基于联邦学习的城市轨道交通列车智能控制方法研究

Intelligent Control Method of Urban Rail Transit Based on Federal Learning

针对城市轨道交通智能驾驶策略优化问题,提出联邦学习方法;学习代理采用基于支持向量机(SVM)的控制模型,通过构建一个多项式和径向基核函数组成的混合核函数,使用随列车速度变化的动态权重因子来提高模型精度;利用北京地铁昌平、燕房、亦庄及16号线数据,在联邦学习的框架下,仿真完成列车智能驾驶模型的训练。仿真结果表明:该方法训练出的模型具有良好的预测效率和预测准确度。基于联邦学习的列车智能控制可为列车自动驾驶的优化与改进提供有力的实践依据。

The federal learning method is proposed for the problem of urban rail transit intelligent driving strategy optimization.The learning agent in this method adopts a control model based on support vector machine(SVM),and improves model accuracy by constructing a hybrid kernel function consisting of a polynomial and a radial basis kernel function,and applying dynamic weight factors that vary with train speed.Using data from the Beijing Metro Changping,Yanfang,Yizhuang and Line 16,the training of the train intelligent driving model is completed by simulation under the framework of federal learning.The simulation results show that the trained model has good prediction efficiency and prediction accuracy.The intelligent train control based on federal learning can provide a strong practical basis for the optimization and improvement of automatic train driving.