模糊控制与神经网络方法研究

【作者】 于洋；

【导师】 章卫国；

【作者基本信息】 西北工业大学 ， 导航、制导与控制， 2001， 硕士

【摘要】 本论文主要在综合介绍模糊逻辑控制原理和神经元网络理论的基础上，对现有的模糊控制方法和神经网络方法进行剖析，在指出了各自不足的同时，也进行了一定的改进。并且针对典型的二阶延迟系统分别采用多种神经元网络和神经模糊集成方法进行控制，通过大量的仿真实验进行了验证。论文的主要工作如下： 1．通过对学习速率选取的改进和控制增益的在线调节，提出了一种简便而快速的具有学习能力的单神经元自适应控制方法，仿真结果验证了其有效性。 2．在原有的误差反向传播（BP）网络的基础上，对其学习算法进行了改进，通过在线调节学习速率，提高了算法的实现速度，并且与传统的比例积分微分（PID）控制方法进行结合，分别实现了两种集成方法：BP网络与PID串行控制方法和基于BP网络的PID参数自整定控制方法。通过对常见的二阶延迟系统进行控制应用，验证了两种方法的有效性。 3．对于传统的史密斯预测控制和模型算法预测控制，分别采用不同的神经模糊集成技术加以改进，得到了具有强鲁棒性的控制系统。（1）对于前者，用改进的带优化修正函数的模糊PI控制器替代原来的PID控制器，并且通过BP神经网络实现了PI参数的在线调节。（2）对模型算法预测控制，在保留其优点的同时，用训练好的神经网络代替旧的预测模型，改进的模糊控制器取代老式的PID控制器，构成了一种基于神经网络的预测模糊控制系统。仿真结果验证了这两种方法的有效性。更多还原

【Abstract】 In this paper, the theories of fuzzy logic control and neural networks is introduced. Then, some shortcomings of the existing methods of fuzzy control and neural networks are pointed out and improved upon. Different methods of neural networks and integral neural-fuzzy technique are used in the control of a typical two-order lingering system. The effectiveness of each method is verified through simulation tests. The main contribution of this dissertation is summarized as follows:1.Through improving the choice of learning rate and implementing online adjustment of control plus, a simple and rapid single-nerve adaptive control method with learning ability is proposed, the effectiveness of which is proved by simulation.2.Based on the original BP network, some improvement on error back propagation arithmetic is made. The executing speed of the algorithm is increased through online adjustment of learning rate. Combined with traditional PID control, this method generated two integral schemes: BP network + PID serial control and self-confirming control of parameters of PID controller based on BP network are constructed. Application results in the control of two-order lingering system demonstrate the effectiveness of the two methods.3.Different integral neural-fuzzy techniques are adopted to improve the traditional smith predictive control and model arithmetic predictive control, which results in control systems with strong robustness. (1) The old PID controller of the anterior scheme is substituted by improved PT fuzzy controller with optimal correction function, and the parameters of the segment of P1 are adjusted online by BP neural network. (2) Preserving the intrinsic advantage of the posterior scheme, a predictive fuzzy control. system on neural network is constructed by replacing the old predictive model and PID controller with?.well-trained neural network and the improved fuzzy controller. The effectiveness of the two schemes is verified by the simulation results.更多还原