NN-Based Iterative Learning Control Under Resource Constraints A Feedback Scheduling Approach英文教材.pdfVIP

NN-Based Iterative Learning Control Under Resource Constraints: A Feedback Scheduling Approach Feng Xia and Youxian Sun National Laboratory of Industrial Control Technology, Institute of Modern Control Engineering Zhejiang University, Hangzhou 310027, China {xia,yxsun}@iipc.zju.edu.cn Abstract. The problem of neural network based iterative learning control (NNILC) in a resource-constrained environment with workload uncertainty is examined from the real-time implementation perspective. Thanks to the itera- tive nature of the NNILC algorithm, it is possible to abort the optimization rou- tine before it reaches the optimum. Taking into account the impact of resource constraints, a feedback scheduling approach is suggested, with the primary goal of maximize the control performance. The execution time of the NNILC task is dynamically adjusted to achieve a desired CPU utilization level. Thus a tradeoff is done between the available CPU time and the control performance. For the sake of easy implementation, a practical solution with a dynamic iteration stop criterion is proposed. Preliminary simulation results argue that the proposed approach is efficient and delivers better performance in the face of workload variations than the traditional NNILC algorithm. 1 Introduction Neural network based iterative learning control (NNILC) has attracted significant attention in many control applications thanks to the successful combination of neural networks and iterative learning control. As perfect nonlinear function approximators, neural networks have good potential for identification and cont