基于LSTM神经网络的地铁车站温度预测

Prediction of subway station temperature based on LSTM neural network

地铁车站温度是影响乘客热舒适性的主要指标,且易受到列车运动、机械风等多种因素的影响而产生较大幅度和较为频繁的波动,需要准确掌握车站温度的变化规律和趋势,以便为合理调控车站热环境舒适性提供科学依据.为此,以北京市某地铁车站的温度实测数据为例,采用小波去噪、数据窗口化处理以及时序数据建模方法,分别建立了车站温度的差分自回归移动平均(ARIMA)预测模型、长短时记忆(LSTM)神经网络预测模型、双向长短时记忆(BiLSTM)神经网络预测模型.通过3种预测模型,得到车站温度的预测值与实测值的对比,研究结果表明:3种模型均具有较好的回归预测性能,适用于宏观掌握地铁车站温度的变化趋势,且BiLSTM模型的预测精度最优,其次是LSTM模型和ARIMA模型.其中,BiLSTM模型的预测精度可达80.58%,能够学习温度时序数据的双向特征,更适用于预测环境状态波动明显的空间温度变化趋势.

Temperature is one of the main indicators affecting the thermal comfort for passengers in the subway station,which is prone to show significant and frequent fluctuations due to the train movement,mechanical ventilation and other factors.Therefore,predicting the station temperature and its change rule is required to provide scientific basis for adjusting the thermal comfort in the subway station.Taking the measured temperature data of a subway station in Beijing as an example,this paper uses wavelet denoising and data windowing processing,as well as the time series data modeling method to establish 3 prediction models including Autoregressive Integrated Moving Average(ARIMA),Long Short-Term Memory(LSTM)neural network and Bi-directional Long Short-Term Memory(BiLSTM)neural network prediction model.Then,the predicted value is compared with the measured value based on these models.The results show that all 3 models perform well in regression prediction,suitable for grasping the variation tendency of subway station temperature from a macro perspective;BiLSTM model has the highest prediction accuracy,followed by LSTM model,and then ARIMA model;the prediction accuracy of BiLSTM model is up to 80.58%,it can achieve bidirectional correlation with subway station temperature,which can be applied to predict temperature data with large amplitude and frequency.