Driver state remote monitoring collaborative alarm platform and method based on Internet of Things

Abstract

The invention discloses a driver state remote monitoring collaborative alarm platform and method based on the Internet of Things, and relates to the technical field of the Internet of Things, and the method comprises the steps: carrying out the channel information entropy analysis of collected state data, adaptively cutting off a low-information-amount channel, carrying out the nonlinear feature mapping to obtain a mapping feature sequence, constructing a space-time coupling information tensor, and carrying out the self-adaptive cutting of a low-information-amount channel; the method comprises the following steps: calculating a multi-mode cooperative surge factor, executing tensor decomposition to obtain a decomposition factor, reconstructing an estimation tensor and calculating a reconstruction error, calculating a multi-mode entropy surge factor, generating an abnormal score, counting and generating a dynamic threshold according to the abnormal score in a sliding window, and judging a real-time score to generate an alarm trigger signal. Through adaptive dimensionality reduction, hybrid mapping, dynamic modulation CP decomposition and adaptive reconstruction error evaluation based on channel information entropy, real-time efficient processing and deep multi-modal fusion of multi-source data are realized, noise and redundancy are effectively eliminated, abnormal features are amplified, space-time coordination is accurately described, and dynamic adaptive anomaly judgment is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and specifically to a driver status remote monitoring and collaborative alarm platform and method based on the Internet of Things. Background Art

[0002] With the rapid development of Internet of Things (IoT) technology and in-vehicle smart terminals, driver status monitoring systems based on multi-source sensors have become an important means of ensuring vehicle safety. Existing technologies primarily rely on in-vehicle cameras, physiological signal acquisition modules, and vehicle dynamic sensors to analyze a single or limited set of data, such as the driver's facial expressions, eyelid closure, heart rate, skin charge, vehicle speed, and acceleration, to determine fatigue, distraction, or abnormal driving behavior. Typical solutions often use threshold comparisons or simple machine learning models to match single-modal features with preset thresholds to trigger local warnings. However, in actual road conditions and complex vehicle environments, such methods are easily affected by factors such as lighting changes, individual differences, and sensor noise, resulting in frequent false alarms and missed alarms, making it difficult to meet the safety requirements of high precision and high reliability.

[0003] To improve monitoring accuracy, some technologies incorporate multimodal data fusion, combining multiple sources of information, such as visual, physiological, and vehicle status, at the feature or decision level. However, most rely on fixed weighting coefficients or simple weighted summation, lacking a dynamic assessment of the amount of information in each channel and the time-varying correlations between the data. Furthermore, existing fusion methods are inefficient for processing high-dimensional data and cannot meet the requirements of real-time analysis. They also often use static thresholds for anomaly detection, ignoring the dynamic changes in driver status under varying road conditions, driving habits, and environmental conditions, making it difficult to achieve remote collaborative alarms and adaptive risk warnings.

[0004] Furthermore, large-scale, multi-source time-series data, after being transmitted to the cloud or edge servers, often faces bandwidth and computing resource bottlenecks, making it difficult to achieve both real-time and robustness. The lack of effective filtering of redundancy and noise in massive channel data leads to poor results in subsequent feature mapping and multimodal fusion. To address these issues, a collaborative driver status remote monitoring and alarm method is urgently needed that can adaptively remove low-information channels, dynamically measure multimodal collaborative features, and achieve accurate alarms based on reconstruction errors and adaptive thresholds. This method can improve the system's detection accuracy, real-time response capabilities, and remote collaborative processing efficiency. Summary of the Invention

[0005] Based on the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a driver status remote monitoring and collaborative alarm platform and method based on the Internet of Things to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a driver status remote monitoring and coordinated alarm method based on the Internet of Things, comprising:

[0007] S1: Collect state data through multi-source sensors, perform channel information entropy analysis on the collected state data, and adaptively truncate low-information channels;

[0008] S2: Perform nonlinear feature mapping on the truncated channel data to obtain the mapping feature sequence of the remaining channels;

[0009] S3: Construct the spatiotemporal coupling information tensor based on the mapping feature sequence and calculate the multi-mode co-surge factor;

[0010] S4: Perform tensor decomposition on the spatiotemporal coupling information tensor to obtain decomposition factors, reconstruct the estimated tensor according to the decomposition factors, and calculate the reconstruction error;

[0011] S5: Calculate the multi-mode entropy surge factor based on the reconstruction error to generate an anomaly score. Generate a dynamic threshold based on the anomaly score statistics within the sliding window, and judge the real-time score to generate an alarm trigger signal.

[0012] The present invention is further configured such that the multi-source sensors include an in-vehicle high-definition visible light camera, an infrared or depth camera, an on-board inertial measurement unit, and a temperature, humidity, and gas concentration sensor; and the status data include a color image sequence, eyelid closure, blinking frequency, line of sight deflection angle, head posture, expression feature vector, depth map, relative displacement and angle change of the trunk and limbs, three-axis acceleration, three-axis angular velocity, quaternion or Euler angle of the vehicle and driver's head posture, air temperature, relative humidity, carbon dioxide concentration, carbon monoxide concentration, volatile organic compound concentration, and alcohol concentration.

[0013] The present invention is further configured such that step S1 comprises:

[0014] Synchronize the state data collected by multiple source sensors according to a unified timestamp to form multi-channel time series data;

[0015] Obtain the value range of each channel's time series data, divide the value range into multiple continuous sub-intervals, and calculate the channel information weight of each sub-interval;

[0016] Calculate the channel information entropy according to the channel information weight, and identify the low-information channels whose information content is lower than the threshold based on the comparison between the channel information entropy and the preset entropy threshold;

[0017] Truncation processing is performed on channels with low information content, and the state data of the subset of channels with information content above a threshold is retained.

[0018] The present invention is further configured to record the multi-channel time series data as is the status data of the qth channel of the i-th sensor, is a set of real numbers, T is the number of time steps, Q i is the total number of sensor channels of the i-th sensor;

[0019] The value range of each channel time series data is recorded as [min(d i,q ),max(d i,q )], the continuous subinterval is recorded as B is the number of subintervals;

[0020] The calculation logic of channel information weight is: h q,m is the channel information value of the subinterval, h q,m =|{t|d i,q (t)∈I m}|, t is the time step index, |·| is the cardinality of the set;

[0021] The calculation logic of channel information entropy is: H i,q is the channel information entropy of the qth channel of the i-th sensor, η i is the entropy order control parameter.

[0022] The present invention is further configured such that step S2 includes:

[0023] Calculate the fractal norm and window entropy aggregation based on the truncated channel data;

[0024] The truncated channel data is subjected to nonlinear feature mapping according to the fractal norm and window entropy aggregation to obtain the mapping feature sequence of the remaining channels.

[0025] The present invention is further configured such that the calculation logic of the fractal norm is: d i,q is the time series state data vector of the qth channel after the truncation of the i-th sensor within the sampling period, recorded as L i is the total number of sampling points, ∥d i,q ∥ Ξi is the fractal norm of the qth channel after truncation of the i-th sensor, p and r are indices, α i is the scale attenuation index;

[0026] The calculation logic of window entropy aggregation is: SH(d i,q ;η i ) is the window entropy aggregation of the qth channel after the truncation of the i-th sensor, η i is the entropy exponential order, K is the total number of windows, S is the window step size, is the channel information entropy on the k-th window, β i is the entropy concentration index;

[0027] The logic of mapping feature sequences is: f i,q (t) is the mapping feature sequence of the qth channel after truncation of the i-th sensor, ξ i is the fractal enhancement index, τ i is the entropy modulation coefficient, κ i is the global nonlinear index.

[0028] The present invention is further configured such that step S3 comprises: mapping the feature sequence f i,q (t) Construct the feature matrix of the i-th sensor T is the number of time steps, Q' i is the total number of channels after truncation;

[0029] Defining the spatiotemporal coupling information tensor Elements in a tensor is the qth sensor after truncation of the i-th sensor i Channel index;

[0030] The calculation logic of the multi-mode surge factor is: Δ t is the multimode interference factor, MI ij (t) is the cross mutual information, C k (t) is the adaptive curvature coefficient.

[0031] The present invention is further configured such that step S4 includes: fitting the spatiotemporal coupling information tensor X in a CP decomposition form with a tensor decomposition rank of R, and the objective function is: L is the objective function, i M is the mode index, corresponding to the channel index of the Mth sensor, R is the tensor decomposition rank, is the i-th spatial factor of the r-th mode j Quantity, v r (t) is the value of the rth time factor at the tth moment, Θ(Δ t ) is the dynamic modulation function, ψ is the decomposition error norm index, μ is the regularization weight, σ j is the j-th mode regularization strength parameter, is the factor vector under the j-th mode Regularization function of ;

[0032] Alternating Optimization Factor Matrix and V=[v1,…,v R ],fixed With {v r},untie U is a temporary placeholder, and the solution result is assigned to U (j) , X (j)is the j-th mode expansion of the spatiotemporal coupling information tensor X, ⊙ is the Khatri–Rao product, ○ k≠j U (k) is the matrix obtained by Khatri–Rao product of all spatial factor matrices except the jth mode, is the Frobenius norm square, R(U;σ j ) is the regularization function of U under the j-th mode; fix all untie X (T) is the time pattern expansion, V⊙Θ(Δ) is the modulation result after multiplying the time factor matrix by the modulation matrix element by element, W is a temporary placeholder, and the solution result is assigned to V⊙Θ(Δ), X (T) is the matrix after the spatiotemporal coupling information tensor X is expanded along the time pattern, To perform Khatri–Rao product on the factor matrices of all spatial modes column by column;

[0033] According to the solved U (j) and V, the reconstructed estimate tensor Estimated tensor for reconstruction;

[0034] The calculation logic of the reconstruction error is: E is the reconstruction error, and ψ2 is the power of the reconstruction error norm.

[0035] The present invention is further configured such that step S5 includes: the calculation logic of the multi-mode entropy surge factor is: IS(E t κ e )=-κ e E t ln(E t +ε0), IS(E t κ e ) is the multimode entropy surge factor, E t is the reconstruction error at time t, κ e is the entropy surge attenuation coefficient, ε0 is a constant;

[0036] The calculation logic of the anomaly score is: S t is the abnormal score at time t, ψ3 is the score sensitivity parameter, τ e is the scoring nonlinear decay index;

[0037] Define the window width W and calculate W anomaly scores within the window {S t-W+1 ,…,S t}, aggregate to generate dynamic thresholds: Λ t is the dynamic threshold at time t, and γ is the threshold aggregation index.

[0038] The present invention also provides a driver status remote monitoring and coordinated alarm platform based on the Internet of Things, which is used to implement the above-mentioned driver status remote monitoring and coordinated alarm method based on the Internet of Things, including:

[0039] Acquisition module: collects status data through multi-source sensors, performs channel information entropy analysis on the collected status data, and adaptively cuts off low-information channels;

[0040] Mapping module: Perform nonlinear feature mapping on the truncated channel data to obtain the mapping feature sequence of the remaining channels;

[0041] Computational module: constructs the spatiotemporal coupling information tensor based on the mapping feature sequence and calculates the multi-mode co-surge factor;

[0042] Reconstruction module: performs tensor decomposition on the spatiotemporal coupling information tensor to obtain decomposition factors, reconstructs the estimated tensor based on the decomposition factors, and calculates the reconstruction error;

[0043] Alarm module: Calculates the multi-mode entropy surge factor based on the reconstruction error, generates anomaly scores, generates dynamic thresholds based on anomaly score statistics within the sliding window, and determines the real-time scores to generate alarm trigger signals.

[0044] The present invention provides a driver status remote monitoring and collaborative alarm platform and method based on the Internet of Things. The method collects status data through multi-source sensors, performs channel information entropy analysis on the collected status data, and adaptively truncates low-information channels; performs nonlinear feature mapping on the truncated channel data to obtain a mapping feature sequence of the remaining channels; constructs a spatiotemporal coupling information tensor based on the mapping feature sequence, and simultaneously calculates a multi-mode collaborative surge factor; performs tensor decomposition on the spatiotemporal coupling information tensor to obtain decomposition factors, reconstructs an estimation tensor based on the decomposition factors, and calculates a reconstruction error; calculates a multi-mode entropy surge factor based on the reconstruction error, generates an anomaly score, generates a dynamic threshold based on the anomaly score statistics within a sliding window, and determines the real-time score to generate an alarm trigger signal. The beneficial effects produced include:

[0045] 1. Adaptive channel dimensionality reduction: Through adaptive channel screening based on channel information entropy, low-information channels are accurately eliminated, significantly reducing data redundancy and computational complexity, improving real-time processing capabilities on the edge, and fundamentally reducing reliance on noise and invalid features, achieving a significant reduction in input dimensions while ensuring the integrity of key information.

[0046] 2. Fractal norm and window entropy aggregation hybrid mapping: By performing a weighted fusion mapping of the fractal norm and window entropy aggregation, the multi-scale self-similar structure and local uncertainty of the signal are simultaneously extracted. This method has higher sensitivity to small mutations and hidden abnormal states, effectively amplifies abnormal components, and enables subsequent multimodal fusion and tensor decomposition to more accurately capture abnormal features.

[0047] 3. Spatiotemporal Coupling Information Tensor and Dynamic Co-surge Modeling: By constructing a spatiotemporal coupling information tensor that integrates multi-source mapping features and introducing a multi-mode co-surge factor, the coupling relationship between sensor data in the spatiotemporal dimension is uniformly represented. This allows for in-depth separation and dynamic characterization of normal and abnormal modes, demonstrating the coordinated evolution of various modes at different time steps and providing a solid data foundation for accurate identification of abnormal modes.

[0048] 4. Dynamically modulated CP decomposition and adaptive reconstruction error: By introducing a dynamic modulation function based on the multi-mode co-surge factor into the CP tensor decomposition, the time factor and the co-surge quantity are adaptively coupled element by element, achieving the beneficial effects of a higher degree of fit of the decomposition model to time-varying characteristics and a stronger amplification ability for abnormal errors. Combined with the nonlinear error norm and factor regularization, it is possible to significantly highlight the abnormal reconstruction residuals while retaining the reconstruction accuracy of the normal mode.

[0049] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without inventive efforts. In the drawings:

[0051] Figure 1 This is a flowchart of a driver status remote monitoring and coordinated alarm method based on the Internet of Things, shown as an exemplary embodiment of the present invention;

[0052] Figure 2 The figure is a schematic structural diagram of a driver status remote monitoring and collaborative alarm platform based on the Internet of Things, showing an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following describes the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art will readily appreciate the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the various details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are intended only to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0054] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0055] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0056] Example 1

[0057] Driver status remote monitoring and collaborative alarm method based on the Internet of Things, such as Figure 1 Shown, including:

[0058] S1: Collect state data through multi-source sensors, perform channel information entropy analysis on the collected state data, and adaptively truncate low-information channels;

[0059] S2: Perform nonlinear feature mapping on the truncated channel data to obtain the mapping feature sequence of the remaining channels;

[0060] S3: Construct the spatiotemporal coupling information tensor based on the mapping feature sequence and calculate the multi-mode co-surge factor;

[0061] S4: Perform tensor decomposition on the spatiotemporal coupling information tensor to obtain decomposition factors, reconstruct the estimated tensor according to the decomposition factors, and calculate the reconstruction error;

[0062] S5: Calculate the multi-mode entropy surge factor based on the reconstruction error to generate an anomaly score. Generate a dynamic threshold based on the anomaly score statistics within the sliding window, and judge the real-time score to generate an alarm trigger signal.

[0063] The present invention is further configured such that the multi-source sensors include an in-vehicle high-definition visible light camera, an infrared or depth camera, an on-board inertial measurement unit, and a temperature, humidity, and gas concentration sensor; and the status data include a color image sequence, eyelid closure, blinking frequency, line of sight deflection angle, head posture, expression feature vector, depth map, relative displacement and angle change of the trunk and limbs, three-axis acceleration, three-axis angular velocity, quaternion or Euler angle of the vehicle and driver's head posture, air temperature, relative humidity, carbon dioxide concentration, carbon monoxide concentration, volatile organic compound concentration, and alcohol concentration. Specifically, in the above parameters, the color image sequence is a continuous frame of RGB images, which is used to extract visual information such as face, gestures, and ambient light. It is collected at a fixed frame rate by the high-definition visible light camera in the car; each frame is stored in the timing buffer after dedistortion, white balance and color correction, providing the original input for the subsequent visual algorithm; eyelid closure is a normalized measure of the degree of eyelid opening and closing in a single frame, reflecting fatigue or sleepiness; face detection is performed on the color image, 68 facial key points are located, and the ratio of the distance between the upper and lower eyelids to the distance between the canthus is calculated (Eye Aspect Blink frequency is the number of times the eyes complete a closing-opening cycle per unit time, which is used to assess fatigue and distraction. Based on the eyelid closure curve of consecutive frames, the event count of the EAR dropping from above the closure threshold to below the closure threshold and then rising again is detected. Frequency = number of events in the window length / time length. The gaze deflection angle is the horizontal / vertical offset angle of the gaze direction relative to the center line of the front of the vehicle, which is used to determine whether attention has strayed from the road. The eyeball center and iris boundary are inferred from the key points of the face, and the gaze vector is calculated based on the calibration model. The angle between the gaze vector and the "front" unit vector in the vehicle camera coordinate system is the deflection angle.The head posture includes the pitch, yaw, and roll angles of the driver's head, reflecting behaviors such as turning or lowering the head. 2D facial key points are detected on color / infrared images, and the SolvePnP algorithm is combined with the 3D facial calibration model to solve the 3D head posture Euler angle. The expression feature vector describes the current facial expression state through a high-dimensional vector, including anger, surprise, and calmness, which is used for emotional and cognitive state judgment. The intermediate layer output of the deep learning model (such as VGG-Face / CNN) is extracted within the ROI (facial region), or a 16-64 dimensional feature vector is produced based on the facial Action Units count. The depth map is the distance matrix from each pixel corresponding to the color image to the camera, in meters, highlighting the 3D structure of the scene, through infrared or TOF (Time of Flight). The 3D Flight depth camera directly outputs calibrated depth frames. Each frame undergoes hole filling and filtering to produce a complete depth map. The relative displacement and angle changes of the torso and limbs represent the relative movement distances and joint angles of key skeletal points such as the shoulder blades, elbows, wrists, and knees in three-dimensional space, reflecting the driver's movement amplitude. A human pose estimation algorithm (such as OpenPose and MediaPipe) is used on the depth map / RGBD image to obtain 3D skeleton key points, and the Euclidean distance difference and vector angle between adjacent joint points are calculated. The three-axis acceleration is the instantaneous linear acceleration of the vehicle or head in the three orthogonal directions of X, Y, and Z, measured in m / s. 2, the accelerometer in the vehicle-mounted inertial measurement unit (IMU) is sampled at a high frequency, and the calibration value after gravity component separation and low-pass filtering is output; the three-axis angular velocity is the instantaneous angular velocity (yaw / pitch / roll rate) of the vehicle-mounted IMU on the X, Y, and Z axes, in rad / s, which is directly sampled and output by the IMU gyroscope sensor, combined with zero-bias calibration and complementary filtering to filter out random noise and fuse with the accelerometer; the attitude quaternion or Euler angle is the three-dimensional orientation representation of the vehicle body and the driver's head obtained based on the IMU, which can be in the form of quaternion or Euler angle, and the accelerometer, gyroscope (and magnetometer) data are fused through filtering algorithms such as Madgwick or Mahony to calculate the attitude quaternion of the body and head in real time, and then converted Converting to Euler angles facilitates analysis; air temperature is the ambient temperature inside the car, in degrees Celsius (℃), reflecting the possible impact of the thermal environment in the cockpit on the driving state; relative humidity is the percentage of water vapor content in the air inside the car relative to the saturated water vapor content (%RH), which affects human comfort and sensor performance; carbon dioxide concentration is the CO2 concentration in the air inside the car, in ppm, used to evaluate the relationship between ventilation conditions and the driver's mental state; carbon monoxide concentration is the concentration of harmful gas CO in the car, in ppm, used to detect exhaust backflow or incomplete combustion leakage, and prevent poisoning hazards; volatile organic compound (VOC) concentration is an indicator of the total amount of VOCs such as air, fuel vapor, formaldehyde, etc., in ppm, used to judge the air quality inside the car; alcohol concentration is the ethanol concentration in the car environment, in mg / m 3 In the logic of obtaining the above parameters, the sensor data is obtained through multi-source sensors and uploaded to the cloud, where the sensor data is processed to obtain the above parameters.

[0064] The present invention is further configured such that step S1 comprises:

[0065] The state data collected by the multi-source sensors are synchronized according to a unified timestamp to form multi-channel time series data; the present invention is further configured to record the multi-channel time series data as is the status data of the qth channel of the i-th sensor, is a set of real numbers, T is the number of time steps, Q i is the total number of sensor channels of the i-th sensor;

[0066] Obtain the range of each channel’s time series data, divide the range into multiple continuous subintervals, and calculate the channel information weight of each subinterval; record the range of each channel’s time series data as [min(d i,q ),max(d i,q )], the continuous subinterval is recorded as B is the number of subintervals; the calculation logic of channel information weight is: h q,mis the channel information value of the subinterval, h q,m =|{t|d i,q (t)∈I m}|, t is the time step index, |·| is the cardinality of the set; specifically, for the time series data of each channel, first determine the range interval consisting of its minimum and maximum values, then divide the interval into B continuous sub-intervals of equal width, and then count the number of samples h falling in each sub-interval q,m , and normalize it to get the weight p q,m , which depicts the "information contribution" of the channel in each amplitude interval. This weight distribution based on binning and normalization provides a discrete probability perspective for subsequent entropy calculation and channel screening;

[0067] The channel information entropy is calculated based on the channel information weight. Based on the comparison between the channel information entropy and the preset entropy threshold, the low-information channels with information content below the threshold are identified. The calculation logic of the channel information entropy is: H i,q is the channel information entropy of the qth channel of the i-th sensor, η i is the entropy control parameter; specifically, the discrete probability weight p of each channel q,m Apply Renyi entropy measurement, first according to the entropy order parameter η i Take the power of each weight and sum it, then transform it by logarithm and normalization factor to get the channel information entropy H i,q , and finally compare it with the preset entropy threshold to identify channels with insufficient information and eliminate them; the entropy step control parameter η i To determine the emphasis of Renyi entropy on the high probability and low probability parts in the weight distribution, the value range is [0.5, 2], η i When >1, more attention is paid to high-probability events, i When <1, the impact of low-probability events is amplified; through the nonlinear measurement of channel weight distribution by Renyi entropy, differentiated evaluation of long-tail and concentrated distribution is achieved, effectively eliminating low-information channels and improving the overall algorithm robustness and response speed.

[0068] Truncation processing is performed on channels with low information content, and the state data of the subset of channels with information content above a threshold is retained.

[0069] The present invention is further configured such that step S2 includes:

[0070] The fractal norm and window entropy aggregation are calculated based on the truncated channel data. The present invention is further configured such that the calculation logic of the fractal norm is: d i,q is the time series state data vector of the qth channel after the truncation of the i-th sensor within the sampling period, recorded as Li is the total number of sampling points, ∥d i,q ∥ Ξi is the fractal norm of the qth channel after truncation of the i-th sensor, p and r are indices, α i is the scale attenuation exponent; the calculation logic of window entropy aggregation is: SH(d i,q ;η i ) is the window entropy aggregation of the qth channel after the truncation of the i-th sensor, η i is the entropy exponential order, K is the total number of windows, S is the window step size, is the channel information entropy on the k-th window, β i is the entropy aggregation index; specifically, the fractal norm is used to quantitatively characterize the self-similarity and local amplitude fluctuations of time series data at different sample intervals. First, all sampling points of the same channel are regarded as sequences, and the square of the amplitude difference is calculated for each pair of different sampling moments; at the same time, a power-law attenuation weight is applied according to the size of the index difference between the two moments - the larger the index difference, the smaller the weight - to highlight the influence of neighboring point pairs and suppress remote point pairs. Then, all weighted difference squares are accumulated and the square root is taken to obtain the fractal norm of the channel; the fractal norm can take into account both global trends and local mutations under the same metric, automatically amplify abnormal fluctuations, and provide information rich in self-similar structures for subsequent feature mapping; the scale attenuation exponent α i The larger the value, the faster the attenuation of distant point pairs, thus focusing more on local structures. The value range is [0.5,3]; Window entropy aggregation is used to measure the information uncertainty of the signal in a short period of time and its prominence in the whole. First, the sequence is divided into several continuous subsequences according to a fixed window length and step size, and the channel information entropy is calculated on each subsequence to quantify the uncertainty of the data distribution in the segment. Subsequently, all window entropy values are aggregated with high powers: that is, the entropy value of each window is first exponentiated, and then the results are accumulated, and finally the accumulated results are restored to the corresponding root to ensure that the overall scale is comparable. The high-power aggregation method amplifies the contribution of a few high-entropy windows to the overall entropy value, while the influence of the stationary window is suppressed. It can keenly capture occasional high-uncertainty segments, suppress the influence of noise and long-term stationary states on the overall entropy, and make the mapping features more discerning of hidden anomalies; Entropy exponentiation order η i Used to calculate the Renyi entropy within the window, usually in the range [0.5, 2], the entropy concentration index β i Used to balance the response to extreme entropy values, the value range is [2,5].

[0071] According to the fractal norm and window entropy aggregation, nonlinear feature mapping is performed on the truncated channel data to obtain the mapping feature sequence of the remaining channels; the logic of the mapping feature sequence is: f i,q (t) is the mapped feature sequence of the q-th channel after truncation of the i-th sensor, ξ i is the fractal enhancement exponent, τ i is the entropy modulation coefficient, κ i is the global non-linear exponent; specifically, the core features are extracted from the data of each truncated channel through two main lines simultaneously: on the one hand, the overall multi-scale self-similar fluctuations of the signal are captured by the fractal norm, and on the other hand, the local uncertainty of the signal is aggregated and measured by the window entropy; subsequently, the coefficient τ i is used to weight and fuse the two, and a global power-law non-linear mapping is applied to the fused result, so as to generate a mapped feature sequence that takes into account both the global structure and local mutations at each time step t; the fractal enhancement exponent ξ i ranges from [1, 3]. When the fractal enhancement exponent is larger, it is more sensitive to the multi-scale structure and can be adjusted according to the signal smoothness; the entropy modulation coefficient τ i ranges from [0.1, 10]. The entropy modulation coefficient determines the degree of emphasis on local uncertainty. A low value emphasizes the surge pattern scale, and a high value emphasizes entropy aggregation; the global non-linear exponent κ i ranges from [1, 5]. The global non-linear exponent controls the mapping curvature, and the larger the value, the more obvious the amplification of anomalies.

[0072] The present invention is further configured such that step S3 includes: constructing a feature matrix of the i-th sensor according to the mapped feature sequence f i,q (t), where T is the number of time steps and Q' is the total number of truncated channels; i

[0073] Define the spatio-temporal coupling information tensor The elements within the tensor are the q-th i channel index after truncation of the i-th sensor;

[0074] The calculation logic of the multi-mode co-surge factor is: Δ t is the multi-mode co-surge factor, MI ij (t) is the cross mutual information, C k (t) is the adaptive curvature coefficient; specifically, the core of the multi-mode co-surge factor Δ t lies in simultaneously measuring the information interaction intensity between the mapped features of each sensor and their respective local fluctuation amplitudes, combining the two in a proportional form, and then normalizing through logarithmic transformation to obtain a single scalar to reflect the coupling emergence degree between multi-modalities at time t; the calculation logic of the cross mutual information MI ij (t): for each pair of sensors i < j, construct a joint probability distribution by binning the channel amplitudes ​and marginal probability and Marginal probability is the marginal distribution obtained by summing the joint probability along one dimension.

[0075] q i and q j are the channel indices of the i-th and j-th sensors, and is the mapping feature sequence of the qth channel of the i-th and j-th sensors; cross mutual information MI ij (t) is used to capture the nonlinear interaction information between multimodal features, which can reflect how the characteristics of each sensor fluctuate synchronously when there are abnormal or coordinated changes; the adaptive curvature coefficient C k The calculation logic of (t): The feature vectors of all channels of the i-th sensor at the same time

[0076] ω i is the curvature decay exponent; cross mutual information MI ij (t) Provides a quantitative benchmark for the "complexity" or "volatility" of each mode to prevent the mutual information from being amplified when a single mode is abnormal; when any sensor pair experiences abnormal synchronous changes, the mutual information increases significantly, and the multi-mode synergy factor Δ t Then it rose rapidly.

[0077] The present invention is further configured such that step S4 includes: fitting the spatiotemporal coupling information tensor X in a CP decomposition form with a tensor decomposition rank of R, and the objective function is: L is the objective function, i M is the mode index, corresponding to the channel index of the Mth sensor, R is the tensor decomposition rank, is the i-th spatial factor of the r-th mode j Quantity, v r (t) is the value of the rth time factor at the tth moment, Θ(Δ t ) is the dynamic modulation function, ψ is the decomposition error norm index, μ is the regularization weight, σ j is the j-th mode regularization strength parameter, is the factor vector under the j-th mode Regularization function; Specifically, by decomposing the spatiotemporal coupling information tensor X into a weighted sum of several rank-1 modes, and introducing a "dynamic modulation function" on the time factor to reflect the time-varying influence of the multi-mode co-surge factor, the reconstruction error is minimized while taking into account the regularization constraints of each spatial factor, and obtaining a decomposition factor that can both faithfully reconstruct the normal state and amplify the abnormal mode; the value range of the decomposition error norm exponent ψ is [1,3], and the value range of the regularization weight μ is [10 -4 ,10 -1 ]; dynamic modulation function Θ(Δ t ) is in the form of ξ is greater than zero and ψ3 is greater than 1. By balancing the nonlinear error norm and regularization, high-precision reconstruction of the normal mode is achieved and missed detection is reduced.

[0078] Alternating Optimization Factor Matrix and V=[v1,…,v R ],fixed With {v r},untie U is a temporary placeholder, and the solution result is assigned to U (j) , X (j) is the j-th mode expansion of the spatiotemporal coupling information tensor X, ⊙ is the Khatri–Rao product, ○ k≠j U (k) is the matrix obtained by Khatri–Rao product of all spatial factor matrices except the jth mode, is the Frobenius norm square, R(U;σ j ) is the regularization function of U under the j-th mode; fix all untie X (T) is the time pattern expansion, V⊙Θ(Δ) is the modulation result after multiplying the time factor matrix by the modulation matrix element by element, W is a temporary placeholder, and the solution result is assigned to V⊙Θ(Δ), X (T) is the matrix after the spatiotemporal coupling information tensor X is expanded along the time pattern, To perform column-wise Khatri–Rao products on the factor matrices of all spatial modes, an alternating least squares (ALS) strategy is employed to optimize the spatial and temporal factor matrices in steps. In each iteration, the remaining factors and dynamic modulation terms are fixed, solving only for the spatial factor matrix of one mode. Furthermore, all spatial factors are fixed, solving only for the modulated temporal factor matrix. Each step aims to minimize the reconstruction error after tensor expansion plus the regularization term until convergence. By splitting the high-dimensional tensor decomposition problem into spatial and temporal subproblems, the solution is simplified and convergence is accelerated. Both the spatial and temporal factors obtained through alternating optimization can be used for subsequent mode analysis and visualization, facilitating safety decision support.

[0079] According to the solution of U( j ) and V, reconstruct the estimated tensor To reconstruct the estimated tensor; the calculation logic of the reconstruction error is: E is the reconstruction error, ψ2 is the power of the reconstruction error norm; specifically, after completing the spatial factor U( j ) and V are solved, the estimated tensor is obtained by reconstructing the tensor in rank-1 mode, that is, superimposing the outer product of each potential mode in each modal space factor component and the weighted sum of the corresponding time factor. Subsequently, the residuals of the original tensor X and the estimated tensor at all indices are compared, and the residuals are accumulated to the power of ψ2 to obtain the global reconstruction error, which is used to quantify the degree of fit of the decomposition model to the original data. The value of the reconstruction error norm power ψ2 is greater than zero; all residuals in time and space are aggregated into a single scalar to facilitate setting a unified threshold or generating anomaly scores. The reconstruction error reflects the difference in model fit between normal and abnormal modes, providing a direct basis for adaptive threshold determination and model iterative optimization.

[0080] The present invention is further configured such that step S5 includes: the calculation logic of the multi-mode entropy surge factor is: IS(E t κ e )=-κ e E t ln(E t +ε0), IS(E t κ e ) is the multimode entropy surge factor, E t is the reconstruction error at time t, κ e is the entropy surge attenuation coefficient, ε0 is a constant; specifically, by reconstructing the error E t Combined with its logarithmic term, a mapping similar to "local entropy" is constructed: first add a small offset ε0 to the reconstructed residual to prevent logarithmic divergence, and then calculate E t ln(E t +ε0) to measure the coupling of error size and uncertainty, and finally multiply by the attenuation coefficient -κ e The multimode entropy surge factor IS(E t κ e ), which can amplify its influence on the judgment when the error is small and appropriately suppress it when the error is too large, thereby forming a nonlinear mapping that is both sensitive and stable to abnormal reconstruction points; the value range of the constant ε0 is [10 -6 ,10 -3 ], entropy surge attenuation coefficient κ eThe value range of is [0.1, 10]. The error magnitude and the logarithm of the information entropy are unified into one formula, taking into account both amplitude and distribution uncertainty, providing a more discriminative signal for subsequent scoring and threshold determination.

[0081] The calculation logic of the anomaly score is: S t is the abnormal score at time t, ψ3 is the score sensitivity parameter, τ e is the nonlinear decay index of the score; specifically, the abnormal score S t By mapping the magnitude of the reconstruction error coupled with its entropy surge modulation factor to the interval (0,1), the exponential decay function is used to achieve nonlinear quantification of the degree of anomaly: First, the composite quantity E is calculated. t IS(E t κ e ), and then divide it by 1 / τ e The power is scaled and multiplied by the sensitivity coefficient ψ3, and finally the exponential value is taken to ensure that the evaluation score increases with E t IS(E t κ e ) increases monotonically, and a slight error corresponds to S t Near 1, significant abnormality corresponds to S t Close to 0; the value range of the scoring sensitivity parameter ψ3 is [0.1,10], and the scoring nonlinear attenuation exponent τ e The value range is [1,5].

[0082] Define the window width W and calculate W anomaly scores within the window {S t-W+1 ,…,S t}, aggregate to generate dynamic thresholds: Λ t is the dynamic threshold at time t, γ is the threshold aggregation index, specifically, to avoid false positives or false negatives caused by using static thresholds, the anomaly scores {S t-W+1 ,…,S t} as a basis, an adaptive dynamic threshold Λ is calculated through the γ norm (power aggregation) t The threshold is updated in a rolling manner as the score distribution changes, allowing the judgment standard to flexibly increase with normal fluctuations and decrease promptly when the overall score declines, thereby improving the timeliness and reliability of the alarm.

[0083] When the abnormality score S t Less than the dynamic threshold Λ t When an accident occurs, an alarm trigger signal is generated and sent to the monitoring center and the driver's mobile terminal in real time through the Internet of Things wireless network to achieve coordinated alarm.

[0084] Example 2

[0085] See also Figure 2 The exemplary IoT-based driver status remote monitoring and collaborative alarm platform is used to implement the aforementioned IoT-based driver status remote monitoring and collaborative alarm method, including:

[0086] Acquisition module: collects status data through multi-source sensors, performs channel information entropy analysis on the collected status data, and adaptively cuts off low-information channels;

[0087] Mapping module: Perform nonlinear feature mapping on the truncated channel data to obtain the mapping feature sequence of the remaining channels;

[0088] Computational module: constructs the spatiotemporal coupling information tensor based on the mapping feature sequence and calculates the multi-mode co-surge factor;

[0089] Reconstruction module: performs tensor decomposition on the spatiotemporal coupling information tensor to obtain decomposition factors, reconstructs the estimated tensor based on the decomposition factors, and calculates the reconstruction error;

[0090] Alarm module: Calculates the multi-mode entropy surge factor based on the reconstruction error, generates anomaly scores, generates dynamic thresholds based on anomaly score statistics within the sliding window, and determines the real-time scores to generate alarm trigger signals.

[0091] It should be noted that the IoT-based remote monitoring and collaborative alarm platform for driver status provided in the above embodiment and the IoT-based remote monitoring and collaborative alarm method for driver status provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs operations has been described in detail in the method embodiment and will not be repeated here. In actual applications, the IoT-based remote monitoring and collaborative alarm platform for driver status provided in the above embodiment can allocate the above functions to different functional modules as needed, that is, divide the internal structure of the system into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0092] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0093] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0094] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0095] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0096] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0097] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0098] In the several embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0099] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0100] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0101] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0102] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A driver status remote monitoring and coordinated alarm method based on the Internet of Things, characterized in that: include: S1: Collect state data through multi-source sensors, perform channel information entropy analysis on the collected state data, and adaptively truncate low-information channels; S2: Perform nonlinear feature mapping on the truncated channel data to obtain the mapping feature sequence of the remaining channels; S3: Construct the spatiotemporal coupling information tensor based on the mapping feature sequence and calculate the multi-mode co-surge factor; S4: Perform tensor decomposition on the spatiotemporal coupling information tensor to obtain decomposition factors, reconstruct the estimated tensor according to the decomposition factors, and calculate the reconstruction error; S5: Calculate the multi-mode entropy surge factor based on the reconstruction error to generate an anomaly score. Generate a dynamic threshold based on the anomaly score statistics within the sliding window, and judge the real-time score to generate an alarm trigger signal.

2. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 1 is characterized in that: The multi-source sensors include in-vehicle high-definition visible light cameras, infrared or depth cameras, on-board inertial measurement units, and temperature, humidity, and gas concentration sensors. The status data include color image sequences, eyelid closure, blinking frequency, gaze deflection angle, head posture, expression feature vector, depth map, relative displacement and angle change of torso and limbs, three-axis acceleration, three-axis angular velocity, vehicle and driver's head posture quaternion or Euler angle, air temperature, relative humidity, carbon dioxide concentration, carbon monoxide concentration, volatile organic compound concentration, and alcohol concentration.

3. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 1 is characterized in that: Step S1 includes: Synchronize the state data collected by multiple source sensors according to a unified timestamp to form multi-channel time series data; Obtain the value range of each channel's time series data, divide the value range into multiple continuous sub-intervals, and calculate the channel information weight of each sub-interval; Calculate the channel information entropy according to the channel information weight, and identify the low-information channels whose information content is lower than the threshold based on the comparison between the channel information entropy and the preset entropy threshold; Truncation processing is performed on channels with low information content, and the state data of the subset of channels with information content above a threshold is retained.

4. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 3 is characterized in that: Multi-channel time series data is recorded as is the status data of the qth channel of the i-th sensor, is a set of real numbers, T is the number of time steps, Q i is the total number of sensor channels of the i-th sensor; The value range of each channel time series data is recorded as [min(d i,q ),max(d i,q )], the continuous subinterval is recorded as B is the number of subintervals; The calculation logic of channel information weight is: h q,m is the channel information value of the subinterval, h q,m =|{t|d i,q (t)∈I m }|, t is the time step index, |·| is the cardinality of the set; The calculation logic of channel information entropy is: H i,q is the channel information entropy of the qth channel of the i-th sensor, η i is the entropy order control parameter.

5. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 1 is characterized in that: Step S2 includes: Calculate the fractal norm and window entropy aggregation based on the truncated channel data; The truncated channel data is subjected to nonlinear feature mapping according to the fractal norm and window entropy aggregation to obtain the mapping feature sequence of the remaining channels.

6. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 5 is characterized in that: The calculation logic of the fractal norm is: d i,q is the time series state data vector of the qth channel after the truncation of the i-th sensor within the sampling period, recorded as L i is the total number of sampling points, is the fractal norm of the qth channel after truncation of the i-th sensor, p and r are indices, α i is the scale attenuation index; The calculation logic of window entropy aggregation is: SH(d i,q ;η i is the window entropy aggregation of the qth channel after truncation of the i-th sensor, η i is the entropy exponential order, K is the total number of windows, S is the window step size, is the channel information entropy on the k-th window, β i is the entropy concentration index; The logic of mapping feature sequences is: f i,q (t) is the mapping feature sequence of the qth channel after truncation of the i-th sensor, ξ i is the fractal enhancement index, τ i is the entropy modulation coefficient, κ i is the global nonlinear index.

7. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 1 is characterized in that: Step S3 includes: according to the mapping feature sequence f i,q (t) Construct the feature matrix of the i-th sensor T is the number of time steps, Q' i is the total number of channels after truncation; Defining the spatiotemporal coupling information tensor Elements in a tensor q i ∈{1,…,Q' i } is the qth sensor after truncation i Channel index; The calculation logic of the multi-mode surge factor is: Δ t is the multimode interference factor, MI ij (t) is the cross mutual information, C k (t) is the adaptive curvature coefficient.

8. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 7 is characterized in that: Step S4 includes fitting the spatiotemporal coupling information tensor X in a CP decomposition form with a tensor decomposition rank of R, and the objective function is: L is the objective function, i M is the mode index, corresponding to the channel index of the Mth sensor, R is the tensor decomposition rank, is the i-th spatial factor of the r-th mode j Quantity, v r (t) is the value of the rth time factor at the tth moment, Θ(Δ t ) is the dynamic modulation function, ψ is the decomposition error norm index, μ is the regularization weight, σ j is the j-th mode regularization strength parameter, is the factor vector under the j-th mode Regularization function of ; Alternating Optimization Factor Matrix and V=[v1,…,v R ],fixed With {v r },untie U is a temporary placeholder, and the solution result is assigned to U (j) , X (j) is the j-th mode expansion of the spatiotemporal coupling information tensor X, ⊙ is the Khatri–Rao product, ○ k≠j U (k) is the matrix obtained by Khatri–Rao product of all spatial factor matrices except the jth mode, is the Frobenius norm square, R(U;σ j ) is the regularization function of U under the j-th mode; fix all untie X( T is the time pattern expansion, V⊙Θ(Δ) is the modulation result after multiplying the time factor matrix by the modulation matrix element by element, W is a temporary placeholder, and the solution result is assigned to V⊙Θ(Δ), X (T) is the matrix after the spatiotemporal coupling information tensor X is expanded along the time pattern, To perform Khatri–Rao product on the factor matrices of all spatial modes column by column; According to the solved U (j) and V, the reconstructed estimate tensor Estimated tensor for reconstruction; The calculation logic of the reconstruction error is: E is the reconstruction error, and ψ2 is the power of the reconstruction error norm.

9. The driver status remote monitoring and coordinated alarm method based on the Internet of Things according to claim 1 is characterized in that: Step S5 includes: the calculation logic of the multi-mode entropy surge factor is: IS(E t κ e )=-κ e E t ln(E t +ε0), IS(E t κ e ) is the multimode entropy surge factor, E t is the reconstruction error at time t, κ e is the entropy surge attenuation coefficient, ε0 is a constant; The calculation logic of the anomaly score is: S t is the abnormal score at time t, ψ3 is the score sensitivity parameter, τ e is the scoring nonlinear decay index; Define the window width W and calculate W anomaly scores within the window {S t-W+1 ,…,S t }, aggregate to generate dynamic thresholds: Λ t is the dynamic threshold at time t, and γ is the threshold aggregation index.

10. A driver status remote monitoring and coordinated alarm platform based on the Internet of Things, used to implement the driver status remote monitoring and coordinated alarm method based on the Internet of Things according to any one of claims 1 to 9, characterized in that: include: Acquisition module: collects status data through multi-source sensors, performs channel information entropy analysis on the collected status data, and adaptively cuts off low-information channels; Mapping module: Perform nonlinear feature mapping on the truncated channel data to obtain the mapping feature sequence of the remaining channels; Computational module: constructs the spatiotemporal coupling information tensor based on the mapping feature sequence and calculates the multi-mode co-surge factor; Reconstruction module: performs tensor decomposition on the spatiotemporal coupling information tensor to obtain decomposition factors, reconstructs the estimated tensor based on the decomposition factors, and calculates the reconstruction error; Alarm module: Calculates the multi-mode entropy surge factor based on the reconstruction error, generates anomaly scores, generates dynamic thresholds based on anomaly score statistics within the sliding window, and determines the real-time scores to generate alarm trigger signals.

Images