Fatigue monitoring system and method based on flexible pressure array of high-speed train driver seat

Abstract

The present application relates to the field of rail transit technology, and in particular to a fatigue monitoring system and method based on a flexible pressure array of a high-speed rail driver's seat, comprising the following steps: S1. Arrangement of a seat flexible pressure array; S2. Synchronous acquisition of multi-dimensional pressure data; S3. Pressure feature extraction; S4. Driver's sitting posture pressure baseline modeling; S5. Working condition coupling correction; S6. Fatigue grade determination; S7. Graded early warning and cloud linkage; S8. Model self iteration. Through "arranging a flexible pressure array in four areas of a seat cushion, a backrest and both side handrails and extracting multi-dimensional fatigue sensitive features", the present application realizes "all-around, multi-modal perception of driver's sitting posture pressure distribution", and solves the problem of "incomplete fatigue representation and low determination sensitivity of existing technologies due to single feature and missing handrail information".

Description

Technical Field

[0001] This invention relates to the field of rail transit technology, specifically to a fatigue monitoring system and a fatigue monitoring method based on a flexible pressure array for a high-speed rail driver's seat. Background Technology

[0002] The fatigue level of high-speed rail drivers is directly related to driving safety. Current driver fatigue monitoring technologies mainly fall into two categories: First, "wearable physiological parameter monitoring" requires drivers to wear smart bracelets, EEG caps, and other devices close to their bodies, which has the problems of "strong restraint and poor comfort". In addition, physiological electrical signals are easily interfered with in the strong electromagnetic environment of locomotives, resulting in "high misjudgment rate".

[0003] The second is "visual-based facial / eye feature analysis," which captures facial images through cameras. However, it suffers from "recognition failure" due to backlighting or mask obstruction, and the 24 / 7 video surveillance raises concerns about "privacy leaks," resulting in significant obstacles to its widespread adoption.

[0004] In recent years, technologies that utilize seat pressure to monitor fatigue have gradually emerged; for example, existing technologies employ seat cushion and backrest pressure matrix sensors and use accelerometers to subtract vehicle vibration components from the pressure signals; however, this approach still suffers from the following "technical shortcomings": (1) The “general fatigue criterion” is adopted, but “individual differences” caused by different body types and sitting habits are not considered, resulting in a high misjudgment rate; (2) The “linear subtraction method” is used for vehicle body vibration, which cannot handle the “nonlinear coupling effect” between vibration and prolonged sitting time, resulting in distortion of fatigue feature extraction under long-term bumpy conditions; (3) It relies solely on “single features” such as changes in the center of gravity of pressure, without involving “pressure monitoring of the handrail area”, and the information collection is incomplete. Summary of the Invention

[0005] This invention addresses the problems in the prior art by providing a fatigue monitoring system and method based on a flexible pressure array for a high-speed rail driver's seat. The specific technical solution is as follows: On the one hand, the present invention provides a fatigue monitoring method based on a flexible pressure array for a high-speed rail driver's seat, comprising the following steps: S1. Flexible pressure array layout of the seat: Soft pressure sensing pads are built into the seat cushion, backrest and armrest areas of the high-speed rail driver's seat to form a distributed pressure array; the soft pressure sensing pads are made of flexible fabric to fit the curved surface of the seat and have embedded pressure-sensitive sensing units. S2. Multi-dimensional pressure data synchronous acquisition: Real-time acquisition of raw data on seat pressure distribution, backrest support pressure, arm support force, and pressure time-series fluctuations at a sampling frequency of no less than 20Hz; S3. Pressure Feature Extraction: Calculate and extract the pressure center of gravity offset, pressure distribution dispersion, body posture adjustment frequency, static pressure steady-state value, pressure time-series fluctuation coefficient, and the symmetrical pressure ratio and pressure fluctuation spectrum energy of the two handrails from the original pressure field data as fatigue-sensitive features; S4. Driver Postural Pressure Baseline Modeling: Collect pressure distribution samples of drivers in a conscious and normal driving state, and establish a personal postural pressure baseline library for each driver to eliminate individual differences caused by different body types and sitting habits. S5. Working Condition Coupling Correction: The duration of continuous sitting and the vibration intensity of the vehicle body are introduced as correction factors to compensate for the extracted fatigue-sensitive features and eliminate non-fatigue pressure disturbances caused by vehicle body bumps and normal sitting posture adjustments. The compensation adopts a pre-calibrated nonlinear mapping function, which includes at least the square term of vibration intensity, the duration of sitting, and the interaction product term of the two. S6. Fatigue level determination: The real-time fatigue sensitivity features after working condition coupling correction are compared with the baseline in the personal sitting pressure benchmark library for spatial distribution deviation and temporal fluctuation deviation. The fatigue level is output through the pre-built fusion fatigue model. The fatigue level is divided into three levels: Level 1 low risk, Level 2 medium risk, and Level 3 high risk. S7. Tiered early warning and cloud linkage: When fatigue is determined to be level 2, a gentle voice prompt is triggered on the vehicle; when fatigue is determined to be level 3, an audible and visual alarm is triggered and the alarm information is automatically reported to the railway dispatch cloud platform for personnel intervention and intelligent scheduling. S8. Model self-iteration: Continuously collect pressure fatigue sample data from multiple drivers and multiple lines, and optimize the feature thresholds and model parameters of the fused fatigue model online.

[0006] As a further technical solution of the present invention, in step S4, a personal exclusive sitting posture pressure benchmark library is established for each driver to form a one-person-one-model discrimination system.

[0007] As a further technical solution of the present invention, in step S6, the fusion fatigue model uses the Mahalanobis distance algorithm to calculate the quantitative deviation between the real-time feature vector and the individual sitting posture pressure baseline, and uses this deviation as the fatigue index; the fatigue level determination is performed by comparing the fatigue index with a preset first-level threshold. and second-level threshold It is achieved through comparison, and ; Among them, when the fatigue index < When the risk level is determined to be low (Level 1), the system remains silent. when ≤ Fatigue Index < At that time, it was determined to be level two, medium-risk fatigue; When fatigue index ≥ At that time, it was determined to be level three high-risk fatigue; The first-level threshold and second-level threshold Adaptive calibration is performed as follows: the 95th percentile of the Mahalanobis distance set of baseline samples collected during the driver's sober adaptation period is used as the initial first-level threshold. Level 2 threshold by Multiply by the preset multiplier factor Obtained in this way, and Synchronous optimization is performed during the cloud-based self-iteration process. When the baseline sample size is less than 100 frames, the default threshold or global statistical threshold is used instead.

[0008] As a further technical solution of the present invention, the calculation formula for the steady-state value of static pressure in step S3 is as follows: ; in, This is the static steady-state pressure value. This represents variance calculation. Indicates from time arrive Pressure centroid offset within the sliding time window Time series, This represents the length of the sliding time window.

[0009] As a further technical solution of the present invention, it also includes: S9. Fatigue Evolution Trend Prediction: Based on the fatigue index within the current window. Based on historical trends, an autoregressive prediction model is used to calculate predicted fatigue index values ​​for future time points. Fatigue risk warnings or alert levels are then triggered in advance based on these predicted values. When the predicted fatigue index reaches or exceeds [a certain value] in the next 10 minutes... And the current index is lower than When the predicted value exceeds a certain threshold, a "fatigue risk warning" voice prompt is triggered; when the predicted value exceeds a certain threshold... At that time, the alert level was directly upgraded to Level 3.

[0010] On the other hand, the present invention also provides a fatigue monitoring system based on a flexible pressure array for a high-speed rail driver's seat, comprising: The flexible pressure sensing pad module for the seat is used to be installed in the seat cushion, backrest and armrests of the driver's seat to collect distributed pressure signals. The pressure data acquisition module is used to synchronously acquire the pressure distribution of the seat cushion, the pressure of the backrest support, the force of the arm support, and the corresponding pressure time series data. An acceleration detection module is used to acquire vertical and / or lateral vibration signals of the vehicle body in real time. The pressure feature extraction module is used to extract pressure centroid offset, pressure distribution dispersion, body posture adjustment frequency, static pressure steady-state value, and pressure time-series fluctuation coefficient from the raw pressure data. The Personal Sitting Posture Baseline Modeling Module is used to build a personal sitting posture pressure baseline library using pressure samples from drivers in a conscious and normal driving state. The working condition coupling correction module is used to correct the extracted features based on the duration of continuous sitting and the intensity of vehicle vibration. The fatigue fusion judgment module is used to compare the corrected real-time features with the individual baseline in terms of spatial and temporal deviations, and output the first, second or third level of fatigue. The vehicle-mounted graded warning module is used to activate voice prompts when fatigue reaches level two and trigger audible and visual alarms when fatigue reaches level three. The dispatch cloud management module is used to receive three-level fatigue alarm information to enable remote personnel intervention and scheduling assistance; All modules of the system are integrated inside the driver's seat or in the vehicle-mounted equipment directly associated with it, enabling seamless deployment.

[0011] As a further technical solution of the present invention, the personal sitting posture baseline modeling module includes a cluster analysis unit, which is used to collect pressure feature vector samples during the driver's sober adaptation period, remove abnormal action segments, perform cluster analysis, and use the mean vector and covariance matrix of the cluster centers as the driver's personal sitting posture pressure baseline; the fatigue fusion judgment module includes a Mahalanobis distance calculation unit, which is used to calculate the Mahalanobis distance between the real-time corrected feature vector and the mean vector and covariance matrix, and output the fatigue index.

[0012] As a further technical solution of the present invention, the pressure feature extraction module also includes a handrail pressure feature extraction unit, which is used to extract the symmetrical pressure ratio and pressure fluctuation spectrum energy of the two handrails as an auxiliary dimension for fatigue criterion.

[0013] As a further technical solution of the present invention, it also includes an abnormal behavior recognition module, which is used to detect the driver's off-seat state and large-scale violent movements in real time, and to adaptively filter out abnormal segments in the feature extraction and baseline modeling stages.

[0014] As a further technical solution of the present invention, it also includes a local data storage module for storing raw pressure data and fatigue judgment records in real time, and supporting real-time download and viewing of historical data through a data interface.

[0015] The beneficial effects of this invention are as follows: (1) In this invention, by “arranging flexible pressure arrays in four areas of the seat cushion, backrest and double armrests and extracting multi-dimensional fatigue-sensitive features”, “all-round, multi-modal perception of driver’s sitting pressure distribution” is realized, solving the problem that “existing technologies have incomplete fatigue characterization and low judgment sensitivity due to single features and omission of armrest information”; at the same time, by “collecting normal driving pressure samples during the driver’s sober adaptation period and performing cluster analysis to construct a personal exclusive sitting pressure benchmark library”, “a personalized discrimination system with one model for each person” is realized, solving the problem that “general fatigue criteria lead to large cross-individual differences and high misjudgment rate”; and by “introducing continuous sitting time and vehicle vibration intensity and using a pre-calibrated nonlinear mapping function to compensate for fatigue features”, “precise removal of the coupling effect between vehicle vibration and sitting time” is realized, solving the problem that “linear subtraction methods cannot handle nonlinear interference and fatigue feature extraction is distorted under long-term bumpy conditions”.

[0016] (2) In this invention, by “using Mahalanobis distance algorithm to fuse five-dimensional fatigue-sensitive features and adaptively calibrating the grading threshold based on the awake adaptation period sample”, “exponential sensitive response to fatigue state and dynamic threshold adjustment” are achieved, solving the problems of “insufficient sensitivity of single feature judgment and poor adaptability of fixed threshold”.

[0017] (3) In this invention, by “adding a fatigue evolution trend prediction step, predicting the future fatigue index based on the autoregressive model and giving early warnings”, a “technical leap from passive real-time monitoring to proactive early warning” is achieved, solving the problem that “existing technologies can only alarm after fatigue occurs and cannot intervene in advance”.

[0018] (4) By “designing a self-iterative closed-loop architecture for edge-cloud collaboration, the cloud gathers massive amounts of data to optimize global parameters and regularly sends out updates”, the system accuracy has been continuously evolved as the data used grows, solving the problem of “statically fixed model parameters and inability to improve performance over long-term use”. Attached Figure Description

[0019] Figure 1 This is a schematic diagram showing the layout of the flexible pressure sensing pad on the high-speed rail driver's seat in this invention; Figure 2 This is a schematic diagram of the layer structure of a flexible pressure sensing pad. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments.

[0021] See Figure 1 and Figure 2This embodiment provides a fatigue monitoring system based on a flexible pressure array for a high-speed train driver's seat. The hardware consists of four customized flexible pressure sensing pads, respectively laid on the seat cushion, backrest support, and the upper surfaces of the left and right armrests. Each sensing pad uses breathable fabric as its surface and inner lining, with multiple piezoresistive or capacitive flexible pressure sensing units arranged in an array in the middle layer. The spacing between the sensing units is 2-3 cm, distributed in a row-column matrix. The entire sensing pad is no more than 5 mm thick and is fixed between the original foam layer and the breathable cover of the seat by wrapping or embedding. Positioning is aided by Velcro or zippers, ensuring an appearance completely consistent with the original seat. The driver experiences no foreign body sensation and it does not affect normal driving operations.

[0022] Each sensing unit is connected to a signal acquisition and processing module hidden under the seat via a flexible cable. This module includes a multi-channel analog front-end, a high-precision analog-to-digital converter, and a microcontroller. It synchronously scans the entire array of pressure values ​​at a sampling frequency of no less than 20Hz and generates a pressure distribution matrix that reflects the contact state between the human body and the seat in real time. At the same time, it acquires vertical and lateral vibration signals of the vehicle body through the vehicle CAN bus or an independently installed accelerometer.

[0023] Before feature extraction, perform abnormal behavior detection: Real-time monitoring of total pressure. If the weight is below the empty seat threshold (e.g., 3 kg) and the duration exceeds 1 second, it is considered an absence from the seat, and fatigue assessment is paused; the current COP is compared with the individual baseline coordinates. Euclidean distance between If the jump exceeds 150mm within two consecutive frames, it is judged as a violent action. The segment is marked and removed from the subsequent feature statistics to ensure that the proportion of effective data in the feature calculation window is not less than 80%. Otherwise, the window is skipped.

[0024] After data collection, the system performs fatigue monitoring according to the following steps: I. Pressure Feature Extraction; Define the seat cushion plane as a two-dimensional coordinate system, and the coordinates of each sensing unit are... , ,in The sampling time.

[0025] (1) Coordinates of the centroid of pressure (COP): The weighted average position of the pressure field in each frame is calculated, i.e., the centroid of pressure. ; ; Pressure center of gravity offset Defined as the current COP and the individual baseline reference coordinates. Euclidean distance between them: ; Should The higher the value, the further the sitting posture deviates from the normal range.

[0026] (2) Pressure distribution dispersion: Calculate the weighted spatial variance of the pressure value of each sensing unit relative to the pressure centroid: ; The larger the value, the more uneven the pressure distribution between the human body and the seat, reflecting the unstable state of the torso support force.

[0027] (3) Posture adjustment frequency: Set the length of the sliding time window (e.g., 60 seconds) and micro-motion detection threshold Statistics within the window Exceed The number of times is used to obtain the frequency of postural adjustments per unit time. : ; (4) Static pressure steady-state value: defined as the sliding time window The reciprocal of the variance of the time series of internal pressure centroid offset: ; As fatigue deepens, fluctuations in the center of gravity intensify. Increase It decreases accordingly.

[0028] (5) Pressure time series fluctuation coefficient: Calculate the energy proportion of the barycenter coordinate time series within the preset frequency band, or simplify it to the energy proportion within the sliding window. Coefficient of variation of the sequence: ; (6) Handrail auxiliary features; Symmetrical pressure ratio: Let the total pressure on the left armrest be... The total pressure on the right armrest is ,definition When fatigued, the force on the left and right armrests tends to be unbalanced. Monotonically decreasing.

[0029] Handrail pressure fluctuation spectrum energy: The proportion of energy in the 0.2-2Hz frequency band after the handrail pressure time series is subjected to fast Fourier transform. The energy in this frequency band increases when fatigue worsens.

[0030] II. Baseline Modeling of Individual Seated Posture Pressure; During the driver's adjustment period after starting work (usually the first 30-60 minutes of their first shift), the system collects their stress data during normal driving, calculates feature vectors using the method described above, and uses an abnormal behavior recognition module to remove abnormal segments such as leaving their seat or bending over to pick up objects. The collected feature vector set is then processed... Cluster analysis (with 3-5 clusters), taking the mean vector of the largest cluster centers. Covariance Matrix To prevent singularities, a regularization term is added to the diagonal elements of Σ. A personal posture pressure baseline for the driver is constructed; this baseline is stored in the personal benchmark library to achieve "one person, one model".

[0031] III. Operating Condition Coupling Correction The root mean square value of the vertical vibration acceleration of the vehicle body is obtained in real time using accelerometers. and continuous driving time ; Utilizing a nonlinear mapping function calibrated beforehand through experiments Calculate the non-fatigue pressure offset term This function can be obtained by collecting data of different vibration levels and different driving durations under non-fatigue conditions and then performing multinomial regression or neural network training. An example of multinomial regression is as follows: ,coefficient , , , The corrected true fatigue-related pressure centroid offset is obtained by fitting calibration data. ; Similarly, based on the influence of vibration and prolonged sitting on the distribution dispersion, the following applies: By performing appropriate compensation, the corrected feature vector is obtained. .

[0032] IV. Fatigue Level Integration Determination; The corrected real-time feature vector The bias was quantified by comparing the data to an individual baseline using Mahalanobis distance. ; in, The fatigue index is the inverse of the individual baseline covariance matrix. With preset hierarchical threshold In comparison, among which, The threshold is adaptively calibrated based on the Mahalanobis distance set of baseline samples during the driver's sober adaptation period. Take the 95th percentile as the initial first-level threshold. Level 2 threshold ,in Preset magnification factor (initial value) =2.5, and will be further optimized through cloud-based self-iterative optimization.

[0033] when < When the risk level is determined to be low (Level 1), the system remains silent. when ≤ < At that time, it was determined to be level two, medium-risk fatigue; when ≥ At that time, it was determined to be level three high-risk fatigue.

[0034] The judgment logic can be further combined with the frequency of body posture adjustments. The trend of change: If fatigue transitions from moderate to severe, sudden drop Maintaining a high level will increase the severity of fatigue assessment.

[0035] V. Prediction of Fatigue Evolution Trend; System maintenance fatigue index Historical time series (the most recent 10-30 time windows). Construct a first-order autoregressive prediction model: ,in , The fatigue index is obtained by online fitting of the most recent N historical points using the least squares method. It predicts the fatigue index for the next 5 and 10 minutes. If: ≥ And currently < The system generates a "fatigue risk warning" in advance, suggesting that drivers take a short break or adjust their posture when convenient.

[0036] VI. Tiered early warning system and cloud-based linkage; Perform the appropriate action based on the fatigue level: Level 2 fatigue: Play a gentle voice message, such as "You have been driving for a long time, please take a rest," through the built-in speaker in the seat headrest or the driver's dashboard audio system.

[0037] Level 3 Fatigue: Triggers a prominent audible and visual alarm on the instrument panel, and simultaneously generates an alarm message containing the driver's employee number, train number, fatigue level, timestamp, and current characteristic summary. This message is automatically sent to the ground railway dispatch cloud platform via the vehicle-to-ground wireless communication network. Dispatchers can intervene based on the alarm information to arrange shift changes, reduce speed, or provide remote voice reminders, thus achieving human-machine joint prevention.

[0038] VII. Self-iterative edge-cloud collaborative model; A lightweight fusion model (model size <5MB, single inference <50ms) is deployed on the local vehicle terminal to perform real-time fatigue monitoring. Simultaneously, the local data storage module continuously records complete stress characteristic data before and after each warning and during normal shifts, and uploads it to the cloud periodically (e.g., after each shift). The cloud utilizes the accumulated massive amounts of real-world data from multiple drivers and routes to adjust threshold values. , Multiplier Mapping function The parameters, personal baseline update rules, and trend prediction model parameters are optimized in batches. After pruning and quantization, the updated model is remotely distributed to each vehicle terminal, completing the self-iterative closed loop of edge-cloud collaboration.

[0039] 8. Adaptive filtering of abnormal behavior; The system monitors the following abnormal states in real time: Seated Detection: Total Pressure If the weight is below the empty seat threshold (e.g., 3kg) and the duration exceeds 1 second, it is determined that the person has left the seat, the fatigue assessment is paused, and the system automatically enters standby mode.

[0040] Large-scale, intense motion detection: If the jump exceeds the large motion threshold (e.g., 150mm) within two consecutive frames, it is determined to be an action such as bending over to pick up an object, and the segment is removed from the feature statistical analysis.

[0041] Vibration interference suppression: When the vertical vibration acceleration RMS value of the vehicle body exceeds 0.3g, the system automatically increases the threshold value ε for judging the body posture adjustment frequency to avoid road bumps being mistaken for body posture adjustment.

[0042] Through the above-described embodiments, the present invention requires no wearable devices or cameras, and under completely imperceptible and privacy-free conditions, it sensitively captures micro-changes in the sitting posture caused by driver fatigue, achieving high-accuracy and high-reliability fatigue classification early warning, and providing strong technical support for high-speed rail driving safety.

[0043] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A fatigue monitoring method based on a flexible pressure array for a high-speed rail driver's seat, comprising the following steps: S1. Flexible pressure array layout of the seat: Soft pressure sensing pads are built into the seat cushion, backrest and armrest areas of the high-speed rail driver's seat to form a distributed pressure array; the soft pressure sensing pads are made of flexible fabric to fit the curved surface of the seat and have embedded pressure-sensitive sensing units. S2. Multi-dimensional pressure data synchronous acquisition: Real-time acquisition of raw data on seat pressure distribution, backrest support pressure, arm support force, and pressure time-series fluctuations at a sampling frequency of no less than 20Hz; S3. Pressure Feature Extraction: Calculate and extract the pressure center of gravity offset, pressure distribution dispersion, body posture adjustment frequency, static pressure steady-state value, pressure time-series fluctuation coefficient, and the symmetrical pressure ratio and pressure fluctuation spectrum energy of the two handrails from the original pressure field data as fatigue-sensitive features; S4. Driver Postural Pressure Baseline Modeling: Collect pressure distribution samples of drivers in a conscious and normal driving state, and establish a personal postural pressure baseline library for each driver to eliminate individual differences caused by different body types and sitting habits. S5. Working Condition Coupling Correction: The duration of continuous sitting and the vibration intensity of the vehicle body are introduced as correction factors to compensate for the extracted fatigue-sensitive features and eliminate non-fatigue pressure disturbances caused by vehicle body bumps and normal sitting posture adjustments. The compensation adopts a pre-calibrated nonlinear mapping function, which includes at least the square term of vibration intensity, the duration of sitting, and the interaction product term of the two. S6. Fatigue level determination: The real-time fatigue sensitivity features after working condition coupling correction are compared with the baseline in the personal sitting pressure benchmark library for spatial distribution deviation and temporal fluctuation deviation. The fatigue level is output through the pre-built fusion fatigue model. The fatigue level is divided into three levels: Level 1 low risk, Level 2 medium risk, and Level 3 high risk. S7. Tiered early warning and cloud linkage: When fatigue is determined to be level 2, a gentle voice prompt is triggered on the vehicle; when fatigue is determined to be level 3, an audible and visual alarm is triggered and the alarm information is automatically reported to the railway dispatch cloud platform for personnel intervention and intelligent scheduling. S8. Model self-iteration: Continuously collect pressure fatigue sample data from multiple drivers and multiple lines, and optimize the feature thresholds and model parameters of the fused fatigue model online.

2. The fatigue monitoring method based on a flexible pressure array for a high-speed train driver's seat according to claim 1, characterized in that: In step S4, a personal sitting posture pressure benchmark library is established for each driver, forming a unique judgment system for each person.

3. The fatigue monitoring method based on a flexible pressure array for a high-speed train driver's seat according to claim 1, characterized in that: In step S6, the fusion fatigue model uses the Mahalanobis distance algorithm to calculate the quantitative deviation between the real-time feature vector and the individual sitting posture pressure baseline, and uses this deviation as the fatigue index; the fatigue level determination is performed by comparing the fatigue index with a preset first-level threshold. and second-level threshold It is achieved through comparison, and ; Among them, when the fatigue index < When the risk level is determined to be low (Level 1), the system remains silent. when ≤ Fatigue Index < At that time, it was determined to be level two, medium-risk fatigue; When fatigue index ≥ At that time, it was determined to be level three high-risk fatigue; The first-level threshold and second-level threshold Adaptive calibration is performed as follows: the 95th percentile of the Mahalanobis distance set of baseline samples collected during the driver's sober adaptation period is used as the initial first-level threshold. Level 2 threshold by Multiply by the preset multiplier. Obtained in this way, and Synchronous optimization is performed during the cloud-based self-iteration process. When the baseline sample size is less than 100 frames, a default threshold or a global statistical threshold is used instead.

4. The fatigue monitoring method based on a flexible pressure array for a high-speed train driver's seat according to claim 1, characterized in that: The formula for calculating the steady-state value of static pressure in step S3 is as follows: ； in, This is the static steady-state pressure value. This represents variance calculation. Indicates from time arrive Pressure centroid offset within the sliding time window Time series, This represents the length of the sliding time window.

5. The fatigue monitoring method based on a flexible pressure array for a high-speed train driver's seat according to claim 1, characterized in that, Also includes: S9. Fatigue Evolution Trend Prediction: Based on the fatigue index within the current window. Based on historical trends, an autoregressive prediction model is used to calculate predicted fatigue index values ​​for future time points. Fatigue risk warnings or alert levels are then triggered in advance based on these predicted values. When the predicted fatigue index reaches or exceeds [a certain value] in the next 10 minutes... And the current index is lower than When the predicted value exceeds a certain threshold, a "fatigue risk warning" voice prompt is triggered; when the predicted value exceeds a certain threshold... At that time, the alert level was directly upgraded to Level 3.

6. A fatigue monitoring system based on a flexible pressure array for a high-speed train driver's seat, characterized in that: include: The flexible pressure sensing pad module for the seat is used to be installed in the seat cushion, backrest and armrests of the driver's seat to collect distributed pressure signals. The pressure data acquisition module is used to synchronously acquire the pressure distribution of the seat cushion, the pressure of the backrest support, the force of the arm support, and the corresponding pressure time series data. An acceleration detection module is used to acquire vertical and / or lateral vibration signals of the vehicle body in real time. The pressure feature extraction module is used to extract pressure centroid offset, pressure distribution dispersion, body posture adjustment frequency, static pressure steady-state value, and pressure time-series fluctuation coefficient from the raw pressure data. The Personal Sitting Posture Baseline Modeling Module is used to build a personal sitting posture pressure baseline library using pressure samples from drivers in a conscious and normal driving state. The working condition coupling correction module is used to correct the extracted features based on the duration of continuous sitting and the intensity of vehicle vibration. The fatigue fusion judgment module is used to compare the corrected real-time features with the individual baseline in terms of spatial and temporal deviations, and output the first, second or third level of fatigue. The vehicle-mounted graded warning module is used to activate voice prompts when fatigue reaches level two and trigger audible and visual alarms when fatigue reaches level three. The dispatch cloud management module is used to receive three-level fatigue alarm information to enable remote personnel intervention and scheduling assistance; All modules of the system are integrated inside the driver's seat or in the vehicle-mounted equipment directly associated with it, enabling seamless deployment.

7. The fatigue monitoring system based on a flexible pressure array for a high-speed train driver's seat according to claim 6, characterized in that: The personal sitting posture baseline modeling module includes a cluster analysis unit, which is used to collect pressure feature vector samples during the driver's sober adaptation period, remove abnormal motion segments, perform cluster analysis, and use the mean vector and covariance matrix of the cluster centers as the driver's personal sitting posture pressure baseline; the fatigue fusion judgment module includes a Mahalanobis distance calculation unit, which is used to calculate the Mahalanobis distance between the real-time corrected feature vector and the mean vector and covariance matrix, and output the fatigue index.

8. The fatigue monitoring system based on a flexible pressure array for a high-speed train driver's seat according to claim 6, characterized in that: The pressure feature extraction module also includes a handrail pressure feature extraction unit, which is used to extract the symmetrical pressure ratio and pressure fluctuation spectrum energy of the two handrails as an auxiliary dimension for fatigue criterion.

9. The fatigue monitoring system based on a flexible pressure array for a high-speed train driver's seat according to claim 6, characterized in that: It also includes an abnormal behavior recognition module, which is used to detect the driver's off-seat state and large-scale violent movements in real time, and to adaptively filter out abnormal segments during the feature extraction and baseline modeling stages.

10. The fatigue monitoring system based on a flexible pressure array for a high-speed train driver's seat according to claim 6, characterized in that: It also includes a local data storage module for storing raw pressure data and fatigue assessment records in real time, and supports real-time downloading and viewing of historical data through a data interface.

Images