Model-Based Policy Search for Automatic Tuning of Multivariate PID Controllers

Proportional, Integral and Derivative (PID) control structures are still the main control tool being used in industrial applications, in particular in the process industry [1], but also in automotive applications [2] and in low-level control in robotics [3]. The large share of PID controlled applications is mainly due to the past record of success, the wide availability, and the simplicity in use of this technique. In practice, control design is still often achieved by tedious manual tuning or by heuristic PID tuning rules [4]. More advanced tuning concepts are most frequently developed for Single-Input-Single-Output (SISO) systems [5], [6]. For Multi-Input-Multi-Output (MIMO) systems, popular tuning methods, such as biggest log-modulus and the dominant pole placement tuning method [7], strive to tune each control loop individually, followed by a collective de-tuning to stabilize the multi-loop system. These tuning methods, however, rely on linear process models and require stable processes. For general PID control structures where multiple controllers act on each input, controller design is usually conducted by decoupling the process, subsequently allowing the design of individual SISO PIDs. One example are online adjusted precompensators, which decouple the process transfer function matrix [8].