论文部分内容阅读
为提高冲击地压危险等级预测模型的泛化性能及预测精度,采用网格搜索法结合十折交叉验证法对极限学习机(ELM)中的隐含层神经元个数及激活函数的类型进行组合优化,进而建立冲击地压危险等级预测模型;选取重庆砚石台煤矿36组实测数据进行试验,对影响因素数据进行标准化处理,选择其中26组样本对模型进行训练,采用该模型对后10组样本中冲击地压危险等级进行预测,并与其他方法作对比。结果显示:经过十折交叉验证,用该模型得到的正确识别率为84.615%,高于朴素贝叶斯及Adaboost M1的76.92%、61.54%,采用该模型对测试样本集中冲击地压危险等级进行预测,预测准确率为90%,高于朴素贝叶斯及Adaboost M1预测准确率80%。
In order to improve the generalization performance and prediction accuracy of predicting models of rock burst, grid search method and ten-fold cross-validation method were used to study the number of hidden neurons in ELM and the type of activation function Combined with optimization, and then to establish a prediction model of rock burst pressure level; 36 groups of measured data from Chongqing Yan Shitai Coal Mine were selected for testing, the influencing factors were standardized and 26 samples were chosen to train the model. Group of samples in the risk of rock burst to predict and compare with other methods. The results show that the correct recognition rate obtained by this model is 84.615% after 10-fold cross-validation, which is higher than that of Naïve Bayesian and Adaboost M1 by 76.92% and 61.54%. The model is used to test the concentration of rock burst The forecasting accuracy is 90%, higher than that predicted by Naive Bayes and Adaboost M1 by 80%.