CAJ | 학술논문

由于BP网络存在学习效率低、收敛速度慢、易陷入局部极小状态、适应能力较差等缺点，而粒子群优化（PSO）算法的收敛速度快（尤其是在进化初始阶段），运算简单、易于实现，又没有遗传算法的编解码和杂交、变异等复杂运算，因此是一种很好的优化算法。但是，PSO算法也存在不足，该算法进化后期存在速度变慢以及早熟的现象。提出一种改进的粒子群BP神经网络对高炉炉温进行预测。通过调整粒子群算法中学习因子的自适应能力，提高算法的收敛速度和搜索全局最优的能力。通过仿真结果说明改进的粒子群算法要优于BP算法和标准的粒子群算法。
유우BP망락존재학습효솔저、수렴속도만、역함입국부겁소상태、괄응능력교차등결점，이입자군우화（PSO）산법적수렴속도쾌（우기시재진화초시계단），운산간단、역우실현，우몰유유전산법적편해마화잡교、변이등복잡운산，인차시일충흔호적우화산법。단시，PSO산법야존재불족，해산법진화후기존재속도변만이급조숙적현상。제출일충개진적입자군BP신경망락대고로로온진행예측。통과조정입자군산법중학습인자적자괄응능력，제고산법적수렴속도화수색전국최우적능력。통과방진결과설명개진적입자군산법요우우BP산법화표준적입자군산법。
The BP network has the disadvantages such as the low learning efficiency, slow convergence, easily falling into local minimum state, and poor adaptability. Particle swarm optimization (PSO) algorithm is fast in convergence especially at the initial stage, simple in the computing and easy to implement. And when compared with the genetic algorithms, it has no complex operations of hybrid codecs or mutation, which proves to be a good optimization algorithm. However, particle swarm optimization (PSO) algorithm also has disadvantages. It is slower in convergence rate at the late evolution of the algorithm. In this paper, a new BP Neural Network based on improved Particle Swarm Optimization(PSO)is proposed. The convergence speed of this algorithm and the capacity of searching global optimum is increased through adjusting the adaptability of learning factor. The simulation results illustrate that the improved PSO is superior to the standard BP algorithm and particle swarm optimization.