论文部分内容阅读
采用了一种基于贝叶斯方法的前向神经网络训练算法以提高网络的泛化能力,并在网络的目标函数中引入了表示网络结构复杂性的惩罚项,避免了网络的过拟合。采用Levenberg-Marquardt算法训练网络,并使用Gauss-Newton的数值方法来近似求解Hessian矩阵,以减少计算量,从而提高了网络的收敛速度。将上述网络应用于冷轧过程的轧制力预报中,预报结果的精度远远高于解析模型,与基于传统BP神经网络的冷轧轧制力预报模型相比,在收敛的速度和预报的精度上均优于后者。
A forward neural network training algorithm based on Bayesian method is adopted to improve the generalization ability of the network, and a penalty item that represents the complexity of the network structure is introduced into the objective function of the network to avoid the network over-fitting. The Levenberg-Marquardt algorithm is used to train the network, and the Gauss-Newton numerical method is used to approximate the Hessian matrix to reduce the computational complexity and improve the network convergence rate. The above network is applied to the prediction of the rolling force in the cold rolling process. The accuracy of the prediction results is much higher than that of the analytical model. Compared with the prediction model of the cold rolling force based on the traditional BP neural network, Accuracy are better than the latter.