Multimodal Data Fusion Analysis and Judgment Method Driven by Environmental Perception

Abstract

The present invention relates to the technical field of multi-modal electrical variable analysis, and discloses a multi-modal data fusion analysis and judgment method driven by environmental perception, including a sensor unit for acquiring environmental parameter data, preliminarily judging data anomalies through environmental data, extracting features of the environmental data, attaching the extracted environmental data features to tags with timestamps, and inputting them into a shifting unit. The shifting unit is provided with multiple levels of analysis and shifting to analyze and judge multi-environment data, output an environmental analysis result, and a dynamic optimization unit for real-time adjusting the shifting step size of the shifting unit. The dynamic optimization unit receives the multi-modal data analysis result of environmental perception, judges whether the enclosed space and the power space need to be separated, realizes the analysis and judgment of multi-environment perception, and improves the automation and intelligence of the separation of the enclosed space and the power space.

Description

Technical Field

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

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

[0003] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title, but such simplifications or omissions shall not be used to limit the scope of the present invention.

[0004] To solve the above technical problems, the main purpose of the present invention is to provide a multimodal data fusion analysis and judgment method driven by environmental perception. Among them, the multimodal data fusion analysis and judgment method driven by environmental perception includes:

[0005] A sensor unit for acquiring environmental parameter data, aligning different data dimensions by establishing a multi-data radiation field, and preliminarily judging data anomalies in the environmental data;

[0006] Extract features from the environmental data, assign the extracted environmental data features with time stamps, and input them into a shifting unit. The shifting unit sets multi-level analysis shifts to analyze and judge multi-environmental data, and outputs an environmental analysis result;

[0007] A dynamic optimization unit for real-time adjusting the shifting step of the shifting unit, receiving the environmental analysis result, and judging whether the closed space and the dynamic space need to be separated.

[0008] As a preferred solution of the multimodal data fusion analysis and judgment method driven by environmental perception of the present invention, among them:

[0009] Establish a multi-data radiation field, the multi-data radiation field includes a pressure field and a voiceprint field, and determine the time-varying correlation between the pressure and the voiceprint signal through alignment coding. The pressure field is constructed based on an unstructured grid discretization method, and the water pressure data is mapped into a three-dimensional radiation field;

[0010] If there are sudden changes in the pressure mapped in the pressure field and anomalies in the high-frequency region of the voiceprint mapped in the voiceprint field, then by dividing the high-resolution grid of the multi-data radiation field and extracting local neighborhood features, the grid density is dynamically optimized.

[0011] As a preferred solution of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein:

[0012] The environmental data baseline is used to compare whether the obtained torque data, water level pressure data, and underwater voiceprint data are within the normal threshold range of the environmental data. If the torque data, water level pressure data, and underwater voiceprint data are not within the normal threshold range of the environmental data, it triggers the judgment of data anomalies, and the judgment of data anomalies includes environmental data baseline, single-modal anomalies, and secondary verification;

[0013] The single-modal anomaly is used to map whether there are sudden changes, imbalances, and decouplings in the torque data, water level pressure data, and underwater voiceprint data. If any one of the torque data, water level pressure data, and underwater voiceprint data has a single data anomaly, a single environmental data anomaly report is output, and whether the sensor unit is abnormal is detected;

[0014] The secondary verification includes time-series correlation verification and spatial propagation verification. The time-series correlation verification is used to establish sudden changes, imbalances, and decouplings of torque data with continuous timestamps, and the spatial propagation verification is used to obtain sudden changes, imbalances, and decouplings of water level pressure data and underwater voiceprint data.

[0015] As a preferred solution of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein:

[0016] The shift unit includes a torque shift unit, a water level pressure shift unit, and an underwater voiceprint shift unit;

[0017] The torque shift unit takes the torque feature with a timestamp label as the input, and by monitoring the torque feature within continuous timestamps in real time, obtains the difference between the torque feature at the previous moment and the current moment and the difference between the torque feature at the next moment and the current moment, and triggers an anomaly mark within continuous time.

[0018] As a preferred solution of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein:

[0019] The water level pressure shift unit sets up 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;

[0020] The mutation detection includes predicting whether there is a mutation in the gradient change between the predicted pressure and the actual value, processing the pressure signal through short-time Fourier transform, and identifying the sudden increase part of the high-frequency signal components;

[0021] The underwater acoustic fingerprint migration unit is provided with an acoustic fingerprint classification model and an underwater acoustic fingerprint recognition unit.

[0022] As a preferred embodiment of the multi-modal data fusion analysis and judgment method based on environmental perception driving of the present invention, wherein:

[0023] The acoustic fingerprint classification model is used to process the underwater acoustic fingerprints obtained by the underwater acoustic fingerprint recognition unit, obtain the spectral features of the acoustic fingerprints through the acoustic fingerprint classification model, and classify the spectral features, including rainstorm impact, mechanical vibration, and human operation;

[0024] The underwater acoustic fingerprint recognition unit is used to obtain multi-environment underwater acoustic data and underwater acoustic fingerprint optimization. The multi-environment underwater acoustics include underwater acoustic fingerprint features affected by weather and underwater acoustic fingerprint features affected by human operation. The underwater acoustic fingerprint optimization is used to optimize the obtained underwater acoustic fingerprints.

[0025] As a preferred embodiment of the multi-modal data fusion analysis and judgment method based on environmental perception driving of the present invention, wherein:

[0026] Obtain the torque change rate, water pressure gradient change rate, and underwater acoustic fingerprint spectral features within a continuous time;

[0027] Through the torque change rate within the continuous time stamp, the torque migration unit is driven to receive the torque change rate within the continuous time stamp and adjust the torque parameter migration step size. If the torque data change rate per unit time is greater than the highest weight, the torque parameter migration step size is shortened. If the torque data change rate per unit time does not exceed the highest weight, it enters the state to be adjusted;

[0028] Adjust the step migration length through the water pressure change acceleration within the continuous time stamp. If the water pressure change acceleration from time stamp t-1 to t rises more than the water pressure change acceleration of the continuous time stamp from t-2 to t-1, and the acceleration rise is greater than the minimum threshold, the step size ratio is compressed;

[0029] Obtain the underwater acoustic signal through the underwater acoustic fingerprint recognition unit, and activate the step size adjustment through comparison with the acoustic feature library. If the acoustic signal captured by the underwater acoustic fingerprint spectrum is bubble rupture and vortex, it is judged as water immersion, and the acoustic fingerprint migration step size of the underwater acoustic fingerprint migration unit is adjusted.

[0030] As a preferred embodiment of the multi-modal data fusion analysis and judgment method based on environmental perception driving of the present invention, wherein:

[0031] By setting step - size boundaries to limit the acoustic fingerprint shift step - size for reception adjustment, the water - pressure compression step - size ratio, and the adjusted torque shift step - size, and setting intensity levels, the intensity levels include low disturbance and medium - high disturbance;

[0032] By adjusting the step - size boundary, trigger the redundant boundary buffer layer to make a temporary tolerance for step - size adjustment. If the step - size briefly exceeds the rated boundary, the total number of actually adjusted step - sizes is not greater than the maximum step - size number of the redundant boundary buffer layer. If the residence time of the buffer layer exceeds the threshold, immediately switch to the minimum safety step - size.

[0033] As a preferred embodiment of the multi - modal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein:

[0034] If the torque change rate, water - pressure gradient change rate, and underwater acoustic fingerprint spectrum characteristics after the adjustment of the shift unit are within the stable domain of the redundant boundary buffer layer, continue to adjust the step - size of the shift of the shift unit according to real - time data;

[0035] If the total compensation number of the step - size adjustment of the shift unit is not greater than the maximum step - size number of the redundant boundary buffer layer, and the residence time of the buffer layer is greater than the threshold, trigger step - size freezing.

[0036] As a preferred embodiment of the multi - modal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein:

[0037] If it is determined that the enclosed space needs to be separated, the hatch of the enclosed space is provided with a sealing strip. When the hatch is closed, it is in a closed state. A position sealing ring is provided at the connection between the enclosed space and the outside. All wire - controlled controls are provided with sensors to convert physical control into digital signals and transmit them to the main controller, and the main controller controls the separation of the mechanical and electrical connectors of the enclosed space;

[0038] When the enclosed space receives the separation signal output from the multi - modal data analysis, trigger the separation of the enclosed space and the power space, trigger the closing of the air - conditioning ventilation duct, trigger the inflation of the inflatable airbag outside the enclosed space, and trigger the separation balancer. The balancer moves the weight upward in the inclined longitudinal and transverse directions to maintain an approximately horizontal state, and finally completes the separation of the enclosed space and the power space.

[0039] As a preferred embodiment of the multi - modal data fusion analysis and judgment method based on environmental perception drive of the present invention, wherein:

[0040] If the enclosed space does not need to be separated, after the enclosed space without a separation mechanism receives the corresponding signal output from the multi - modal data analysis, trigger the closing of the air - conditioning ventilation duct, trigger the inflation of the inflatable airbag outside the enclosed space, and trigger the balancer. The balancer moves the weight upward in the inclined longitudinal and transverse directions to maintain an approximately horizontal state, trigger the electrical isolation between the enclosed space and the power space, and finally complete the overall floating.

[0041] Advantages of the present invention:

[0042] Through the dual verification of time series correlation (continuous torque mutation) and spatial propagation (pressure wave and acoustic fingerprint time difference matching), this application improves the confidence level of immersion judgment. The redundant boundary buffer layer allows the step size to temporarily exceed the boundary (such as 10% beyond the theoretical value), reducing the action lag caused by excessive constraints while ensuring safety. By setting multiple shifting units, parallel and efficient individual processing of multiple environmental parameters is achieved, improving the analysis efficiency and accuracy of multi-modal fusion analysis for enclosed spaces and dynamic spaces. By setting the boundary redundancy and step size adjustment freezing mechanism, the maximum adjustment of the step size for multi-environment perception is realized, improving the multi-modal response speed. Description of the Drawings

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0044] Figure 1 It is a flowchart of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention;

[0045] Figure 2 It is the environmental perception multi-modal analysis method of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention;

[0046] Figure 3 It is a schematic diagram of the underwater acoustic fingerprint recognition unit in the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention;

[0047] Figure 4 It is a schematic diagram of the principle of the shifting unit of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention;

[0048] Figure 5 It is a schematic diagram of the enclosed space of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention;

[0049] Figure 6 It is a connection diagram of the passenger compartment and the vehicle frame in the practical application of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention;

[0050] Figure 7 It is a schematic diagram of the lower positioning bolt of the passenger compartment in the practical application of the multi-modal data fusion analysis and judgment method based on environmental perception drive of the present invention;

[0051] 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

[0052] 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.

[0053] 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.

[0054] 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.

[0055] Embodiment 1

[0056] like Figure 1 As shown, the multimodal data fusion analysis and judgment method based on environmental perception driving includes:

[0057] 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.

[0058] 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;

[0059] The determination of data anomalies includes environmental data baseline, single-mode anomalies and secondary verification.

[0060] 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;

[0061] The single-modal anomaly is used to map whether there are mutations, imbalances, and decouplings in torque data, water level pressure data, and underwater acoustic fingerprint data. If there is a single-data anomaly in any of the torque data, water level pressure data, and underwater acoustic fingerprint data, a single environmental data anomaly report is output, and whether the sensor unit is abnormal is detected;

[0062] The secondary verification includes time-series correlation verification and spatial propagation verification. The time-series correlation verification is used to establish mutations, imbalances, and decouplings of torque data with continuous timestamps, and the spatial propagation verification is used to obtain mutations, imbalances, and decouplings of water level pressure data and underwater acoustic fingerprint data.

[0063] Specifically, install a 4-channel MEMS microphone acoustic fingerprint sensor array to cover the target frequency band of 500 - 800 Hz, support capturing sound wave incident angles of ±30°, integrate a pressure sensor with a range of 0 - 100 kPa and an accuracy of ±0.1% FS, use a spiral channel to enhance the sensitivity of the fluid pressure gradient, deploy a strain torque sensor (non-contact, range 0 - 500 Nm), and isolate mechanical vibration interference through a three-axis shock absorption system.

[0064] Eliminate high-frequency vibration noise through moving average filtering (window = 10 data points), smooth turbulent fluctuations through Kalman filtering, and reconstruct through wavelet packet decomposition (retain the bubble rupture characteristic frequency band of 500 - 800 Hz).

[0065] The specific implementation method for establishing the environmental data baseline includes:

[0066] Based on historical data statistics, the normal fluctuation range = rated value ± 5% (for example, for a 500 Nm system, the baseline is 475 - 525 Nm). Divide the steady-state pressure interval through clustering, the threshold = mean ± 3σ, where σ is the standard deviation of the gradient. Construct an MFCC feature library (acoustic fingerprint template under normal working conditions), the matching degree threshold = 85% (abnormal if lower than 70%). Each node uploads encrypted features (such as torque statistical distribution, MFCC mean of acoustic fingerprint) every 24 hours, and the cloud aggregates and then distributes new thresholds. Simulate interferences such as heavy rain (200 - 500 Hz broadband noise) and mechanical shock (800 Hz harmonic) to verify the robustness of the baseline.

[0067] Single-modal anomaly detection includes: torque mutation, pressure mutation, and acoustic fingerprint mutation.

[0068] The calculation gradient of torque mutation Δτ / Δt > 25% / s (for example, in a 500 Nm system, the change within 5 seconds > 125 Nm). If it exceeds the limit continuously for 3 times, an alarm will be triggered. Here, Δτ is the torque change rate, Δt is the time interval, and the meaning of the calculation gradient of torque mutation is the time change rate of torque (τ, unit: Nm), which reflects the relative change amplitude of torque per unit time. Whether torque mutation occurs is judged by the change amplitude, (current torque - torque at the previous moment) / rated torque of the system × 100% / time interval.

[0069] For the analysis of pressure mutation, the high-frequency energy increase in the range of 500 - 800 Hz > 30% (compared with the baseline), or the gradient ΔP / Δt > 10 kPa / s. Similarly, here, ΔP is the pressure change rate, Δt is the time interval, and it is the time change rate of pressure (P, unit: kPa), which directly reflects the sudden change of pressure.

[0070] For voiceprint mutation, it is judged by the spectral variance > 15% (window = 1 s), or the main frequency deviation > 10%.

[0071] The imbalance of data is judged by a single data continuously exceeding the baseline threshold (for example, pressure > 105 kPa for more than 10 seconds).

[0072] The decoupling judgment includes the cross-covariance of torque - pressure < 0.6 (under normal working conditions, τ - P should be strongly correlated), and the deviation of the propagation time delay between voiceprint - pressure > 5% (theoretical time delay = distance / 1480 m / s), where τ is the torque data and P is the pressure data.

[0073] For the secondary verification, it is verified whether there is data disconnection through time series correlation verification, and the data in the mutation area output by the multi-data radiation field analysis is verified through space propagation verification.

[0074] Specifically, the strain torque sensor adopts a non-contact design, detects torque changes through strain gauges on the elastic shaft, outputs a 0 - 10V analog signal, and ensures that the axis concentricity < Φ0.05 mm during deployment, and realizes real-time data transmission through wireless telemetry.

[0075] The water level pressure sensor selects a MEMS resonant pressure sensor with a range of 0 - 100 kPa, an accuracy of ±0.1% FS, supports 4 - 20 mA output, and needs to be deployed according to the triangular topology during installation, with a spacing not exceeding 3 times the pipe diameter to eliminate turbulent interference.

[0076] The underwater voiceprint sensor is configured as a capacitive microphone array with a sampling rate ≥ 1 kHz, combines morphological filtering to eliminate rainstorm noise (200 - 500 Hz), and extracts the characteristics of bubble rupture in the range of 500 - 800 Hz.

[0077] such as Figure 3As shown, it includes a top acoustic wave inlet 101, a spiral acoustic wave processing structure 202, and a support unit 303. Among them, the top acoustic wave inlet 101 adopts a horn-shaped acoustic waveguide structure with an opening diameter matching the acoustic wave wavelength. For example, the designed opening diameter is about 6 - 15 cm, and the interior is coated with sound-absorbing material to suppress high-frequency reflected noise. Through the design of gradual change in acoustic impedance, the free-field acoustic wave is gradually transitioned to the interior of the device. A dust-proof net (mesh number ≥ 2000) is set at the inlet to filter out interfering substances such as dust and water droplets while maintaining a sound transmission rate (> 90%), optimize the acoustic wave incident angle (reception cone angle ± 30°), improve the capture efficiency of the target sound source (such as the underwater bubble rupture sound), and suppress environmental noise (such as wind noise and mechanical vibration) from entering the subsequent processing link.

[0078] The spiral acoustic wave processing structure 202 is composed of multiple-stage spiral waveguides, made of 3D printed acoustic resin material. The spiral pitch decreases according to an exponential law (the starting pitch is λ / 2, where λ is the target acoustic wave wavelength). The end of each stage of the spiral is connected to a resonance cavity (the volume matches 1 / 4 wavelength of the target frequency). Through the acoustic lens effect of the spiral structure, the incident acoustic wave energy is focused on the central axis, and the sound pressure level gain can reach 12 dB (compared with the non-guided structure). Further, different pitches correspond to different frequency bands (for example, the first-stage spiral processes low frequencies of 80 - 120 Hz, and the second stage processes medium frequencies of 350 - 500 Hz), realizing the physical pre-separation of the acoustic fingerprint spectrum. The serrated texture (depth ≈ λ / 10) on the inner wall of the spiral destroys the acoustic wave reflection path, suppresses the formation of standing waves, reduces the probability of spectral aliasing, guides the broadband acoustic wave to different processing channels according to the frequency band, enhances the target characteristics (such as the 500 - 800 Hz pulse group of the bubble rupture sound), and simultaneously attenuates the interfering frequency bands (such as the 20 - 50 Hz low-frequency vibration of the ship engine).

[0079] The support unit 303 integrates a three-axis shock absorption system with adjustable damping coefficient, internally embedded with a MEMS capacitive microphone array (4 channels, SNR > 70 dB), and equipped with a threaded interface (such as M18×1.5) at the bottom to support rapid installation and attitude calibration. Through the mass-spring-damping system (m-c-k model), the resonance frequency of the device is reduced to below 5 Hz to avoid external vibration coupling to the acoustic components. The time delay difference (TDOA) of the 4-channel microphones is used to achieve sound source localization (accuracy ±0.5°), and combined with the adaptive beamforming algorithm (such as MVDR) to improve the signal-to-noise ratio, ensuring the stable operation of the acoustic components in a complex environment, and at the same time enhancing the spatial selectivity of the target sound source through array processing.

[0080] Establish a multi-data radiation field, where the multi-data radiation field includes a pressure field and an acoustic fingerprint field, and determine the time-varying correlation between the pressure and acoustic fingerprint signals through alignment coding;

[0081] Specifically, the construction of the pressure field includes:

[0082] The construction of the pressure field is based on the unstructured grid discretization method. The water pressure data is mapped into a three-dimensional radiation field, and the spatial distribution characteristics of the pressure gradient mutation region are captured through finite element analysis. For example, the sudden change in pressure (ΔP / Δt > 10 kPa / s) caused by fluid impact at the immersion point. The calculation accuracy can be improved by setting high-resolution grids.

[0083] When the detected pressure gradient exceeds the threshold (10 kPa / s), the high-resolution grid is encrypted to an accuracy of 0.1 m, and the other areas maintain a sparse grid of 0.5 m to ensure that computing resources are preferentially allocated to key areas.

[0084] The grid density is dynamically adjusted according to the amplitude of the pressure fluctuation to balance the computing efficiency and accuracy, and at the same time ensure the rationality of the adjustment. For example, the grid resolution is enhanced in the pressure wave propagation path (such as at the pipe bend).

[0085] Specifically, the establishment of the acoustic fingerprint field includes:

[0086] The construction of the acoustic fingerprint field focuses on the high-frequency components of 500 - 800 Hz (corresponding to the characteristics of bubble rupture and vortex cavitation). Its physical basis is the dispersion characteristics of acoustic wave propagation. High-frequency acoustic fingerprints attenuate faster in liquids, and local transient characteristics are captured through high-precision grids.

[0087] The acoustic fingerprint signal is decomposed by five-layer wavelet packet decomposition, and the energy distribution of the key frequency band of 2 - 5 kHz is extracted to drive the grid to be encrypted to an accuracy of 0.05 m in the sound source area (such as the immersion point).

[0088] Environmental interference (such as low-frequency noise in heavy rain) is suppressed through anti-noise methods of eliminating impulse noise and filling spectral holes (such as obtaining the anti-noise coefficient to eliminate noise pollution).

[0089] Based on the time difference of arrival (TDOA) positioning method, the time difference of arrival of acoustic waves is calculated through a multi-sensor array, and the spatial coordinates of the sound source are inverted in combination with the wave speed (1480 m / s) and spatially matched and verified with the pressure field mutation region.

[0090] Specific alignment coding includes:

[0091] The input layer receives the four-dimensional data (x, y, z, t) of the multi-data radiation field with time-stamped labels, and models the propagation time delay relationship between the pressure wave and the acoustic fingerprint signal through a gating mechanism. For example, it takes Δt1 for the pressure wave to propagate to the sensor, while the corresponding acoustic fingerprint signal is delayed by Δt2 due to medium differences. LSTM learns the non-linear mapping of Δt1 - Δt2 and enforces wave speed matching through the LSTM loss function.

[0092] If there are mutations in the pressure mapped in the pressure field and abnormalities in the high-frequency region of the acoustic fingerprint mapped in the acoustic fingerprint field, then by dividing the high-resolution grid of the multi-data radiation field and extracting local neighborhood features, the grid density is dynamically optimized.

[0093] By setting a multi - data radiation field to dynamically allocate computing resources to abnormal data and output accurate processed water pressure and acoustic fingerprint data, it reduces waste of computing resources while ensuring data accuracy.

[0094] As Figure 2 shown, feature extraction is performed on environmental data, the extracted environmental data features are given tags with timestamps, and are input into a shifting unit. The shifting unit sets multi - level analysis and shifting to analyze and judge multi - environmental data, and outputs an environmental analysis result;

[0095] As Figure 4 shown, a specific implementation method of a shifting unit:

[0096] The shifting unit includes a torque shifting unit, a water level pressure shifting unit, and an acoustic fingerprint shifting unit.

[0097] Among them, the torque shifting unit starts with torque input, outputs torque anomalies through the first torque detection and the second torque detection, and is denoted as the torque analysis result.

[0098] The water level pressure shifting unit starts with water pressure input, outputs water pressure anomalies through the first water pressure detection and the second water pressure detection, and is denoted as the water pressure analysis result.

[0099] The acoustic fingerprint shifting unit starts with acoustic fingerprint input, outputs acoustic fingerprint anomalies through the first acoustic fingerprint detection and the second acoustic fingerprint detection, and is denoted as the acoustic fingerprint analysis result.

[0100] Through parallel processing of the shifting unit, combining the torque, water pressure, and acoustic fingerprint analysis results, it is judged whether the enclosed space and the power system need to be separated. If all three meet the separation conditions simultaneously, the separation of the enclosed space and the power system is responded to. If the three do not meet the separation conditions simultaneously, the separation of the enclosed space and the power system is not triggered, and the detection continues.

[0101] By receiving the high - frequency region output by the multi - data radiation field, dividing the high - resolution network, dynamically adjusting the shifting step, and limiting the shifting step adjustment range by the dynamically adjusted redundant buffer layer to prevent distortion.

[0102] The immediate separation of three - mode synchronization anomalies, short separation trigger time, single / double anomaly start verification, low false - trigger situation, greatly improves the separation accuracy of the enclosed space and the power system, and greatly reduces the response time, increasing the system's linkage.

[0103] Specifically, the feature extraction of environmental data includes multi - dimensional feature extraction and timestamp tags.

[0104] Furthermore, the multi-dimensional feature extraction technology includes time-domain feature extraction, frequency-domain feature extraction, deep learning feature extraction, and network feature extraction.

[0105] 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, and waveform energy, reflecting the instantaneous change characteristics of the signal (such as abnormal torque fluctuations and sudden pressure increases).

[0106] Frequency-domain feature extraction extracts the spectral energy distribution and main frequency components of underwater acoustic fingerprint data through fast Fourier transform (FFT) or wavelet transform to capture the underwater environmental acoustic fingerprint characteristics (such as water flow impact sounds and mechanical vibration noises).

[0107] Deep learning feature extraction uses convolutional neural network (CNN) or recurrent neural network (RNN) to automatically learn features from underwater acoustic fingerprint data and extract high-order abstract features (such as periodic or mutation features in the acoustic fingerprint pattern).

[0108] Complex network features are used to construct a multi-sensor data association network, and the interaction between environmental factors (such as the correlation between torque and pressure changes) is analyzed through indicators such as node degree and clustering coefficient.

[0109] The timestamp tags include data acquisition synchronization and feature serialization.

[0110] Furthermore, data acquisition synchronization adds millisecond-level timestamps to the original data through a hardware clock synchronization protocol during the sensor data acquisition stage to ensure the time alignment of multi-source data.

[0111] Feature serialization binds the extracted feature vectors with timestamps to generate a time-tagged time series feature matrix for subsequent frame-by-frame analysis by the shift unit.

[0112] The acoustic fingerprint classification model is used to process the acoustic fingerprints obtained by the acoustic fingerprint recognition unit, obtain the spectral features of the acoustic fingerprints through the acoustic fingerprint classification model, and classify the spectral features, including rainstorm impacts, mechanical vibrations, and human operations;

[0113] The acoustic fingerprint classification model includes data preprocessing and feature extraction, classification model architecture, and classification decision fusion. A specific implementation method includes:

[0114] Adaptive filtering uses the NLMS (Normalized Least Mean Square) algorithm, with the reference signal being the environmental noise floor (acquired through silent segments), to cancel steady-state noise (such as water flow sounds) in real time.

[0115] Wavelet threshold denoising is used to perform 5-layer Db4 wavelet decomposition on the voiceprint signal. Soft threshold processing is applied to the detail coefficients (threshold = σ√(2logN), where σ is the noise standard deviation). The characteristics of bubble rupture at 500 - 800 Hz are retained. The frame length is 25 ms (corresponding to the underwater acoustic wave propagation characteristics), the frame shift is 10 ms, and the Hamming window is used to suppress spectral leakage. For voiceprints of mechanical vibration type (zero-crossing rate < 20 times / frame), dynamic frame length adjustment (15 - 40 ms) is enabled to avoid harmonic truncation. Through a 40-channel Mel filter bank, the linear frequency (0 - 8 kHz) is mapped to the Mel scale (0 - 1125 mel), and the target frequency band of 80 - 500 Hz is emphasized. The logarithm of the spectral energy (log(1 + E)) is taken to enhance the visibility of weak high-frequency signals. For mechanical vibration type data, the harmonic product spectrum (HPS) characteristics are superimposed to capture the harmonic structure of the engine.

[0116] Furthermore, the specific implementation method of the classification model architecture includes:

[0117] The input layer is a 64×64 Mel-Spectrogram (time-frequency diagram) with 3 channels (original spectrum, reference spectrum, real-time spectrum).

[0118] The convolution module is used to perform convolution operations on the input voiceprint frequency-domain time-frequency diagram and output the operation result. Through the analysis of the operation result, the feature channels are squeezed to enhance the weight of the high-frequency components of the voiceprint.

[0119] The classification model architecture outputs the voiceprint features affected by weather and the voiceprint features affected by human operations to judge the underwater sound in multiple environments.

[0120] Obtain the torque change rate, water pressure gradient change rate, and voiceprint spectrum features within a continuous time period;

[0121] Through the torque change rate within the continuous time stamp, the torque shift unit receives the torque change rate within the continuous time stamp and adjusts the torque parameter shift step. If the torque data change rate per unit time is greater than the highest weight, the torque parameter shift step is shortened. If the torque data change rate per unit time does not exceed the highest weight, it enters the state to be adjusted;

[0122] Adjust the step shift length through the water pressure change acceleration within the continuous time stamp. If the water pressure change acceleration from time stamp t - 1 to t rises more than the water pressure change acceleration of the continuous time stamp from t - 2 to t - 1, and the acceleration rise is greater than the minimum threshold, the step ratio is compressed;

[0123] The underwater acoustic signal is obtained through the underwater acoustic voiceprint recognition unit, and through comparison with the acoustic feature library, the step adjustment is activated. If the acoustic signals captured in the frequency band through the underwater acoustic voiceprint spectrum are bubble rupture and vortex, it is judged as immersion, and the voiceprint shift step of the underwater acoustic voiceprint shift unit is adjusted.

[0124] Example Two

[0125] The multi-modal data fusion analysis and judgment method based on environmental perception driving further includes a dynamic optimization unit for real-time adjusting the pushing step size of the pushing unit, which receives the analysis results of multi-modal data of environmental perception and judges whether the closed space and the power space need to be separated.

[0126] By setting step boundary limits for receiving the adjusted voiceprint pushing step size, water pressure compression step size ratio, and adjusted torque pushing step size, and setting intensity levels, the intensity levels include low disturbance and medium-high disturbance;

[0127] By adjusting the step boundary, triggering the redundant boundary buffer layer to make a temporary tolerance for step adjustment. If the step briefly exceeds the rated boundary, the total number of actually adjusted step sizes is not greater than the maximum step size of the redundant boundary buffer layer. If the residence time of the buffer layer exceeds the threshold, immediately switch to the minimum safety step size.

[0128] If the torque change rate, water pressure gradient change rate, and underwater acoustic spectrum characteristics after the adjustment of the pushing unit are within the stable domain of the redundant boundary buffer layer, continue to adjust the step size of the pushing unit according to real-time data;

[0129] If the total compensation quantity of the step size adjusted by the pushing unit is not greater than the maximum step size of the redundant boundary buffer layer and the residence time of the buffer layer is greater than the threshold, trigger step freezing.

[0130] A specific implementation method of a dynamic optimization unit includes:

[0131] Multi-modal data perception and feature mapping capture environmental parameters in real time through three types of sensors.

[0132] Furthermore, monitor the mechanical load change of the power system of the equipment through torque data to identify abnormal fluctuations of sudden increase or sudden drop.

[0133] Perceive the gradient change of water level pressure through water pressure data to capture instantaneous water pressure impacts (such as pipeline rupture or flood peak).

[0134] Analyze underwater acoustic signals through voiceprint data to identify immersion characteristics such as bubble rupture and vortex cavitation.

[0135] These three types of data respectively correspond to mechanical force, hydrodynamics, and acoustic wave propagation characteristics in the physical space. When an abnormality occurs in a certain parameter (such as sudden increase in torque, drastic fluctuation in water pressure, or specific voiceprint characteristics), the system will initially judge the environmental disturbance level (low / medium-high disturbance) and trigger the step adjustment mechanism.

[0136] The logic of step dynamic adjustment includes low disturbance scenarios and medium-high disturbance scenarios:

[0137] 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.

[0138] 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.

[0139] 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).

[0140] A specific implementation method of a redundant boundary buffer layer includes:

[0141] 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).

[0142] 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.

[0143] 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.

[0144] Embodiment 3

[0145] Methods for separating enclosed space from power space include:

[0146] The enclosed space can be separated from the power system. After separation, the enclosed space can float on the water surface. The enclosed space is a space device that floats independently on the water surface after being separated from the power system. The door of the enclosed space is equipped with a sealing strip and is in a sealed state after the door is closed. The control module of the enclosed space (steering wheel, controller connection line, etc.) has a controlled separable mechanism outside the cabin. There is a sealing ring at the connection position between the control module and the enclosed space. The steering column has a separable mechanism that can be controlled to separate from the external steering column. The ventilation opening of the enclosed space has a sealed partition door. The enclosed space is controlled to be airtight after separation. A skylight is provided in the enclosed space for entry and exit when necessary. An inflatable airbag is installed at the bottom outside the enclosed space. Water level sensors are installed on both outer sides of the enclosed space. A horizontal sensor is installed inside the enclosed space. The enclosed space has a battery that can be used after separation. The enclosed space has a control main board.

[0147] Furthermore, the power system consists of components such as a vehicle frame, suspension, machine, motor, battery, transmission system, and wheels. The vehicle frame connects the entire enclosed space and the power system into an integrated whole and supports the mass of the entire equipment. A positioning connection seat is provided at the load-bearing connection position of the power system. 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 spring pin or tightened. After being fixed, they are connected into an integrated equipment. After the spring pin is controlled to be pulled out or unscrewed, the cabin positioning bolt can freely disengage 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. The connections between the enclosed space and each connector and the outlet of the lead wire are all sealed.

[0148] The software of the control main board determines to control the separation of the connection mechanism after detecting the water level information and controls the air-conditioning air outlet to close.

[0149] When the software of the control main board generates an instruction to control the disconnection of the electrical docking parts between the enclosed space and the power system, it controls the disconnection of the lines inside the enclosed space from the plugs, cuts off the interference source at the plug water entry point, and enables the power supply and the control main board inside the separated enclosed space to work normally.

[0150] After detecting that the enclosed space has entered the water, the software of the control main board generates an instruction to control the inflatable 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 strip-shaped long cylinders after inflation. There is a hard protective layer at the bottom. The inflation method is controlled by the scene requirements, including slow inflation with an air pump or fast inflation with gas generated by a chemical reaction. If gas is generated by a chemical reaction, the proportion of the reaction substances is adjusted. And the chemical gas generation device is installed outside the cabin and is waterproofed by resin sealing. The trigger wire required for the chemical reaction inflation device is sealed through the cabin wall. The fast inflation time is greater than 1 s. If an air pump is used for inflation, the air pump is installed inside the enclosed space, and the pipeline leads to the airbag. The pipeline interface is sealed with a sealing ring.

[0151] The software of the control mainboard receives the changing information of the cabin level, judges the tilting direction according to the algorithm, and generates instructions based on the judgment status to control the operation of the seat moving motor, moving the weight upward in the longitudinal and transverse directions of the tilt to maintain a nearly horizontal state.

[0152] After the above instructions are executed, the software of the control mainboard detects and calculates that when a large amount of water will not pour into the separated enclosed space level state, it opens the skylight.

[0153] Small micro thrusters can be added outside the enclosed space and are controlled from inside the cabin.

[0154] The above automatic separation control program can also be switched to manual operation.

[0155] The specific operation is selected according to actual needs.

[0156] Embodiment 4

[0157] After the multi-modal data fusion analysis and judgment based on environmental perception drive, the methods for the enclosed space and the power space not to be separated include:

[0158] The enclosed space and the power space can be used as an integral device that can float on the water surface. The door of the enclosed space is equipped with a sealing strip and is in a sealed state after the door is closed. There are sealing rings at the connection positions where the control module of the enclosed space (steering wheel, controller connection line, etc.) is installed with the enclosed space. The ventilation opening of the enclosed space has a sealed partition door, which is controlled to be airtight after entering the water. The enclosed space is provided with a skylight for entry and exit when necessary. An inflatable airbag is installed at the bottom of the vehicle frame. Water level sensors are installed on both outer sides of the enclosed space. A horizontal sensor is installed inside the enclosed space. There is a storage battery in the enclosed space that can be used after entering the water. The enclosed space has a control mainboard.

[0159] Furthermore, the power system consists of components such as the vehicle frame, suspension, machine, motor, battery, transmission system, and wheels. The vehicle frame connects the entire enclosed space and the power system into an integral whole, becoming an integral device without a separation mechanism. The enclosed space and all connectors and lead-out wire installation outlets are sealed.

[0160] The software of the control mainboard determines to control the air-conditioning air outlet to close after judging the water entry according to the water level information.

[0161] When the software of the control mainboard instructs the electrical isolation between the enclosed space and the power system, it controls the disconnection of the lines inside the enclosed space from the external electrical connection, cuts off the interference source at the water entry, so that the power supply and the control mainboard inside the separated enclosed space can work normally.

[0162] After detecting that the enclosed space is flooded, the control mainboard software generates an instruction to control the inflatable device to inflate the inflatable airbags installed on both sides of the bottom of the vehicle frame. Before inflation, the inflatable airbags are flat strips, and after inflation, they become multiple strip-shaped long cylinders with a hard protective layer at the bottom. They are quickly inflated by the gas generated by a chemical reaction. By adjusting the proportion of the reaction substances, and the chemical gas generation device is installed outside the cabin. It is waterproofed by means of resin sealing. The trigger wires required for the chemical reaction inflator are sealed at the cabin wall, and the quick inflation time is greater than 1 s.

[0163] The control mainboard software receives the changing cabin level information, judges the tilting direction according to the tilting algorithm, and generates an instruction based on the judgment status to control the seat movement motor to work, moving the weight upward in the longitudinal and transverse directions of the tilt to maintain an approximately horizontal state.

[0164] Specifically, a specific implementation method of a tilting algorithm includes:

[0165] The control mainboard software continuously obtains the current level status information of the cabin. The level status information is used to reflect the tilting conditions of the cabin in the transverse (left - right direction) and longitudinal (front - back direction). For example, the level status information includes: the left side of the current cabin is higher than the right side, or the front is lower than the back, etc., similar descriptions of the tilting status information.

[0166] Judging the tilting direction through the tilting algorithm includes judging the transverse tilting direction and the longitudinal tilting direction.

[0167] Specifically, the judgment of the transverse tilting direction includes: when the received cabin level information shows that the height on the left side of the cabin is higher than the right side, it is judged that the cabin is tilted to the left transversely; conversely, if the height on the right side is higher than the left side, it is judged that the cabin is tilted to the right transversely. If the heights on both the left and right sides are equal, it is considered that there is no tilt transversely.

[0168] When the height in the front of the cabin is higher than the back, it is judged that the cabin is tilted forward longitudinally; if the height in the back is higher than the front, it is judged that the cabin is tilted backward longitudinally. If the heights in the front and back are the same, there is no tilt longitudinally.

[0169] If it is judged that the cabin is tilted to the left, an instruction is generated to let the seat movement motor move the weight of the seat to the right, that is, upward in the transverse tilt direction; if it is judged that the cabin is tilted to the right, the motor moves the seat weight to the left. If there is no tilt transversely, no movement instruction is sent.

[0170] If it is judged that the cabin is tilted forward, the seat movement motor sends an instruction to move the weight of the seat backward, that is, upward in the longitudinal tilt direction; if it is judged that the cabin is tilted backward, the motor moves the seat weight forward. When there is no tilt longitudinally, no movement instruction is sent.

[0171] The control mainboard software continuously receives new information about the cabin level, repeats the above steps of tilt direction judgment and motor control instruction generation. As the cabin level state changes continuously, the seat moving motor can continuously make corresponding actions, so that the cabin can be maintained as close to the horizontal state as possible.

[0172] After the above instructions are executed, the control mainboard software detects and calculates that when a large amount of water will not flood into the closed space level state, it opens the skylight.

[0173] A thruster can be added outside the closed space and controlled from inside the cabin.

[0174] The control program can also be switched to manual operation.

[0175] The specific operation is selected according to actual needs.

[0176] For specific applications in automobiles, such as Figure 5 shown, Figure 5 For the vehicle occupant cabin (closed space), it includes: skylight 1, steering column 2, air-conditioning ventilation pipe 3, accelerator 4, brake 5, cabin positioning bolt 6.

[0177] Among them, it should be noted that Figure 5 the cabin positioning bolt 6 in Figure 6 is the same unit as the positioning bolt described in Figure 5 and has the same function. However, Figure 6 the cabin positioning bolt 6 in

[0178] is applied to automobiles, while the positioning bolt described in

[0179] is the application principle.

[0180] Furthermore, the separable steering column is composed of a universal joint and the upper part of the steering column connected to the split lower part;

[0181] As Figure 7 shown, among which, the universal joint and the upper part of the steering column include a universal joint 8, an upper separable shaft 9, and an inter-axis groove 10.

[0182] The split lower part includes a protruding body 11, a connecting sleeve 12, a lower part of the steering column 13, and a lower end connected to the steering mechanism 14.

[0183] Among them, the universal joint and the upper part of the steering column are installed outside the passenger compartment, connecting the steering wheel assembly. The lower end of the upper separable shaft 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 protruding body 11 and a connecting sleeve 12, which are connected to the lower end of the upper separable shaft 9. The separable lower part of the steering column is installed at the steering mechanism 14 of the driving system.

[0184] Furthermore, the protruding body 11 and the inter-axis groove 10 are connected to form a complete separable steering column. If the passenger compartment floats up, the upper part of the universal joint and the steering column will separate from the lower part, causing the passenger compartment to separate from the driving system.

[0185] By setting the universal joint 8, it is used to adjust the angle of the separable steering column to the vertical state.

[0186] It transforms the passive protection of traditional vehicles into a dynamic risk avoidance mechanism. Through the closed-loop control of the water level sensor, control main board, and separation actuator, it creates an independent living space in the event of a waterlogging accident. Compared with the traditional way of breaking the window to escape, it greatly improves the survival rate and has a wide range of applications.

[0187] Furthermore, by adding buoyancy through a hidden airbag, driving a counterweight slide rail (maximum adjustment torque 500 N·m) and a micro thruster (bidirectional vector nozzle) by a horizontal sensor (±0.1° accuracy), it can achieve precise obstacle avoidance within a radius of 5 m, and provides a large buoyancy to ensure the smooth separation of the passenger compartment and the driving system, and has a large load-bearing capacity.

[0188] The battery in the passenger compartment is connected in parallel with the main battery of the driving system. The signal lines and control lines of other driving systems are all electrically connected to the passenger compartment and the driving system through integrated plugs and sockets, forming a separable system.

[0189] The control main board software determines to control the separation of the connection mechanism after detecting the water level, and controls the air-conditioning air outlet to close.

[0190] When the control main board software generates an instruction to control the disconnection of the electrical plug and socket between the passenger compartment and the driving system, it controls the disconnection of the in-cabin circuit from the plug, cutting off the interference source at the plug entry into the water, so that the power supply and control main board in the separated passenger compartment can work normally.

[0191] Furthermore, modules such as the brake 5, the accelerator 4, and the air-conditioning ventilation pipe 3 that are electrically connected to the vehicle driving system are all provided with sealing rings or controlled sealing doors to prevent water from entering the passenger compartment after separation.

[0192] The controlled sealing door is used for sealing the air-conditioning ventilation opening.

[0193] As Figure 6 shown, it is a connection diagram of the passenger compartment and the vehicle frame, which includes: the passenger compartment floor 15, the retaining pin 17, the positioning seat 18, and the vehicle frame plate 19. A positioning connection seat is provided at the load-bearing connection position of the vehicle frame structure of the vehicle driving system, and a positioning bolt is provided at the corresponding position of the passenger compartment. After the positioning bolt is inserted into the positioning seat 18, it is fixed by the retaining pin 17, and they are connected into a whole after being fixed. Furthermore, the passenger compartment floor 15 and the retaining pin groove 16 form a lower positioning bolt, where the retaining pin groove 16 is used to embed the retaining pin 17, and the retaining pin 17 is controlled to be electrically pulled out.

[0194] The vehicle frame plate 19 is used to support the vehicle frame, and the positioning seat 18 is used to fix the passenger compartment and the vehicle frame.

[0195] It is important to note that the construction and arrangement of the present application shown in multiple different exemplary embodiments are merely illustrative. Although only multiple embodiments are described in detail in this disclosure, those who refer to this disclosure should easily understand that many modifications are possible without substantially departing from the novel teachings and advantages of the subject matter described in this application. For example, changes in the dimensions, scales, structures, shapes and proportions of various elements, as well as parameter values (such as temperature, pressure, etc.), installation arrangements, use of materials, colors, orientations, etc. For example, an element shown as integrally formed may be composed of multiple parts or elements, the position of the element may be inverted or otherwise changed, and the nature, number or position of discrete elements may be changed or altered. Therefore, all such modifications are intended to be included within the scope of the present invention. The order or sequence of any process or method steps may be changed or reordered according to alternative embodiments. Any "means-plus-function" clause is intended to cover the structure that performs the function described herein, and not only structural equivalents but also equivalent structures. Other substitutions, modifications, changes and omissions may be made in the design, operating conditions and arrangement of the exemplary embodiments without departing from the scope of the present invention. Therefore, the present invention is not limited to a specific embodiment, but extends to various modifications that still fall within the scope of the appended claims.

[0196] In addition, in order to provide a concise description of the exemplary embodiments, not all features of the actual embodiments may be described (i.e., those features that are not relevant to the currently considered best mode of implementing the present invention, or those features that are not relevant to the implementation of the present invention).

[0197] It should be understood that, during the development of any actual implementation, such as in any engineering or design project, a large number of specific implementation decisions can be made. Such development efforts may be complex and time-consuming, but for those of ordinary skill in the art who benefit from this disclosure, without undue experimentation, the development efforts will be routine work of design, fabrication, and production.

[0198] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within 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