论文部分内容阅读
在标准BLISS方法基础上,依据神经网络的全局映射性,提出了基于BP神经网络建立学科层优化目标函数与系统变量之间的响应面近似模型,并以绝对和相对形式描述的误差函数替代了传统BP算法中单一的绝对形式描述的误差函数。选用样本点在参数空间分布均匀程度更高的试验设计方法——CVT(Centroidal Voronoi Tessellations)试验设计方法来产生训练样本和测试样本,从理论上保证了近似模型的精度。最后利用多属性决策法从算法实施的难易度、优化结果准确性、系统级计算量、算法鲁棒性及收敛性5个方面来评估多学科可行方法(MDF)、改进二级系统合成一体化优化方法(BLISS)的综合性能,定量说明改进BLISS方法更加适合YM160锚杆钻机动力头优化设计。
Based on the standard BLISS method and based on the global mapping of neural network, a model of response surface approximation between objective function and system variable based on BP neural network is proposed. The error function described by absolute and relative form is replaced by error function The Error Function of a Single Absolute Form Described in Traditional BP Algorithm. The experimental design method of CVT (Centroidal Voronoi Tessellations) test is used to generate training samples and test samples, and the accuracy of the approximate model is theoretically guaranteed. Finally, the multi-attribute decision-making method is used to evaluate the multi-disciplinary feasible method (MDF) from five aspects of the difficulty of algorithm implementation, the accuracy of the optimization results, the system-level computation, the robustness of the algorithm and the convergence. The comprehensive performance of Bliss optimization method (BLISS) shows that the improved BLISS method is more suitable for the optimization design of power head of YM160 bolt drilling rig.