An intelligent perception and prediction system for the physical and mechanical state of roadbed in high-cold areas
By establishing an intelligent perception and forecasting system on railway subgrades in high-altitude areas, and using sensor networking and neural network models, the problem of low roadbed disease monitoring efficiency in the existing technology is solved, and efficient monitoring and forecasting of the physical and mechanical state of the roadbed is achieved. Ensures the service life of railway safe operation and rail structure.
Patent Information
- Application Number
- CN202110982878.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-25
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-08-25
AI Technical Summary
The existing technology is inefficient in monitoring and handling railway subgrade diseases and cannot meet the needs of high-speed railway development. Especially in high-altitude areas, the problem of freezing and thawing seriously affects driving safety and the service life of the track structure.
A smart perception and forecasting system for physical and mechanical states of roadbeds in high-altitude areas is designed, including roadbed catastrophe information database, data transmission and analysis system and disaster forecasting and early warning system. Through sensor networking, the roadbed hidden danger points are monitored in real time, intelligently transmit and analyze data, and provide Timely rescue of early warning level, and use IoT technology and neural network models to predict and warn data.
It realizes efficient monitoring and forecasting of the physical and mechanical state of the roadbed, improves data transmission efficiency, can quickly identify roadbed disasters, make timely forecasts and early warnings, ensures safe operation of railways, and improves the service life of the rail structure.
Smart Images

Figure CN113657041B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of intelligent monitoring and early warning of roadbed diseases, and relates to an intelligent perception and forecasting system for the physical and mechanical state of roadbed in high-cold areas. Background Art
[0002] In recent years, high-speed railways have developed rapidly in my country, and are particularly known for their speed, safety, and efficiency. It is estimated that by the end of 2021, the total mileage of high-speed railways in China will be close to 40,000 kilometers, ranking first in the world in total operating mileage.
[0003] Railway subgrade is a geotechnical structure built to meet the conditions for track laying and operation. It is a structure that bears and transmits the gravity of the track and the dynamic action of the train. It bears the deadweight of the soil and the weight of the pavement structure, and also bears the traffic load transmitted from the pavement. Under the influence of traffic load and environmental factors, the subgrade will suffer from uneven settlement, slurry overflow, mud and other diseases. In addition, due to the increasing construction mileage of high-speed railways in cold regions, the problem of frost heave and thaw settlement of high-speed railway subgrades in cold regions seriously affects driving safety, and the mismatch of nominal strength and stiffness of the subgrade affects driving comfort. With the rapid development of heavy-duty railway seamless lines, foundation deformation caused by subgrade frost damage has become a major engineering problem that needs to be urgently solved in the construction and operation management of heavy-duty railways in cold regions. On the one hand, subgrade frost damage will cause the overall unevenness of the rails, threatening the safety of driving. On the other hand, it will aggravate the damage of the line structure itself, affect the safety of the seamless line, and reduce the service life of the track structure.
[0004] At present, railway subgrade disease monitoring adopts a variety of means such as light dynamic probing, geological radar, transient surface wave method and soil sampling test to test the subgrade and obtain the subgrade disease information. However, this monitoring method has a large workload and low monitoring efficiency, and cannot perform intelligent processing on the monitoring data, which lags behind the needs of high-speed railway development. Summary of the invention
[0005] In view of the problems existing in the prior art, the present invention provides an intelligent perception and prediction system for the physical and mechanical state of roadbed in high-cold areas. The present invention constructs an information platform for emergency rescue of roadbed disasters, monitors roadbed hidden dangers in real time through sensor networking, and intelligently transmits, processes and analyzes them, evaluates the current warning level, and promptly mobilizes relevant management personnel to carry out real-time rescue and repair roadbed damage, providing reliable guarantee for the normal and safe operation of high-speed railways in cold regions.
[0006] The objective of the present invention is achieved through the following technical solutions:
[0007] An intelligent perception and prediction system for the physical and mechanical state of roadbed in high-cold areas, including three parts: roadbed disaster information database, data transmission and analysis system, and disaster forecasting and early warning system, among which:
[0008] The roadbed disaster information database is composed of a static information database and a dynamic information database;
[0009] The static information database is used to provide basic data information during the service period of roadbed construction, and the basic data information includes: vehicle condition information, geographic information, engineering geological information, historical disease information, and indoor and outdoor test information during the operation period;
[0010] The dynamic information library is used to provide physical state indicators and mechanical state indicators. The physical state indicators include: electric field, moisture field, and temperature field data of the physical state of the roadbed. The mechanical state indicators include: acceleration, dynamic strain, and dynamic stress data of the mechanical state of the roadbed.
[0011] The data transmission and analysis system has two functions: on the one hand, it has the function of close-range integration and long-range wireless transmission of sensor data; on the other hand, it has the function of predicting the stability service index of the roadbed based on the physical and mechanical perception data of the roadbed, specifically, the function of establishing a relationship model between physical and mechanical indicators based on relevant data analysis methods;
[0012] The disaster prediction and early warning system is used to predict the service status of the roadbed at different times in the future. The prediction and early warning system focuses on the evaluation of the service status of the roadbed in the time-varying process.
[0013] The roadbed disaster information database and data transmission and analysis system provide forecast and warning indicators and thresholds of each indicator for the disaster forecast and warning system. The roadbed disaster information database provides a basis for the establishment of the forecast and warning model in the data transmission and analysis system. The disaster forecast and warning system judges the results of the forecast and warning model, that is, whether the forecast or warning value is reached compared with the threshold.
[0014] In the present invention, the static information database includes a vehicle condition information database, a geographic information database, an engineering geological information database, a historical disease information database, and an indoor and outdoor test information database, wherein:
[0015] The train condition information database is used to provide train condition information such as running train model, speed, time, etc.
[0016] The geographic information database and engineering geological information database are used to provide hydrogeological overview, climate distribution characteristics, rainfall and precipitation data along the roadbed project;
[0017] The historical disease information database is used to provide data such as regular disease detection results;
[0018] The indoor and outdoor test information database is used to provide a multi-field mutual feedback empirical model related to the physical state of the roadbed and a dynamic characteristic evolution law model related to the mechanical state of the roadbed, wherein:
[0019] The multi-field mutual feedback empirical model is obtained through indoor tests, specifically referring to a model that can reflect the relationship between roadbed deformation, temperature, and unfrozen water content. The roadbed deformation is divided into two parts, one part is frost heave deformation, and the other part is thaw settlement deformation. The general rule for frost heave deformation is that the lower the unfrozen water content, the higher the temperature, the greater the frost heave; the general rule for thaw settlement deformation is that the higher the unfrozen water content, the higher the temperature, the greater the thaw settlement;
[0020] The model of the evolution law of dynamic characteristics is obtained through indoor experiments and previous research literature. It specifically refers to a model that can reflect the relationship between roadbed deformation, temperature, and unfrozen water content under the action of train load. In this model, the train load effect is characterized by dynamic parameters, and the specific parameters are dynamic acceleration, dynamic stress, and dynamic strain. The dynamic acceleration reflects the overall stiffness of the roadbed. The more obvious the roadbed acceleration response, the lower the temperature, the lower the unfrozen water content, and the greater the overall stiffness, and vice versa. The dynamic parameters have an important influence on the stability of train driving.
[0021] In the present invention, the dynamic information library includes a physical state sensing system and a mechanical state sensing system, wherein:
[0022] The physical state sensing system is used to characterize the following physical state indicators: electric field, moisture field, and temperature field data of the physical state of the roadbed;
[0023] The mechanical state sensing system is used to characterize the following mechanical state indicators: acceleration, dynamic strain, and dynamic stress data of the mechanical state of the roadbed.
[0024] In the present invention, the data transmission and analysis system includes three parts: a data integration wireless transmission module based on the Internet of Things (IOT) technology, a monitoring data preprocessing module, and a data intelligent analysis and processing model, wherein:
[0025] The data integrated wireless transmission module is used to monitor the data remote transmission network;
[0026] The monitoring data preprocessing module is used for abnormal data processing and data denoising;
[0027] The data intelligent analysis and processing model is used to clarify two aspects of data obtained by intelligent perception: 1. The intrinsic relationship between the physical field information of the cold-region roadbed; 2. The intrinsic relationship between the dynamic parameters of the cold-region roadbed.
[0028] In the present invention, the data intelligent analysis and processing model includes a roadbed physical field prediction model and a roadbed safety service multi-field mutual feedback experience model, wherein:
[0029] The roadbed physical field prediction model is obtained through field data, specifically refers to a model that can reflect the relationship between roadbed deformation, temperature, and unfrozen water content. The roadbed deformation is divided into two parts, one is frost heave deformation, and the other is thaw settlement deformation. The general rule for frost heave deformation is that the lower the unfrozen water content, the higher the temperature, the greater the frost heave; the general rule for thaw settlement deformation is that the higher the unfrozen water content, the higher the temperature, the greater the thaw settlement. The above rules are determined for the soil in the same area.
[0030] The multi-field mutual feedback empirical model for roadbed safe service is obtained through field data. Specifically, it refers to a model that can reflect the relationship between roadbed stiffness, deformation, temperature, and unfrozen water content under the action of train load. In this model, the train load effect is characterized by dynamic parameters, and the specific parameters are dynamic acceleration, dynamic stress, and dynamic strain. The dynamic acceleration reflects the overall stiffness of the roadbed. The more obvious the roadbed acceleration response, the lower the temperature, the lower the unfrozen water content, and the greater the overall stiffness, and vice versa. The dynamic parameters have an important influence on the stability of train driving.
[0031] In the present invention, the disaster forecasting and early warning system includes a long-term forecasting module, a medium-term forecasting module, a short-term forecasting module and an automatic alarm module, wherein:
[0032] The long-term forecast module is used to forecast the stable service status of the roadbed one year later.
[0033] The medium-term forecast module is used to forecast the stable service status of the roadbed six months later.
[0034] The short-term forecast module is used to forecast the stable service status of the roadbed one month later;
[0035] The automatic alarm module function is used to predict the stable service status of the roadbed one day later;
[0036] The prediction accuracy of the long-term prediction module, the medium-term prediction module, the short-term prediction module and the automatic alarm module are 60%, 70%, 90% and 95% respectively.
[0037] Compared with the prior art, the present invention has the following advantages:
[0038] (1) The present invention uses a roadbed physical and mechanical sensing system and electrical resistivity tomography technology to achieve a more comprehensive and efficient monitoring of the physical and mechanical state of the roadbed;
[0039] (2) The data transmission and analysis system of the present invention is composed of a multi-dimensional heterogeneous network, which integrates and encodes the data of the sensor network, improves the efficiency of data transmission, and realizes the collaborative cooperation of multiple types of sensors;
[0040] (3) The present invention uses the cold region roadbed safety service and disaster forecasting and warning platform to identify the tiny variables of the roadbed physical and mechanical state, build a prediction model based on big data learning, and make forecasts and warnings of roadbed disasters in a timely manner;
[0041] (4) The various technologies proposed in the present invention can quickly obtain the roadbed status over a large range (within the range of 0.5km to 5km), and can realize the forecast and early warning of roadbed disasters over a large range. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 This is a framework diagram for intelligent perception and disaster forecasting and warning;
[0043] Figure 2 Build a framework for roadbed disaster information database;
[0044] Figure 3 Layout diagram of measuring points for mechanical state perception;
[0045] Figure 4 It is a system framework for data transmission and analysis;
[0046] Figure 5 Provide ideas for obtaining physical field data under full convolutional networks;
[0047] Figure 6 Flow chart of parameter model for BP neural network training dynamics;
[0048] Figure 7 It is the framework of roadbed disaster forecasting and early warning system. DETAILED DESCRIPTION
[0049] The technical solution of the present invention is further described below in conjunction with the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.
[0050] The present invention provides an intelligent sensing and forecasting system for the physical and mechanical state of roadbed in high-cold areas. Figure 1 As shown in the figure, the overall structure is mainly composed of three parts: cold region rail subgrade disaster information database, data transmission and analysis system, and disaster forecasting and early warning system, and the three parts are coordinated with each other.
[0051] like Figure 2 As shown in Figure 1, the roadbed disaster information database consists of a static information database and a dynamic information database, where:
[0052] The static information database is the basic data information of the roadbed construction service period, including the following information databases: vehicle condition information database, geographic information database, engineering geological information database, historical disease information database, indoor and outdoor test information database during the operation period;
[0053] The dynamic information library includes a physical state sensing system and a mechanical state sensing system. The physical state sensing system is used to characterize the following physical state indicators: electric field, moisture field, and temperature field data of the physical state of the roadbed; the mechanical state sensing system is used to characterize the following mechanical state indicators: acceleration, dynamic strain, and dynamic stress data of the mechanical state of the roadbed;
[0054] The basic information in the static information database provides a basis for establishing the physical field model in the dynamic information database, and the physical and mechanical state indicators are used to predict the deformation characteristics of the roadbed during its operation and service.
[0055] In the present invention, the vehicle condition information database in the static information database is established and maintained by the railway management department, and the data in the vehicle condition information database includes vehicle condition information such as the running train model, speed, time, etc.; the geographic information database, engineering geological information database, and historical disease information database are established and maintained by the survey unit of the line construction, and the geographic information database and engineering geological information include the hydrogeological overview, climate distribution characteristics, rainfall, and precipitation data along the roadbed project, and the historical disease information database includes data such as regular disease detection results; the indoor and outdoor test information database is updated and maintained by the scientific research unit, and the indoor and outdoor test information database includes a multi-field mutual feedback experience model related to the physical state of the roadbed, and a dynamic characteristic evolution law model related to the mechanical state of the roadbed.
[0056] In the present invention, the construction method of the roadbed static information database is as follows: collect and organize the train condition information (train model, speed, time) during the train operation service period, obtain geographic information and engineering geological information from the survey unit of the line construction, and the historical disease information during the line operation should also be stored in an exclusive database. The hydrogeological overview, climate distribution characteristics, rainfall, precipitation data, regular disease detection results and other data along the roadbed project are all part of the content established by the static database.
[0057] In the present invention, the indoor and outdoor test information database refers to a relationship model of relevant indicators established by means of indoor and outdoor tests.
[0058] In the present invention, a relationship model of mutual transformation between the subgrade electric field, moisture field and temperature field is obtained by analyzing the field data. The relationship model of mutual transformation between the subgrade electric field, moisture field and temperature field is shown in formula (1):
[0059]
[0060] Where m, n, a, b, c, and d are all empirical parameters that can be calibrated for different soils. ρ is the resistivity of frozen soil, and θ uis the unfrozen water content, and T is the freezing temperature. The establishment method is as follows: first, take the original soil sample from the subgrade site to carry out the soil freezing characteristic curve test, that is, test the unfrozen water content of the soil at different freezing temperatures, that is, (a) in formula (1). The test equipment adopts the currently more mature low-field nuclear magnetic resonance technology (NMR). Then, based on the laboratory test equipment for soil resistivity, including constant voltage power supply and ammeter, the resistivity value of the soil at different freezing temperatures is tested, that is, (b) in formula (1). Finally, with the same freezing temperature as a bridge, a laboratory model of resistivity and unfrozen water content is established, that is, (c) in formula (1).
[0061] In the present invention, the physical state indicators in the dynamic information library mainly include the electric field information of the roadbed site tested based on the high-density electrical instrument, and the temperature field information and moisture field information of the roadbed tested based on the wireless Bluetooth temperature sensor and the wireless Bluetooth moisture sensor. The mechanical state indicators are realized by measuring the roadbed dynamic parameters (acceleration, dynamic strain, dynamic stress) on site.
[0062] In the present invention, the specific method for constructing the physical state indicator is as follows:
[0063] (1) Electric field
[0064] A high-density electrical instrument is used to test the change in site resistivity during the operation of the roadbed, and then the resistivity value of the roadbed site tested by the high-density electrical instrument is inverted to obtain a resistivity cloud map. The test system of the high-density electrical instrument selects a multi-channel distributed measurement method, which can conduct regular monitoring of areas where roadbed diseases frequently occur in cold regions. Multiple electrodes can be buried during the initial layout (one electrode is buried every 1m within the retest range). When collecting monitoring data, the electrode shielding and electrode deletion functions are used. It is not necessary to connect all electrodes to the line, and the electrodes at the corresponding positions can be enabled as needed. The location, measurement frequency and measurement time period of electrode monitoring are set based on engineering experience and survey data, and the sensing system monitors for one or more cold and warm season cycles.
[0065] (2) Moisture field and temperature field
[0066] For the acquisition of moisture field and temperature field data, wireless Bluetooth temperature sensor and wireless Bluetooth moisture sensor are used to collect temperature field and moisture field data and compare and verify with electric field information.
[0067] The specific verification process is as follows: the resistivity data obtained from the high-density electrical instrument electrode test is first applied to the laboratory model of formula 1(c) to obtain the unfrozen water content data or temperature field data. This data is then used to compare with the data collected by the temperature and moisture sensors deployed on site. The two refer to each other. If there is a deviation between the two sets of data, the temperature field shall be based on the on-site temperature sensor data, and the moisture field shall be based on the resistivity data and the moisture field data calculated in formula 1(c).
[0068] (3) Deformation field
[0069] The test of roadbed deformation field data will be carried out based on synthetic aperture radar (InSRA) technology to monitor the surface deformation along the roadbed, obtain the law of roadbed deformation over time, and use roadbed deformation as the final indicator of disease evaluation to describe the frost heave and thaw settlement phenomenon of the roadbed. A prediction model for roadbed frost heave and thaw settlement deformation will be established based on the roadbed physical state data (electric field, temperature field, moisture field).
[0070] In the present invention, the method for obtaining the mechanical state index is as follows: a number of test sections are set up, and each test point is mainly equipped with a vertical dynamic response test element, including a dynamic earth pressure gauge and a dynamic strain gauge, an acceleration pickup to test the vertical acceleration, a vibration pickup to test the vertical vibration displacement, and a collector to collect the vibration signal; the dynamic stress response is monitored and data is collected using a strain type earth pressure box, see Figure 3 .
[0071] In the present invention, the data transmission and analysis system includes three parts: a data integration wireless transmission module based on the Internet of Things (IOT) technology, a monitoring data preprocessing module, and a data intelligent analysis and processing model. Figure 4 ,in:
[0072] (1) Data integration wireless transmission module
[0073] The data integration wireless transmission module is used to monitor the remote data transmission network. The integration and transmission of the collected data are coordinated by short-distance wireless Bluetooth transmission and long-distance packet radio service technology (GPRS). The data is wirelessly transmitted in the form of packets. When the GPRS signal is poor during on-site testing, the spare Beidou-1 satellite is selected to fill the network to ensure that the measured data can be transmitted to the monitoring and early warning center in real time.
[0074] (2) Monitoring data preprocessing module
[0075] The monitoring data preprocessing module includes two preprocessing methods for optimizing data, namely, abnormal data processing and data denoising, wherein: the abnormal data processing includes two methods, namely, trend method and manual elimination. The trend method refers to eliminating abnormal data outside the trend according to the overall law of existing data, and retaining data near the trend for subsequent analysis. The manual elimination refers to removing abnormal data appearing in the data according to the previous research experience of data processing personnel; the data denoising is aimed at on-site monitoring dynamic acceleration data, and uses Kalman filtering and empirical mode decomposition method to perform denoising analysis on the data. The Kalman filtering method establishes a linear system state equation by compiling a computer program, and optimally estimates the data, thereby filtering out the noise components in the data. The empirical mode decomposition method decomposes the acceleration time history data obtained from on-site monitoring into several intrinsic mode functions by compiling a calculation program, and converts the decomposed intrinsic mode function curve into a power spectral density (PSD) curve, wherein the last intrinsic mode function is a residual term.
[0076] (3) Data intelligent analysis and processing model
[0077] The data intelligent analysis and processing model includes image data analysis based on convolutional neural network (CNN) technology to obtain resistivity cloud maps, unfrozen water content cloud maps, and temperature field cloud maps corresponding to the physical field, as well as a roadbed physical field prediction model constructed based on BP neural network technology.
[0078] a. Roadbed physical field data analysis method:
[0079] The construction method of the roadbed physical field prediction model is based on the electric field, moisture field, temperature field, deformation field in the dynamic information library of the roadbed disaster information library and the indoor test model information in the static information library, combined with the full convolution (FCN) neural network algorithm. The electric field, moisture field, temperature field, and deformation field data information in the dynamic information library are all presented in the form of images.
[0080] The present invention innovatively proposes a multi-physical field information extraction technology for image data. Specifically, first, the resistivity cloud map information is input into the convolutional neural network, and then the full convolutional network algorithm is implemented with the help of the TensorFlow program terminal. By reading the pixel information in the image, the abnormal high resistance value or extremely low resistance value is reasonably eliminated, and the specific electric field information in the resistivity cloud map obtained by the resistivity tomography technology is quickly obtained. Then, a 1*1 convolution kernel is designed, and the indoor test model (Formula 1) based on the static information library in the roadbed disaster information library is written into a matrix form related to the formation position. The convolutional network will automatically learn and output a cloud map that can characterize the size of the unfrozen water content of the site, which can be compared and verified with the moisture sensor monitoring data, and the data error is analyzed. In addition, another 1*1 convolution kernel is designed, and the temperature and unfrozen water content relationship model in the freezing characteristic curve of the roadbed site is written into a matrix form related to the formation position. The full convolutional network will automatically learn and output a cloud map that can characterize the ground temperature of the roadbed site, which can be compared and verified with the temperature sensor data, and the data error is analyzed. Figure 5 As shown in the figure, the resistivity cloud is extracted in the full convolutional network framework after designing a 1*1 convolution kernel, and then the unfrozen water content cloud map and temperature cloud map are converted with the help of the resistivity model and freezing characteristic curve.
[0081] b. Construction method of roadbed safety service prediction model:
[0082] In part a, a method for obtaining the moisture field and temperature field data of the roadbed from the roadbed resistivity cloud map was invented. Based on the acquired resistivity cloud map, unfrozen water content cloud map, and temperature field cloud map data, combined with the roadbed field displacement cloud map obtained by InSAR testing, and the dynamic parameter monitoring results, the dynamic parameters mainly include acceleration, dynamic strain, and dynamic stress, and the BP neural network algorithm is used to establish a roadbed safety service prediction model.
[0083] The present invention creatively proposes a method for constructing a multivariate data source correlation analysis model. The processing flow is as follows: Figure 6 As shown, LSTM is an improved model for processing recurrent deep neural networks (RNNs) with time dimensions and similar time dimension information, which can analyze the changing trends of the electric field, moisture field, temperature field, and deformation field data of the roadbed site on a time scale. The specific steps of implementation are as follows:
[0084] First, write the BP neural network algorithm implementation program, and then Figure 5The roadbed site resistivity data, unfrozen water content data, temperature field data obtained by full convolutional neural network processing, roadbed deformation data obtained by InSAR testing, and acceleration, dynamic strain, and dynamic stress data obtained by dynamic parameter monitoring are divided into two parts: training data and verification data. Then data training is carried out. The specific steps of network training are as follows:
[0085] The input and output relationship between neurons in the hidden layer under the BP network model framework can be expressed by formulas (2) and (3):
[0086]
[0087] O f =f(v j )=1 / [1-exp(-v j )] (3);
[0088] In the formula, v j and O j Represents the input and output of each neuron in the hidden layer, x i represents the input of the i-th node in the input layer, w ij represents the connection weight between this node and the jth node in the hidden layer, θ j represents the threshold of the jth node in the hidden layer, f uses the Sigmoid function, m represents the number of nodes in the input layer, and i represents the node sequence number of the input layer.
[0089] Then, the input and output of each neuron in the output layer of the BP neural network model are calculated by formula (4) and formula (5):
[0090]
[0091] y k =f(s k )=1 / [1-exp(-s k )] (5);
[0092] In the formula, s k represents the input of each neuron in the output layer, w jk represents the connection weight between the jth node in the input layer and the kth node in the hidden layer, o j represents the output of each neuron in the hidden layer, θ j represents the threshold of the jth node in the hidden layer, η k represents the threshold of the kth node in the output layer, y k represents the output of each neuron in the output layer, n represents the number of nodes in the hidden layer, and j represents the node sequence number of the hidden layer.
[0093] Then, the weight error on the input layer unit connected to the BP neural network model is calculated by formula (6):
[0094] δ k =(1+y k )(d k -y k )y k (6);
[0095] In the formula, δ k represents the weight error connected to the input layer unit, y k represents the output of each neuron in the output layer, d k represents the sample expected value.
[0096] Then, the weight error on the hidden layer unit connected to the BP neural network model is calculated by formula (7):
[0097]
[0098] In the formula, σ j represents the weight error connected to the hidden layer unit, o j represents the output of each neuron in the hidden layer, δ k represents the weight error connected to the input layer unit, w jk represents the connection weight between the jth node in the hidden layer and the kth node in the output layer, q represents the number of nodes in the output layer, and k represents the node sequence number of the output layer.
[0099] Update the current connection weight and threshold through formula (8) and formula (9):
[0100] w jk (N+1)=w jk (N)+αδ k o j (8);
[0101] η k (N+1)=η k (N)-βδ k (9);
[0102] In the formula, w jk (N+1) represents the updated connection weight between the jth node in the hidden layer and the kth node in the output layer, w jk (N) represents the connection weight between the hidden j-th node and the output layer k-th node before the update, α represents the weight adjustment parameter, δ k represents the weight error connected to the input layer unit, o j Represents the output of each neuron in the hidden layer; η k (N+1) represents the threshold of the kth node in the output layer after update, ηk (N) represents the threshold of the kth node in the output layer before updating, and β represents the threshold adjustment parameter.
[0103] Update the next connection weight and threshold using formula (10) and formula (11):
[0104] w ij (N+1)=w ij (N)+ασ j x i (10);
[0105] θ j (N+1)=θ j (N)-βσ j (11);
[0106] In the formula, w ij (N+1) represents the connection weight between the i-th node in the input layer and the j-th node in the hidden layer after the update, w ij (N) represents the connection weight between the i-th node in the input layer and the j-th node in the hidden layer before the update, α represents the weight adjustment parameter, and σ j represents the weight error connected to the hidden layer unit, x i represents the input of the i-th node in the input layer; θ j (N+1) represents the threshold of the jth node in the hidden layer after the update, θ j (N) represents the threshold of the j-th node in the hidden layer before updating, and β represents the threshold adjustment parameter.
[0107] When all samples in the training set are trained, the mean square error is calculated. If the mean square error meets the output condition, new roadbed frost heave prediction and thaw settlement prediction values are output.
[0108] The mean square error is calculated by formula (12):
[0109]
[0110] In the formula, E t represents the mean square error, C k Represents the expected output value, y k represents the output of each neuron in the output layer, and q represents the number of nodes in the output layer.
[0111] When the mean square error satisfies Then the mean square error satisfies the output condition, where ε represents the learning accuracy, t represents the training samples in the training set, and z represents the total number of training samples.
[0112] In the present invention, the disaster forecast and warning system includes a long-term forecast module, a medium-term forecast module, a short-term forecast module and an automatic alarm module. The long-term forecast module, the medium-term forecast module, the short-term forecast module and the automatic alarm module serve as the data processing and intelligent analysis center of the system of the present invention, and coordinate the data encoding, integration and decoding process of the perception system and the sensor network, wherein: the long-term forecast module, the medium-term forecast module, the short-term forecast module and the automatic alarm module respectively discriminate the data by corresponding theoretical analysis methods. The long-term forecast module, the medium-term forecast module, the short-term forecast module and the automatic alarm module respectively indicate that the service status of the roadbed will face challenges in a certain period of time in the future (different times), that is, the safe operation of the train is challenged. The disaster forecast and warning system sequentially designs the main program, subprogram, and subprocess, and then defines and debugs the input and output link relationship between each module, and performs specific module design with function as the guide.
[0113] The above-mentioned data intelligent analysis and processing model proposed by the present invention can output specific results of indicators related to the service status of the roadbed, that is, the deformation of the roadbed within a certain period of time. Four different warning levels are designed respectively. Specifically, the track unevenness is first predicted according to the deformation of the roadbed within a certain period of time, and then the driving safety under the track unevenness is evaluated. If the model prediction results show that the safe operation of the roadbed will face challenges in the next year, it means that the long-term forecast index threshold of the system has been reached. If the model prediction results show that the safe operation of the roadbed will face challenges in the next six months, it means that the medium-term forecast index threshold of the system has been reached. If the model prediction results show that the safe operation of the roadbed will face challenges in the next month, it means that the short-term forecast index threshold of the system has been reached. If the model prediction results show that the safe operation of the roadbed will face challenges in the next day, it means that the automatic alarm index threshold of the system has been reached, and the test should immediately notify the railway department to stop the operation of the line. The above-mentioned disaster forecast and early warning system is judged and fed back according to the intelligent analysis and processing model of monitoring data. Furthermore, the data integration wireless transmission module in the data transmission and analysis system performs data fusion on the collected data, realizes the coordinated work of various sensors, and then transmits the data to the GPRS terminal remotely through the data transmission and analysis system, and then transmits it to the disaster forecasting and early warning system through the public network. Figure 7 .
Claims
1. An intelligent perception and prediction system for the physical and mechanical state of track subgrade in high-cold areas, characterized by The system includes three parts: track subgrade disaster information database, data transmission and analysis system, and disaster forecasting and early warning system, among which: The track subgrade disaster information database is composed of a static information database and a dynamic information database; The static information database is used to provide basic data information during the service life of the track subgrade construction; the static information database includes a vehicle condition information database, a geographic information database, an engineering geological information database, a historical disease information database, and an indoor and outdoor test information database, wherein: the vehicle condition information database is used to provide the model, speed, and time of the running train; the geographic information database and the engineering geological information database are used to provide the hydrogeological overview, climate distribution characteristics, rainfall, and precipitation data along the track subgrade project; the historical disease information database is used to provide regular disease detection result data; the indoor and outdoor test information database is used to provide a multi-field mutual feedback empirical model related to the physical state of the track subgrade and a dynamic characteristic evolution law model related to the mechanical state of the subgrade; The dynamic information library is used to provide physical state indicators and mechanical state indicators; the dynamic information library includes a physical state sensing system and a mechanical state sensing system, wherein: the physical state sensing system is used to characterize the following physical state indicators: electric field, moisture field, temperature field, and deformation field data of the physical state of the track subgrade; The relationship model of the mutual transformation of the electric field, moisture field and temperature field of the physical state of the track subgrade is shown in formula (1): (1); In the formula, , , , , , These are all empirical parameters, calibrated for different soil types. is the resistivity of frozen soil, is the unfrozen water content, is the freezing temperature; The mechanical state sensing system is used to characterize the following mechanical state indicators: acceleration, dynamic strain, and dynamic stress data of the mechanical state of the track subgrade; The data transmission and analysis system is used for short-range integration and long-range wireless transmission of sensor data, and predicts the stability service index of the subgrade based on the physical and mechanical perception data of the track subgrade; the data transmission and analysis system includes three parts: a data integration wireless transmission module, a monitoring data preprocessing module, and a data intelligent analysis and processing model, among which: the data integration wireless transmission module is used for monitoring the data remote transmission network; the monitoring data preprocessing module is used for abnormal data processing and data denoising; the data intelligent analysis and processing model is used to clarify the intrinsic correlation between the physical field information of the cold-region track subgrade and the dynamic parameters of the cold-region track subgrade The data intelligent analysis and processing model includes the resistivity cloud map, unfrozen water content cloud map, temperature field cloud map corresponding to the physical field obtained by image data analysis based on convolutional neural network technology, and the track subgrade physical field prediction model constructed based on BP neural network technology; the construction method of the track subgrade physical field prediction model is based on the electric field, moisture field, temperature field, deformation field in the dynamic information library of the track subgrade disaster information library and the indoor test model information in the static information library, combined with the full convolutional neural network algorithm for construction, and the electric field, moisture field, temperature field, deformation field data information in the dynamic information library are all presented in the form of images; The disaster forecasting and early warning system is used to forecast the service status of the track subgrade at different times in the future; the track subgrade disaster information database and the data transmission and analysis system provide the disaster forecasting and early warning system with forecasting and early warning indicators and thresholds of various indicators; the track subgrade disaster information database provides a basis for the establishment of a prediction model in the data transmission and analysis system; the disaster forecasting and early warning system determines the prediction model result, that is, whether it reaches the forecast or early warning value compared with the threshold; The disaster forecast and warning system includes a long-term forecast module, a medium-term forecast module, a short-term forecast module and an automatic alarm module, wherein: the long-term forecast module is used to forecast the stable service status of the roadbed one year later; the medium-term forecast module is used to forecast the stable service status of the roadbed six months later; the short-term forecast module is used to forecast the stable service status of the roadbed one month later; and the automatic alarm module function is used to forecast the stable service status of the roadbed one day later.
