Finite impulse response optimizers for solving optimization problems / Tasiransurini Ab Rahman

Faculty of Mechanical & Manufacturing Engineering

Thesis (Doctor of Philosophy) -- Universiti Malaysia Pahang – 2019

Includes bibliographical references

Optimization problems are frequently found in various fields. The classification of estimation-based metaheuristic algorithms has been introduced for solving optimization problems. Simulated Kalman filter (SKF) algorithm is one of the algorithms under this classification. SKF is inspired by the framework of Kalman filter (KF) which is a popular estimator for solving estimation problems. SKF needs parameters of the initial error covariant, measurement noise, and process noise to operate. Nonetheless, no study on parameter tuning being carried out for all SKF’s parameters. Selecting optimal parameters’ values may improve an algorithm’s performance. This can be done through parameter tuning experiment. However, tuning several parameters is a challenging task and time-consuming. Thus, this study attempts to adopt a new search strategy from another popular estimator, named the Ultimate iterative unbiased finite impulse response (UFIR) filter which works with only one parameter. UFIR filter is one of the variants of the finite impulse response (FIR) filter. FIR filter is introduced to overcome the limitation in KF filter which has several parameters that difficult to be determined in a real application. In this work, three new estimation-based metaheuristic algorithms are introduced. The first algorithm is a single-agent-based algorithm, named Single-agent FIR optimizer (SAFIRO). The second algorithm is a multi-agent-based algorithm with synchronous update mechanism, named Multi-agent FIR optimizer (MAFIRO). The third algorithm is a multi-agent-based algorithm with asynchronous update mechanism, named Asynchronous FIR optimizer (AFIRO). SAFIRO differs from MAFIRO in term of the number of agents. Meanwhile, MAFIRO differs from AFIRO in terms of the iteration search strategy. These three algorithms are called in short as FIR optimizers (FIROs). Each agent in FIROs responsible for searching a solution by performing the measurement and estimation. During measurement, FIROs employ a random mutation of the best-sofar solution with local neighbourhood method to balance between the exploration and exploitation process. This measurement value is then used in the estimation to improve the solution iteratively. The performances of FIROs are tested by solving the CEC 2014 benchmark suite. The competencies of FIROs are statistically compared with four existing metaheuristic algorithms: the SKF, single-solution SKF (ssSKF), Particle swarm optimization (PSO), and Genetic algorithm (GA). Statistical analysis using the Friedman test and Holm post hoc test are performed to rank the performances of FIROs. Friedman test shows that SAFIRO has the highest rank, followed by MAFIRO, AFIRO, ssSKF, SKF, PSO, and GA. Holm post hoc test reveals SAFIRO performed significantly better than SKF, ssSKF, PSO, and GA. Whereas, both MAFIRO and AFIRO performed significantly better than PSO and GA, but equivalent to SKF and ssSKF. SAFIRO, MAFIRO, and AFIRO provide on par performances. However, SAFIRO can be regarded as the best algorithm with the highest ranking of Friedman and the highest number of best performances in solving the CEC 2014 benchmark suite. Findings show that the concept of UFIR filter is a good inspiration for metaheuristic algorithm. These newly estimationbased metaheuristic algorithms can offer promising results for solving optimization problems.