基于MLPs-dynFWA模型的高速铁路短时客流预测方法研究

Short-term Passenger Flow Prediction of High Speed Railway Based on MLPs-dyn FWA Model

针对传统深度学习模型在预测高速铁路短时客流时难以确定时间阈值、网络参数选取有难度等问题,提出一种基于多层感知器时间序列网络与动态搜索烟花算法的高速铁路预测模型。首先以高速铁路短时客流预测作为研究对象,将MLPs网络中每个节点作为一个感知器,模拟生物神经网络中神经元基础功能,对时间变化特征进行建模;再将dynFWA算法应用到神经网络参数多样性选择中,利用爆炸算子搜索机制对网络超参数组合进行优化,以高速铁路历史真实客流系数为基础,部分数据作为数据源,部分数据作为验证组,通过MLPs-dyn FWA模型进行预测并将结果与其他预测模型进行比较,得到不同数据组在不同模型优化策略下的性能指标。通过实验结果得知,MLPs-dynFWA模型对于高速铁路短时客流预测性能最优。

Traditional deep learning models are difficult to determine the time threshold and select network parameters when predicting the short-term passenger flow of high speed railways. A high speed railway prediction model based on the multi-layer perceptron time series network and the dynamic search fireworks algorithm was proposed. First, the short-term passenger flow prediction of high speed railways was taken as the research object. Each node in the MLPs network was taken as a perceptron to simulate the basic function of neurons in the biological neural network and model the temporal variation characteristics. Then the dynFWA algorithm was applied to the diversity selection of neural network parameters, and the explosion operator search mechanism was used to optimize the network hyper-parameter combination. On the basis of the historical real passenger flow coefficient of high speed railways, part of the data was taken as sources, while other as validation groups. The MLPs-dynFWA model was built to predict and compare the results with other models, and the performance indicators of different models with various optimization strategies were obtained. The experimental results show that the MLPs-dynFWA model performs best in short-term passenger flow prediction of high speed railways.