Train automatic rehooking control system and method based on positioning recognition

Abstract

The application discloses a train automatic rehooking control system and method based on positioning recognition, and relates to the technical field of train rehooking.The method is characterized in that a multi-point acquisition device is arranged around the hook head of adjacent car bodies to obtain a hook head spacing variation rate HVR, a proximity speed three-dimensional vector RVV and a posture angular velocity three-dimensional vector RAV, and data preprocessing and standardization operations in the rehooking control server are combined, so that a high-time-efficiency and high-precision closed trend characteristic vector can be constructed.The closed trend score value CTS calculated based on the characteristic vector further considers the dynamic trend of the adjacent car bodies in the straight-line proximity direction and introduces the constraint of the posture disturbance factor, thereby avoiding misjudgment caused by single parameter determination.The method effectively improves the intelligent recognition ability of the rehooking time, realizes high-precision judgment of the closed trend under complex dynamic working conditions, and makes the rehooking process have higher precision and stability.

Description

Technical Field

[0001] This invention relates to the field of train recoupling technology, specifically to an automatic train recoupling control system and method based on positioning recognition. Background Technology

[0002] This paper discusses vehicle connection and decoupling control technology during railway train operation, particularly an automated recoupling control method applied to train marshalling, demarcation, and operation. In this field, train cars are mechanically connected via couplers. However, after train marshalling adjustments or decoupling due to faults, recoupling often needs to be performed again. Traditional recoupling relies on manual operation or limited automated control methods, resulting in insufficient environmental adaptability, low motion judgment accuracy, and lagging control feedback. Therefore, in the scenario of automatic train car recoupling, achieving high-precision and stable automatic recoupling through positioning recognition has become an important research and application direction in this field.

[0003] Currently, in existing car body re-coupling control processes, although some systems can assist in closure judgment through position detection or speed sensing, they mostly remain at the level of identifying static position errors, lacking real-time modeling and analysis of dynamic closure trends. This method may misjudge when the hook head approaches due to attitude disturbances, uneven rail surfaces, or lateral offsets, causing the system to believe that the closure conditions have been met when in fact the effective closure requirement has not been achieved. Especially when the hook head has not yet completed attitude stabilization, prematurely triggering the closure action can easily cause closure impact, misalignment, or re-coupling failure. Existing technologies, when dealing with such dynamic and complex environments, exhibit inaccurate timing of actions and poor robustness to sudden disturbances, making it difficult to guarantee the reliability and safety of the re-coupling action.

[0004] The aforementioned situation and shortcomings mainly stem from the imperfections of existing technology in handling the transition state during the recoupling process. When the hook head is relatively close but has not yet reached the closing condition, due to the limited detection methods, the system is prone to misinterpreting short-term proximity signals as recoupling completion signals, thus incorrectly issuing closing execution commands. Such misjudgments often lead to a series of abnormal effects: on the one hand, it may cause a hard impact on the hook head, resulting in mechanical wear or even damage to the hook body and coupler seat; on the other hand, after a failed closing action, it is necessary to re-unlock and adjust, increasing train scheduling delays and affecting train formation efficiency. Simultaneously, due to the lack of dynamic trend recognition and feedback mechanisms, the system cannot extract disturbance features from the failure process for adaptive optimization, causing similar problems to repeatedly occur in subsequent operations, seriously affecting the reliability and intelligence level of automatic recoupling control. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a train automatic recoupling control system and method based on positioning recognition, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution, comprising the following steps:

[0007] S1. Deploy multiple data acquisition devices around the hooks of adjacent cars to be re-coupled to collect sensing data in real time; transmit the collected sensing data to the re-coupling control server wirelessly in real time, and preprocess the data in the re-coupling control server to construct a standardized closed trend feature vector.

[0008] S2. Calculate the Closure Trend Score (CTS) based on the standardized closure trend feature vector, and then classify the current re-hook state into closure level I, compensation level II, and blocking level III based on the closure trend score (CTS).

[0009] S3. When the Closed Trend Score (CTS) is at Compensation Level II, the Attitude Disturbance Compensation Control Mechanism is triggered. The Attitude Disturbance Guiding Three-Dimensional Vector (PAV) is calculated. Based on the Attitude Disturbance Guiding Three-Dimensional Vector (PAV), the control command is output to fine-tune the pushing direction of the traction device and perform the hook alignment attitude compensation operation.

[0010] S4. After the hook alignment posture compensation operation is completed, the longitudinal axis acceleration a during the closing process is collected, the motion stability score STS is calculated, and the stability is evaluated.

[0011] Preferably, S1 includes S11;

[0012] S11. The multi-point acquisition device includes a laser rangefinder sensor deployed at the front end of the coupler of adjacent car bodies, a multi-axis velocity sensor deployed on the side wall of the end of the car body, and an inertial measurement unit (IMU) installed on the coupler seat; it acquires the sensing data of the train coupler in real time.

[0013] The sensing data includes the hook-head spacing change rate HVR, the three-dimensional approach velocity vector RVV, and the three-dimensional attitude angular velocity vector RAV;

[0014] The laser rangefinder sensor monitors the hook head spacing in real time by deploying a laser rangefinder sensor at the front end of the hook head, and calculates the hook head spacing change rate HVR through continuous time difference calculation.

[0015] The multi-axis velocity sensor is used to collect relative velocity components in real time along the longitudinal, transverse and vertical axes of the vehicle body, forming a three-dimensional approach velocity vector RVV between adjacent vehicle bodies;

[0016] The inertial measurement unit (IMU) is used to acquire the three-dimensional vector (RAV) of the vehicle's attitude angular velocity in the pitch, yaw, and roll directions in real time.

[0017] The multi-point acquisition device performs collaborative acquisition according to a synchronous triggering sequence.

[0018] Preferably, S1 further includes S12 and S13;

[0019] S12. The multi-point acquisition device is connected to the hook control server via a data transmission module. The data transmission module includes an edge computing node, a temporary cache unit and a main station communication interface.

[0020] Among them: the edge computing node is located inside the vehicle body and is used to compress and encode the sensing data collected by the multi-point acquisition device;

[0021] Temporary buffer units provide redundancy compensation for abnormal packet loss;

[0022] The main station communication interface uploads the compressed data in real time to the hook control server located in the control center in the form of protocol-based data frames;

[0023] S13. In the double hook control server, the received sensing data is preprocessed to obtain a standardized closed trend feature vector;

[0024] The preprocessing includes time alignment, noise filtering, and feature normalization.

[0025] The time alignment process calibrates the timestamps of each acquisition device by configuring a unified clock synchronization protocol, so that all sensing data have a unified time reference benchmark; and constructs a time sliding buffer queue based on the sampling time window, which automatically aligns the sensing data according to the time axis after it is received.

[0026] The noise removal process employs a two-stage filtering mechanism to denoise the original signal. Specifically, a 3-point sliding window mid-range filter is used to remove short-time pulses from the hook spacing change rate (HVR) and the approach velocity 3D vector (RVV), eliminating spike anomalies caused by instantaneous track surface disturbances and encoder jumps. A 5-point Gaussian weighted filter is used to perform smooth convolution processing on the attitude angular velocity 3D vector (RAV).

[0027] Among them, the weight coefficients of the five-point Gaussian weight are {0.0625, 0.25, 0.375, 0.25, 0.0625};

[0028] The feature normalization process involves normalizing the noise-filtered sensing data according to the historical operating conditions range. Each parameter in the sensing data is normalized using the extreme value normalization method. The normalized result is a dimensionless value, which is then combined into a standardized closed-trend feature vector.

[0029] Preferably, S2 includes S21;

[0030] S21. A closed-loop trend scoring calculation model is pre-constructed in the hook control server. The closed-loop trend scoring calculation model consists of an input layer, a feature operation layer, and an output layer, wherein:

[0031] The input layer is used to receive the standardized closed-trend feature vector in real time; the transmitted hook spacing change rate HVR, approach velocity three-dimensional vector RVV, and attitude angular velocity three-dimensional vector RAV;

[0032] The feature operation layer is calculated according to the following rules: First, the absolute value of the hook spacing change rate HVR is introduced into the logarithmic function for nonlinear amplification to optimize the trend recognition sensitivity when the hook approaches; then, the magnitude of the approach velocity three-dimensional vector RVV is used as a factor of trend intensity to quantify the closed potential energy of the relative motion of the vehicle body; at the same time, the magnitude of the attitude angular velocity three-dimensional vector RAV is placed as a perturbation factor in the denominator to suppress the trend score under unstable attitude conditions.

[0033] The output layer generates a Closed Trend Score (CTS) based on the output results of the feature operation layer, and comprehensively analyzes the approach trend and attitude stability of adjacent car body hooks during the re-coupling process.

[0034] Preferably, S2 further includes S22;

[0035] S22. Set closure interval thresholds based on historical data. The closure interval thresholds include a closure confirmation threshold F1 and a closure validity lower limit threshold F2. Compare the closure trend score value CTS output in real time by the closure trend score calculation model with the closure interval thresholds, and classify the current re-hook status into closure level I, compensation level II, and blocking level III. The specific comparison content is as follows:

[0036] When the closing trend score CTS is greater than or equal to the closing confirmation threshold F1, it is classified as closing level I, and it is determined that there is a stable and effective closing trend between the adjacent car body hooks. The system immediately issues a closing execution command.

[0037] When the lower limit threshold of closure effectiveness F2 ≤ closure trend score CTS < closure confirmation threshold F1, it is classified as compensation level II, closure is suspended, and attitude disturbance compensation mechanism is triggered.

[0038] When the Closure Trend Score (CTS) is less than the lower limit of closure effectiveness (F2), it is classified as Blockage Level III, and the next cycle is initiated for re-collection and judgment. If the Blockage Level III is maintained for three consecutive cycles, an early warning message is sent to the operation interface of the re-hook control server.

[0039] Preferably, S3 includes S31;

[0040] S31. After triggering the attitude disturbance compensation mechanism, the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV are input into the compensation calculation model of the compound hook control server. In the compensation calculation model, a vector cross product operation is performed on the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV to obtain the disturbance direction relationship between the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV. The calculation result is defined as the attitude disturbance guidance three-dimensional vector PAV. Then, the output attitude disturbance guidance three-dimensional vector PAV is stored in the compound hook control server in the form of a three-dimensional vector (PAVx, PAVy, PAVz).

[0041] Preferably, S3 further includes S32;

[0042] S32. Based on the attitude disturbance, guide the three-dimensional vector PAV to output control commands and execute the hook head alignment attitude compensation operation;

[0043] The hook alignment attitude compensation operation calculates the adjustment amount △Jd of the pushing direction angle and the adjustment amount △Sd of the pushing speed by using the attitude disturbance-guided three-dimensional vector PAV.

[0044] Among them, the adjustment amount △Jd of the pushing direction angle is obtained by comparing the attitude disturbance-guided three-dimensional vector PAV with the hook reference pushing direction vector Dref and using the vector angle formula;

[0045] The adjustment amount △Sd of the push speed is derived from the magnitude of the attitude perturbation-guided three-dimensional vector PAV;

[0046] During the hook alignment attitude compensation operation, the pushing device corrects the vehicle's propulsion direction according to the adjustment amount △Jd of the pushing direction angle, so that the hook turns from the offset state to the closing direction; at the same time, it finely adjusts the propulsion speed according to the adjustment amount △Sd of the pushing speed; after the hook alignment attitude compensation operation is completed, the system recalculates the closing trend score value CTS. If the closing trend score value CTS is improved to the closing level I, the closing action is executed.

[0047] Preferably, S4 includes S41;

[0048] S41. After the hook alignment posture compensation operation is completed, within the time window from the start time t0 to the start time t0+0.3 seconds after the closing action is completed, the continuous data of the longitudinal axis acceleration a(t) at time t within the 0.3-second time window are collected by the multi-axis velocity sensor, and the mean value aj of the continuous data of the longitudinal axis acceleration a is calculated. The maximum deviation between the longitudinal axis acceleration a and the mean value aj of the continuous data is calculated, and the acceleration stability score STS is obtained.

[0049] Then, a stability assessment is performed based on the Acceleration Stability Score (STS) to determine the stability of the hook re-hook after the hook alignment attitude compensation operation is completed; the specific assessment content is as follows:

[0050] When the acceleration stability score (STS) is less than 0.2 m / s² 2 At that time, the closing action is determined to be stable and successful;

[0051] When the acceleration stability score (STS) is greater than 0.5 m / s² 2 If the closure action is deemed abnormal, the closure is considered a failure.

[0052] When the acceleration stability score (STS) is 0.2 m / s² 2 up to 0.5m / s 2 If there is uncertainty in the determination of the closed state, a mechanism for re-collection and re-evaluation is triggered.

[0053] Preferably, S4 further includes S42;

[0054] S42. When the stability assessment indicates that the closure has failed, the hook control server executes an action feedback process, the specific content of which is as follows:

[0055] The original standardized closure trend feature vector during the re-hook process and the acceleration stability score (STS) value at the time of failure are sent back to the initial acquisition node in order to reconstruct the closure trend score (CTS) value and the attitude disturbance guidance three-dimensional vector (PAV) and re-initiate the re-hook action.

[0056] If closure failure occurs in three consecutive judgment periods and is manifested as repeated attitude disturbances, the standardized closure trend feature vector of the closure failure is written into the disturbance feature database.

[0057] In the subsequent re-hooking process, the disturbance feature database is called as part of the trend feature database. The scoring function is corrected and iteratively optimized using historical disturbance features, so that the Closed Trend Score (CTS) and Acceleration Stability Score (STS) can adaptively identify similar disturbance situations in the next judgment.

[0058] An automatic train recoupling control system based on positioning recognition includes a multi-point acquisition module, a closure trend analysis module, a hook alignment compensation module, and a stability optimization module.

[0059] The multi-point acquisition module collects sensing data in real time by deploying multi-point acquisition devices around the hooks of adjacent cars to be re-coupled; the collected sensing data is transmitted wirelessly to the re-coupling control server in real time, and the data is preprocessed in the re-coupling control server to construct a standardized closed trend feature vector.

[0060] The closure trend analysis module calculates the closure trend score (CTS) based on the standardized closure trend feature vector, and then classifies the current re-hook state into closure level I, compensation level II, and blocking level III based on the closure trend score (CTS).

[0061] The hook alignment compensation module triggers the attitude disturbance compensation control mechanism when the closure trend score value CTS is at compensation level II, calculates the attitude disturbance guidance three-dimensional vector PAV, outputs control commands based on the attitude disturbance guidance three-dimensional vector PAV to fine-tune the pushing direction of the traction device, and performs hook alignment attitude compensation operation.

[0062] The stability optimization module collects the longitudinal axis acceleration 'a' during the closing process after the hook alignment posture compensation operation is completed, calculates the motion stability score (STS), and performs a stability assessment.

[0063] This invention provides a train automatic recoupling control system and method based on positioning recognition. It has the following beneficial effects:

[0064] (1) This method acquires the hook spacing change rate (HVR), the three-dimensional approach velocity vector (RVV), and the three-dimensional attitude angular velocity vector (RAV) by deploying multiple acquisition devices around the hooks of adjacent car bodies. Combined with data preprocessing and standardization operations in the re-coupling control server, this method can construct a highly timely and accurate closed-loop trend feature vector. The closed-loop trend score (CTS) calculated based on this feature vector not only considers the dynamic trend of adjacent car bodies in the straight approach direction but also introduces the constraint of attitude disturbance factors, thereby avoiding misjudgment caused by a single parameter. This method effectively improves the intelligent recognition capability of re-coupling timing, realizes high-precision judgment of closed-loop trends under complex dynamic working conditions, and makes the re-coupling process more accurate and stable.

[0065] (2) When the closing trend score value (CTS) is at compensation level II, this method triggers the attitude disturbance compensation control mechanism. It uses the approach velocity three-dimensional vector (RVV) and the attitude angular velocity three-dimensional vector (RAV) to perform vector operations to generate the attitude disturbance guidance three-dimensional vector (PAV). Based on this, it calculates the push direction angle adjustment amount (ΔJd) and the push speed adjustment amount (ΔSd), thus achieving fine-tuning of the traction device's push direction and speed. This dynamic compensation strategy can correct the push path in real time when the vehicle body has lateral axis offset, vertical axis bumps, or unstable attitude, allowing the hook head to gradually approach the ideal closed attitude. Compared with the traditional re-coupling method that relies on manual or static threshold judgment, this mechanism significantly reduces the probability of impact and misalignment during the re-coupling process, thereby enhancing the overall stability and controllability of the re-coupling action.

[0066] (3) After the recoupling operation is completed, this method quantitatively evaluates the success of the closing action by collecting longitudinal axis acceleration and calculating the action stability score (STS). When the closing action fails, the system feeds back the standardized closing trend feature vector, the attitude disturbance guidance three-dimensional vector (PAV), and the acceleration stability score (STS) to the trend feature database, and stores and iteratively optimizes the failed features. In subsequent recoupling processes, the system can call historical disturbance features to correct the scoring function, gradually achieving adaptive optimization for multiple working conditions. This mechanism enables the recoupling control method to have self-learning capabilities, continuously improving its ability to identify and respond to abnormal disturbances in complex environments during multiple runs, thereby achieving long-term evolution of system performance and significantly improving train formation and operation efficiency. Attached Figure Description

[0067] Figure 1 This is a schematic diagram illustrating the steps of an automatic train recoupling control method based on positioning recognition according to the present invention.

[0068] Figure 2 This is a schematic diagram of the automatic train recoupling control system based on positioning recognition according to the present invention.

[0069] Figure 3 This is a schematic diagram of the multi-point data acquisition device deployment. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1, please refer to Figure 1 This invention provides a method for automatic recoupling control of trains based on positioning recognition. To achieve the above objectives, this invention is implemented through the following technical solution, including the following steps:

[0072] S1. Deploy multiple data acquisition devices around the hooks of adjacent cars to be re-coupled to collect sensing data in real time; transmit the collected sensing data to the re-coupling control server wirelessly in real time, and preprocess the data in the re-coupling control server to construct a standardized closed trend feature vector.

[0073] S2. Calculate the Closure Trend Score (CTS) based on the standardized closure trend feature vector, and then classify the current re-hook state into closure level I, compensation level II, and blocking level III based on the closure trend score (CTS).

[0074] S3. When the Closed Trend Score (CTS) is at Compensation Level II, the Attitude Disturbance Compensation Control Mechanism is triggered. The Attitude Disturbance Guidance Three-Dimensional Vector (PAV) is calculated. Based on the Attitude Disturbance Guidance Three-Dimensional Vector (PAV), the control command is output to fine-tune the pushing direction of the traction device and execute the hook alignment attitude compensation operation.

[0075] S4. After the hook alignment posture compensation operation is completed, the longitudinal axis acceleration a during the closing process is collected, the motion stability score STS is calculated, and the stability is evaluated.

[0076] In this embodiment, S1 is the core starting point. The reason for deploying multiple acquisition devices around the hook head and simultaneously acquiring the hook head spacing change rate (HVR), the approach velocity three-dimensional vector (RVV), and the attitude angular velocity three-dimensional vector (RAV) is that relying solely on single position information is prone to misjudgment. For example, when the hook heads of two car bodies briefly approach each other due to uneven rail surface or inertial disturbance, without real-time speed and attitude parameters, it might be mistakenly assumed that the closure condition has been met. This multi-point collaborative acquisition ensures the integrity and real-time nature of the information. After preprocessing into a standardized closure trend feature vector, it can provide stable and reliable input for subsequent trend judgment. In S2, the closure trend score value (CTS) is obtained by modeling and calculating the standardized closure trend feature vector. Its significance lies in transforming the "whether it is closed" from a vague empirical judgment into a quantifiable score value. For example, when RVV is large but RAV fluctuates violently at the same time, the system will automatically determine it as "compensation level II" instead of "closure level I," avoiding the impact problem caused by premature closure due to attitude instability. This grading not only dynamically reflects the true physical trend of the approach process but also provides a precise basis for subsequent control actions. In S3, when in "compensation level II", a three-dimensional attitude disturbance guidance vector (PAV) is introduced to correct the pushing direction. Its design purpose is very clear: if the car body hook head deviates due to crosswinds, uneven rail gaps, or operational inertia, simply continuing to push forward often leads to misalignment or top deviation. However, the pushing direction angle adjustment ΔJd and speed adjustment ΔSd calculated by PAV can allow the traction device to "straighten itself" before "closes" to the target position. Thus, the re-coupling process changes from a simple direct push to a flexible push with self-correction, significantly reducing the closure failure rate. Finally, in S4, by collecting the longitudinal axis acceleration during the closure process and calculating the action stability score (STS), it is equivalent to performing a "quality check" on the entire action. The real physical meaning is: if the closure action is stable, the acceleration curve will be relatively smooth; if an impact or disturbance occurs, the standard deviation will be instantly amplified. Setting a threshold to judge the STS allows for rapid detection of closure anomalies and triggers a feedback optimization mechanism. This not only improves the safety of actions, but also enables the system to learn and evolve on its own through the accumulation of failure data.

[0077] Example 2, please refer to Figure 1 and Figure 3 Specifically: S1 includes S11;

[0078] S11. The multi-point acquisition device includes a laser rangefinder sensor deployed at the front end of the coupler of adjacent car bodies, a multi-axis velocity sensor deployed on the side wall of the end of the car body, and an inertial measurement unit (IMU) installed on the coupler seat; it acquires the sensing data of the train coupler in real time.

[0079] The sensing data includes the hook-nose distance change rate (HVR), the three-dimensional approach velocity vector (RVV), and the three-dimensional attitude angular velocity vector (RAV).

[0080] The laser rangefinder sensor monitors the hook head spacing in real time by deploying a laser rangefinder sensor at the front end of the hook head, and calculates the hook head spacing change rate HVR through continuous time difference calculation;

[0081] Multi-axis velocity sensors are used to collect relative velocity components in real time along the longitudinal, transverse, and vertical axes of the vehicle body, forming a three-dimensional approach velocity vector RVV between adjacent vehicle bodies;

[0082] The inertial measurement unit (IMU) is used to acquire the three-dimensional vector (RAV) of the vehicle's attitude angular velocity in the pitch, yaw, and roll directions in real time.

[0083] Multi-point acquisition devices collect data collaboratively according to a synchronous triggering sequence to ensure the timeliness consistency of dynamic parameters;

[0084] It should be noted that the approach velocity 3D vector RVV and the attitude angular velocity 3D vector RAV are both 3D vectors, specifically in the form of RVV=(RVVx, RVVy, RVVz) and RAV=(RAVx, RAVy, RAVz); where: describing the speed and direction of the vehicle body hook in linear motion;

[0085] RVV = (RVVx, RVVy, RVVz) represents the relative velocity between the two car body hooks in the three-dimensional coordinate system. It describes the speed and direction of the car body hooks in linear motion and is used to determine whether the hooks are "approaching forward". It is the direct driving factor of the closing trend.

[0086] RAV = (RAVx, RAVy, RAVz) represents the angular velocity of the car body hook around its own coordinate axis, describing the speed and direction of the car body hook's rotation, and is used to determine whether the hook is "stable in attitude". If the angular velocity is large, it means that the hook is swaying or rotating, which may lead to closure failure even if the linear velocity is appropriate.

[0087] RVVx represents the horizontal component of the approach velocity, reflecting whether there is lateral offset;

[0088] RVVy represents the longitudinal component of the approach velocity, which mainly determines whether the hook is approaching;

[0089] RVVz represents the vertical component of the approach velocity, reflecting the presence of vertical axis turbulence;

[0090] RAVx represents the roll rate of the attitude angular velocity about the horizontal axis;

[0091] RAVy represents the pitch velocity of the attitude angular velocity about the longitudinal axis;

[0092] RAVz represents the yaw rate of the attitude angular velocity about the vertical axis.

[0093] S1 also includes S12 and S13;

[0094] S12. The multi-point acquisition device is connected to the control server via a data transmission module. The data transmission module includes an edge computing node, a temporary cache unit and a main station communication interface.

[0095] Among them: the edge computing node is located inside the vehicle body and is used to compress and encode the sensing data collected by the multi-point acquisition device;

[0096] Temporary buffer units provide redundancy compensation for abnormal packet loss;

[0097] The main station communication interface uploads the compressed data in real time to the hook control server located in the control center in the form of protocol-based data frames;

[0098] S13. In the double hook control server, the received sensing data is preprocessed to obtain a standardized closed trend feature vector;

[0099] Preprocessing includes time alignment, noise filtering, and feature normalization.

[0100] Time alignment processing calibrates the timestamps of each acquisition device by configuring a unified clock synchronization protocol, so that all sensing data have a unified time reference benchmark; and constructs a time sliding buffer queue based on the sampling time window, automatically aligning the sensing data according to the time axis after the sensing data is received, ensuring that the sensing data are jointly calculated at the same time node.

[0101] The noise removal process employs a two-stage filtering mechanism to denoise the original signal. Specifically, a 3-point sliding window mid-range filter is used to remove short-time pulses from the hook spacing change rate (HVR) and the approach velocity 3D vector (RVV), eliminating spike anomalies caused by instantaneous rail surface disturbances and encoder jumps. A 5-point Gaussian weighted filter is used to perform smooth convolution processing on the attitude angular velocity 3D vector (RAV).

[0102] Among them, the weight coefficients of the five-point Gaussian weight are {0.0625, 0.25, 0.375, 0.25, 0.0625}, in order to improve the stability and trend expression ability of the angular velocity signal;

[0103] Feature normalization is performed by normalizing the noise-filtered sensing data according to the historical operating conditions. Each parameter in the sensing data is normalized using the extreme value normalization method. The normalized result is a dimensionless value and is combined into a standardized closed trend feature vector.

[0104] In this embodiment, the deployment design of method S11 directly solves the problem of easy loss of key information in traditional single-point acquisition. For example, if only laser ranging sensors are used to collect distance, slight track fluctuations or slight vehicle body sway will cause short-term anomalies in distance changes, leading to system misjudgment. By coordinating the deployment of laser ranging sensors, multi-axis velocity sensors, and inertial measurement units (IMUs), not only can the hook head spacing change rate (HVR) be obtained in real time, but the approach velocity three-dimensional vector (RVV) and attitude angular velocity three-dimensional vector (RAV) can also be captured simultaneously, forming a complete closed-loop monitoring of "distance-velocity-attitude". This multi-dimensional perception avoids the deviation caused by single-parameter distortion and improves the real-time stability of the re-coupling process. The core purpose of the data transmission module design in S12 is to ensure the continuity and reliability of dynamic data. For example, in the presence of interference in the re-coupling environment, the vehicle body wireless communication link is prone to data packet loss due to interference. Without redundancy compensation, data interruption may occur, leading to delays in the re-coupling control server's judgment. Real-time compression and encoding through edge computing nodes can reduce the transmission bandwidth pressure; the temporary buffer unit performs redundancy compensation for lost data packets to ensure that key data is delivered completely. In this way, the input data received by the control server is continuous, complete, and protocol-compliant, ensuring the accuracy of the input to the subsequent calculation model. In the preprocessing stage of S13, the combination of time alignment, noise filtering, and feature normalization solves the problems of multi-source asynchrony and signal fluctuations. For example, if the timestamp of the laser ranging sensor deviates from that of the IMU by tens of milliseconds, the calculated trend vector will be "misaligned," thus affecting the scoring results. By unifying clock synchronization and interpolation alignment, it is ensured that all sensing data are jointly calculated under the same time reference, avoiding the hidden danger of "time misalignment." Simultaneously, a two-stage filtering mechanism is employed: on the one hand, median filtering removes abnormal pulses caused by sudden disturbances on the track surface; on the other hand, Gaussian weighted smoothing suppresses attitude signal jitter, making the input data smoother and more stable. Finally, extreme value normalization unifies data of different dimensions into a dimensionless vector, thereby ensuring the computability and consistency of the trend scoring model.

[0105] Example 3, please refer to Figure 1 Specifically: S2 includes S21;

[0106] S21. A closed-loop trend scoring calculation model is pre-built in the hook control server. The closed-loop trend scoring calculation model consists of an input layer, a feature operation layer, and an output layer, wherein:

[0107] The input layer is used to receive the standardized closed-trend feature vector in real time; the transmitted hook spacing change rate HVR, approach velocity three-dimensional vector RVV, and attitude angular velocity three-dimensional vector RAV;

[0108] The feature operation layer is calculated according to the following rules: First, the absolute value of the hook spacing change rate HVR is introduced into the logarithmic function for nonlinear amplification to optimize the trend recognition sensitivity when the hook approaches; then, the magnitude of the approach velocity three-dimensional vector RVV is used as a factor of trend intensity to quantify the closed potential energy of the relative motion of the vehicle body; at the same time, the magnitude of the attitude angular velocity three-dimensional vector RAV is placed as a perturbation factor in the denominator to suppress the trend score under unstable attitude conditions.

[0109] The output layer generates a Closed Trend Score (CTS) based on the output of the feature operation layer, and comprehensively analyzes the approach trend and attitude stability of adjacent car body hooks during the re-coupling process.

[0110] The specific form of the closed-trend scoring calculation model is as follows: Where ln represents the logarithmic function;

[0111] The approach velocity 3D vector RVV is a vector that typically contains three components (RVVx, RVVy, RVVz), representing the relative velocities of adjacent vehicle bodies in three coordinate directions. The modulus is calculated as follows: Physical meaning: Represents the "actual approach speed" between two vehicles; if the RVV modulus is large, it means the vehicles are approaching quickly; if the RVV modulus is close to 0, it means the vehicles are almost stationary or there is no obvious approach.

[0112] The attitude angular velocity three-dimensional vector RAV is also a vector, typically containing three components (RAVx, RAVy, RAVz), representing the angular velocities of the hook in the pitch, yaw, and roll directions, respectively; Modulus calculation method: Physical meaning: It represents the overall attitude change rate of the hook head, that is, the intensity of "jitter" or "offset"; if the RAV modulus is large, it indicates that the hook head attitude is unstable and there is a strong disturbance; if the RAV modulus is close to 0, it indicates that the hook head attitude is stable.

[0113] Fundamental source: The magnitude of the three-dimensional velocity vector RVV, ||RVV||, originates from the classical dynamics formula for relative motion and is a fundamental definition of velocity vectors in physics;

[0114] The magnitude of the three-dimensional angular velocity vector RAV, ||RAV||, is derived from the kinematics formula of rigid bodies. Taking the Euclidean magnitude of the three-axis angular velocity vector is a basic method for calculating mechanical angular velocity. The rate of change of distance, |HVR|, is derived from the derivative formula of kinematic displacement with respect to time, which is the differential definition of position changing with time. The introduction of the logarithmic function ln(1+|HVR|) is borrowed from the logarithmic amplification algorithm in signal processing, and is used to enhance trend sensitivity when there are small changes.

[0115] Improvement process: Based on the existing "approach velocity / angular velocity" ratio, a logarithmic transformation of the distance change rate is introduced for the first time to characterize the approach trend of the hook; the three-dimensional vector of attitude angular velocity RAV is placed in the denominator to form a penalty mechanism to avoid erroneous high scores under unstable attitude conditions;

[0116] Dimensionality analysis:

[0117] The absolute value of the hook spacing change rate HVR, |HVR|, is dimensionless (derivative of the distance change rate, approximately equivalent acceleration).

[0118] ln(1+|HVR|): The logarithmic function outputs a dimensionless value;

[0119] The magnitude of the approximate velocity three-dimensional vector RVV is ||RVV|| (relative velocity magnitude).

[0120] The magnitude of the attitude angular velocity three-dimensional vector RAV is ||RAV|| (angular velocity magnitude).

[0121] The "1" in the denominator 1+||RAV|| represents a dimensionless reference constant, and it remains dimensionless after being added to ||RAV||. In order to maintain dimensional consistency, normalization is performed in practice to convert the denominator into a dimensionless quantity.

[0122] Therefore, the entire formula outputs a closed-trend score (CTS), which is a dimensionless score and suitable for judgment and comparison.

[0123] S2 also includes S22;

[0124] S22. Set closure interval thresholds based on historical data. The closure interval thresholds include a closure confirmation threshold F1 and a closure effectiveness lower limit threshold F2. Compare the closure trend score (CTS) output in real time by the closure trend score calculation model with the closure interval thresholds, and classify the current re-hook status into closure level I, compensation level II, and blocking level III. The specific comparison content is as follows:

[0125] When the closing trend score CTS is greater than or equal to the closing confirmation threshold F1, it is classified as closing level I, and it is determined that there is a stable and effective closing trend between the adjacent car body hooks. The system immediately issues a closing execution command.

[0126] When the lower limit threshold of closure effectiveness F2 ≤ closure trend score CTS < closure confirmation threshold F1, it is classified as compensation level II, the hook approach trend is determined to be in a fuzzy or disturbed state, closure is suspended, and the attitude disturbance compensation mechanism is triggered.

[0127] When the Closure Trend Score (CTS) is less than the lower limit threshold (F2) of closure effectiveness, it is classified as Blockage Level III. The hook movement trend is deemed insufficient to support the closure action. The system suspends the current closure process and enters the next cycle for re-collection and judgment. If the Blockage Level III occurs for three consecutive cycles, an early warning message is sent to the operation interface of the re-hook control server. The early warning message is "Re-hook Failure Message," prompting manual intervention to avoid re-hook failure due to incorrect closure.

[0128] In this embodiment, the key design of S21 lies in introducing the hook spacing change rate HVR into a logarithmic function for nonlinear amplification. This is because during the gradual approach of the train hooks, the initial small changes of a few millimeters are often most easily masked by rail surface disturbances and signal noise. If a linear value is used directly, the approach trend may be underestimated. By transforming ln(1+|HVR|), small changes can be amplified in the early stages of approach, improving the sensitivity of trend recognition, thereby avoiding misjudgments of "unperceived approach" and ensuring a more accurate timing for the re-coupling action. In the same calculation model, the magnitude of the three-dimensional vector of approach velocity RVV is used as the closing trend strength factor, whose physical meaning is to characterize the "actual approach potential energy" between the two car bodies. If RVV is too small, even if HVR shows a certain trend, it may be an illusion caused by random vibration. Therefore, introducing RVV|| as a weighting factor can ensure that the driving force of the closing trend is real and reliable. The purpose of doing this is to prevent instantaneous fluctuations from being misjudged as effective closure when in fact the two car bodies are still relatively stationary. On the other hand, placing the magnitude of the attitude angular velocity three-dimensional vector RAV in the denominator creates a disturbance penalty mechanism. This is because during actual re-hooking, if the hook head sways excessively in the yaw or pitch directions, even with a good linear approach trend, it is highly susceptible to "angle jamming" at the moment of closure. By suppressing this instability in the scoring model, false signals of "dangerous closure" can be effectively filtered out. For example, when RAV suddenly increases, CTS will rapidly decrease, causing the system to enter blocking level III, thus avoiding untimely closure actions. The threshold settings in S22 further ensure the robustness of the decision. F1 is the closure confirmation threshold, determined based on the "lower statistical limit of CTS when the re-hooking success rate exceeds 95%" in historical re-hooking data samples, and can be dynamically corrected through long-term running datasets to ensure that closure actions are only allowed under high-confidence trends. F2 is the lower limit threshold for closure effectiveness, determined based on the "upper statistical limit of CTS when the re-hooking failure rate exceeds 80%" in historical re-hooking data samples, and calibrated through experimental conditions to ensure that closure actions are prevented when the trend is insufficient. The closure confirmation threshold F1 acts as a "safety threshold." The system only allows closure to be executed when the CTS is consistently higher than F1, preventing premature actions. The lower limit threshold F2 for closure validity prevents blind compensation when the trend is insufficient. For example, if the CTS remains below F2 for an extended period, the system directly blocks the process and prompts for manual intervention, avoiding an ineffective cycle of repeated compensation without success. This hierarchical judgment mechanism allows the re-hook action to be performed at the "appropriate time and under appropriate conditions," avoiding blindness and improving both safety and efficiency. Through the above implementation methods, the design of S2 achieves three objectives: first, it amplifies small-amplitude approach trends, enhancing sensitivity in the initial stage; second, it introduces dual constraints of speed and attitude, improving the reliability and stability of the scoring; and third, through threshold hierarchical control, it effectively avoids erroneous actions and closure failures.Therefore, the re-hook process not only makes the judgment more accurate, but also significantly reduces the probability of impact, blockage and failure caused by misjudgment.

[0129] Example 4, please refer to Figure 1 Specifically: S3 includes S31;

[0130] S31. After triggering the attitude disturbance compensation mechanism, the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV are input into the compensation calculation model of the double hook control server. In the compensation calculation model, a vector cross product operation is performed on the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV to obtain the disturbance direction relationship between the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV. The calculation result is defined as the attitude disturbance guiding three-dimensional vector PAV. The magnitude of the attitude disturbance guiding three-dimensional vector PAV is used to characterize the intensity of the attitude disturbance. The magnitude of the attitude disturbance guiding three-dimensional vector PAV is proportional to the hook head attitude offset amplitude. Then, the output attitude disturbance guiding three-dimensional vector PAV is stored in the double hook control server in the form of a three-dimensional vector (PAVx, PAVy, PAVz) for the generation of subsequent compensation commands and closure control determination of the vehicle body pushing device.

[0131] The attitude perturbation-guided 3D vector PAV is specifically calculated and output through the following compensation calculation model: ;

[0132] The cross product is a fundamental operation in classical mathematics and physics, and is widely used in mechanics, rigid body kinematics, and engineering control. It is defined as follows: if there are two vectors A and B, then the result of A × B is a vector whose direction is perpendicular to the plane containing A and B.

[0133] S3 also includes S32;

[0134] S32. Based on the attitude disturbance, guide the three-dimensional vector PAV to output control commands and execute the hook head alignment attitude compensation operation;

[0135] The hook alignment attitude compensation operation uses attitude disturbance to guide the three-dimensional vector PAV to calculate the adjustment amount △Jd of the pushing direction angle and the adjustment amount △Sd of the pushing speed.

[0136] Among them, the adjustment amount △Jd of the pushing direction angle is obtained by comparing the attitude disturbance-guided three-dimensional vector PAV with the hook reference pushing direction vector Dref and using the vector angle formula, which is used to correct the pushing direction of the traction device.

[0137] The adjustment amount △Sd of the push speed is derived from the magnitude of the attitude perturbation-guided three-dimensional vector PAV and is used to adapt to the output speed of the push device.

[0138] During the hook alignment attitude compensation operation, the pushing device corrects the vehicle's propulsion direction according to the adjustment amount △Jd of the pushing direction angle, so that the hook turns from the offset state to the closing direction; at the same time, it finely adjusts the propulsion speed according to the adjustment amount △Sd of the pushing speed to avoid re-hooking impact or re-hooking failure due to pushing too fast or too slow; after the hook alignment attitude compensation operation is completed, the system recalculates the closing trend score value CTS. If the closing trend score value CTS is improved to the closing level I, the closing action is executed.

[0139] In this embodiment, the core design of S31 is to generate the attitude disturbance guidance three-dimensional vector PAV by cross-producting the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV. The physical significance of this is that RVV describes the linear trend of the hook's approach, while RAV reflects the hook's rotational instability. When there is a directional difference between the two, the cross-product result PAV can intuitively reveal this "misalignment direction," and its magnitude directly corresponds to the offset amplitude. In S32, the attitude disturbance guidance three-dimensional vector PAV is further used to calculate the push direction angle adjustment ΔJd and the push speed adjustment ΔSd to achieve precise compensation. ΔJd is obtained by comparing it with the re-hook reference push direction vector Dref, ensuring that the push direction always points to the closed optimal trajectory. Without this correction mechanism, even at the appropriate distance, the hook may misalign due to excessive angular deviation, leading to re-hook failure. ΔSd is calculated from the PAV magnitude and is used to control the push speed strength, its function being to avoid "over-push" or "under-push." For example, when the disturbance is large, the system reduces the pushing speed to minimize the risk of impact; when the disturbance is small and the direction has been corrected, it maintains or slightly increases the speed to accelerate the closing rhythm. Through this mechanism, the compensation process not only eliminates attitude deviations but also ensures the matching of pushing speed and direction. In a real physical sense, this is equivalent to providing "automatic navigation" for the hook in a dynamic environment, allowing it to enter the closed state stably and smoothly even in the presence of disturbances. The ultimate effect is to significantly improve the fault tolerance and stability of the re-hooking action, avoid impact damage caused by excessive speed or multiple attempts due to angular deviations, thereby improving the success rate of re-hooking and overall operational efficiency.

[0140] Example 5, please refer to Figure 1 Specifically: S4 includes S41;

[0141] S41. After the hook alignment attitude compensation operation is completed, within the time window from the start time t0 to the start time t0+0.3 seconds after the closing action is completed, continuous data of the longitudinal axis acceleration a(t) at time t within the 0.3-second time window are collected by a multi-axis velocity sensor, and the mean value a(t) of the continuous data of the longitudinal axis acceleration a is calculated. The maximum deviation between the longitudinal axis acceleration a and the mean value a(t) of the continuous data is calculated to obtain the acceleration stability score STS. The specific calculation formula for the acceleration stability score STS is as follows: ;

[0142] Then, a stability assessment is performed based on the Acceleration Stability Score (STS) to determine the stability of the hook re-hook after the hook alignment attitude compensation operation is completed; the specific assessment content is as follows:

[0143] When the acceleration stability score (STS) is less than 0.2 m / s² 2 At that time, the closing action is determined to be stable and successful;

[0144] When the acceleration stability score (STS) is greater than 0.5 m / s² 2 If the closure action is deemed abnormal, the closure is considered a failure.

[0145] When the acceleration stability score (STS) is 0.2 m / s² 2 up to 0.5m / s 2 If there is uncertainty in the determination of the closed state, a mechanism for re-collection and re-evaluation is triggered.

[0146] S4 also includes S42;

[0147] S42. When the stability assessment indicates a closure failure, the hook control server executes an action feedback process, the specific content of which is as follows:

[0148] The original standardized closure trend feature vector during the re-hook process and the acceleration stability score (STS) value at the time of failure are sent back to the initial acquisition node in order to reconstruct the closure trend score (CTS) value and the attitude disturbance guidance three-dimensional vector (PAV) and re-initiate the re-hook action.

[0149] If closure failure occurs in three consecutive judgment periods and is manifested as repeated attitude disturbances, the standardized closure trend feature vector of the closure failure is written into the disturbance feature database.

[0150] In the subsequent re-hooking process, the disturbance feature database is called as part of the trend feature database. The historical disturbance features are used to correct and iteratively optimize the scoring function, so that the Closed Trend Score (CTS) and Acceleration Stability Score (STS) can adaptively identify similar disturbance situations in the next judgment.

[0151] The update mechanism of the trend feature database ensures that the system has self-learning capabilities, enabling the double hook control strategy to gradually improve the recognition accuracy of complex working conditions and the success rate of closing actions in multiple runs.

[0152] In this embodiment, the design focus of method S41 is to calculate the acceleration stability score (STS) using longitudinal axis acceleration data within a 0.3-second time window. The 0.3-second timeframe is chosen because the impact and jitter during the closing of the hook action typically occur within a very short timescale. If the sampling time is too long, it will mask transient disturbances; if the sampling time is too short, it will not reflect the overall trend. By averaging the collected acceleration a(t) and then calculating the maximum deviation from the average, it is possible to effectively identify whether there are impact fluctuations during the closing process. For example, if the longitudinal axis acceleration fluctuates significantly after the hook closes, it indicates that the system has experienced a significant impact or turbulence, and the hook action is likely unstable. The physical significance of this step is to transform "whether the closure is stable" into a quantifiable dynamic indicator, avoiding the one-sided judgment based solely on position alignment. In S42, the action feedback mechanism ensures the system's self-learning and self-optimization capabilities. If the closure fails, the system not only immediately returns the feature vector and acceleration stability score (STS) value at the time of failure but also records the disturbance characteristics. This design aims to avoid repeating the same mistakes. For example, when the track surface is uneven or there are periodic disturbances, without a feature database, the system will repeatedly attempt and fail. However, by writing these failure patterns into a trend feature database, the system can directly lower the CTS score or trigger an additional compensation mechanism when encountering similar disturbances again, thereby avoiding risks in advance. Its physical significance lies in transforming repeated failures into subsequent empirical parameters, making the system increasingly "intelligent" with use. Therefore, the beneficial effects of this implementation process are: not only can it accurately identify whether the closure is stable after the double-coupling operation is completed, avoiding misjudgments that could lead to the hook loosening during train operation, but it also enables the system to dynamically adapt to complex environments through feedback and learning mechanisms. Ultimately, this improves the stability, reliability, and intelligence level of the double-coupling operation, reduces the need for manual intervention, and significantly reduces safety hazards caused by unstable closure.

[0153] Example 6, please refer to Figure 1 and Figure 2 An automatic train recoupling control system based on positioning recognition includes a multi-point acquisition module, a closure trend analysis module, a hook alignment compensation module, and a stability optimization module.

[0154] The multi-point acquisition module collects sensing data in real time by deploying multi-point acquisition devices around the hooks of adjacent cars to be re-coupled; the collected sensing data is transmitted wirelessly to the re-coupling control server in real time, and the data is preprocessed in the re-coupling control server to construct a standardized closed trend feature vector.

[0155] The Closure Trend Analysis module calculates the Closure Trend Score (CTS) based on the standardized Closure Trend Feature Vector, and then classifies the current hook state into Closure Level I, Compensation Level II, and Blockage Level III based on the Closure Trend Score (CTS).

[0156] The hook alignment compensation module triggers the attitude disturbance compensation control mechanism when the closure trend score value (CTS) is at compensation level II. It calculates the attitude disturbance guidance three-dimensional vector (PAV), outputs control commands based on the attitude disturbance guidance three-dimensional vector (PAV) to fine-tune the pushing direction of the traction device, and performs the hook alignment attitude compensation operation.

[0157] The stability optimization module collects the longitudinal axis acceleration 'a' during the closing process after the hook alignment posture compensation operation is completed, calculates the motion stability score (STS), and performs a stability assessment.

[0158] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for automatic recoupling control of trains based on positioning recognition, characterized in that: Includes the following steps: S1. Deploy multiple data acquisition devices around the hooks of adjacent cars to be re-coupled to collect sensing data in real time; transmit the collected sensing data to the re-coupling control server wirelessly in real time, and preprocess the data in the re-coupling control server to construct a standardized closed trend feature vector. S2. Calculate the Closure Trend Score (CTS) based on the standardized closure trend feature vector, and then classify the current re-hook state into closure level I, compensation level II, and blocking level III based on the closure trend score (CTS). S3. When the Closed Trend Score (CTS) is at Compensation Level II, the Attitude Disturbance Compensation Control Mechanism is triggered. The Attitude Disturbance Guiding Three-Dimensional Vector (PAV) is calculated. Based on the Attitude Disturbance Guiding Three-Dimensional Vector (PAV), the control command is output to fine-tune the pushing direction of the traction device and perform the hook alignment attitude compensation operation. S4. After the hook alignment posture compensation operation is completed, the longitudinal axis acceleration a during the closing process is collected, the motion stability score STS is calculated, and the stability is evaluated.

2. The automatic train recoupling control method based on positioning recognition according to claim 1, characterized in that: S1 includes S11; S11. The multi-point acquisition device includes a laser rangefinder sensor deployed at the front end of the coupler of adjacent car bodies, a multi-axis velocity sensor deployed on the side wall of the end of the car body, and an inertial measurement unit (IMU) installed on the coupler seat; it acquires the sensing data of the train coupler in real time. The sensing data includes the hook-head spacing change rate HVR, the three-dimensional approach velocity vector RVV, and the three-dimensional attitude angular velocity vector RAV; The laser rangefinder sensor monitors the hook head spacing in real time by deploying a laser rangefinder sensor at the front end of the hook head, and calculates the hook head spacing change rate HVR through continuous time difference calculation. The multi-axis velocity sensor is used to collect relative velocity components in real time along the longitudinal, transverse and vertical axes of the vehicle body, forming a three-dimensional approach velocity vector RVV between adjacent vehicle bodies; The inertial measurement unit (IMU) is used to acquire the three-dimensional vector (RAV) of the vehicle's attitude angular velocity in the pitch, yaw, and roll directions in real time. The multi-point acquisition device performs collaborative acquisition according to a synchronous triggering sequence.

3. The automatic train recoupling control method based on positioning recognition according to claim 2, characterized in that: S1 also includes S12 and S13; S12. The multi-point acquisition device is connected to the hook control server via a data transmission module. The data transmission module includes an edge computing node, a temporary cache unit and a main station communication interface. Among them: the edge computing node is located inside the vehicle body and is used to compress and encode the sensing data collected by the multi-point acquisition device; Temporary buffer units provide redundancy compensation for abnormal packet loss; The main station communication interface uploads the compressed data in real time to the hook control server located in the control center in the form of protocol-based data frames; S13. In the double hook control server, the received sensing data is preprocessed to obtain a standardized closed trend feature vector; The preprocessing includes time alignment, noise filtering, and feature normalization. The time alignment process calibrates the timestamps of each acquisition device by configuring a unified clock synchronization protocol, so that all sensing data have a unified time reference benchmark; and constructs a time sliding buffer queue based on the sampling time window, which automatically aligns the sensing data according to the time axis after it is received. The noise removal process employs a two-stage filtering mechanism to denoise the original signal. Specifically, a 3-point sliding window mid-range filter is used to remove short-time pulses from the hook spacing change rate (HVR) and the approach velocity 3D vector (RVV), eliminating spike anomalies caused by instantaneous track surface disturbances and encoder jumps. A 5-point Gaussian weighted filter is used to perform smooth convolution processing on the attitude angular velocity 3D vector (RAV). Among them, the weight coefficients of the five-point Gaussian weight are {0.0625, 0.25, 0.375, 0.25, 0.0625}; The feature normalization process involves normalizing the noise-filtered sensing data according to the historical operating conditions range. Each parameter in the sensing data is normalized using the extreme value normalization method. The normalized result is a dimensionless value, which is then combined into a standardized closed-trend feature vector.

4. The automatic train recoupling control method based on positioning recognition according to claim 3, characterized in that: S2 includes S21; S21. A closed-loop trend scoring calculation model is pre-constructed in the hook control server. The closed-loop trend scoring calculation model consists of an input layer, a feature operation layer, and an output layer, wherein: The input layer is used to receive the standardized closed-trend feature vector in real time; the transmitted hook spacing change rate HVR, approach velocity three-dimensional vector RVV, and attitude angular velocity three-dimensional vector RAV; The feature operation layer is calculated according to the following rules: First, the absolute value of the hook spacing change rate HVR is introduced into the logarithmic function for nonlinear amplification to optimize the trend recognition sensitivity when the hook approaches; then, the magnitude of the approach velocity three-dimensional vector RVV is used as a factor of trend intensity to quantify the closed potential energy of the relative motion of the vehicle body; at the same time, the magnitude of the attitude angular velocity three-dimensional vector RAV is placed as a perturbation factor in the denominator to suppress the trend score under unstable attitude conditions. The output layer generates a Closed Trend Score (CTS) based on the output results of the feature operation layer, and comprehensively analyzes the approach trend and attitude stability of adjacent car body hooks during the re-coupling process.

5. The automatic train recoupling control method based on positioning recognition according to claim 4, characterized in that: S2 further includes S22; S22. Set a closure interval threshold based on historical data, wherein the closure interval threshold includes a closure confirmation threshold F1 and a closure validity lower limit threshold F2; The Closure Trend Score (CTS) value output in real time by the Closure Trend Score Calculation Model is compared with the Closure Interval Threshold to classify the current complex hook state into Closure Level I, Compensation Level II, and Blockage Level III; the specific comparison is as follows: When the closing trend score CTS is greater than or equal to the closing confirmation threshold F1, it is classified as closing level I, and it is determined that there is a stable and effective closing trend between the adjacent car body hooks. The system immediately issues a closing execution command. When the lower limit threshold of closure effectiveness F2 ≤ closure trend score CTS < closure confirmation threshold F1, it is classified as compensation level II, closure is suspended, and attitude disturbance compensation mechanism is triggered. When the Closure Trend Score (CTS) is less than the lower limit of closure effectiveness (F2), it is classified as Blockage Level III, and the next cycle is initiated for re-collection and judgment. If the Blockage Level III is maintained for three consecutive cycles, an early warning message is sent to the operation interface of the re-hook control server.

6. The automatic train recoupling control method based on positioning recognition according to claim 5, characterized in that: S3 includes S31; S31. After triggering the attitude disturbance compensation mechanism, the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV are input into the compensation calculation model of the compound hook control server. In the compensation calculation model, a vector cross product operation is performed on the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV to obtain the disturbance direction relationship between the approach velocity three-dimensional vector RVV and the attitude angular velocity three-dimensional vector RAV. The calculation result is defined as the attitude disturbance guidance three-dimensional vector PAV. Then, the output attitude disturbance guidance three-dimensional vector PAV is stored in the compound hook control server in the form of a three-dimensional vector (PAVx, PAVy, PAVz).

7. The automatic train recoupling control method based on positioning recognition according to claim 6, characterized in that: S3 further includes S32; S32. Based on the attitude disturbance, guide the three-dimensional vector PAV to output control commands and execute the hook head alignment attitude compensation operation; The hook alignment attitude compensation operation calculates the adjustment amount △Jd of the pushing direction angle and the adjustment amount △Sd of the pushing speed by using the attitude disturbance-guided three-dimensional vector PAV. Among them, the adjustment amount △Jd of the pushing direction angle is obtained by comparing the attitude disturbance-guided three-dimensional vector PAV with the hook reference pushing direction vector Dref and using the vector angle formula; The adjustment amount △Sd of the push speed is derived from the magnitude of the attitude perturbation-guided three-dimensional vector PAV; During the hook alignment attitude compensation operation, the pushing device corrects the vehicle's propulsion direction according to the adjustment amount △Jd of the pushing direction angle, so that the hook turns from the offset state to the closing direction; at the same time, it finely adjusts the propulsion speed according to the adjustment amount △Sd of the pushing speed; after the hook alignment attitude compensation operation is completed, the system recalculates the closing trend score value CTS. If the closing trend score value CTS is improved to the closing level I, the closing action is executed.

8. The automatic train recoupling control method based on positioning recognition according to claim 6, characterized in that: S4 includes S41; S41. After the hook alignment posture compensation operation is completed, within the time window from the start time t0 to the start time t0+0.3 seconds after the closing action is completed, the continuous data of the longitudinal axis acceleration a(t) at time t within the 0.3-second time window are collected by the multi-axis velocity sensor, and the mean value aj of the continuous data of the longitudinal axis acceleration a is calculated. Then, the maximum deviation value between the longitudinal axis acceleration a and the mean value aj of the continuous data is calculated to obtain the acceleration stability score STS. Then, a stability assessment is performed based on the Acceleration Stability Score (STS) to determine the stability of the hook re-hook after the hook alignment attitude compensation operation is completed; the specific assessment content is as follows: When the acceleration stability score (STS) is less than 0.2 m / s² 2 At that time, the closing action is determined to be stable and successful; When the acceleration stability score (STS) is greater than 0.5 m / s² 2 If the closure action is deemed abnormal, the closure is considered a failure. When the acceleration stability score (STS) is 0.2 m / s² 2 up to 0.5m / s 2 If there is uncertainty in the determination of the closed state, a mechanism for re-collection and re-evaluation is triggered.

9. The automatic train recoupling control method based on positioning recognition according to claim 8, characterized in that: S4 also includes S42; S42. When the stability assessment indicates that the closure has failed, the hook control server executes an action feedback process, the specific content of which is as follows: The original standardized closure trend feature vector during the re-hook process and the acceleration stability score (STS) value at the time of failure are sent back to the initial acquisition node in order to reconstruct the closure trend score (CTS) value and the attitude disturbance guidance three-dimensional vector (PAV) and re-initiate the re-hook action. If closure failure occurs in three consecutive judgment periods and is manifested as repeated attitude disturbances, the standardized closure trend feature vector of the closure failure is written into the disturbance feature database. In the subsequent re-hooking process, the disturbance feature database is called as part of the trend feature database. The scoring function is corrected and iteratively optimized using historical disturbance features, so that the Closed Trend Score (CTS) and Acceleration Stability Score (STS) can adaptively identify similar disturbance situations in the next judgment.

10. A train automatic recoupling control system based on positioning recognition, applied to the train automatic recoupling control method based on positioning recognition as described in any one of claims 1-9, characterized in that: It includes a multi-point acquisition module, a closure trend analysis module, a hook alignment compensation module, and a stability optimization module; The multi-point acquisition module collects sensing data in real time by deploying multi-point acquisition devices around the hooks of adjacent cars to be re-coupled; the collected sensing data is transmitted wirelessly to the re-coupling control server in real time, and the data is preprocessed in the re-coupling control server to construct a standardized closed trend feature vector. The closure trend analysis module calculates the closure trend score (CTS) based on the standardized closure trend feature vector, and then classifies the current re-hook state into closure level I, compensation level II, and blocking level III based on the closure trend score (CTS). The hook alignment compensation module triggers the attitude disturbance compensation control mechanism when the closure trend score value CTS is at compensation level II, calculates the attitude disturbance guidance three-dimensional vector PAV, outputs control commands based on the attitude disturbance guidance three-dimensional vector PAV to fine-tune the pushing direction of the traction device, and performs hook alignment attitude compensation operation. The stability optimization module collects the longitudinal axis acceleration 'a' during the closing process after the hook alignment posture compensation operation is completed, calculates the motion stability score (STS), and performs a stability assessment.

Images