Genetic Algorithm with Gene Regulatory Networks Based Optimization Method for Distributed Video Analysis System

To construct a digital twin, we have to analyze a large amount of video data obtained from the real world in real-time to perceive an accurate current situation. The increase in the video traffic flowing through networks and the accompanying increase in power consumption are problems. One solution is edge computing to distribute video analyzing tasks; it requires dynamic control of task distribution and selecting analyzing models according to the network status and application requirements. Our research group has formulated an optimization problem for dynamic control of distributed video analysis systems and proposed a method to optimize power consumption using a genetic algorithm. However, as the scale of the problem expands, there are cases where the genetic algorithm cannot solve the problem because it falls into the local solution. Therefore, we focus on gene regulatory networks that map a genotype of an organism to its phenotype. The characteristics of gene regulatory networks are to memorize specific phenotypes, to mutate multiple phenotype bits from one-bit perturbation in genotype simultaneously, and to generate new phenotypes similar to the memorized ones by the network structure. In this study, we propose a dynamic control method for distributed video analysis systems by applying gene regulatory networks to genetic algorithms. We used a mathematical model of gene regulatory networks and made it memorized several optimal solutions as phenotypes. We confirmed the escape from local solutions by recalling the stored solutions and generating multiple phenotype bits mutations. We also verified the adaptability improvement to the dynamic network environment and a new application requirement by using the gene regulatory network.