基于模型降阶的时滞多变量系统动态解耦

Dynamic Decoupling of Multivariable Systems with Time-delay Based on Reduced Order Models

针对工业生产过程一类时滞多变量系统,采用BP神经网络实现时滞多变量系统的动态解耦。时滞环节增加了解耦器的设计难度,并导致解耦器在物理上难以实现。针对该问题,对时滞多变量离散系统的解耦进行了讨论,为降低神经网络解耦器的规模,采用离散状态方程模型的均衡实现降阶算法对神经网络解耦器进行降维。以典型的火电机组协调系统进行解耦仿真试验,并采用PID控制器实现控制。结果表明,采用的离散化方法对时滞多变量系统具有良好的解耦效果,合理解决了时滞对解耦过程的影响,并通过模型降阶技术降低了神经网络解耦器的规模,便于神经网络解耦实现在线学习。

This paper discusses BP neural network based on dynamic decoupling for a type of multivariable system with time-delay in the process industry, whose decoupler is difficult to be designed and realized in physics. According to the discussion about the decoupling of multivariable discrete system with time-delay, dimensions of the neural network decoupler is reduced by the balance reduction technology for the discrete state model. A decoupling simulation is taken for a typical coordinate system of a thermal power unit, which is controlled by the PID controller. The simulation results show that the model order reduction technique has good decoupling effect for a type of multivariable discrete system with time-delay, which is effective to reduce the influence of time-delay in the decoupling process and the scale of the neural network decoupler. So it is easier to realize the online learning algorithm of the neural network decoupler.