基于SKPCA与NEAT算法的煤与瓦斯突出危险性预测  

Prediction and forecast of the SKPCA with NEAT coal and gas outburst risks

作　　者：徐耀松[1] 程业伟 XU Yao-song;CHENG Ye-wei(Faculty of Electrical and Engineering Control,Liaoning Technical University,Huludao 125105,Liaoning,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,辽宁葫芦岛125105

出　　处：《安全与环境学报》2021年第4期1427-1433,共7页Journal of Safety and Environment

基　　金：辽宁省教育厅辽宁省高等学校基本科研项目(LJ2017QL012);辽宁省自然科学基金指导计划项目(20180550438)。

摘　　要：针对煤与瓦斯突出预测精度不足的问题,为了减少突出灾害的危害,提出了基于稀疏核主成分分析法(SKPCA)与增强拓扑神经进化算法(NEAT)的煤与瓦斯突出危险性预测方法。通过核主成分分析法对突出非线性数据进行降维,然后对主成分数据进行稀疏化,减少原始数据中不重要元素对降维后主成分的影响。搭建增强拓扑神经进化网络(NEAT)预测模型,采用进化算法同时优化神经网络的权值和拓扑结构,得到最佳神经网络预测模型。将处理后的主成分数据输入模型预测突出危险指数,结果表明,该方法危险等级预测准确度达到98%,SKPCA-NEAT模型在煤与瓦斯突出预测上相比PCA和BP神经网络具有优势。The present paper is to devote itself to the investigation and examination of the coal and gas outbursts,one of the main accidents in the coal mine production process.To reduce the harmful gas explosion caused by the coal and gas outburst accidents,this paper intends to propose a prediction method of coal and gas outburst risks based on the sparse kernel principal component analysis(SKPCA)and the neuron-evolution of the augmentation topologies(NEAT).The kernel principal component analysis method uses the kernel function to map the original data to high dimensions,and then to reduce the dimension of the data.For the non-linear data of the coal and gas outburst,it is usually recommended to turn the kernel principal component analysis method for processing.And,then,the principal component data that can be obtained can be sparsely processed so as to reduce the loading factors of the insignificant elements in the main components to zero so as to retain the main elements.Neuron-evolution of the augmentation topologies(NEAT)is introduced to predict gas outburst.NEAT algorithm uses the principle of evolutionary algorithm to optimize the weight and connection topology of neural network simultaneously.The network structure is genetically encoded.And,then,genetic variation is carried out to retain individuals with high fitness(prediction accuracy)and eliminate individuals with low fitness.After hundreds of generations of evolution,the best neural network prediction model is obtained.And,then,the principal component data processed by SKPCA method were input into the model to predict the outburst risk index.The experimental results we have gained through the above tests show that the prediction accuracy of the method we have suggested can be expected to reach the efficiency up to 98%.The PCA-BP method is only 93%,and PCA-NEAT method accuracy is 96%.Thus,it can be concluded that the SKPCANEAT model can be expected to enjoy a better prediction accuracy in forecasting the probability rate of the coal and gas outbursts.And,of the

关 键 词：安全工程 煤突出 瓦斯突出 稀疏核主成分分析 NEAT算法 

分 类 号：X936[环境科学与工程—安全科学]