Multi-modal data fusion analysis and judgment method based on environmental perception driving

Abstract

The invention relates to the technical field of multi-modal electrical variable analysis, and discloses a multi-modal data fusion analysis and judgment method based on environmental perception driving, which comprises a sensor unit for acquiring environmental parameter data, preliminarily judging data exception through environmental data, performing feature extraction on the environmental data, and obtaining a feature extraction result; the extracted environment data features are endowed with labels with timestamps and are input into a pushing unit, the pushing unit is provided with a dynamic adjusting and optimizing unit for performing multi-stage analysis and pushing on multi-environment data, outputting an environment analysis result and adjusting the pushing step length of the pushing unit in real time, and the dynamic adjusting and optimizing unit is used for receiving a multi-modal data analysis result of environment perception and outputting the multi-modal data analysis result. Whether the closed space and the power space need to be separated or not is judged, analysis and judgment of multi-environment perception are achieved, and automation and intelligence of separation of the closed space and the power space are improved.

Description

Technical Field

[0001] The invention relates to the technical field of multimodal electrical variable analysis, and discloses a multimodal data fusion analysis and judgment method based on environmental perception drive. Background Art

[0002] Data heterogeneity and insufficient fusion efficiency. The data types (images, point clouds, time series signals) and sampling rates generated by different sensors (such as cameras, LiDAR, sonar) vary significantly, resulting in difficulties in data alignment and feature extraction. The system's misjudgment rate increases in complex scenarios. For example, an autonomous vehicle may misidentify obstacles in a rainstorm. In extreme weather (rain, snow, fog) or sensor failure (occlusion, noise), the quality of some modal data drops sharply, resulting in unreliable fusion results. Summary of the invention

[0003] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0004] In order to solve the above technical problems, the main purpose of the present invention is to provide a multimodal data fusion analysis and judgment method based on environmental perception driving, wherein the multimodal data fusion analysis and judgment method based on environmental perception driving includes: The sensor unit that obtains environmental parameter data aligns different data dimensions by establishing a multi-data radiation field and makes a preliminary judgment on data anomalies in the environmental data; Extract features from environmental data, assign the extracted environmental data features to tags with timestamps, and input them into a shifting unit, wherein the shifting unit sets a multi-level analysis shift to analyze and judge multiple environmental data, and outputs environmental analysis results; The dynamic tuning unit adjusts the moving step length of the moving unit in real time, receives the environmental analysis result, and determines whether the closed space and the power space need to be separated.

[0005] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: Establish a multi-data radiation field, which includes a pressure field and a voiceprint field, and determine the time-varying correlation between the pressure and voiceprint signals through alignment coding. The pressure field is constructed based on an unstructured grid discretization method to map the water pressure data into a three-dimensional radiation field; If the pressure mapped in the pressure field changes suddenly and the high-frequency area of ​​the voiceprint mapped in the voiceprint field is abnormal, the grid density is dynamically optimized by dividing the multi-data radiation field into high-resolution grids and extracting local neighborhood features.

[0006] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: The environmental data baseline is used to compare whether the acquired torque data, water level pressure data and water sound pattern data are within the normal threshold range of the environmental data. If the torque data, water level pressure data and water sound pattern data are not within the normal threshold range of the environmental data, a judgment of data abnormality is triggered, and the judgment of data abnormality includes environmental data baseline, single mode abnormality and secondary verification; The single-mode anomaly is used to map whether there is a mutation, imbalance and decoupling in the torque data, water level pressure data and water sound pattern data. If there is a single data anomaly in any of the torque data, water level pressure data and water sound pattern data, a single environmental data anomaly data report is output, and whether the sensor unit is abnormal is detected; The secondary verification includes time series correlation verification and space propagation verification. The time series correlation verification is used to establish continuous timestamp torque data mutation, imbalance and decoupling. The space propagation verification is used to obtain water level pressure data and water sound pattern data mutation, imbalance and decoupling.

[0007] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: The pushing unit includes a torque pushing unit, a water level pressure pushing unit and a water sound pattern pushing unit; The torque shift unit takes the torque feature with a timestamp tag as input, obtains the difference between the torque feature at the previous moment and the torque feature at the current moment, and obtains the difference between the torque feature at the next moment and the torque feature at the current moment by real-time monitoring of the torque features within continuous timestamps, and triggers an abnormal mark within a continuous time.

[0008] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: The water level pressure shift unit sets a water level pressure prediction model, predicts the water level pressure data at the next moment through the water level pressure prediction model, and marks the flooding risk through mutation detection; The mutation detection includes predicting whether a sudden change occurs in the gradient change between the pressure and the actual value, processing the pressure signal by short-time Fourier transform, and identifying the sudden increase of the high-frequency signal component; The water soundprint shifting unit is provided with a soundprint classification model and a water soundprint recognition unit.

[0009] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: The voiceprint classification model is used to process the water voiceprint obtained by the water voiceprint recognition unit, obtain the frequency spectrum characteristics of the voiceprint through the voiceprint classification model, and classify the frequency spectrum characteristics, including rainstorm impact, mechanical vibration and human operation; The water soundprint recognition unit is used to obtain multi-environment water sound data and water soundprint optimization. The multi-environment water sound includes water soundprint features affected by weather and water soundprint features affected by human operation. The water soundprint optimization is used to optimize the acquired water soundprint.

[0010] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: Obtain the torque change rate, water pressure gradient change rate and water sound pattern spectrum characteristics in continuous time; The torque shifting unit is driven to receive the torque shifting rate in the continuous timestamps through the torque shifting rate in the continuous timestamps, and the torque parameter shifting step is adjusted. If the torque data shifting rate per unit time is greater than the highest weight, the torque parameter shifting step is shortened. If the torque data shifting rate per unit time does not exceed the highest weight, the state to be adjusted is entered. The step length is adjusted by the acceleration of the water pressure change at the continuous time stamp. If the acceleration of the water pressure change at the time stamp from t-1 to t is greater than the acceleration of the water pressure change at the continuous time stamp from t-2 to t-1, and the acceleration increase is greater than the minimum threshold, the step length ratio is compressed. The water acoustic signal is obtained through the water soundprint recognition unit, and the step size adjustment is activated through comparison with the acoustic feature library. If the acoustic signal in the frequency band captured by the water soundprint spectrum is bubble burst and vortex, it is judged as water immersion, and the soundprint shift step size of the water soundprint shift unit is adjusted.

[0011] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: By setting step boundary limits for receiving the adjusted soundprint shift step, water pressure compression step ratio and adjusted torque shift step, and setting intensity levels, the intensity levels include low disturbance and medium-high disturbance; By adjusting the step size boundary, the redundant boundary buffer layer is triggered to make temporary tolerance for the step size adjustment. If the step size temporarily exceeds the rated boundary, the total number of steps after actual adjustment is no greater than the maximum number of steps in the redundant boundary buffer layer. If the buffer layer stays for more than the threshold, it immediately switches to the minimum safe step size.

[0012] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: If the torque change rate, water pressure gradient change rate and water sound pattern spectrum characteristics of the pushing unit are in the stable domain of the redundant boundary buffer layer after adjustment, the step size of the pushing unit continues to be adjusted according to the real-time data; If the total compensation quantity of the step length adjustment of the push unit is not greater than the maximum step length quantity of the redundant boundary buffer layer, and the buffer layer residence time is greater than a threshold, the step length freezing is triggered.

[0013] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: If it is determined that the enclosed space needs to be separated, a sealing strip is provided on the hatch of the enclosed space, and the hatch is in a closed state when it is closed. A position sealing ring is provided at the place where the enclosed space is connected to the outside world, and all sensors are provided for wire control to convert physical control into digital signals, and transmit them to the main controller, which controls the separation of the enclosed space mechanism connector and the electrical connector; The closed space receives the separation signal output by the multimodal data analysis, which triggers the separation of the closed space and the power space, triggers the closing of the air-conditioning ventilation duct, triggers the inflation of the inflatable airbag outside the closed space, and triggers the separation balancer, which moves the weight upward in the inclined longitudinal and lateral directions to maintain a nearly horizontal state, ultimately completing the separation of the closed space and the power space.

[0014] As a preferred solution of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein: If the enclosed space does not need to be separated, after the enclosed space without a separation mechanism receives the corresponding signal output by the multimodal data analysis, it triggers the closing of the air-conditioning ventilation duct, triggers the inflation of the inflatable airbag outside the enclosed space, and triggers the balancer. The balancer moves the weight upward in the inclined longitudinal and lateral directions to maintain a nearly horizontal state, triggers the electrical isolation of the enclosed space and the power space, and finally completes the overall floating.

[0015] Beneficial effects of the present invention: The present application improves the confidence of water immersion judgment through dual verification of time association (continuous mutation of torque) and space propagation (time difference matching of pressure wave and soundprint). The redundant boundary buffer layer allows the step length to temporarily exceed the boundary (such as exceeding the theoretical value by 10%), thereby reducing the action lag caused by excessive constraints while ensuring safety. By setting up multiple shifting units, multiple environmental parameters can be processed efficiently and separately in parallel, thereby improving the efficiency and accuracy of multimodal fusion analysis for closed spaces and dynamic spaces. By setting up boundary redundancy and step length adjustment freezing mechanism, the maximum step length adjustment of multi-environment perception is achieved, thereby improving the multimodal response speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them: Figure 1 This is a flow chart of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention; Figure 2 The present invention is a multi-modal data fusion analysis and judgment method based on environmental perception driven by the environmental perception multi-modal analysis method; Figure 3 It is a schematic diagram of a water soundprint recognition unit in the multimodal data fusion analysis and judgment method based on environment perception drive of the present invention; Figure 4 This is a schematic diagram of a shift unit of a multimodal data fusion analysis and judgment method based on environment perception drive of the present invention; Figure 5 It is a schematic diagram of a closed space of a multimodal data fusion analysis and judgment method based on environment perception drive of the present invention; Figure 6 A connection diagram of a passenger compartment and a vehicle frame in practical application of the multimodal data fusion analysis and judgment method driven by environmental perception of the present invention; Figure 7 It is a schematic diagram of the positioning bolt at the bottom of the passenger compartment in the practical application of the multimodal data fusion analysis and judgment method based on environmental perception drive of the present invention; Explanation of the reference numerals in the accompanying drawings: 1. sunroof; 2. steering column; 3. air conditioning vent; 4. throttle; 5. brake; 6. cabin locating bolt; 8. universal joint; 9. upper part of detachable shaft; 10. inter-shaft groove; 11. protrusion; 12. connecting sleeve; 13. lower part of steering column; 14. steering mechanism; 15. passenger compartment floor; 16. bayonet pin groove; 17. bayonet pin; 18. locating seat; 19. frame plate; 101. top sound wave inlet; 202. spiral sound wave processing structure; 303. support unit. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0018] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0020] Embodiment 1 like Figure 1 As shown, the multimodal data fusion analysis and judgment method based on environmental perception driving includes: The sensor unit that obtains environmental parameter data aligns different data dimensions by establishing a multi-data radiation field, and makes a preliminary judgment on data anomalies in the environmental data.

[0021] Specifically, the sensor unit includes a strain gauge torque sensor for obtaining torque, a pressure sensor for obtaining water level pressure, and a sound print sensor for obtaining water sound print; The determination of data anomalies includes environmental data baseline, single-mode anomalies and secondary verification.

[0022] The environmental data baseline is used to compare whether the acquired torque data, water level pressure data and water sound pattern data are within the normal threshold of the environmental data; The single-mode anomaly is used to map whether there is a mutation, imbalance and decoupling in the torque data, water level pressure data and water sound pattern data. If there is a single data anomaly in any of the torque data, water level pressure data and water sound pattern data, a single environmental data anomaly data report is output, and whether the sensor unit is abnormal is detected; The secondary verification includes time series correlation verification and space propagation verification. The time series correlation verification is used to establish continuous timestamp torque data mutation, imbalance and decoupling. The space propagation verification is used to obtain water level pressure data and water sound pattern data mutation, imbalance and decoupling.

[0023] Specifically, a 4-channel MEMS microphone of the voiceprint sensor array is installed, covering the target frequency band of 500-800Hz, supporting the capture of sound wave incident angle of ±30°, integrating a pressure sensor with a range of 0-100kPa and an accuracy of ±0.1%FS), using a spiral channel to enhance the sensitivity of fluid pressure gradient, deploying a strain torque sensor (non-contact, range 0-500Nm), and isolating mechanical vibration interference through a three-axis shock absorption system.

[0024] High-frequency vibration noise was eliminated by moving average filtering (window = 10 data points), turbulent fluctuations were smoothed by Kalman filtering, and reconstructed by wavelet packet decomposition (retaining the characteristic frequency band of bubble rupture of 500-800 Hz).

[0025] The specific implementation methods for establishing an environmental data baseline include: Based on historical data statistics, the normal fluctuation range = rated value ± 5% (such as a 500Nm system, the baseline is 475-525Nm). The steady-state pressure interval is divided by clustering, the threshold = mean ± 3σ, σ is the gradient standard deviation, and the MFCC feature library (normal working condition voiceprint template) is constructed. The matching threshold = 85% (below 70% is considered abnormal). Every 24 hours, each node uploads encrypted features (such as torque statistical distribution, voiceprint MFCC mean), and the new threshold is issued after cloud aggregation. Heavy rain (200-500Hz broadband noise), mechanical shock (800Hz harmonics) and other interference are simulated to verify the robustness of the baseline.

[0026] Single-modal anomaly detection includes: torque mutation, pressure mutation and voiceprint mutation.

[0027] The torque mutation calculation gradient Δτ / Δt>25% / s (for example, for a 500Nm system, the change is>125Nm within 5 seconds). An alarm is triggered for three consecutive over-limits, where Δτ is the torque change rate and Δt is the time interval. The torque mutation calculation gradient means the time change rate of the torque (τ, unit Nm), which reflects the relative change amplitude of the torque per unit time. The change amplitude is used to determine whether the torque has mutated: (current torque - torque at the previous moment) / system rated torque × 100% / time interval.

[0028] Pressure mutation analysis 500-800Hz high-frequency energy increase > 30% (compared to baseline), or gradient ΔP / Δt > 10kPa / s. Similarly, ΔP is the pressure change rate, Δt is the time interval, and the time change rate of pressure (P, unit kPa) directly reflects the sudden change in pressure.

[0029] Voiceprint mutation is detected when the spectrum variance is greater than 15% (window = 1s) or the main frequency deviation is greater than 10%.

[0030] Data imbalance is determined when a single data continuously exceeds the baseline threshold (such as pressure > 105 kPa for more than 10 seconds).

[0031] The decoupling judgment includes torque-pressure mutual covariance <0.6 (τ-P should be strongly correlated under normal working conditions), and soundprint-pressure propagation delay deviation >5% (theoretical delay = distance / 1480m / s), where τ is the torque data and P is the pressure data.

[0032] The second level verification verifies whether the data is disconnected through time series correlation and verifies the mutation area data output by multi-data radiation field analysis through spatial propagation verification.

[0033] Specifically, the strain-type torque sensor adopts a non-contact design, detects torque changes through strain gauges on the elastic shaft, outputs a 0-10V analog signal, ensures the concentricity of the axis is less than Φ0.05mm during deployment, and realizes real-time data transmission through wireless telemetry.

[0034] The water level pressure sensor uses a MEMS resonant pressure sensor with a range of 0-100kPa, an accuracy of ±0.1%FS, and supports 4-20mA output. It needs to be deployed in a triangular topology during installation, and the spacing should not exceed 3 times the pipe diameter to eliminate turbulent interference.

[0035] The water sound pattern sensor is configured as a capacitive microphone array with a sampling rate ≥ 1kHz. It combines morphological filtering to eliminate rainstorm noise (200-500Hz) and extract 500-800Hz bubble rupture features.

[0036] like Figure 3 As shown, it includes a top sound wave inlet 101, a spiral sound wave processing structure 202 and a support unit 303, wherein the top sound wave inlet 101 adopts a trumpet-shaped sound waveguide structure, and the opening diameter is matched with the sound wave wavelength. For example, the opening diameter is designed to be about 6-15 cm, and the interior is coated with sound-absorbing material to suppress high-frequency reflected noise. Through the gradual change of acoustic impedance design, the free-field sound waves are gradually transferred to the inside of the device. A dust-proof net (mesh number ≥ 2000) is set at the entrance to filter out interferences such as sand, dust, and water droplets while maintaining the sound transmittance (> 90%), optimize the sound wave incident angle (receiving cone angle ± 30°), improve the capture efficiency of the target sound source (such as the sound of underwater bubble bursting), and suppress environmental noise (such as wind noise, mechanical vibration) from entering the subsequent processing link.

[0037] The spiral sound wave processing structure 202 is composed of a multi-stage spiral waveguide and adopts 3D printed acoustic resin material. The spiral pitch decreases exponentially (the starting pitch is λ / 2, λ is the target sound wave wavelength), and the end of each spiral is connected to a resonance cavity (the volume matches 1 / 4 wavelength of the target frequency). The spiral structure focuses the incident sound wave energy to the central axis through the acoustic lens effect, and the sound pressure level gain can reach 12dB (compared with the unguided structure). Furthermore, different pitches correspond to different frequency bands (such as the first-stage spiral processes 80-120Hz low frequency, and the second-stage spiral processes 350-500Hz medium frequency), realizing physical pre-separation of the soundprint spectrum. The serrated texture of the inner wall of the spiral (depth ≈λ / 10) destroys the sound wave reflection path, suppresses the formation of standing waves, reduces the probability of spectrum aliasing, and guides broadband sound waves to different processing channels according to frequency bands, thereby enhancing target features (such as 500-800Hz pulse groups of bubble burst sounds) and attenuating interference frequency bands (such as 20-50Hz low-frequency vibrations of ship engines).

[0038] The support unit 303 integrates a three-axis shock absorption system with an adjustable damping coefficient. It is embedded with a MEMS capacitive microphone array (4 channels, SNR>70dB) and is equipped with a threaded interface (such as M18×1.5) at the bottom to support quick installation and posture calibration. The mass-spring-damper system (mck model) is used to reduce the resonance frequency of the device to below 5Hz to prevent external vibration from coupling to the acoustic components. The time delay difference (TDOA) of the 4-channel microphone is used to achieve sound source positioning (accuracy ±0.5°), and the adaptive beamforming algorithm (such as MVDR) is combined to improve the signal-to-noise ratio, ensuring the stable operation of the acoustic components in complex environments. At the same time, the spatial selectivity of the target sound source is enhanced through array processing.

[0039] Establishing a multi-data radiation field, the multi-data radiation field includes a pressure field and a voiceprint field, and determining a time-varying correlation between pressure and voiceprint signals through alignment coding; Specifically, the pressure field construction includes: The pressure field is constructed based on the unstructured grid discretization method. The water pressure data is mapped into a three-dimensional radiation field. Finite element analysis is used to capture the spatial distribution characteristics of the pressure gradient mutation area, such as the sudden pressure change caused by fluid impact at the immersion point (ΔP / Δt>10 kPa / s). The calculation accuracy can be improved by setting a high-resolution grid.

[0040] When the pressure gradient is detected to exceed the threshold (10 kPa / s), the high-resolution grid is refined to 0.1m accuracy, and other areas maintain a 0.5m sparse grid to ensure that computing resources are allocated preferentially to key areas.

[0041] Dynamically adjust the mesh density according to the pressure fluctuation amplitude to balance computational efficiency and accuracy while ensuring the rationality of the adjustment. For example, enhance the mesh resolution in the pressure wave propagation path (such as pipe bends).

[0042] Specifically, establishing a voiceprint field includes: The construction of the soundprint field focuses on the high-frequency components of 500-800Hz (corresponding to bubble bursting and vortex cavitation characteristics). Its physical basis is the dispersion characteristics of sound wave propagation. High-frequency soundprints decay faster in liquids, and local transient characteristics are captured through high-precision grids.

[0043] The voiceprint signal is decomposed into 5 layers of wavelet packets to extract the energy distribution of the key frequency band of 2-5kHz, and the driving grid is encrypted to an accuracy of 0.05m in the sound source area (such as the water immersion point).

[0044] Environmental interference (such as low-frequency noise from heavy rain) is suppressed by using anti-noise methods that eliminate pulse noise and fill spectrum holes (such as obtaining anti-noise coefficients and eliminating noise pollution).

[0045] Based on the time difference of orientation (TDOA) method, the time difference of sound wave arrival is calculated through a multi-sensor array, and the spatial coordinates of the sound source are inverted by combining the wave speed (1480m / s), and the spatial matching verification is carried out with the pressure field mutation area.

[0046] Specific alignment codes include: The input layer receives four-dimensional data (x, y, z, t) of the multi-data radiation field with timestamp labels, and models the propagation delay relationship between the pressure wave and the voiceprint signal through the gating mechanism. For example, it takes Δt1 for the pressure wave to propagate to the sensor, while the corresponding voiceprint signal is delayed by Δt2 due to the medium difference. LSTM learns the nonlinear mapping of Δt1-Δt2 and enforces wave speed matching through the LSTM loss function.

[0047] If the pressure mapped in the pressure field changes suddenly and the high-frequency area of ​​the voiceprint mapped in the voiceprint field is abnormal, the grid density is dynamically optimized by dividing the multi-data radiation field into high-resolution grids and extracting local neighborhood features.

[0048] By setting up a multi-data radiation field to dynamically allocate computing resources to abnormal data and output accurate processed water pressure and voiceprint data, the waste of computing resources is reduced while ensuring data accuracy.

[0049] like Figure 2 As shown, feature extraction is performed on environmental data, the extracted environmental data features are assigned tags with timestamps, and are input into a shift unit, wherein the shift unit sets a multi-level analysis shift to analyze and judge multiple environmental data, and outputs environmental analysis results; like Figure 4 As shown, a specific implementation method of a push unit: The pushing unit includes a torque pushing unit, a water level pressure pushing unit and a water sound pattern pushing unit.

[0050] The torque shifting unit starts with the torque input, and outputs the torque abnormality through the first torque detection and the second torque detection, which is recorded as the torque analysis result.

[0051] The water level pressure shift unit starts with water pressure input, and outputs water pressure anomalies through first water pressure detection and second water pressure detection, which are recorded as water pressure analysis results.

[0052] The water voiceprint shift unit starts with the voiceprint input, and outputs the voiceprint anomaly through the first voiceprint detection and the second voiceprint detection, which is recorded as the voiceprint analysis result.

[0053] Through parallel processing of the pushing unit, combined with the torque, water pressure and soundprint analysis results, it is judged whether the enclosed space and the power system need to be separated. If the three meet the separation conditions at the same time, the enclosed space and the power system will be separated. If the three do not meet the separation conditions at the same time, the enclosed space and the power system will not trigger separation and continue detection.

[0054] By receiving the high-frequency area of ​​the multi-data radiation field output, the high-resolution network is divided, the shift step is dynamically adjusted, and the dynamic adjustment redundant buffer layer limits the shift step adjustment range to prevent distortion.

[0055] The three-mode synchronization abnormality is separated immediately, the separation trigger time is short, the single / double abnormality start verification, and the false triggering situation is low, which greatly improves the separation accuracy of the enclosed space and the power system, greatly reduces the response time, and increases the linkage of the system.

[0056] Specifically, feature extraction of environmental data includes multi-dimensional feature extraction and timestamp labeling.

[0057] Furthermore, multi-dimensional feature extraction technology includes time domain feature extraction, frequency domain feature extraction, deep learning feature extraction and network feature extraction.

[0058] Furthermore, time domain feature extraction is performed on torque data and water level pressure data to extract basic statistics such as mean, variance, peak-to-peak value, waveform energy, etc., reflecting the instantaneous change characteristics of the signal (such as abnormal torque fluctuations and sudden pressure increases).

[0059] Frequency domain feature extraction uses fast Fourier transform (FFT) or wavelet transform to extract the spectrum energy distribution and main frequency components of water soundprint data, and capture the soundprint characteristics of the underwater environment (such as water flow impact sound and mechanical vibration noise).

[0060] Deep learning feature extraction uses convolutional neural networks (CNN) or recurrent neural networks (RNN) to automatically learn features of water voiceprint data and extract high-order abstract features (such as periodic or mutation features in voiceprint patterns).

[0061] The multi-sensor data association network is constructed using complex network features, and the interactions between environmental factors (such as the correlation between torque and pressure changes) are analyzed through indicators such as node degree and clustering coefficient.

[0062] Timestamp tags include data collection synchronization and feature serialization.

[0063] Furthermore, during the sensor data acquisition phase, data acquisition synchronization adds millisecond timestamps to the raw data through a hardware clock synchronization protocol to ensure time alignment of multi-source data.

[0064] Feature serialization binds the extracted feature vectors to timestamps to generate a time-stamped temporal feature matrix for subsequent frame-by-frame analysis of the push unit.

[0065] The voiceprint classification model is used to process the water voiceprint obtained by the water voiceprint recognition unit, obtain the frequency spectrum characteristics of the voiceprint through the voiceprint classification model, and classify the frequency spectrum characteristics, including rainstorm impact, mechanical vibration and human operation; The voiceprint classification model includes data preprocessing and feature extraction, classification model architecture and classification decision fusion. A specific implementation method includes: The adaptive filtering adopts the NLMS (normalized least mean square) algorithm, and the reference signal is the ambient noise floor (collected through the silent period), which offsets the steady-state noise (such as the sound of water flow) in real time.

[0066] Wavelet threshold denoising performs 5-layer Db4 wavelet decomposition on the soundprint signal, applies soft threshold processing to the detail coefficient (threshold = σ√(2logN), σ is the noise standard deviation), retains the 500-800Hz bubble burst characteristics, the frame length is 25ms (corresponding to the underwater sound wave propagation characteristics), the frame shift is 10ms, and the Hamming window suppresses spectrum leakage. For mechanical vibration soundprints (zero crossing rate <20 times / frame), dynamic frame length adjustment (15-40ms) is enabled to avoid harmonic truncation. The linear frequency (0-8kHz) is mapped to the Mel scale (0-1125 mel) through a 40-channel Mel filter group, focusing on enhancing the 80-500Hz target frequency band, taking the logarithm of the spectrum energy (log(1+E)) to improve the visibility of high-frequency weak signals, and for mechanical vibration data, the harmonic product spectrum (HPS) feature is superimposed to capture the engine harmonic structure.

[0067] Furthermore, the specific implementation method of the classification model architecture includes: Input layer 64×64 Mel-Spectrogram (time-frequency diagram), 3 channels (original spectrum, control spectrum, real-time spectrum).

[0068] The convolution module is used to process the input voiceprint frequency domain time-frequency diagram for convolution operation and output the operation result. Through the analysis of the operation result, the feature channel is squeezed to enhance the weight of the high-frequency component of the voiceprint.

[0069] The classification model architecture outputs the water sound pattern characteristics affected by weather and human operation, and judges the water sound in multiple environments.

[0070] Obtain the torque change rate, water pressure gradient change rate and water sound pattern spectrum characteristics in continuous time; The torque shifting unit is driven to receive the torque shifting rate in the continuous timestamps through the torque shifting rate in the continuous timestamps, and the torque parameter shifting step is adjusted. If the torque data shifting rate per unit time is greater than the highest weight, the torque parameter shifting step is shortened. If the torque data shifting rate per unit time does not exceed the highest weight, the state to be adjusted is entered. The step length is adjusted by the acceleration of the water pressure change at the continuous time stamp. If the acceleration of the water pressure change at the time stamp from t-1 to t is greater than the acceleration of the water pressure change at the continuous time stamp from t-2 to t-1, and the acceleration increase is greater than the minimum threshold, the step length ratio is compressed. The water acoustic signal is obtained through the water soundprint recognition unit, and the step size adjustment is activated through comparison with the acoustic feature library. If the acoustic signal in the frequency band captured by the water soundprint spectrum is bubble burst and vortex, it is judged as water immersion, and the soundprint shift step size of the water soundprint shift unit is adjusted.

[0071] Embodiment 2 The multimodal data fusion analysis and judgment method based on environmental perception also includes: a dynamic tuning unit that adjusts the moving step of the moving unit in real time, receives the multimodal data analysis results of environmental perception, and judges whether the closed space and the power space need to be separated.

[0072] By setting step boundary limits for receiving the adjusted soundprint shift step, water pressure compression step ratio and adjusted torque shift step, and setting intensity levels, the intensity levels include low disturbance and medium-high disturbance; By adjusting the step size boundary, the redundant boundary buffer layer is triggered to make temporary tolerance for the step size adjustment. If the step size temporarily exceeds the rated boundary, the total number of steps after actual adjustment is no greater than the maximum number of steps in the redundant boundary buffer layer. If the buffer layer stays for more than the threshold, it immediately switches to the minimum safe step size.

[0073] If the torque change rate, water pressure gradient change rate and water sound pattern spectrum characteristics of the pushing unit are in the stable domain of the redundant boundary buffer layer after adjustment, the step size of the pushing unit continues to be adjusted according to the real-time data; If the total compensation quantity of the step length adjustment of the push unit is not greater than the maximum step length quantity of the redundant boundary buffer layer, and the buffer layer residence time is greater than a threshold, the step length freezing is triggered.

[0074] A specific implementation method of a dynamic tuning unit includes: Multimodal data perception and feature mapping capture environmental parameters in real time through three types of sensors.

[0075] Furthermore, the torque data is used to monitor the mechanical load changes of the equipment power system and identify abnormal fluctuations such as sudden increases or decreases.

[0076] Water pressure data is used to sense gradient changes in water level pressure and capture instantaneous water pressure shocks (such as pipe ruptures or flood peaks).

[0077] Underwater acoustic signals are analyzed through voiceprint data to identify water immersion characteristics such as bubble burst and vortex cavitation.

[0078] These three types of data correspond to mechanical forces, fluid dynamics, and sound wave propagation characteristics in physical space. When a parameter is abnormal (such as a sudden increase in torque, drastic fluctuations in water pressure, or specific soundprint characteristics), the system will preliminarily determine the level of environmental disturbance (low / medium-high disturbance) and trigger the step adjustment mechanism.

[0079] The logic of dynamic adjustment of step size includes low disturbance scenarios and medium and high disturbance scenarios: Low-disturbance scenarios include: if the data fluctuations are within the safety threshold (such as smooth torque changes and periodic fluctuations in water pressure), the system will advance the analysis process with a benchmark step size (such as 0.5 seconds) and gradually optimize the parameters.

[0080] Medium and high disturbance scenarios include: when a certain parameter significantly exceeds the limit (such as a sudden increase of 25% in torque within 3 seconds or water pressure acceleration exceeding the threshold), the system will compress the step size (such as shortening it to 0.1 seconds) and speed up the response frequency to deal with emergencies.

[0081] For example, when the sound of water immersion (the sound of bubbles bursting) is detected, the system will simultaneously shorten the time window of voiceprint analysis and update the water immersion risk assessment results at a high frequency to buy time for subsequent actions (such as isolating closed spaces).

[0082] A specific implementation method of a redundant boundary buffer layer includes: To avoid system oscillation caused by frequent step size adjustment, the system sets an "elastic range" for step size adjustment. The normal range of step size (such as 0.1 seconds to 2.0 seconds) has the highest adjustment priority within this range, allowing the step size to temporarily exceed the rated upper limit (such as temporarily extending to 2.5 seconds). The trigger conditions include the cumulative number of adjustments not exceeding the preset value (such as 5 times) and the duration of exceeding the rated boundary being less than the threshold (such as 3 times the current step time). When a sudden disturbance causes the step size to exceed the rated boundary, the system will not immediately trigger an alarm, but allow a short-term "over-limit operation" while monitoring whether the disturbance continues. If the over-limit time is too long or the adjustment is too frequent (such as 3 consecutive over-limit times), the system will force a return to a safe step size (such as 0.1 seconds) and start the fault self-check process (such as sensor calibration).

[0083] For example, under the impact of a flood peak, the water pressure gradient may temporarily exceed the processing capacity of the conventional analysis step size. At this time, the system will temporarily expand the step size to the redundant boundary, but if the flood peak lasts for more than 10 seconds, it will be forced to switch to the minimum step size to ensure real-time performance.

[0084] The dynamic tuning unit achieves high efficiency and robustness of the monitoring system in complex environments through collaborative analysis of multimodal data, elastic step adjustment and buffer fault tolerance mechanism.

[0085] Embodiment 3 Methods for separating enclosed space from power space include: The enclosed space can be separated from the power system, and after separation, the enclosed space can float on the water surface. The enclosed space is a space equipment that floats on the water surface independently after being separated from the power system. The enclosed space door is equipped with a sealing strip and is in a sealed state after the door is closed. The enclosed space control module (steering wheel, controller connecting wire, etc.) has a controlled detachable mechanism outside the cabin, and a sealing ring is provided at the connection position where the control module is installed with the enclosed space. The steering column has a detachable mechanism that can be controlled to be separated from the external steering column. The enclosed space vent has a sealed partition door, and the enclosed space is controlled and sealed after separation. A skylight is provided in the enclosed space for entry and exit when necessary. A hidden inflatable airbag is installed on the outer bottom of the enclosed space, water level sensors are installed on both outer sides of the enclosed space, and a level sensor is installed inside the enclosed space. The enclosed space has a battery that can be used after separation, and the enclosed space has a control motherboard.

[0086] Furthermore, the power system is composed of components such as a frame, suspension, machine, motor, battery, transmission system and wheels. The frame connects the entire enclosed space and the power system into a whole and supports the overall equipment quality. A positioning connection seat is provided at the load-bearing connection position of the power system, and a cabin positioning bolt is provided at the corresponding position of the enclosed space. After the cabin positioning bolt is inserted into the positioning seat, it is fixed by a bayonet or tightened to form an integral device. After the bayonet is controlled to be pulled out or unscrewed, the cabin positioning bolt can be freely separated from the positioning seat, and the enclosed space can be separated from the power system. The connection between the enclosed space and the power system is a docking component connection and a docking electrical connection, and the enclosed space and each connector and lead-out wire installation outlet are all sealed.

[0087] The control mainboard software determines the water level and controls the connection mechanism to separate and close the air conditioning vents after falling into the water.

[0088] When the control mainboard software generates an instruction to disconnect the electrical connection between the control enclosed space and the power system, the circuit in the control enclosed space is disconnected from the plug, cutting off the interference source at the point where the plug enters the water, so that the power supply and control mainboard in the separated enclosed space can work normally.

[0089] After detecting that the enclosed space has entered water, the control mainboard software generates instructions to control the inflation device to inflate the inflatable airbags installed on both sides of the bottom of the enclosed space. The inflatable airbags are flat strips before inflation and become long strip cylinders after inflation. There is a hard protective layer on the bottom. The inflation method is controlled by the scene requirements, including slow filling with an air pump or fast filling with gas produced by the reaction of chemical substances. If the chemical substance reacts to produce gas, the proportion of the reaction substance components is adjusted, and the chemical gas production device is installed outside the cabin. Resin sealing is used for waterproofing. The trigger wire required for the chemical reaction inflation device is sealed through the bulkhead, and the fast charging time is greater than 1s. If an air pump is used for inflation, the air pump is installed in the enclosed space, the pipe leads to the airbag, and the pipe interface is sealed with a sealing ring.

[0090] The control motherboard software receives the changing cabin level information, determines the tilt direction according to the algorithm, and generates instructions to control the seat movement motor based on the judgment state, moving the weight upward in the longitudinal and lateral directions of the tilt to maintain a nearly horizontal state.

[0091] After the above instructions are executed, the control main board software detects and calculates that the closed space after separation is horizontal and a large amount of water will not flow in, and then opens the skylight.

[0092] Small micro thrusters can be added to the outside of the enclosed space and controlled from inside the cabin.

[0093] The above automatic separation control program can also be switched to manual operation.

[0094] The specific operation is selected according to actual needs.

[0095] Embodiment 4 After the multi-modal data fusion analysis and judgment based on environmental perception drive, the closed space and dynamic space do not need to be separated and processed, including: The enclosed space and the power space can be used as an integral device that can float on the water surface. The enclosed space door is equipped with a sealing strip and is in a sealed state after the door is closed. There is a sealing ring at the connection position between the enclosed space control module (steering wheel, controller connecting wire, etc.) and the enclosed space installation. The enclosed space vent has a sealed partition door and is controllably closed after entering water. A skylight is set in the enclosed space for entry and exit when necessary. A hidden inflatable airbag is installed at the bottom of the frame, water level sensors are installed on both sides of the enclosed space, and a horizontal sensor is installed inside the enclosed space. The enclosed space has a battery that can be used after entering water, and the enclosed space has a control motherboard.

[0096] Furthermore, the power system is composed of components such as a frame, suspension, machine, motor, battery, transmission system and wheels. The frame connects the entire enclosed space and the power system into a whole, becoming an integrated device without a separation mechanism. The enclosed space and each connector and lead wire installation outlet are all sealed.

[0097] The control mainboard software determines the air conditioner vents will be closed after water falls based on the water level information.

[0098] When the control mainboard software instructs the enclosed space to be electrically isolated from the power system, the circuits in the control enclosed space are disconnected from the external electrical connection, cutting off the interference source at the water inlet, so that the power supply and control mainboard in the separated enclosed space can work normally.

[0099] After detecting water entering the enclosed space, the control main board software generates instructions to control the inflation device to inflate the inflatable airbags installed on both sides of the bottom of the frame. The inflatable airbags are flat strips before inflation and multiple long strip cylinders after inflation. There is a hard protective layer on the bottom. The gas produced by the reaction of chemical substances is quickly charged, and the proportion of the reaction substance components is adjusted. The chemical gas production device is installed outside the cabin and waterproofed by resin sealing. The trigger wires required for the chemical reaction inflation device are sealed through the cabin wall, and the fast charging time is greater than 1s.

[0100] The control motherboard software receives the changing cabin level information, determines the tilt direction according to the tilt algorithm, and generates instructions to control the seat movement motor according to the judgment state, moving the weight upward in the longitudinal and lateral directions of the tilt to maintain a nearly horizontal state.

[0101] Specifically, a specific implementation method of the tilt algorithm includes: The control mainboard software continuously obtains the current horizontal status information of the cabin. The horizontal status information is used to reflect the cabin's tilt in the lateral (left-right direction) and longitudinal (front-back direction) directions. For example, the horizontal status information includes: the left side of the cabin is higher than the right side, or the front is lower than the back, and other similar tilt status description information.

[0102] Determining the tilt direction through the tilt algorithm includes determining the horizontal tilt direction and determining the vertical tilt direction.

[0103] The specific judgment of the lateral tilt direction includes: when the received cabin level information shows that the height of the left side of the cabin is higher than that of the right side, it is judged that the cabin is tilted to the left in the lateral direction; conversely, if the height of the right side is higher than that of the left side, it is judged that the cabin is tilted to the right in the lateral direction. If the heights of the left and right sides are equal, it is considered that there is no lateral tilt.

[0104] When the height of the front of the cabin is higher than that of the rear, it is judged that the cabin is tilted forward in the longitudinal direction; if the height of the rear is higher than that of the front, it is judged that the cabin is tilted backward in the longitudinal direction; if the height of the front and rear is the same, there is no tilt in the longitudinal direction.

[0105] If it is determined that the cabin is tilted to the left, an instruction is generated to let the seat moving motor move the weight of the seat to the right, that is, move it above the lateral tilt; if it is determined that the cabin is tilted to the right, the motor moves the weight of the seat to the left. If there is no lateral tilt, no movement instruction is sent.

[0106] If it is determined that the cabin is tilted forward, the seat movement motor sends a command to move the weight of the seat backward, that is, toward the upper side of the longitudinal tilt; if it is determined that the cabin is tilted backward, the motor moves the weight of the seat forward, and when there is no longitudinal tilt, no movement command is sent.

[0107] The control motherboard software continuously receives new cabin level information and repeats the above steps of tilt direction judgment and motor control command generation. As the cabin level continues to change, the seat movement motor can continue to make corresponding actions to keep the cabin as close to a horizontal state as possible.

[0108] After the above instructions are executed, the control main board software detects and calculates that the enclosed space is in a horizontal state and will not flood with a large amount of water, so the skylight is opened.

[0109] Thrusters can be added to the outside of the enclosed space and controlled from inside the cabin.

[0110] The control program can also be switched to manual operation.

[0111] The specific operation is selected according to actual needs.

[0112] Specific application of automobiles such as Figure 5 As shown, Figure 5 The vehicle passenger compartment (enclosed space) includes: a skylight 1, a steering column 2, an air conditioning vent pipe 3, an accelerator 4, a brake 5, and a cabin positioning bolt 6.

[0113] Among them, it should be noted that Figure 5 The cabin positioning bolt 6 and Figure 6 The positioning bolts described are the same unit and perform the same functions, but Figure 5 The middle cabin positioning bolt 6 is applied to automobiles, and Figure 6 The positioning bolt described in is the application principle.

[0114] Specifically, a specific implementation method for separating a closed space from a power system includes: The passenger compartment can be separated from the driving system, and after separation, the passenger compartment can float on the water. The passenger compartment is a passenger and luggage compartment that floats on the water independently after being separated from the vehicle driving system. The passenger compartment door is equipped with a sealing strip, which is in a sealed state after the door is closed. The accelerator 4 pedals and brake 5 pedals of the passenger compartment have a controlled detachable mechanism outside the cabin. The connection position where the accelerator 4 pedals and brake 5 pedals are installed with the passenger compartment has a sealing ring, the steering wheel assembly has a sealing ring at the connection position where it is installed with the passenger compartment, and the detachable steering column has a detachable mechanism. The structure can be controllably separated from the external steering column 2, the air-conditioning vents in the passenger compartment are provided with sealed partition doors, the passenger compartment is controlled and closed after being separated from the frame, the passenger compartment is provided with a large skylight 1 for the passengers to enter and exit when necessary, there are longitudinal and transverse slides under the passenger seats for controlled movement as needed, hidden inflatable airbags are installed on both sides of the bottom outside the passenger compartment, the outward connecting pipelines are sealed at the bulkhead, water level sensors are installed on both sides of the passenger compartment, a horizontal sensor is installed in the passenger compartment, the passenger compartment has a battery that can be used after separation, and the passenger compartment has a control motherboard.

[0115] Further, the detachable steering column is composed of a universal joint and a steering column upper portion connected to a split lower portion; like Figure 7 As shown, the universal joint and the upper part of the steering column include a universal joint 8, a separable shaft upper part 9, and an inter-shaft groove 10.

[0116] The lower part of the split body includes a protruding body 11 , a connecting sleeve 12 , a lower part of a steering column 13 , and a lower end connected to a steering mechanism 14 .

[0117] The universal joint and the upper part of the steering column are installed outside the passenger compartment and connected to the steering wheel assembly. The lower end of the detachable shaft upper part 9 is installed vertically downward and connected to the upper end of the vertically installed split lower part. The upper end of the split lower part includes a protrusion 11 and a connecting sleeve 12, which are connected to the lower end of the detachable shaft upper part 9. The detachable steering column split lower part is installed at the connecting steering mechanism 14 of the traveling system.

[0118] Furthermore, the protrusion 11 is connected with the inter-axle groove 10 to form a complete detachable steering column. If the passenger compartment floats up, the universal joint and the upper part of the steering column are separated from the lower part, so that the passenger compartment is separated from the driving system.

[0119] The universal joint 8 is provided to adjust the angle of the detachable steering column to a vertical state.

[0120] The passive protection of traditional vehicles is transformed into a dynamic risk avoidance mechanism. Through the closed-loop control of water level sensors, control motherboards, and separation actuators, an independent living space is created in the event of a fall into water. Compared with the traditional window-breaking escape method, it greatly improves the survival rate and has a wide range of applications.

[0121] Furthermore, hidden airbags are used to increase buoyancy, and horizontal sensors (±0.1° accuracy) are used to drive the counterweight slide rail (maximum adjustment torque 500N·m) and micro-thrusters (bidirectional vector nozzles) to achieve precise obstacle avoidance within a radius of 5m. It also provides greater buoyancy to ensure smooth separation of the passenger compartment and the driving system, and has a large load-bearing capacity.

[0122] The battery in the passenger compartment is connected in parallel with the main battery of the driving system vehicle, and other driving system signal lines and driving system control lines are electrically connected to the passenger compartment and the driving system through integrated plugs and sockets to form a detachable system.

[0123] The control mainboard software determines the water level and controls the connection mechanism to separate and close the air conditioning vents after falling into the water.

[0124] When the control main board software generates instructions to disconnect the electrical plug-in connectors between the passenger compartment and the driving system, the circuit in the control compartment is disconnected from the plug, cutting off the interference source at the point where the plug enters the water, allowing the power supply and control main board in the separated passenger compartment to work normally.

[0125] Furthermore, modules electrically connected to the vehicle driving system, such as the brake 5, the accelerator 4 and the air conditioning vent 3, are all provided with sealing rings or controlled sealing doors to prevent water from entering the passenger compartment after separation.

[0126] Controlled sealing doors are used for air conditioning vent sealing.

[0127] like Figure 6 As shown, it is a connection diagram between the passenger compartment and the frame, which includes: a passenger compartment floor 15, a bayonet 17, a positioning seat 18 and a frame plate 19. A positioning connection seat is provided at the load-bearing connection position of the frame structure of the automobile running system, and a positioning bolt is provided at the corresponding position of the passenger compartment. The positioning bolt is inserted into the positioning seat 18 and fixed by the bayonet 17. After being fixed, they are connected into a whole. Furthermore, the passenger compartment floor 15 and the bayonet pin groove 16 form a lower positioning bolt, wherein the bayonet pin groove 16 is used to embed the bayonet pin 17, and the bayonet pin 17 is controlled to be pulled out electrically.

[0128] The frame plate 19 is used to support the frame, and the positioning seat 18 is used to fix the passenger compartment and the frame.

[0129] Importantly, it should be noted that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are only exemplary. Although only a plurality of embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible, for example, the size, scale, structure, shape and ratio of various elements, and parameter values ​​(e.g., temperature, pressure, etc.), installation arrangement, use of materials, color, directional changes, etc., without substantially departing from the novel teachings and advantages of the subject matter described in the application. For example, the element shown as integrally formed can be composed of a plurality of parts or elements, the position of the element can be inverted or otherwise changed, and the nature or number or position of the discrete element can be changed or changed. Therefore, all such modifications are intended to be included in the scope of the present invention. The order or sequence of any process or method steps can be changed or reordered according to alternative embodiments. Any "device plus function" clause is intended to cover the structure of the execution function described herein, and is not only structurally equivalent but also equivalent structure. Without departing from the scope of the present invention, other replacements, modifications, changes and omissions can be made in the design, operating conditions and arrangement of the exemplary embodiments. Therefore, the invention is not limited to a specific embodiment, but extends to numerous modifications still falling within the scope of the appended claims.

[0130] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment may not be described (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those features that are not relevant to implementing the invention).

[0131] It should be understood that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multimodal data fusion analysis and judgment method based on environmental perception, characterized in that: include: The sensor unit that obtains environmental parameter data aligns different data dimensions by establishing a multi-data radiation field and makes a preliminary judgment on data anomalies in the environmental data; The data abnormality judgment includes environmental data baseline, single-mode abnormality and secondary verification; The secondary verification includes time series correlation verification and space propagation verification. The time series correlation verification is used to establish the sudden change, imbalance and decoupling of the continuous time stamp torque data. The space propagation verification is used to obtain the sudden change, imbalance and decoupling of the water level pressure data and the water sound pattern data. Extract features from environmental data, assign the extracted environmental data features to tags with timestamps, and input them into a shifting unit, wherein the shifting unit sets a multi-level analysis shift to analyze and judge multiple environmental data, and outputs environmental analysis results; The dynamic tuning unit adjusts the moving step length of the moving unit in real time, receives the environmental analysis result, and determines whether the closed space and the power space need to be separated.

2. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 1 is characterized by: Establish a multi-data radiation field, which includes a pressure field and a voiceprint field, and determine the time-varying correlation between the pressure and voiceprint signals through alignment coding. The pressure field is constructed based on an unstructured grid discretization method to map the water pressure data into a three-dimensional radiation field; If the pressure mapped in the pressure field changes suddenly and the high-frequency area of ​​the voiceprint mapped in the voiceprint field is abnormal, the grid density is dynamically optimized by dividing the multi-data radiation field into high-resolution grids and extracting local neighborhood features.

3. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 2 is characterized by: The environmental data baseline is used to compare whether the acquired torque data, water level pressure data and water sound pattern data are within the normal threshold range of the environmental data. If the torque data, water level pressure data and water sound pattern data are not within the normal threshold range of the environmental data, the data abnormality is judged. The single-mode anomaly is used to map whether there are mutations, imbalances and decoupling in the torque data, water level pressure data and water sound pattern data. If there is a single data anomaly in any of the torque data, water level pressure data and water sound pattern data, a single environmental data anomaly report is output and whether the sensor unit is abnormal is detected.

4. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 3 is characterized by: The pushing unit includes a torque pushing unit, a water level pressure pushing unit and a water sound pattern pushing unit; The torque shift unit takes the torque feature with a timestamp tag as input, obtains the difference between the torque feature at the previous moment and the torque feature at the current moment, and obtains the difference between the torque feature at the next moment and the torque feature at the current moment by real-time monitoring of the torque features within continuous timestamps, and triggers an abnormal mark within a continuous time.

5. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 4 is characterized by: The water level pressure shift unit sets a water level pressure prediction model, predicts the water level pressure data at the next moment through the water level pressure prediction model, and marks the flooding risk through mutation detection; The mutation detection includes predicting whether a sudden change occurs in the gradient change between the pressure and the actual value, processing the pressure signal by short-time Fourier transform, and identifying the sudden increase of the high-frequency signal component; The water soundprint shifting unit is provided with a soundprint classification model and a water soundprint recognition unit.

6. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 5 is characterized by: The voiceprint classification model is used to process the water voiceprint obtained by the water voiceprint recognition unit, obtain the frequency spectrum characteristics of the voiceprint through the voiceprint classification model, and classify the frequency spectrum characteristics, including rainstorm impact, mechanical vibration and human operation; The water soundprint recognition unit is used to obtain multi-environment water sound data and water soundprint optimization. The multi-environment water sound includes water soundprint features affected by weather and water soundprint features affected by human operation. The water soundprint optimization is used to optimize the acquired water soundprint.

7. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 6 is characterized by: Obtain the torque change rate, water pressure gradient change rate and water sound pattern spectrum characteristics in continuous time; The torque shifting unit is driven to receive the torque shifting rate in the continuous timestamps through the torque shifting rate in the continuous timestamps, and the torque parameter shifting step is adjusted. If the torque data shifting rate per unit time is greater than the highest weight, the torque parameter shifting step is shortened. If the torque data shifting rate per unit time does not exceed the highest weight, the state to be adjusted is entered. The step length is adjusted by the acceleration of the water pressure change at the continuous time stamp. If the acceleration of the water pressure change at the time stamp from t-1 to t is greater than the acceleration of the water pressure change at the continuous time stamp from t-2 to t-1, and the acceleration increase is greater than the minimum threshold, the step length ratio is compressed. The water acoustic signal is obtained through the water soundprint recognition unit, and the step size adjustment is activated through comparison with the acoustic feature library. If the acoustic signal in the frequency band captured by the water soundprint spectrum is bubble burst and vortex, it is judged as water immersion, and the soundprint shift step size of the water soundprint shift unit is adjusted.

8. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 1 is characterized by: By setting step boundary limits for receiving the adjusted soundprint shift step, water pressure compression step ratio and adjusted torque shift step, and setting intensity levels, the intensity levels include low disturbance and medium-high disturbance; By adjusting the step size boundary, the redundant boundary buffer layer is triggered to make temporary tolerance for the step size adjustment. If the step size temporarily exceeds the rated boundary, the total number of steps after actual adjustment is no greater than the maximum number of steps in the redundant boundary buffer layer. If the buffer layer stays for more than the threshold, it immediately switches to the minimum safe step size.

9. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 8 is characterized by: If the torque change rate, water pressure gradient change rate and water sound pattern spectrum characteristics of the pushing unit are in the stable domain of the redundant boundary buffer layer after adjustment, the step size of the pushing unit continues to be adjusted according to the real-time data; If the total compensation quantity of the step length adjustment of the push unit is not greater than the maximum step length quantity of the redundant boundary buffer layer, and the buffer layer residence time is greater than a threshold, the step length freezing is triggered.

10. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 9 is characterized in that: If it is determined that the enclosed space needs to be separated, a sealing strip is provided on the hatch of the enclosed space, and the hatch is in a closed state when it is closed. A position sealing ring is provided at the place where the enclosed space is connected to the outside world, and all sensors are provided for wire control to convert physical control into digital signals, and transmit them to the main controller, which controls the separation of the enclosed space mechanism connector and the electrical connector; The closed space receives the separation signal output by the multimodal data analysis, which triggers the separation of the closed space and the power space, triggers the closing of the air-conditioning ventilation duct, triggers the inflation of the inflatable airbag outside the closed space, and triggers the separation balancer, which moves the weight upward in the inclined longitudinal and lateral directions to maintain a nearly horizontal state, ultimately completing the separation of the closed space and the power space.

11. The multimodal data fusion analysis and judgment method based on environmental perception drive according to claim 9 is characterized in that: If the enclosed space does not need to be separated, after the enclosed space without a separation mechanism receives the corresponding signal output by the multimodal data analysis, it triggers the closing of the air-conditioning ventilation duct, triggers the inflation of the inflatable airbag outside the enclosed space, and triggers the balancer. The balancer moves the weight upward in the inclined longitudinal and lateral directions to maintain a nearly horizontal state, triggers the electrical isolation of the enclosed space and the power space, and finally completes the overall floating.

Images