WO2021013190A1 - Meteorological parameter-based high-speed train positioning method and system in navigation blind zone - Google Patents

Meteorological parameter-based high-speed train positioning method and system in navigation blind zone Download PDF

Info

Publication number
WO2021013190A1
WO2021013190A1 PCT/CN2020/103576 CN2020103576W WO2021013190A1 WO 2021013190 A1 WO2021013190 A1 WO 2021013190A1 CN 2020103576 W CN2020103576 W CN 2020103576W WO 2021013190 A1 WO2021013190 A1 WO 2021013190A1
Authority
WO
WIPO (PCT)
Prior art keywords
tunnel
model
samples
sample
cluster
Prior art date
Application number
PCT/CN2020/103576
Other languages
French (fr)
Chinese (zh)
Inventor
刘辉
吴海平
李燕飞
李烨
Original Assignee
中南大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中南大学 filed Critical 中南大学
Priority to RU2021122153A priority Critical patent/RU2771515C1/en
Priority to SG11202103075TA priority patent/SG11202103075TA/en
Priority to AU2020316538A priority patent/AU2020316538B2/en
Publication of WO2021013190A1 publication Critical patent/WO2021013190A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/14—Image acquisition
    • G06V30/146—Aligning or centring of the image pick-up or image-field
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/19—Recognition using electronic means

Definitions

  • the invention relates to a train positioning technology in a long and large tunnel, in particular to a method and system for positioning a blind area of high-speed train navigation based on meteorological parameters.
  • the tunnel navigation information simulation system obtains information through the satellite signal simulator and then generates the simulated navigation information and sends it to the target train through the optical cable. It can realize the continuous navigation and positioning simulation in the tunnel, reduce the cost, and solve the problem of information loss and time discontinuity in the tunnel. the goal of.
  • optical cable groups required, high requirements for hardware conditions, and special design for different terrain locations in practical applications, which has poor universality.
  • Trackside train positioning device acquires trackside equipment images during train operation, and realizes train positioning through trackside intelligent identification device combined with electronic maps.
  • the recognition accuracy will increase with the increase of the camera's frame rate, but it is necessary to arrange trackside equipment along the route, and the maintenance cost is high.
  • the technical problem to be solved by the present invention is to provide a high-speed train navigation blind zone positioning method and system based on meteorological parameters in order to solve the problem of difficult train positioning in the typical navigation blind zone of a long tunnel and reduce positioning cost.
  • the technical solution adopted by the present invention is: a method for locating blind areas of high-speed train navigation based on meteorological parameters, including the following steps:
  • HSV color space template library uses the HSV color space template library to train an HSV template matching model; use the data of similar tunnel groups to train a correlation vector machine to establish a tunnel mileage prediction model;
  • the invention uses artificial intelligence big data analysis technology to establish a mileage prediction model. After the modeling is completed, only the vehicle-mounted sensor is needed to realize the input data collection without any trackside equipment, and the system construction cost and maintenance cost are reduced.
  • the specific process of constructing the tunnel meteorological parameter database includes: collecting temperature series, humidity series, longitude, latitude, time-average temperature, time-average humidity, and time-average solar radiation collected when the train passes through a tunnel at one time.
  • the predicted values constitute a set of samples of tunnel meteorological parameters; the samples of tunnel meteorological parameters collected during the operation of all trains in the jurisdiction within one year constitute the tunnel meteorological parameter database.
  • the database construction process of the present invention can ensure the acquisition of a sufficient number of sample data, and improve the prediction accuracy of the prediction fusion model.
  • step S2 includes:
  • the input attributes are roughly divided into the tunnel category, and the OPTICS algorithm is used to sort the output of the tunnel group samples, and the reachable distance of the sorted sequence is compared with the set neighborhood distance parameter ⁇ .
  • the reachable distance in the sequence is less than the neighborhood distance parameter Consecutive samples of ⁇ form a cluster sample cluster; obtain the cluster center of each cluster sample cluster And the T1 samples closest to the cluster center among them Corresponds to the processed longitude and latitude, Corresponds to the processed weather parameters of the tunnel; the cluster center And T1 samples are defined as the representative samples of the current type of tunnel population, and T1 samples in the representative samples are the typical tunnel samples of the current type of tunnel population.
  • step S2 includes:
  • step b) Taking the depth feature attribute set as input, follow the process of step b) to obtain the cluster center of each cluster sample cluster of the secondary clustering And the T1 samples closest to the cluster center; the cluster center And T1 samples are defined as the representative samples of the secondary clustering of the current type of tunnel population, and T1 samples in the secondary clustering of the representative samples are the typical tunnel samples of the current type of tunnel population.
  • the implementation process of the above step S2 realizes the accurate classification of the tunnel group, which can ensure that the data of the subsequent HSV color space template library is more accurate, thereby improving the prediction accuracy of the prediction fusion model.
  • step S3 the specific implementation process of constructing a typical sequence HSV color space template library using the typical tunnel sample includes: taking the temperature and humidity parameter sequence that changes with the tunnel mileage in the typical tunnel sample as the template sequence of the corresponding type of tunnel group, and setting Delay time and window length, use the delay coordinate method to reconstruct the phase space of the time series of temperature, humidity and temperature difference sequence, obtain three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity, and reconstitute the three two-dimensional reconstruction matrices The combination is combined according to the HSV color space to form a color image, that is, the template image in the typical sequence HSV color space template library is obtained.
  • This realization constructs a two-dimensional reconstruction matrix through temperature, humidity and temperature difference, and the calculation process is simple.
  • step S4 the specific implementation process of training the HSV template matching model includes:
  • the best matching position of the current train position in the template library can be determined, and the prediction accuracy can be further improved.
  • step S4 the specific implementation process of training the correlation vector machine and establishing the tunnel mileage prediction model includes:
  • Binary encoding is performed on the features of each dimension in the input sample I.
  • the encoding value corresponding to the feature of a certain dimension is 1, the feature is selected as the input variable of the RVM model.
  • the encoding value corresponding to the feature of a certain dimension is At 0, the feature of this dimension is discarded;
  • the RVM model is the tunnel mileage prediction model; among them, To verify the true output value in the set.
  • Using the correlation vector machine (RVM) to obtain the tunnel mileage prediction model can improve the prediction accuracy of the tunnel mileage prediction model.
  • the specific implementation process of predicting the train position includes: calculating the RVM model input vector of the current sample point, substituting the RVM model input vector into the target model, and obtaining the tunnel mileage prediction model output value; obtaining the HSV template matching of the current sample point
  • the model input value is substituted into the HSV template matching model to obtain the HSV template matching model output value; the mileage prediction model output value and the HSV template matching model output value are substituted into the mileage prediction fusion model to obtain the final train position prediction result;
  • the target template is Refers to the model trained under the secondary clustering target sample cluster; secondary clustering target sample cluster refers to the sample cluster corresponding to the minimum value between the secondary clustering representation samples from the current sample point to all primary clustering target sample clusters;
  • the first clustering target sample cluster refers to the sample cluster corresponding to the minimum value between the current sample point and all the cluster characterization samples;
  • the cluster characterization sample refers to the samples in the cluster sample cluster, and the cluster sample cluster refers to The input attributes are roughly divided by the tunnel category
  • the present invention also provides a high-speed train navigation blind spot positioning system based on meteorological parameters, including a sensor installed on the train for collecting meteorological parameters in the tunnel; the sensor communicates with the computer equipment; the computer The device is programmed or configured to perform the steps of the method described in the present invention.
  • the present invention has the following beneficial effects:
  • the present invention makes full use of artificial intelligence big data analysis technology, fully excavates the potential law of tunnel environment parameters changing with tunnel depth, and solves the problem of difficult train positioning in the typical navigation blind zone of long tunnel from the perspective of data-driven modeling ;
  • the present invention only needs the vehicle-mounted temperature and humidity sensor to realize the input data collection after the modeling is completed, without any trackside equipment, and reduces the system construction cost and maintenance cost.
  • Figure 1 is a flowchart of data collection and rough classification of categories
  • Figure 2 is a flow chart of the secondary classification of categories and model establishment
  • Figure 3 shows the combination of time series phase space reconstruction and HSV color space
  • Figure 4 is a flow chart of the test process model call.
  • Embodiment 1 of the present invention provides a method for locating blind spots in high-speed train navigation, which specifically includes the following steps:
  • Step 1 Collect the weather parameters of the tunnel and construct the weather parameter database of the tunnel
  • the on-board sensor collects the temperature, humidity and mileage time series in the tunnel when the train passes in real time, and the sampling interval is 0.1s.
  • the temperature series, humidity series, the latitude and longitude of the area where the train passes through a tunnel at one time, and the predicted values of time-average temperature, time-average humidity, and time-average solar radiation constitute a set of tunnel meteorological parameter samples.
  • the samples of tunnel meteorological parameters collected during the operation of all trains within one year constitute the tunnel meteorological parameter database.
  • Step 2 Multi-scale and hierarchical classification of tunnel meteorological parameters
  • the temperature series, humidity series, latitude and longitude (longitude and latitude) of the area collected when passing through a tunnel as well as hourly average temperature, hourly average humidity, hourly average solar radiation (corresponding to hourly average temperature, Hourly average humidity, hourly average solar radiation, that is, the predicted value of the hourly average) is used as information input, and multi-type attributes, multi-feature scales, and multi-level categories are divided to realize the classification of tunnel groups with similar attributes .
  • the specific implementation process is as follows:
  • Step A1 Use the spherical coordinate conversion relationship to convert the latitude and longitude coordinates into plane coordinates. Normalize the converted latitude and longitude, hourly average temperature, hourly average humidity, and hourly average solar radiation to form a set of input attributes for the rough classification of tunnel categories.
  • Step A2 Set the initial neighborhood distance parameter ⁇ to 0.1, the initial neighborhood sample number parameter MinPts to 5, and use the OPTICS algorithm to sort the output of the tunnel group samples with the input attributes of the tunnel category coarsely divided.
  • the reachable distance of the sequence (Reachability-distance, a concept defined by the OPTIC algorithm, is the minimum value of the core distance and the Euclidean distance, see https://blog.csdn.net/han1202012/article/details/105936710)
  • the set neighborhood distance parameter ⁇ consecutive samples whose reachable distance is less than the set value in the sequence constitute a sample cluster.
  • Cluster centers And 5 samples Defined as a representative sample of the current category.
  • Step A3 further dividing each cluster sample cluster obtained by roughly dividing the input attribute set based on the tunnel category. Specifically, it includes the following sub-steps: 1 Mirror and extend the temperature time sequence and humidity sequence when the train passes through the tunnel in the sample cluster, and convert the temperature sequence and humidity sequence in the sample cluster into a sequence whose length is equal to the longest sample length. 2Set the delay time to 1, the window length to 5, and use the Delay-coordinate method to reconstruct the temperature sequence and humidity sequence in phase space to obtain a two-dimensional reconstruction matrix representing the evolution characteristics of temperature and humidity. (3) Input the reconstruction matrix of the sample into the training autoencoder to obtain the depth feature attribute set for further division.
  • step A2 uses the depth feature attribute set as input, follow the procedure in step A2 to obtain the cluster center of each cluster sample cluster of the secondary clustering And the 5 samples closest to the cluster center among them Corresponding to the 5 dimensional variables in the depth feature attribute set.
  • Cluster centers And 5 samples Defined as the representative sample of the secondary clustering of the current category.
  • Step 3 Build a typical sequence HSV color space template library
  • the 5 tunnel samples closest to the cluster center corresponding to the group are selected As a typical tunnel sample under this category.
  • the temperature and humidity parameter sequence that varies with the tunnel mileage in the typical sample is used as the template sequence of this category, the delay time is set to 1, the window length is 5, and the delay coordinate method is used to determine the temperature, humidity and temperature difference.
  • Step 4 Train the HSV template matching model
  • phase space reconstruction and HSV color space combination are performed to construct a current position feature image.
  • Step B1 Collect the current sample point and the previous 19 sample points in the temperature, humidity, and temperature differential time series.
  • Step B2 Set the delay time to 1, the window length to 5, and use the delay coordinate method to reconstruct the phase space, obtain three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity, and combine the three matrices according to the HSV color space , Forming the current location feature image h.
  • Step B3 Convolve the current position feature image with the image in the template library
  • Each g i is a one-dimensional sequence.
  • Step 5 Train the RVM recognition model
  • the output sample is the mileage value O corresponding to the current position.
  • Step C2 Divide training samples, verification samples, and test samples. Using random sampling without replacement, 60% of the M samples are selected as the training set, 30% as the verification set, and 10% as the test set.
  • Step C3 Determine the optimization object and initialize the optimization value.
  • the binary whale algorithm is used to optimize the input features of the model, that is, the feature of each dimension in the input sample I is binary coded.
  • the code value corresponding to the feature of a certain dimension is 1, the feature is selected as the input variable of the RVM model.
  • the code value corresponding to a feature of a certain dimension is 0, the feature of this dimension will be discarded.
  • the 43 dimensional features are randomly initialized and coded as 0 or 1.
  • Step C5 Output the optimized prediction model.
  • the binary whale algorithm is used for iterative optimization operations to determine the optimal input features and RVM model, which is the RVM mileage prediction model.
  • Step 6 Construct HSV template matching and RVM recognition fusion model
  • Step 7 Obtain input data, call mileage prediction fusion model
  • the current temperature, air pressure and solar radiation data outside the tunnel are obtained from the weather station in the area.
  • the temperature and humidity sensors installed at the head and tail of the train are used to obtain the current temperature and humidity sequence. Specifically include the following steps:
  • Step D1 Refer to the process of step 2 to obtain the input attributes used for a clustering among them Corresponding to the five attributes of the converted longitude, latitude, hourly average temperature, hourly average humidity, and hourly average solar radiation. Refer to the process of step 2 to obtain the input attributes used for secondary clustering among them Corresponds to 5 attributes in the depth feature attribute set.
  • Step D2 Calculate the feature value of the current sample point The distance from the first clustering characterization sample
  • Step D3 Calculate the feature value of the current sample point The distance from the second clustering characterization sample
  • the sample cluster corresponding to the minimum value among the secondary clustering representation samples from the current sample point to all primary clustering target sample clusters is taken as the secondary clustering target sample cluster.
  • the model and template library trained under this sample cluster are the target model and the target template library.
  • Step 8 Predict the train position
  • step 5 to calculate the RVM model input vector of the current sample point, and substitute the RVM model input vector into the target model to obtain the target RVM model output value.
  • step 4 to obtain the template matching model input value of the current sample point, bring it into the target template matching model, and obtain the target template matching model output value.
  • Equation 6 to obtain the final train position prediction result.
  • Embodiment 2 of the present invention provides a high-speed train navigation blind spot positioning system based on meteorological parameters, including a sensor installed on the train for collecting meteorological parameters in the tunnel; the sensor communicates with a computer device; the computer device is It is programmed or configured to execute the steps of the method of Embodiment 1 of the present invention.

Abstract

Disclosed is a meteorological parameter-based high-speed train positioning method in a navigation blind zone, comprising: collecting tunnel's meteorological parameters; classifying the collected tunnel's meteorological parameters; constructing a typical sequence HSV color space template library by using the classified tunnel's meteorological parameters; training the typical sequence HSV color space template library; training an HSV template matching model; training an RVM recognition model; constructing a fusion model of the HSV template matching model and the RVM recognition model to obtain a mileage prediction fusion model; and acquiring input data, and calling the mileage prediction fusion model to predict the location of the train. According to the present invention, the artificial intelligence big data analysis technology is fully utilized, and the potential law that the environmental parameters in the tunnel change with the tunnel depth is fully mined. From the perspective of data-driven modeling, the problem of train positioning in a long tunnel, i.e., a typical navigation blind zone, is solved.

Description

基于气象参数的高速列车导航盲区定位方法及系统Method and system for positioning blind area of high-speed train navigation based on meteorological parameters 技术领域Technical field
本发明涉及长大隧道内列车定位技术，具体是一种基于气象参数的高速列车导航盲区定位方法及系统。The invention relates to a train positioning technology in a long and large tunnel, in particular to a method and system for positioning a blind area of high-speed train navigation based on meteorological parameters.
背景技术Background technique
近年来，越来越多铁路线路铺往中国的西部地区，随着地势和地形的变化，线路不免经过很多高山地区，在隧道内搜索不到导航卫星的信号，会造成短暂信息失联，形成导航盲区，危及行车安全。隧道内列车准确定位对保证列车行车安全具有重要意义。In recent years, more and more railway lines have been laid to the western region of China. With the changes in topography and topography, the lines will inevitably pass through many mountainous areas, and the navigation satellite signals cannot be searched in the tunnel, which will cause temporary information loss. Navigation blind spots endanger driving safety. The accurate positioning of the train in the tunnel is of great significance to ensure the safety of the train.
目前国内对于长大隧道(单洞长度10千米以上的隧道)这一典型导航盲区内的列车定位方向的研究尚处于初步阶段，为避免由于信号缺失而造成的定位盲区和行车事故，需开发定位准确、成本适合且易实施的定位装备及方法。目前的列车隧道定位方法有以下几种：At present, the domestic research on the train positioning direction in the typical navigation blind zone of long tunnels (tunnels with a single tunnel length of more than 10 kilometers) is still in the preliminary stage. In order to avoid positioning blind spots and traffic accidents caused by signal loss, development is needed Positioning equipment and methods that are accurate, cost-effective, and easy to implement. The current train tunnel positioning methods are as follows:
隧道导航信息仿真系统，通过卫星信号模拟器获取信息再生成仿真导航信息经过光缆发送至目标列车，能够实现载隧道内的连续导航定位仿真，达到降低成本，解决隧道内信息丢失和时间不连续问题的目的。但所需光缆组较多，对硬件条件需求高，且在实际应用中需要针对不同地形位置特别设计，普适性差。The tunnel navigation information simulation system obtains information through the satellite signal simulator and then generates the simulated navigation information and sends it to the target train through the optical cable. It can realize the continuous navigation and positioning simulation in the tunnel, reduce the cost, and solve the problem of information loss and time discontinuity in the tunnel. the goal of. However, there are many optical cable groups required, high requirements for hardware conditions, and special design for different terrain locations in practical applications, which has poor universality.
轨旁列车定位装置，在列车运行过程中获取轨旁设备图像，通过轨旁智能识别装置结合电子地图，实现对列车的定位。识别精度会随相机的帧率的提升而提升，但需要在沿线布置轨旁设备，维护成本较高。Trackside train positioning device acquires trackside equipment images during train operation, and realizes train positioning through trackside intelligent identification device combined with electronic maps. The recognition accuracy will increase with the increase of the camera's frame rate, but it is necessary to arrange trackside equipment along the route, and the maintenance cost is high.
此外还有测速计算式定位和应答式定位等技术，但同样存在定位精度不高或者维护成本过高的问题，综上可知现有的隧道内列车定位技术难以在保证较高的定位精度下实现大面积推广。In addition, there are technologies such as speed measurement calculation type positioning and response type positioning, but there are also problems of low positioning accuracy or high maintenance costs. In summary, it can be seen that the existing tunnel positioning technology is difficult to achieve with high positioning accuracy. Large-scale promotion.
发明内容Summary of the invention
本发明所要解决的技术问题是，针对现有技术不足，提供一种基于气象参数的高速列车导航盲区定位方法及系统，解决长大隧道典型导航盲区内列车定位难的问题，降低定位成本。The technical problem to be solved by the present invention is to provide a high-speed train navigation blind zone positioning method and system based on meteorological parameters in order to solve the problem of difficult train positioning in the typical navigation blind zone of a long tunnel and reduce positioning cost.
为解决上述技术问题，本发明所采用的技术方案是：一种基于气象参数的高速列车导航盲区定位方法，包括以下步骤：In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for locating blind areas of high-speed train navigation based on meteorological parameters, including the following steps:
S1、实时采集列车通过时隧道内的隧道气象参数，构建隧道气象参数数据库；S1. Collect the tunnel meteorological parameters in the tunnel in real time when the train passes, and construct the tunnel meteorological parameter database;
S2、基于所述隧道气象参数数据库，对属性相近的隧道群体进行归类，获取各类别 隧道群体的典型隧道样本；S2, based on the tunnel meteorological parameter database, classify tunnel groups with similar attributes, and obtain typical tunnel samples of each type of tunnel group;
S3、利用所述典型隧道样本构建典型序列HSV颜色空间模板库；S3. Use the typical tunnel samples to construct a typical sequence HSV color space template library;
S4、利用所述HSV颜色空间模板库训练HSV模板匹配模型；采用同类隧道群体的数据，训练相关向量机，建立隧道里程预测模型；S4. Use the HSV color space template library to train an HSV template matching model; use the data of similar tunnel groups to train a correlation vector machine to establish a tunnel mileage prediction model;
S5、构建HSV模板匹配模型与隧道里程预测模型的融合模型，得到里程预测融合模型；S5. Construct a fusion model of the HSV template matching model and the tunnel mileage prediction model to obtain the mileage prediction fusion model;
S6、获取列车运行过程中的隧道气象参数数据，调用里程预测融合模型，预测列车位置。S6. Obtain the tunnel meteorological parameter data during train operation, call the mileage prediction fusion model, and predict the train position.
本发明利用人工智能大数据分析技术，建立里程预测模型，建模完成后只需要车载传感器即可实现输入数据采集，无需任何轨旁设备，降低了系统建设成本和维护成本。The invention uses artificial intelligence big data analysis technology to establish a mileage prediction model. After the modeling is completed, only the vehicle-mounted sensor is needed to realize the input data collection without any trackside equipment, and the system construction cost and maintenance cost are reduced.
步骤S1中，构建隧道气象参数数据库的具体过程包括：采集列车一次通过某隧道时采集的温度序列、湿度序列、所在区域的经度、纬度以及时平均气温、时平均湿度、时平均太阳辐射量的预测值，构成1组隧道气象参数样本；辖区内所有列车1年内运行采集的隧道气象参数样本构成隧道气象参数数据库。本发明的数据库构建过程可以保证获取足够数量的样本数据，提高预测融合模型的预测精度。In step S1, the specific process of constructing the tunnel meteorological parameter database includes: collecting temperature series, humidity series, longitude, latitude, time-average temperature, time-average humidity, and time-average solar radiation collected when the train passes through a tunnel at one time. The predicted values constitute a set of samples of tunnel meteorological parameters; the samples of tunnel meteorological parameters collected during the operation of all trains in the jurisdiction within one year constitute the tunnel meteorological parameter database. The database construction process of the present invention can ensure the acquisition of a sufficient number of sample data, and improve the prediction accuracy of the prediction fusion model.
步骤S2的具体实现过程包括：The specific implementation process of step S2 includes:
a)将隧道所在区域经度、纬度坐标转换为平面坐标；对隧道气象参数进行归一化处理，处理后的经度、纬度坐标和隧道气象参数组成隧道类别粗划分输入属性集合；a) Convert the longitude and latitude coordinates of the area where the tunnel is located into plane coordinates; normalize the tunnel meteorological parameters, and the processed longitude, latitude coordinates and tunnel meteorological parameters form a set of input attributes for the rough division of tunnel categories;
b)以隧道类别粗划分输入属性为对象，采用OPTICS算法进行隧道群样本输出排序，比较排序后的序列的可达距离与设定邻域距离参数ε，序列中可达距离小于邻域距离参数ε的连续样本构成一个聚类样本簇；获取每个聚类样本簇的聚类中心
Figure PCTCN2020103576-appb-000001
及距离聚类中心最近的T1个样本
Figure PCTCN2020103576-appb-000002
其中
Figure PCTCN2020103576-appb-000003
对应于处理后的经度和纬度，
Figure PCTCN2020103576-appb-000004
对应于处理后的隧道气象参数；将聚类中心
Figure PCTCN2020103576-appb-000005
和T1个样本定义为当前类别隧道群体的表征样本，该表征样本中的T1个样本即为当前类别隧道群体的典型隧道样本。 b) The input attributes are roughly divided into the tunnel category, and the OPTICS algorithm is used to sort the output of the tunnel group samples, and the reachable distance of the sorted sequence is compared with the set neighborhood distance parameter ε. The reachable distance in the sequence is less than the neighborhood distance parameter Consecutive samples of ε form a cluster sample cluster; obtain the cluster center of each cluster sample cluster
Figure PCTCN2020103576-appb-000001
And the T1 samples closest to the cluster center
Figure PCTCN2020103576-appb-000002
among them
Figure PCTCN2020103576-appb-000003
Corresponds to the processed longitude and latitude,
Figure PCTCN2020103576-appb-000004
Corresponds to the processed weather parameters of the tunnel; the cluster center
Figure PCTCN2020103576-appb-000005
And T1 samples are defined as the representative samples of the current type of tunnel population, and T1 samples in the representative samples are the typical tunnel samples of the current type of tunnel population.
通过上述过程，实现了属性相近的隧道群体的粗分类。Through the above process, the rough classification of tunnel groups with similar attributes is realized.
步骤S2的具体实现过程包括：The specific implementation process of step S2 includes:
1)将聚类样本簇中列车通过隧道时的温度时间序列和湿度序列进行镜像延拓，将聚类样本簇内的温度序列、湿度序列转化为长度分别等于各自最长样本长度的序列；1) Mirror and extend the temperature time sequence and humidity sequence when the train passes through the tunnel in the cluster sample cluster, and convert the temperature sequence and humidity sequence in the cluster sample cluster into a sequence whose length is equal to the length of the longest sample;
2)设定延迟时间和窗口长度，采用延迟坐标法对温度、湿度时间最长样本长度的序列进行相空间重构，获取代表温度、湿度演化特性的二维重构矩阵；2) Set the delay time and window length, and use the delay coordinate method to reconstruct the phase space of the sequence with the longest sample length of temperature and humidity time to obtain a two-dimensional reconstruction matrix representing the evolution characteristics of temperature and humidity;
3)将所述二维重构矩阵输入训练自动编码器，获取用于进一步划分的深度特征属性集合；3) Input the two-dimensional reconstruction matrix into the training autoencoder to obtain the depth feature attribute set for further division;
4)将深度特征属性集合作为输入，按照步骤b)的流程获取二次聚类的每个聚类样本簇的聚类中心
Figure PCTCN2020103576-appb-000006
及距离聚类中心最近的T1个样本；将聚类中心
Figure PCTCN2020103576-appb-000007
和T1个样本定义为当前类别隧道群体二次聚类的表征样本，该二次聚类的表征样本中的T1个样本即为当前类别隧道群体的典型隧道样本。 4) Taking the depth feature attribute set as input, follow the process of step b) to obtain the cluster center of each cluster sample cluster of the secondary clustering
Figure PCTCN2020103576-appb-000006
And the T1 samples closest to the cluster center; the cluster center
Figure PCTCN2020103576-appb-000007
And T1 samples are defined as the representative samples of the secondary clustering of the current type of tunnel population, and T1 samples in the secondary clustering of the representative samples are the typical tunnel samples of the current type of tunnel population.
上述步骤S2的实现过程实现了隧道群体的精确分类，可以保证后续HSV颜色空间模板库的数据更加精确，进而提高预测融合模型的预测精度。The implementation process of the above step S2 realizes the accurate classification of the tunnel group, which can ensure that the data of the subsequent HSV color space template library is more accurate, thereby improving the prediction accuracy of the prediction fusion model.
步骤S3中，利用所述典型隧道样本构建典型序列HSV颜色空间模板库的具体实现过程包括：将典型隧道样本中随隧道里程变化的气温和湿度参数序列作为对应类别隧道群体的模板序列，设定延迟时间和窗口长度，采用延迟坐标法对温度、湿度和温度差分序列时间序列进行相空间重构，获取三个代表温度、湿度演化特性的二维重构矩阵，将三个二维重构矩阵组合按照HSV颜色空间组合，形成彩色图像，即得到典型序列HSV颜色空间模板库中的模板图像。该实现通过温度、湿度和温度差构建二维重构矩阵，计算过程简单。In step S3, the specific implementation process of constructing a typical sequence HSV color space template library using the typical tunnel sample includes: taking the temperature and humidity parameter sequence that changes with the tunnel mileage in the typical tunnel sample as the template sequence of the corresponding type of tunnel group, and setting Delay time and window length, use the delay coordinate method to reconstruct the phase space of the time series of temperature, humidity and temperature difference sequence, obtain three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity, and reconstitute the three two-dimensional reconstruction matrices The combination is combined according to the HSV color space to form a color image, that is, the template image in the typical sequence HSV color space template library is obtained. This realization constructs a two-dimensional reconstruction matrix through temperature, humidity and temperature difference, and the calculation process is simple.
步骤S4中，训练HSV模板匹配模型的具体实现过程包括：In step S4, the specific implementation process of training the HSV template matching model includes:
1)采集温度、湿度和温度差分时间序列中当前样本点及当前样本点往前的N个采样点；1) Collect the current sample point and N sampling points before the current sample point in the temperature, humidity and temperature differential time series;
2)设定延迟时间和窗口长度，采用延迟坐标法进行相空间重构，获取三个代表温度、湿度演化特性的二维重构矩阵，将三个矩阵组合按照HSV颜色空间组合，形成当前位置特征图像；2) Set the delay time and window length, use the delay coordinate method to reconstruct the phase space, obtain three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity, and combine the three matrices according to the HSV color space to form the current position Feature image
3)将当前位置特征图像与典型序列HSV颜色空间模板库中的模板图像进行卷积运算，得到多个一维序列；3) Convolve the current position feature image with the template image in the HSV color space template library of the typical sequence to obtain multiple one-dimensional sequences;
4)对所有一维序列中的元素进行由大到小的排序，确定的最大的T2个元素为 候选元素，候选元素对应的排序前所在位置为候选位置，获取候选位置对应的里程值；4) Sort all elements in the one-dimensional sequence from big to small, determine the largest T2 elements as candidate elements, and the position corresponding to the candidate element before sorting is the candidate position, and obtain the mileage value corresponding to the candidate position;
5)对所有候选位置对应的所有里程值取均值，得到当前的HSV模板匹配模型输出序列。5) Take the average of all mileage values corresponding to all candidate positions to obtain the current HSV template matching model output sequence.
通过上述过程可以确定列车当前位置在模板库中的最佳匹配位置，进一步提高预测精度。Through the above process, the best matching position of the current train position in the template library can be determined, and the prediction accuracy can be further improved.
步骤S4中，训练相关向量机，建立隧道里程预测模型的具体实现过程包括：In step S4, the specific implementation process of training the correlation vector machine and establishing the tunnel mileage prediction model includes:
1)定义输入样本I＝(T,H,t 0,h 0,r 0),其中T＝(t 1,t 2…,t M)为隧道内当前样本点及往前M个样本点的温度时间序列，H＝(h 1,h 2…,h 19,h M)为隧道内的当前样本点及往前N个样本点的湿度序列，t 0,h 0和r 0分别为从气象站获取的时平均气温、时平均湿度和时平均太阳辐射量的预测值；定义输出样本为当前位置对应的里程值O，输入和输出组合Y＝{I,O}构成建模样本；针对每个同类隧道群体，选取M个建模样本； 1) Define the input sample I = (T, H, t 0 , h 0 , r 0 ), where T = (t 1 , t 2 …, t M ) is the current sample point in the tunnel and the previous M sample points Temperature time series, H = (h 1 , h 2 …, h 19 , h M ) is the current sample point in the tunnel and the humidity sequence of the previous N sample points, t 0 , h 0 and r 0 are the meteorological The predicted values of hourly average temperature, hourly average humidity, and hourly average solar radiation obtained by the station; define the output sample as the mileage value O corresponding to the current position, and the input and output combination Y={I,O} constitutes a modeling sample; A group of similar tunnels, select M modeling samples;
2)将M个建模样本随机划分训练集、验证集和测试集；2) Randomly divide M modeling samples into training set, validation set and test set;
3)对输入样本I中每个维度的特征进行二进制编码，当某维度的特征对应的编码值为1时，该特征被选择作为RVM模型的输入变量，当某维度的特征对应的编码值为0时，该维度的特征被舍弃；3) Binary encoding is performed on the features of each dimension in the input sample I. When the encoding value corresponding to the feature of a certain dimension is 1, the feature is selected as the input variable of the RVM model. When the encoding value corresponding to the feature of a certain dimension is At 0, the feature of this dimension is discarded;
4)基于当前特征编码值，确定新的输入特征，更新训练集、验证集和测试集，采用更新后的训练集数据训练RVM模型，将更新后的验证集数据输入训练好的RVM模型，获取模型输出序列
Figure PCTCN2020103576-appb-000008
M 1＝0.3M； 4) Determine the new input feature based on the current feature code value, update the training set, validation set, and test set, use the updated training set data to train the RVM model, and input the updated validation set data into the trained RVM model to obtain Model output sequence
Figure PCTCN2020103576-appb-000008
M 1 =0.3M;
5)重复步骤)和步骤4)，确定使优化目标函数
Figure PCTCN2020103576-appb-000009
最小的最优的输入特征和RVM模型，该RVM模型即为隧道里程预测模型；其中，
Figure PCTCN2020103576-appb-000010
为验证集中的真实输出值。 5) Repeat steps) and step 4) to determine the optimization objective function
Figure PCTCN2020103576-appb-000009
The smallest and optimal input features and RVM model, the RVM model is the tunnel mileage prediction model; among them,
Figure PCTCN2020103576-appb-000010
To verify the true output value in the set.
利用相关向量机(RVM)获取隧道里程预测模型，可以提高隧道里程预测模型的预测精度。Using the correlation vector machine (RVM) to obtain the tunnel mileage prediction model can improve the prediction accuracy of the tunnel mileage prediction model.
HSV模板匹配模型与隧道里程预测模型的融合模型
Figure PCTCN2020103576-appb-000011
其中， k1＝1,2…M 2，M 2＝0.1M；
Figure PCTCN2020103576-appb-000012
Figure PCTCN2020103576-appb-000013
为将测试集数据输入隧道里程预测模型后得到的输出序列；
Figure PCTCN2020103576-appb-000014
为测试集中的真实输出结果；
Figure PCTCN2020103576-appb-000015
为将测试集输入HSV模板匹配模型后得到的输出序列。 Fusion model of HSV template matching model and tunnel mileage prediction model
Figure PCTCN2020103576-appb-000011
Among them, k1 = 1, 2...M 2 , M 2 =0.1M;
Figure PCTCN2020103576-appb-000012
Figure PCTCN2020103576-appb-000013
It is the output sequence obtained after inputting the test set data into the tunnel mileage prediction model;
Figure PCTCN2020103576-appb-000014
Real output results in the test set;
Figure PCTCN2020103576-appb-000015
It is the output sequence obtained after inputting the test set into the HSV template matching model.
步骤S6中，预测列车位置的具体实现过程包括：计算当前样本点的RVM模型输入向量，将该RVM模型输入向量代入目标模型中，获取隧道里程预测模型输出值；获取当前样本点的HSV模板匹配模型输入值，代入HSV模板匹配模型中，获取HSV模板匹配模型输出值；将里程预测模型输出值和HSV模板匹配模型输出值代入里程预测融合模型，获取最终列车位置预测结果；其中，目标模板是指二次聚类目标样本簇下训练的模型；二次聚类目标样本簇是指当前样本点到所有一次聚类目标样本簇下属的二次聚类表征样本间最小值对应的样本簇；所述一次聚类目标样本簇是指当前样本点到所有聚类表征样本间最小值对应的样本簇；所述聚类表征样本是指聚类样本簇中的样本，所述聚类样本簇是指以隧道类别粗划分输入属性为对象，采用OPTICS算法进行隧道群样本输出排序得到的序列的可达距离与设定邻域距离参数ε之间的值小于邻域距离参数ε的连续样本构成的样本簇。In step S6, the specific implementation process of predicting the train position includes: calculating the RVM model input vector of the current sample point, substituting the RVM model input vector into the target model, and obtaining the tunnel mileage prediction model output value; obtaining the HSV template matching of the current sample point The model input value is substituted into the HSV template matching model to obtain the HSV template matching model output value; the mileage prediction model output value and the HSV template matching model output value are substituted into the mileage prediction fusion model to obtain the final train position prediction result; where the target template is Refers to the model trained under the secondary clustering target sample cluster; secondary clustering target sample cluster refers to the sample cluster corresponding to the minimum value between the secondary clustering representation samples from the current sample point to all primary clustering target sample clusters; The first clustering target sample cluster refers to the sample cluster corresponding to the minimum value between the current sample point and all the cluster characterization samples; the cluster characterization sample refers to the samples in the cluster sample cluster, and the cluster sample cluster refers to The input attributes are roughly divided by the tunnel category as the object, and the OPTICS algorithm is used to sort the output of the tunnel group samples. The range between the reachable distance of the sequence and the set neighborhood distance parameter ε is less than the neighborhood distance parameter ε. cluster.
相应地，本发明还提供了一种基于气象参数的高速列车导航盲区定位系统，包括安装于列车上的、用于采集隧道内的气象参数的传感器；所述传感器与计算机设备通信；所述计算机设备被编程或配置为执行本发明所述方法的步骤。Correspondingly, the present invention also provides a high-speed train navigation blind spot positioning system based on meteorological parameters, including a sensor installed on the train for collecting meteorological parameters in the tunnel; the sensor communicates with the computer equipment; the computer The device is programmed or configured to perform the steps of the method described in the present invention.
与现有技术相比，本发明所具有的有益效果为：Compared with the prior art, the present invention has the following beneficial effects:
1、本发明充分利用人工智能大数据分析技术，充分挖掘隧道内环境参数随隧道深度变化的潜在规律，从数据驱动建模的角度解决了长大隧道这一典型导航盲区内列车定位难的问题；1. The present invention makes full use of artificial intelligence big data analysis technology, fully excavates the potential law of tunnel environment parameters changing with tunnel depth, and solves the problem of difficult train positioning in the typical navigation blind zone of long tunnel from the perspective of data-driven modeling ；
2、本发明在建模完成后只需要车载温湿度传感器即可实现输入数据采集，无需任何轨旁设备，降低了系统建设成本和维护成本。2. The present invention only needs the vehicle-mounted temperature and humidity sensor to realize the input data collection after the modeling is completed, without any trackside equipment, and reduces the system construction cost and maintenance cost.
附图说明Description of the drawings
图1为数据采集及类别粗划分流程图；Figure 1 is a flowchart of data collection and rough classification of categories;
图2为类别二次划分及模型建立流程图；Figure 2 is a flow chart of the secondary classification of categories and model establishment;
图3为时间序列相空间重构与HSV颜色空间组合；Figure 3 shows the combination of time series phase space reconstruction and HSV color space;
图4为测试过程模型调用流程图。Figure 4 is a flow chart of the test process model call.
具体实施方式Detailed ways
本发明实施例1提供了一种高速列车导航盲区定位方法，具体包括以下步骤：Embodiment 1 of the present invention provides a method for locating blind spots in high-speed train navigation, which specifically includes the following steps:
步骤1：采集隧道气象参数，构建隧道气象参数数据库Step 1: Collect the weather parameters of the tunnel and construct the weather parameter database of the tunnel
通过车载传感器实时采集列车通过时隧道内的温度和湿度和里程时间序列，采样间隔为0.1s。利用隧道所在地区的气象站获取当前位置的经纬度、时平均气温、时平均湿度、时平均太阳辐射量的预测值。列车一次通过某隧道时采集的温度序列、湿度序列、所在区域的经纬度以及时平均气温、时平均湿度、时平均太阳辐射量的预测值构成1组隧道气象参数样本。辖区内所有列车1年内运行采集的隧道气象参数样本构成隧道气象参数数据库。The on-board sensor collects the temperature, humidity and mileage time series in the tunnel when the train passes in real time, and the sampling interval is 0.1s. Use the weather station in the area where the tunnel is located to obtain the predicted value of the current location's latitude and longitude, hourly average temperature, hourly average humidity, and hourly average solar radiation. The temperature series, humidity series, the latitude and longitude of the area where the train passes through a tunnel at one time, and the predicted values of time-average temperature, time-average humidity, and time-average solar radiation constitute a set of tunnel meteorological parameter samples. The samples of tunnel meteorological parameters collected during the operation of all trains within one year constitute the tunnel meteorological parameter database.
步骤2：隧道气象参数多尺度分层分类Step 2: Multi-scale and hierarchical classification of tunnel meteorological parameters
基于隧道气象参数数据库，以通过某隧道时采集的温度序列、湿度序列、所在区域的经纬度(经度和纬度)以及时平均气温、时平均湿度、时平均太阳辐射量(分别对应逐小时平均气温，逐小时平均湿度，逐小时平均太阳辐射量，即每小时的平均值)的预测值作为信息输入，进行多类型属性、多特征尺度、多层次的类别划分，实现属性相近的隧道群体的归类。如图2和图3所示，具体实现流程如下：Based on the tunnel weather parameter database, the temperature series, humidity series, latitude and longitude (longitude and latitude) of the area collected when passing through a tunnel, as well as hourly average temperature, hourly average humidity, hourly average solar radiation (corresponding to hourly average temperature, Hourly average humidity, hourly average solar radiation, that is, the predicted value of the hourly average) is used as information input, and multi-type attributes, multi-feature scales, and multi-level categories are divided to realize the classification of tunnel groups with similar attributes . As shown in Figure 2 and Figure 3, the specific implementation process is as follows:
步骤A1:采用球坐标转换关系，将经纬度坐标转换为平面坐标。对转换后的经纬度、时平均气温、时平均湿度、时平均太阳辐射量进行归一化处理，形成隧道类别粗划分输入属性集合。Step A1: Use the spherical coordinate conversion relationship to convert the latitude and longitude coordinates into plane coordinates. Normalize the converted latitude and longitude, hourly average temperature, hourly average humidity, and hourly average solar radiation to form a set of input attributes for the rough classification of tunnel categories.
步骤A2:设定初始邻域距离参数ε为0.1，初始邻域样本数参数MinPts为5，以隧道类别粗划分输入属性为对象，采用OPTICS算法进行隧道群样本输出排序。将序列的可达距离(Reachability-distance，OPTIC算法定义的一个概念，是核心距离与欧几里得距离的最小值，见https://blog.csdn.net/han1202012/article/details/105936710)与设定邻域距离参数ε相比较，序列中可达距离小于设定值的连续样本构成一个样本簇。获取每个聚类样本簇的聚类中心
Figure PCTCN2020103576-appb-000016
及距离聚类中心最近的5个样本
Figure PCTCN2020103576-appb-000017
其中
Figure PCTCN2020103576-appb-000018
对应于转换后的经度、纬度、时平均气温、时平均湿度、时平均太阳辐射量等5个属性。将聚类中心
Figure PCTCN2020103576-appb-000019
和5个样本
Figure PCTCN2020103576-appb-000020
定义为当前类别的表征样本。 Step A2: Set the initial neighborhood distance parameter ε to 0.1, the initial neighborhood sample number parameter MinPts to 5, and use the OPTICS algorithm to sort the output of the tunnel group samples with the input attributes of the tunnel category coarsely divided. The reachable distance of the sequence (Reachability-distance, a concept defined by the OPTIC algorithm, is the minimum value of the core distance and the Euclidean distance, see https://blog.csdn.net/han1202012/article/details/105936710) Compared with the set neighborhood distance parameter ε, consecutive samples whose reachable distance is less than the set value in the sequence constitute a sample cluster. Get the cluster center of each cluster sample cluster
Figure PCTCN2020103576-appb-000016
And the 5 samples closest to the cluster center
Figure PCTCN2020103576-appb-000017
among them
Figure PCTCN2020103576-appb-000018
Corresponding to the five attributes of the converted longitude, latitude, hourly average temperature, hourly average humidity, and hourly average solar radiation. Cluster centers
Figure PCTCN2020103576-appb-000019
And 5 samples
Figure PCTCN2020103576-appb-000020
Defined as a representative sample of the current category.
步骤A3:针对每个基于隧道类别粗划分输入属性集合获得的聚类样本簇进行进一 步划分。具体包括以下子步骤：①针对样本簇中列车通过隧道时的温度时间序列和湿度序列进行镜像延拓，将样本簇内的温度序列、湿度序列转化为长度分别等于各自最长样本长度的序列。②设定延迟时间为1，窗口长度为5，采用延迟坐标法(Delay-coordinate method)对温度序列、湿度序列进行相空间重构，获取代表温度湿度演化特性的二维重构矩阵。③将样本的重构矩阵输入训练自动编码器，获取用于进一步划分的深度特征属性集合。④采用深度特征属性集合作为输入，按照步骤A2中的流程获取二次聚类的每个聚类样本簇的聚类中心
Figure PCTCN2020103576-appb-000021
及距离聚类中心最近的5个样本
Figure PCTCN2020103576-appb-000022
其中
Figure PCTCN2020103576-appb-000023
对应于深度特征属性集合中的5个维度变量。将聚类中心
Figure PCTCN2020103576-appb-000024
和5个样本
Figure PCTCN2020103576-appb-000025
定义为当前类别二次聚类的表征样本。 Step A3: further dividing each cluster sample cluster obtained by roughly dividing the input attribute set based on the tunnel category. Specifically, it includes the following sub-steps: ① Mirror and extend the temperature time sequence and humidity sequence when the train passes through the tunnel in the sample cluster, and convert the temperature sequence and humidity sequence in the sample cluster into a sequence whose length is equal to the longest sample length. ②Set the delay time to 1, the window length to 5, and use the Delay-coordinate method to reconstruct the temperature sequence and humidity sequence in phase space to obtain a two-dimensional reconstruction matrix representing the evolution characteristics of temperature and humidity. (3) Input the reconstruction matrix of the sample into the training autoencoder to obtain the depth feature attribute set for further division. ④Using the depth feature attribute set as input, follow the procedure in step A2 to obtain the cluster center of each cluster sample cluster of the secondary clustering
Figure PCTCN2020103576-appb-000021
And the 5 samples closest to the cluster center
Figure PCTCN2020103576-appb-000022
among them
Figure PCTCN2020103576-appb-000023
Corresponding to the 5 dimensional variables in the depth feature attribute set. Cluster centers
Figure PCTCN2020103576-appb-000024
And 5 samples
Figure PCTCN2020103576-appb-000025
Defined as the representative sample of the secondary clustering of the current category.
步骤3：构建典型序列HSV颜色空间模板库Step 3: Build a typical sequence HSV color space template library
将各类别隧道群体中距离该群体对应的聚类中心最近的5个隧道样本
Figure PCTCN2020103576-appb-000026
作为该类别下的典型隧道样本。如图3所示，将典型样本中随隧道里程变化的气温和湿度参数序列作为该类别的模板序列，设定延迟时间为1，窗口长度为5，采用延迟坐标法对温度、湿度和温度差分序列时间序列进行相空间重构，获取三个代表温度湿度演化特性的二维重构矩阵，将三个矩阵组合按照HSV颜色空间组合，形成彩色图像即为模板图像f i,i＝1,2…5。 In each type of tunnel group, the 5 tunnel samples closest to the cluster center corresponding to the group are selected
Figure PCTCN2020103576-appb-000026
As a typical tunnel sample under this category. As shown in Figure 3, the temperature and humidity parameter sequence that varies with the tunnel mileage in the typical sample is used as the template sequence of this category, the delay time is set to 1, the window length is 5, and the delay coordinate method is used to determine the temperature, humidity and temperature difference. The sequence time sequence is reconstructed in phase space, and three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity are obtained, and the three matrices are combined according to the HSV color space to form a color image, which is the template image f i , i=1, 2 …5.
步骤4：训练HSV模板匹配模型Step 4: Train the HSV template matching model
基于当前位置样本点及往前的一段样本序列，进行相空间重构和HSV颜色空间组合，构建当前位置特征图像。将当前位置特征模块与模板库中图像进行相关性计算，确定当前位置在模板库中的最佳匹配位置。具体包括以下步骤：Based on the current position sample points and the previous sample sequence, phase space reconstruction and HSV color space combination are performed to construct a current position feature image. Calculate the correlation between the current location feature module and the image in the template library to determine the best matching position of the current location in the template library. Specifically include the following steps:
步骤B1:采集温度、湿度和温度差分时间序列中当前样本点及往前的19个采样点。Step B1: Collect the current sample point and the previous 19 sample points in the temperature, humidity, and temperature differential time series.
步骤B2:设定延迟时间为1，窗口长度为5，采用延迟坐标法进行相空间重构，获取三个代表温度湿度演化特性的二维重构矩阵，将三个矩阵组合按照HSV颜色空间组合，形成当前位置特征图像h。Step B2: Set the delay time to 1, the window length to 5, and use the delay coordinate method to reconstruct the phase space, obtain three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity, and combine the three matrices according to the HSV color space , Forming the current location feature image h.
步骤B3:将当前位置特征图像与模板库中的图像进行卷积运算
Figure PCTCN2020103576-appb-000027
其中每个g i均为一个一维序列。 Step B3: Convolve the current position feature image with the image in the template library
Figure PCTCN2020103576-appb-000027
Each g i is a one-dimensional sequence.
步骤B4:对所有g i序列中的元素进行由大到小的排序，确定的最的5个元素为候选 元素，候选元素对应的排序前所在位置为候选位置，候选位置对应的里程值为l j,j＝1,2,…5。 Step B4: Sort all the elements in the g i sequence from big to small, the five most determined elements are candidate elements, the position corresponding to the candidate element before sorting is the candidate position, and the mileage value corresponding to the candidate position is l j ,j=1,2,...5.
步骤B5:将候选位置对应的里程值取均值确定为当前的模板匹配输出值，即模板匹配输出值O M＝mean(l j),j＝1,2,…5。 Step B5: The average mileage value corresponding to the candidate position is determined as the current template matching output value, that is, the template matching output value O M =mean(l j ), j=1, 2, ...5.
步骤5：训练RVM识别模型Step 5: Train the RVM recognition model
采用同类隧道群体的数据，以温度序列、湿度序列、时平均气温、时平均湿度、时平均太阳辐射量为输入，以当前隧道内的里程数据为输出，训练相关向量机(RVM)，建立隧道里程预测模型。具体包括以下步骤：Using the data of similar tunnel groups, taking temperature sequence, humidity sequence, hourly average temperature, hourly average humidity, and hourly average solar radiation as input, and the current mileage data in the tunnel as output, train the correlation vector machine (RVM) to build the tunnel Mileage prediction model. Specifically include the following steps:
步骤C1:定义训练样本，定义输入样本I＝(T,H,t 0,h 0,r 0),其中T＝(t 1,t 2…,t 19,t 20)为隧道内当前样本点及往前19个样本点的温度时间序列，其中H＝(h 1,h 2…,h 19,h 20)为隧道内的当前样本点及往前19个样本点的湿度序列，t 0,h 0和r 0分别为从气象站获取的时平均气温、时平均湿度和时平均太阳辐射量的预测值。输出样本为当前位置对应的里程值O。输入和输出组合Y＝{I,O}构成建模样本。针对每个二次聚类形成的同类隧道群体，选取M个(本发明中，M取5000)样本用于建立里程预测模型。 Step C1: Define the training sample, define the input sample I=(T,H,t 0 ,h 0 ,r 0 ), where T=(t 1 ,t 2 …,t 19 ,t 20 ) is the current sample point in the tunnel And the temperature time series of the previous 19 sample points, where H=(h 1 ,h 2 …,h 19 ,h 20 ) is the current sample point in the tunnel and the humidity sequence of the previous 19 sample points, t 0 , h 0 and r 0 are the predicted values of hourly average temperature, hourly average humidity, and hourly average solar radiation obtained from the weather station. The output sample is the mileage value O corresponding to the current position. The input and output combination Y={I, O} constitutes a modeling sample. For each group of similar tunnels formed by the secondary clustering, M (in the present invention, M takes 5000) samples are selected to build a mileage prediction model.
步骤C2:划分训练样本和验证样本和测试样本。采用无放回随机采样的方式选取M个样本中60％作为训练集，30％作为验证集，10％作为测试集。Step C2: Divide training samples, verification samples, and test samples. Using random sampling without replacement, 60% of the M samples are selected as the training set, 30% as the verification set, and 10% as the test set.
步骤C3:确定优化对象，初始化优化值。采用二进制鲸鱼算法优化模型的输入特征，即对输入样本I中每个维度的特征进行二进制编码，当某维度的特征对应的编码值为1时，该特征被选择作为RVM模型的输入变量，当某维度的特征对应的编码值为0时，该维度的特征将被舍弃。将43个维度特征随机初始化编码为0或1。Step C3: Determine the optimization object and initialize the optimization value. The binary whale algorithm is used to optimize the input features of the model, that is, the feature of each dimension in the input sample I is binary coded. When the code value corresponding to the feature of a certain dimension is 1, the feature is selected as the input variable of the RVM model. When the code value corresponding to a feature of a certain dimension is 0, the feature of this dimension will be discarded. The 43 dimensional features are randomly initialized and coded as 0 or 1.
步骤C4:确定优化目标函数。基于当前特征编码值，确定输入特征，采用训练集数据训练RVM模型。将验证集数据输入训练好的RVM模型，获取模型输出序列为
Figure PCTCN2020103576-appb-000028
其中M 1＝0.3M。定义优化目标函数 Step C4: Determine the optimization objective function. Based on the current feature code value, the input feature is determined, and the training set data is used to train the RVM model. Input the validation set data into the trained RVM model, and obtain the model output sequence as
Figure PCTCN2020103576-appb-000028
Where M 1 =0.3M. Define optimization objective function
Figure PCTCN2020103576-appb-000029
Figure PCTCN2020103576-appb-000029
式中
Figure PCTCN2020103576-appb-000030
验证集的真实输出值。 Where
Figure PCTCN2020103576-appb-000030
The true output value of the validation set.
步骤C5:输出优化预测模型。采用二进制鲸鱼算法进行迭代优化运算，确定最优的 输入特征和RVM模型，该模型为RVM里程预测模型。Step C5: Output the optimized prediction model. The binary whale algorithm is used for iterative optimization operations to determine the optimal input features and RVM model, which is the RVM mileage prediction model.
步骤6：构建HSV模板匹配与RVM识别融合模型Step 6: Construct HSV template matching and RVM recognition fusion model
将测试集数据代入RVM里程预测模型，获取模型输出序列为
Figure PCTCN2020103576-appb-000031
其中M 2＝0.1M。以测试集数据为输入，按照步骤4的操作流程，获取模板匹配的模型输出结果为
Figure PCTCN2020103576-appb-000032
测试集数据中的真实输出结果为
Figure PCTCN2020103576-appb-000033
计算RVM模型输出值与真实值的误差为 Substitute the test set data into the RVM mileage prediction model to obtain the model output sequence as
Figure PCTCN2020103576-appb-000031
Where M 2 =0.1M. Take the test set data as input and follow the operation process of step 4 to obtain the model output result of template matching as
Figure PCTCN2020103576-appb-000032
The real output result in the test set data is
Figure PCTCN2020103576-appb-000033
Calculate the error between the output value of the RVM model and the true value as
Figure PCTCN2020103576-appb-000034
Figure PCTCN2020103576-appb-000034
计算模板匹配模型输出值与真实值的误差为Calculate the error between the output value of the template matching model and the true value as
Figure PCTCN2020103576-appb-000035
Figure PCTCN2020103576-appb-000035
获取同时计算当前若干模板样本与当前样本的距离。则RVM模型的模型融合系数的定义为Obtain and calculate the distance between several current template samples and the current sample at the same time. Then the model fusion coefficient of the RVM model is defined as
Figure PCTCN2020103576-appb-000036
Figure PCTCN2020103576-appb-000036
则模板匹配模型的模型融合系数的定义为Then the model fusion coefficient of the template matching model is defined as
Figure PCTCN2020103576-appb-000037
Figure PCTCN2020103576-appb-000037
则模型的最终输出结果为Then the final output of the model is
Figure PCTCN2020103576-appb-000038
Figure PCTCN2020103576-appb-000038
步骤7：获取输入数据，调用里程预测融合模型Step 7: Obtain input data, call mileage prediction fusion model
列车运行过程中，通过区域内的气象站获取当前、隧道外气温、气压和太阳辐射数据。利用安装在列车头部和尾部的温度湿度传感器，获取当前的温度湿度序列。具体包括以下步骤：During the train operation, the current temperature, air pressure and solar radiation data outside the tunnel are obtained from the weather station in the area. The temperature and humidity sensors installed at the head and tail of the train are used to obtain the current temperature and humidity sequence. Specifically include the following steps:
步骤D1.参照步骤二的流程，获取用于一次聚类时的输入属性
Figure PCTCN2020103576-appb-000039
其中
Figure PCTCN2020103576-appb-000040
对应于转换后的经度、纬度、时平均气温、时平均湿度、时平均太阳辐射量等5个属性。参照步骤二的流程，获取用于二次次聚类时的输入属性
Figure PCTCN2020103576-appb-000041
其 中
Figure PCTCN2020103576-appb-000042
对应于深度特征属性集合中的5个属性。 Step D1. Refer to the process of step 2 to obtain the input attributes used for a clustering
Figure PCTCN2020103576-appb-000039
among them
Figure PCTCN2020103576-appb-000040
Corresponding to the five attributes of the converted longitude, latitude, hourly average temperature, hourly average humidity, and hourly average solar radiation. Refer to the process of step 2 to obtain the input attributes used for secondary clustering
Figure PCTCN2020103576-appb-000041
among them
Figure PCTCN2020103576-appb-000042
Corresponds to 5 attributes in the depth feature attribute set.
步骤D2:计算当前样本点特征值
Figure PCTCN2020103576-appb-000043
与第一次聚类表征样本间的距离 Step D2: Calculate the feature value of the current sample point
Figure PCTCN2020103576-appb-000043
The distance from the first clustering characterization sample
Figure PCTCN2020103576-appb-000044
Figure PCTCN2020103576-appb-000044
取当前样本点到所有聚类表征样本间最小值对应的样本簇为一次聚类目标样本簇。Take the sample cluster corresponding to the minimum value between the current sample point and all cluster characterization samples as the primary clustering target sample cluster.
步骤D3:计算当前样本点特征值
Figure PCTCN2020103576-appb-000045
与第二次聚类表征样本间的距离 Step D3: Calculate the feature value of the current sample point
Figure PCTCN2020103576-appb-000045
The distance from the second clustering characterization sample
Figure PCTCN2020103576-appb-000046
Figure PCTCN2020103576-appb-000046
取当前样本点到所有一次聚类目标样本簇下属的二次聚类表征样本间最小值对应的样本簇为二次聚类目标样本簇。该样本簇下训练的模型和模板库为目标模型和目标模板库。The sample cluster corresponding to the minimum value among the secondary clustering representation samples from the current sample point to all primary clustering target sample clusters is taken as the secondary clustering target sample cluster. The model and template library trained under this sample cluster are the target model and the target template library.
步骤8：预测列车位置Step 8: Predict the train position
参照步骤5的流程计算当前样本点的RVM模型输入向量，将RVM模型输入向量代入目标模型中，获取目标RVM模型输出值。参照步骤4的流程获取当前样本点的模板匹配模型输入值，带入目标模板匹配模型中，获取目标模板匹配模型输出值。参照式6获取最终列车位置预测结果。Refer to the process of step 5 to calculate the RVM model input vector of the current sample point, and substitute the RVM model input vector into the target model to obtain the target RVM model output value. Refer to the process of step 4 to obtain the template matching model input value of the current sample point, bring it into the target template matching model, and obtain the target template matching model output value. Refer to Equation 6 to obtain the final train position prediction result.
本发明实施例2提供了一种基于气象参数的高速列车导航盲区定位系统，包括安装于列车上的、用于采集隧道内的气象参数的传感器；所述传感器与计算机设备通信；该计算机设备被编程或配置为执行本发明实施例1方法的步骤。Embodiment 2 of the present invention provides a high-speed train navigation blind spot positioning system based on meteorological parameters, including a sensor installed on the train for collecting meteorological parameters in the tunnel; the sensor communicates with a computer device; the computer device is It is programmed or configured to execute the steps of the method of Embodiment 1 of the present invention.

Claims (10)

  1. 一种基于气象参数的高速列车导航盲区定位方法，其特征在于，包括以下步骤：A method for locating blind spots in high-speed train navigation based on meteorological parameters is characterized in that it includes the following steps:
    S1、实时采集列车通过时隧道内的隧道气象参数，构建隧道气象参数数据库；S1. Collect the tunnel meteorological parameters in the tunnel in real time when the train passes, and construct the tunnel meteorological parameter database;
    S2、基于所述隧道气象参数数据库，对属性相近的隧道群体进行归类，获取各类别隧道群体的典型隧道样本；S2, based on the tunnel meteorological parameter database, classify tunnel groups with similar attributes, and obtain typical tunnel samples of each type of tunnel group;
    S3、利用所述典型隧道样本构建典型序列HSV颜色空间模板库；采用同类隧道群体的数据，训练相关向量机，建立隧道里程预测模型；S3. Use the typical tunnel samples to construct a typical sequence HSV color space template library; use the data of similar tunnel groups to train a correlation vector machine to establish a tunnel mileage prediction model;
    S4、利用所述HSV颜色空间模板库训练HSV模板匹配模型；S4. Use the HSV color space template library to train an HSV template matching model;
    S5、构建HSV模板匹配模型与隧道里程预测模型的融合模型，得到里程预测融合模型；S5. Construct a fusion model of the HSV template matching model and the tunnel mileage prediction model to obtain the mileage prediction fusion model;
    S6、获取列车运行过程中的隧道气象参数数据，调用里程预测融合模型，预测列车位置。S6. Obtain the tunnel meteorological parameter data during train operation, call the mileage prediction fusion model, and predict the train position.
  2. 根据权利要求1所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，步骤S1中，构建隧道气象参数数据库的具体过程包括：采集列车一次通过某隧道时采集的温度序列、湿度序列、所在区域的经度、纬度以及时平均气温、时平均湿度、时平均太阳辐射量的预测值，构成1组隧道气象参数样本；辖区内所有列车1年内运行采集的隧道气象参数样本构成隧道气象参数数据库。The method for locating blind spots in high-speed train navigation based on meteorological parameters according to claim 1, characterized in that, in step S1, the specific process of constructing a tunnel meteorological parameter database includes: collecting temperature series and humidity series collected when the train passes through a certain tunnel at one time , The longitude, latitude, and the predicted values of the time average temperature, time average humidity, and time average solar radiation of the area where they are located, constitute a set of tunnel meteorological parameter samples; the tunnel meteorological parameter samples collected within one year of all trains in the jurisdiction constitute the tunnel meteorological parameters database.
  3. 根据权利要求1所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，步骤S2的具体实现过程包括：The method for locating blind spots in high-speed train navigation based on meteorological parameters according to claim 1, wherein the specific implementation process of step S2 includes:
    a)将隧道所在区域经度、纬度坐标转换为平面坐标；对隧道气象参数进行归一化处理，处理后的经度、纬度坐标和隧道气象参数组成隧道类别粗划分输入属性集合；a) Convert the longitude and latitude coordinates of the area where the tunnel is located into plane coordinates; normalize the tunnel meteorological parameters, and the processed longitude, latitude coordinates and tunnel meteorological parameters form a set of input attributes for the rough division of tunnel categories;
    b)以隧道类别粗划分输入属性为对象，采用OPTICS算法进行隧道群样本输出排序，比较排序后的序列的可达距离与设定邻域距离参数ε，序列中可达距离小于邻域距离参数ε的连续样本构成一个聚类样本簇；获取每个聚类样本簇的聚类中心
    Figure PCTCN2020103576-appb-100001
    及距离聚类中心最近的T1个样本
    Figure PCTCN2020103576-appb-100002
    其中
    Figure PCTCN2020103576-appb-100003
    对应于处理后的经度和纬度，
    Figure PCTCN2020103576-appb-100004
    对应于处理后的隧道气象参数；将聚类中心
    Figure PCTCN2020103576-appb-100005
    和T1个样本定义为当前类别隧道群体的表征样本，该表征样本中的T1个样本即为当前类别隧道群体的 典型隧道样本。     b) The input attributes are roughly divided into the tunnel category, and the OPTICS algorithm is used to sort the output of the tunnel group samples, and the reachable distance of the sorted sequence is compared with the set neighborhood distance parameter ε. The reachable distance in the sequence is less than the neighborhood distance parameter Consecutive samples of ε form a cluster sample cluster; obtain the cluster center of each cluster sample cluster
    Figure PCTCN2020103576-appb-100001
    And the T1 samples closest to the cluster center
    Figure PCTCN2020103576-appb-100002
    among them
    Figure PCTCN2020103576-appb-100003
    Corresponds to the processed longitude and latitude,
    Figure PCTCN2020103576-appb-100004
    Corresponds to the processed weather parameters of the tunnel; the cluster center
    Figure PCTCN2020103576-appb-100005
    And T1 samples are defined as the representative samples of the current type of tunnel population, and T1 samples in the representative samples are the typical tunnel samples of the current type of tunnel population.
  4. 根据权利要求3所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，步骤S2的具体实现过程包括：The method for locating blind areas in high-speed train navigation based on meteorological parameters according to claim 3, wherein the specific implementation process of step S2 includes:
    1)将聚类样本簇中列车通过隧道时的温度时间序列和湿度序列进行镜像延拓，将聚类样本簇内的温度序列、湿度序列转化为长度分别等于各自最长样本长度的序列；1) Mirror and extend the temperature time sequence and humidity sequence when the train passes through the tunnel in the cluster sample cluster, and convert the temperature sequence and humidity sequence in the cluster sample cluster into a sequence whose length is equal to the length of the longest sample;
    2)设定延迟时间和窗口长度，采用延迟坐标法对温度、湿度时间最长样本长度的序列进行相空间重构，获取代表温度、湿度演化特性的二维重构矩阵；2) Set the delay time and window length, and use the delay coordinate method to reconstruct the phase space of the sequence with the longest sample length of temperature and humidity time to obtain a two-dimensional reconstruction matrix representing the evolution characteristics of temperature and humidity;
    3)将所述二维重构矩阵输入训练自动编码器，获取用于进一步划分的深度特征属性集合；3) Input the two-dimensional reconstruction matrix into the training autoencoder to obtain the depth feature attribute set for further division;
    4)将深度特征属性集合作为输入，按照步骤b)的流程获取二次聚类的每个聚类样本簇的聚类中心
    Figure PCTCN2020103576-appb-100006
    及距离聚类中心最近的T1个样本；将聚类中心
    Figure PCTCN2020103576-appb-100007
    和T1个样本定义为当前类别隧道群体二次聚类的表征样本，该二次聚类的表征样本中的T1个样本即为当前类别隧道群体的典型隧道样本。     4) Taking the depth feature attribute set as input, follow the process of step b) to obtain the cluster center of each cluster sample cluster of the secondary clustering
    Figure PCTCN2020103576-appb-100006
    And the T1 samples closest to the cluster center; the cluster center
    Figure PCTCN2020103576-appb-100007
    And T1 samples are defined as the representative samples of the secondary clustering of the current type of tunnel population, and T1 samples in the secondary clustering of the representative samples are the typical tunnel samples of the current type of tunnel population.
  5. 根据权利要求1～4之一所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，步骤S3中，利用所述典型隧道样本构建典型序列HSV颜色空间模板库的具体实现过程包括：将典型隧道样本中随隧道里程变化的气温和湿度参数序列作为对应类别隧道群体的模板序列，设定延迟时间和窗口长度，采用延迟坐标法对温度、湿度和温度差分序列时间序列进行相空间重构，获取三个代表温度、湿度演化特性的二维重构矩阵，将三个二维重构矩阵组合按照HSV颜色空间组合，形成彩色图像，即得到典型序列HSV颜色空间模板库中的模板图像。The method for locating blind spots in high-speed train navigation based on meteorological parameters according to any one of claims 1 to 4, characterized in that, in step S3, the specific implementation process of constructing a typical sequence HSV color space template library using the typical tunnel samples includes: The temperature and humidity parameter sequence that changes with the length of the tunnel in the typical tunnel sample is used as the template sequence of the corresponding type of tunnel group, the delay time and window length are set, and the delay coordinate method is used to reconstruct the phase space of the temperature, humidity and temperature difference sequence time series. The three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity are obtained, and the three two-dimensional reconstruction matrices are combined according to the HSV color space to form a color image, that is, the template image in the typical sequence HSV color space template library is obtained .
  6. 根据权利要求1～5之一所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，步骤S4中，训练HSV模板匹配模型的具体实现过程包括：The method for locating blind spots in high-speed train navigation based on meteorological parameters according to any one of claims 1 to 5, wherein, in step S4, the specific implementation process of training the HSV template matching model includes:
    1)采集温度、湿度和温度差分时间序列中当前样本点及当前样本点往前的N个采样点；1) Collect the current sample point and N sampling points before the current sample point in the temperature, humidity and temperature differential time series;
    2)设定延迟时间和窗口长度，采用延迟坐标法进行相空间重构，获取三个代表温度、湿度演化特性的二维重构矩阵，将三个矩阵组合按照HSV颜色空间组合，形成当前位置特征图像；2) Set the delay time and window length, use the delay coordinate method to reconstruct the phase space, obtain three two-dimensional reconstruction matrices representing the evolution characteristics of temperature and humidity, and combine the three matrices according to the HSV color space to form the current position Feature image
    3)将当前位置特征图像与典型序列HSV颜色空间模板库中的模板图像进行卷积运算，得到多个一维序列；3) Convolve the current position feature image with the template image in the HSV color space template library of the typical sequence to obtain multiple one-dimensional sequences;
    4)对所有一维序列中的元素进行由大到小的排序，排序后的序列中最大的T2个元素为候选元素，候选元素对应的排序前所在位置为候选位置，获取候选位置对应的里程值；4) Sort all the elements in the one-dimensional sequence from big to small, the largest T2 element in the sorted sequence is the candidate element, the position corresponding to the candidate element before sorting is the candidate position, and the mileage corresponding to the candidate position is obtained value;
    5)对所有候选位置对应的所有里程值取均值，得到当前的HSV模板匹配模型输出序列。5) Take the average of all mileage values corresponding to all candidate positions to obtain the current HSV template matching model output sequence.
  7. 根据权利要求1～6之一所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，步骤S4中，训练相关向量机，建立隧道里程预测模型的具体实现过程包括：The method for locating blind spots in high-speed train navigation based on meteorological parameters according to any one of claims 1 to 6, characterized in that, in step S4, the specific implementation process of training a correlation vector machine and establishing a tunnel mileage prediction model includes:
    1)定义输入样本I＝(T,H,t 0,h 0,r 0),其中T＝(t 1,t 2…,t M)为隧道内当前样本点及往前M个样本点的温度时间序列，H＝(h 1,h 2…,h 19,h M)为隧道内的当前样本点及往前N个样本点的湿度序列，t 0,h 0和r 0分别为从气象站获取的时平均气温、时平均湿度和时平均太阳辐射量的预测值；定义输出样本为当前位置对应的里程值O，输入和输出组合Y＝{I,O}构成建模样本；针对每个同类隧道群体，选取M个建模样本； 1) Define the input sample I = (T, H, t 0 , h 0 , r 0 ), where T = (t 1 , t 2 …, t M ) is the current sample point in the tunnel and the previous M sample points Temperature time series, H = (h 1 , h 2 …, h 19 , h M ) is the current sample point in the tunnel and the humidity sequence of the previous N sample points, t 0 , h 0 and r 0 are the meteorological The predicted values of hourly average temperature, hourly average humidity, and hourly average solar radiation obtained by the station; define the output sample as the mileage value O corresponding to the current position, and the input and output combination Y={I,O} constitutes a modeling sample; A group of similar tunnels, select M modeling samples;
    2)将M个建模样本随机划分训练集、验证集和测试集；2) Randomly divide M modeling samples into training set, validation set and test set;
    3)对输入样本I中每个维度的特征进行二进制编码，当某维度的特征对应的编码值为1时，该特征被选择作为RVM模型的输入变量，当某维度的特征对应的编码值为0时，该维度的特征被舍弃；3) Binary encoding is performed on the features of each dimension in the input sample I. When the encoding value corresponding to the feature of a certain dimension is 1, the feature is selected as the input variable of the RVM model. When the encoding value corresponding to the feature of a certain dimension is At 0, the feature of this dimension is discarded;
    4)基于当前特征编码值，确定新的输入特征，更新训练集、验证集和测试集，采用更新后的训练集数据训练RVM模型，将更新后的验证集数据输入训练好的RVM模型，获取模型输出序列
    Figure PCTCN2020103576-appb-100008
    k＝1,2…M 1；M 1＝0.3M；     4) Determine the new input feature based on the current feature code value, update the training set, validation set, and test set, use the updated training set data to train the RVM model, and input the updated validation set data into the trained RVM model to obtain Model output sequence
    Figure PCTCN2020103576-appb-100008
    k=1, 2...M 1 ; M 1 ＝0.3M;
    5)重复步骤3)和步骤4)，确定使优化目标函数
    Figure PCTCN2020103576-appb-100009
    最小的最优的输入特征和RVM模型，该RVM模型即为隧道里程预测模型；其中，
    Figure PCTCN2020103576-appb-100010
    为验证集中的真实输出值。     5) Repeat steps 3) and 4) to determine the optimization objective function
    Figure PCTCN2020103576-appb-100009
    The smallest and optimal input features and RVM model, the RVM model is the tunnel mileage prediction model; among them,
    Figure PCTCN2020103576-appb-100010
    To verify the true output value in the set.
  8. 根据权利要求7所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，HSV模板匹配模型与隧道里程预测模型的融合模型
    Figure PCTCN2020103576-appb-100011
    其中，k1＝1,2…M 2，M 2＝0.1M；
    Figure PCTCN2020103576-appb-100012
    Figure PCTCN2020103576-appb-100013
    为将测试集数据输入隧道里程预测模型后得到的输出序列；
    Figure PCTCN2020103576-appb-100014
    为测试集中的真实输出结果；
    Figure PCTCN2020103576-appb-100015
    为将测试集输入HSV模板匹配模型后得到的输出序列。     The method for locating the blind zone of high-speed train navigation based on meteorological parameters according to claim 7, wherein the HSV template matching model and the tunnel mileage prediction model are a fusion model
    Figure PCTCN2020103576-appb-100011
    Among them, k1 = 1, 2...M 2 , M 2 =0.1M;
    Figure PCTCN2020103576-appb-100012
    Figure PCTCN2020103576-appb-100013
    It is the output sequence obtained after inputting the test set data into the tunnel mileage prediction model;
    Figure PCTCN2020103576-appb-100014
    Real output results in the test set;
    Figure PCTCN2020103576-appb-100015
    It is the output sequence obtained after inputting the test set into the HSV template matching model.
  9. 根据权利要求1～8之一所述的基于气象参数的高速列车导航盲区定位方法，其特征在于，步骤S6中，预测列车位置的具体实现过程包括：计算当前样本点的RVM模型输入向量，将该RVM模型输入向量代入目标模型中，获取隧道里程预测模型输出值；获取当前样本点的HSV模板匹配模型输入值，代入HSV模板匹配模型中，获取HSV模板匹配模型输出值；将里程预测模型输出值和HSV模板匹配模型输出值代入里程预测融合模型，获取最终列车位置预测结果；其中，目标模板是指二次聚类目标样本簇下训练的模型；二次聚类目标样本簇是指当前样本点到所有一次聚类目标样本簇下属的二次聚类表征样本间最小值对应的样本簇；所述一次聚类目标样本簇是指当前样本点到所有聚类表征样本间最小值对应的样本簇；所述聚类表征样本是指聚类样本簇中的样本，所述聚类样本簇是指以隧道类别粗划分输入属性为对象，采用OPTICS算法进行隧道群样本输出排序得到的序列的可达距离与设定邻域距离参数ε之间的值小于邻域距离参数ε的连续样本构成的样本簇。The method for locating blind spots in high-speed train navigation based on meteorological parameters according to any one of claims 1 to 8, characterized in that, in step S6, the specific implementation process of predicting the train position includes: calculating the RVM model input vector of the current sample point, and The input vector of the RVM model is substituted into the target model to obtain the output value of the tunnel mileage prediction model; the input value of the HSV template matching model of the current sample point is obtained, and it is substituted into the HSV template matching model to obtain the output value of the HSV template matching model; and the mileage prediction model is output Value and the output value of HSV template matching model are substituted into the mileage prediction fusion model to obtain the final train position prediction result; among them, the target template refers to the model trained under the secondary clustering target sample cluster; the secondary clustering target sample cluster refers to the current sample From the point to the sample cluster corresponding to the minimum value between the secondary clustering characterization samples under all primary clustering target sample clusters; the primary clustering target sample cluster refers to the sample corresponding to the minimum value between the current sample point and all the clustering characterization samples Cluster; The cluster characterization sample refers to the samples in the cluster sample cluster, and the cluster sample cluster refers to the sequence of the sequence obtained by using the OPTICS algorithm to sort the output of the tunnel group sample with the input attributes of the tunnel category as the object. A cluster of samples composed of continuous samples whose value between the reach distance and the set neighborhood distance parameter ε is less than the neighborhood distance parameter ε.
  10. 一种基于气象参数的高速列车导航盲区定位系统，其特征在于，包括安装于列车上的、用于采集隧道内的气象参数的传感器；所述传感器与计算机设备通信；所述计算机设备被编程或配置为执行权利要求1～9之一所述方法的步骤。A high-speed train navigation blind spot positioning system based on meteorological parameters is characterized in that it includes a sensor installed on the train for collecting meteorological parameters in the tunnel; the sensor communicates with a computer device; the computer device is programmed or It is configured to perform the steps of the method described in one of claims 1-9.
PCT/CN2020/103576 2019-07-24 2020-07-22 Meteorological parameter-based high-speed train positioning method and system in navigation blind zone WO2021013190A1 (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
RU2021122153A RU2771515C1 (en) 2019-07-24 2020-07-22 Method and system for determining the location of a high-speed train in a navigation blind spot based on meteorological parameters
SG11202103075TA SG11202103075TA (en) 2019-07-24 2020-07-22 Method and system for locating high-speed train in navigation blind zone based on meteorological parameters
AU2020316538A AU2020316538B2 (en) 2019-07-24 2020-07-22 Meteorological parameter-based high-speed train positioning method and system in navigation blind zone

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910670376.6A CN110427993B (en) 2019-07-24 2019-07-24 High-speed train navigation blind area positioning method based on meteorological parameters
CN201910670376.6 2019-07-24

Publications (1)

Publication Number Publication Date
WO2021013190A1 true WO2021013190A1 (en) 2021-01-28

Family

ID=68410557

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/103576 WO2021013190A1 (en) 2019-07-24 2020-07-22 Meteorological parameter-based high-speed train positioning method and system in navigation blind zone

Country Status (5)

Country Link
CN (1) CN110427993B (en)
AU (1) AU2020316538B2 (en)
RU (1) RU2771515C1 (en)
SG (1) SG11202103075TA (en)
WO (1) WO2021013190A1 (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113822473A (en) * 2021-09-03 2021-12-21 浙江浙大中控信息技术有限公司 Traction energy consumption reasonable interval prediction method based on multidimensional data
CN114559439A (en) * 2022-04-27 2022-05-31 南通科美自动化科技有限公司 Intelligent obstacle avoidance control method and device for mobile robot and electronic equipment
CN115690747A (en) * 2022-12-30 2023-02-03 天津所托瑞安汽车科技有限公司 Vehicle blind area detection model test method and device, electronic equipment and storage medium
CN117251748A (en) * 2023-10-10 2023-12-19 中国船舶集团有限公司第七〇九研究所 Track prediction method, equipment and storage medium based on historical rule mining

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110427993B (en) * 2019-07-24 2023-04-21 中南大学 High-speed train navigation blind area positioning method based on meteorological parameters
CN113610117B (en) * 2021-07-19 2024-04-02 上海德衡数据科技有限公司 Underwater sensing data processing method and system based on depth data

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2272731C2 (en) * 2002-01-21 2006-03-27 Игорь Николаевич Сушкин Method to check location of railway train
JP2018044843A (en) * 2016-09-14 2018-03-22 三菱電機株式会社 Position location device
CN109738927A (en) * 2018-12-14 2019-05-10 湖南卫导信息科技有限公司 Navigation signal continuously-positioning system and method inside and outside tunnel
CN109828289A (en) * 2019-03-06 2019-05-31 湖南卫导信息科技有限公司 Navigation information analogue system for tunnel
CN110427993A (en) * 2019-07-24 2019-11-08 中南大学 Bullet train navigational gap localization method based on meteorologic parameter

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9720069B2 (en) * 2012-10-10 2017-08-01 Lg Electronics Inc. Apparatus and method for measuring location of user equipment located indoors in wireless network
WO2018082085A1 (en) * 2016-11-07 2018-05-11 中国科学院自动化研究所 Microscope image acquisition method based on sequence slice
CN107423412B (en) * 2017-07-28 2018-03-27 中南大学 A kind of method of the carrying robot Intelligent Recognition floor based on meteorological sensing time series pattern
CN109902881A (en) * 2019-03-19 2019-06-18 武汉乐易创想科技有限公司 The PM2.5 concentration prediction method merged based on multi-variate statistical analysis and LSTM

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2272731C2 (en) * 2002-01-21 2006-03-27 Игорь Николаевич Сушкин Method to check location of railway train
JP2018044843A (en) * 2016-09-14 2018-03-22 三菱電機株式会社 Position location device
CN109738927A (en) * 2018-12-14 2019-05-10 湖南卫导信息科技有限公司 Navigation signal continuously-positioning system and method inside and outside tunnel
CN109828289A (en) * 2019-03-06 2019-05-31 湖南卫导信息科技有限公司 Navigation information analogue system for tunnel
CN110427993A (en) * 2019-07-24 2019-11-08 中南大学 Bullet train navigational gap localization method based on meteorologic parameter

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113822473A (en) * 2021-09-03 2021-12-21 浙江浙大中控信息技术有限公司 Traction energy consumption reasonable interval prediction method based on multidimensional data
CN113822473B (en) * 2021-09-03 2023-12-26 浙江中控信息产业股份有限公司 Traction energy consumption reasonable interval prediction method based on multidimensional data
CN114559439A (en) * 2022-04-27 2022-05-31 南通科美自动化科技有限公司 Intelligent obstacle avoidance control method and device for mobile robot and electronic equipment
CN114559439B (en) * 2022-04-27 2022-07-26 南通科美自动化科技有限公司 Mobile robot intelligent obstacle avoidance control method and device and electronic equipment
CN115690747A (en) * 2022-12-30 2023-02-03 天津所托瑞安汽车科技有限公司 Vehicle blind area detection model test method and device, electronic equipment and storage medium
CN117251748A (en) * 2023-10-10 2023-12-19 中国船舶集团有限公司第七〇九研究所 Track prediction method, equipment and storage medium based on historical rule mining
CN117251748B (en) * 2023-10-10 2024-04-19 中国船舶集团有限公司第七〇九研究所 Track prediction method, equipment and storage medium based on historical rule mining

Also Published As

Publication number Publication date
RU2771515C1 (en) 2022-05-05
CN110427993B (en) 2023-04-21
AU2020316538A1 (en) 2021-04-15
AU2020316538B2 (en) 2022-07-07
CN110427993A (en) 2019-11-08
SG11202103075TA (en) 2021-04-29

Similar Documents

Publication Publication Date Title
WO2021013190A1 (en) Meteorological parameter-based high-speed train positioning method and system in navigation blind zone
CN110264709B (en) Method for predicting traffic flow of road based on graph convolution network
CN106228808B (en) City expressway travel time prediction method based on Floating Car space-time grid data
CN114092832B (en) High-resolution remote sensing image classification method based on parallel hybrid convolutional network
JP7226855B2 (en) Calculation method, system and storage medium for train passing through tunnel
Yan et al. Spatial-temporal chebyshev graph neural network for traffic flow prediction in iot-based its
CN111027430B (en) Traffic scene complexity calculation method for intelligent evaluation of unmanned vehicles
CN111985325B (en) Aerial small target rapid identification method in extra-high voltage environment evaluation
CN110991523A (en) Interpretability evaluation method for unmanned vehicle detection algorithm performance
CN111383273B (en) High-speed rail contact net part positioning method based on improved structure reasoning network
CN112884014A (en) Traffic speed short-time prediction method based on road section topological structure classification
CN112862171A (en) Flight arrival time prediction method based on space-time neural network
CN114882069A (en) Taxi track abnormity detection method based on LSTM network and attention mechanism
CN112990282A (en) Method and device for classifying fine-grained small sample images
CN115099328A (en) Traffic flow prediction method, system, device and storage medium based on countermeasure network
CN116662468A (en) Urban functional area identification method and system based on geographic object space mode characteristics
CN109147322B (en) Multi-source data self-adaptive fusion method in urban traffic big data processing
CN113159371B (en) Unknown target feature modeling and demand prediction method based on cross-modal data fusion
CN112101132B (en) Traffic condition prediction method based on graph embedding model and metric learning
CN114898089B (en) Functional area extraction and classification method fusing high-resolution images and POI data
Qing et al. Using feature interaction among GPS Data for road intersection detection
Gao An image recognition method for speed limit plate based on deep learning algorithm
CN115482666B (en) Multi-graph convolution neural network traffic prediction method based on data fusion
CN116976526B (en) Land utilization change prediction method coupling ViViViT and ANN
CN116597315A (en) Infrared remote sensing image runway detection method based on YOLO v3-tiny

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20843123

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020316538

Country of ref document: AU

Date of ref document: 20200722

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20843123

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20843123

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC , EPO FORM 1205A DATED 03.11.22.

122 Ep: pct application non-entry in european phase

Ref document number: 20843123

Country of ref document: EP

Kind code of ref document: A1