Fitness-evaluated Adaptive Switching Simulated Kalman Filter Algorithm with Randomness

Abstract

The original Simulated Kalman Filter (SKF) is an optimizer that employs synchronous update mechanism. The agents in SKF update their solutions after all fitness calculations, prediction process, and measurement process are completed. An alternative to synchronous update is asynchronous update. In asynchronous update, only one agent does fitness calculation, prediction, measurement, and estimation processes at one time. A recent study shows that the asynchronous SKF outperforms synchronous SKF. In this study, synchronous and asynchronous mechanisms are combined in SKF. At first, the SKF starts with either synchronous or asynchronous update. By evaluating the fitness, if no improved solution is found, the SKF changes its update mechanism. The decision to switch from synchronous to asynchronous or vice versa is made randomly. Using the CEC2014 benchmark test suite, experimental results indicate that the proposed adaptive switching SKF randomness outperforms the original SKF algorithm.