步长BP人工神经网络的轧制力模型研究

现有的轧制力数学模型大多是在多种假设的条件下,通过一系列简化推导出来的,从而决定了其模型的不准确性,所以常规轧制力模型本身不能提供足够精确的预报值.神经网络技术提供了一个崭新的建模工具,此模型采用了动态变步长BP算法.为了使模型得到最佳的迭代计算速度和预报精度,对隐含层单元数、权重初始值范围、学习速率等参数进行了优化,同时对变步长参数的选择范围进行了探讨和研究,对于更好地理解掌握和应用邯钢薄板坯连铸连轧厂的原设计轧制力模型,具有重要的实用和参考价值.

Study of Rolling Force Model for Neural Networks with Changing Rate

Because most of the existing models of rolling force are deduced according to a series of simplification and predigesting which have many inaccuracy factors, traditional models can't give satisfied results.The technology of neural networks has been provided as a new tool.The model use BP algorithm with changing rate of training and optimized parameters such as the number of hidden nodes, the scope of the beginning weight factors, etc. The result of this model greatly approaches to the set values of rolling schedule. The work will put forward the research of modeling of rolling load based on neural network. Meanwhile, it will help to use the former model perfectly.