An AI-based collision mitigation braking method for buses

Abstract

This invention discloses an artificial intelligence-based collision mitigation braking method for buses, belonging to the field of collision mitigation technology. The method includes: collecting the forward approach state of the bus and the longitudinal response state of the passenger compartment; performing difference correction to form target approach data and passenger compartment sway data; determining the collision mitigation demand intensity based on the target approach data; extracting the main vibration phase and main vibration period based on the passenger compartment sway data; calculating the phase difference of the passenger compartment's overall sway and the consistency of the longitudinal sway direction; inputting these values ​​into a passenger compartment inertial phase recognition model to generate the passenger compartment inertial phase state; constructing a forward tilt peak avoidance window under the forward tilt phase state and calculating the phase peak acceptance criterion; determining the braking peak release acceptance phase under the swing phase state, generating a unique braking release curve; and executing bus collision mitigation braking through the unique braking release curve. This invention enables the braking peak to avoid the forward tilt peak, reducing the risk caused by the simultaneous superposition of passenger forward tilt inertia and braking impact.

Description

Technical Field

[0001] This invention relates to the field of collision mitigation technology, and in particular to an artificial intelligence-based collision mitigation braking method for buses. Background Technology

[0002] Collision mitigation braking technology for public transport vehicles primarily targets operational scenarios such as following other vehicles on urban roads, entering and exiting bus stops, navigating intersections, and mixing with non-motorized vehicles. It uses forward-facing sensing devices to acquire the approach status of a target and combines this with vehicle speed, braking response characteristics, and the relative motion of the target to determine whether to initiate a collision mitigation control process. Conventional methods typically generate braking requests based on target approach time, relative speed, or a safe distance threshold. The vehicle's brake controller then executes brake pressure build-up, target deceleration tracking, and brake force release to reduce the vehicle's contact speed in a potential collision. This represents an important technological direction in vehicle driving safety control.

[0003] In the context of public transportation vehicles, conventional collision mitigation braking methods mainly focus on the matching relationship between the risk of approaching external targets and the vehicle's braking capacity, without giving sufficient consideration to the relationship between the longitudinal sway phase of the passenger compartment and the inertial bearing state of standing passengers. At the same time, the peak braking release time usually advances or intensifies as the collision risk increases, lacking a phase staggered control mechanism that links the forward tilting phase and the swaying phase of the passenger compartment with the peak braking release process, making it difficult to simultaneously reduce collision speed and suppress secondary impacts inside the passenger compartment. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based bus collision mitigation braking method to solve the problems of existing technologies, such as the difficulty in avoiding the peak inertia of the forward tilt of the carriage during the braking peak release time and insufficient utilization of the carriage swing phase.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an artificial intelligence-based collision mitigation braking method for buses, comprising: collecting the forward target approach state and the longitudinal response state of the bus body, and performing difference correction on the longitudinal response state of the bus body to form target approach data and front-to-back sway data of the bus body; determining the collision mitigation demand intensity based on the target approach data, and extracting the main vibration phase and main vibration period of the bus body from the front-to-back sway data, calculating the phase difference of the bus body's overall sway and the consistency of the longitudinal sway direction of the bus body, and inputting them into the bus body inertial phase recognition model to generate the bus body inertial phase state; when the bus body... When the inertial phase state is in the forward tilt phase, a forward tilt peak avoidance window is constructed based on the main vibration period of the front and rear of the carriage. The phase peak acceptance criterion is calculated by considering the collision mitigation demand intensity, the forward tilt peak avoidance window, the phase difference of the carriage group sway, and the consistency of the longitudinal sway direction of the carriage. It is then determined whether the braking peak can be released after avoiding the forward tilt peak of the carriage. When the inertial phase state of the carriage is in the swing phase, the swing phase is taken as the acceptance phase for allowing the release of the braking peak, and a unique braking release curve is generated. The bus collision mitigation braking is performed and a closed-loop update is performed through the unique braking release curve.

[0008] As a preferred embodiment of the AI-based bus collision mitigation braking method of the present invention, the formation of target approach data and front-to-back sway data of the bus includes: collecting the forward target distance and the forward target relative velocity, and using the forward target distance and the forward target relative velocity as target approach data; collecting the longitudinal response quantity at the front and rear of the bus to form the front-to-back longitudinal response state of the bus; reading the longitudinal motion state of the bus, and using the current longitudinal acceleration of the bus in the longitudinal motion state as the common mode motion quantity, performing difference correction on the longitudinal response quantity at the front and rear of the bus to obtain the longitudinal sway quantity at the front and rear of the bus; performing short-time window processing on the longitudinal sway quantity at the front and rear of the bus to form the longitudinal sway sequence at the front and rear of the bus; and using the longitudinal sway sequence at the front and rear of the bus as the front-to-back sway data of the bus.

[0009] As a preferred embodiment of the AI-based bus collision mitigation braking method of the present invention, determining the collision mitigation demand intensity includes: determining a safe reserve distance based on the bus's current speed, braking pressure build-up time, and front-end safety margin; determining the target approach time based on the forward target distance, the forward target relative speed, the safe reserve distance, and the minimum approach speed protection value; when the target approach time is greater than the safe approach time boundary, the collision mitigation demand intensity is determined as a soft start demand; when the target approach time is less than or equal to the safe approach time boundary and greater than the strong deceleration time boundary, the collision mitigation demand intensity is determined as a continuous deceleration demand; when the target approach time is less than or equal to the strong deceleration time boundary, the collision mitigation demand intensity is determined as an immediate strong deceleration demand.

[0010] As a preferred embodiment of the artificial intelligence-based bus collision mitigation braking method of the present invention, the extraction of the main vibration phase and main vibration period of the front and rear of the bus includes: performing the same main vibration extraction process on the longitudinal sway sequence of the front of the bus and the longitudinal sway sequence of the rear of the bus respectively; the main vibration extraction process includes identifying adjacent zero-crossing points and local peak points of the longitudinal sway in the corresponding longitudinal sway sequence; when the corresponding longitudinal sway sequence has no less than two adjacent zero-crossing points in the same direction and no less than one local peak point within a short time window, determining the main vibration period of the corresponding position according to the time interval between the two adjacent zero-crossing points in the same direction, and determining the main vibration phase of the corresponding position according to the relative time position between the two adjacent zero-crossing points in the same direction at the current sampling time; the corresponding position includes the front of the bus and the rear of the bus; the corresponding longitudinal sway sequence includes the longitudinal sway sequence of the front of the bus and the longitudinal sway sequence of the rear of the bus.

[0011] As a preferred embodiment of the artificial intelligence-based bus collision mitigation braking method of the present invention, the calculation of the phase difference of the swaying of the entire carriage and the consistency of the longitudinal swaying direction of the carriage includes: performing phase reflection processing based on the phase difference between the main vibration phase of the front of the carriage and the main vibration phase of the rear of the carriage to calculate the phase difference of the swaying of the entire carriage; calculating the consistency of the longitudinal swaying direction of the carriage based on the longitudinal swaying amount of the front and rear of the carriage and the effective amplitude boundary of the swaying; when the absolute values ​​of the longitudinal swaying amount of the front and rear of the carriage are both less than the effective amplitude boundary of the swaying, the front and rear swaying amplitude state of the carriage is determined to be a low amplitude state; when at least one of the absolute values ​​of the longitudinal swaying amount of the front and rear of the carriage is not less than the effective amplitude boundary of the swaying, the front and rear swaying amplitude state of the carriage is determined to be an effective swaying state.

[0012] As a preferred embodiment of the AI-based bus collision mitigation braking method of the present invention, the inertial phase recognition model of the bus includes an input normalization layer, an amplitude-phase embedding layer, a short-time gating layer, and a state discrimination layer. The input normalization layer includes scaling the phase difference of the bus group sway, the consistency of the longitudinal sway direction of the bus, the main vibration period of the front of the bus, the main vibration period of the rear of the bus, and the amplitude state of the front and rear sway. The amplitude-phase embedding layer includes establishing the correspondence between the amplitude state of the front and rear sway, the phase difference of the bus group sway, and the consistency of the longitudinal sway direction of the bus. The short-time gating layer includes reading the amplitude-phase embedding data within a fixed number of consecutive braking control cycles to suppress state jumps caused by fluctuations at a single sampling point. The state discrimination layer includes outputting a forward tilt in-phase state, a sway in-phase state, a phase discrete state, or a low-amplitude stable state as the inertial phase state of the bus.

[0013] As a preferred embodiment of the AI-based bus collision mitigation braking method of the present invention, the step of constructing a forward tilting peak avoidance window based on the main vibration periods of the front and rear of the bus when the inertial phase state of the bus is in a forward tilting phase state is as follows: when the inertial phase state of the bus is in a forward tilting phase state, the current braking control moment is determined as the starting point of the forward tilting peak avoidance window; the length of the forward tilting peak avoidance window is determined based on the main vibration periods of the front and rear of the bus and the forward tilting peak avoidance ratio; the ending point of the forward tilting peak avoidance window is determined based on the starting point and the length of the forward tilting peak avoidance window; when the inertial phase state of the bus is in a swinging phase state, a phase discrete state, or a low-amplitude stable state, the length of the forward tilting peak avoidance window is set to zero.

[0014] As a preferred embodiment of the AI-based bus collision mitigation braking method described in this invention, the phase staggering acceptance criterion includes: determining the forward tilting in-phase suppression amount based on the phase difference of the carriage group swaying and the consistency of the carriage's longitudinal swaying direction; calculating the phase staggering acceptance criterion based on the target approach time, braking pressure build-up time, the length of the carriage's forward tilting peak avoidance window, and the forward tilting in-phase suppression amount; when the phase staggering acceptance criterion is greater than or equal to the staggering acceptance boundary, it is determined to be a staggering acceptance state; when the phase staggering acceptance criterion is less than the staggering acceptance boundary but greater than the forced release boundary, it is determined to be a limited staggering state; when the phase staggering acceptance criterion is less than or equal to the forced release boundary, it is determined to be a non-staggering state; the staggering acceptance state, the limited staggering state, and the non-staggering state are taken as the phase staggering acceptance states.

[0015] As a preferred embodiment of the AI-based bus collision mitigation braking method of the present invention, the generation of a unique brake release curve includes: when the collision mitigation demand intensity is a slow start demand, determining the target deceleration peak as a low-intensity target deceleration peak, setting the target deceleration peak release time as the slow start release time after brake pressure build-up is completed, and outputting the target deceleration according to the slow start incremental rate, so that the target deceleration increases to the low-intensity target deceleration peak; when the phase staggered peak acceptance state is a staggered peak acceptance state, the carriage inertial phase state is a forward tilting phase state, and the collision mitigation demand intensity is a continuous deceleration demand, setting the target deceleration peak release time after the end of the carriage forward tilting peak avoidance window, and outputting the target deceleration according to the low-intensity incremental rate within the carriage forward tilting peak avoidance window; when the phase staggered peak acceptance state is... When the phase shift is limited and the inertial phase of the carriage is in a forward-leaning phase, and the collision mitigation demand intensity is a continuous deceleration demand, the target deceleration peak value is determined as the limited target deceleration peak value, and the target deceleration is output according to the limited incremental rate. When the phase shift is not allowed, or the collision mitigation demand intensity is an immediate strong deceleration demand, the waiting constraint on the release time of the target deceleration peak value is released, and the release time of the target deceleration peak value is set as the brake pressure build-up completion time. When the inertial phase of the carriage is in a swing phase, the swing phase is used as the phase for allowing the release of the brake peak value, and the target deceleration peak value is released after the brake pressure build-up is completed. A unique brake release curve is generated based on the brake pressure build-up start point, the target deceleration incremental rate, the target deceleration peak value release time, the target deceleration peak value, and the target deceleration retraction rate.

[0016] As a preferred embodiment of the AI-based bus collision mitigation braking method of the present invention, the specific steps of performing bus collision mitigation braking and closed-loop updating through a unique brake release curve are as follows: Performing bus collision mitigation braking according to the unique brake release curve; rereading target approach data and vehicle sway data in each braking control cycle, performing closed-loop updates, and obtaining the updated vehicle inertial phase state and phase peak acceptance state; adjusting the target deceleration peak release time, target deceleration peak value, and target deceleration change rate according to the updated collision mitigation demand intensity, vehicle inertial phase state, and phase peak acceptance state; when the target approach risk is eliminated, entering the braking force retreat phase, and gradually reducing the target deceleration to zero according to the target deceleration retreat rate.

[0017] The beneficial effects of this invention are as follows: By generating the inertial phase state of the carriage, intelligent differentiation is achieved among forward tilting phase state, swinging phase state, phase discrete state, and low-amplitude stable state, enabling bus braking control to simultaneously consider external collision risks and passenger inertial bearing state within the carriage; through the carriage forward tilting peak avoidance window and phase staggered peak bearing criterion, the determination of whether the braking peak is released after avoiding the carriage forward tilting peak is realized, reducing the risk of falls caused by the simultaneous superposition of braking peak and passenger forward tilting inertial peak; by generating a unique braking release curve, layered control of the braking peak release time and target deceleration increase process is achieved, reducing secondary impacts within the carriage while maintaining collision speed reduction capability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of an AI-based bus collision mitigation braking method.

[0020] Figure 2 A schematic diagram generated from the target proximity data and the inertial phase state of the carriage.

[0021] Figure 3 This is a schematic diagram of the phase stagger criterion that can be used to generate the criteria.

[0022] Figure 4 This is a schematic diagram illustrating the execution and closed-loop update of the unique brake release curve. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides an artificial intelligence-based bus collision mitigation braking method, comprising the following steps:

[0027] S1. Collect the forward target approach status and the longitudinal response status of the bus carriage, and perform differential correction on the longitudinal response status of the bus carriage to form target approach data and front and rear sway data of the bus carriage.

[0028] A forward-facing millimeter-wave radar is installed at the front of the bus, with its detection direction aligned with the longitudinal centerline of the bus, to collect data on the approach of forward targets. A first longitudinal inertial measurement unit (IMU) is installed on the floor frame near the front wheel axle at the front of the passenger compartment, and a second longitudinal IMU is installed on the floor frame near the rear wheel axle at the rear of the passenger compartment to collect data on the longitudinal response of the passenger compartment. The longitudinal motion state of the bus is then read through the vehicle chassis controller.

[0029] Among them, the forward target approach state includes the forward target distance and the forward target relative speed; the bus longitudinal motion state includes the bus current speed and the bus current longitudinal acceleration; and the front and rear longitudinal response state of the carriage includes the front longitudinal response amount and the rear longitudinal response amount of the carriage.

[0030] It should be noted that the forward target distance represents the distance between the front of the bus and the target in the longitudinal direction of the vehicle, in meters; the forward target relative speed represents the longitudinal approach speed of the target relative to the bus, in meters per second, with a positive value when the distance between the target and the bus decreases; the bus current speed represents the speed of the bus in the current braking control cycle (e.g., 20ms), in meters per second; the bus current longitudinal acceleration represents the acceleration of the bus as a whole in the longitudinal direction, in meters per second squared; the longitudinal response of the front of the carriage represents the response of the front of the carriage in the longitudinal direction of the vehicle collected by the first longitudinal inertial measurement unit, in meters per second squared; and the longitudinal response of the rear of the carriage represents the response of the rear of the carriage in the longitudinal direction of the vehicle collected by the second longitudinal inertial measurement unit, in meters per second squared.

[0031] Furthermore, to ensure that the forward target approach status, the longitudinal motion status of the bus, and the longitudinal response status of the front and rear of the carriage can be judged within the same braking control cycle, the vehicle braking controller reads the data uploaded by the forward millimeter-wave radar, the vehicle chassis controller, the first longitudinal inertial measurement unit, and the second longitudinal inertial measurement unit at a fixed control cycle, and performs time alignment on the data within the same control cycle using the timestamp of the vehicle chassis controller as a unified time reference; when the forward millimeter-wave radar does not form a stable target point within two consecutive control cycles, the forward target approach status of the previous control cycle is maintained and marked as a pending confirmation status; when the forward target distance continuously decreases and the relative velocity of the forward target remains positive, the forward target approach status is marked as a valid approach status.

[0032] It should be noted that a stable target point is defined as a forward target distance and relative velocity that are output in two consecutive braking control cycles, with the direction of distance change being consistent with the sign of the relative velocity.

[0033] Furthermore, to avoid misinterpreting the overall acceleration and deceleration of the bus as localized inertial swaying of the passenger compartment, the current longitudinal acceleration of the bus is used as the common mode motion quantity. A difference correction is applied to the longitudinal response quantities at the front and rear of the passenger compartment to generate the longitudinal swaying quantities at the front and rear of the passenger compartment, expressed as follows:

[0034] ;

[0035] ;

[0036] in, This indicates the amount of longitudinal sway at the front of the carriage. This indicates the amount of longitudinal sway at the rear of the carriage. This indicates the longitudinal response at the front of the carriage. This indicates the longitudinal response at the rear of the carriage. This indicates the current longitudinal acceleration of the bus.

[0037] It should be noted that, , as well as The unit is meters per second squared.

[0038] It should be noted that the longitudinal sway at the front and rear of the carriage is not used to identify passenger identity or posture, but rather to characterize the local longitudinal dynamic response formed by the combined effects of passenger inertial load, suspension response, and vehicle body structural flexibility at the front and rear of the carriage. After differential correction using the current longitudinal acceleration of the bus, the interference of the vehicle's acceleration and deceleration on the judgment of local sway at the front and rear of the carriage can be reduced, making the phase difference of the subsequent carriage group sway and the consistency of the longitudinal sway direction closer to the inertial response state inside the carriage.

[0039] Furthermore, short-term window adjustments are made to monitor the longitudinal sway at the front and rear of the carriage.

[0040] The short-time window length is 8 to 12 times the braking control cycle of the bus. Within the short-time window, abrupt sampling points that exceed the range boundary of the inertial measurement unit are removed, and the remaining sampling points are arranged in chronological order to form the longitudinal sway sequence at the front of the carriage and the longitudinal sway sequence at the rear of the carriage.

[0041] It should be noted that the braking control cycle of a bus is usually used to generate a braking command update. If the short window length is less than 8 braking control cycles, the number of samples that can be used to identify zero-crossing points and local peak points in the longitudinal sway sequence of the front and rear of the carriage is insufficient, which may easily misjudge the minor vibrations of the road surface as inertial sway of the carriage. If the short window length is greater than 12 braking control cycles, the inertial phase state of the carriage will lag behind the current target approach state, affecting the real-time performance of the subsequent carriage forward tilt peak avoidance window.

[0042] S2. Determine the collision mitigation demand intensity based on target proximity data, and extract the main vibration phase and main vibration period of the front and rear of the carriage based on the front and rear sway data of the carriage. Calculate the phase difference of the carriage group sway and the consistency of the longitudinal sway direction of the carriage, and input them into the carriage inertial phase recognition model to generate the carriage inertial phase state.

[0043] The vehicle braking controller reads the forward target distance, the forward target relative speed, the current bus speed, the longitudinal sway sequence at the front of the bus, and the longitudinal sway sequence at the rear of the bus.

[0044] Furthermore, the vehicle's brake controller determines the safe holding distance based on the bus's current speed and brake pressure build-up time, expressed as:

[0045] ;

[0046] in, Indicates maintaining a safe distance. Indicates the current speed of the bus. Indicates the braking pressure build-up time. This indicates the safety margin at the front of the vehicle.

[0047] It should be noted that the safe distance is in meters; the current speed of the bus is in meters per second; the braking pressure build-up time is in seconds, obtained by recording the time interval between the moment the braking command is issued and the moment the effective deceleration is formed during the calibration phase of the bus on the brake test bench; the front safety margin is in meters, determined by the outer contour of the front structure of the bus, the installation position of the forward millimeter-wave radar, and the minimum longitudinal interval allowed to be maintained during the low-speed close-up test of the vehicle, with a value range of 0.5m to 1.5m.

[0048] Furthermore, the target approach time is calculated based on the forward target distance, the forward target relative velocity, and the safe maintenance distance, expressed as:

[0049] ;

[0050] in, Indicates the time approaching the target. Indicates the distance to the target ahead. Indicates the relative velocity of the forward target. This indicates the minimum approach speed protection value.

[0051] It should be noted that the minimum approach speed protection value is in meters per second. It is obtained by using a forward millimeter-wave radar to stably resolve the minimum longitudinal speed during low-speed approach calibration, and the value ranges from 0.1 m / s to 0.3 m / s.

[0052] Furthermore, the intensity of collision mitigation requirements is determined based on the target approach time.

[0053] Specifically, when the target approach time is greater than the safe approach time boundary, the collision mitigation requirement intensity is determined as a soft start requirement; when the target approach time is less than or equal to the safe approach time boundary and the target approach time is greater than the strong deceleration time boundary, the collision mitigation requirement intensity is determined as a continuous deceleration requirement; when the target approach time is less than or equal to the strong deceleration time boundary, the collision mitigation requirement intensity is determined as an immediate strong deceleration requirement.

[0054] It should be noted that the safe approach time boundary is obtained by statistically analyzing the historical records of the time required for the same vehicle model to complete the pre-braking pressure build-up and low-intensity deceleration within the current speed range, with a value range of 1.2 seconds to 2.5 seconds; the strong deceleration time boundary is obtained by statistically analyzing the time required for the same vehicle model to complete the forced braking pressure build-up and minimum collision speed reduction within the current speed range, with a value range of 0.4 seconds to 1.2 seconds.

[0055] Furthermore, the main vibration extraction was performed on the longitudinal sway sequence at the front of the carriage and the longitudinal sway sequence at the rear of the carriage, respectively.

[0056] Specifically, the vehicle brake controller reads adjacent zero-crossing points and local peak points of the longitudinal sway sequence at the front of the carriage within a short time window.

[0057] Among them, the zero-crossing point is the sampling position where the sign of the longitudinal sway at the front of the carriage changes from positive to negative or from negative to positive. The local peak point is the sampling position where the absolute value of the longitudinal sway at the front of the carriage is greater than the absolute value of the previous sampling value and not less than the absolute value of the next sampling value in adjacent sampling intervals. The lower limit of the number of zero-crossing points is set to two because a single zero-crossing point can only represent one change in the direction of the longitudinal sway at the front of the carriage, and cannot form a time interval that can be used to estimate the vibration period. Two adjacent zero-crossing points in the same direction can correspond to the completion of a complete reciprocating cycle of the longitudinal sway at the front of the carriage, and can determine the main vibration cycle of the front of the carriage. The lower limit for the number of local peak points is set to one because at least one position with maximum amplitude is required within a complete cyclic period to confirm the existence of effective inertial swaying rather than low-amplitude noise crossing zero within the short-time window. Therefore, when the longitudinal swaying sequence at the front of the carriage has at least two adjacent zero-crossing points in the same direction and at least one local peak point within the short-time window, the main vibration period of the front of the carriage is determined based on the time interval between the two adjacent zero-crossing points in the same direction, and the phase of the main vibration of the front of the carriage is determined based on the relative time position of the current sampling time between the two adjacent zero-crossing points in the same direction.

[0058] Specifically, the vehicle brake controller reads the zero-crossing point and local peak point of the longitudinal sway sequence at the rear of the carriage within the same short time window.

[0059] The longitudinal swaying sequence at the rear of the carriage adopts the same zero-crossing point identification rules and local peak point identification rules as the longitudinal swaying sequence at the front of the carriage. The lower limit for the number of zero-crossing points is also set to two, and the lower limit for the number of local peak points is also set to one. This is because the main vibration period and main vibration phase of the front and rear of the carriage need to be extracted under the same judgment scale to ensure that the phase difference of the subsequent carriage group swaying is comparable. Therefore, when the longitudinal swaying sequence at the rear of the carriage has no less than two adjacent zero-crossing points in the same direction and no less than one local peak point within a short time window, the main vibration period of the rear of the carriage is determined according to the time interval between the two adjacent zero-crossing points in the same direction, and the main vibration phase of the rear of the carriage is determined according to the relative time position of the current sampling time between the two adjacent zero-crossing points in the same direction.

[0060] Furthermore, when the number of zero-crossing points or local peak points within the short-time window is insufficient, forced phase extrapolation is not performed; instead, a low-amplitude stable state or a phase-discrete state is generated by the car's inertial phase recognition model. When the number of zero-crossing points and local peak points within the short-time window meets the requirements for extracting the main vibration, the phase difference of the car group's swaying is calculated, expressed as:

[0061] ;

[0062] in, This indicates the phase difference of the swaying of the carriages. Indicates the phase of the main vibration at the front of the carriage. Indicates the phase of the main vibration at the rear of the carriage. This indicates that the phase difference is reflected back to... Phase reversal processing within the interval.

[0063] It should be noted that the phase difference of the swaying of the carriages It is an angular quantity, and its value range is [0, ...]. ].

[0064] Furthermore, the amount of longitudinal sway in the carriage is calculated in the same direction, expressed as:

[0065] ;

[0066] in, This indicates the amount of longitudinal sway in the carriage, which is in the same direction. Represents a symbolic function. This indicates the effective amplitude boundary of the sway.

[0067] It should be noted that the effective amplitude boundary of the sway is measured in meters per second squared, and is obtained through statistics on the upper limit of the longitudinal vibration amplitude of the bus carriage under the condition of straight-line uniform speed and no load, with a value range of 0.03. Up to 0.08 .

[0068] Furthermore, the vehicle brake controller generates the front-to-back sway amplitude state of the carriage based on the longitudinal sway of the front and rear of the carriage.

[0069] Specifically, when the absolute values ​​of the longitudinal sway at the front and rear of the carriage are both less than the effective amplitude boundary of the sway, the sway amplitude state of the carriage is recorded as a low amplitude state; when at least one of the absolute values ​​of the longitudinal sway at the front and rear of the carriage is not less than the effective amplitude boundary of the sway, the sway amplitude state of the carriage is recorded as an effective sway state.

[0070] Furthermore, the vehicle braking controller inputs the phase difference of the swaying of the car body, the consistency of the longitudinal swaying direction of the car body, the main vibration period of the front of the car body, the main vibration period of the rear of the car body, and the amplitude of the front and rear swaying of the car body into the car body inertial phase recognition model to generate the inertial phase state of the car body.

[0071] The inertial phase recognition model for the carriage is a lightweight time-series classification model deployed within the vehicle's braking controller. It includes an input normalization layer, an amplitude-phase embedding layer, a short-time gating layer, and a state discrimination layer. The input normalization layer receives the phase difference of the carriage group's swaying, the consistency of the carriage's longitudinal swaying direction, the main vibration period of the front of the carriage, the main vibration period of the rear of the carriage, and the amplitude state of the carriage's front and rear swaying, and unifies the value scale of different input quantities. The amplitude-phase embedding layer establishes the correspondence between the amplitude state of the carriage's front and rear swaying, the phase difference of the carriage group's swaying, and the consistency of the carriage's longitudinal swaying direction, in order to distinguish between low-amplitude noise, synchronous forward tilting, synchronous swaying, and asynchronous front and rear swaying. The short-time gating layer reads the amplitude-phase embedding data within a fixed number (e.g., 10) of braking control cycles to suppress state jumps caused by fluctuations at a single sampling point. The state discrimination layer outputs the forward tilting in-phase state, the swaying in-phase state, the phase discrete state, or the low-amplitude stable state.

[0072] It should be noted that the training process of the carriage inertial phase recognition model is as follows: Training is performed using historical braking segments of the same vehicle model. During training, the model first generates the carriage group sway phase difference, the consistency of the carriage's longitudinal sway direction, the main vibration period of the front and rear of the carriage, and the amplitude of the carriage's front and rear sway from the historical braking segments, according to the correction method for the difference in the longitudinal sway of the carriage's front and rear, and the main vibration extraction method. This data from multiple consecutive braking control cycles is then compiled into time-series samples. Next, the forward tilting in-phase state, the swaying in-phase state, the phase discrete state, and the low-amplitude stable state are used as sample labels to supervise the training of the carriage inertial phase recognition model. During training, the carriage inertial phase recognition model outputs the confidence scores of four states. First, the state classification loss constraint model outputs the state consistent with the sample labels, and then... The adjacent cycle stability loss constrains the change in state confidence of adjacent braking control cycles within the same historical braking segment, avoiding frequent state switching of the model due to disturbances at a single sampling point. The state classification loss adopts cross-entropy loss, and the adjacent cycle stability loss adopts the average of the absolute values ​​of the state confidence differences of adjacent braking control cycles. The total model loss is composed of the weighted sum of the state classification loss and the adjacent cycle stability loss. After training, the model is validated using historical braking segments that were not used in the training. When the model's recognition accuracy for forward tilting in phase and swaying in phase reaches the recognition accuracy boundary, and the number of state jumps within a continuous braking control cycle (e.g., 10) does not exceed the state stability boundary, the car body inertial phase recognition model is written into the vehicle braking controller to generate the car body inertial phase state in real-time braking control.

[0073] It should be noted that the accuracy boundary is determined by manual verification of historical braking segments of the same vehicle model that were not used in training. The ratio of the number of manually verified forward tilting and swaying phase states to the total number of corresponding states is used as the basis for determining the accuracy boundary, with a value range of 0.88 to 0.96. The stability boundary is determined by statistically analyzing historical operating segments of the same vehicle model on straight roads without emergency braking and with the amplitude of the front and rear sway of the carriage within the normal range. The upper limit of the number of carriage inertial phase state switching allowed within a unit short-term window is used as the stability boundary, with a value range of 1 to 3 times per short-term window.

[0074] Furthermore, when the inertial phase recognition model of the carriage outputs a forward tilting in-phase state, it indicates that the front and rear of the carriage are jointly moving towards the front of the vehicle to form an inertial response; when the inertial phase recognition model of the carriage outputs a swinging in-phase state, it indicates that the front and rear of the carriage are jointly moving towards the rear of the vehicle to form a recovery response; when the inertial phase recognition model of the carriage outputs a phase discrete state, it indicates that there is effective swaying in front and behind the carriage but no synchronous inertial phase is formed; when the inertial phase recognition model of the carriage outputs a low-amplitude stable state, it indicates that the longitudinal swaying in front and behind the carriage has not reached the effective amplitude that affects the release of the braking peak.

[0075] S3. When the inertial phase state of the car body is in the forward tilt phase state, the forward tilt peak avoidance window of the car body is constructed according to the main vibration period of the front and rear of the car body. The phase peak acceptance criterion is calculated by the collision mitigation demand intensity, the forward tilt peak avoidance window of the car body, the phase difference of the car body group swaying, and the consistency of the longitudinal swaying direction of the car body. It is determined whether the braking peak can be released after avoiding the forward tilt peak of the car body. When the inertial phase state of the car body is in the swing phase state, the swing phase state is taken as the acceptance phase for the release of the braking peak and a unique braking release curve is generated.

[0076] Furthermore, when the vehicle's inertial phase state is in a forward-tilting phase, the vehicle braking controller takes the current braking control moment as the starting point of the forward-tilting peak avoidance window, and determines the length of the forward-tilting peak avoidance window based on the main vibration period of the front and rear of the vehicle. The expression is as follows:

[0077] ;

[0078] in, This indicates the length of the window used to avoid the forward tilt peak of the carriage. This indicates the proportion of forward-leaning peak avoidance. Indicates the period of the main vibration at the front of the carriage. This indicates the period of the main vibration at the rear of the carriage.

[0079] It should be noted that the forward tilt peak avoidance ratio is obtained by statistically analyzing the proportion of the duration of the forward tilt peak of the passenger compartment to the main vibration cycle of the passenger compartment in historical braking segments of the same vehicle model. The value ranges from 0.18 to 0.35. When the forward tilt peak avoidance ratio is less than 0.18, the forward tilt peak avoidance window of the passenger compartment is difficult to cover the main inertial response stage formed by the front and rear of the passenger compartment facing the front of the vehicle. When the forward tilt peak avoidance ratio is greater than 0.35, the forward tilt peak avoidance window of the passenger compartment is prone to occupying too much collision mitigation time, causing the release of the braking peak to lag behind the target approach state. Therefore, limiting the forward tilt peak avoidance ratio to 0.18 to 0.35 can balance the sufficiency of the forward tilt peak coverage of the passenger compartment and the real-time nature of collision mitigation.

[0080] Furthermore, the endpoint of the forward tilting peak avoidance window is recorded as... The expression is:

[0081] ;

[0082] in, This indicates the end point of the window for avoiding the forward tilting peak of the train carriage. Indicates the current braking control moment. This indicates the length of the window used to avoid the forward tilt peak of the carriage.

[0083] When the inertial phase state of the carriage is in a swing-in-phase state, a phase-discrete state, or a low-amplitude stable state, the length of the carriage's forward tilt peak avoidance window is... Set to 0; the swing-back phase state serves as the transition phase for the release of the braking peak, indicating that the front and rear of the carriage jointly form a recovery response towards the rear of the vehicle, and the braking peak can be released during the swing-back phase of the passenger's body, thereby weakening the superposition of braking inertia and forward tilting inertia.

[0084] Furthermore, the vehicle brake controller calculates the forward tilt and in-phase suppression amount, expressed as:

[0085] ;

[0086] In the formula, This indicates the forward-leaning, in-phase suppression amount.

[0087] When the forward tilt phase state is more obvious, the longitudinal sway of the carriage is closer to the positive direction, the phase difference of the carriage group sway is smaller, and the forward tilt phase suppression amount is greater. The forward tilt phase suppression amount is used to suppress the release of the braking peak within the forward tilt peak of the carriage, so that the release time of the braking peak is not solely determined by the approach time of the external target.

[0088] Furthermore, the vehicle brake controller reads the brake pressure build-up time and calculates the phase misalignment acceptance criterion, expressed as:

[0089] ;

[0090] in, This indicates that phase misalignment can be accepted as a criterion.

[0091] It should be noted that the phase staggering criterion can be used to determine whether a bus, under the current target approach pressure, still has the conditions to release the braking peak after avoiding the forward tilting peak of the passenger compartment. This is achieved by first determining whether there is still usable time after deducting the braking pressure build-up time and the time spent avoiding the forward tilting peak during the target approach process. The degree of staggering is then compressed based on the obviousness of the passenger compartment's forward and rearward swaying towards the front of the bus. The reason for this setting is that bus collision mitigation braking cannot only consider whether the target ahead is dangerous, nor can it only consider whether the passenger compartment is tilting forward; it must simultaneously assess how much waiting time is allowed due to the external collision risk. The timing of braking is determined by several factors, including how long it takes for the braking actuator to achieve effective deceleration, how long the forward tilt peak of the passenger compartment needs to be avoided, and whether the forward tilt phase is significant. Only when all these conditions are met can the braking peak be delayed until after the forward tilt peak of the passenger compartment ends. Through phase staggering, the vehicle can avoid the simultaneous superposition of the braking peak and the forward tilt inertia peak of standing passengers under continuous deceleration requirements, and can promptly release waiting constraints and prioritize reducing collision speed under immediate strong deceleration requirements. Thus, without simply reducing braking intensity, the vehicle can achieve a synergistic improvement in external collision mitigation capability and internal passenger stability.

[0092] When the phase shifting acceptance criterion is greater than or equal to the shifting acceptance boundary, it is determined to be a shifting acceptance state; when the phase shifting acceptance criterion is less than the shifting acceptance boundary but greater than the forced release boundary, it is determined to be a limited shifting state; when the phase shifting acceptance criterion is less than or equal to the forced release boundary, it is determined to be a non-shifting state; the shifting acceptance state, the limited shifting state, and the non-shifting state are taken as the phase shifting acceptance state.

[0093] It should be noted that the staggered peak acceptance boundary is obtained through statistics of braking segments in the historical braking segments of the same vehicle model that meet the collision speed reduction requirements and do not trigger the condition of the forward tilt peak amplification, with a value range of 0.18 to 0.35; the condition of the forward tilt peak amplification is determined by whether the forward tilt peak value of the longitudinal sway of the front and rear of the vehicle after the release of the braking peak value exceeds the upper limit of the historical normal forward tilt peak value, and the upper limit of the historical normal forward tilt peak value is obtained through statistics of the front and rear longitudinal sway peak values ​​of the same vehicle model under normal passenger carrying and non-emergency braking conditions; the forced release boundary is obtained through statistics of the time margin that must be released immediately in the historical emergency approach segments of the same vehicle model, with a value range of -0.05 to 0.08.

[0094] Furthermore, the vehicle braking controller reads the corresponding target deceleration peak value according to the collision mitigation demand intensity. When the collision mitigation demand intensity is a slow start demand, it reads the low-intensity target deceleration peak value. When the collision mitigation demand intensity is a continuous deceleration demand, it reads the continuous target deceleration peak value. When the collision mitigation demand intensity is an immediate strong deceleration demand, it reads the forced target deceleration peak value.

[0095] It should be noted that the peak values ​​of low-intensity target deceleration, continuous target deceleration, and forced target deceleration were all determined jointly using the same vehicle model's full-load braking calibration, wheel-end adhesion estimation boundaries, and passenger standing stability test segments. The peak value of low-intensity target deceleration ranged from 0.5. Up to 1.2 The peak value of the continuous target deceleration ranges from 1.2. Up to 2.5 The forced target deceleration peak value range is 2.5. Up to 4.5 When the speed is below the corresponding range, it is difficult to achieve effective collision speed reduction; when it is above the corresponding range, it is easy to cause significant inertial impact to standing passengers in the bus.

[0096] Furthermore, when the collision mitigation demand intensity is the slow start demand, the vehicle brake controller determines the target deceleration peak value as the low intensity target deceleration peak value, determines the release time of the target deceleration peak value as the brake pressure build-up completion time, and outputs the target deceleration according to the slow start incremental growth rate, so that the target deceleration increases from zero to the low intensity target deceleration peak value.

[0097] It should be noted that the gradual acceleration rate of the slow start is determined by the peak value of the low-intensity target deceleration and the slow start release time. The slow start release time is obtained by statistically analyzing braking segments of the same vehicle model that complete low-intensity smooth deceleration within the current speed range without amplifying the longitudinal sway of the vehicle body.

[0098] Furthermore, when the phase shifting and peak acceptance state is a shifting and peak acceptance state, and the inertial phase state of the carriage is a forward tilting phase state, and the collision mitigation demand intensity is a continuous deceleration demand, the vehicle brake controller sets the target deceleration peak release time after the end of the carriage forward tilting peak avoidance window. Within the carriage forward tilting peak avoidance window, the vehicle brake controller first controls the brake actuator to complete brake pressure build-up, and then outputs the target deceleration according to a low-intensity incremental rate, so that the target deceleration increases from zero to the low-intensity target deceleration peak and is maintained until the end of the carriage forward tilting peak avoidance window; thus, the brake actuator enters a rapid response state in advance, the bus generates controlled deceleration in advance, and the continuous target deceleration peaks are not superimposed within the carriage forward tilting peak.

[0099] It should be noted that the low-intensity incremental growth rate is determined by the low-intensity target deceleration peak value and the length of the car's forward tilting peak avoidance window, so that the target deceleration only reaches the low-intensity target deceleration peak value within the car's forward tilting peak avoidance window, without prematurely releasing the continuous target deceleration peak value.

[0100] Furthermore, when the phase shifting state is a limited-amplitude shifting state, and the inertial phase state of the carriage is a forward-tilting in-phase state, and the collision mitigation demand intensity is a continuous deceleration demand, the vehicle braking controller does not wait for the complete carriage forward-tilting peak avoidance window to end. Instead, it first controls the braking actuator to complete the braking pressure build-up, and then outputs the target deceleration according to the limited incremental rate, so that the target deceleration increases from zero to the limited target deceleration peak value. Thus, even when the bus cannot wait for the carriage forward-tilting peak avoidance window to end completely, it still avoids the target deceleration from directly jumping to the continuous target deceleration peak value, thereby reducing the impact superposition in the carriage forward-tilting in-phase state while retaining the continuous deceleration capability.

[0101] It should be noted that the peak value of the limited target deceleration is determined by the intermediate range between the peak value of the low-intensity target deceleration and the peak value of the continuous target deceleration. Specifically, the difference between the peak value of the low-intensity target deceleration and the peak value of the continuous target deceleration is read, and half of the difference is added to the peak value of the low-intensity target deceleration to form the peak value of the limited target deceleration. Half of the difference is used to place the peak value of the limited target deceleration at an equidistant midpoint between the peak value of the low-intensity target deceleration and the peak value of the continuous target deceleration. The reason for choosing an equidistant midpoint is that the limited peak staggering state is between the staggerable peak carrying state and the non-staggerable peak state. It is necessary to avoid the target deceleration directly jumping to the peak value of the continuous target deceleration while retaining the collision speed reduction capability, thereby forming a definite, reproducible transition braking intensity that does not depend on subjective weights. The limited incremental rate is determined by the available time between the peak value of the limited target deceleration and the time corresponding to the time boundary of the strong deceleration, and is constrained by the continuous braking incremental boundary.

[0102] Furthermore, when the phase shifting state is a non-shifting state, or when the collision mitigation demand intensity is an immediate strong deceleration demand, the vehicle brake controller releases the waiting constraint on the target deceleration peak release time, sets the target deceleration peak release time as the brake pressure build-up completion time, and limits the target deceleration increment rate within the forced braking increment boundary.

[0103] It should be noted that the mandatory braking incremental boundary is obtained through calibration of braking segments of the same vehicle under full load that do not trigger wheel instability and do not exceed the permissible range of longitudinal impact on the vehicle body. The value range is 4. up to 8 .

[0104] Furthermore, when the inertial phase state of the carriage is in the same phase as the swing, the vehicle brake controller releases the target deceleration peak value corresponding to the collision mitigation demand intensity after the brake pressure is built up, and controls the target deceleration increment rate within the continuous braking increment boundary.

[0105] It should be noted that the continuous braking increment boundary was obtained through statistics of braking segments of the same vehicle model under normal passenger-carrying conditions that did not cause significant forward pitch amplification, and the value range was 1.5. Up to 4 .

[0106] It should be noted that releasing the braking peak during the same phase of the swing allows the braking inertia to act on the carriage during the recovery response phase, reducing the risk of peak superposition when standing passengers transition from swing to forward tilt.

[0107] Furthermore, when the inertial phase state of the carriage is in a phase-discrete state, the vehicle brake controller does not set a carriage forward tilt peak avoidance window, determines the release time of the target deceleration peak as the brake pressure build-up completion time, determines the target deceleration peak as the target deceleration peak corresponding to the collision mitigation demand intensity, and determines the target deceleration increment rate as half of the continuous braking increment boundary, so that the target deceleration continuously increases from zero to the target deceleration peak. When the inertial phase state of the carriage is in a low-amplitude stable state, the vehicle brake controller does not set a carriage forward tilt peak avoidance window, determines the release time of the target deceleration peak as the brake pressure build-up completion time, determines the target deceleration peak as the target deceleration peak corresponding to the collision mitigation demand intensity, and determines the target deceleration increment rate as a normal increment rate, so that the target deceleration continuously increases from zero to the target deceleration peak. Thus, in the phase-discrete state, the asynchronous swaying of the front and rear carriages is suppressed by reducing the increment rate, and in the low-amplitude stable state, the collision mitigation response efficiency is maintained by the normal increment rate.

[0108] It should be noted that the conventional rate of increase is obtained through statistical analysis of historical braking segments of the same vehicle model under low-amplitude stable conditions that do not amplify longitudinal sway of the passenger compartment. Furthermore, the conventional rate of increase is not greater than the continuous braking increase boundary, and the preferred value range is 1.0 m / s³ to 3.5 m / s³.

[0109] Furthermore, the unique brake release curve is generated by five parameters: the brake pressure build-up start point, the target deceleration rate, the target deceleration peak release time, the target deceleration peak value, and the target deceleration retraction rate.

[0110] Specifically, the vehicle brake controller first determines the current braking control cycle as the starting point of brake pressure build-up, then determines the target deceleration peak value based on the collision mitigation demand intensity, determines the target deceleration peak value release time based on the vehicle's inertial phase state and phase misalignment acceptance criteria, determines the target deceleration increment rate based on the vehicle's inertial phase state, and determines the target deceleration retraction rate based on the target approach time recovery. Each of the five parameters corresponds to only one value within the same braking control cycle. The vehicle brake controller generates a single target deceleration curve over time according to the five parameters and uses this single target deceleration curve over time as the unique brake release curve.

[0111] S4. Perform bus collision mitigation braking and close-loop update through the unique brake release curve.

[0112] Furthermore, the vehicle brake controller performs collision mitigation braking for the bus according to the unique brake release curve, and rereads the target approach data and the front and rear sway data of the carriage in each braking control cycle, updates the collision mitigation demand intensity, and re-inputs the phase difference of the carriage group sway and the consistency of the longitudinal sway direction of the carriage into the carriage inertial phase recognition model to update the carriage inertial phase state and update the phase staggered acceptance state.

[0113] Furthermore, when the updated car body inertial phase state is a forward-tilting in-phase state, and the updated phase peak avoidance state is a peak avoidance state, the vehicle brake controller recalculates the length of the car body forward-tilting peak avoidance window, delays the release time of the target deceleration peak to the end of the recalculated car body forward-tilting peak avoidance window, and maintains the target deceleration within the low-intensity target deceleration peak during the delay period; when the updated phase peak avoidance state is a limited peak avoidance state, the vehicle brake controller no longer waits for the complete car body forward-tilting peak avoidance window, and releases the target deceleration peak... The target deceleration peak value is adjusted to a limited target deceleration rate, and the target deceleration increment rate is adjusted to a limited increment rate, so that the target deceleration continuously increases from the current target deceleration to the limited target deceleration peak value; when the updated collision mitigation demand intensity turns into an immediate strong deceleration demand, or the updated phase staggering can be accepted as a non-staggering state, the vehicle brake controller releases the waiting constraint of the target deceleration peak value release time, adjusts the target deceleration peak value to a forced target deceleration peak value, and performs bus collision mitigation braking according to the target deceleration increment rate not exceeding the forced braking increment boundary.

[0114] Furthermore, when the target approach time recovers to above the safe approach time boundary, and the distance to the forward target continues to increase within a fixed number (e.g., 3) of braking control cycles, while the relative speed of the forward target is not greater than zero, the vehicle brake controller determines that the target approach risk has been eliminated and enters the braking force retraction phase.

[0115] Furthermore, the vehicle brake controller records the target deceleration at the start of the brake force retraction phase as the initial target deceleration. It then gradually reduces the target deceleration to zero according to the target deceleration retraction rate. During brake force retraction, the vehicle brake controller continues to update the car's inertial phase state based on the phase difference of the car group's swaying and the consistency of the car's longitudinal swaying direction. When the car's inertial phase state transitions to a forward-tilting phase state, the vehicle brake controller adjusts the target deceleration retraction rate to half of the current target deceleration retraction rate, causing the target deceleration to decrease at a slower rate until the car's inertial phase state exits the forward-tilting phase state and then returns to the original target deceleration retraction rate. Therefore, the braking force will not disappear rapidly when the front and rear of the car form an inertial response towards the front of the vehicle, preventing secondary swaying of standing passengers caused by a sudden drop in deceleration.

[0116] It should be noted that the target deceleration retraction rate is obtained through statistics of brake release segments that do not cause amplification of longitudinal swaying of the passenger compartment under normal passenger-carrying conditions for the same vehicle model. Furthermore, the target deceleration retraction rate is not greater than the continuous braking increment boundary, and the preferred value range for the target deceleration retraction rate is 0.8 m / s³ to 3.0 m / s³. The reason for adjusting the target deceleration retraction rate to half of the current target deceleration retraction rate is to form a gradual release transition without stopping the release of braking force, so as to avoid secondary swaying of standing passengers in the passenger compartment caused by a rapid decrease in the target deceleration. When the inertial phase state of the passenger compartment exits the forward tilting phase state, the vehicle brake controller restores the original target deceleration retraction rate.

[0117] In summary, this invention achieves intelligent differentiation between forward tilting phase state, swaying phase state, phase discrete state, and low-amplitude stable state by generating the inertial phase state of the passenger compartment. This enables bus braking control to simultaneously consider external collision risks and the inertial bearing state of passengers inside the passenger compartment. By using the forward tilting peak avoidance window and the phase staggered peak bearing criterion, it enables the determination of whether the braking peak is released after avoiding the forward tilting peak of the passenger compartment, reducing the risk of falls caused by the simultaneous superposition of the braking peak and the passenger forward tilting inertial peak. By generating a unique braking release curve, it achieves layered control of the braking peak release time and the target deceleration increase process, reducing secondary impacts inside the passenger compartment while maintaining the collision speed reduction capability.

[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based bus collision mitigation braking method, characterized by, include: The system collects the forward target approach status and the longitudinal response status of the bus carriage, and performs differential correction on the longitudinal response status of the bus carriage to form target approach data and front and rear sway data of the bus carriage. Based on the target proximity data, the collision mitigation demand intensity is determined, and based on the front and rear sway data of the carriage, the front and rear main vibration phases and main vibration periods of the carriage are extracted. The phase difference of the carriage group sway and the consistency of the longitudinal sway direction of the carriage are calculated and input into the carriage inertial phase recognition model to generate the carriage inertial phase state. When the inertial phase state of the car body is in the forward tilt phase state, the forward tilt peak avoidance window of the car body is constructed according to the main vibration period of the front and rear of the car body. The phase peak acceptance criterion is calculated by the collision mitigation demand intensity, the forward tilt peak avoidance window of the car body, the phase difference of the car body group swaying and the consistency of the longitudinal swaying direction of the car body. It is determined whether the braking peak can be released after avoiding the forward tilt peak of the car body. When the inertial phase state of the car body is in the swing phase state, the swing phase state is taken as the acceptance phase that allows the braking peak to be released, and a unique braking release curve is generated. The bus collision mitigation braking is performed using a unique brake release curve, and a closed-loop update is then implemented.

2. The artificial intelligence-based bus collision mitigation braking method of claim 1, wherein, The target proximity data and the front-to-back swaying data of the carriage include: Collect the forward target distance and forward target relative velocity, and use the forward target distance and forward target relative velocity as target approach data; Collect the longitudinal response values ​​at the front and rear of the carriage to form the longitudinal response status of the carriage at the front and rear. Read the longitudinal motion state of the bus, and use the current longitudinal acceleration of the bus in the longitudinal motion state as the common mode motion quantity to perform difference correction on the longitudinal response quantity at the front of the carriage and the longitudinal response quantity at the rear of the carriage, and obtain the longitudinal sway quantity at the front of the carriage and the longitudinal sway quantity at the rear of the carriage. The longitudinal sway at the front and rear of the carriage are processed in a short time window to form a longitudinal sway sequence at the front and rear of the carriage. The longitudinal sway sequence at the front and rear of the carriage are used as the front-to-back sway data of the carriage. 3.The AI-based bus collision mitigation braking method of claim 2, wherein, The determination of the collision mitigation demand intensity includes: Determine the safe distance based on the bus's current speed, braking pressure build-up time, and the safety margin of the bus front. The target approach time is determined based on the forward target distance, the forward target relative speed, the safe retention distance, and the minimum approach speed protection value. When the target approach time is greater than the safe approach time boundary, the collision mitigation requirement intensity is determined as the soft start requirement; When the target approach time is less than or equal to the safe approach time boundary and greater than the strong deceleration time boundary, the collision mitigation requirement intensity is determined as continuous deceleration requirement. When the target approach time is less than or equal to the strong deceleration time boundary, the collision mitigation requirement intensity is determined as an immediate strong deceleration requirement.

4. The artificial intelligence-based bus collision mitigation braking method as described in claim 2, characterized in that, The extraction of the front and rear principal vibration phases and principal vibration periods of the carriage includes: The same principal vibration extraction process was performed on the longitudinal sway sequence at the front of the carriage and the longitudinal sway sequence at the rear of the carriage, respectively; The main vibration extraction process includes identifying adjacent zero-crossing points and local peak points of the longitudinal sway in the corresponding longitudinal sway sequence. When the corresponding longitudinal sway sequence has no less than two adjacent zero-crossing points in the same direction and no less than one local peak point within a short time window, the main vibration period at the corresponding position is determined according to the time interval between the two adjacent zero-crossing points in the same direction, and the main vibration phase at the corresponding position is determined according to the relative time position between the two adjacent zero-crossing points in the same direction at the current sampling time. The corresponding locations include the front and rear of the carriage; The corresponding longitudinal sway sequence includes a longitudinal sway sequence at the front of the carriage and a longitudinal sway sequence at the rear of the carriage. 5.The AI-based bus collision mitigation braking method of claim 4, wherein, The calculation of the phase difference of the swaying of the carriage group and the consistency of the longitudinal swaying direction of the carriages include: Based on the phase difference between the main vibration phase at the front of the carriage and the main vibration phase at the rear of the carriage, phase reversal processing is performed to calculate the phase difference of the carriage group swaying. Based on the longitudinal sway at the front and rear of the carriage and the effective amplitude boundary of the sway, calculate the longitudinal sway direction consistency of the carriage. When the absolute values ​​of the longitudinal sway at the front and rear of the carriage are both less than the effective amplitude boundary of the sway, the sway amplitude state of the carriage is determined to be a low amplitude state. When at least one of the absolute values ​​of the longitudinal sway at the front of the carriage and the absolute value of the longitudinal sway at the rear of the carriage is not less than the effective sway amplitude boundary, the sway amplitude state of the carriage is determined to be an effective sway state.

6. The artificial intelligence-based bus collision mitigation braking method of claim 1, wherein, The inertial phase recognition model for the carriage includes an input normalization layer, an amplitude-phase embedding layer, a short-time gating layer, and a state discrimination layer. The input normalization layer includes scaling the phase difference of the swaying of the carriage group, the consistency of the longitudinal swaying direction of the carriage, the main vibration period of the front of the carriage, the main vibration period of the rear of the carriage, and the amplitude of the swaying of the carriage at the front and rear. The amplitude and phase embedding layer includes establishing a correspondence between the front and rear sway amplitude state of the carriage, the phase difference of the carriage group sway, and the consistency of the longitudinal sway direction of the carriage. The short-time gating layer includes reading amplitude and phase embedded data within a fixed number of consecutive braking control cycles to suppress state jumps caused by fluctuations at a single sampling point; The state discrimination layer includes outputting forward tilt in-phase state, sway in-phase state, phase discrete state, or low-amplitude stable state as the inertial phase state of the carriage.

7. The artificial intelligence-based bus collision mitigation braking method as described in claim 6, characterized in that, When the inertial phase state of the carriage is in a forward-tilting phase state, a forward-tilting peak avoidance window is constructed based on the main vibration period of the carriage. The specific steps are as follows: When the inertial phase state of the carriage is in the forward tilt phase state, the current braking control moment is determined as the starting point of the carriage forward tilt peak avoidance window; The length of the forward tilt peak avoidance window of the car is determined based on the main vibration period of the front of the car, the main vibration period of the rear of the car, and the forward tilt peak avoidance ratio. The endpoint of the forward tilting peak avoidance window is determined based on the starting point and length of the forward tilting peak avoidance window. When the inertial phase state of the carriage is in a swing-in-phase state, a phase-discrete state, or a low-amplitude stable state, the length of the carriage's forward tilt peak avoidance window is set to zero. 8.The AI-based bus collision mitigation braking method of claim 5, wherein, The phase misalignment criteria include: The forward tilting and in-phase suppression amount is determined based on the phase difference of the swaying of the carriage group and the consistency of the longitudinal swaying direction of the carriage. Based on the target approach time, braking pressure build-up time, length of the forward tilting peak avoidance window of the carriage, and forward tilting in-phase suppression amount, the phase stagger acceptance criterion is calculated. When the phase stagger acceptance criterion is greater than or equal to the stagger acceptance boundary, it is determined to be in a stagger acceptance state. When the phase stagger acceptance criterion is less than the stagger acceptance boundary but greater than the forced release boundary, it is determined to be a limited stagger state. When the phase stagger acceptance criterion is less than or equal to the forced release boundary, it is determined to be a non-stagger state; The states of being able to take on peak loads, being limited to take on peak loads, and being unable to take on peak loads are defined as phase-based peak load states. 9.The AI-based bus collision mitigation braking method of claim 6 or 8, wherein, The generation of the unique brake release curve includes: When the collision mitigation demand intensity is the slow start demand, the target deceleration peak value is determined as the low intensity target deceleration peak value, the target deceleration peak value release time is set as the slow start release time after the brake pressure build-up is completed, and the target deceleration is output according to the slow start incremental growth rate, so that the target deceleration is incrementally increased to the low intensity target deceleration peak value. When the phase staggered acceptance state is a staggered acceptance state, the inertial phase state of the carriage is a forward tilting phase state, and the collision mitigation demand intensity is a continuous deceleration demand, the target deceleration peak release time is set after the end of the carriage forward tilting peak avoidance window, and the target deceleration is output at a low intensity incremental rate within the carriage forward tilting peak avoidance window. When the phase shifting state is the limited shifting state, the inertial phase state of the carriage is the forward tilting phase state, and the collision mitigation demand intensity is the continuous deceleration demand, the target deceleration peak value is determined as the limited target deceleration peak value, and the target deceleration is output according to the limited incremental growth rate. When the phase shifting state is a non-shifting state, or the collision mitigation demand intensity is an immediate strong deceleration demand, the waiting constraint of the target deceleration peak release time is released, and the target deceleration peak release time is set as the braking pressure build-up completion time. When the inertial phase state of the carriage is in the same phase as the swing, the same phase as the swing is taken as the transition phase that allows the release of the peak braking force. The target deceleration peak is released after the brake pressure is built up. A unique brake release curve is generated based on the brake pressure build-up start point, the target deceleration rate of increase, the target deceleration peak release time, the target deceleration peak value, and the target deceleration retraction rate.

10. The artificial intelligence-based bus collision mitigation braking method as described in claim 9, characterized in that, The specific steps for performing collision mitigation braking on the bus using a unique brake release curve and then updating the loop are as follows: The bus collision mitigation braking was performed according to the unique brake release curve. During each braking control cycle, target approach data and car front-to-back sway data are reread and closed-loop updates are performed to obtain the updated car inertial phase state and phase peak misalignment acceptance state. Based on the updated collision mitigation demand intensity, the inertial phase state of the carriage, and the phase shift acceptance state, the release time of the target deceleration peak, the target deceleration peak value, and the target deceleration change rate are adjusted. When the target is close to the risk being eliminated, it enters the braking force withdrawal phase, and gradually reduces the target deceleration to zero according to the target deceleration withdrawal rate.

Images