基于时间进化反传学习算法的控制器设计

Design of Controller Based on BPTT Learning Algorithm

时间进化反传算法(BPTT)引入离散时间变量,把网络各个时刻的处理发展成前向网络。其计算的导数能用于模式识别及随机、确定控制。基于BPTT的控制器仿真用一单层网络作为仿真器模拟系统的运动学特性,用BPTT算法按时间展开仿真器模型,再运用BP算法进行训练,以求解动态系统诸问题。

A discrete time variable was introduced in the back-propagation through time (BPTT) learning algorithms. This recurrent network is processed into a feed-forward network by unfolding the network at the time steps. The derivatives of calculation for BPTT are used in pattern recognition, stochastic and deterministic control. For BPTT-based controller simulation, a layer network is adopted as the emulator, and the kinematics characteristic of a simulating system is simulated by it. The simulative model was unfolded with BPTT algorithm according to the time steps. As the same time, BPTT was trained with BP algorithm to solve the problem of identifying a dynamic system and so on.