矿井淋水井筒风温PSO-SVR预测方法

PSO-SVR prediction method of airflow temperature of shaft with water dropping in mine

矿井进风井筒风温的准确预测对于井下风流的热计算至关重要。为提高矿井井筒风温预测精度,在结合矿井生产特点和参考有关淋水井筒风温预测研究的基础上,采用粒子群算法(PSO)对支持向量回归(SVR)参数进行优化,建立矿井淋水井筒风温PSO-SVR预测模型,并与利用同样的训练和测试样本建立的常规SVR预测模型和多元线性回归(MLR)预测模型进行比较。结果表明:对于训练和测试样本,MLR预测模型的预测与观测值散点分散于标准线四周,相比于MLR预测模型,常规SVR预测模型的散点较集中于标准线周围,而经过PSO优化后的SVR预测模型的散点均紧密分布在标准线附近,说明PSO-SVR预测模型具有更好的预测精度,更强的泛化性;MLR预测模型、常规SVR预测模型和PSO-SVR预测模型的测试样本预测结果的平均绝对百分比误差分别为3.43%,1.27%和0.37%,常规SVR预测模型较MLR预测模型的预测结果改进比约63%,PSO-SVR预测模型较常规SVR预测模型的预测结果改进比约71%,表明PSO-SVR预测模型的预测效果显著优于MLR预测模型和常规SVR预测模型,该模型适用于矿井淋水井筒风温的预测。

The accurate prediction of airflow temperature in air intake shaft of mine of great significance for the thermal calculation of underground mine airflow.In order to improve the prediction accuracy of airflow temperature in air intake shaft of mine,based on the characteristics of mine production and the study of the prediction of airflow temperature of shaft with water dropping,particle swarm optimization(PSO)is used to optimize the parameters of support vector regression(SVR),and the PSO-SVR prediction model of airflow temperature of shaft with water dropping in mine is established.The conventional SVR prediction and multiple linear regression(MLR)models are established by using the same training and testing samples,with the predicted results of PSO-SVR model compared.It is found that for training and testing samples,the scatter points of prediction and observation values of MLR prediction model are scattered around the standard line.Compared with MLR prediction model,the scatter points of conventional SVR prediction model are more concentrated around the standard line,while the scatter points of SVR prediction model after PSO optimization are closely distributed near the standard line,which indicates that PSO-SVR prediction model has better prediction accuracy and stronger generation capacity.The mean absolute percentage errors of MLR prediction model,conventional SVR prediction model and PSO-SVR prediction model are 3.43%,1.27%and 0.37%,respectively.The improvement ratio of conventional SVR prediction model is about 63%compared with MLR prediction model,and the improvement ratio of PSO-SVR prediction model is about 71%compared with conventional SVR prediction model.The prediction effect of PSO-SVR prediction model is better than MLR prediction model and conventional SVR prediction model.The model is suitable for the prediction of airflow temperature of shaft with water dropping in mine.