基于人工神经网络的PC钢棒力学性能预报模型

Forecast Model of Mechanical Properties of PC Steel Bar Based on Artificial Neural Networks

提出一种PC钢棒抗拉强度的人工神经网络模型方法,采用4×9×1的三层前向BP网络结构,模型主要因素为淬火温度、回火温度、含碳量和单位长度质量。经1500余次训练后,误差平方和<0.001。训练出来的神经网络可以预报生产现场的PC钢棒的抗拉强度σb,预报值与实测值相对偏差在±3%以内,大于93%的预报值与实测值的绝对偏差在±10MPa之间。

A artificial neural networks (ANN) model to predict the tensile strength of PC bar is put forward. The model is based on a backpropagation BP network algorithm with three levels of 4×9×1. The controlling parameters in the model are the quenching temperature, annealing temperature, the carbon content and the mass per unit length. The model is trained after 1500 times and the standard tolerance reduce below 0.001. The trained ANN model can be used to forecast the tensile strength σb of the PC bar on site, the error between the predicted and the measured tensile strength is below ±3%, and more than 93% of the ANN forecasted strength has absolute error below ±10MPa compared to the measured tensile strength.