论文部分内容阅读
对一种在Elman动态递归网络基础上发展而来的复合输入动态递归网络 (CIDRNN)作了改进 ,提出一种新的动态递归神经网络结构 ,称为状态延迟动态递归神经网络 (StateDelayInputDynamicalRecurrentNeuralNetwork) .具有这种新的拓扑结构和学习规则的动态递归网络 ,不仅明确了各权值矩阵的意义 ,而且使权值的训练过程更为简洁 ,意义更为明确 .仿真实验表明 ,这种结构的网络由于增加了网络输入输出的前一步信息 ,提高了收敛速度 ,增强了实时控制的可能性 .然后将该网络用于机器人未知非线性动力学的辨识中 ,使用辨识实际输出与机理模型输出之间的偏差 ,来识别机理模型或简化模型所丢失的信息 ,既利用了机器人现有的建模方法 ,又可以减小网络运算量 ,提高辨识速度 .仿真结果表明了这种改进的有效性 .
An improved dynamic input recursive network (CIDRNN) based on Elman dynamic recursive network is proposed, and a new structure of dynamic recursive neural network is proposed, which is called state delay dynamic recurrent neural network (StateDelayInputDynamicalRecurrentNeuralNetwork) This new topology structure and dynamic recursive network of learning rules not only clarify the significance of each weight matrix, but also make the training process of weight value more concise and more explicit.The simulation results show that the network of this kind of structure Which can increase the convergence speed and enhance the possibility of real-time control.And then, the network is used in the identification of the robot’s unknown nonlinear dynamics, and the method is used to identify the difference between the actual output and the output of mechanism model To identify the mechanism model or simplify the information lost by the model, both the existing modeling method of the robot can be used, the amount of computation and the speed of identification can be reduced, and the simulation results show the effectiveness of this improvement.