A Novel, Low Computational Complexity, Parallel Swarm Algorithm for Application in Low-Energy Devices

Area of Science:

• Computer Engineering
• Artificial Intelligence
• Embedded Systems

Background:

• Particle Swarm Optimization (PSO) is widely used for optimization problems but faces challenges in hardware implementation, especially for low-power applications.
• Miniaturized devices like wireless sensor networks (WSNs) and wireless body area sensors (WBANs) demand algorithms with reduced computational complexity and energy consumption.
• Efficient hardware implementation of PSO is hindered by challenges, particularly in the randomization function, limiting its use in energy-constrained systems.

Purpose of the Study:

• To propose a novel metaheuristic algorithm based on PSO, specifically designed to reduce computational complexity and energy consumption for hardware implementation.
• To address the challenges of implementing PSO in hardware, focusing on efficient randomization techniques.
• To enhance the suitability of swarm intelligence algorithms for energy-limited embedded systems.

Main Methods:

• Modified a conventional particle swarm optimization (PSO) algorithm by replacing the random value generation block with deterministic methods.
• Developed novel techniques to differentiate particle trajectories within the swarm using deterministic approaches.
• Investigated software models of the modified algorithm and presented hardware implementations of key modified blocks, focusing on reduced complexity and energy.

Main Results:

• The modified PSO algorithm demonstrated comparable or superior performance to the conventional PSO across various scenarios in software simulations.
• The algorithm proved flexible and adaptive when tested with numerous fitness functions.
• Hardware implementation focused on reducing complexity and energy consumption while maintaining high-speed operation.

Conclusions:

• The proposed deterministic PSO variant offers a viable solution for low-energy consumption in miniaturized devices and systems.
• The modifications successfully reduced computational complexity, leading to significant energy savings in hardware implementations.
• The algorithm's performance and adaptability make it suitable for resource-constrained applications like WSNs and WBANs.