基于竞争学习的粒子群优化算法设计及应用

DESIGN AND APPLICATIONS OF PARTICLE SWARM OPTIMIZATION BASED ON COMPETITIVE LEARNING

摘要:

针对传统PSO算法容易陷入局部最优的问题，提出一种基于竞争学习的粒子群优化算法(competitive learning-based PSO, CLPSO)。在CLPSO中，首先通过动态计算粒子的适应度值将种群分成优选、合理和疏离三个子群；其次，根据三个子群中粒子的进化特性，为三个子群分别设计了不同的更新变异方式。然后，利用12个基准测试函数对算法的性能进行了验证。实验结果表明，所提的竞争学习策略能够有效克服经典PSO算法在处理复杂多峰问题时容易陷入局部最优的缺陷。最后，利用CLPSO算法优化模糊神经网络的参数设计CLPSO-FNN算法，并利用其建立出水氨氮软测量模型，实验表明，CLPSO-FNN软测量模型能够更精确更实时地测量出水氨氮浓度。

Abstract:

To solve the problem that traditional PSO algorithm is easy to fall into local optimization, a competitive learning-based particle swarm optimization (CLPSO) algorithm is proposed. In CLPSO, first, by dynamically calculating the fitness value of particles, the population is divided into three subgroups: the optimal region, the reasonable region, and the alienated region. Secondly, according to the evolutionary characteristics of the particles in the three subgroups, different updating and variation modes are designed for the three subgroups respectively. Then, 12 benchmark functions are used to verify the performance of the algorithm. The experimental results show that the proposed competitive learning strategy can effectively overcome the premature convergence shortcoming of classical PSO algorithm in dealing with complicated optimization problems. Finally, the CLPSO algorithm was used to optimize the parameters of the fuzzy neural network, and the CLPSO-FNN algorithm was designed, and the soft measurement model of effluent ammonia nitrogen was established. The experiment showed that the CLPSO-FNN soft measurement model could measure the effluent ammonia nitrogen concentration more accurately and in real time.