Railway risk disaster three-dimensional multi-stage monitoring system

Abstract

The invention discloses a three-dimensional multi-stage monitoring system for railway risk disasters. Comprising a third monitoring unit, a second monitoring unit, a first monitoring unit, a gateway unit and a cloud computing platform. The third monitoring unit, the second monitoring unit and the first monitoring unit respectively obtain third monitoring data, second monitoring data and first monitoring data, and the third monitoring area, the second monitoring area and the first monitoring area are gradually far away from the rail surface of the railway; the third monitoring data, the second monitoring data and the first monitoring data are transmitted to a cloud computing platform through a gateway unit, a data fusion model is preset in the cloud computing platform, the data fusion model fuses the third monitoring data, the second monitoring data and / or the first monitoring data to obtain fused data, and the cloud computing platform sends the fused data to the cloud computing platform based on the fused data. And early warning signals of different grades are given. According to the method, the risk level of the abnormity can be accurately judged according to the position of the abnormity, and a proper coping strategy is formulated according to the risk level.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk disaster monitoring, and particularly relates to a railway risk disaster three-dimensional multi-level monitoring system. BACKGROUND

[0002] Railway transportation is the most important transportation mode in China. The railway has the characteristics of fast speed, long distance, complex geological and topographical conditions, high bridge-tunnel ratio, and the safe operation of the railway faces various risks caused by geological disasters, structural health, external environment and human factors.

[0003] The railway foreign matter invasion risk has the characteristics of various scenes, various types and great monitoring difficulty. The existing foreign matter invasion monitoring methods can be divided into contact type and non-contact type. The contact type monitoring method mainly installs a protective net in the monitoring section and installs a sensing device on the protective net to judge whether foreign matter falls on the protective net. The non-contact type monitoring method mainly uses infrared, microwave, radar or video to detect the size and position of the foreign matter, especially to scan the track position to monitor whether foreign matter invades. However, the existing monitoring system has a single function, cannot realize hierarchical alarm, and cannot accurately judge the railway foreign matter invasion risk. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a railway risk disaster three-dimensional multi-level monitoring system, which solves the problem that the existing monitoring system has a single function, cannot realize hierarchical alarm, and cannot accurately judge the railway foreign matter invasion risk.

[0005] To solve the above technical problem, one technical solution of the present application is to provide a railway risk disaster three-dimensional multi-level monitoring system, which comprises a third monitoring unit, a second monitoring unit, a first monitoring unit, a gateway unit and a cloud computing platform. The third monitoring unit, the second monitoring unit and the first monitoring unit are respectively arranged in a third monitoring area, a second monitoring area and a first monitoring area of a railway, and acquire third monitoring data, second monitoring data and first monitoring data, respectively. The third monitoring area, the second monitoring area and the first monitoring area are gradually away from the track surface of the railway. The third monitoring data, the second monitoring data and the first monitoring data are transmitted to the cloud computing platform through the gateway unit. A data fusion model is pre-set in the cloud computing platform. The data fusion model fuses the third monitoring data, the second monitoring data and / or the first monitoring data to obtain fusion data. The cloud computing platform gives different levels of early warning signals based on the fusion data.

[0006] The beneficial effects of the present application are: in the present application, the third monitoring area, the second monitoring area and the first monitoring area on the railway are monitored by the third monitoring unit, the second monitoring unit and the first monitoring unit respectively. When the railway has an anomaly, it will pass through the third monitoring area, the second monitoring area and the first monitoring area in space, and trigger the third monitoring unit, the second monitoring unit or the first monitoring unit in time. Thus, the hierarchical alarm of the anomaly can be realized. When the anomaly is located in the third monitoring area, the third monitoring unit is triggered, and the third monitoring unit sends a third-level alarm signal. At this time, the anomaly is located outside the guardrail, the risk is low, and whether there is an anomaly can be confirmed within a preset third time. When the anomaly passes through the third monitoring area and enters the second monitoring area, the second monitoring unit is triggered, and the second monitoring unit sends a second-level alarm signal. At this time, the anomaly is located inside the guardrail, the risk is high, and whether there is an anomaly can be confirmed within a preset second time. When the anomaly passes through the second monitoring area and enters the first monitoring area, the first monitoring unit is triggered, and the first monitoring unit sends a first-level alarm signal. At this time, the anomaly is located on the track surface, the risk is extremely high, and whether there is an anomaly can be confirmed within a preset first time. Thus, the risk level of the anomaly can be accurately judged according to the position of the anomaly, and appropriate response strategies can be developed according to the risk level. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is a schematic diagram of an application scenario of the multi-level monitoring system according to the present application;

[0008] Figure 2 is a schematic diagram of another application scenario of the multi-level monitoring system according to the present application;

[0009] Figure 3 is a schematic diagram of the layout of the monitoring optical fiber of the multi-level monitoring system according to the present application;

[0010] Figure 4 is another schematic diagram of the layout of the monitoring optical fiber of the multi-level monitoring system according to the present application;

[0011] Figure 5 is still another schematic diagram of the layout of the monitoring optical fiber of the multi-level monitoring system according to the present application;

[0012] Figure 6 is still another schematic diagram of the layout of the monitoring optical fiber of the multi-level monitoring system according to the present application;

[0013] Figure 7 is a flowchart of the response to the early warning signal of the multi-level monitoring system according to the present application;

[0014] Figure 8 is a flowchart of the response to the third-level alarm signal of the multi-level monitoring system according to the present application;

[0015] Figure 9is a flow chart of the response of the multi-level monitoring system according to the present application to a secondary alarm signal;

[0016] Figure 10 is a flow chart of the response of the multi-level monitoring system according to the present application to a primary alarm signal;

[0017] Figure 11 is a schematic diagram of data level fusion of the multi-level monitoring system according to the present application;

[0018] Figure 12 is a schematic diagram of feature level fusion of the multi-level monitoring system according to the present application;

[0019] Figure 13 is a schematic diagram of decision level fusion of the multi-level monitoring system according to the present application;

[0020] Reference signs: 10, perimeter fence net, 20, monitoring optical fiber, 201, optical fiber composite sensor, 30, sensing device, 40, protective net. DETAILED DESCRIPTION

[0021] In order to facilitate the understanding of the present application, the present application will be described in more detail below in conjunction with the drawings and specific embodiments. The preferred embodiments of the present application are shown in the drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described in the specification. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0022] It should be noted that unless otherwise defined, all technical and scientific terms used in the specification are the same as those commonly understood by those skilled in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application. The term "and / or" used in the specification includes any and all combinations of one or more related listed items.

[0023] Figures 1-2 As shown, an embodiment of the railway risk disaster stereoscopic multi-level monitoring system of the present application is shown, which includes: taking the area outside the railway perimeter fence net 10 as a third monitoring area A3, taking the area between the railway track surface and the perimeter fence net 10 as a second monitoring area A2, and taking the track surface of the railway as a first monitoring area A1;

[0024] The third monitoring unit, the second monitoring unit and the first monitoring unit are arranged to monitor whether there is an anomaly in the third monitoring area A3, the second monitoring area A2 and the first monitoring area A1 respectively, and if the third monitoring unit, the second monitoring unit or the first monitoring unit detects an anomaly in the third monitoring area A3, the second monitoring area A2 or the first monitoring area A1, a third-level alarm signal, a second-level alarm signal or a first-level alarm signal is sent respectively, and in response to the third-level alarm signal, the second-level alarm signal or the first-level alarm signal, the anomaly in the third monitoring area A3, the second monitoring area A2 and the first monitoring area A1 is re-confirmed within a preset third time, a preset second time and a preset first time respectively.

[0025] In the present application, the third monitoring area A3, the second monitoring area A2 and the first monitoring area A1 on the railway are monitored by the third monitoring unit, the second monitoring unit and the first monitoring unit respectively, and when there is an anomaly, the railway will pass through the third monitoring area A3, the second monitoring area A2 and the first monitoring area A1 in space and will trigger the third monitoring unit, the second monitoring unit or the first monitoring unit in time, thereby realizing the hierarchical alarm of the anomaly. When the anomaly is located in the third monitoring area A3, the third monitoring unit is triggered, and the third monitoring unit sends a third-level alarm signal. At this time, the anomaly is located outside the guardrail, and the risk is low. Whether there is an anomaly can be confirmed within a preset third time. When the anomaly passes through the third monitoring area A3 and enters the second monitoring area A2, the second monitoring unit is triggered, and the second monitoring unit sends a second-level alarm signal. At this time, the anomaly is located inside the guardrail, and the risk is high. Whether there is an anomaly can be confirmed within a preset second time. When the anomaly passes through the second monitoring area A2 and enters the first monitoring area A1, the first monitoring unit is triggered, and the first monitoring unit sends a first-level alarm signal. At this time, the anomaly is located on the track surface, and the risk is extremely high. Whether there is an anomaly can be confirmed within a preset first time. Thus, the risk level of the anomaly can be accurately judged according to the position of the anomaly, and appropriate countermeasures can be developed according to the risk level.

[0026] The method can be used for railways and highways. Taking the railway as an example, the foreign object intrusion monitoring scene along the railway mainly includes tunnel entrances, slopes and overpass railway bridges. These sections are prone to accidents such as rockfall, sliding, debris flow, projectile and falling objects that endanger train safety.

[0027] The abnormal risk in the railway is caused by the complex environment-structure-operation interaction along the railway. Including foreign invasion, geological disasters, structural failure, equipment failure, fire and flood, etc. These disasters are not isolated, and their coupling effect forms a risk transmission chain through vulnerable nodes. The occurrence of various risks, the cause mechanism and evolution law show strong coupling correlation and multi-dimensional nonlinear interaction characteristics in the space-time dimension. For example, in the typical disaster scenario at the tunnel entrance, there is a coupling relationship from the spatial and temporal dimensions: the third monitoring unit in the third monitoring area A3 may be deformed due to rain-caused floods, landslides, or malicious damage, which is an earlier risk and does not affect driving; The second monitoring area A2, i.e. the inside of the perimeter fence net 10, may have risks such as personnel crossing, terrorist damage, and large animals entering, from approaching the fence to crossing and destroying, both spatially and temporally, there is a time-space evolution process, and the risk is further increased; When the anomaly breaks through the perimeter fence net 10 to the slope area, it may be deformed or landslided due to geological disasters, and finally intrudes into the first monitoring area A1, i.e. the track surface of the railway, directly affecting the safety of driving.

[0028] In some embodiments, the third monitoring area A3 can be the slope outside the railway guardrail, or the area on the top of the tunnel entrance prone to rockfall, the second monitoring area A2 is the slope between the track surface and the guardrail, and the first monitoring area A1 is the track surface of the railway. The third monitoring unit is arranged in the third monitoring area A3 to determine whether the third monitoring area A3 has an anomaly, and the third monitoring unit can be a monitoring optical fiber 20 to determine whether the third monitoring area A3 deforms or has a falling object by monitoring the change of the monitoring optical fiber 20. Mainly prevent the risk of dangerous stones and rockfalls.

[0029] The second monitoring unit is arranged in the second monitoring area A2 to determine whether the second monitoring area A2 has an anomaly, and the second monitoring unit can be a monitoring optical fiber 20 and / or a sensing device 30 to determine whether the second monitoring area A2 deforms or has a falling object by monitoring the change of the monitoring optical fiber 20, or to determine whether the second monitoring area A2 has an anomaly by the image obtained by the sensing device 30. Mainly prevent the risk of slope sliding.

[0030] The first monitoring unit is arranged in the first monitoring area A1 to determine whether the first monitoring area A1 has an anomaly, and the first monitoring unit can be a sensing device 30 to determine whether there is a falling object or other abnormal conditions in the first monitoring area A1 by the image obtained by the sensing device 30.

[0031] In some embodiments, the monitoring optical fiber 20 includes a stress monitoring optical fiber 20 and an acoustic monitoring optical fiber 20, etc., and whether there is an intruder in the third monitoring area A3 or the second monitoring area A2 is monitored according to the change of the monitoring optical fiber 20. The monitoring optical fiber 20 can utilize the redundant core of the existing communication optical cable along the railway as the main sensing unit, and can monitor the vibration, temperature and strain along the line in real time while constructing the transmission channel along the line, thereby realizing one-core dual-use of the existing communication optical cable of the railway.

[0032] In some embodiments, as shown in Figure 3 and Figure 4 , the monitoring optical fiber 20 includes a plurality of optical fiber composite sensors 201. In some embodiments, the optical fiber composite sensors 201 are arranged transversely along the railway slope anchor cable, and each anchor cable is provided with a set of optical fiber composite sensors 201.

[0033] In some embodiments, as shown in Figure 4 and Figure 5 , when the slope protection mode of the third monitoring area A3 is the protective net 40, the optical fiber composite sensors 201 are transversely laid along the protective net 40 reinforced anchor cable; the number of lanes is determined according to the number of active net reinforced anchor cable lanes.

[0034] In some embodiments, a plurality of optical fiber composite sensors 201 are installed along the cutting protective net 40 or the tunnel entrance, the first optical fiber composite sensor 201 is 0.5 meters away from the ground, and the distance between adjacent optical fiber composite sensors 201 is 0.5 meters.

[0035] In some embodiments, as shown in Figure 6 , the third monitoring unit can also be arranged on the perimeter fence net 10, and when arranged on the perimeter fence net 10, the optical fiber composite sensors 201 are transversely and uniformly laid in two lanes along the perimeter fence net 10, the first optical fiber composite sensor 201 is installed at the starting position of 10 m of the perimeter fence net 10, and then one is installed on the support column of the perimeter fence net 10 every 10 meters. The optical fiber composite sensor 201 is installed on the support column of the perimeter fence net 10 by using the force rod fixing plate.

[0036] In some embodiments, the second monitoring unit can include the above-mentioned monitoring optical fiber 20, and can also include a sensing device 30, which includes a monitoring camera, a monitoring radar or a combination of a monitoring camera and a monitoring radar. The monitoring camera and / or the monitoring radar can monitor whether the second monitoring area A2 is abnormal through video images and / or radar images.

[0037] In some embodiments, the monitoring camera and / or monitoring radar are installed on a stand, the monitoring camera and / or monitoring radar can monitor a range of 200 meters, the monitoring camera and / or monitoring radar are located not less than 3 meters away from the perimeter fence net 10, and the number and interval of the monitoring camera and / or monitoring radar are set according to the monitoring range covering the entire second monitoring area A2.

[0038] In some embodiments, the first monitoring unit includes the monitoring camera and / or monitoring radar described above, the monitoring camera and / or monitoring radar are installed on the roadbed section, the stand of the monitoring camera and / or monitoring radar is located 4.5 meters away from the center of the line, and the monitoring camera and / or monitoring radar are located 2 meters above the railway plane. The height and installation position of the stand do not intrude into the railway construction clearance, and each set of monitoring camera and / or monitoring radar covers a range of 60 meters.

[0039] In some embodiments, as shown in Figure 1 The fourth monitoring area A4 can be a 50-meter range of the railway, and a region that needs to be focused on due to personnel activities and the like. In the fourth monitoring area A4, a fourth monitoring unit is arranged, the fourth monitoring unit is used to monitor whether the fourth monitoring area A4 has an anomaly, and the fourth monitoring unit is the perception device 30. If the fourth monitoring unit detects that the fourth monitoring area A4 has an anomaly, a warning signal is sent, and within a preset warning time, the anomaly in the fourth monitoring area A4 is confirmed based on the warning signal.

[0040] In some embodiments, as shown in Figure 7 If the fourth monitoring unit detects that the fourth monitoring area A4 has an anomaly, a warning signal is sent, and in response to the warning signal, the fourth monitoring area A4 is remotely viewed through a perception device. The perception device can be the first monitoring unit monitoring the first monitoring area, the monitoring camera and / or monitoring radar arranged at the second monitoring area, the monitoring camera and / or monitoring radar arranged at the third monitoring area, or the monitoring camera and / or monitoring radar installed at other positions. If the perception device finds that the fourth monitoring area A4 has an anomaly, on-site confirmation is performed within a warning time, which can be 12 to 24 hours. The on-site evaluation of whether the anomaly in the fourth monitoring area A4 will affect the track can predict the risk in advance and develop a strategy to deal with the risk.

[0041] In some embodiments, as shown in Figure 8As shown, if the third monitoring unit monitors that the third monitoring area A3 has an abnormality, a third-level alarm signal is sent out, in response to the third-level alarm signal, the third monitoring area A3 is immediately remotely viewed and confirmed by a sensing device, which can be the first monitoring unit monitoring the first monitoring area, a monitoring camera and / or a monitoring radar arranged at the second monitoring area, a monitoring camera and / or a monitoring radar arranged at the third monitoring area, a monitoring camera and / or a monitoring radar installed at other positions, etc. If the sensing device finds that the third monitoring area A3 has an abnormality, on-site confirmation is immediately performed, after the staff arrives at the scene, according to the abnormal situation, a driving restriction condition is determined. If the sensing device does not find that the third monitoring area A3 has an abnormality, the staff can perform on-site confirmation within a third time, which can be 6 hours to 12 hours. It is determined whether the abnormality in the third monitoring area A3 will affect the rail in the on-site evaluation, so as to predict the risk in advance and develop a strategy to cope with the risk.

[0042] In some embodiments, as shown in FIG. 1, the first monitoring unit monitors the first monitoring area A1, the second monitoring unit monitors the second monitoring area A2, and the third monitoring unit monitors the third monitoring area A3. Figure 9 As shown, if the second monitoring unit monitors that the second monitoring area A2 has an abnormality, a second-level alarm signal is sent out, in response to the second-level alarm signal, the second monitoring area A2 is immediately remotely viewed and confirmed by a sensing device, which can be the first monitoring unit monitoring the first monitoring area, a monitoring camera and / or a monitoring radar arranged at the second monitoring area, a monitoring camera and / or a monitoring radar arranged at the third monitoring area, a monitoring camera and / or a monitoring radar installed at other positions, etc. If the sensing device finds that the second monitoring area A2 has an abnormality, on-site confirmation is immediately performed, after the staff arrives at the scene, according to the abnormal situation, a driving restriction condition is determined. The railway is blocked or the train is slowed down, and the abnormal situation is immediately handled. If the sensing device does not find that the second monitoring area A2 has an abnormality, the staff can perform on-site confirmation within a second time, which can be 2 hours to 6 hours. It is determined whether the abnormality in the second monitoring area A2 will affect the rail in the on-site evaluation, so as to predict the risk in advance and develop a strategy to cope with the risk.

[0043] In some embodiments, as shown in FIG. 1, the first monitoring unit monitors the first monitoring area A1, the second monitoring unit monitors the second monitoring area A2, and the third monitoring unit monitors the third monitoring area A3. Figure 10As shown, if the sensing device detects an anomaly in the first monitoring area A1, a first alarm signal is sent. The sensing device can be a first monitoring unit monitoring the first monitoring area, a monitoring camera and / or a monitoring radar arranged at the second monitoring area, a monitoring camera and / or a monitoring radar arranged at the third monitoring area, a monitoring camera and / or a monitoring radar installed at other positions, etc. In response to the first alarm signal, a staff member can perform on-site confirmation within a first time period, which can be 5 minutes to 1 hour. The staff member immediately goes to the site to confirm the abnormal situation. After confirming the abnormality, the staff member dispatches all trains on the railway to stop running and immediately handles the abnormality.

[0044] In some embodiments, a first interval time threshold is preset between the third alarm signal and the second alarm signal. If the interval time between the third alarm signal and the second alarm signal is less than the first interval time threshold, on-site confirmation of the abnormal situation is performed within the interval time. For example, the first interval time threshold is set to 1 hour to 2 hours, and the interval time between the third alarm signal and the second alarm signal is 0.5 hours. Since the interval time is less than the first interval time threshold, it is indicated that the abnormality may be moving from the third monitoring area A3 to the second monitoring area A2, and there is a risk that the abnormality may invade the first monitoring area A1 within 0.5 hours. Therefore, on-site confirmation of the abnormal situation is required within the interval time of 0.5 hours. Thus, the risk can be predicted in advance according to the spatio-temporal changes of the abnormality, and a strategy for coping with the risk can be developed.

[0045] The application also provides a railway risk disaster stereoscopic multi-level monitoring system, which comprises a third monitoring unit, a second monitoring unit, a first monitoring unit, a gateway unit and a cloud computing platform. The third monitoring unit, the second monitoring unit and the first monitoring unit are interconnected with the cloud computing platform through the gateway unit.

[0046] The third monitoring unit is arranged in a third monitoring area of the railway, is used for monitoring the third monitoring area to obtain third monitoring data, and sends the third monitoring data to the cloud computing platform through the gateway unit. The cloud computing platform analyzes the third monitoring data, and if the third monitoring data is found to be abnormal, a third-level early warning signal is sent. The second monitoring unit is arranged in a second monitoring area of the railway, is used for monitoring the second monitoring area to obtain second monitoring data, and sends the second monitoring data to the cloud computing platform through the gateway unit. The cloud computing platform analyzes the second monitoring data, and if the second monitoring data is found to be abnormal, a second-level early warning signal is sent. The first monitoring unit is arranged in a first monitoring area of the railway, is used for monitoring the first monitoring area to obtain first monitoring data, and sends the first monitoring data to the cloud computing platform through the gateway unit. The cloud computing platform analyzes the first monitoring data, and if the first monitoring data is found to be abnormal, a first-level early warning signal is sent.

[0047] In some embodiments, the cloud computing platform can also compare the interval time between the third-level warning signal and the second-level warning signal with the first interval time threshold value, predict the risk in advance, and develop a strategy to deal with the risk.

[0048] The first monitoring unit, the second monitoring unit, and the third monitoring unit can be a monitoring camera, a monitoring radar, and a monitoring optical fiber, respectively. The first monitoring unit and the second monitoring unit can also be a monitoring camera, a monitoring radar, or a combination thereof. The railway risk disaster stereoscopic multi-level monitoring system adopts a multi-source perception scheme. The monitoring optical fiber, the monitoring camera, and the monitoring radar have different use conditions and perception capabilities. For example, the monitoring optical fiber has highly sensitive perception capability and can accurately capture physical changes such as vibration, strain, and temperature. The monitoring camera has good visualization capability and can directly observe the monitoring object. By using multi-source fusion comprehensive identification, the performance of the monitoring system can be improved, and the shortcomings of single perception technology can be overcome.

[0049] The application can realize information complementation in different scenarios, improve the overall monitoring accuracy and adaptability, especially in extreme weather or harsh environment, and the use of multi-source information comprehensive analysis and judgment can significantly improve the system's perception and understanding ability of the environment and events, effectively reducing the occurrence rate of false positives and false negatives. In complex and variable scenarios, multi-source perception technology integrates multi-source perception data from different sources, such as optical fiber data as the third monitoring data, video data and radar data as the second monitoring data or the first monitoring data, etc., thereby forming a more comprehensive and accurate perception ability. These multi-source perception data can be verified and supplemented after synchronous collection, verification and fusion, thereby improving the reliability and integrity of the data. On this basis, a multi-dimensional and multi-level analysis method such as data mining, artificial intelligence algorithm, expert knowledge base, etc. is used to process and interpret complex multi-source perception data. Not only can potential trends and correlations be discovered, but also invalid data can be quickly filtered and key information can be accurately identified. The synergistic effect of multi-source perception data fusion and comprehensive analysis and judgment helps to reduce false positives caused by single data misleading, avoids missing key information that leads to false negatives, and improves the system's response speed and decision reliability.

[0050] In some embodiments, the raw data obtained by the third monitoring unit is directly transmitted to the cloud computing platform through the gateway unit, thereby ensuring the integrity of the data obtained by the third monitoring unit. The cloud computing platform can directly access the raw data obtained by the third monitoring unit and directly lock the monitoring area corresponding to the third monitoring unit, thereby ensuring the accurate positioning of the predicted risk position.

[0051] In some embodiments, the raw data obtained by the second monitoring unit and / or the first monitoring unit can be processed by a processor arranged in the second monitoring unit and / or the first monitoring unit to extract feature data from the raw data as feature data, and then the feature data is transmitted to the cloud computing platform through the gateway unit, and the cloud computing platform predicts risks according to the feature data. Thus, the system reliability and fault tolerance can be improved; the communication bandwidth requirement is low, the calculation speed is fast, and the reliability and continuity are good.

[0052] In the cloud computing platform, the raw data obtained by the third monitoring unit and the feature data of the second monitoring unit and / or the first monitoring unit are fused together as multi-source perception data. The form of multi-source perception data fusion includes data level fusion, feature level fusion and decision level fusion.

[0053] In some embodiments, as Figure 11As shown, the third detection unit adopts data-level fusion, which is to directly fuse the raw data obtained by the third monitoring unit, and then extract feature vectors from the fused data for judgment and recognition. Data-level fusion can ensure the comprehensiveness of the data, thereby ensuring the accuracy of the decision.

[0054] In some embodiments, as shown in Figure 12 and Figure 13 As shown, the second monitoring unit and the first monitoring unit adopt feature-level fusion or decision-level fusion. Feature-level fusion is to first extract representative feature data from the raw observation data provided by each second monitoring unit or first monitoring unit, and then fuse these feature data into a single feature vector. Feature data includes edges, directions, speeds, shapes, etc. Feature-level fusion has relatively reduced requirements for computing capacity and communication bandwidth, and can efficiently process data.

[0055] In some embodiments, decision-level fusion is a high-level abstraction of data, and the output is a joint decision result. The joint decision should be more accurate or more explicit than any single second monitoring unit and first monitoring unit decision. Decision-level fusion has high flexibility in information processing, and the system has low requirements for information transmission bandwidth, can effectively fuse different types of information reflecting various aspects of the environment or target, and can process non-synchronous information.

[0056] The multi-source perception data is fused through a data fusion model, which includes weighted average method, filtering method estimation method, neural network method, etc.

[0057] In some embodiments, the weighted average method is suitable for multi-source perception decision problems. In this method, different multi-source perception data is given different weights, and the final decision result is based on the weighted average value of each factor. The weight reflects the importance of the multi-source perception data in the entire decision-making process, and can effectively handle multi-dimensional, multi-source multi-source perception data fusion problems. The core idea is:

[0058]

[0059] where D is the decision value, is the result of the oth multi-source perception data, is the weight of the multi-source perception data, satisfying , O is the number of multi-source perception data. First, the various multi-source perception data that affect the decision need to be determined. According to the importance of each multi-source perception data, each factor is given a weight. The weight can be determined by expert evaluation, statistical analysis or experience estimation. The allocation of weights should ensure that the sum of all weights is 1. Multiply the result of each multi-source perception data by its corresponding weight, and then sum all the weighted results to get the final decision value.

[0060] In the multi-source perception system, the data of the third monitoring unit, the second monitoring unit and the first monitoring unit can have different reliability or importance. By using the weighted average method, the multi-source perception data from different sources can be weighted and fused to generate a comprehensive perception result. The method is intuitive and easy to understand, and the weighted average method is simple and intuitive, easy to understand and operate, and suitable for processing problems in multiple dimensions or standards; the weight distribution can be adjusted according to specific needs, so that the method is highly adaptable; the calculation process is simple and the result can be quickly obtained.

[0061] In some embodiments, the filtering method is mainly used to fuse low-level real-time dynamic multi-sensor redundant data. The method recursively determines the optimal fusion and data estimation in a statistical sense using the statistical characteristics of the measurement model. If the system has a linear dynamic model and the errors of the system and the sensors conform to the Gaussian white noise model, the filter will provide the optimal estimation in a unique statistical sense for the fused data.

[0062] The filtering method updates the optimal estimation of the system state in the process of continuously obtaining monitoring data. It is particularly suitable for state estimation of linear, Gaussian noise systems. The filter updates the estimation through two main steps:

[0063] ① Prediction step: predict the state at the current time based on the state at the last time and the model. This step estimates the current state and its uncertainty based on the dynamic model of the system. The definition is as follows:

[0064]

[0065]

[0066] where, is the state prediction at time k, A is the state transition matrix, is the state prediction at time k-1, B is the control matrix, is the control input, is the predicted value of the state covariance matrix at time k, is the predicted value of the state covariance matrix at time k-1, T is the transpose, Q is the process noise covariance matrix, which describes the uncertainty in the model.

[0067] ② Update step: once there is new measurement data, the predicted value and the actual measurement value are combined through this step to generate a more accurate state estimation. This process balances the uncertainty of the prediction and the uncertainty of the measurement to obtain a more reliable estimation, which is defined as follows:

[0068]

[0069]

[0070]

[0071] where, is the gain, representing the trade-off between prediction and measurement, is the observation matrix, is the observation noise covariance matrix, is the state at time k, is the observation data, is the updated covariance matrix.

[0072] Filtering methods can update state estimates every time new data is obtained, making them very suitable for real-time applications, especially in scenarios that require continuous dynamic monitoring and estimation. Filters can effectively handle noise and uncertainty by dynamically adjusting the balance between prediction and measurement, generating more accurate estimates. The recursive nature of filtering allows the system to process without requiring large amounts of data storage and computation.

[0073] In some embodiments, estimation methods derive more accurate posterior probabilities by updating prior probabilities combined with new observation data, providing the basis for decision-making. Estimation methods are suitable for handling uncertainty problems, and can continuously correct and optimize decisions through dynamic updates of information.

[0074] The formula for estimation is as follows:

[0075]

[0076] where, The posterior probability of hypothesis H being true given event E occurs, i.e. the updated probability. The prior probability of hypothesis H being true, i.e. the probability before event occurs, The likelihood of event E occurring given hypothesis H is true, The total probability of event E occurring.

[0077] Estimation methods dynamically update probabilities by combining prior probabilities and events to generate new posterior probabilities for decision-making. Each monitoring data is treated as an estimate, and the correlation probability distribution of each individual object is combined into a joint posterior probability distribution function. By minimizing the likelihood function of the joint distribution function, the final fusion value of multi-source perception data is provided.

[0078] ​Estimation methods can effectively handle uncertainty by combining prior knowledge with new information to generate updated probability estimates. This is particularly effective for scenarios that require decision-making in dynamic environments. Estimation methods can handle complex decision-making problems and can continuously update and adjust models, making them suitable for scenarios with continuous data streams or gradual discovery of new information. Estimation methods can combine multiple information sources to optimize decisions from different perspectives. Whether it is historical data or real-time data, estimation methods can be flexibly adapted.

[0079] Neural network decision fusion refers to the process of combining the decision results from multiple independent decision models or information sources through a neural network to improve the accuracy, robustness, and overall performance of the decision. This method is particularly suitable for multi-source perception data fusion scenarios, where the output results from different sources are integrated to obtain a comprehensive final decision. Common neural networks include MLP, CNN, RNN, ENN (Ensemble Neural Network), etc. Neural networks have strong non-linear mapping capabilities and can capture complex relationships between multiple input signals. Especially when the input decisions or data sources are nonlinear, neural networks can learn these complex patterns through training to achieve more accurate fusion. In addition, neural networks can effectively handle noise and uncertainty. In some cases where there are errors in input decisions or information sources, neural networks can still effectively fuse and derive reasonable final decisions. In multi-source perception data fusion, neural networks can learn the reliability and complementarity of different sensors to reduce the impact of errors.

[0080] In order to fully fuse the multi-source perception data of the third monitoring unit, the second monitoring unit, and the first monitoring unit in the three-dimensional multi-level monitoring system of railway risk disasters, the data fusion model uses the following method to fuse and make decisions on the multi-source perception data.

[0081] In some embodiments, the third monitoring unit is used as a sensing unit, and the second monitoring unit and the first monitoring unit are used as a sensing unit. The analysis is completed inside the sensing unit, and the analysis results are transmitted to the cloud computing platform as real-time analysis data. The cloud computing platform comprehensively analyzes and judges the real-time analysis data.

[0082] In order to fully analyze the risk, the data fusion model proposes a spatio-temporal comprehensive analysis mechanism, which makes decisions on multi-source perception data to determine whether there are abnormal events of various concerns.

[0083] The main and passive nets of the tunnel portal slope are monitored by monitoring optical fibers, and the track surface is monitored by monitoring cameras and / or monitoring radars. Each distinguishable monitoring point of each monitoring optical fiber is considered as an independent sensing unit, and the track surface is considered as an independent sensing unit.

[0084] In some embodiments, it is assumed that the analysis result of the monitoring area is X={F,RC}, where F represents the analysis result of the monitoring fiber optic sensing unit, and RC represents the analysis result of the monitoring camera and / or monitoring radar sensing unit. The analysis result of the monitoring fiber optic monitoring point i at time t is denoted as... The analysis results of monitoring camera and / or radar monitoring point j at time t are as follows: The k multi-source observation analysis values ​​at the tunnel entrance are expressed as follows:

[0085]

[0086]

[0087] in, These represent the monitoring points i in the monitoring fiber optic cable. , ,..., , ,..., The results of the analysis at any given moment. These represent monitoring points j of the monitoring camera and / or monitoring radar, respectively. , ,..., , ,..., The analysis results at each moment. i=1,...,M represents one of the M monitoring points in the first sensing unit, and j=1,...,N represents one of the N monitoring points in the second sensing unit.

[0088] assumed This indicates the presence of an unusual event requiring attention, such as a landslide or falling rocks from the rails. If no abnormal events require attention, then based on the observations of the slope and track surface over k iterations, the final fusion result will be... for:

[0089]

[0090] in, This indicates an abnormal event that requires attention. Given that the event occurs, assume that X is the true posterior probability. This indicates that there are no abnormal events that require attention. Given that the event occurs, assume that X is the true posterior probability.

[0091] Considering the independence between sensing units, then fusion It can be calculated as:

[0092]

[0093] Wherein, r is 0 or 1, s = 1,...,T, T represents the total monitoring time length, s represents a time or a period in the total monitoring time length, Indicates the analysis result corresponding to s time or period.

[0094] In some embodiments, considering the different costs caused by error fusion, i.e. the cost of false alarm is greater than that of false alarm, a cost matrix is introduced, and the matrix element C(u,v) represents the cost of classifying u as v; u represents that there is an abnormal event that needs attention, and v represents that there is no abnormal event that needs attention, and the decision risk is defined as:

[0095]

[0096] Wherein, Indicates that there is no abnormal event that needs attention Occur, the posterior probability of assuming X is true.

[0097] The final category judgment is calculated as follows:

[0098]

[0099] Wherein, Indicates the final category judgment, which refers to whether there is an event of interest, Indicates the parameter value when the decision risk reaches the minimum value, not the minimum value itself. Indicates the hypothesis when there is no abnormal event that needs attention.

[0100] At this time, the category calculated is the final fusion category, and this method can make more stable decisions, especially in high error cost scenarios. Similarly, for other monitoring scenarios such as perimeter, using this data fusion model can realize comprehensive identification of multiple monitoring areas and ensure system monitoring accuracy.

[0101] In the present application, the railway is divided into a third monitoring area, a second monitoring area and a first monitoring area from far to near, the monitoring optical fiber is arranged in the third monitoring area, and the rockfall impact event is monitored, and three-level alarm is realized. The monitoring optical fiber and / or sensing device is arranged in the second monitoring area, the subtle changes of the dangerous rock, slope and active net structure in the second monitoring area in the stress and strain are monitored according to the monitoring optical fiber and / or sensing device, the potential landslide, collapse and other geological disaster risks are identified in advance through data analysis and pattern recognition technology, and two-level early warning is realized for the deformation of the active and passive nets. The sensing device is arranged in the first monitoring area, the sensing device has high resolution and long-distance detection capability, can monitor the track area all-weather and uninterruptedly, and realizes one-level alarm for the rockfall intrusion into the track surface event. The present application carries out hierarchical monitoring around the railway, and realizes real-time monitoring on the risk source itself and its development process, and constructs a three-dimensional, multi-level and multi-level foreign matter intrusion warning monitoring mechanism.

[0102] From the space of the slope and the track surface to the time of early warning and alarm, multi-level three-dimensional monitoring is realized. For example, the stress and strain of the early development of the dangerous rock, slope and active net, the medium-term sliding, rockfall, falling object impact passive net, fence net and the last line of defense of the track surface foreign matter, the accurate monitoring and real-time alarm of the infrastructure and external environment risks such as the trench slope rockfall, slope sliding and personnel intrusion are realized, the intelligent monitoring of the whole line, multiple scenes, multiple targets, three-dimensional and all-weather in the key risk section along the line is implemented, the role of "early prevention, small prevention and timely early warning" is played, the safety risks that are difficult to be found by personnel patrol at present are actively and timely monitored. The blind area, error and negligence of manual inspection are avoided, the golden time for disaster emergency is won, and the railway safety guarantee level is effectively improved.

[0103] The above is only an embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structural transformation using the content of the present application specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A railway risk disaster three-dimensional multi-level monitoring system, characterized in that, The system comprises a third monitoring unit, a second monitoring unit, a first monitoring unit, a gateway unit and a cloud computing platform; the third monitoring unit, the second monitoring unit and the first monitoring unit are respectively arranged in a third monitoring area, a second monitoring area and a first monitoring area of a railway, and acquire third monitoring data, second monitoring data and first monitoring data; the third monitoring area, the second monitoring area and the first monitoring area are gradually away from the track surface of the railway; the third monitoring data, the second monitoring data and the first monitoring data are transmitted to the cloud computing platform through the gateway unit; a data fusion model is preset in the cloud computing platform; the data fusion model fuses the third monitoring data, the second monitoring data and / or the first monitoring data to obtain fusion data; and the cloud computing platform gives different levels of early warning signals based on the fusion data.

2. The railway risk disaster stereoscopic multi-level monitoring system according to claim 1, characterized in that, The first monitoring unit and the second monitoring unit are monitoring cameras and / or monitoring radars, and the third monitoring unit is a monitoring optical fiber.

3. The railway risk disaster stereoscopic multi-level monitoring system according to claim 1, characterized in that, The raw data obtained by the third monitoring unit is directly transmitted to the cloud computing platform through the gateway unit; the cloud computing platform directly fuses the raw data obtained by the third monitoring unit, extracts a feature vector from the fused data for judgment and identification, directly accesses the raw data obtained by the third monitoring unit, and further directly locks the monitoring point corresponding to the third monitoring unit.

4. The railway risk disaster stereoscopic multi-level monitoring system according to claim 3, characterized in that, A processor is arranged in the second monitoring unit and / or the first monitoring unit to extract features in the raw data obtained by the second monitoring unit and / or the first monitoring unit to obtain feature data; the feature data is transmitted to the cloud computing platform through the gateway unit; and the cloud computing platform fuses the feature data into a single feature vector to predict risks.

5. The railway risk hazard stereoscopic multi-level monitoring system according to claim 1, characterized in that, The data fusion model derives an accurate posterior probability by combining the updated prior probability with new observation data, and provides a basis for decision-making; the formula of the data fusion model is as follows: ； wherein, represents the posterior probability that the hypothesis H is true given that the event E has occurred, represents the hypothesis is true, represents the likelihood that the event E has occurred given that the hypothesis H is true, represents the total probability that the event E has occurred.

6. The railway risk hazard stereoscopic multi-level monitoring system according to claim 1, characterized in that, The data fusion model takes the third monitoring unit as a first sensing unit, and takes the second monitoring unit and the first monitoring unit as second sensing units; A monitoring area analysis result is X={F, RC}, wherein F represents an analysis result of the first sensing unit, RC represents an analysis result of the second sensing unit, an analysis result of a monitoring point i of the first sensing unit at time t is denoted as , an analysis result of a monitoring point j of the second sensing unit at time t is denoted as , and the monitoring area k-time multi-source observation analysis value is denoted as ； ； wherein respectively represent the analysis result of the monitoring point i of the first sensing unit at , , , , time; respectively represent the analysis result of the monitoring point j of the second sensing unit at , , , , time, i = 1,..., M represents one of the M monitoring points in the first sensing unit, and j = 1,..., N represents one of the N monitoring points in the second sensing unit.

7. The railway risk hazard stereoscopic multi-level monitoring system according to claim 1, characterized in that, The data fusion model sets indicates that there is an abnormal event that needs attention, indicates that there is no abnormal event that needs attention, then the final fusion result is: ； wherein, represents the posterior probability that the hypothesis X is true given that an abnormal event of interest occurred, and represents the posterior probability that the hypothesis X is true given that an abnormal event of interest did not occur. Fusion may be calculated as: ； Wherein, r is 0 or 1, s = 1,...,T, T represents the total monitoring time length, and s represents a time or a period in the total monitoring time length, represents the analysis result corresponding to the time or period s.

8. The railway risk hazard stereoscopic multi-level monitoring system according to claim 1, wherein, The data fusion model introduces a cost matrix, and a matrix element C(u, v) represents a cost of misclassifying u as v; u represents that there is an abnormal event that needs attention, and v represents that there is no abnormal event that needs attention, at which time a decision risk is defined as: ； wherein, represents the posterior probability of X being true given that no abnormal event of interest occurred; and wherein, The final category judgment is calculated as follows: ； wherein, represents the final category determination, represents the parameter value at which the decision risk takes a minimum value, represents the hypothesis that there is no abnormal event that needs attention.

9. The railway risk hazard stereoscopic multi-level monitoring system according to claim 1, characterized in that, The cloud computing platform analyzes the third monitoring data, the second monitoring data, the first monitoring data and / or the fusion data; if the third monitoring data is found to be abnormal, a third-level early warning signal is issued; if the second monitoring data is found to be abnormal, a second-level early warning signal is issued; If the first monitoring data is found to be abnormal, a first-level early warning signal is issued; and the cloud computing platform predicts the risk level according to the interval time between the third-level early warning signal and the second-level early warning signal.

10. The railway risk hazard stereoscopic multi-level monitoring system according to claim 1, characterized in that, A fourth monitoring area is divided around the railway; the fourth monitoring area is located outside the third monitoring area or the second monitoring area; a fourth monitoring unit is arranged in the fourth monitoring area, and fourth monitoring data is acquired; and the cloud computing platform determines whether to issue an early warning signal based on the fourth monitoring data.

Images