【摘要】 基于精确的铁路客运量预测对于国家和企业的规划管理非常重要,为提高预测的精度,提出改进粒子群算法(IPSO)和将粒子群算法(PSO)与长短时记忆神经网络相结合的预测模型(IPSO-LSTM)。LSTM与传统的全连接神经网络不同,其避免梯度消失,具有记忆过去信息的能力。由于LSTM的神经元数量、学习率和迭代次数难以确定,利用IPSO对这些参数进行优化。提出利用非线性惯性权重变化来提高PSO的全局寻优能力和收敛速度。将相关性分析得到的铁路营业里程、国家铁路客车拥有量、国内生产总值和年末总人口作为铁路客运量的影响因素并对铁路客运量进行预测。预测结果表明,当LSTM具有2层隐含层时,IPSO-LSTM具有更高的精确度。更多还原

【Abstract】 Accurate railway passenger volume forecasting and passenger turnover volume forecasting are very important to the planning and management for state and enterprises. In order to improve the predictive accuracy, the improved particle swarm optimization(IPSO) and a combination model(IPSO-LSTM) combining particle swarm optimization(PSO) with the long-short term memory neural network are proposed. LSTM avoids gradient disappearance and has the ability of remembering past information. The difference between LSTM and the traditional fully connected neural network is that it avoids the vanishing of gradient and has the ability to memorize the past information. Since the number of neurons, the learning rate and the number of iterations of LSTM are difficult to determine, IPSO is used to optimize these parameters. In addition, the nonlinear inertia weight is proposed to improve the global search ability and convergence speed of PSO. The railway mileage, the ownership of the national railway passenger train, the gross domestic product and the year-end total population are considered as the influencing factors of the railway passenger volume, and the railway passenger volume is predicted. The prediction results show that when the LSTM has two hidden layers, IPSO-LSTM has higher accuracy.更多还原