Design of a data-driven expert controller

This paper proposes a design method for an expert adaptive controller that stores operating data from experts in a database and designs a local proposal-integral-derivative (PID) controller at each step by referencing applicable data from this database. In recent years, the retirements of senior experts have created a serious social problem owing to the loss of their expertise. A skill-based PID controller has been proposed to solve this problem. However, the method uses only one expert´s skills. According to the proposed method, data vectors that are composed of a number of experts´ skills and evaluation scores are stored in a database. Data vectors that are close to a query are designated as neighbor data; a local PID controller is designed by using these neighbor data. Data vectors including low evaluation scores are excluded from the neighbor data by adding evaluation scores to their distances. Thus, the controller is designed by using neighbor data that include high evaluation scores, so that superior control can be obtained. The effectiveness of the proposed method is evaluated by a simulated example.