基于IPSO-Capsule-NN模型的中欧班列出口需求量预测

为提高中欧班列出口需求量的预测精度,提出将改进粒子群算法(IPSO)与胶囊神经网络(Capsule-NN)相结合的预测模型(IPSO-Capsule-NN)。与全连接神经网络不同,胶囊神经网络通过动态路由算法增强了模型的拟合能力和泛化能力。利用改进粒子群算法优化胶囊神经网络的神经元数量、迭代次数以及学习率,以克服人为设定模型参数随机性较大导致模型精确度不高的不足之处。此外,针对标准粒子群算法存在的缺点,提出一种非线性递减惯性权重并引入Levy飞行对粒子群算法的全局寻优能力和收敛速度进行优化。将采用spearman秩相关性分析得到的11个因素作为中欧班列出口需求量的影响因素并对其进行预测,结果表明:胶囊神经网络具有2层隐含层时,IPSO-Capsule-NN模型预测精度更高。

Prediction of Export Demand of China Railway Express Based on IPSO-Capsule-NN Model

In order to improve the prediction accuracy of the export demand of the China Railway Express,the prediction model(IPSO-Capsule-NN)combining the improved particle swarm optimization(IPSO)and capsule neural network(Capsule-NN)is proposed.The difference between Capsule-NN and fully connected neural network is that Capsule-NN uses the dynamic routing algorithm to enhance the fitting ability and generalization ability of the model.The high randomness of manually setting model parameters will lead to low prediction accuracy.In order to overcome this deficiency,IPSO algorithm is used to optimize the parameters of Capsule-NN which including the number of neurons,the number of iterations and the learning rate.In addition,aiming at the shortcomings of the standard PSO algorithm,a nonlinear decreasing inertia weight is proposed and Levy flight is introduced to optimize the ability of searching for globally optimal and convergence speed of PSO algorithm.Spearman rank correlation analysis method is used to screen the influencing factors of the export demand of the China Railway Express,and 11 factors are used to predict the export demand.Results show that when the Capsule-NN has two hidden layers,IPSO-Capsule-NN has higher prediction accuracy.