变邻域宽度的爬山微粒群优化算法及其应用

Hill-climbing particle swarm optimization algorithm with variable width neighborhood and its application

This paper proposes a hill-climbing particle swarm optim ization algorithm with variable width neighborhood (vwnHCPSO).The new method ass umes that some stochastic particles are produced in an initial neighborhood of t he best particle P_g at the first iteration of PSO.Then the best individual P_gn of the stochastic particles is found. If P_gn is better than P_g,P_g is replaced with P_gn and the next iteration of PSO goes on. If P_gn is not better than P_g,the neighborhood wid th of the best particle P_g is broadened, the stochastic particles producti on is renewed and the best individual P_gn of stochastic particles is f ound again. If P_gn can be better than P_g now, P_g is repla ced with P_gn and the next iteration of PSO can go on. Otherwise, the neighborhood width of the best particle P_g is broadened again and the next iteration of PSO does not go on until the best individual P_gn of stoc hastic particles is found or the neighborhood width exceeds the scheduled width . Then, vwnHCPSO, hill-climbing particle swarm optimization algorithm with inv ariable width neighborhood (HCPSO) and PSO are used to resolve several well-kno wn and widely used test functions’ optimization problems. Results show t hat vwnHCPSO has greater efficiency, better performance and more advantages in m any aspects than HCPSO and PSO. Next, vwnHCPSO is used to train artificial neur al network (NN) to construct a practical soft-sensor of light diesel oil flash point of the main fractionator of fluid catalytic cracking unit (FCCU).The obtai ned results and comparison with actual industrial data indicate that the new met hod proposed in this paper is feasible and effective in soft-sensor of light di esel oil flash point.

This paper proposes a hill-climbing particle swarm optimization algorithm with variable width neighborhood （vwnHCPSO）. The new method assumes that some stochastic particles are produced in an initial neighborhood of the best particle Pg at the first iteration of PSO. Then the best individual Pgn of the stochastic particles is found. If Pgn is better than Pg, Pg is replaced with Pgn and the next iteration of PSO goes on. If Pgo is not better than Pg, the neighborhood width of the best particle Pg is broadened, the stochastic particles production is renewed and the best individual Pgn of stochastic particles is found again. If Pgn can be better than Pg now, Pg is replaced with Pgn and the next iteration of PSO can go on. Otherwise, the neighborhood width of the best particle Pg is broadened again and the next iteration of PSO does not go on until the best individual Pgn of stochastic particles is found or the neighborhood width exceeds the scheduled width. Then, vwnHCPSO, hill-climbing particle swarm optimization algorithm with invariable width neighborhood （HCPSO） and PSO are used to resolve several well-known and widely used test functions＇ optimization problems. Results show that vwnHCPSO has greater efficiency, better performance and more advantages in many aspects than HCPSO and PSO. Next, vwnHCPSO is used to train artificial neural network （NN） to construct a practical soft-sensor of light diesel oil flash point of the main fractionator of fluid catalytic cracking unit （FCCU）. The obtained results and comparison with actual industrial data indicate that the new method proposed in this paper is feasible and effective in softsensor of light diesel oil flash point.