kernel mapping theory are optimized by an improved Mayfly algorithm (MA) based on adaptive weights. Finally, MA-DKELM model is constructed; the data after dimension reduction are used as input to predict the heavy metal content in soil. The results show that the MA-DKELM model has strong prediction stability and can be applied to the prediction of heavy metal content in soil. Compared with PSO-DELM and DELM prediction models, MA-DKELM model has the highest accuracy and the best comprehensive performance in predicting soil heavy metals. MA-DKELM model is practical in predicting heavy metal content in soil.