【作者】 梁健宇；

【导师】 凌呼君；

【作者基本信息】 内蒙古工业大学 ， 控制理论与控制工程， 2009， 硕士

【摘要】 单元机组协调控制系统的被控对象之间存在强耦合,且响应特性差异巨大,常规机炉协调控制系统的控制策略一般不能满足电网对单元机组协调控制的要求。常规PID控制算法原理简明,参数物理意义明确,理论分析体系完整且应用经验丰富,但鲁棒性比较差。而神经网络具有很强的逼近任意非线性函数的能力,并具有自适应学习、并行分布处理和较强的鲁棒性及容错性等特点。将传统PID控制与现代控制和智能控制理论相结合,能在很大程度上改善控制对象的控制品质。基于PID和神经网络的上述特点,将神经网络与PID控制相结合,提出了一种神经网络解耦PID在线调整控制器参数的控制策略。结合BP神经网络的学习算法和工作原理,详细分析了利用神经网络自动调整PID控制器参数的算法、原理和实现步骤,以及利用神经网络对多变量强耦合系统进行分散解耦的算法、原理及其实现步骤。神经网络通过自身在线学习对强耦合系统进行解耦,神经网络PID控制器根据对象参数发生变化时,系统误差的变化来调整神经网络的权值,以此来改变网络中比例、积分和微分作用的相对强弱,使系统具备较好的动态和静态性能,实现系统解耦控制的要求。最后,用所设计的控制系统对单元机组进行了大量的仿真研究。仿真结果表明:该控制系统响应速度快、超调量小、稳态精度高,能够快速跟踪系统输出并进行有效控制,且具有一定的自适应性和鲁棒性,控制方案切实可行。有效地避免了单纯PID控制鲁棒性差的不足,满足实时控制的要求,对于研究非线性、参数时变的强耦合控制系统提供了一种新的思路,具有一定的理论意义和广阔的应用前景。更多还原

【Abstract】 As the strong coupling among the controlled object of unit coordinated control system and the large differences in response characteristics,the control strategy of conventional boiler-turbine coordinated control system can not meet the requirements of coordination and control for power grid unites. The principium of conventional PID algorithm is simple and clear in physical parameters meaning. The system of theoretical analysis is integrated and abundant application experienced. But it has poor robustness. The neural network approach has a strong ability to close with any nonlinear function, and has more advantages such as self-adaptive learning, parallel distributed processing, strong robustness, fault-tolerance and so on. The quality of the controlling object can be great improved due to the combination of traditional PID control and modern control theoryBased on the above-mentioned characteristics of the PID and neural network, and the combination of them, a new control strategy is advanced which is adjusting controller parameters online by neural network decoupling PID. According to BP neural network learning algorithm and the working principle, this report indicates a detailed analysis of the algorithm, principle and realization for adjusting controller parameters automatically by neural network, moreover, the algorithm, principle and its implementation steps for multivariable strong coupling system decentralized-decoupling by neural network are also shown. Neural network can decouples strong-coupling through its online learning, while the neural network PID can adjust the weights of neural networks based on the change of system error due to object parameters change. Because of the above-mentioned characteristics, system have preferable dynamic and static performance, achieves the requirements of decoupling control.Finally, a large number of simulation studies are implemented by this control system. The result indicates: this control system has fast response, small overshoot, high steady state precision, moreover, it can track system output quickly and control effectively, additionally, it has a certain self-adaptability and robustness. So, the control program is feasible. This project effectively avoids the robustness of a simple PID controlling, satisfy the requirements of real-time control, provides a new approach on strong decoupling control system research of nonlinear, parameters time-variant, has a certain theoretical significance and wide application prospects.更多还原