基于烟花粒子群算法优化BP神经网络的轨道客流量预测

Prediction of Railway Passenger Flow Based on Fireworks Particle Swarm Optimization Algorithm Optimized BP Neural Network

为进一步提高轨道客流量预测的精确度,提出一种基于烟花算法(fireworks algorithm, FWA)搜索机制下的FWA-PSO-BP轨道客流预测模型。粒子群算法(particle swarm optimization, PSO)通过将随机因素引入进化方程中实现,不过,由于这种随机搜索模式会导致粒子群算法的局部搜索功能减弱,很容易出现早熟收敛现象和寻优力不足的情况。为了改进这一问题,通过引入烟花算法中的爆炸火花和突变火花,对粒子的搜索范围和粒子数量进行动态调节,增强粒子群的多样性,使粒子群算法具有局部搜索能力和全局搜索能力的自我调节机制,从而改善粒子群算法的早熟收敛问题,对反向传播(back propagation, BP)神经网络的初始权值与阈值进行更好的优化。以重庆轨道客流数据进行实例验证,结果表明:FWA-PSO-BP模型的平均绝对百分比误差(mean absolute percentage error, MAPE)为2.54%,优于所有其他对比模型。

To further enhance the accuracy of orbit passenger flow prediction,a fireworks algorithm(FWA)-particle swarm optimization(PSO)-back propagation(BP)orbit passenger flow prediction model was proposed.Random factors were introduced through PSO algorithm into the evolutionary equation.However,the local search function of PSO algorithm might be weakened by this random search mode,easily leading to premature convergence and insufficient optimization power.In order to improve this problem,the explosion spark and mutation spark in the fireworks algorithm were introduced to dynamically adjust the search range and number of particles,enhance the diversity of the particle swarm,and make the particle swarm algorithm have a self-regulation mechanism of local search ability and global search ability,thereby improving the premature convergence problem of the particle swarm algorithm.The initial weights and thresholds of the backpropagation(BP)neural network were better optimized.Using Chongqing rail passenger flow data as an example,the results show that the FWA-PSO-BP model has a mean absolute percentage error(MAPE)of 2.54%,which is superior to all other compared models.