【作者】 李博；

【导师】 曲洪权；

【作者基本信息】 北方工业大学 ， 信息与通信工程， 2021， 硕士

【摘要】 随着地铁数量的不断增加,地铁车站工作人员和乘客的舒适性和安全性越来越受到人们的重视。对于地铁站内的环境状况进行分析和预测已成为决策和行动的重要依据。传统的点预测方法的输出结果是未来一段时间内的环境参数的具体数值,如果预测模型输出的结果是基于不同置信度的概率值而不是精确数值的话,可能更接近实际情况的要求。因此,本文提出了基于自回归长短时记忆(LSTM)神经网络的地铁车站环境参数概率预测方法,对地铁车站环境参数趋势的不确定性进行量化,经过验证具有较好的预测准确率。本文针对于地铁环境研究的具体内容如下:1、地铁车站环境参数相关性及特征分析。首先通过7天的现场试验,获得了北京地铁部分车站环境参数和室外大气环境参数的实测数据。然后,分析了站内环境参数之间的相关性,对站内的部分环境参数在站内的分布状况进行调研分析,得到了站内环境状况的基本规律。2、本文首先对传统的时间序列预测方法做了调研分析,同时针对神经网络方法进行时间序列预测的研究方向进行分析讨论。采用斯皮尔曼(Spearman)相关系数对输入参数做了相关性分析讨论,提出了基于神经网络(ANN)的地铁车站环境参数预测方法,拟合外部变量与站内环境参数之间的非线性关系,预测未来一段时间内环境参数值。最后在地铁车站进行现场测试,将采集到的观测数据输入提出的ANN模型中,得到ANN模型的预测平均相对误差为9%,能够较为准确的预测地铁车站环境参数在未来一段时间内的变化。3、本文提出了一种基于自回归LSTM神经网络的地铁车站环境参数概率预测方法。利用一些外部变量和预测变量的历史数据,可以较为准确得预测地铁车站环境参数在未来一段时间内变化情况的概率分布。最后采用地铁站实测得到的2800个观测数据样本验证了提出的概率预测模型的准确性。该模型可以在过去数据和未来数据之间建立条件分布,其结果是一系列具有均值和标准差的分布。最后,通过计算实际值在三个区间内的下降频率,验证了模型的可靠性。综上,与传统的ANN方法相比,概率预测方法还可以为预测参数提供上、下边界等具有相应概率的信息。另外,使用预测的概率分布的平均值来比较,该模型的RMSE平均比ANN提高了58%。为保证预测性能,采用随机森林算法选择关键输入变量进行概率预测。选取关键输入变量和前期预测值作为输入,建立概率预测模型。该方法能合理地实现对地铁车站环境参数概率分布的预测。这项工作有助于预防突发事件的发生,一定程度上可以为合理调控地铁车站环境提供理论依据。更多还原

【Abstract】 With the increasing number of subway stations,people pay more and more attention to the comfort and safety of staff and passengers at a subway station.It has become an important basis for decision-making and action to analyze and predict the environmental conditions in subway stations.The output of traditional point prediction method is the specific value of environmental parameters in the future.However,if the output of the prediction model is based on the probability value of different confidence rather than the accurate value,it may be closer to the requirements of the actual situation.Therefore,this paper proposes a probabilistic prediction model of subway station environmental parameters based on the long-term and short-term memory(LSTM)neural network,which quantifies the uncertainty of subway station environmental trend,and has a good effect after verification.Based on this,the specific research of this paper is as follows:1.Correlation and characteristic analysis of subway environmental parameters.In this paper,through the analysis of the correlation of environmental parameters in the station,the distribution of some environmental parameters in the station is investigated and analyzed,and the basic law of environmental conditions in the station is determined.Through 7-day field test,the measured data of 10 environmental parameters and 8atmospheric environmental parameters of a subway station are obtained.2.The traditional time series prediction methods are investigated and analyzed,and the research direction of neural network method for time series prediction is analyzed and discussed.Spearman correlation coefficient is used to analyze and process the input parameters,and a prediction method of subway station environmental parameters based on neural network is proposed.The nonlinear relationship between external variables and station environmental parameters is fitted to predict the environmental parameters in the future.The model is verified by the observation data of subway station environment.The average relative error of ANN model is 9%,which can accurately predict the environmental parameters of subway station.3.In this paper,an environment probability prediction method of subway station based on Autoregressive LSTM network is proposed.Using the historical data of some external variables and predictive variables,the probability distribution of environmental parameters of subway station can be predicted.The model is validated by 2800 observed data of subway stations.The model can establish conditional distribution between past data and future data,and the result is a series of distributions with mean and standard deviation.The falling frequency of the actual value in three intervals is calculated to verify the reliability of the model.In conclusion,compared with the traditional artificial neural network model,the probabilistic prediction method can also provide the upper and lower boundary information with corresponding probability for the prediction parameters.In addition,the RMSE of the model is 0.58 higher than that of ANN.In order to ensure the prediction performance,random forest algorithm is used to select key input variables for probability prediction.The key input variables and early prediction values are selected as inputs to establish the probability prediction model.This method can reasonably predict the probability distribution of environmental parameters of subway station.This work will help to prevent the occurrence of emergencies and reasonably regulate the environmental control system.更多还原