A Machine Learning based Closed-Loop Efficiency Optimization Strategy for Data-Driven Optimization Methods

Data-driven optimization methods play a major role in industrial system performance optimization. As the requirements for performance optimization efficiency become more and more stringent, the efficiency of data-driven methods becomes the key to their application. The efficiency of the data-driven methods mainly depends on their algorithm parameter settings. However, the current parameters settings scheme optimization process is usually conducted in trial-and-error mode, which depends on the experience of engineers or researchers, and is cumbersome and time-consuming. To address this problem, this paper proposes a closed-loop datadriven optimization method, which can guide the improvement of the next optimization process through the morphological characteristics of the current optimization process. In this method, the Visual Geometry Group(VGG) neural network is used to obtain the morphological characteristics of the current optimization process to guide the data-driven algorithm based on Simultaneous Perturbation Stochastic Approximation (SPSA) to improve the current optimization process, realize the auxiliary optimization of control parameters, and show the role of using multidimensional data to assist the traditional control method to achieve rapid convergence and achieve high efficiency and economy. Simulation experiments were conducted on the performance optimization of the steam generator level control system. The simulation results showed that this method is feasible and could be assisted in optimizing the control parameters of the data-driven optimization methods.