Multi-agent simulated annealing algorithm based on differential perturbation for protein structure prediction problems

Simulated annealing SA algorithm is extremely slow in convergence, and the implementation and efficiency of parallel SA algorithms are typically problem-dependent. To overcome such intrinsic limitations, this paper presents a multi-agent SA MSA algorithm to address protein structure prediction problems based on the 2D off-lattice model. Inspired by the learning ability of the mutation operators in differential evolution algorithm, three differential perturbation DP operators are defined to generate candidate solutions collaboratively. This paper also analyses the effect of different sampling grain, which determines how many dimensions will be perturbed when a candidate solution is generated. The proposed MSA algorithm can achieve better intensification ability by taking advantage of the learning ability from DP operators, which can adjust its neighbourhood structure adaptively. Simulation experiments were carried on four artificial Fibonacci sequences, and the results show that the performance of MSA algorithm is promising.