核k均值RBFNN的煤与瓦斯突出预测研究

Study on the RBFNN Prediction Method of Coal and Gas Outburst Based on Kernel k-means Algorithm

针对BPNN模型在煤与瓦斯突出预测中存在收敛慢、误差较大等问题,建立了RBFNN模型对煤与瓦斯突出进行预测。采用对样本具有普适性的核k均值聚类算法来确定RBF的中心、梯度下降自适应算法优化网络宽度参数和递推最小二乘法算法调整网络权值。并用国内煤矿的煤与瓦斯突出实测数据对该混合算法及模型进行了验证。实验结果表明,本研究的方法在预测精度和收敛速度上均优于BPNN和基于经典k均值聚类算法的RBFNN,具有良好的实用性和有效性。

In view of the problems existing in the prediction models of coal and gas outburst based on BPNN, and in order to get a fast convergence and more accurate prediction before the outburst accidents, the RBFNN model is built to predict it. The kernel k-means clustering algorithm, which is universal for the samples, is used to determine the central value of the basis function. Its width and its weight are optimized and adjusted by the gradient descent adaptive algorithm and the recursive least square algorithm respectively. And then, the hybrid algorithm and the model are verified with the measured data of coal and gas outburst in China. The simulation results show that the method in the paper has better forecasting accuracy and superior convergence rate than the BPNN and the improved RBFNN based on the classical k-means clustering algorithm, and indicate the practicability and the efficiency of the new model.