RU2773332C1 - Method, system and storage medium for calculating the duration of a train's passage through a tunnel - Google Patents

Abstract

FIELD: computer technology.

SUBSTANCE: invention relates to computer technology, namely to determining the time of passage of a train through a tunnel. A method is described for calculating the duration of the passage of a train through a tunnel, while on-board temperature, humidity and air pressure sensors installed at both ends of the train receive sequences of barometric and meteorological parameters when the train passes through the tunnel, and pre-installed responders receive time sequences of the path in miles from the current position to the tunnel exit in order to create a database of meteorological parameters of the tunnel. The barometric and meteorological parameters of the tunnel in the database are classified by means of architectural zoning and an average temperature distribution model and a comparison of typical sequences of each category is carried out, which are extracted to form a combined model for matching RGB color image templates.

EFFECT: increase in the accuracy of determining the time of passage of a train in a tunnel.

10 cl, 2 dwg

Description

FIELD OF TECHNOLOGY

The present invention relates to the field of artificial intelligence training, and more particularly to a method, system, and storage medium for calculating the duration of a train passing through a tunnel.

PRIOR ART

Improving the passenger experience of using passenger trains while traveling is conducive to increasing the competitiveness of high-speed rail passenger transport. Comprehensive information cues help to enable passengers to quickly perceive the current status of train operation and improve the passenger experience. At present, train operational information includes temperature, train speed, etc., but it lacks relevant information prompts for special conditions such as a train passing through a tunnel.

After the train enters the tunnel, the transmitted information may be lost, so that passengers, in the age of the Internet, find themselves out of work, and the passenger's sense of comfort quickly declines. In addition, the dark hermetic environment in the tunnel, the discomfort in the eardrum caused by the deteriorating situation of the hermetic environment of the train, and the unknown length of the tunnel cause a noticeable feeling of anxiety. For this reason, efficient and accurate positioning methods for tunnel passage times are being developed in order to inform passengers in a timely manner of the time they will need to wait for tunnel passage, which is very important to improve the passenger experience. The essence of calculating the duration of the passage of the tunnel of trains is the exact positioning of the train in the tunnel. The usual ways to position a train are as follows:

1. Positioning type based on Beidou/GPS; Beidou satellite positioning can provide all-round, all-weather and 24-hour positioning information, but is not suitable for train positioning in a tunnel.

2. Use of wireless base stations; wireless stations at both ends of the tunnel provide information about the train, which effectively reduces the use of track equipment along the line, but is not suitable for train positioning in the tunnel.

3. Use of defendants; a plurality of track equipments are arranged along the railway track, and the corresponding on-board equipments are installed on the train, which has better accuracy, but the development cost and maintenance cost are very high, so that the practical benefit is small.

Brief description of the essence of the invention

The technical problem solved by the present invention is to provide, in response to the shortcomings of the prior art, a method, system and storage medium for calculating the duration of the passage of the train through the tunnel so as to realize the intelligent perception of the distance between the train in the tunnel and the exit of the tunnel , and accurately estimate the time of train movement in the tunnel.

In order to solve the above technical problem, the technical solution adopted by the present invention is as follows: a method for calculating the duration of the passage of a train through a tunnel, and the remaining duration for which the train passes through this tunnel is estimated using the following equation: t = O ( i )/ f _Hz ( O ( i )- O ( i -1)); moreover, the distance in miles from the current position of the train to the exit of the tunnel is O ( i )= O ^M ( i )+ ε ( i ); O ( i -1) represents the distance in miles from the last position of the current position of the train to the exit of the tunnel; O ^M ( i ) represents the position of the best match of the current position in the library of combined type samples patterns, which is obtained after calculating the correlations between the images of the features of the current position and the images in the library of the combined type samples patterns, the image of the current position features is obtained by reconstructing the phase space and combining the color RGB spaces of received sequences of meteorological parameters of sample points at the head and tail of the train at the current position and meteorological parameters with a period of time before the current position, and template images in the library of combined templates of type samples are color images regarding the change of meteorological parameters at the head and tail parts of the train, and the meteorological parameters in the head and tail parts of the train are obtained by reconstructing the phase space of typical samples of each category and groups of tunnels; sample samples of each category of tunnel groups refer to samples obtained by classifying meteorological parameters according to the geographic location of the tunnel and meteorological conditions inside and outside the tunnel and further classifying the classified homogeneous samples in the area; ε ( i ) is the compensation error obtained by inputting the meteorological parameters obtained during the operation of the train into the tunnel mile prediction error compensation model, and the tunnel mile prediction error compensation model refers to the model obtained by training the support vector machine of the smallest squares using uniform tunnel group data, with temperature sequence, humidity sequence, and air pressure sequence as input and combined RGB color matching model prediction error as output; and f _Hz is the position update rate.

The present invention realizes the intelligent perception of the distance between the train in the tunnel and the tunnel exit using artificial intelligence big data analysis technology through the data of the sequence of temperature, humidity sequence and air pressure sequence at both ends of the train, and the present invention does not require any track equipment, so that the cost is low and practical utility high.

The specific process of determining the final output value O ( i ) of the distance between the train's current position and the tunnel exit includes:

1) obtaining the meteorological parameters of the tunnel in order to create a database of the meteorological parameters of the tunnel;

2) classification, based on the tunnel meteorological parameters database, the meteorological parameters in the tunnel meteorological parameters database according to the geographic location of the tunnel and the meteorological conditions inside and outside the tunnel, so as to obtain uniform samples in the area so as to realize the classification of groups of tunnels with similar attributes ;

3) additional classification of homogeneous samples in the area to obtain sample samples of each category of tunnel groups;

4) performing phase space reconstruction on sample samples of each category of tunnel groups to obtain color images of changes in barometric and meteorological parameters at the head and tail of the train, i.e. pattern images; wherein the template images corresponding to all sample samples constitute a library of combined sample sample templates;

5) collecting sample points at the head and tail of the train at the current position and sequences of samples with a time period before the current position, respectively, and performing phase space and RGB color space combination reconstruction to create images { c _h , c _f } of the current position features ; calculating correlations between the current position feature images and the template images in the combined type sample template library to determine the position of the best current position match in the combined type sample template library;

6) training the least squares support vector machine using uniform tunnel group data, with temperature sequence, humidity sequence, and air pressure sequence as input and best match position prediction error as output, to establish a tunnel mile path prediction error compensation model; and

7) obtaining meteorological parameters during the operation of the train in real time and obtaining the final mileage to the tunnel exit using the tunnel mileage prediction error compensation model, that is, obtaining O ( i ).

The method according to the present invention can make full use of changes in meteorological parameters in the tunnel to realize intelligent perception in the position of the train, does not require track equipment, and therefore has the advantages of simplicity and practical usefulness.

The specific implementation process for step 2) includes:

1) classification of groups of tunnels into N categories in accordance with the geographical location of the samples of tunnels; and

2) obtaining distributions of average temperatures within T min before the train enters the tunnel for one year in a given area, approximating the distributions of average temperatures in the tunnel of this area using a Gaussian distribution function, dividing the approximated distributions of average temperatures into K equal sections in accordance with the probability , defining temperature samples that belong to the same area as uniform temperature samples in the area, and defining a temperature sequence, a humidity sequence, and an air pressure sequence corresponding to the uniform temperature samples as homogeneous samples.

The present invention uses geographic location to roughly classify tunnels, and then uses meteorological parameters to accurately classify tunnels in the same area, thereby ensuring the reliability of pattern samples.

The specific implementation process for step 3) includes:

1) creating an autoregressive integrated moving average model for each time series of homogeneous samples in a given area, and extracting the characteristic values of each time series, that is, the parameters of autoregressive terms, integrated terms and moving regression terms of each time series; moreover, all the characteristic values of homogeneous samples in the region constitute the characteristic matrix A ; wherein the time sequences include a temperature sequence, a humidity sequence, and an air pressure sequence;

2) reducing the size of the feature matrix A , and selecting the M major components with the largest contribution to characterize the information of the original feature matrix, thus obtaining the transformed matrix A' ;

3) definition of the kernel function k = αk _rbf + βk _linear +(1- α - β ) k _laplace ; wherein k _rbf is the radial basis kernel function, k _linear is the linear kernel, and k _laplace is the Laplace kernel function; arbitrarily determining the initial values of the coefficients α , β and the number n ₀ of the categories of the kernel function, and α , β ∈[0, 1], and n ₀ is a positive integer less than 30;

, причем avg(C _i ) представляет собой среднее расстояние выборок в кластере C _i выборок, avg(C _j ) представляет собой среднее расстояние выборок в кластере C _j выборок, и d _cen (C _i , C _j ) представляет собой расстояние между центральными точками кластера C _i выборок и кластера C _j выборок;4) mapping the characteristic values in matrix A' onto the characteristic space corresponding to the kernel function k to obtain a new characteristic matrix A", according to the initial number n ₀ of categories, using the characteristic values in A" as an input to cluster the characteristic values in A" through a k -means clustering algorithm to classify homogeneous samples in a region into n categories, with each category of samples forming a cluster of samples; computing an optimization objective function fitness =

, where avg ( C _i ) is the average distance of the samples in the cluster C _i of samples, avg ( C _j ) is the average distance of the samples in the cluster C _j of samples, and d _cen ( C _i , C _j ) is the distance between the center points cluster C _i samples and cluster C _j samples;

5) update, in accordance with the current value of the fitness optimization objective function, the values of α , β and n ₀ by means of optimization according to the “gray wolf” algorithm;

6) repeating step 4) and step 5) until the set number m =100 optimizations is reached, wherein the values of α , β and n ₀ at this time represent the final optimized parameter values, and the clustering results at the final optimized parameter values are final clustering results; and

} (i=0, 1, 2…P, n=0, 1, 2…E), соответствующей Р выборкам, ближайшим к центру кластера, при этом

соответствует последовательности температуры, полученной в головной части поезда, последовательности влажности, полученной в головной части поезда, последовательности воздушного давления, полученной в головной части поезда, последовательности температуры, полученной в хвостовой части поезда, последовательности влажности, полученной в хвостовой части поезда, и последовательности воздушного давления, полученной в хвостовой части поезда, соответственно; и определение исходной временной последовательности, соответствующей (P+1) выборкам, как типовой выборки текущей категории групп тоннелей.7) classifying the homogeneous samples in the given area into E sample clusters according to the final clustering results, and obtaining the cluster center of each sample cluster and the original time sequence {

} ( i =0, 1, 2… P , n =0, 1, 2… E ) corresponding to P samples closest to the center of the cluster, while

corresponds to the sequence of temperature obtained at the head of the train, the sequence of humidity obtained at the head of the train, the sequence of air pressure obtained at the head of the train, the sequence of temperature obtained at the tail of the train, the sequence of humidity obtained at the tail of the train, and the sequence of air pressure obtained in the tail section of the train, respectively; and determining the initial time sequence corresponding to the (P+1) samples as a typical sample of the current category of tunnel groups.

The present invention determines the type samples by an adaptive method and optimizes the feature mapping space and the feature clustering parameters, which is beneficial in obtaining more representative type samples and guarantees the efficiency of the templates.

The specific implementation process for step 5) includes:

1) determining the current sample points and the previous G sample points in the sequence of temperature, humidity sequence and air pressure sequence of the head and tail of the train at the current position;

2) performing a phase space reconstruction using the delay coordinate method to obtain six two-dimensional reconstruction matrices representing the characteristics of temperature, humidity and air pressure changes at the head and tail of the train, dividing the six matrices into matrices corresponding to the head of the train, and matrices, corresponding to the rear of the train, and combining the six matrices according to the RGB color space to form images { c _h , c _f } of the current position indicia;

3) performing a convolution operation on the images of current position features and images in the library of combined templates of type samples to obtain a plurality of one-dimensional sequences;

4) sorting the elements in all one-dimensional sequences from large to small to determine the L largest elements as candidate elements, where the positions corresponding to the candidate elements before sorting are the candidate positions, and the mileage to the tunnel exit corresponding to the positions -candidates are s _j ( j =1, 2... L ); and

текущего положения в библиотеке объединенных шаблонов типовых выборок.5) averaging the mileage to the tunnel exit corresponding to the candidate locations, with the average being the location of the best match

the current position in the library of merged type selection templates.

The present invention converts one-dimensional time sequences into two-dimensional images by phase space reconstruction, which is beneficial in finding detailed information in time sequences, and is beneficial for exact pattern matching in multiple time sequences.

The specific implementation process for step 6) includes:

1) definition of input samplesI=(T _h ,H _h ,P _h ,T _f ,H _f ,P _f ), andT _h =(t_one , t ₂…t _G ) is a sequence of temperature of the current sample points andG previous sample points at the head of the train in the tunnel,H _h is a sequence of humidity at the head of the train with a lengthG,P _h is the sequence of air pressure at the head of the train with a lengthG,T _f is the sequence of temperature at the rear of the train with lengthG,H _f is a sequence of humidity in the tail section of the train with lengthG, andP _f is the sequence of air pressure in the tail section of the train with lengthG; defining output samples as valuesε mistakes path in miles corresponding to the current position; with combinationsY={I,ε} input samples and output samples constitute simulation samples;

2) classification of simulation samples training set and validation set;

3) performing binary encoding on a feature of each dimension in the input samples I , wherein when the code value corresponding to a feature of some dimension is 1, the feature is selected as an input variable of the least squares support vector machine; when the code value corresponding to a feature of a certain dimension is 0, the feature of that dimension is discarded;

машины опорных векторов наименьших квадратов, причем k=1, 2..., N ₁, и N ₁ представляет собой число пар выборок в наборе валидации; и4) updating the input samples based on the current feature code values to obtain an updated training set and validation set, and training the least squares support vector machine with the updated training set data; inputting the updated validation set data into the trained least squares support vector machine to obtain the output sequence

least squares support vector machines, where k =1, 2..., N ₁ , and N ₁ is the number of pairs of samples in the validation set; and

, то есть получение оптимальных входных признаков и оптимальной машины опорных векторов наименьших квадратов, причем оптимальная машина опорных векторов наименьших квадратов представляет собой модель компенсации ошибки предсказания пути в милях; причем

представляет собой действительное выходное значение в наборе валидации.5) repeating step 3) and step 4) to determine the input features and the least squares support vector machine that minimize the optimization objective function

, that is, obtaining optimal input features and an optimal least squares support vector machine, the optimal least squares support vector machine being a path prediction error compensation model in miles; and

represents the actual output value in the validation set.

The present invention establishes an error correction model by obtaining data that is beneficial in reducing the error caused by the deviation of patterns and actual values, and further improves the accuracy of mileage prediction.

The specific implementation process for step 7) includes:

1) obtaining data on the distribution of average temperatures within T min before the train enters the tunnel, obtaining the temperature and humidity sequences at the head and tail of the train at the current position, obtaining the output value of matching with the template O ^M ( i ), and obtaining the output result compensation errors ε ( i ) using the tunnel mileage path prediction error compensation model; and

2) obtaining the distance in miles from the current position of the train to the exit of the tunnel using the following equation: O ( i )= O ^M ( i )+ ε ( i ).

The present invention calculates the mileage of a tunnel by a combination of approximate pattern matching prediction and error correction compensation, which is advantageous for improving the accuracy of mileage calculation.

The present invention further provides a system for calculating the duration of a train passing through a tunnel, including:

a best match position obtaining module, configured to calculate correlations between current position feature images and patterns in the combined type sample template library, obtain the best match position of the current position in the combined type sample template library; at the same time, images of signs of the current position are obtained by reconstructing the phase space and combining the RGB color space of the obtained sequences of meteorological parameters of sample points in the head and tail of the train in the current position and meteorological parameters with a time period before the current position, images of templates in the library of combined templates of typical samples are color images regarding the change of meteorological parameters at the head and tail of the train, and the meteorological parameters at the head and tail of the train are obtained by reconstructing the phase space of the type samples of each category of tunnel groups, and the type samples of each category of tunnel groups refer to the samples obtained by classifying meteorological parameters according to the geographic location of the tunnel and the meteorological conditions inside and outside the tunnel, and further classifying the class ssifiable homogeneous samples in the region;

tunnel mile path prediction error compensation model obtained by training a least squares support vector machine using uniform tunnel group data, with temperature sequence, humidity sequence, and air pressure sequence as input and best match position prediction error as output, with the output of this model is the compensation error ε ( i ); and

a remaining duration calculation module configured to estimate the remaining duration t that a train travels in a given tunnel according to the following equation: t=O ( i )/( f _Hz (O(i)-O(i- 1))) ; where O ( i )= O ^M ( i )+ ε ( i ), and O ( i -1) is the distance in miles from the previous position relative to the current position of the train to the exit of the tunnel; O ^M ( i ) is the position of the best match; and f _Hz is the train position update rate.

The module for obtaining the position of the best match includes:

a data acquisition unit configured to determine current sample points and previous G sample points in a temperature sequence, a humidity sequence, and a train head and tail air pressure sequence at a current position;

a feature image computing unit, configured to perform phase space reconstruction with the delay coordinate method to obtain six two-dimensional reconstruction matrices representing temperature, humidity, and air pressure change characteristics at the head and tail of the train, divide the six matrices into matrices corresponding to the head the train part, and matrices corresponding to the rear of the train, combine six matrices according to the RGB color space to form images { c _h , c _f } of current position indicia;

a convolution unit configured to perform a convolution operation on the current position feature images and the images in the combined sample sample library to obtain a plurality of one-dimensional sequences;

a sorter configured to sort the elements in all one-dimensional sequences from large to small to determine the L largest elements as candidate elements, the positions corresponding to the candidate elements before sorting being the candidate positions, and the path values in miles to the exit the tunnels corresponding to candidate positions are s _j ( j =1, 2... L ); and

текущего положения в библиотеке объединенных шаблонов типовых выборок.an output unit configured to average the mileage to the tunnel exit corresponding to the candidate positions, the average being the position of the best match

the current position in the library of merged type selection templates.

The training process of the tunnel mileage prediction error compensation model according to the present invention includes:

1) definition of input samplesI=(T _h ,H _h ,P _h ,T _f ,H _f ,P _f ), andT _h =(t_one , t ₂…t _G ) is a sequence of temperature of the current sample points andG previous sample points at the head of the train in the tunnel,H _h is a sequence of humidity at the head of the train with a lengthG,P _h is the sequence of air pressure at the head of the train with a lengthG,T _f is the sequence of temperature at the rear of the train with lengthG,H _f is a sequence of humidity in the tail section of the train with lengthG, andP _f is a sequence of air pressure in the tail section with a lengthG; defining output samples as valuesε mistakes path in miles corresponding to the current position; with combinationsY={I,ε} input samples and output samples constitute simulation samples;

2) classifying the simulation samples into a training set and a validation set;

3) performing binary encoding on a feature of each dimension in the input samples I , wherein when the code value corresponding to a feature of some dimension is 1, the feature is selected as an input variable of the least squares support vector machine; when the code value corresponding to a feature of a certain dimension is 0, the feature of that dimension is discarded;

машины опорных векторов наименьших квадратов, причем k=1, 2..., N ₁ и N ₁ представляет собой число пар выборок в наборе валидации; и4) updating the input samples based on the current feature code values to obtain an updated training set and validation set, and training the least squares support vector machine with the updated training set data; inputting the updated validation set data into the trained least squares support vector machine to obtain the output sequence

least squares support vector machines, where k =1, 2..., N ₁ and N ₁ is the number of pairs of samples in the validation set; and

, то есть получение оптимальных входных признаков и оптимальной машины опорных векторов наименьших квадратов, причем оптимальная машина опорных векторов наименьших квадратов представляет собой модель компенсации ошибки предсказания пути в милях; причем

представляет собой действительное выходное значение в наборе валидации.5) repeating step 3) and step 4) to determine the input features and the least squares support vector machine that minimize the optimization objective function

, that is, obtaining optimal input features and an optimal least squares support vector machine, the optimal least squares support vector machine being a path prediction error compensation model in miles; and

represents the actual output value in the validation set.

As an inventive concept, the present invention further provides a computer-readable storage medium storing a program; and the program is configured to execute the steps of the method in accordance with the present invention.

Compared with the prior art, the present invention has the following beneficial results:

1. The present invention uses artificial intelligence big data analysis technology to fully explore the potential patterns of changes in barometric and meteorological parameters in a tunnel with a change in tunnel depth. Through the data of temperature sequence, humidity sequence and air pressure sequence obtained at both ends of the train, intelligent perception of the distance between the train in the tunnel and the exit of the tunnel is realized, and based on this, an effective estimate of the time spent by the train in the tunnel is realized.

2. The present invention can provide passengers with accurate informational indications (such as temperature, humidity, and remaining time in the tunnel), which is useful for improving the experience of passengers.

3. After the simulation is completed, only temperature, humidity and air pressure sensors installed on the train are required to receive input data, without track equipment, so the present invention is of great importance to promote development.

Brief description of the drawings

Fig. 1 is a basic flow chart of the present invention;

Fig. 2 is a schematic diagram of phase space reconstruction and combined RGB pattern matching in accordance with the present invention.

DETAILED DESCRIPTION OF THE EMBODIMENTS

As shown in FIG. 1, the specific steps of Embodiment 1 of the present invention are as follows:

Stage 1: Obtaining barometric and meteorological parameters of the tunnel and creating a database of barometric and meteorological parameters of the tunnel.

On-board sensors distributed at both ends of the train receive, in real time, sequences of temperature, humidity and air pressure as the train passes through the tunnel, and transponder devices pre-installed in the tunnel and on the train receive time sequences of the train's mileage in the tunnel , with a sampling rate of 10 Hz. The average temperature and humidity values obtained at T min before the train enters the tunnel are used as estimated local average temperature and humidity values. The sequence of temperature, sequence of humidity and sequence of air pressure obtained by the sensors at both ends of the train when the train passes through the tunnel at the same time, and the estimated values obtained before entering the tunnel, the average temperature and average humidity of the area where the tunnel is located, form a set of samples meteorological parameters of the tunnel. The samples of meteorological parameters of the tunnel, obtained during the operation of all trains in the given area in one year, form a database of meteorological parameters of the tunnel. The range of T values is [10, 20] and T =10 in the present invention.

Step 2: Classification of barometric and meteorological parameters of tunnel groups.

The meteorological parameters in the database are classified based on the tunnel meteorological parameters database according to the geographic location of the tunnel and the meteorological conditions inside and outside the tunnel, so as to realize the classification of groups of tunnels with similar attributes. The specific implementation process is as follows:

Step A1: Tunnel groups are classified into 7 categories according to the classification of China's climatic characteristics through China's architectural zoning and according to the geographic location of the tunnel samples in the tunnel meteorological parameters database.

Step A2: Average temperature distributions are obtained for T min before the train enters the tunnel for one year in the given area, and the average temperature distributions are approximated using a Gaussian distribution function. The average temperature distributions are divided into 10 equal sections according to probability, the temperature samples that belong to the same section are defined as uniform temperature samples in a given area, and the temperature sequence, humidity sequence, and air pressure sequence corresponding to uniform temperature samples are defined as homogeneous samples.

Stage 3: Representation of homogeneous samples and sample samples in the area.

Each homogeneous sample in the area, which is obtained on the basis of an approximate classification into categories of tunnel groups and by entering a set of attributes, is further classified. More specifically, this process includes the following sub-steps:

Step B1: Characterize the change in temperature, humidity and air pressure sequences in homogeneous samples as the train passes through the tunnel. The specific step involves modeling an autoregressive integrated moving average (ARIMA) model for each time series, and extracting parameters of the autoregressive terms, integrated terms, and moving regression terms of each series. More specifically, 6 time sequences can be obtained when the train passes through the tunnel at the same time, then the characteristic value extracted when the train passes through the tunnel at the same time can be expressed as α =( p ₁ , d ₁ , q ₁ , p _{2 ,} d _{2 ,} q ₂ , . _ _{_} _ _{_} _ _{_} _ _{_} time sequence 1. All characteristic quantities of a homogeneous sample in the region form a characteristic matrix A .

Step B2: Dimensional constraint is performed by principal component analysis (PCA) on the characteristic matrix A formed by the homogeneous sample characteristic values in the region. The 5 principal components with the largest contribution are selected to characterize the information of the original characteristic matrix A , and the transformed matrix A' is obtained.

Stage B3: The kernel function is defined as:

k = αk _rbf + βk _linear +(1- α - β ) k _laplace (1)

In this equation, k _rbf is the radial basis kernel function, k _linear is the linear kernel, and k _laplace is the Laplace kernel function. The characteristic values in the matrix A' are mapped onto the characteristic space corresponding to the kernel function k.

Step B4: Optimized objects are determined, i.e., the coefficients α , β and the number of categories of the kernel function are optimized by gray wolf optimization (GWO), where α , β ∈[0, 1] and the number of categories is a positive integer less than 20.

Step B5: The optimization objective function is defined as:

(2) fitness =

(2)

In this equation, avg ( C _i ) is the average distance of samples in cluster C _i , and d _cen ( C _i , C _j ) is the distance between the center points of cluster C _i and cluster C _j .

} (i=0, 1, 2, …, 5) соответствующую 5 выборкам, ближайшим к центру кластера.

соответствует последовательности температуры, полученной в головной части поезда, последовательности влажности, полученной в головной части поезда, последовательности воздушного давления, полученной в головной части поезда, последовательности температуры, полученной в хвостовой части поезда, последовательности влажности, полученной в хвостовой части поезда, и последовательности воздушного давления, полученной в хвостовой части поезда.Step B6: Characteristics after dimension reduction are grouped into clusters according to the parameters set above by k -means clustering algorithm optimized by gray wolf optimization. Get the cluster centers of each cluster of samples and the original time sequence {

} ( i =0, 1, 2, …, 5) corresponding to 5 samples closest to the cluster center.

corresponds to the sequence of temperature obtained at the head of the train, the sequence of humidity obtained at the head of the train, the sequence of air pressure obtained at the head of the train, the sequence of temperature obtained at the tail of the train, the sequence of humidity obtained at the tail of the train, and the sequence of air pressure received in the tail section of the train.

Stage 4: Creation of a library of combined type selection templates.

The phase space reconstruction is performed on the sample sample, the delay time is set to 1, the window length is set to 5, the phase space reconstruction is performed on the temperature, humidity, and air pressure sequences using the delay coordinate method to obtain three two-dimensional reconstruction matrices representing temperature change characteristics, humidity and air pressure at the head of the train, and three two-dimensional reconstruction matrices representing the characteristics of change in temperature, humidity and air pressure at the tail of the train, the corresponding three matrices are combined in accordance with the RGB color space to form color images representing the characteristics of changes in barometric and meteorological parameters at the head or tail of the train, and the color images are template images { h _i , f _i } ( i= 0, 1, 2…5).

Step 5: Training a combined RGB pattern color matching model (as shown in FIG. 2).

The meteorological data of the sample points at the head and tail of the train at the current position and the sequence of meteorological samples for the time period ahead are obtained respectively, and phase space reconstruction and RGB color space combination are performed to create images { c _h , c _f } of the current position features. Correlations between the current position feature module and the images in the template library are calculated to determine the best match position for the current position in the template library. This process includes the following steps:

Step C1: Relevant meteorological data of the current sample points in the sequences of temperature, humidity and air pressure at the head and tail of the train at the current position and the previous 19 sample points are obtained.

Step C2: The delay time is set to 1, the window length is set to 5, the phase space reconstruction is performed using the delay coordinate method to obtain six two-dimensional reconstruction matrices representing the characteristics of temperature, humidity and air pressure changes at the head and tail of the train, six matrices are divided into matrices corresponding to the head of the train and matrices corresponding to the tail of the train, and the six matrices are combined according to the RGB color space to form images { c _h , c _f } of the current position features.

Step C3: The convolution operation g _i =h _i Υ c _h +f _i Υ c _f ( i =0, 1, 2…5) is performed on the current position feature images and template library images, with each g _i being a one-dimensional sequence.

Step C4: The elements in all sequences g _i are sorted from largest to smallest to determine the 5 largest elements as candidate elements, with the positions corresponding to the candidate elements before sorting being the candidate positions, and the mileage to the tunnel exit corresponding to candidate positions are s _j ( j =1, 2…5).

Step C5: The tunnel exit miles corresponding to the candidate locations are averaged, with the average being the current pattern match output, i.e., the formula for calculating the pattern match output is:

Step 6: Establishment of a joint pattern matching error compensation model.

A least squares support vector machine (LSSVM) is trained using uniform tunnel group data, with temperature sequence, humidity sequence, and air pressure sequence as input and pattern matching model prediction error as output, to establish a tunnel mileage prediction error compensation model . This process includes the following steps:

Step D1: Training samples are determined and input samples I =( T _h , H _h , P _h , T _f , H _f , P _f ) are determined, with T _h = (t ₁ , t ₂ …, t ₁₉ , t ₂₀ ) is the sequence of the temperature of the current sample points and the 19 previous sample points at the head of the train in the tunnel. Similarly, H _h is the train head humidity sequence with length 20, P _h is the train head air pressure sequence with length 20, T _f is the train tail temperature sequence with length 20, H _f is the sequence humidity at the tail of a train with a length of 20, and P _f is the sequence of air pressure at the tail of a train with a length of 20. The output samples are the mileage track error ε values corresponding to the current position. The input and output combinations Y ={ I , ε } constitute the simulation samples. For each homogeneous group of tunnels, 3000 samples are chosen to establish a mileage prediction error compensation model.

Step D2: The simulation samples are classified into a training set and a validation set. By random sampling without replacement, 70% of the 3000 pairs of samples among the simulation samples are chosen as the training set and 30% as the validation set.

Step D3: Optimization objects are defined and optimization values are initialized. The input features of the model are optimized by means of a binary antlion optimizer algorithm, that is, binary encoding is performed on a feature of each dimension in the input samples I . When the code value corresponding to a feature of some dimension is 1, the feature is selected as an input variable of the LSSVM model. When the code value corresponding to a feature of a certain dimension is 0, the feature of that dimension is discarded. The 60 dimension features are randomly initialized and encoded as 0 or 1.

модели (k=1, 2…900). Целевая функция оптимизации определяется как:Step D4: The optimization objective function is determined. The input features are determined based on the code values of the current features, and the LSSVM model is trained on the training set data. Validation set data is fed into the trained LSSVM model to produce an output sequence

models ( k =1, 2…900). The optimization objective function is defined as:

, (k=1, 2…900) представляет собой действительное выходное значение набора валидации.In the equation

, ( k =1, 2…900) is the actual output value of the validation set.

Step D5: An optimized prediction model is derived. Iterative optimization operations are performed through a binary antlion algorithm to determine the optimal input features and the LSSVM model, the LSSVM model is the LSSVM mileage prediction error compensation model.

Step 7: Get Test Data and Call Mileage Prediction Model

During the operation of the train, real-time barometric and meteorological parameters are obtained to form images of the signs of the current status, search and match 3D image patterns; and the error compensation input variables and the error compensation model are determined. The pattern match output is combined with the error compensation output to obtain the final mileage to the tunnel exit. This process includes the following steps:

Step E1: Get the average temperature distribution within T min before the train enters the tunnel. The category of the current status sample is determined (that is, the meteorological parameters of the current position of the train). The current temperature sequence and the current humidity sequence are obtained using temperature and humidity sensors installed at the head and tail of the train.

Step E2: The corresponding template library is retrieved according to the sampling category determined according to the current average temperature, the current status.

Step E3: Determine the position of the best match in the template library with reference to the process in step 3, and output values of the match with the template O ^M ( i ) of the current sample points.

Step E4: The current status model input vectors I are obtained with reference to process C1 in step 6 and substituted into the LSSVM to be trained to obtain compensation error outputs ε ( i ).

Step E5: Combine the 3D pattern match outputs and the LSSVM model outputs to get the final distance output between the current position and the tunnel exit:

Step 8: Calculate the predicted remaining duration until the tunnel exits

The remaining duration for which the train passes through this tunnel can be estimated according to the following equation:

In the equation, f _Hz is the position update rate, and f _Hz =50 Hz in the present invention.

According to the method in Embodiment 1, Embodiment 2 of the present invention provides a system for calculating the duration of a train passing through a tunnel, including:

a best match position obtaining module, configured to calculate correlations between current position feature images and patterns in the combined type sample pattern library to obtain the best match position of the current position in the combined type sample template library; moreover, the images of the signs of the current position are obtained by reconstructing the phase space and the combination of the RGB color space of the obtained sequences of meteorological parameters of sample points at the head and tail of the train at the current position and meteorological parameters with a time period before the current position, the template images in the library of combined type sample templates represent are color images regarding the change of meteorological parameters at the head and tail of the train, and the meteorological parameters at the head and tail of the train are obtained by reconstructing the phase space of the type samples of each category of tunnel groups, and the type samples of each category of tunnel groups refer to the samples obtained by classification of meteorological parameters according to the geographic location of the tunnel and meteorological conditions inside and outside the tunnel and additional classification class fifiable homogeneous samples in the region;

tunnel mile path prediction error compensation model obtained by training a least squares support vector machine using uniform tunnel group data, with temperature sequence, humidity sequence, and air pressure sequence as input and best match position prediction error as output, with the output of this model is the compensation error ε ( i ); and

a remaining duration calculation module configured to estimate the remaining duration t that the train passes in a given tunnel according to the following equation: t = O ( i )/ f _Hz ( O ( i ) - O ( i -1))); where O ( i )= O ^M ( i )+ ε ( i ), and O ( i -1) is the distance in miles from the previous position relative to the current position of the train to the exit of the tunnel; O ^M ( i ) is the position of the best match; f _Hz is the train position update rate.

The module for obtaining the position of the best match includes:

a data acquisition unit configured to determine current sample points and previous G sample points in a temperature sequence, a humidity sequence, and a train head and tail air pressure sequence at a current position;

a feature image computing unit, configured to perform phase space reconstruction with the delay coordinate method to obtain six two-dimensional reconstruction matrices representing temperature, humidity, and air pressure change characteristics at the head and tail of the train, divide the six matrices into matrices corresponding to the head train part and train tail matrices, and combine the six train head and tail matrices according to the RGB color space to form current position indicia { c _h , c _f } images;

a convolution unit configured to perform a convolution operation on the current position feature images and the images in the combined sample sample library to obtain a plurality of one-dimensional sequences;

a sorter configured to sort the elements in all one-dimensional sequences from large to small to determine the L largest elements as candidate elements, the positions corresponding to the candidate elements before sorting being the candidate positions, and the path values in miles to the exit the tunnels corresponding to candidate positions are s _j ( j =1, 2... L ); and

наилучшего совпадения текущего положения в библиотеке объединенных шаблонов типовых выборок.an output unit configured to average the mileage to the tunnel exit corresponding to the candidate positions, the average being the position

the best match of the current position in the library of merged pattern samples.

The data acquisition unit in this embodiment includes various sensors installed at the head and tail of the train, such as temperature sensors, humidity sensors, and air pressure sensors.

The process of training the tunnel mileage prediction error compensation model includes the following:

1) Input samples are determinedI=(T _h ,H _h ,P _h ,T _f ,H _f ,P _f ), andT _h =(t_one , t ₂…t _G ) is a sequence of temperature of the current sample points andG previous sample points at the head of the train in the tunnel,H _h is a sequence of humidity at the head of the train with a lengthG,P _h is the sequence of air pressure at the head of the train with a lengthG,T _f is the sequence of temperature at the rear of the train with lengthG,H _f is a sequence of humidity in the tail section of the train with lengthG, andP _f is the sequence of air pressure in the tail section of the train with lengthG; output samples are defined as valuesε mistakes paths in miles corresponding to the current position; and combinationsY={I,ε} input samples and output samples constitute simulation samples;

2) The simulation samples are classified into a training set and a validation set;

3) Binary coding is performed on the feature of each dimension in the input samples I , wherein when the code value corresponding to the feature of some dimension is 1, the feature is selected as the input variable of the least squares support vector machine; when the code value corresponding to a feature of a certain dimension is 0, the feature of that dimension is discarded;

машины опорных векторов наименьших квадратов, причем k=1, 2..., N ₁ и N ₁ представляет собой число пар выборок в наборе валидации; 4) The input samples are updated based on the current feature code values to obtain an updated training set and a validation set, and the least squares support vector machine is trained on the updated training set data; the updated validation set data is fed into the trained least squares support vector machine to produce the output sequence

least squares support vector machines, where k =1, 2..., N ₁ and N ₁ is the number of pairs of samples in the validation set;

, то есть получают оптимальные входные признаки и оптимальную машину опорных векторов наименьших квадратов, причем оптимальная машина опорных векторов наименьших квадратов представляет собой модель компенсации ошибки предсказания пути в милях; причем

представляет собой действительное выходное значение в наборе валидации.5) Step 3) and step 4) are repeated to determine the input features and the least squares support vector machine that minimize the optimization objective function

, that is, the optimal input features and the optimal least squares support vector machine are obtained, the optimal least squares support vector machine being the path prediction error compensation model in miles; and

represents the actual output value in the validation set.

Embodiment 3 of the present invention provides a computer-readable storage medium storing a program; and the program is configured to execute the steps of the method according to Embodiment 1 of the present invention.

Claims (52)

1. A method for calculating the duration of the passage of a train through a tunnel, and the remaining duration for which the train passes through this tunnel is estimated using the following equation: t = O ( i ) / f _Hz ( O ( i ) - O ( i -1) ); moreover, the distance in miles from the current position of the train to the exit of the tunnel is O ( i )= O ^M ( i )+ ε ( i ); O ( i -1) represents the distance in miles from the last position of the current position of the train to the exit of the tunnel; O ^M ( i ) is the position of the best match of the current position in the combined type sample template library, which is obtained after calculating the correlations between the current position feature images and the images in the combined type sample template library; images of signs of the current position are obtained by reconstructing the phase space and the combination of the RGB color space of the obtained sequences of meteorological parameters of sample points at the head and tail of the train at the current position and meteorological parameters with a time period before the current position; template images in the library of combined templates of type samples are color images regarding the change in meteorological parameters at the head and tail of the train; meteorological parameters at the head and tail of the train are obtained by reconstructing the phase space of typical samples of each category of tunnel groups; sample samples of each category of tunnel groups refer to samples obtained by classifying meteorological parameters according to the geographic location of the tunnel and meteorological conditions inside and outside the tunnel and further classifying the classified homogeneous samples in the area; ε ( i ) is the compensation error obtained by inputting the meteorological parameters obtained during the operation of the train into the tunnel mile prediction error compensation model, and the tunnel mile prediction error compensation model refers to the model obtained by training the support vector machine of the smallest squares using uniform tunnel group data, with temperature sequence, humidity sequence, and air pressure sequence as input and combined RGB color matching model prediction error as output; and f _Hz is the position update rate; and meteorological parameters refer to the sequence of temperature corresponding to the head of the train, the sequence of humidity corresponding to the head of the train, the sequence of air pressure corresponding to the head of the train, the sequence of temperature corresponding to the tail of the train, the sequence of humidity corresponding to the tail of the train, the sequence of air pressure, corresponding tail section of the train. 2. The method for calculating the duration of the passage of the train through the tunnel according to claim 1, and the specific process of determining the value O ( i ) of the path in miles from the current position of the train to the exit of the tunnel contains: 1) obtaining the meteorological parameters of the tunnel in order to create a database of the meteorological parameters of the tunnel; 2) classification, based on the tunnel meteorological parameters database, the meteorological parameters in the tunnel meteorological parameters database according to the geographic location of the tunnel and the meteorological conditions inside and outside the tunnel, so as to obtain uniform samples in the area so as to realize the classification of groups of tunnels with similar attributes ; 3) additional classification of homogeneous samples in the area to obtain sample samples of each category of tunnel groups; 4) performing a phase space reconstruction on typical samples of each category of tunnel groups to obtain color images of changes in barometric and meteorological parameters at the head and tail of the train, i.e. pattern images; wherein the template images corresponding to all sample samples constitute a library of combined sample sample templates; 5) collecting sample points at the head and tail of the train at the current position and sequences of samples with a time period before the current position, respectively, and performing phase space and RGB color space combination reconstruction to create current position feature images { c _h , c _f } ; calculating correlations between the current position feature images and the template images in the combined type sample template library to determine the position of the best current position match in the combined type sample template library; 6) training the least squares support vector machine using uniform tunnel group data, with temperature sequence, humidity sequence, and air pressure sequence as input and best match position prediction error as output, to establish a tunnel mile path prediction error compensation model; and 7) Obtaining meteorological parameters during the operation of the train in real time and obtaining the final mileage to the exit of the tunnel using the tunnel mileage prediction error compensation model, that is, obtaining O ( i ). 3. The method for calculating the duration of the passage of the train through the tunnel according to claim 2, and the specific process for implementing step 2) contains: 1) classification of tunnel groups into N categories according to the geographic location of the tunnel samples; and 2) obtaining distributions of average temperatures within T min before the train enters the tunnel for one year in a given area, approximating the distributions of average temperatures in the tunnel of this area using a Gaussian distribution function, dividing the approximated distributions of average temperatures into K equal sections in accordance with the probability , determining temperature samples that belong to the same area as uniform temperature samples in the area, and determining a temperature sequence, a humidity sequence, and an air pressure sequence corresponding to the uniform temperature samples as homogeneous samples; where T is the set time. 4. Method for calculating the duration of the passage of the train through the tunnel according to any one of paragraphs. 1-3, and the specific process for implementing step 3) contains: 1) creating an autoregressive integrated moving average model for each time series of homogeneous samples in a given area and extracting the characteristic values of each time series, that is, the parameters of autoregressive terms, integrated terms and moving regression terms of each time series; moreover, all the characteristic values of homogeneous samples in the region constitute the characteristic matrix A; wherein the time sequences comprise a temperature sequence, a humidity sequence, and an air pressure sequence; 2) reducing the size of the feature matrix A and selecting the M major components with the largest contribution to characterize the information of the original feature matrix, thus obtaining the transformed matrix A' ; 3) definition of the kernel function k = αk _rbf + βk _linear +(1- α - β ) k _laplace ; wherein k _rbf is the radial basis kernel function, k _linear is the linear kernel, and k _laplace is the Laplace kernel function; determining arbitrarily the initial values of the coefficients α , β and the number n ₀ categories of the kernel function, and α , β ∈[0, 1], and n ₀ is a positive integer less than 30; 4) mapping the characteristic values in matrix A' onto the characteristic space corresponding to the kernel function k to obtain a new characteristic matrix A" ; according to the initial number n ₀ of categories, using the characteristic values in A" as an input to cluster the characteristic values in A" by means of a k -means clustering algorithm, i.e. to classify homogeneous samples in an area into n categories, with each category of samples forming a cluster of samples; calculating the optimization objective function fitness =

, where avg ( C _i ) is the average distance of the samples in the cluster C _i of samples, avg ( C _j ) is the average distance of the samples in the cluster C _j of samples, and d _cen ( C _i , C _j ) is the distance between the center points of the cluster C _i samples and cluster C _j samples; 5) updating, in accordance with the current value of the fitness optimization objective function, the values of α , β and n ₀ by means of optimization according to the “gray wolf” algorithm; 6) repeating step 4) and step 5) until the set number m =100 optimizations is reached, where α , β and n ₀ at this time represent the final optimized parameter values, and the clustering results at the final optimized parameter values represent the final clustering results; and 7) classifying homogeneous samples in a given area into E sample clusters according to the final clustering results; and obtaining the cluster center of each cluster of samples and the original time sequence {

} ( i =0, 1, 2… P , n =0, 1, 2… E ) corresponding to P samples closest to the center of the cluster, while

corresponds to the sequence of temperature obtained at the head of the train, the sequence of humidity obtained at the head of the train, the sequence of air pressure obtained at the head of the train, the sequence of temperature obtained at the tail of the train, the sequence of humidity obtained at the tail of the train, and the sequence of air pressure obtained in the tail section of the train, respectively; and determining the initial time sequence corresponding to the (P+1) samples as a typical sample of the current category of tunnel groups. 5. A method for calculating the duration of the passage of the train through the tunnel according to any one of paragraphs. 1-4, and the specific process for implementing step 5) contains: 1) determining the current sample points and the previous G sample points in the sequence of temperature, humidity sequence and air pressure sequence of the head and tail of the train at the current position; 2) performing a phase space reconstruction with the delay coordinate method to obtain six two-dimensional reconstruction matrices representing temperature, humidity, and air pressure characteristics at the head and tail of the train; dividing the six matrices into matrices corresponding to the head of the train and matrices corresponding to the tail of the train, and combining the six matrices according to the RGB color space to form images { c _h , c _f } of the current position indicia; 3) performing a convolution operation on the images of current position features and images in the library of combined templates of type samples to obtain a plurality of one-dimensional sequences; 4) sorting the elements in all one-dimensional sequences from large to small to determine the L largest elements as candidate elements, where the positions corresponding to the candidate elements before sorting are the candidate positions, and the mileage to the tunnel exit corresponding to the positions -candidates are s _j ( j =1, 2... L ); and 5) averaging the mileage to the tunnel exit corresponding to the candidate locations, with the average being the location of the best match

the current position in the library of merged type selection templates. 6. Method for calculating the duration of the passage of the train through the tunnel according to any one of paragraphs. 1-5, and the specific process for implementing step 6) contains: 1) definition of input samplesI=(T _h ,H _h ,P _h ,T _f ,H _f ,P _f ), andT _h =(t_one , t ₂…t _G ) is a sequence of temperature of the current sample points andG previous sample points at the head of the train in the tunnel,H _h is a sequence of humidity at the head of the train with a lengthG,P _h is the sequence of air pressure at the head of the train with a lengthG,T _f is the sequence of temperature at the rear of the train with lengthG,H _f is a sequence of humidity in the tail section of the train with lengthG, andP _f is the sequence of air pressure in the tail section of the train with lengthG; defining output samples as valuesε mistakes path in miles corresponding to the current position; with combinationsY={I,ε} input samples and output samples constitute simulation samples; 2) classifying the simulation samples into a training set and a validation set; 3) performing binary encoding on the feature of each dimension in the input samples I ; moreover, when the code value corresponding to a feature of a certain dimension is equal to 1, the feature is selected as an input variable of the least squares support vector machine; when the code value corresponding to a feature of a certain dimension is 0, the feature of that dimension is discarded; 4) updating the input samples based on the code values of the current features to obtain an updated training set and validation set; and training the least squares support vector machine with the updated training set data; inputting the updated validation set data into the trained least squares support vector machine to obtain the output sequence

least squares support vector machines, where k =1, 2..., N ₁ , and N ₁ is the number of pairs of samples in the validation set; and 5) repeating step 3) and step 4) to determine the input features and the least squares support vector machine that minimize the optimization objective function

, that is, obtaining optimal input features and an optimal least squares support vector machine, the optimal least squares support vector machine being a path prediction error compensation model in miles; and

represents the actual output value in the validation set. 7. A method for calculating the duration of the passage of the train through the tunnel according to any one of paragraphs. 1-6, and the specific process for implementing step 7) contains: 1) obtaining data on the distribution of average temperatures within T min before the train enters the tunnel; obtaining temperature sequences and humidity sequences at the head and tail of the train at the current position; getting an output value O ^M ( i ) matching the template; and obtaining a compensation error output ε ( i ) using the tunnel mileage prediction error compensation model; and 2) obtaining the distance in miles from the current position of the train to the exit of the tunnel using the following equation: O ( i )= O ^M ( i )+ ε ( i ). 8. A system for calculating the duration of a train passing through a tunnel, comprising: a best match position obtaining module, configured to calculate correlations between current position feature images and patterns in the combined type sample pattern library to obtain the best match position of the current position in the combined type sample template library; the images of the signs of the current position are obtained by reconstructing the phase space and the combination of the RGB color space of the obtained sequences of meteorological parameters of sampling points in the head and tail of the train in the current position and meteorological parameters with a time period before the current position; the template images in the combined type sample template library are color images relative to the change of meteorological parameters at the head and tail of the train, and the meteorological parameters at the head and tail of the train are obtained by reconstructing the phase space of the type samples of each category of tunnel groups; and the sample samples of each category of tunnel groups refer to the samples obtained by classifying the meteorological parameters according to the geographic location of the tunnel and the meteorological conditions inside and outside the tunnel, and further classifying the classified homogeneous samples in the area; tunnel mileage prediction error compensation model obtained by training a least squares support vector machine using uniform tunnel group data with temperature sequence, humidity sequence, and air pressure sequence as input and best match position prediction error as output, the output of this model being is an error ε ( i ) compensation; and a remaining duration calculation module configured to estimate the remaining duration t that a train passes through a given tunnel according to the following equation: t=O ( i )/( f _Hz (O(i)-O(i- 1))) ; where O ( i )= O ^M ( i )+ ε ( i ), and O ( i -1) is the distance in miles from the previous position relative to the current position of the train to the exit of the tunnel; O ^M ( i ) is the position of the best match; and f _Hz is the train position update rate. 9. The system according to claim 8, wherein the module for obtaining the position of the best match contains: a data acquisition unit configured to determine current sample points and previous G sample points in a temperature sequence, a humidity sequence, and a train head and tail air pressure sequence at a current position; a feature image computing unit configured to perform phase space reconstruction with the delay coordinate method to obtain six two-dimensional reconstruction matrices representing temperature, humidity, and air pressure change characteristics at the head and tail of the train; divide the six matrices into matrices corresponding to the head of the train and matrices corresponding to the tail of the train, combine the six matrices according to the RGB color space to form images { c _h , c _f } of the current position indicia; a convolution unit configured to perform a convolution operation on the current position feature images and the images in the combined sample sample library to obtain a plurality of one-dimensional sequences; a sorter configured to sort the elements in all one-dimensional sequences from large to small, to determine the L largest elements as candidate elements, the positions corresponding to the candidate elements before sorting being the candidate positions, and the path values in miles to the exit the tunnels corresponding to candidate positions are s _j ( j =1, 2... L ); and an output unit configured to average the mileage to the tunnel exit corresponding to the candidate positions, the average being the position of the best match

current position in the library of merged pattern samples; wherein, preferably, the process of training the tunnel mileage prediction error compensation model according to the present invention comprises: 1) definition of input samplesI=(T _h ,H _h ,P _h ,T _f ,H _f ,P _f ), andT _h =(t_one , t ₂…t _G ) is a sequence of temperature of the current sample points andG previous sample points at the head of the train in the tunnel,H _h is a sequence of humidity at the head of the train with a lengthG,P _h is the sequence of air pressure at the head of the train with a lengthG,T _f is the sequence of temperature at the rear of the train with lengthG,H _f is a sequence of humidity in the tail section of the train with lengthG, andP _f is the sequence of air pressure in the tail section of the train with lengthG; defining output samples as valuesε mistakes path in miles corresponding to the current position; with combinationsY={I,ε} input samples and output samples constitute simulation samples; 2) classifying the simulation samples into a training set and a validation set; 3) performing binary encoding on a feature of each dimension in the input samples I , wherein when the code value corresponding to a feature of some dimension is 1, the feature is selected as an input variable of the least squares support vector machine; when the code value corresponding to a feature of a certain dimension is 0, the feature of that dimension is discarded; 4) updating the input samples based on the current feature code values to obtain an updated training set and validation set, and training the least squares support vector machine with the updated training set data; inputting the updated validation set data into the trained least squares support vector machine to obtain the output sequence

least squares support vector machines, where k =1, 2,..., N ₁ , and N ₁ is the number of pairs of samples in the validation set; and 5) repeating step 3) and step 4) to determine the input features and the least squares support vector machine that minimize the optimization objective function

, that is, obtaining optimal input features and an optimal least squares support vector machine, wherein the optimal least squares support vector machine is a path prediction error compensation model in miles, where

represents the actual output value in the validation set. 10. A computer-readable storage medium storing a program, the program being configured to execute the steps of a method according to any one of claims. 1-7.

Images