关键词: 开采沉陷, 概率积分法, RW降噪, 蚁群算法, ENN神经网络

Abstract: In order to improve the accuracy of the Elman neural network (ENN)to obtain probability integral prediction parameters.Taking the measured data of 30 surface observation stations in China as the sample data for learning training and testing,and the strong robust local weighted regression method (rlowess,RW) for noise reduction of the 30 surface observation stations data was adopted.Ant colony optimization (ACO) was used to optimize the weights and thresholds of the ENN neural network to construct the ACOENN probability integral prediction parameters solving method.The results show that comparing the measured data before and after the ACOENN model solved RW noise reduction treatment found that the RW noise reduction treatment significantly improved the data quality and the prediction accuracy of the solved model.The average relative errors of the subsidence coefficient,horizontal movement coefficient,main influence angle tangent and inflection point offset distance solved by using ACOENN were 2.41%,3.48%,6.11%,and 1.67%,respectively.The probability integral prediction parameters solved by the ACOENN model are better than those solved by the traditional ENN algorithm in terms of accuracy,which provides a new idea to obtain probability integral prediction parameters with higher accuracy.