Study Of Initial Iterative Learning Control Strategy Based On Adaptive Neural-Fuzzy Inference System

Iterative learning control is a technique for improving the transient response and tracking performance of processes, machines, equipment, or systems that execute the same trajectory, motion, or operation repetitively. The idea behind Iterative learning control is that if the system's operating conditions are the same each times it executes then any errors in the tracking response will be repeated during each operation. These errors and the control input signal can be recorded during every repetition and this information can be used to compute a new input signal that will be applied to the system during the next repetition so that the tracking accuracy would be increased as the number of repetitions increases. With the emerging of the cyclic of repetitive systems or processes in industry, Iterative learning control has become a key technique of advanced controls.How to choose the appropriate initial control input value for the system (the control input when the first iteration) is one effective way to make the system achieve the desired trajectory tracking in high-precision using less iterations. However, previous studies on the selected initial control input value is subjective, blind, and generally set it as zero or a bounded random quantity.In order to avoid the blend choice of the initial control input in iterative learning control when the control system faced a new desired trajectory-tracked task or a new environment lead to the existence of tasks such as: 1) the convergence speed is affected because of the re-experienced iterative learning proces; 2) in the first cycle of the new task, the initial control input has no function to the volume output of the system, or even has counterproductive because of the absence of amendments to control the importation of the initial value, increasing the learning time and the number of cycles and so on. An improved algorithm is proposed to obtain the initial value of the iterative learning control based on BP neural network and Adaptive Neural-fuzzy Inference System. Desired control input of iterative learning control is estimated based on the experience of the past control experience using the new desired trajectory-tracked task, the system output, system state and its derivative as its input. The simulation results show that: 1) these methods is feasible and effective for different non-linear object; 2) for different iterative learning algorithm, using the above-mentioned method to determine the initial control input can track the changed expectations trajectory in a smaller tracking error than the zero initial value control input, reduce the number of iterations in the same request to the tracking accuracy.