Automobile immersion monitoring system and method

Abstract

The invention relates to the technical field of vehicle monitoring, and discloses an automobile immersion monitoring system and method, and the method comprises the steps: building a multi-source sensor time sequence collection frame based on a dual-sensor combination detection device and a heavy hammer switch device, and obtaining an original sensor signal sequence with a timestamp; performing targeted filtering processing on the original signal to obtain a filtered sensor signal sequence; performing space-time and trend analysis on the filtered signal to obtain a left side signal feature vector, a right side signal feature vector, a cross-side consistency index and a signal change rate trend feature; constructing a hierarchical state transition model based on the features; scene state transition, differential self-rescue action execution and hysteresis control are implemented based on the model; a dynamic wading decision confusion trap is effectively cracked, and accurate, robust and delay-free identification and decision of the wading state are realized.

Description

Technical Field

[0001] This invention relates to the field of vehicle monitoring technology, and more specifically, to a vehicle flooding monitoring system and method. Background Technology

[0002] With the increasing frequency of vehicle flooding accidents, vehicle immersion monitoring technology is crucial for ensuring the safety of occupants. Existing technologies mostly rely on single sensors or simple logic judgments, which are difficult to cope with complex and ever-changing flooding environments. This can lead to misjudgments or response delays in monitoring systems at critical moments, seriously threatening the lives of occupants.

[0003] Chinese patent application CN114821475A discloses a method, device, equipment, and storage medium for monitoring vehicles falling into water. It collects the vehicle's geographical location at a preset frequency to determine if the vehicle is in an unconventional location, such as an unpaved road or near a river. If an unconventional location is identified, a water-fall detection is initiated. Once the vehicle is confirmed to be in water, an onboard terminal controls a water electrolysis device to generate oxygen and hydrogen. The oxygen is delivered to the passenger compartment to ensure breathing needs and balance the pressure inside and outside the vehicle, while the hydrogen is used to inflate pre-installed airbags or tires at the bottom of the vehicle, increasing buoyancy and slowing sinking. This solution focuses on survival after falling into water, providing oxygen and increased buoyancy to create an escape window for the occupants. Chinese patent CN118636846B, authorized by the patent authority, proposes a vehicle control method, storage medium, and vehicle. This method acquires the vehicle's wading status (including at least three states: wading in water, floating, and wading out of water), and matches differentiated braking strategies to each state. For example, in the wading in water state, the braking assist intensity is adjusted to prevent brake failure due to water resistance; in the floating state, the braking intervention is limited to prevent vehicle instability. This solution focuses on braking safety control during wading, optimizing the control strategy through state classification.

[0004] However, existing technologies have not effectively addressed the spatiotemporal decoupling and decision-making confusion traps of multimodal sensing information in dynamic wading events. A vehicle wading through water is a high-dimensional nonlinear dynamic event, its state space defined by numerous variables, including the vehicle's three-dimensional position, attitude (roll, pitch, yaw), linear velocity, angular velocity, and the time-varying water pressure / level distribution in the external flow field. The sensor system (float, immersion device, and weight) in this invention is essentially a low-dimensional, discrete, and asynchronous measurement projection onto the high-dimensional state space. This projection is not unique; highly dynamic events with drastically different properties (such as the violent splashing of water when passing through a deep puddle at high speed versus a vehicle rolling over and sliding into shallow water) may produce extremely similar "projection characteristics" in the low-dimensional sensing data space. Specifically, this manifests as: distributed sensor installation leading to spatiotemporal decoupling; differences in the response characteristics of different sensors leading to modal asynchrony; and highly similar signal combination characteristics. Existing technologies cannot distinguish these similar features, falling into the "feature confusion trap," which leads to decision paralysis, missing the golden escape window, or generating suppression alarms, resulting in occupants being trapped; or even implementing incorrect self-rescue strategies (such as keeping the windows closed), blocking the escape route and causing irreparable casualties.

[0005] It is particularly important to note that existing vehicle immersion monitoring solutions all employ a single type of water level sensor. They either use only a float-type sensor for mechanical triggering or only a conductivity sensor for water contact detection. Currently, no solution integrates both a float-type water sensor and a water immersion sensor to form a dual verification mechanism. This single-sensor approach has fundamental flaws: when used alone, the float-type sensor is prone to false triggering or missed triggering due to mechanical jamming or changes in float density; when used alone, the water immersion sensor is prone to generating false signals due to differences in water quality and electromagnetic interference. These inherent limitations of single sensors prevent current technologies from reliably identifying wading conditions, becoming a core bottleneck restricting the development of vehicle immersion monitoring technology. Summary of the Invention

[0006] This invention is applicable to real-time monitoring of various vehicles in water-filled environments, especially in scenarios such as urban road flooding, river overflows, and flooded sections caused by heavy rain, providing occupants with timely and accurate assessment of water conditions and escape guidance. To overcome the aforementioned shortcomings of existing technologies, this invention provides a vehicle immersion monitoring system and method. It proposes for the first time a combined configuration of a float-type water sensor and a water immersion sensor, addressing the inherent limitations of single sensors. By establishing a multi-source sensor timing acquisition framework based on the aforementioned combined configuration—a dual-sensor detection device and a counterweight switch device—targeted filtering, spatiotemporal and trend analysis of sensor signals are performed, and a hierarchical state transition model is constructed to achieve accurate identification and decision-making for dynamic water-filled events. This invention effectively distinguishes between complex water-filled scenarios such as normal entry into water, side-tied shallow water, inverted water, and splashing interference, significantly improving the accuracy, robustness, and real-time performance of vehicle immersion monitoring, providing occupants with safer and more reliable escape protection.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for monitoring vehicle flooding includes:

[0009] A dual-sensor combined detection device and a counterweight switch device are set up to establish a multi-source sensor timing acquisition framework based on the dual-sensor combined detection device and the counterweight switch device, and the original sensor signal sequence with timestamps is obtained; targeted filtering processing is performed on the original sensor signal sequence with timestamps to obtain the filtered sensor signal sequence.

[0010] Spatiotemporal and trend analysis was performed on the filtered sensor signal sequence to obtain the left signal feature vector, the right signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics.

[0011] A hierarchical state transition model is constructed based on the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics.

[0012] Based on the hierarchical state transition model, scenario state transitions, differentiated self-rescue actions, and hysteresis control are implemented.

[0013] The dual-sensor combined detection device includes both a float-type water sensor and a water immersion sensor.

[0014] The method for performing targeted filtering on the raw sensor signal sequence with timestamps includes:

[0015] The switching signal of the float-type water sensor, the conductivity analog signal of the water immersion sensor, and the mechanical trigger signal of the counterweight switch were separated from the original sensor signal sequence with timestamps.

[0016] The switching signals are processed using a debouncing filtering algorithm to obtain the trigger time and duration of each float-type water sensor.

[0017] The method for performing targeted filtering on the raw sensor signal sequence with timestamps further includes:

[0018] The conductivity analog signal is processed by a medium filtering algorithm to generate a filtered conductivity value. When the filtered conductivity value exceeds a preset conductivity threshold, the trigger time and duration of the water immersion sensor are recorded.

[0019] The mechanical trigger signal is filtered to maintain its execution status, and the triggering time and duration of the counterweight switch are confirmed.

[0020] The method for performing spatiotemporal and trend analysis on the filtered sensor signal sequence includes:

[0021] Based on the filtered sensor signal sequence and the installation location of each sensor, construct the left-side signal feature vector and the right-side signal feature vector;

[0022] The cross-side consistency index is calculated using the feature vectors of the left and right signals.

[0023] Trend analysis was performed on the filtered sensor signal sequence within a time window to obtain the trend characteristics of the signal change rate.

[0024] The method for constructing the left-side signal feature vector and the right-side signal feature vector includes:

[0025] The filtered sensor signal sequence is spatially grouped according to the physical installation position of the sensors on the vehicle, into a left sensor group and a right sensor group.

[0026] Extract the trigger time and duration of each sensor in the left sensor group to form the left signal feature vector;

[0027] Extract the trigger time and duration of each sensor in the right-hand sensor group to form the right-hand signal feature vector.

[0028] The method for calculating the cross-side consistency index includes:

[0029] Extract the trigger times of similar sensors from the left and right signal feature vectors respectively, and calculate the absolute value ΔTF of the difference between the average trigger times of the left float-type water sensor and the average trigger times of the right float-type water sensor, as well as the absolute value ΔTW of the difference between the average trigger times of the left water immersion sensor and the average trigger times of the right water immersion sensor.

[0030] Based on ΔTF and ΔTW, the cross-side consistency index is obtained.

[0031] The method for obtaining the cross-side consistency index based on ΔTF and ΔTW includes:

[0032] ΔTF and ΔTW are converted into time consistency coefficients using a Gaussian function;

[0033] By comparing the sensor triggering combination modes in the left and right signal feature vectors, the consistency of the triggering modes is obtained;

[0034] Multiplying the time consistency coefficient by the trigger mode consistency coefficient yields the cross-side consistency index.

[0035] The method for constructing the hierarchical state transition model includes:

[0036] By combining the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics, a confidence scoring model for four immersion scenarios is constructed, and the confidence values ​​for the four immersion scenarios are calculated. The four immersion scenarios include normal water entry scenario, side-overturned shallow water scenario, upside-down water scenario, and splashing interference scenario.

[0037] Based on the confidence values ​​of four immersion scenarios, a three-state master state machine is established, which includes three main states: normal state, suspected state, and water danger state. Under the water danger state, scenario sub-state diagrams are nested to form a hierarchical state transition model. The scenario sub-state diagrams include the water entry sub-state diagram, the side capsizing shallow water sub-state diagram, and the inverted water sub-state diagram.

[0038] A vehicle flooding monitoring system is provided for implementing the aforementioned vehicle flooding monitoring method. The system includes:

[0039] Signal preprocessing module: used to set up a dual-sensor combined detection device and a counterweight switch device, establish a multi-source sensor timing acquisition framework based on the dual-sensor combined detection device and the counterweight switch device, and acquire the original sensor signal sequence with timestamps; perform targeted filtering processing on the original sensor signal sequence with timestamps to obtain the filtered sensor signal sequence; the dual-sensor combined detection device includes both a float-type water sensor and a water immersion sensor;

[0040] Feature extraction module: used to perform spatiotemporal and trend analysis on the filtered sensor signal sequence to obtain the left signal feature vector, the right signal feature vector, the cross-side consistency index, and the signal change rate trend features;

[0041] State transition model construction module: Based on the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics, a hierarchical state transition model is constructed;

[0042] Execution module: Based on a hierarchical state transition model, it implements scenario state transitions, differentiated self-rescue actions, and hysteresis control;

[0043] The progressively activated buoyancy airbag system includes a progressively activated buoyancy airbag under the vehicle body and a progressively activated buoyancy airbag on the roof. Both the progressively activated buoyancy airbag under the vehicle body and the progressively activated buoyancy airbag on the roof contain multiple airbag igniters, pressure stabilizing tanks, one-way valves, and folding airbags.

[0044] When the car enters the submerged state of water danger, the buoyancy airbags under the vehicle body are activated in stages; when the car enters the inverted submerged state of water danger, the buoyancy airbags on the roof are activated in stages.

[0045] When the buoyancy airbags at the bottom of the vehicle body or at the top of the vehicle are activated in stages, the airbag igniters are detonated one by one at preset time intervals.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] This invention effectively avoids data distortion caused by signal asynchrony and noise interference by using multi-source sensor time-series acquisition and targeted filtering, providing high-quality data support for subsequent analysis. Through spatiotemporal and trend analysis, it extracts multi-dimensional features to comprehensively characterize the spatial distribution, temporal synchronization, and trend evolution of dynamic water-related events, overcoming the limitations of single-feature characterization. Utilizing a hierarchical state transition model and scene-adaptive execution mechanism, it achieves accurate differentiation and rapid response to different water-related scenarios, avoiding decision paralysis and catastrophic misjudgments caused by spatiotemporal decoupling, modal asynchrony, and feature confusion. This significantly improves the accuracy, robustness, and real-time performance of vehicle flooding monitoring, providing occupants with crucial escape time and minimizing the risk of injury or death from flooding accidents. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0049] Figure 1 A flowchart illustrating the principle of a vehicle immersion monitoring method provided in an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the dual-sensor combined detection device provided in an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the principle of the hammer switch device provided in an embodiment of the present invention;

[0052] Figure 4This is a schematic diagram of the layout of the dual-sensor combined detection device in the vehicle body according to an embodiment of the present invention;

[0053] Figure 5 A flowchart illustrating the method for constructing left-side signal feature vectors and right-side signal feature vectors provided in this embodiment of the invention;

[0054] Figure 6 This is a schematic diagram illustrating the principle of the step-by-step activation of the buoyancy airbag provided in an embodiment of the present invention;

[0055] Figure 7 This is a functional block diagram of a car flooding monitoring system provided in an embodiment of the present invention.

[0056] Reference numerals: 1. Float; 2. Cage-shaped cylindrical outer casing; 3. Waterproof micro switch; 4. Water immersion sensor probe; 5. Spherical weight; 6. Cylindrical outer casing; 7. Micro switch; 8. Dual-sensor combination detection device at the front of the vehicle; 9. Dual-sensor combination detection device at the rear of the vehicle; 10. Sequentially activated buoyancy airbag at the lower part of the vehicle body; 11. Sequentially activated buoyancy airbag on the roof; 12. Roof rack; 13. Multiple airbag igniters; 14. Pressure stabilizing tank; 15. One-way valve; 16. Folding airbag. Detailed Implementation

[0057] 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.

[0058] Example 1

[0059] Please see Figure 1 As shown, this embodiment provides a method for monitoring vehicle flooding, including:

[0060] Step S10: Establish a multi-source sensor timing acquisition framework based on a dual-sensor combined detection device and a counterweight switch device, and acquire the original sensor signal sequence with timestamps; perform targeted filtering processing on the original sensor signal sequence with timestamps to obtain the filtered sensor signal sequence.

[0061] Step S10 aims to establish a multi-source sensor timing acquisition framework and perform targeted filtering to solve the problem of basic data distortion caused by asynchronous sensor signals and noise interference during vehicle wading. This is a prerequisite for spatiotemporal decoupling and modal asynchrony. If it is not solved, subsequent feature extraction and state judgment will be based on distorted data, inevitably falling into the decision confusion trap.

[0062] Further, step S10 includes:

[0063] Step S11: Set up a dual-sensor combined detection device and a counterweight switch device, establish a multi-source sensor timing acquisition framework based on the dual-sensor combined detection device and the counterweight switch device, and assign a unique sensor identifier to each sensor in the multi-source sensor timing acquisition framework. The dual-sensor combined detection device includes both a float-type water sensor and a water immersion sensor.

[0064] When constructing the multi-source sensor time-series acquisition framework, four dual-sensor combined detection devices are used. Figure 2 ) and 1 counterweight switch device ( Figure 3 This forms the core monitoring network. The dual-sensor combined detection device consists of a float-type water sensor and a water immersion sensor, installed on the inside of the hood on both the left and right sides of the front and rear of the vehicle. Figure 4 It includes two dual-sensor combination detection devices 8 and 9 on the front of the vehicle, with an unobstructed lower section to allow water to seep in from the bottom. (See also...) Figure 2 The float 1 rises upon contact with water, triggering the waterproof microswitch 3. The water immersion sensor probe 4 generates a signal through direct contact with the water. This dual verification mechanism effectively eliminates false triggering by a single sensor. The cage-like cylindrical outer casing 2 protects the internal components while allowing water flow. The waterproof microswitch is a triggering element with a waterproof and sealed structure that can accurately switch between "on" and "off" states in water-related scenarios. Its core principle revolves around "converting external physical triggering into an electrical signal." Common implementation principles include mechanical contact, Hall effect sensor, and photoelectric principles. The mechanical contact principle refers to a switch with an internal spring contact structure. When an external trigger (such as a floating float) presses against the contact, the contact closes and conducts the circuit. After the triggering force disappears, the spring resets, causing the contact to open. The switch outputs a switching signal through the change in mechanical contact state. The Hall effect sensor principle refers to a switch that integrates a Hall element and a magnet (the magnet can be integrated into the float). When the trigger moves the magnet close to the Hall element, the Hall element outputs an electrical signal due to the magnetic field. The signal disappears when the magnet moves away. This method involves no mechanical wear and offers enhanced waterproofing, making it suitable for high-frequency triggering or complex wading environments with a lot of mud and sand. The photoelectric principle refers to a switch that integrates an infrared transmitter and receiver. When the trigger moves, it blocks or reflects the infrared light path, causing a change in the intensity of the light signal received by the receiver. When the change exceeds a threshold, a switching signal is output. This is also non-contact, avoiding mechanical failure.

[0065] See Figure 3The counterweight switch device determines the vehicle's attitude by monitoring the displacement of a spherical counterweight 5 within a cylindrical outer casing 6. When the car overturns, the counterweight presses against a microswitch 7, generating a signal to verify the system's attitude. The sensor layout in the multi-source sensor timing acquisition framework covers wading conditions at different locations on the left and right sides of the vehicle, avoiding the omission of scenarios like single-side tilting into water by a single-location sensor. The counterweight switch device is installed inside the passenger compartment. As it is an inertial mechanical sensor, it is highly sensitive to sudden changes in vehicle attitude (such as rollover or overturning). Sensors on the hood, however, cannot directly reflect the overall vehicle attitude. The combination of these two sensors enables multi-dimensional information acquisition of "vehicle wading status + vehicle attitude," overcoming the shortcomings of existing technologies that only monitor water level and neglect attitude correlation. Existing technologies often misjudge rollover into water as normal splashing due to a lack of attitude information, or miss early warning opportunities for tilting into water due to insufficient monitoring on one side. The float-type water sensor triggers a waterproof microswitch to generate a signal when the float rises upon contact with water. The immersion sensor generates a signal based on the change in conductivity after the probe contacts the water. This dual verification mechanism effectively eliminates false triggering caused by mechanical failure or environmental interference from a single sensor. The cage-like cylindrical enclosure serves as the external structure for the dual-sensor combination device. It protects the internal float and probe elements from external impacts, while the perforated design allows for smooth water flow, ensuring timely sensor contact and preventing signal delays caused by obstruction from the enclosure. The counterweight switch device houses a spherical counterweight within its cylindrical enclosure. When the vehicle's posture is abnormal, the counterweight's inertial displacement compresses the microswitch, generating a signal. The enclosure also protects the internal counterweight and switch elements, preventing non-postural displacement of the counterweight due to vehicle vibrations.

[0066] When assigning a unique identifier to each sensor, it is essential to strictly associate it with the sensor type and installation location. For example, the left front float water sensor is designated FLF, the left front water immersion sensor is designated WLF, the left rear float water sensor is designated FLR, the left rear water immersion sensor is designated WLR, the right front float water sensor is designated FRF, the right front water immersion sensor is designated WRF, the right rear float water sensor is designated FRR, the right rear water immersion sensor is designated WRR, and the passenger compartment counterweight switch is designated HG. The technical purpose of this identification rule is to ensure that each sensor signal can be accurately located. The source clearly identifies both the vehicle body location corresponding to the signal (front left / rear left / front right / rear right / passenger compartment) and the signal type (buoy / water immersion / weight). This provides a clear location basis for the spatial grouping of sensor signals (left sensor group, right sensor group) and cross-side consistency analysis in subsequent steps. If the identifier lacks location or type information, the front left water immersion signal may be mistakenly classified as front right, or the float and water immersion signal may be confused, resulting in spatial grouping disorder and loss of accuracy in cross-side consistency analysis. Table 1 is a table of sensor identifiers and corresponding information.

[0067] Table 1 Sensor Identifiers and Corresponding Information

[0068]

[0069] It is particularly important to emphasize that the combined use of the float-type water sensor and the immersion sensor is a necessary condition for the reliable water monitoring achieved by this invention; neither can be dispensed with. This necessity is reflected in three aspects: First, the complementarity of their physical detection principles: The float-type water sensor performs binary state detection based on the principle of buoyancy, sensitive to water level height but insensitive to water quality and temperature; the immersion sensor performs continuous detection based on the principle of conductivity, sensitive to water contact but susceptible to electromagnetic interference. The detection principles of the two sensors are completely different, and their detection blind zones overlap; only by using them simultaneously can comprehensive monitoring be achieved. Second, the mutual verification of signal characteristics: When the float-type sensor triggers but the immersion sensor does not, it can be determined that the float is mechanically faulty; when the immersion sensor triggers but the float-type sensor does not, it can be determined that there is electromagnetic interference or the water level has not reached the threshold. This cross-verification mechanism cannot be achieved by a single sensor and must rely on the coordinated work of the two sensors. Third, the fundamental guarantee of decision reliability: The hierarchical state transition model of this invention relies on the comprehensive analysis of signals from two sensors. The absence of either sensor will result in incomplete information in the subsequent left-side signal feature vector SL and right-side signal feature vector SR, thus rendering the calculation of the cross-side consistency index CLR meaningless and ultimately causing the scene confidence score to fail. Therefore, the combined use of the float-type water sensor and the water immersion sensor is not an optional configuration, but a fundamental prerequisite for the implementation of the technical solution of this invention. In contrast, if only the float-type water sensor is used, the system will lose the ability to continuously monitor the degree of water contact and will be unable to distinguish between instantaneous splashing and continuous immersion; if only the water immersion sensor is used, the system will lose the ability to clearly determine the water level threshold and will be unable to determine the danger level. Only when both sensors work simultaneously can the accurate water wading state identification and decision-making described in this invention be achieved. It should be noted that the sensors used in this invention for vehicle immersion monitoring are not limited to float-type water sensors, immersion sensors, and counterweight switch devices. Other types of sensors, such as ultrasonic level sensors, can be added according to the actual usage scenarios of the vehicle, such as off-road vehicles, urban commuter vehicles, and vehicles with high frequency of water wading, in order to expand the monitoring dimensions, improve data redundancy, and enhance adaptability to complex environments.

[0070] Step S12: Based on the multi-source sensor timing acquisition framework, acquire the original sensor signal sequences with timestamps from all sensors, and perform targeted filtering on the original sensor signal sequences with timestamps to obtain the filtered sensor signal sequences.

[0071] Further, step S12 includes:

[0072] Step S121: Separate the switching signal of the float-type water sensor, the conductivity analog signal of the water immersion sensor, and the mechanical trigger signal of the counterweight switch from the original sensor signal sequence with timestamps.

[0073] Step S122: Apply a debouncing filtering algorithm to the switch signal to obtain the trigger time and duration of each float-type water sensor.

[0074] Step S123: Apply a medium-range filtering algorithm to process the conductivity analog signal to generate a filtered conductivity value. When the filtered conductivity value exceeds a preset conductivity threshold, record the trigger time and duration of the water immersion sensor.

[0075] Step S124: Perform state holding filtering on the mechanical trigger signal to confirm the triggering time and duration of the counterweight switch;

[0076] Step S125: Integrate the trigger time and duration of each sensor to form a complete filtered sensor signal sequence.

[0077] Each sampling data point in the timestamped raw sensor signal sequence uniformly contains four core data categories: "unique sensor identifier, sampling timestamp, raw signal value, and supplementary remarks." The raw signal value matches the sensor's physical characteristics, while the supplementary remarks record key environmental or state information during signal acquisition. As shown in Table 2, the raw signal of the float-type water sensor is a mechanically triggered switching signal, which is a discrete signal with only two stable states and no intermediate transition values. The core signal is generated by the float rising / sinking, causing the waterproof microswitch to open and close. It reflects a binary state of "whether the float trigger water level has been reached." The raw signal value contains only two distinct level values: a low level (0V) records the raw signal value as "0," indicating that the float has not risen, the waterproof microswitch is open, and there is no effective trigger signal. A high level, such as 5V, records the raw signal value as "1," indicating that the float has risen, the waterproof microswitch is closed, and an effective trigger signal is generated. The original signal from the water immersion sensor is a continuous analog conductivity signal, whose value changes linearly with the water's conductivity. The core of the signal is generated by converting the conductivity change after the probe contacts the water into an analog voltage, reflecting the continuous state of water contact and conductivity magnitude. The analog voltage value (unit: V) is the direct original signal value, and the voltage range has a linear relationship with conductivity. When not in contact with water, the original signal value is approximately "0.1V", corresponding to a conductivity of ≈5μS / cm, which is the basic conductivity in air. As the water level rises and the water conductivity increases, the voltage value increases synchronously. The original signal from the counterweight switch device is a discrete inertial mechanical trigger switch signal. The core of the signal is generated by the displacement of a spherical counterweight with abrupt changes in vehicle posture, pressing and switching a microswitch on and off, reflecting a binary state indicating whether the vehicle is in an abnormal posture. The original signal value is also represented by a level value, corresponding to the displacement of the spherical hammer and the on / off state of the switch. A low level (0V) indicates a "0" original signal value, meaning the vehicle's posture is normal, the spherical hammer has not moved, and it is not pressing the microswitch. A high level, such as 5V, indicates a "1" original signal value, meaning the vehicle's posture is abnormal, such as rollover or overturning, and the spherical hammer has moved due to inertia, pressing the microswitch to close. A synchronous sampling mode triggered by a hardware timer is used, with the reference sampling frequency set to 100Hz. At the beginning of the i-th sampling period, the main control circuit board simultaneously triggers the sampling actions of all sensors through an interrupt signal, and adds a microsecond-level timestamp to each sampled data.

[0078] Table 2 Sensor Signal Type Table

[0079] Sensor type Signal type Raw signal characteristics Float-type water sensor On-off signal Binary level Water immersion sensor Conductivity analog signal Continuous voltage value Weighted switch device Mechanical trigger on-off signal Binary level

[0080] For example, at the sampling time of 2025-05-06 15:20:30.000, when the vehicle was initially in a non-water-crossing state, the original signal sequence data of the left front float water sensor FLF was "Identifier: FLF, Timestamp: 2025-05-06 15:20:30.000, Original Signal Value: 0 (Low Level, 0V), Supplementary Remarks: Not Triggered, Float Not Floating, Waterproof Microswitch in the Off State"; the original signal sequence data of the left front water immersion sensor WLF was "Identifier The original signal sequence data for the crew compartment counterweight switch device HG is: "Identifier: HG, timestamp: 2025-5-06 15:20:30.000, original signal value: 0.1V, supplementary remarks: not triggered, probe not in contact with water, vehicle power supply voltage 12.3V";

[0081] The switching signal of the float-type water sensor, the conductivity analog signal of the water immersion sensor, and the mechanical trigger signal of the counterweight switch are separated from the original sensor signal sequence with timestamps. For the switching signal of the float-type water sensor, a debouncing filtering algorithm is applied. Specifically, based on the sampling timestamp, five sampling points are continuously collected. A valid trigger is determined only when at least four sampling points are high level. The trigger time and duration are recorded. The trigger time is the valid trigger start timestamp, and the duration is the time interval from the trigger time to the current time or the time when the signal recovers to a non-triggered state. The selection of 5 sampling points and the setting of "at least 4 high levels" as the judgment condition is due to the mechanical response characteristics of the float-type water sensor. The float needs to overcome its own weight and switching resistance to float, and during this process, it may experience momentary jitter due to water flow impact, resulting in false high levels. Monitoring of 5 consecutive sampling points can cover the typical duration of jitter, 10-20ms. The "at least 4 high levels" threshold can filter out single or two momentary jitter interferences while ensuring response speed, avoiding misjudging the brief displacement of the float caused by water flow impact as a valid trigger. For the conductivity analog signal of the water immersion sensor, the analog voltage signal is first converted into a digital signal by ADC conversion, and then a medium-range filtering algorithm is applied. Specifically, 7 consecutive sampling values ​​are taken, the maximum and minimum values ​​are removed, and the average value is calculated to obtain the filtered conductivity value. When the filtered conductivity value exceeds the preset conductivity threshold, the trigger time and duration are recorded. Seven sampled values ​​were chosen for median filtering because seven is an odd number, making it easier to remove the maximum and minimum values ​​and retain five valid data points. This effectively suppresses pulse noise caused by water flow impact and electromagnetic interference while preserving the continuous trend of conductivity changes with water level. The conductivity threshold was determined by collecting the average conductivity values ​​of various common water-related scenarios, such as rainwater, river water, and stagnant water. Half of this average value was taken as the conductivity threshold to ensure triggering as soon as the water touches the probe, while avoiding false triggering caused by the baseline conductivity in the air. For the mechanical trigger signal of the hammer switch, a state-holding filter was implemented. Specifically, based on the sampling timestamp, the state switch was only confirmed and the trigger time and duration were recorded when the signal state (high level / low level) remained unchanged for 30 consecutive sampling cycles (corresponding to 300 milliseconds). The selection of 30 sampling periods is based on the inertial characteristics of the counterweight switch. Bumps during normal vehicle operation cause brief displacement of the counterweight, with the duration of such displacement typically less than 200 milliseconds. The 300-millisecond state retention time filters out false triggers caused by bumps, while ensuring accurate capture of sustained abnormal postures such as rollovers and overturning, avoiding subsequent decision-making errors due to attitude misjudgment. The trigger times and durations of each sensor are integrated to form a complete filtered sensor signal sequence. During integration, the filtered data from each sensor is associated with its unique identifier and sampling timestamp, forming a structured data format of "identifier-timestamp-trigger time-duration-signal type".Structured data can also enable signal tracing. When abnormal signals are found in subsequent analysis, they can be traced back to specific sensors through identifiers to check for sensor failures and improve the maintainability of the system.

[0082] The multi-source sensor timing acquisition framework constructed in step S11 provides a physical location basis for the spatial grouping (left / right sensor group) in step S20. Without this framework, step S20 cannot effectively group the sensor signals, and the cross-side consistency index calculation will lose its foundation. The targeted filtering process in step S12 provides reliable data for the feature extraction in step S20. If the original signal is not filtered, the signal containing jitter and pulse interference will cause the feature extraction result to be distorted, which will cause the hierarchical state transition model in step S30 to fall into decision confusion. For example, the un-de-jittered float signal may be misjudged as a valid trigger, resulting in confusion of the feature vectors of the side-overturning shallow water scene and the splashing interference scene. Step S10 achieves full-dimensional monitoring of "vehicle wading status + vehicle attitude" through multi-source sensor layout and labeling rules. This effectively avoids missed detections due to the limitations of single-position sensor monitoring, such as unilateral tilting into the water. Simultaneously, it ensures the effective correlation between wading scenarios and vehicle attitude information, compensating for information gaps caused by missing attitude monitoring and providing a more comprehensive monitoring dimension for subsequent scene recognition. Targeted filtering algorithms suppress float jitter, water immersion pulse interference, and false triggering by the weighted hammer, significantly improving the reliability of the original sensor signals. This provides high-quality data support for feature extraction in step S20, preventing signal distortion from affecting the accuracy of feature extraction results and reducing the possibility of misjudgments in subsequent hierarchical state transition models due to data errors. Structured data integration provides a standardized data interface for subsequent steps, avoiding low processing efficiency caused by chaotic data formats.

[0083] Step S20: Perform spatiotemporal and trend analysis on the filtered sensor signal sequence to obtain the left signal feature vector, the right signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics;

[0084] Step S20 aims to perform multi-dimensional feature extraction and consistency analysis on the filtered sensor signal sequence, transforming the discrete, time-series-attributed sensor data output from step S12 into a structured, quantifiable feature set, providing core input for constructing the hierarchical state transition model in step S30. This step addresses the shortcomings of existing technologies that rely solely on a single sensor trigger state while ignoring signal temporal sequence, spatial distribution, and trend characteristics, resulting in an inability to distinguish complex wading scenarios, such as full vehicle immersion versus single-sided splashing. Through the collaborative extraction of multi-dimensional features, it lays a data foundation for subsequent accurate state identification.

[0085] Further, step S20 includes:

[0086] Step S21: Based on the filtered sensor signal sequence and the installation position of each sensor, construct the left signal feature vector and the right signal feature vector;

[0087] See Figure 5 Furthermore, step S21 includes:

[0088] Step S211: The filtered sensor signal sequence is spatially grouped according to the physical installation position of the sensor on the vehicle, into a left sensor group and a right sensor group.

[0089] Step S212: Extract the trigger time and duration of each sensor in the left sensor group to form the left signal feature vector;

[0090] Step S213: Extract the trigger time and duration of each sensor in the right sensor group to form the right signal feature vector.

[0091] Step S211 first spatially groups the sensors according to their physical installation positions on the vehicle, dividing them into a left-side sensor group and a right-side sensor group. The grouping is based on the unique sensor identifier assigned in S11: the left-side sensor group includes the left front float-type water sensor FLF, left front water immersion sensor WLF, left rear float-type water sensor FLR, and left rear water immersion sensor WLR; the right-side sensor group includes the right front float-type water sensor FRF, right front water immersion sensor WRF, right rear float-type water sensor FRR, and right rear water immersion sensor WRR. The reason for dividing them into left and right-side sensor groups is that dynamic wading events often exhibit spatial differences, such as side-entry into water (e.g., rollover) or side-entry into water (e.g., normal full-vehicle immersion). Grouping by position allows for direct capture of these differences, avoiding the loss of spatial information caused by disordered sensor signal arrangement. Existing technologies often fail to spatially group sensor signals, relying solely on global AND / OR logic for judgment. This approach cannot identify asymmetrical scenarios such as unilateral tilting into water. In contrast, this grouping method clarifies spatial attribution, enabling subsequent feature extraction to focus on the correlation between unilateral signal characteristics and signals from both sides. This provides a spatial dimension for analyzing the difference between unilateral splashing and side-tipping into water.

[0092] Each sensor's state information must completely include three core elements: a trigger flag (0 indicates no trigger, 1 indicates trigger), trigger time, and duration. For example, the left-hand signal feature vector can be represented as SL={(FLF: trigger flag=1, trigger time=TFLF, duration=DFLF),(WLF: trigger flag=1, trigger time=TWLF, duration=DWLF),(FLR: trigger flag=1, trigger time=TFLR, duration=DFLR),(WLR: trigger flag=1, trigger time=TWLR, duration=DWLR)}. The right-hand signal feature vector SR has the same structure as SL, except that the sensor identifier corresponds to the right-hand sensor. This structured vector construction breaks through the limitations of existing technologies that only focus on the binary state of "whether it is triggered," by incorporating the trigger time and duration, giving the feature vector a temporal attribute. The triggering time reflects the spatial synchronicity of the signal, such as whether the sensors on both sides trigger simultaneously. The duration can distinguish between transient interference, such as short-duration triggering caused by splashing water and continuous danger, and long-duration triggering caused by immersion in water. The combination of these two factors allows the feature vector to characterize the dynamic evolution of the signal, rather than a static snapshot. If only the triggering flag is included, the system cannot distinguish between "simultaneous triggering of four sensors on one side for 1 second" and "simultaneous triggering of four sensors on one side for 0.1 seconds." The former may be due to a side-climbing into water, while the latter may be due to splashing water interference. The feature vector in step S21, by incorporating timing information, directly solves this distinction problem, providing necessary timing data support for subsequent cross-side consistency analysis and signal trend analysis.

[0093] In step S21, a signal quality score Q is calculated for each sensor. This score is calculated by dividing the sensor's duration by the total observation time, then multiplying by the reciprocal of the number of signal changes. The total observation time refers to the time interval from the start of monitoring to the current moment, and the number of signal changes refers to the number of times the sensor switches from triggered to non-triggered or vice versa within the observation time. For example, the signal quality score of the left front float sensor... , among which, T obs N represents the total observation time. FLF D represents the number of signal changes in the FLF. FLF This represents the duration of the FLF (Fluid Form). Duration percentage. This is used to eliminate "transient interference," such as a 0.1-second trigger caused by splashing water. Only signals with a high proportion of continuous triggers are likely to correspond to real wading events. The more switching times, such as frequent on / off cycles caused by water flow impact, the worse the stability. (Inverse form) "Stability" can be converted into a coefficient between 0 and 1, working in conjunction with the duration ratio. If only duration is considered, fault signals that "frequently switch but have a total duration exceeding the target" might be misjudged as valid; conversely, if only the number of changes is considered, interference signals that "continuously trigger but have an extremely low percentage" might be overlooked. This two-dimensional combination can cover more abnormal scenarios. The signal quality score ranges from 0 to 1; the closer to 1, the more stable the signal, indicating a high duration ratio and few switches. The signal quality score is introduced because sensor signals may become unstable due to mechanical faults, such as a stuck float, or environmental interference, such as frequent switching caused by water flow. Directly incorporating such signals into subsequent analysis can easily lead to misjudgments. Existing technologies do not quantify signal quality and often treat faulty signals the same as normal signals, leading to decision-making bias. This step, however, uses signal quality scoring to preprocess and assess the reliability of sensor signals. Subsequent steps, such as the confidence calculation in S31, can assign differentiated weights to different sensor signals based on this score. For example, sensor signals with low signal quality scores have reduced weights in the confidence calculation, thus avoiding interference from unstable signals in decision-making.

[0094] Step S22: Calculate the cross-side consistency index using the left-side signal feature vector and the right-side signal feature vector;

[0095] Further, step S22 includes:

[0096] Step S221: Extract the trigger times of similar sensors from the left signal feature vector and the right signal feature vector respectively, and calculate the absolute value ΔTF of the difference between the average trigger times of the left float-type water sensor and the average trigger times of the right float-type water sensor, as well as the absolute value ΔTW of the difference between the average trigger times of the left water immersion sensor and the average trigger times of the right water immersion sensor.

[0097] Step S222: Convert ΔTF and ΔTW into time consistency coefficients using a Gaussian function;

[0098] Step S223: Compare the trigger combination modes of the sensors in the left signal feature vector and the right signal feature vector to obtain the consistency of the trigger modes;

[0099] Step S224: Multiply the time consistency coefficient by the trigger mode consistency to obtain the cross-side consistency index.

[0100] The Cross-Side Consistency Index (CLR) is used to quantify the synchronization of sensor signals on both sides in terms of timing and mode. Its core solution addresses the technical challenge of existing technologies being unable to distinguish between "full vehicle immersion" and "single-side splashing." When the entire vehicle is submerged, the signals on both sides should exhibit high consistency; when only one side is splashing, the consistency is low. The CLR directly quantifies this difference. Step S221 first extracts the trigger times of similar sensors, calculating the absolute value ΔTF of the difference between the average trigger times of the left and right float-type water sensors, and the absolute value ΔTW of the difference between the average trigger times of the left and right water immersion sensors. Choosing to average the values ​​of similar sensors before calculating the differences, rather than directly comparing individual sensors, avoids misjudgments of timing differences caused by a single sensor failure. For example, if the left front float sensor triggers late due to a fault, comparing only the FLF and FRF times would lead to a misjudgment of a large timing difference. Taking the average of the two left floats smooths out anomalies in individual sensors, improving the robustness of timing difference calculation and reflecting the actual scenario where sensor signals may have local anomalies in dynamic wading events.

[0101] Step S222 converts ΔTF and ΔTW into time consistency coefficients Ct using a Gaussian function. The specific form of the Gaussian function is as follows: ,in, Let T be an exponential function with base e of the natural logarithm. ref The reference time window was determined based on statistical analysis of a large amount of water immersion test data: the time differences between two typical scenarios—normal full vehicle immersion and single-sided splashing—were collected, and the critical value of the time difference distribution between the two scenarios was taken as T. ref This ensures that under normal whole-vehicle immersion scenarios, ΔTF and ΔTW are small, and Ct is close to 1; under unilateral splashing scenarios, ΔTF and ΔTW are large, and Ct is significantly reduced. For example, T... ref By collecting time-series difference data from over 100 sets of normal vehicle immersion and single-sided splash scenarios, the critical values ​​of ΔTF and ΔTW distributions for the two scenarios were statistically analyzed, and the median of the two critical values ​​was taken as T. ref For example, 500ms, but the specific value needs to be adjusted based on the sensor installation spacing and wading dynamics characteristics of the actual vehicle model. The core is to ensure that Ct can accurately characterize the degree of timing synchronization. The technical reason for choosing the Gaussian function is that the Gaussian function has the characteristics of continuous smoothness and monotonically decreasing. When ΔTF and ΔTW are 0 (completely synchronized), Ct=1; as ΔTF and ΔTW increase, Ct gradually decreases without abrupt changes, which can accurately match the gradual change characteristics of signal timing differences in dynamic wading events. If a step function is used, it is easy to be judged as completely out of sync due to minute timing differences, leading to misjudgment. The smoothness of the Gaussian function effectively avoids this problem.

[0102] Step S223 compares the trigger combination modes of the signal feature vectors on the left and right sides and calculates the trigger mode consistency (MLR). The trigger combination mode refers to the combination of trigger states of each sensor in the sensor group. For example, if the left mode is "FLF trigger, WLF trigger, FLR trigger, WLR trigger" and the right mode is "FRF trigger, WRF trigger, FRR trigger, WRR trigger", then the trigger combination modes are completely identical, and MLR=1. If the left mode is "FLF trigger, WLF trigger, FLR not triggered, WLR not triggered" and the right mode is "FRF trigger, WRF not triggered, FRR trigger, WRR not triggered", then the number of sensors with the same trigger is 2 (FLF and FRF, FLR and FRR), and the total number of sensors is 4, so MLR=2 / 4=0.5. If there are no identical trigger sensors on either side, then MLR=0. The logic behind introducing trigger mode consistency lies in the fact that time consistency only reflects timing synchronization and cannot reflect whether the spatial distribution of signals is reasonable. For example, if two sensors trigger simultaneously (Ct=1), but the left side is a dual trigger of "float + water immersion" while the right side is a single trigger of "water immersion only," then the timing is synchronized but the mode is abnormal. This may actually be due to a malfunction of the right-side float sensor, rather than normal vehicle immersion. Existing technologies do not consider trigger mode consistency and rely solely on timing synchronization, which can easily misjudge such abnormal scenarios as normal. This step, however, uses MLR to quantify the mode matching degree and works in conjunction with Ct to form a comprehensive assessment of cross-side consistency. This ensures that cross-side consistency is high only when the timing is synchronized and the mode matches, avoiding the limitations of a single-dimensional judgment.

[0103] Step S224 multiplies the time consistency coefficient by the trigger mode consistency coefficient to obtain the cross-side consistency index CLR. The technical logic of multiplication is that Ct and MLR are two independent and necessary dimensions of cross-side consistency. Only when both are high can the signals on both sides be determined to be truly consistent: if Ct is high but MLR is low, it may be due to sensor failure or local interference; if MLR is high but Ct is low, it may be the initial stage of unilateral tilting into water; only when both are high does it conform to the signal characteristics of normal whole vehicle immersion in water. This multiplication combination allows CLR to comprehensively reflect the synergistic effect of the two dimensions, avoiding misjudgment of high cross-side consistency due to the achievement of a single dimension. The cross-side consistency index quantifies the consistency of the signals on both sides and provides key input for the scenario confidence calculation in step S31.

[0104] Step S23: Perform trend analysis on the filtered sensor signal sequence within the time window to obtain the trend characteristics of the signal change rate.

[0105] Step S23 performs trend analysis on the filtered sensor signal sequence within a time window to obtain the signal change rate trend feature. This feature is used to characterize the dynamic evolution direction of the sensor signal, such as water level rise, stabilization, and decline. The core solution addresses the deficiency of existing technologies that only focus on the current signal state and cannot predict the development trend of danger, leading to missed early warning opportunities. First, a system is established to store the most recent time period T. buf The circular buffer of the sensor signal sequence after internal filtering, T buf The determination is based on the typical timescale of water level changes in water-related incidents. Through experiments collecting data on different water-related scenarios, such as rapid falls into the river and slow immersion, the minimum time required to cover the "initial trigger to escalation of danger" phase is determined. This ensures the buffer can store sufficient historical data for trend analysis while avoiding excessively long buffers that could lead to data redundancy and reduced real-time performance. For example, T... buf The timeout can be set to 10 seconds, but the specific value needs to be adjusted based on the vehicle's wading risk level. For high-risk vehicles, such as SUVs, the timeout can be appropriately extended to capture a slower immersion process. The circular buffer stores the left-side signal feature vector (SL), the right-side signal feature vector (SR), the signal quality scores Q for each sensor, and the cross-side consistency index (CLR), stored in timestamp order. New data overwrites the oldest expired data, ensuring the buffer always maintains the latest T. buf Durational data provides real-time updated historical data for trend analysis.

[0106] Subsequently, the historical trigger state sequence of each sensor is extracted from the circular buffer, and the signal change rate of the float-type water sensor is calculated. Taking the left front float sensor as an example, its signal change rate RFLF is calculated as follows: extract the trigger state SFLF(t) at the current moment and t-ΔT. trend Triggering state SFLF(t-ΔT) at time t trend The difference between the two is RFLF = SFLF(t) - SFLF(t - ΔT). trend ), where ΔT trend The trend analysis time interval is determined based on the response characteristics of the float sensor and the minimum time for state changes in wading scenarios. It must be greater than the de-jitter filtering time in step S12, such as 10-20ms, to avoid misjudging jitter as a state change, and less than T. buf 1 / 5, such as T buf =10 seconds, ΔT trend=2 seconds, ensuring the capture of moderate-rate state changes. The trigger state is a binary discrete value of 0 or 1, where a value of 1 indicates that the float sensor has been triggered, and a value of 0 indicates that the float sensor has not been triggered. The meaning of RFLF values ​​is as follows: a positive value indicates that the sensor has changed from "not triggered" to "triggered", i.e., a new trigger; a negative value indicates that it has changed from "triggered" to "not triggered"; and zero indicates that the state remains unchanged. The purpose of calculating the float signal change rate is that the float sensor is a switching signal, and its state change directly reflects whether the water level has reached the float trigger threshold. The float signal change rate can determine whether the water level is continuously rising or falling, providing a switching dimension basis for water level trend judgment.

[0107] For water immersion sensors, the signal is a continuous conductivity signal. The conductivity value sequence needs to be fitted using the least squares method, and the derivative of conductivity with respect to time needs to be calculated as the signal rate of change. Taking the left front water immersion sensor as an example, the most recent ΔT is extracted from the circular buffer. fit A sequence of conductivity values ​​over a time period, where ΔT fit To fit the time window, and ΔT trend Similar values ​​are used to ensure consistency in the time scale of trend analysis. The least squares method is used to fit the conductivity value sequence as a linear function. The slope of this linear function, i.e., the derivative of conductivity with respect to time, is used as the signal change rate RWLF of the left front water immersion sensor. The reason for choosing the least squares method is that the conductivity signal of the water immersion sensor is easily affected by water flow fluctuations and electromagnetic interference, which generate impulse noise. Directly calculating the difference between adjacent time points will amplify the noise, resulting in drastic fluctuations in the change rate and failing to reflect the true trend. The least squares method, by fitting a linear function, can smooth out the noise, extract the overall trend of the signal change, and make the signal change rate more stable and reliable. The sign of the signal change rate of the water immersion sensor directly reflects the direction of change in conductivity: a positive value indicates an increase in conductivity, i.e., a rise in water level and an increase in the water contact area; a negative value indicates a decrease in conductivity, i.e., a drop in water level; and zero indicates stable conductivity and unchanged water level. This provides a continuous dimension for judging water level trends and complements the signal change rate of the float sensor. The float sensor's signal change rate reflects whether the water level has reached the float trigger threshold, while the water immersion sensor's signal change rate reflects the rate of change. The combination of the two makes trend judgment more comprehensive.

[0108] Finally, based on the positive / negative and increasing characteristics of the rate of change of all sensor signals, the water level change trend is determined. The specific determination logic is as follows: when the rate of change (RW) of all water immersion sensors is consistently positive and increasing, and the rate of change (RF) of the float sensors is mostly positive (meaning the number of newly triggered floats is greater than the number of recovered floats), it is determined as a "water level rising trend." When all RW oscillates slightly around zero, with a fluctuation range less than the set stability threshold, and all RF are zero, it is determined as a "stable immersion state." The stability threshold is determined by collecting conductivity fluctuation data from 50+ sets of normal stable immersion scenarios and taking the 95th percentile as the stability threshold. When all RW is consistently negative, and most RF are negative (meaning the number of recovered floats is greater than the number of newly triggered floats), it is determined as a "water level falling or drainage state." This trend indicator can predict the direction of danger development in advance: for example, a "water level rising trend" indicates that the danger is escalating, requiring accelerated self-rescue actions; a "stable immersion state" indicates that the current danger is stable, and existing self-rescue measures can be maintained; a "water level falling state" indicates that the danger is easing, and a downgrade can be considered. Existing technologies cannot predict such trends and often only trigger actions when the danger intensifies to a certain extent, missing the golden escape window. However, the trend characteristics of this step provide a key basis for the hysteresis control in step S40. For example, under the "water level rising trend", the system will shorten the fusion window and accelerate the transition from the suspected state to the water danger state, avoiding delayed decision-making.

[0109] Step S20 constructs a complete feature system through multi-dimensional feature extraction and consistency analysis, encompassing spatial (left and right feature vectors), temporal (cross-side consistency indicators), and trend (signal change rate trend features). This feature system overcomes the limitations of single features in existing technologies. Through multi-dimensional collaboration, the system can characterize dynamic wading events from different perspectives. Spatial features capture differences between unilateral and bilateral wading, temporal features capture differences between synchronous and asynchronous wading, and trend features capture differences in development direction. The combination of these three features forms a comprehensive description of the essence of the event, effectively overcoming the "feature confusion trap," such as distinguishing between side-lying shallow water and unilateral splashing: when side-lying shallow water, the unilateral feature vector is continuously triggered, and the trend is stable immersion; when unilateral splashing occurs, the unilateral feature vector is briefly and continuously triggered, and the trend is downward. The feature comparison clearly distinguishes features; the integration of signal quality scoring and robust calculation methods during feature extraction makes the features resistant to interference, avoiding feature distortion caused by sensor failure or environmental interference, and improving the reliability of subsequent decisions; this step provides standardized, high-information-density input for steps S30 and S40, enabling the hierarchical state transition model to calculate scene confidence based on quantified features, and scene adaptive control to dynamically adjust decision strategies based on trend features, realizing a logical closed loop of "feature extraction - state recognition - action execution". If step S20 is missing, step S30 will be unable to build a confidence model due to the lack of effective feature input, and step S40 will fall into static decision-making due to the lack of trend basis, and the dynamic adaptability and accuracy of the entire scheme will be completely lost.

[0110] Step S30: Based on the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics, construct a hierarchical state transition model;

[0111] The core of step S30 lies in transforming multi-dimensional sensing features into a quantifiable and transferable state recognition system, solving the problem that existing technologies, which rely on single signals or simple logical combinations, cannot distinguish complex wading scenarios. Existing technologies often fall into decision-making dilemmas due to the similarity of signal features, such as being unable to distinguish between splashing water on one side and capsizing in shallow water. This step, however, achieves accurate characterization of dynamic wading events through confidence quantification and state machine transitions.

[0112] Further, step S30 includes:

[0113] Step S31: Combining the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics, construct a confidence scoring model for four immersion scenarios and calculate the confidence values ​​for the four immersion scenarios; the four immersion scenarios include normal water entry scenario, side-overturning shallow water scenario, upside-down water scenario, and splashing interference scenario;

[0114] Step S31 combines the left-side signal feature vector SL, the right-side signal feature vector SR, the cross-side consistency index CLR, and the signal change rate trend characteristics to construct confidence scoring models for four immersion scenarios and calculate the corresponding confidence values. The four scenarios include normal immersion, side-crawling in shallow water, inverted in water, and splashing interference. The confidence calculation for each scenario is designed based on its unique combination of signal features to ensure that the quantification results highly match the essential attributes of the scenario.

[0115] Confidence level R for normal water entry scenario in The calculation is based on high cross-side consistency as the core criterion, because when the entire vehicle is uniformly submerged in water, the triggering states of the sensors on both sides should exhibit high synchronization. When the cross-side consistency index CLR is greater than the preset high consistency threshold, and the trigger flags of all float sensors and water immersion sensors in the left signal feature vector SL and the right signal feature vector SR are both 1, and the duration D of each sensor exceeds the preset stable triggering duration, then R... inThe high consistency threshold is equal to the CLR multiplied by the geometric mean of all sensor signal quality scores. The stable trigger duration is determined based on the minimum time for the water to stably submerge the sensor, and must be greater than the duration of instantaneous interference such as splashing. The high consistency threshold is determined by collecting CLR data from 500+ sets of scenarios: uniform vehicle immersion (target scenario) and single-sided / asymmetrical immersion (interference scenario). The critical values ​​of the CLR distribution for both scenarios are statistically analyzed, and the median value between the "minimum CLR value for the target scenario" and the "maximum CLR value for the interference scenario" is taken as the high consistency threshold. Example: The high consistency threshold is set to 0.85 to ensure that normal immersion scenarios with only simultaneous triggering on both sides meet the requirements. The geometric mean is used instead of the arithmetic mean because the signal quality score reflects sensor reliability, and the geometric mean is more sensitive in penalizing low-quality signals, preventing a single abnormal sensor from raising the overall score. The cross-side consistency index CLR directly participates in the calculation, ensuring that scenarios with poor signal synchronization on both sides are not misjudged as normal immersion. The introduction of the signal quality score filters out false triggers caused by sensor failures. The two work synergistically to ensure that R... in It can accurately reflect the likelihood of the entire vehicle entering the water stably.

[0116] Confidence R of the side-capsizing scenario in shallow water side The calculation is based on the scenario where only one side of the vehicle body is in contact with water and the vehicle has not overturned. When all sensor trigger flags in only the left signal feature vector SL or only the right signal feature vector SR are 1, and the corresponding duration D exceeds the stable trigger duration, while only some or no sensors in the other side signal feature vector are triggered, and the trigger flag of the counterweight switch signal HG is 0, meaning the vehicle has not overturned or engaged in other extreme postures, then R... side This is equal to the average of the signal quality scores of all sensors on the triggering side multiplied by the proportion of sensors that did not trigger on the non-triggering side. The average signal quality score on the triggering side ensures the stability of the signal on that side and avoids misjudgment due to transient interference; the proportion of sensors that did not trigger on the non-triggering side quantifies the asymmetry of the signals on both sides. A higher proportion indicates a more obvious unilateral water inflow characteristic. For example, if all sensors on the right side did not trigger, this proportion is 1. side If some sensors on the right side are triggered when the maximum value is reached, the ratio decreases, R side The corresponding decrease. This calculation method effectively distinguishes between side-dive shallow water, normal water entry, and single-sided splashing, because in normal water entry, signals from both sides are triggered, while in single-sided splashing, the duration is insufficient; the condition of the counterweight switch signal being 0 eliminates interference from the inverted scenario, making R... side It can accurately depict scenarios where water enters steadily on one side and the posture is not extremely abnormal.

[0117] Confidence R of the inverted underwater scene invertThe calculation is based on the special condition of a vehicle overturned. In this situation, the water immersion sensor is triggered because the vehicle body may come into contact with water due to the inverted position. However, the float sensor, due to its unidirectional triggering design, although the float still moves upwards under buoyancy when the vehicle is overturned, the direction of the float's displacement is opposite to the switch's triggering direction because the triggering mechanism of the waterproof microswitch has reversed 180 degrees. Therefore, an effective mechanical trigger cannot be formed, and the trigger flag is 0. Simultaneously, the counterweight switch is continuously triggered due to the drastic change in attitude. When all water immersion sensor trigger flags are 1 and the float sensor trigger flag is 0, and the duration DHG of the counterweight switch exceeds the stable triggering duration, then R... invert The ratio is equal to DHG divided by the observation time and then multiplied by the average of all water immersion sensor signal quality scores. The ratio of DHG to observation time reflects the continuity of the counterweight switch trigger state; a higher ratio indicates a more stable vehicle tilting posture. The average water immersion sensor signal quality score ensures reliable water contact. This combination utilizes the state contrast between the float and water immersion sensors during tilting, and also incorporates the counterweight switch's attitude signal, effectively distinguishing between tilting, normal water entry, and rollover. This is because during normal water entry, the float-type water sensor triggers, while during rollover, the counterweight switch may not trigger or may not trigger for a sufficient duration, causing R... invert It can accurately identify scenarios where vehicles are overturned and submerged in water.

[0118] Confidence R of water splash interference scenario splash The calculation is based on the instantaneous signal triggering caused by water flow impact. When the signal change rate of any water immersion sensor frequently alternates between positive and negative and the amplitude exceeds the fluctuation threshold, while the corresponding duration is less than the duration of the instantaneous interference, then R... splash The fluctuation threshold is calculated by dividing the number of signal changes by the observation duration and then multiplying by the reciprocal of the average duration. The fluctuation threshold is determined by collecting the conductivity change rate of water immersion sensors in over 100 splash scenarios and using the 95th percentile as the fluctuation threshold. The instantaneous interference duration is calculated by statistically analyzing the average duration of splash triggers and using 1.2 times that duration as the threshold (e.g., 0.8 seconds, less than the stable trigger duration, such as 2 seconds), to distinguish between instantaneous interference and continuous immersion. The average duration is calculated by extracting the single-round trigger duration data from all trigger sensors in the splash scenario, i.e., the time interval from triggering to non-triggering for each sensor, summing all single-round trigger durations, and dividing by the total number of triggers. The number of signal changes divided by the observation duration reflects the frequency of signal fluctuations; a higher ratio indicates more severe interference. The reciprocal of the average duration assigns higher weight to short-duration signals because splash-induced triggers are typically short-lived. This calculation method effectively distinguishes between splash and stable immersion by leveraging the high-frequency fluctuations and short-duration characteristics of the signal, making R... splashIt can accurately identify interfering scenarios and prevent them from being misjudged as dangerous states. The confidence scores of the four scenarios transform qualitative scenario features into quantitative indicators between 0 and 1, enabling the system to clearly identify the scenario with the highest confidence score through numerical comparison when multiple scenario features coexist, thus avoiding decision ambiguity caused by feature confusion.

[0119] Step S32: Based on the confidence values ​​of the four water immersion scenarios, establish a three-state master state machine containing three main states: normal state, suspected state, and water danger state. Nest the scenario sub-state diagram under the water danger state to form a hierarchical state transition model.

[0120] The three-state master machine is designed following a logical progression from no danger to potential danger to confirmed danger, conforming to the evolutionary laws of dynamic events. The normal state indicates the vehicle is not in water, at which point the confidence levels of all scenarios are below the low confidence threshold (determined by the background noise characteristics during normal driving). The suspected state indicates a detected water crossing signal but no confirmation; the transition condition is that the confidence level of any scenario exceeds the low confidence threshold. This state provides a buffer for observation and judgment, preventing the system from directly entering a dangerous state due to instantaneous signals. The water hazard state indicates the vehicle is confirmed to be in a dangerous water crossing state; the transition condition is that the confidence level of the instantaneous leading scenario (highest confidence value) exceeds the high confidence threshold and its duration is greater than the first duration threshold, such as 1 second, ensuring the stability of the danger judgment. Returning from the water hazard state to the suspected state requires that the confidence levels of all scenarios are below the medium confidence threshold and their duration is greater than the second duration threshold. The second duration threshold is greater than the first duration threshold, forming a hysteresis characteristic of "easy to enter, difficult to exit," avoiding frequent switching between dangerous states. The method for setting the low confidence threshold is as follows: Collect confidence data from 1000+ sets of scenarios where vehicles are driving normally, i.e., without water wading or interference. Calculate the 95th percentile of all data as the threshold to filter false signals caused by background noise such as road bumps and electromagnetic interference. For example, a low confidence threshold of 0.3 is set to ensure that the vehicle only enters a suspected state when a genuine water-related signal is detected. The method for setting the high confidence threshold is as follows: Collect instantaneous leading scenario confidence data from 500+ sets of real dangerous water-wading scenarios (normal entry into water, rollover, capsizing), and take the 90th percentile to ensure the stability of the hazard assessment. For example, a high confidence threshold of 0.7 is set to avoid misjudging dangerous states due to signal fluctuations. The method for setting the medium confidence threshold is as follows: Based on the distribution characteristics of the low and high confidence thresholds, take the midpoint between the two, while also referring to the critical confidence data when downgrading dangerous scenarios, balancing the sensitivity and stability of the downgrading process. An example is set to 0.5 as the core criterion for transitioning from a water hazard state to a suspected state. The first duration threshold is set by statistically analyzing the typical duration of instantaneous interference (such as splashing water or sensor malfunction), and taking 1.5 times its maximum value to avoid instantaneous signals triggering dangerous states. An example is set to 1 second to ensure that the high-confidence state of the instantaneous leading scenario stabilizes before entering the water-hazardous state. The second duration threshold is set by collecting data on the duration for which confidence remains low after the dangerous scenario has subsided, and taking its 95th percentile to reinforce the "easy to enter, difficult to exit" hysteresis characteristic. An example is set to 3 seconds to filter out frequent state switching caused by brief signal recovery, ensuring decision-making stability.

[0121] Nested scenario sub-state diagrams under water hazard conditions allow the system to further refine scenario types after confirming danger, providing a basis for differentiated self-rescue. These scenario sub-state diagrams include a water entry sub-state diagram, a side-captured shallow water sub-state diagram, and an inverted water sub-state diagram. The entry condition for the water entry sub-state diagram is R. in The highest CLR is above the high consistency threshold, ensuring entry only occurs when the entire vehicle is stably submerged; the entry condition for the rollover shallow water sub-state diagram is R.side The highest value and stable signal on one side, with a missing signal on the other side, accurately correspond to a single-sided immersion scenario; the entry condition for the sub-state diagram in the inverted water is R. invert The highest setting triggers the hammer switch and the water immersion sensor, but the float does not, precisely matching the inverted scenario. Each scenario sub-state diagram includes entry and exit conditions, ensuring that the transition of sub-states is consistent with actual scenario changes. Exit conditions are based on signal recovery or the emergence of a new scenario, such as R. side More than R in Then exit from the water entry state.

[0122] The hierarchical state transition model transforms complex signal combinations into clear state transition logic, avoiding the combinatorial explosion problem of conditional judgments. The main state machine is responsible for macroscopic control of the danger level, while the sub-state diagrams are responsible for microscopic differentiation of scene types. Their collaboration enables the system to respond quickly to dangers and accurately identify scenes. For example, when the system transitions from a normal state to a suspected state, the buffer provided by the main state machine allows sufficient time to accumulate evidence; upon entering a water hazard state, the refinement of the sub-state diagrams allows for the execution of specific actions for different scenarios. Without step S30, the system cannot transform multi-dimensional features into actionable decision-making criteria, rendering the features extracted in step S20 useless, and the self-rescue actions in step S40 ineffective due to a lack of clear scene judgment. By using confidence quantification and hierarchical state machine design, step S30 effectively overcomes the "feature confusion trap," enabling the system to distinguish fundamentally different wading scenarios from similar signal features. This lays the foundation for subsequent accurate self-rescue. It not only achieves accurate scenario identification but also naturally avoids misjudgments caused by sudden changes in a single signal through the logical constraints of state transitions, significantly improving the system's robustness in complex dynamic environments.

[0123] Step S40: Based on the hierarchical state transition model, implement scenario state transition, differentiated self-rescue action execution, and hysteresis control.

[0124] The core of step S40 is to transform the accurate state recognition results output in step S30 into a safe, efficient, and robust execution strategy. Through dynamic decision-making mechanisms, scenario-based action mapping, and hysteresis constraints, it completely solves the problems of decision paralysis and catastrophic misjudgment in dynamic water-related events, forming a complete technical closed loop of "feature extraction - state recognition - action execution".

[0125] Further, step S40 includes:

[0126] Step S41: For the suspected state in the hierarchical state transition model, perform scenario-adaptive evidence fusion and conflict arbitration to drive the hierarchical state transition model to the water danger state.

[0127] Further, step S41 includes:

[0128] Step S411: Dynamically set the scene-adaptive fusion window based on the signal change rate trend characteristics;

[0129] Step S412: Integrate the confidence values ​​of the four immersion scenarios over time within the fusion window to obtain the cumulative evidence for each immersion scenario.

[0130] Step S413: By comparing the cumulative amount of evidence in each immersion scenario, conflict arbitration is performed to determine the dominant scenario. The dominant scenario is then used as a mandatory input to drive the hierarchical state transition model to the water hazard state.

[0131] The suspected state is a transitional state where the system detects water-related signals but has not yet formed a clear danger judgment. Existing technologies often stall in this stage due to signal ambiguity and asynchronous modes, missing the golden response time. This step breaks this deadlock by dynamically adjusting the decision time scale, accumulating temporal evidence, and implementing a mandatory arbitration mechanism. First, the signal change rate trend feature output in step S23 is extracted. This feature directly reflects the dynamic characteristics of water level changes, and based on this, a scenario-adaptive fusion time window Tw is dynamically set. In the signal change rate trend feature, when the absolute value of the signal change rate of the water immersion sensor is greater than the preset high dynamic threshold, it indicates that the water level is changing rapidly, corresponding to emergency scenarios such as rapid immersion. In this case, the fusion time window Tw is set to a shorter duration, such as 0.5-1 second, to achieve agile response by shortening the decision cycle. When all signal change rates are lower than the low dynamic threshold, it indicates that the water level is changing slowly, corresponding to scenarios such as slow immersion. In this case, the fusion time window Tw is extended, such as 2-3 seconds, to accumulate more effective evidence and avoid misjudgment due to insufficient signal. The high dynamic threshold is determined based on statistical data of signal change rates from numerous rapid wading scenarios, taking the minimum signal change rate in this type of scenario to ensure accurate identification of the characteristics of rapid water level changes. The low dynamic threshold, on the other hand, is based on the signal change rate distribution of slow immersion scenarios, taking the maximum signal change rate in this type of scenario to clearly distinguish between the two types of dynamic scenarios. The dynamic adjustment of the fusion time window allows the system's decision-making to adapt to the time scale of different wading events, avoiding the shortcomings of fixed time windows in responding late in rapid events and lacking anti-interference ability in slowly changing events. This ensures that the decision time scale is precisely matched with the dynamic characteristics of the event, guaranteeing rapid response in emergency scenarios and improving the accuracy of judgment in slowly changing scenarios, thus overcoming the decision ambiguity caused by modal asynchrony from the time dimension.

[0132] Within a defined fusion time window Tw, the confidence values ​​for the four immersion scenarios, i.e., the confidence value R for the normal immersion scenario, are determined. in Confidence R of the side-capsulation scenario in shallow water side Confidence R of inverted underwater scene invert Confidence R of water splashing interference scenario splash By performing time integration, the cumulative evidence quantity E for each immersion scenario is obtained. in Eside E invert E splash The calculation logic of cumulative evidence quantity involves continuously summing the confidence values ​​within the fusion time window. Essentially, it transforms the static confidence of a single time slice into a dynamic sum of evidence in a time-series dimension, avoiding interference from signal fluctuations or noise at a single point in time. Existing technologies rely on single-moment signal judgments, making them susceptible to being deceived by transient interference. Cumulative evidence quantity, however, reflects the continuous characteristics of confidence. In real dangerous scenarios, confidence remains high within the time window, resulting in a naturally larger cumulative evidence quantity. Conversely, confidence from transient interference decays rapidly, leading to a smaller cumulative evidence quantity. The calculation of cumulative evidence quantity enables the system to penetrate transient signal interference, capture the essential characteristics of the scene, strengthen the evidence weight of real-world scenarios, weaken the impact of interfering signals, improve the anti-interference capability of decisions, and avoid misjudgments caused by modal asynchrony or signal jitter.

[0133] The system then performs conflict arbitration, comparing the values ​​of the four accumulated evidence quantities to determine the scenario with the largest accumulated evidence quantity as the dominant scenario within the current fusion window. The determination logic for the dominant scenario is based on the core principle that "evidence from truly dangerous scenarios has persistence and consistency." Its function is to force a unique and clear decision direction in the ambiguity of multiple scenario features, breaking the deadlock of decision paralysis and ensuring that the system can form clear state transition instructions in the suspected state stage. The arbitration result, as a mandatory input, drives the hierarchical state transition model to transition from the suspected state to the corresponding scenario sub-state diagram under the dangerous state, for example, the accumulated evidence quantity E. side At its maximum, the system directly transitions to the sub-state diagram of the capsized shallow water state. Steps S41 and S23, along with the signal change rate trend characteristics and the confidence scoring model in step S31, form a deep synergy. The trend characteristics in step S23 provide a basis for the dynamic adjustment of the fusion window, enabling evidence fusion to adapt to the dynamic characteristics of the event. The confidence value in step S31 provides the basic data for calculating the accumulated evidence amount, ensuring that evidence accumulation accurately reflects the scenario's probability. With these three elements working together, the decision-making process not only possesses time-dimensional adaptability but also evidence-dimensional reliability. Compared to a single static decision-making mechanism, its beneficial effect is the realization of a logical closed loop of "dynamic adaptation - evidence strengthening - mandatory arbitration," avoiding both decision lag and blind transitions. Without step S41, the system will remain in a stagnant "wait-observation" state in the suspected state. The hierarchical state transition model constructed in step S30 will be ineffective, and the multi-dimensional features extracted in step S20 will lose their value due to the lack of an execution path, leading to response delays in rapid wading scenarios or misjudgment risks in gradually changing scenarios.

[0134] Step S42: Based on the water hazard state and its scenario sub-state diagram determined in the hierarchical state transition model, generate a hierarchical decision instruction sequence and execute differentiated self-rescue actions according to the hierarchical decision instruction sequence.

[0135] Further, step S42 includes:

[0136] Step S421: Generate a unique final state confirmation code based on the determined water hazard state and its scene sub-state diagram in the hierarchical state transition model.

[0137] Step S422: Query the preset scene-action mapping table through the final status confirmation code to generate a hierarchical decision instruction sequence containing first-level instructions, second-level instructions and third-level instructions.

[0138] The "water hazard state" is the core state in which the system confirms a vehicle is in a dangerous wading situation. The scene sub-state diagram precisely identifies the specific type of hazard. This step avoids secondary dangers caused by the "one-size-fits-all" self-rescue mode of existing technologies by accurately mapping scenes and actions and planning the timing of hierarchical instructions. First, based on the scene sub-state diagram locked under the water hazard state, a unique final state confirmation code is generated. This final state confirmation code consists of a scene identifier and a state credibility score. The scene identifier directly corresponds to the sub-state diagrams of entering water, overturning in shallow water, and capsized in water. The state credibility score is associated with the scene's confidence value and signal quality score, ensuring that the confirmation code comprehensively reflects the scene type and signal reliability. The generation logic of the final state confirmation code is to transform abstract state information into a standardized execution index, establishing a precise correlation between state recognition and action execution, avoiding mismatches between scenes and actions, and providing clear instruction basis for differentiated self-rescue.

[0139] The system queries a pre-defined scenario-action mapping table using the final status confirmation code. This table is a standardized execution library built upon numerous water-related accident cases, ergonomic escape data, and vehicle structural characteristics. It includes the correspondence between scenario identifiers, first-level commands, second-level commands, and third-level commands. First-level commands have the highest priority and must be executed immediately. Their design is based on the common safety requirements of all dangerous water-related scenarios: triggering an audible and visual alarm to warn occupants and surrounding personnel; sending a distress message containing the vehicle's GPS location, final status confirmation code, and timestamp to pre-defined emergency contacts and rescue centers via the vehicle's communication module to ensure accurate external rescue location; and unconditionally unlocking all central locking systems to eliminate basic obstacles to escape. Second-level commands are scenario-dependent and conditionally executed. Their core design is to adapt to the unique risk points of different scenarios and integrate the precise control of a progressively activated buoyancy airbag system. This invention innovatively introduces a progressively activated buoyancy airbag system made of high-strength, thin-film rubber material. When folded, it is thin and compact, allowing it to be pre-embedded in specific locations within the vehicle structure without affecting normal vehicle use. (See reference...) Figure 4The step-by-step buoyancy airbag system includes a step-by-step buoyancy airbag 10 under the vehicle body and a step-by-step buoyancy airbag 11 on the roof. The initial state is folded, and its layout design fully considers the differences in buoyancy requirements under different wading postures. The underbody stepped buoyancy airbags 10 are pre-embedded in recessed spaces in the chassis structure, such as the inner side of the longitudinal beams, the subframe cavity, and the area around the spare tire compartment—locations with high structural strength and less susceptible to impact. Their shape is customized according to the installation space, either as a flat, elongated strip or a segmented modular structure. The number, volume, and specific installation location need to be optimized and calculated based on the weight distribution, center of gravity, and wading depth requirements of different vehicle bodies to ensure that after inflation, they can form evenly distributed buoyancy support points under the vehicle body, effectively lifting the vehicle body and preventing it from sinking. The roof stepped buoyancy airbags 11 are pre-embedded in the annular cavity inside the roof rack 12. They adopt a U-shaped or circular continuous airbag design, which not only ensures the structural stability after inflation but also provides sufficient upward buoyancy when the vehicle is upside down.

[0140] See Figure 6 The core actuator for the step-by-step activation of the buoyancy airbag comprises four key components: multiple airbag igniters 13, a pressure stabilizing tank 14, a one-way valve 15, and a folding airbag 16. The multiple airbag igniters 13 are the gas generation source, each containing a gas-generating agent such as sodium azide or guanidine nitrate. The amount of propellant in each igniter is precisely calculated to ensure that the generated gas volume increases the airbag volume by a preset proportion without causing overpressure. The igniters employ a multiple small-dose design rather than a single large-dose design. This is because, on the one hand, it allows for a gradual increase in pressure within the airbag, avoiding tearing damage to the airbag material from instantaneous high pressure; on the other hand, by controlling the detonation interval of each igniter, the inflation rate of the airbag can be precisely adjusted to adapt to changes in buoyancy requirements under different water depths and current conditions. The pressure stabilizing tank 14 is a gas buffer device. When the high-temperature, high-pressure gas generated by the igniters enters the pressure stabilizing tank, it is... The heat capacity absorption and pressure fluctuation smoothing effect of the tank body transforms the pulsed gas release into a relatively stable airflow output, preventing gas from directly impacting the airbag and causing local stress concentration. At the same time, the metal filter in the pressure stabilizing tank can also trap solid particles generated by the igniter combustion, protecting the inner surface of the airbag from being burned by high-temperature particles. The one-way valve 15 is installed between the outlet of the pressure stabilizing tank and the inlet of the airbag. Its opening pressure threshold is set slightly higher than the normal working pressure of the airbag, ensuring that gas can only flow from the pressure stabilizing tank to the airbag and will not flow back. This can prevent the gas already filled into the airbag from flowing back and being lost during the intermittent period, ensuring that the pressure in the airbag increases in a stepwise manner after each ignition. When not inflated, the folded airbag 16 is tightly stored in the installation cavity according to a specific folding process. Its folding method adopts a Z-shaped or roll folding to ensure that it can be evenly unfolded during inflation without local entanglement or twisting.

[0141] When the normal water entry state diagram is detected, the secondary command first initiates the step-by-step inflation procedure of the buoyancy airbags under the vehicle body. The controller detonates each airbag igniter sequentially at preset intervals (e.g., 1.5 seconds). When the vehicle body is lifted by buoyancy in the water, it will cause disturbance and waves in the surrounding water. An excessively rapid lifting rate will generate large hydrodynamic resistance and unstable torque, which may lead to loss of vehicle attitude control. By controlling the time interval between the detonation of each igniter, the vehicle body's rising speed is controlled within a safe range, which can quickly leave the dangerous water level without causing violent shaking. In the step-by-step activation mode, even if some igniters fail, the remaining igniters can still provide some buoyancy, which has higher reliability than detonating all igniters at once. In a normal water entry scenario, as the buoyancy airbags under the vehicle body are gradually inflated, the buoyancy at the bottom of the vehicle gradually increases. When the buoyancy balances the weight of the vehicle in the water, the vehicle stops sinking and floats stably. At this time, the secondary command will automatically lower all windows after a preset delay (such as 1.5 seconds). This delay is coordinated with the timing of the airbag inflation to ensure that the escape route is opened only after the vehicle body has been stably raised. This avoids the situation where opening the windows before the water level reaches a dangerous height will cause water to enter more quickly, and also creates a sufficient and relatively safe escape window for the occupants.

[0142] When the vehicle is identified as being submerged in water, the undercarriage buoyancy airbags will no longer function, as their buoyancy direction is opposite to the requirements after the vehicle is submerged. Instead, the roof-mounted buoyancy airbags will be activated in a step-by-step inflation process, detonating each airbag igniter at preset intervals (e.g., 1.5 seconds). In its normal position, the roof-mounted buoyancy airbags are at the highest point of the vehicle; after being submerged, they become the lowest point and are the first to contact the water surface. The buoyancy generated by their inflation can effectively support the inverted vehicle body, preventing the vehicle from sinking completely and buying time for a relatively stable air cavity to form inside the vehicle. In this scenario, the secondary command strictly prohibits the automatic lowering of any windows to maintain air supply. At the same time, the interior lighting is turned on and a window break warning is broadcast repeatedly to guide the occupants to escape scientifically with the buoyancy support provided by the airbags. In the case of a rollover in shallow water, one side of the vehicle is in contact with the water while the other side is in a safe state. No buoyancy airbags are activated. The secondary command is to automatically lower the window on the side not in contact with the water, and at the same time, the escape direction is clearly indicated through the voice prompts inside the vehicle to prevent passengers from accidentally escaping through the window on the water-submerged side, which could lead to danger.

[0143] The introduction of a tiered buoyancy airbag system fundamentally solves the problem in existing automotive wading self-rescue technologies of how to prevent rapid vehicle sinking while ensuring occupant escape time. Traditional passive safety measures, such as strengthening the vehicle body's sealing, can only slow down the rate of water ingress but cannot change the eventual sinking of the vehicle. Active drainage systems are limited by power and drainage efficiency, and their effectiveness is limited in deep water or rapid wading scenarios. The buoyancy airbag system of this invention, by actively providing upward buoyancy, upgrades "delaying sinking" to "preventing sinking," achieving a fundamental breakthrough in the concept of wading safety protection. Compared with traditional one-time inflation, the tiered activation method can dynamically adjust the buoyancy according to the actual water depth and vehicle weight, avoiding vehicle instability caused by over-inflation; the gradual inflation reduces the impact stress on the airbag material, significantly improving system reliability and service life; it provides a buffer time for occupants to adapt to changes in vehicle posture, reducing the negative impact of panic on escape efficiency; even if some ignition units fail, the remaining ignition units can still provide basic buoyancy support, possessing good degraded operation capability.

[0144] Level 3 commands are information-assisted commands that are continuously executed. A graphical escape guide is displayed on the top-level UI of the vehicle's central control screen. This guide matches the current scenario and updates the airbag inflation status information in real time, including key information such as escape routes, window break locations, breathing techniques, and the remaining buoyancy time of the airbag. In particular, after the airbag is deployed, the system will dynamically display the current buoyancy status, estimated floating time, and suggested escape window period to help occupants quickly master the escape method and seize the best escape opportunity.

[0145] The scenario sub-state diagram in step S32 provides accurate scenario input for the final state confirmation code, ensuring the accuracy of action mapping, including buoyancy airbag control. The signal quality score in step S21 indirectly influences the redundant design of action execution through the state confirmation code. For example, when the signal quality score is low, the secondary command adds a secondary confirmation mechanism for airbag activation, avoiding false airbag triggering due to sensor misjudgment. The signal change rate trend feature in step S23 provides a basis for the dynamic adjustment of the airbag inflation rate. When a rapid rise in water level is detected, the system appropriately shortens the ignition interval to 1 second, accelerating the inflation speed to cope with emergency situations. This multi-dimensional collaboration not only makes action execution scenario-specific and fault-resistant, but also provides reliable buoyancy support for vehicles navigating deep water through the innovative mechanism of tiered buoyancy airbag activation. Compared to the fixed action modes of existing technologies, it completely avoids dangerous actions caused by catastrophic misjudgments. Furthermore, the organic combination of tiered commands, buoyancy support, and information assistance significantly improves the escape success rate, transforming the accuracy of state recognition into the safety and feasibility of actual escape.

[0146] Step S43: After executing the hierarchical decision instruction sequence, enter the water hazard state-maintenance mode, set triple degradation conditions, and when the triple degradation conditions are met simultaneously, drive the hierarchical state transition model to degrade from the water hazard state to the suspected state or normal state, thereby achieving hysteresis control of state maintenance and degradation.

[0147] The core design of the water hazard-maintain mode is to avoid the high-frequency oscillation of "water hazard-suspected state" caused by factors such as water flow fluctuations, slight changes in vehicle posture, or brief recovery of sensor signals. This ensures that once a danger is confirmed, the alarm will not be easily deactivated without clear safety evidence. Existing technologies often cause alarms to repeatedly start and stop due to signal fluctuations, interfering with occupants' escape decisions. The three degrading conditions set in this step are independent and indispensable. The first condition is a sharp drop in confidence: the confidence value of all dangerous scenarios, such as normal water entry, rollover, and overturning, is lower than the degrading threshold for a continuous preset time (e.g., 5 seconds). The degrading threshold is determined based on the confidence distribution data of a large number of normal driving scenarios, taking the lowest quantile of the distribution to ensure that the condition is met only when the dangerous features have completely disappeared. The second condition is a clear confirmation of dehydration trend: the trend feature of the signal change rate output in step S23 must be clearly marked as "water level drop or drainage state", and the duration of this state exceeds the preset time (e.g., 3 seconds). This condition verifies whether the danger has been truly alleviated by the direction of water level change, avoiding misjudgment of safety due to instantaneous water level fluctuations. The third condition is large-scale recovery of sensor signals: more than a preset proportion (e.g., 80%) of the triggered sensors in the left and right signal feature vectors output in step S21 recover to the non-triggered state. This proportion is determined based on the recovery law of sensors after the vehicle is completely out of the water, ensuring that the condition is met only when most of the vehicle body is out of the water.

[0148] The logic behind the triple downgrade condition is "simultaneous fulfillment." Through cross-verification of multi-dimensional safety evidence, it strengthens the robustness of the decision-making process, avoids misjudgments caused by single-dimensional signal deception, ensures the stability of the water hazard state, and provides continuous and reliable safety assurance for occupant escape. For example, if some sensor signals briefly recover due to water flow impact, satisfying only a single downgrade condition, the system remains in the water hazard state. Only when the confidence level of the dangerous scenario remains low, the water level clearly drops, and most sensors return to their non-triggered state is the danger deemed resolved, and the downgrade process initiated. Without step S43, the system may frequently switch between the water hazard state and the suspected state due to instantaneous signal fluctuations, causing repeated start-stop self-rescue actions. This would disrupt the occupant escape rhythm and could also lead to secondary dangers due to repeated window raising and lowering accelerating water ingress. Step S40 breaks through the decision-making stagnation in the suspected state through scene adaptive evidence fusion and conflict arbitration, gaining golden response time for rapid water wading scenarios; through scene-action precise mapping, it avoids catastrophic misjudgments such as opening windows when the vehicle is upside down or opening flooded side windows when it rolls over, ensuring the safety of occupants during escape; through hysteresis control of triple degradation conditions, it ensures the stability of the water hazard state and avoids frequent alarm switching that interferes with escape; through graded instructions and information assistance, it improves the efficiency of occupant escape and reduces the risk of secondary injury during the escape process.

[0149] Example 2

[0150] This embodiment, based on Embodiment 1, provides a vehicle flooding monitoring system, such as... Figure 7 As shown, it includes:

[0151] Signal preprocessing module: used to set up a dual-sensor combined detection device and a counterweight switch device, establish a multi-source sensor timing acquisition framework based on the dual-sensor combined detection device and the counterweight switch device, and acquire the original sensor signal sequence with timestamps; perform targeted filtering processing on the original sensor signal sequence with timestamps to obtain the filtered sensor signal sequence; the dual-sensor combined detection device includes both a float-type water sensor and a water immersion sensor;

[0152] Feature extraction module: used to perform spatiotemporal and trend analysis on the filtered sensor signal sequence to obtain the left signal feature vector, the right signal feature vector, the cross-side consistency index, and the signal change rate trend features;

[0153] State transition model construction module: Based on the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics, a hierarchical state transition model is constructed;

[0154] Execution module: Based on a hierarchical state transition model, it implements scenario state transitions, differentiated self-rescue actions, and hysteresis control;

[0155] The progressively activated buoyancy airbag system includes a progressively activated buoyancy airbag 10 under the vehicle body and a progressively activated buoyancy airbag 11 on the roof. Both the progressively activated buoyancy airbag 10 under the vehicle body and the progressively activated buoyancy airbag 11 on the roof include multiple airbag igniters, pressure stabilizing tanks, one-way valves, and folding airbags.

[0156] It should be noted that the vehicle immersion monitoring system provided by this invention has the flexibility of modular configuration, and its functions can be tailored or expanded according to actual application needs: When the full configuration is adopted, it includes all the above-mentioned functional modules, realizing a complete closed loop from signal acquisition, feature extraction, state recognition to self-rescue action execution; when only monitoring function is required, only the monitoring module including the signal preprocessing module, feature extraction module, and state transition model construction module can be configured to output accurate immersion state recognition results without executing self-rescue actions, which is suitable for application scenarios that require manual decision-making or have other existing safety systems; when the monitoring configuration is used in conjunction with third-party actuators, the hierarchical state transition model results output by the state transition model construction module are sent to other vehicle actuators, such as third-party emergency drainage systems, external alarm devices, and independent escape assistance systems, through standardized interfaces such as CAN bus and vehicle Ethernet, to achieve collaborative work of heterogeneous systems. This modular design enables this invention to flexibly adapt to application scenarios with different vehicle models, different safety level requirements, and different cost budgets, serving as both an independent complete solution and a monitoring front-end or decision-making center for vehicle safety systems.

[0157] Furthermore, in the signal preprocessing module, the method for performing targeted filtering on the timestamped raw sensor signal sequence includes:

[0158] Step S121: Separate the switching signal of the float-type water sensor, the conductivity analog signal of the water immersion sensor, and the mechanical trigger signal of the counterweight switch from the original sensor signal sequence with timestamps.

[0159] Step S122: Apply a debouncing filtering algorithm to the switch signal to obtain the trigger time and duration of each float-type water sensor.

[0160] Step S123: Apply a medium-range filtering algorithm to process the conductivity analog signal to generate a filtered conductivity value. When the filtered conductivity value exceeds a preset conductivity threshold, record the trigger time and duration of the water immersion sensor.

[0161] Step S124: Perform state holding filtering on the mechanical trigger signal to confirm the triggering time and duration of the counterweight switch;

[0162] Step S125: Integrate the trigger time and duration of each sensor to form a complete filtered sensor signal sequence.

[0163] Furthermore, the methods for performing spatiotemporal and trend analysis on the filtered sensor signal sequence in the feature extraction module include:

[0164] Step S21: Based on the filtered sensor signal sequence and the installation position of each sensor, construct the left signal feature vector and the right signal feature vector;

[0165] Step S22: Calculate the cross-side consistency index using the left-side signal feature vector and the right-side signal feature vector;

[0166] Step S23: Perform trend analysis on the filtered sensor signal sequence within the time window to obtain the trend characteristics of the signal change rate.

[0167] Further, step S22 includes:

[0168] Step S221: Extract the trigger times of similar sensors from the left signal feature vector and the right signal feature vector respectively, and calculate the absolute value ΔTF of the difference between the average trigger times of the left float-type water sensor and the average trigger times of the right float-type water sensor, as well as the absolute value ΔTW of the difference between the average trigger times of the left water immersion sensor and the average trigger times of the right water immersion sensor.

[0169] Step S222: Convert ΔTF and ΔTW into time consistency coefficients using a Gaussian function;

[0170] Step S223: Compare the trigger combination modes of the sensors in the left signal feature vector and the right signal feature vector to obtain the consistency of the trigger modes;

[0171] Step S224: Multiply the time consistency coefficient by the trigger mode consistency to obtain the cross-side consistency index.

[0172] The methods and systems of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0173] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0174] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring vehicle flooding, characterized in that, The method includes: A dual-sensor combined detection device and a counterweight switch device are set up, and a multi-source sensor timing acquisition framework based on the dual-sensor combined detection device and the counterweight switch device is established to obtain the original sensor signal sequence with timestamps; targeted filtering processing is performed on the original sensor signal sequence with timestamps to obtain the filtered sensor signal sequence; the dual-sensor combined detection device includes both a float-type water sensor and a water immersion sensor. Spatiotemporal and trend analysis was performed on the filtered sensor signal sequence to obtain the left signal feature vector, the right signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics. A hierarchical state transition model is constructed based on the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics. Based on the hierarchical state transition model, scenario state transitions, differentiated self-rescue actions, and hysteresis control are implemented.

2. The method for monitoring vehicle flooding according to claim 1, characterized in that, The method for performing targeted filtering on the raw sensor signal sequence with timestamps includes: The switching signal of the float-type water sensor, the conductivity analog signal of the water immersion sensor, and the mechanical trigger signal of the counterweight switch were separated from the original sensor signal sequence with timestamps. The switching signals are processed using a debouncing filtering algorithm to obtain the trigger time and duration of each float-type water sensor.

3. The method for monitoring vehicle flooding according to claim 2, characterized in that, The method for performing targeted filtering on the raw sensor signal sequence with timestamps further includes: The conductivity analog signal is processed by a medium filtering algorithm to generate a filtered conductivity value. When the filtered conductivity value exceeds a preset conductivity threshold, the trigger time and duration of the water immersion sensor are recorded. The mechanical trigger signal is filtered to maintain its execution status, and the triggering time and duration of the counterweight switch are confirmed.

4. The method for monitoring vehicle flooding according to claim 3, characterized in that, The method for performing spatiotemporal and trend analysis on the filtered sensor signal sequence includes: Based on the filtered sensor signal sequence and the installation location of each sensor, construct the left-side signal feature vector and the right-side signal feature vector; The cross-side consistency index is calculated using the feature vectors of the left and right signals. Trend analysis was performed on the filtered sensor signal sequence within a time window to obtain the trend characteristics of the signal change rate.

5. The method for monitoring vehicle flooding according to claim 4, characterized in that, The method for constructing the left-side signal feature vector and the right-side signal feature vector includes: The filtered sensor signal sequence is spatially grouped according to the physical installation position of the sensors on the vehicle, into a left sensor group and a right sensor group. Extract the trigger time and duration of each sensor in the left sensor group to form the left signal feature vector; Extract the trigger time and duration of each sensor in the right-hand sensor group to form the right-hand signal feature vector.

6. The method for monitoring vehicle flooding according to claim 5, characterized in that, The method for calculating the cross-side consistency index includes: Extract the trigger times of similar sensors from the left and right signal feature vectors respectively, and calculate the absolute value ΔTF of the difference between the average trigger times of the left float-type water sensor and the average trigger times of the right float-type water sensor, as well as the absolute value ΔTW of the difference between the average trigger times of the left water immersion sensor and the average trigger times of the right water immersion sensor. Based on ΔTF and ΔTW, the cross-side consistency index is obtained.

7. A method for monitoring vehicle flooding according to claim 6, characterized in that, The method for obtaining the cross-side consistency index based on ΔTF and ΔTW includes: ΔTF and ΔTW are converted into time consistency coefficients using a Gaussian function; By comparing the sensor triggering combination modes in the left and right signal feature vectors, the consistency of the triggering modes is obtained; Multiplying the time consistency coefficient by the trigger mode consistency coefficient yields the cross-side consistency index.

8. A method for monitoring vehicle flooding according to claim 7, characterized in that, The method for constructing the hierarchical state transition model includes: By combining the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics, a confidence scoring model for four immersion scenarios is constructed, and the confidence values ​​for the four immersion scenarios are calculated. The four immersion scenarios include normal water entry scenario, side-overturned shallow water scenario, upside-down water scenario, and splashing interference scenario. Based on the confidence values ​​of four immersion scenarios, a three-state master state machine is established, which includes three main states: normal state, suspected state, and water danger state. Under the water danger state, scenario sub-state diagrams are nested to form a hierarchical state transition model. The scenario sub-state diagrams include the water entry sub-state diagram, the side capsizing shallow water sub-state diagram, and the inverted water sub-state diagram.

9. A vehicle flooding monitoring system, used to implement the vehicle flooding monitoring method according to any one of claims 1-8, characterized in that, The system includes: Signal preprocessing module: used to set up a dual-sensor combined detection device and a counterweight switch device, establish a multi-source sensor timing acquisition framework based on the dual-sensor combined detection device and the counterweight switch device, and acquire the original sensor signal sequence with timestamps; perform targeted filtering processing on the original sensor signal sequence with timestamps to obtain the filtered sensor signal sequence; the dual-sensor combined detection device includes both a float-type water sensor and a water immersion sensor; Feature extraction module: used to perform spatiotemporal and trend analysis on the filtered sensor signal sequence to obtain the left signal feature vector, the right signal feature vector, the cross-side consistency index, and the signal change rate trend features; State transition model construction module: Based on the left-side signal feature vector, the right-side signal feature vector, the cross-side consistency index, and the signal change rate trend characteristics, a hierarchical state transition model is constructed; Execution module: Based on a hierarchical state transition model, it implements scenario state transitions, differentiated self-rescue actions, and hysteresis control; The step-by-step buoyancy airbag system includes a step-by-step buoyancy airbag (10) under the vehicle body and a step-by-step buoyancy airbag (11) on the roof. Both the step-by-step buoyancy airbag (10) under the vehicle body and the step-by-step buoyancy airbag (11) on the roof include multiple airbag igniters (13), pressure tanks (14), one-way valves (15), and folding airbags (16).

10. A vehicle flooding monitoring system according to claim 9, characterized in that, When the car enters the submerged state of water danger, the buoyancy airbags under the car body are activated step by step (10); when the car enters the inverted submerged state of water danger, the buoyancy airbags on the roof are activated step by step (11). When the vehicle body's lower part buoyancy airbag (10) or the vehicle roof's buoyancy airbag (11) is activated, the airbag igniters are detonated one by one according to a preset time interval.

Images