竖井钻速ANFIS预测模型训练参数正交试验优化

Optimization on training parameters of shaft-drilled speed prediction model based on ANFIS by orthogonal test

训练参数的设置是影响ANFIS预测模型预测精度的关键因素。针对某竖井钻速ANFIS预测模型中隶属函数类型、个数及训练次数3个训练参数的设置问题,笔者开展了正交优化试验,并分析了各训练参数对训练误差和预测误差的影响规律。研究结果表明,与隶属函数类型和训练次数相比,隶属函数个数对训练误差及预测误差的影响程度最为显著。在试验范围内,优化的训练参数设置组合为隶属函数类型,设三角形函数、隶属函数的个数为2,且训练次数为100。优化参数后的ANFIS模型训练误差为0.007 17,预测值总体平均相对误差为6.37%,预测值均方误差为0.016。与传统的钻速方程预测法及单一的BP神经网络预测法相比,ANFIS钻速预测模型预测精度更高,对优化钻井参数、提高钻井效率具有参考意义。

The setting of the training parameters are the key factor affecting the prediction precision of ANFIS prediction model. To research the setting about three training parameters such as the type of membership functions, the number of membership functions and the time of training in a drilling speed prediction model based on ANFIS, an orthogonal test with three factors and three levels was applied to investigate the influence of each training parameter on training error and prediction error. The test showed that the number of membership functions was the most significant factor affecting the training error and the prediction error compared with the type of membership functions and the time of training. Within the test range, the optimal training parameter setting combination was that the type of membership functions was the triangle function, and the number of membership functions was two, and the time of training was 100. As a result, the training error of the ANFIS model with the optimal training parameters was 0.007 17, the overall average relative error of the predicted drilling speed was 6.37%, and the mean square error of the predicted drilling speed was 0.016. Compared with the drilling speed equation forecast method and the neural network prediction method with BP arithmetic, the ANFIS drilling speed prediction model had better prediction precision. It had significance for optimizing drilling parameters and improving drilling efficiency.