The proportional-integral-derivative (PID) control is the most common algorithm in the ship power plant control system. There are as many as dozens of control loops in the whole simulation system including auxiliary power unit systems. It is hard to debug parameters for complicated system containing many control loops, and therefore it is necessary to study auto-tuning system for PID control parameters. An advantage of simulation is that simulation experiments can be repeated with a very small price. Complicated simulation system with data processing functions can easily obtain the output response curve of control system and access feature parameters. According to characteristics of simulation system and analysis in tuning method of PID controller, the tuning method based on pattern recognition of output response is determined. The variation of the control parameters can be directly observed from the response curve with the tuning method based on pattern recognition of output response. The changes between control parameters and control performance indexes, such as overshoot and decay rate, can be obtained by simulation experiment, and tuning rules can be drawn from these changes. Because the rules summarized from observation have the characters of vague, fuzzy theory is adopted to deal with these rules and fuzzy tuning rules are attained. The parameters obtained by fuzzy tuning rules can not guarantee control quality is optimal as a result of experience based on human. Genetic algorithm (GA) is a better optimization algorithm with global search but it is difficult to define the optimization space. In this paper, a new tuning method is established with the advantages of the two algorithms above, in which the overshoot and decay rate are obtained from output of the response caused by step disturbance and compared with the indexes defined by user; then the scopes of the parameters are given by fuzzy rules, which are regarded as the initial optimization space of GA; so the overshoot, decay rate and integrated time and absolute error (ITAE) of output response is defined as the objective function of GA; finally the optimal solution of control parameter are determined by GA. In this process, the control parameters are optimized within the scope defined by fuzzy rules, and control performance indexes are expected to alter in appointed scopes. The efficiency of optimization is determined by several key factors including scope of search space, size of initial population, termination of evolutionary algebra and operation of genetic operators. Basic GA is improved to reduce calculation, increase ability of optimizing and avoid premature. The initial space of parameter optimized in GA is given by fuzzy rules mentioned above; Decision mechanism is introduced to terminate optimizing process when convergence rate becomes slow in order to improve the real-time of the algorithm; Mutation probability is adjusted in the process of genetic manipulation, in which individuals with higher degree of adaptation have smaller probability of mutation and vice versa. With the integration simulation environment, the controller parameter auto-tuning is realized, which significantly reduce the modeling time and the result of performance of the software package is satisfied.
PID controller auto-tuning for ship power plant simulation system
2010
6 Seiten, 4 Bilder, 4 Tabellen, 7 Quellen
Aufsatz (Konferenz)
Datenträger
Englisch
Auto-Tuning PID Controller on Electromechanical Actuators Using Machine Learning
SAE Technical Papers | 2023
|