论文部分内容阅读
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 training parameters is the key factor that affects the prediction accuracy of ANFIS prediction model. Aiming at the setting of three training parameters, the types of membership functions and the number of training times in ANFIS prediction model for drilling speed of a shaft, the author carried out orthogonal optimization experiments and analyzed the influence of each training parameter on training error and prediction error. The results show that the number of membership functions has the most significant impact on training errors and prediction errors compared with the membership functions and training times. Within the experimental range, the optimal combination of training parameters is set as the membership function type, and the triangle function is set. The number of membership functions is 2 and the number of training times is 100. The training error of ANFIS model after optimizing parameters was 0.00717, the overall average relative error of prediction was 6.37%, and the mean square error of prediction was 0.016. Compared with the traditional prediction method of drilling speed equation and the single BP neural network prediction method, the prediction accuracy of the ANFIS drilling rate prediction model is higher, which is of reference value to optimize the drilling parameters and improve the drilling efficiency.