A plate-tectonics based neighborhood search optimizer and its application for fault monitoring in IoT systems

摘要

Plate tectonics is a scientific theory that describes large scale motions of plates in Earth’s lithosphere. This paper presents a new optimization technique inspired by the tectonic forces that drive continental drift: Plate Tectonics based Neighborhood Search Optimization (PBO). The proposed evolutionary algorithm simulates tectonic forces on different points of the plate using nearest neighborhood search to exploit local information available in the objective function’s gradients while simultaneously exploring for better solutions throughout the search space. PBO is a versatile algorithm whose behavior can be altered in accordance to domain specific requirements on the objective function. Such tuning can lead to higher accuracy and faster convergence rate than other optimization algorithms. According to the CEC 2018 benchmark, it performs better than other top-performing evolutionary optimization methods in the CEC 2018 special session in all dimensions (D) where D = 10, 30, 50, and 100. It is resistant to getting stuck at local-minima due to increased sampling density of stable points even in vanishingly small gradients. This property, along with its customizability in search behavior, make it a compelling choice for adaptation in industrial uses such as those in mechanical engineering, communication, and aviation. Towards the end, this paper discusses and adapts PBO for application in one such industrial domain: Fault Monitoring in IoT systems. It also compares the results with other evolutionary algorithms like, GA, ACO, PSO, Simulated Annealing, etc. From the results, it can be seen that PBO outperforms other contemporary algorithms for the same application on standard datasets of different sizes.