Research On Adaptive PID Control Without Modeling

Due to the simplicity, good robustness and high reliability, PID controller is widely used in industrial process. However, industrial process is always accompanied with long time-delay, nonlinearity, time-variance and uncertainty, which is difficult to model precisely. So the conventional PID control usually couldn't achieve expected effect. Furthermore, online setting PID parameters results in non-adaptation to complex situation and consequently the bad performance. Therefore, it is necessary to design an adaptive controller, which is simple and does not depend on accurate mathematical model of plant. The main research work in this thesis is as following.First, to modify expanding response curve method (ERCM) for PID, an open-loop recursive algorithm for ERCM is proposed, which simplifies the parameter setting of PID. Keeping the zeros and poles of the PID unchanged, an adaptive PID controller is presented using the proposed recursive algorithm in the closed-loop system. The robustness and practicability of the system are demonstrated in simulation.Second, to overcome the shortcoming of the original identification-free PSD algorithm, a synchronal adaptation for PSD parameters is presented which is of less computation and more insensitive to zero-mean noise in measurement. In order to apply well to the process with time delay, a model-free predictive control law for the PSD is proposed which scheme is based on the asymptotical stability in delay-free PSD system. Simulation results show the validity of the proposed method.Third, the ANN supervised Hebb learning algorithm is applied to tuning the gain of single-parameter-single-neuron PID, which not only adapts the gain, but also reduces the number of controller parameters to be set compared with the other single-neuron PID. Simulation shows that good performance of the presented algorithm can be hold and the control result is satisfactory in process control.Fourth, to avoid inconvenience of modeling, based on the desired performance index and structure of controller, the concept of desired model and method to determine the model structure are proposed, which makes parameters of controller self-tune through online identification. The presented algorithm does not assume the mathematical model of controlled plant in advance. Thus it could be applied to engineering very conveniently.At last, experiments to control the water level of double tank are conducted to test the former three algorithms presented above.