A Multi-Agent Evolution Algorithm Used for Input Shaping of a Repetitive Non-Linear Dynamic System

This paper explores the challenges regarding designing a heuristic control algorithm for a dynamic non-linear system with multiple inputs and outputs. The presented algorithm aims to shape the voltage input (both magnitude and timing) applied to fast switching valves in a Digital Displacement^® unit. This consists of multiple sub-systems, where optimal decisions must be made, based on the system design and performance criteria. In this regard good performance are defined as: low electrical energy required for switching, accurate switching timing and low plunger velocity near the seat. The proposed algorithm examines the design-space in a user-defined manner combined with stochastic decision making. The randomness of the algorithm is based on the standard deviation between located elite designs. This reveals several feasible input sequences to achieve the goal, and the optimums are benchmarked with a differential evolution algorithm. The techniques are demonstrated by simulation and the results compared showing similar performance of the optimums.