论文部分内容阅读
给出了神经网络趋化性算法的一种新的实现策略，在此基础上，提出了一种动态递归神经网络建模方法和一种控制作用受限的自学习非线性控制方法。将其用于连续搅拌釜式发酵器的状态变量的在线预测和优化控制，仿真结果表明，预测精度高，控制效果好，具有强抗扰和强鲁棒性。在不知道生化过程模型结构的情况下，神经网络模型，可以很容易地通过在线或离线法学习到高度复杂的非线性生化过程的输入、输出关系。对于经过最优操作点，稳态增益的符号会发生变化的这类难以控制的生化过程，神经网络非线性控制策略，可以使生化反应器始终维持在最优状况。本方法有望在实际工业过程中得到应用。
A new implementation strategy of neural network chemotaxis algorithm is given. Based on this, a dynamic recurrent neural network modeling method and a self-learning nonlinear control method with limited control effect are proposed. It is used to predict and optimize the state variables of continuous stirred tank fermentor. The simulation results show that the method has high prediction accuracy, good control effect, strong anti-interference and strong robustness. Without knowing the structure of the biochemical process model, the neural network model can easily learn the input and output relations of highly complex non-linear biochemical processes through online or offline methods. For biochemical processes that are difficult to control and whose sign of steady-state gain will change after the optimal operating point, neural network nonlinear control strategy can keep the bioreactor always in the best condition. The method is expected to be applied in practical industrial processes.