论文部分内容阅读
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。
The existing rolling force mathematical models are mostly derived from a series of simplifications under various assumptions to determine their model inaccuracies, so the conventional rolling force model itself can not provide sufficiently accurate forecast values . Neural network technology provides a brand new modeling tool, which uses a dynamic variable step size BP algorithm. In order to get the best iterative calculation speed and prediction accuracy, the parameters of hidden layer unit number, weight initial value range, learning rate and so on are optimized. At the same time, the range of variable step parameter selection is discussed and studied. To better understand and grasp the original design rolling force model of Handan Iron and Steel Company Slab CSP has important practical and reference value.