A sightseeing vehicle running state monitoring method based on multi-dimensional perception data
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 湖南省特种设备检验检测研究院
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-09
AI Technical Summary
Existing sightseeing vehicle operation monitoring technologies suffer from problems such as insufficient multi-source data coordination, lack of path-state coupling, failure to avoid time lag effects, disconnect between fault diagnosis and energy efficiency monitoring, and fixed early warning thresholds. These issues lead to monitoring bias, high false alarm rates, increased energy consumption, and increased risk of safety accidents.
By employing multi-dimensional sensing data synchronization and fusion, dynamic weight adjustment and feature selection, combined with the ORC_SOM algorithm to predict path features, and using the adaptive UKF-TDC collaborative estimation model, the warning threshold is dynamically adjusted to achieve fault identification and root cause diagnosis, triggering multi-level warnings.
It improved the accuracy of operational status monitoring, reduced parameter estimation errors and accident risks, optimized energy efficiency management, and enhanced safety control capabilities and operational efficiency.
Smart Images
Figure CN122174079A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of vehicle operation monitoring technology, and in particular to a method for monitoring the operation status of sightseeing vehicles based on multi-dimensional sensing data. Background Technology
[0002] With the increasing number of sightseeing vehicles in scenic areas, the demand for their operational safety and reliability is becoming increasingly prominent. Existing sightseeing vehicle operation monitoring technologies have the following key problems: 1. Insufficient multi-source data collaboration: Traditional monitoring relies on a single sensor (such as GPS or IMU), lacking unified synchronization and dynamic fusion of positioning, braking, environment, vehicle body, and battery data. This can easily lead to monitoring deviations due to data asynchrony (such as sensor acquisition time lag of 0.05s-0.1s). 2. Lack of path-state coupling: The estimated operating state is not dynamically corrected based on the complex path characteristics of the scenic area (such as sharp bends and steep slopes). Only a fixed critical speed threshold is used, which results in delayed warnings or a high false alarm rate (false alarm rate exceeding 20%) in sharp bend / steep slope scenarios. 3. Time lag effect not avoided: There is a time lag between sensor acquisition and model calculation (e.g., a 0.03s time lag in lidar point cloud processing), and existing Kalman filter algorithms do not specifically compensate for this, resulting in estimation errors of more than 10% for core parameters such as driving speed and center of gravity sideslip angle; 4. Disconnect between fault diagnosis and energy efficiency monitoring: Fault classification only covers mechanical faults (such as brake failure) without considering path and environmental factors (such as sideslip caused by low-adhesion road surfaces), and the battery state of health (SOH) estimation does not take into account driving energy efficiency, thus failing to reflect the actual impact of battery degradation on operation; 5. Fixed warning thresholds: The warning thresholds are not dynamically adjusted according to the path complexity, resulting in redundancy of thresholds in straight-path scenarios and insufficient thresholds in complex-path scenarios. Summary of the Invention
[0003] The purpose of this invention is to provide a method for monitoring the operating status of sightseeing vehicles based on multi-dimensional sensing data, thereby solving the aforementioned technical problems.
[0004] To achieve the above objectives, the present invention provides a method for monitoring the operating status of a sightseeing vehicle based on multi-dimensional sensing data, comprising the following steps: S1. Synchronously collect sightseeing vehicle positioning data, braking data, environmental road condition data, vehicle body status data, and battery status data to construct an enhanced original multidimensional perception dataset; among which, the vehicle body status data includes three-axis acceleration, three-axis angular velocity, steering angle data, roll angle data, driving speed, and wheel rotation pulse signals; environmental road condition data includes road surface friction coefficient and ambient temperature; S2. Time synchronization and outlier removal are performed on the enhanced original multidimensional sensing dataset output by S1. The dynamic weight fusion algorithm is used to adjust the fusion weights of path, dynamics, and energy features. After LGBM-Boruta feature filtering, a standardized fusion dataset is output. S3. Based on the standardized fusion dataset output by S2, the ORC_SOM algorithm is used to predict subsequent path features and calculate path complexity coefficients. An adaptive UKF-TDC collaborative estimation model is used to couple kinematic and dynamic dual models to estimate driving speed, center of gravity sideslip angle, rollover-side slip critical speed, energy efficiency parameters and battery health state (SOH), and output the core parameter set of path-state-energy efficiency coupling. S4, based on the coupled core parameter set output by S3, combined with the six-category fault classification system constructed by Ishikawa diagram, dynamically calibrates the warning threshold through the LGBM risk prediction model, and adopts an edge-cloud layered architecture to realize fault identification and root cause diagnosis, triggering multi-level offline voice warnings and generating handling suggestions.
[0005] Therefore, the above-mentioned method for monitoring the operating status of sightseeing vehicles based on multi-dimensional sensing data has the following beneficial effects: 1. Improve the accuracy of operation status monitoring and reduce parameter estimation errors. Redundant features are eliminated by LGBM-Boruta feature screening, and core operation parameters (driving speed, center of gravity sideslip angle, etc.) are retained. TDC time delay compensation is integrated into the UKF iteration process to offset the time delay between sensor acquisition and model calculation (0.01~0.05s). At the same time, the noise covariance of the UKF process is dynamically adjusted based on the path complexity coefficient, so that the core parameter estimation error is reduced by more than 30%, providing high-precision data support for safety early warning. 2. Dynamically adapt to complex paths and accurately calculate safety boundaries. A feature library containing 15 types of paths is pre-built. Through ORC_SOM, three-source matching of lidar, Beidou historical data and feature library is achieved to predict the curvature, slope and complexity coefficient of the path from 0-500m. Combined with path features, the critical speed of sideslip / rollover is corrected. The warning threshold is adjusted 500ms in advance for sharp bends / sharp slopes (reduced by 10-15%), avoiding the problem of poor adaptability of fixed thresholds and significantly reducing the risk of sideslip and rollover accidents. 3. Achieve coordinated management of energy efficiency and battery status to extend operating range. Construct an energy efficiency-battery status coupled estimation model, correct the energy efficiency value calculation based on the path complexity coefficient, and estimate the current battery capacity and SOH through EKF to monitor the battery health status in real time; provide data basis for energy consumption optimization and battery maintenance of sightseeing vehicles, reduce ineffective energy consumption by more than 15%, and extend battery life and operating range per charge; 4. Layered fault diagnosis and multi-level early warning enhance safety management capabilities. Combining Ishikawa diagrams, six major categories of operational anomalies are classified. At the edge, TinyMLCNN quickly identifies faults and triggers offline voice warnings (synchronous sound and light reminders inside and outside the vehicle). In the cloud, BiLSTM-XGBoost locates the root cause of the fault and generates handling suggestions. At the same time, early warning data and operation videos are automatically recorded (local + cloud dual storage), providing a complete chain of evidence for accident analysis and driver training, and reducing the time spent on handling safety accidents. 5. Adapts to scenic area operation scenarios and supports intelligent management decisions. It is compatible with commonly used sensing devices in scenic areas such as GPS / BeiDou dual-mode positioning and LiDAR path perception. The output path-state-energy efficiency coupling parameters can be connected to the scenic area dispatch system. It supports real-time monitoring of vehicle location, driving trajectory and operation status, and automatically counts operational data such as passenger flow and vehicle turnover rate. It provides data-driven decision-making basis for scenic area capacity adjustment and route optimization, and improves fleet management efficiency by more than 20%.
[0006] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0007] Figure 1 This is a flowchart of a sightseeing vehicle operation status monitoring method based on multi-dimensional sensing data, as described in this invention. Detailed Implementation
[0008] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.
[0009] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.
[0010] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0011] like Figure 1As shown, a method for monitoring the operational status of a sightseeing vehicle based on multi-dimensional sensing data includes the following steps: S1. Synchronously collect sightseeing vehicle positioning data, braking data, environmental road condition data, vehicle body status data, and battery status data to construct an enhanced original multidimensional perception dataset; among which, the vehicle body status data includes three-axis acceleration, three-axis angular velocity, steering angle data, roll angle data, driving speed, and wheel rotation pulse signals; environmental road condition data includes road surface friction coefficient and ambient temperature; S2. Time synchronization and outlier removal are performed on the enhanced original multidimensional sensing dataset output by S1. The dynamic weight fusion algorithm is used to adjust the fusion weights of path, dynamics, and energy features. After LGBM-Boruta feature filtering, a standardized fusion dataset is output. S3. Based on the standardized fusion dataset output by S2, the ORC_SOM algorithm is used to predict subsequent path features and calculate path complexity coefficients. An adaptive UKF-TDC collaborative estimation model is used to couple kinematic and dynamic dual models to estimate driving speed, center of gravity sideslip angle, rollover-side slip critical speed, energy efficiency parameters and battery health state (SOH), and output the core parameter set of path-state-energy efficiency coupling. S4, based on the coupled core parameter set output by S3, combined with the six-category fault classification system constructed by Ishikawa diagram, dynamically calibrates the warning threshold through the LGBM risk prediction model, and adopts an edge-cloud layered architecture to realize fault identification and root cause diagnosis, triggering multi-level offline voice warnings and generating handling suggestions.
[0012] Step S2 specifically includes the following steps: S21. Data synchronization and anomaly removal optimization; among which, data synchronization adopts a GPS time synchronization mechanism to uniformly calibrate the timestamps of all sensors to the Beidou system time. Anomaly removal combines the fault classification logic of the Ishikawa diagram, categorizing abnormal data into three types: sensor faults, environmental interference, and operational anomalies, and employing different removal strategies for each. Sensor failure: cross-validation using data from multiple sensors; Environmental interference: Transient interference was eliminated using the moving average method; Operational anomalies: Retain operational anomaly data for subsequent early warning analysis; S22, Multi-dimensional Feature Fusion: Spatial feature fusion: The latitude and longitude of the sightseeing vehicle's positioning, path pre-sensing data, and road surface friction coefficient are fused into a path feature vector. ; Dynamic feature fusion: This involves fusing triaxial acceleration, triaxial angular velocity, steering angle data, roll angle data, vehicle speed, and wheel rotation pulse signals into a dynamic vector. ; Energy Feature Fusion: This involves fusing battery voltage, current, and energy consumption data from battery state data into an energy vector. ; S23. Calculate the optimal radius coefficient: ; In the formula, Indicates the first The optimal radius coefficient of the eigenvector, and These correspond to the path feature vector, dynamics vector, and energy vector, respectively. Represents the current input vector With feature center Euclidean distance, Indicates the first The current input vector of the class feature. Indicates the first Feature center vector of class features; Indicates the adaptive radius threshold, and , This represents the total number of iterations of the ORC_SOM algorithm. This indicates the current iteration number of the ORC_SOM algorithm; Indicates the initial radius threshold; S24, Based on the optimal radius coefficient Dynamic weight allocation: ; in, ; ; ; ; In the formula, This represents the final feature vector after dynamic weight fusion; Indicates the first The fusion weights of feature vectors; Indicates the first The correction coefficients of the eigenvectors; , These represent the road curvature correction factor and the road slope correction factor, respectively. and These represent real-time road curvature and road slope, respectively. This represents the roll angle correction factor; Indicates the vehicle body roll angle; Indicates the centroid sideslip angle; This represents the sideslip angle correction factor; This represents the correction factor for the battery current variation coefficient; Indicates the coefficient of variation of battery current; This indicates the battery health correction factor; This indicates an estimated battery health value, and , , , , , , This indicates the current actual effective remaining capacity of the battery. Indicates the battery's rated capacity. Indicates the basic remaining capacity of the battery. The table shows the temperature correction factor. This represents the current ratio correction factor. This represents the correction factor for the number of iterations. Indicates the real-time temperature of the battery. Indicates the real-time battery current. Indicates the battery's rated discharge current. This represents the current value corresponding to 0.2 times the battery's rated capacity. Represents the cyclic decay fitting coefficients. This indicates the cumulative number of charge-discharge cycles of the battery. Indicates the rated number of charge-discharge cycles of the battery. Indicates battery charge / discharge efficiency. Represents the time variable of integration. Indicates the remaining capacity compensation item. and These represent the start and end times of the integration process, respectively. S25 and SRG-TSCGM modules enhance noise reduction; S251, Feature Input and Validity Verification: Receive the final feature vector after dynamic weight fusion. The integrity of the feature data is verified using the CRC32 checksum algorithm, and invalid data that fails the checksum is removed. At the same time, extreme outliers of single-dimensional features are removed using the 3σ criterion, resulting in an effective fused feature vector. ; S252, Temporal Feature Reconstruction and Dimensional Normalization: Effectively fusing feature vectors Reconstructing the temporal feature matrix, with a temporal window length of 5 consecutive time steps, yields... 3D time series feature matrix Then, Min-Max standardization is used to eliminate the dimensional differences between the feature dimensions, completing the feature dimension normalization process and obtaining the normalized time series feature matrix. ; S253. The normalized time series feature matrix Perceptual feature attention weighting is performed, followed by 1D lightweight temporal convolution to achieve temporal smoothing within the feature dimension and cross-dimensional feature interaction. Finally, batch normalization is applied to eliminate feature distribution offset, resulting in the SRG pre-enhanced feature matrix. ; S254. Level 3 Temporal Context Feature Extraction: Using SRG Pre-enhanced Feature Matrix As input, local, surrounding, and global temporal context features are extracted through 1D convolution, 1D dilated convolution with a dilation rate of 2, and global average pooling, respectively, to obtain a three-level context feature set. ; S255, Dynamic Weighted Fusion and Lightweight Dimensionality Reduction: Calculate the fluctuation coefficient of the current feature data, dynamically allocate the fusion weights of the three-level context features based on the fluctuation coefficient, and after completing the weighted fusion of the three-level features, reduce the dimensionality of the temporal feature matrix to a static feature vector through a 1×1 1D convolution. Then, after 8-bit fixed-point quantization, obtain the enhanced feature vector after denoising and enhancement by the SRG-TSCGM module. ; S26. Use Z-score normalization to enhance feature vectors. Normalization is performed to obtain standardized feature vectors. ; S27. Using the LGBM-Boruta feature selection algorithm, with the core objective of sightseeing vehicle operation status monitoring as the evaluation criterion, calculate the standardized feature vector. The importance scores of each feature dimension are calculated, redundant features with importance scores below a threshold are removed, and the core features that contribute the most to the monitoring of operational status are retained to obtain the final preprocessed feature vector. Among them, the core features include driving speed, center of gravity sideslip angle, yaw rate, effective steering angle, longitudinal acceleration, and lateral acceleration.
[0013] In step S255, the feature vector is enhanced. The calculation formula is as follows: ; in, ; ; ; ; ; ; ; In the formula, This represents the rounding operator; This represents the time-series feature matrix after dimensionality reduction; Indicates the use of 1 Dimensionality reduction is achieved using 1D convolutions with a size of 1, a stride of 1, and no padding. This represents the temporal feature matrix after three-level context fusion; , and These represent the fusion weights for local, surrounding, and global features, respectively. This represents the fluctuation coefficient of the SRG pre-enhancement feature matrix; Represents the SRG pre-enhanced feature matrix Standard deviation; Represents the SRG pre-enhanced feature matrix The mean.
[0014] Step S3 specifically includes the following steps: S31. Path feature prediction based on ORC_SOM; S311. Use LiDAR to scan the road ahead in real time and output raw point cloud data; then use a statistical filtering algorithm to remove outliers, and then use the RANSAC algorithm to fit the ground plane, segment the point cloud of the road area, and extract the point sets of the left and right boundaries of the road. and ; S312. For each frame of point cloud, calculate the midpoint of the corresponding points on the left and right boundaries to form the point set of the road centerline; S313. Fit the equation of the circular arc to three consecutive adjacent points on the centerline, and solve for the center of the circle using the least squares method. and radius ; Simultaneously, a continuous pair of points along the road centerline direction of vehicle travel is selected, and the horizontal distance between the two points is calculated. and vertical height ; S314, Calculate road curvature and road slope : ; ; S315, based on road curvature and road slope The scenic route is divided into flat straight road, flat gentle curve road, flat sharp curve road, up gentle slope straight road, up sharp slope straight road, down gentle slope straight road, down sharp slope straight road, up gentle slope turning into gentle curve, up sharp slope turning into gentle curve, up gentle slope turning into sharp curve, up sharp slope turning into sharp curve, down gentle slope turning into gentle curve, down sharp slope turning into gentle curve, down gentle slope turning into gentle curve, down gentle slope turning into sharp curve, and down sharp slope turning into sharp curve. in, when At that time, it was divided into flat roads and straight sections; when At that time, it was divided into flat roads and gentle curves; when At that time, it was divided into flat roads and sharp curves; when At that time, it was divided into a gentle uphill straight road; when At that time, it was divided into a straight road with a steep uphill slope; when At that time, it was divided into a gentle downhill straight road; when At that time, it was divided into a straight road with a steep downhill slope; when At that time, it is divided into a gentle uphill slope turning into a gentle curve; when At that time, it was divided into a steep uphill turn followed by a gentle curve; when At that time, it was divided into a gentle uphill slope turning into a sharp curve; when At that time, it was divided into a steep uphill section followed by a sharp curve; when At that time, it is divided into a gentle downhill slope turning into a gentle curve; when At that time, it was divided into a steep downhill slope turning into a gentle curve; when At that time, it was divided into a gentle downhill slope turning into a sharp curve; when At that time, it was divided into a steep downhill section followed by a sharp curve; S316. Convert the pre-built scenic area path feature library into a structured feature matrix that can be recognized by the ORC_SOM algorithm. , Indicates the first The median of the road curvature threshold for path-like routes. Indicates the first The median of the slope threshold for similar paths. Indicates the first Classpath type, Indicates the first Baseline value for classpath transition segment length. Indicates the first Class path complexity coefficient; S317. Extract the latitude and longitude sequence of the vehicle's historical travel from the historical data cached by BeiDou, convert it to plane coordinates through Gaussian projection, and calculate the curvature sequence and slope sequence of the historical BeiDou path. ; S318, Current data from lidar Beidou historical path sequence and path feature library median Perform Min-Max standardization and output the standardized data road curvature. and road slope ; S319. Using the path feature library as a template benchmark, the competitive learning process of the ORC_SOM algorithm is integrated to achieve three-source data matching of current lidar data, BeiDou historical data, and path feature library template. S3191. Based on the 15 types of paths in the path feature library, construct an ORC_SOM neural network, with 2 neurons corresponding to each type of path, and set the total number of neurons. Initialize the weight vector for each neuron. , , , Indicates the first The weight vector of each neuron; and They represent the first The standardized curvature and slope weights corresponding to each neuron; Indicates the transpose operation; Indicates the first The path feature library type of each neuron mapping; and These represent the initial weight values being the median of the standardized road curvature and road slope from the path feature library, respectively. S3192, Use the standardized current lidar data as the input feature vector of ORC_SOM. Calculate the input vector Euclidean distance to the weights of each neuron And calculate the adaptive radius threshold by combining the number of iterations. Finally, the optimal radius coefficient for each neuron is output: ; In the formula, Indicates the first During the nth iteration The optimal radius coefficient of each neuron; S3193. Selecting the optimal matching neuron through competitive learning. And the optimal neuron Mapping back to the path feature library yields the most similar path feature library template, and simultaneously matching the best-matching BeiDou historical path segment from the historical BeiDou paths. , This represents the calibrated historical road curvature sequence and road slope sequence. This represents a fragment of BeiDou's historical path. The start and end timestamps; in, ; S3194. Update neuron weights using gradient descent based on the optimal matching result: ; in, ; In the formula, Indicates the updated number The weight vector of each neuron; Indicates the adaptive learning rate; Indicates the initial learning rate; Indicates the current iteration number; Indicates the total number of iterations; S3110. Based on the template constraints of the path feature library, boundary constraints are imposed on the short-term (0-50m) and long-term (51m-500m) prediction results. Among them, short-term prediction: based on the best-matched historical path segments of BeiDou. Using a path feature library template, extract path features within that range and output the short-term predicted road curvature. and road slope And match the corresponding type in the path feature library to ensure that the short-term predicted value does not exceed the threshold range of the corresponding type in the path feature library; Medium- to long-term forecasts: Combining the matching results of historical BeiDou paths, a linear interpolation algorithm is used to expand path features: ; ; In the formula, This represents the rate of curvature change within the historical matching segment from 50m to 500m. This represents the rate of change of slope from 50m to 500m within the historical matching segment; and These represent the standardized road curvature and road slope at 500m in the historical matching segment, respectively. and These represent the standardized road curvature and road slope at 50m in the historical matching segment, respectively; The calculated and The rate of change thresholds of the path feature library, respectively and To make a comparison, if or Then take , Otherwise, retain the calculation. and ; Denormalize the standardized predictions to the actual physical values, and output the curvature for medium to long term: ; ; In the formula, and These represent distances from vehicles within 50m-500m. The actual road curvature and road slope at the location; and These represent the maximum and minimum values of road curvature in the path feature database, respectively. and These represent the maximum and minimum road gradient values in the path feature database, respectively. S3111, For each prediction point within 50m-500m and The path feature library is rematched, and the corresponding transition segment length and path complexity coefficient calculation rules are extracted to output medium- and long-term prediction features in the range of 50m to 500m. S3112, Predicted curvature for 0-500m and Perform boundary correction and smoothing correction; Boundary correction: If the predicted value exceeds the threshold range of the path feature library, it is corrected to the maximum or minimum value of the path feature library. Smoothing correction: The moving average method is used to smooth the predicted sequence and eliminate abrupt changes caused by interpolation; S3113. Based on the complexity coefficient calculation rules for each type of path in the path feature library, combined with the predicted curvature... and Calculate the path complexity coefficient for each prediction point within the range of 0-500m. : ; S3114. Integrate all features of the 0-50m short-term prediction and the 50-500m medium- and long-term prediction to output a set of path features that fully cover the 0-500m ahead of the vehicle. S32. Adaptive UKF-TDC Cooperative Parameter Estimation: Based on the path feature set output from S31, multi-dimensional perception fusion data, and the inherent parameters of the sightseeing vehicle, the time delay compensation mechanism of the time delay control (TDC) is integrated into the full iterative process of the unscented Kalman filter (UKF), and based on the path complexity coefficient... To achieve adaptive adjustment of the path between UKF process noise covariance and TDC compensation gain; S321. Construct a joint state model of sightseeing vehicle dynamics and kinematics that couples path characteristics. Adapt the low-speed driving characteristics of the sightseeing vehicle to a two-wheeled bicycle model, incorporating the path gradient output from S31. coefficient of friction with road surface Construct a nonlinear state equation and a linear observation equation based on multidimensional sensing fusion data; S322. Design an adaptive TDC time delay compensation term. To address the time delay deviation caused by sensor acquisition and model calculation, a 6-dimensional TDC time delay compensation term is designed, based on the path complexity coefficient output in S31. The TDC compensation gain is dynamically adjusted, and the compensation term is integrated into the UKF state prediction stage to offset the impact of time delay on parameter estimation. S323. Adaptive UKF algorithm initialization: Based on the static or low-speed start-up characteristics of the sightseeing vehicle, the initial state values, initial error covariance matrix, and unscented transformation of the UKF algorithm are initialized, and the initial path complexity coefficient output from S31 is used. Initialize the noise covariance matrix of the path adaptation process : ; In the formula, Represents the basic process noise covariance matrix; S324. Based on the state estimate and error covariance matrix from the previous time step, generate the following through UT transformation: A set of Sigma points, covering the probability distribution of the state vector; S325, Sigma point prediction with TDC compensation in one step, substitutes the Sigma point set into the nonlinear state equation of S321 for further prediction, and then incorporates the TDC time delay compensation term designed in S322 to complete the Sigma point correction. S326. Based on the predicted Sigma point set with fused TDC compensation, the mean and covariance weights of the predicted state are calculated by combining the mean weights and covariance weights of the UT transform, and the real-time path complexity coefficient output from S31 is used as the basis. To achieve dynamic updating of the process noise covariance matrix; S327. Based on the observation vector of multi-dimensional sensing fusion, the predicted value is updated, the Kalman gain is calculated, and a TDC observation residual compensation term is designed to offset the residual bias caused by the observation time lag, finally obtaining... The state estimate and error covariance matrix at time t; S328, Yes The state estimates at each time point are smoothed using a moving average to eliminate parameter abrupt changes caused by nonlinear models and sensor noise. The core operating state parameters of the sightseeing vehicle are then extracted from the smoothed state estimates. ; in, These represent the estimated values of driving speed, center of gravity sideslip angle, yaw rate, effective steering angle, longitudinal acceleration, lateral acceleration, error covariance matrix, and trace of the error covariance matrix, respectively. S33. Calculate the dynamic critical velocity; By incorporating the curvature and slope of path prediction, the critical speed for sideslip is corrected: ; In the formula, This represents the critical speed at which the sightseeing vehicle can sideslip under the current path conditions; Indicates the predicted slope; Indicates the predicted road curvature; Indicates the center distance of the main sales point of the sightseeing vehicle; Indicates the wheelbase of the sightseeing vehicle; This represents the sideslip angle correction factor; Indicates the wheelbase of the sightseeing vehicle; Represents gravitational acceleration; Indicates the vehicle body roll angle; Incorporating path complexity coefficient Correcting the critical rollover speed: ; in, ; In the formula, This indicates the critical speed at which the sightseeing vehicle will overturn under the current path conditions; Indicates the wheelbase correction term; This represents the roll angle correction factor; This indicates the dynamic center of gravity height of the sightseeing vehicle; Indicates the height of the base center of gravity; S34. Dynamic threshold adjustment: Based on the path prediction results, the critical speed threshold is adjusted 500ms in advance. The threshold is reduced by 10%-15% on sharp bends or steep slopes, and the threshold is restored to the basic value on straight roads. S35, Energy Efficiency-Battery State Coupling Estimation: ; ; in, ; ; In the formula, This indicates the energy efficiency rating of the sightseeing vehicle; This refers to the mechanical work output by the wheels of the sightseeing vehicle during its operation. This indicates the energy consumption of the sightseeing vehicle during operation; Indicates the battery's health status; This represents the current battery capacity estimated by EKF; Indicates the battery's rated capacity; , , and These represent rolling resistance, air resistance, the component of gravity, and inertial force, respectively. This indicates the real-time battery voltage.
[0015] Step S321 specifically includes the following steps: S3211, Define the state vector Input vector and observation vector ,in, , , , , and They represent At any given moment, the vehicle's speed, sideslip angle, yaw rate, effective steering angle, longitudinal acceleration, and lateral acceleration are measured. and They represent The steering angle and wheel angular velocity after instantaneous integration; , , and They represent The observed velocity, yaw rate, longitudinal acceleration, and lateral acceleration after time-mapping; S3212, Combining the dynamics of sightseeing vehicles with the slope of the path. Road surface friction coefficient Construct a nonlinear state equation that couples path features: ; in, ; ; In the formula, Represents a nonlinear state transition function; express The speed at any given moment; and They represent The longitudinal and lateral accelerations at time t; Represents gravitational acceleration; Indicates the calculation step size; express The centroid sideslip angle at any given moment; express The yaw rate at any given moment; This indicates the distance from the front axle of the sightseeing vehicle to its center of gravity. and These represent the slip angles of the front and rear tires, respectively. Indicates the effective slip angle of the front tire; This indicates the distance from the rear axle of the sightseeing vehicle to its center of gravity. Indicates the effective slip angle of the rear tire; This represents the moment of inertia of the sightseeing vehicle about the z-axis; express Steering angle after moment fusion; This represents the path complexity turning angle correction factor; Indicates the torque of the drive wheels; Indicates the rolling resistance coefficient; Indicates the overall vehicle weight; Indicates the road surface friction correction coefficient; express Effective steering angle at any given time; S3213, Constructing the linear observation equation , This represents the observation matrix.
[0016] Step S322 specifically includes the following steps: S3221. Based on the model-free adaptive compensation principle of TDC, and taking the deviation between the control quantity at the previous moment and the state at the current moment as a basis, design a time delay compensation term for the motion state of the sightseeing car. : ; in, ; In the formula, express Time-based TDC compensation control quantity; express Time-based TDC compensation control quantity; Let represent the pseudo-input matrix, and ; express The rate of change of state at time t, and , Indicates the sampling interval. and They represent Time and The state vector at any given time; Indicates tracking reference items, and , This represents the rate of change of the ideal reference value. Represents the proportional gain matrix; This represents the path adaptive TDC compensation gain matrix, and , Represents the TDC basic compensation gain matrix; Indicates a status reference value; S3222, Add the 6-dimensional TDC time delay compensation term Additive fusion with UKF state predictions: ; In the formula, This represents the state prediction value after incorporating TDC compensation; This represents the UKF original state prediction value.
[0017] Step S4 specifically includes the following steps: S41. Drawing on the causal analysis method of Ishikawa's illustration, abnormal sightseeing bus operations are categorized into 6 main types: Human factors: operational errors, improper maintenance; Machine-related: Braking system failure, steering system failure, battery system failure, sensor failure; Environmental factors: low road surface adhesion, bumpy road surface, extreme temperatures; Route-related offenses: speeding on sharp bends, overloading on steep slopes; Process-related issues: failure to follow designated routes, failure to conduct regular maintenance; Materials-related: Battery aging, brake pad wear; S42. Dynamically adjust the early warning threshold based on risk level and path characteristics: ; in, ; ; ; In the formula, This indicates the dynamically adjusted warning threshold; The base value representing the warning threshold. Pick or ; This indicates the risk level output by LGBM; These represent the skid risk level and the rollover risk level, respectively. S43. Based on the TinyML CNN model, the fault is quickly identified and offline voice warning is triggered. The vehicle terminal uploads the abnormal data and preliminary warning information to the cloud. The cloud uses the BiLSTM-XGBoost hybrid model to locate the root cause of the fault and generate targeted suggestions.
[0018] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for monitoring the operational status of a sightseeing vehicle based on multi-dimensional sensing data, characterized in that: Includes the following steps: S1. Synchronously collect sightseeing vehicle positioning data, braking data, environmental road condition data, vehicle body status data, and battery status data to construct an enhanced original multidimensional perception dataset; among which, the vehicle body status data includes three-axis acceleration, three-axis angular velocity, steering angle data, roll angle data, driving speed, and wheel rotation pulse signals; environmental road condition data includes road surface friction coefficient and ambient temperature; S2. Time synchronization and outlier removal are performed on the enhanced original multidimensional sensing dataset output by S1. The dynamic weight fusion algorithm is used to adjust the fusion weights of path, dynamics, and energy features. After LGBM-Boruta feature filtering, a standardized fusion dataset is output. S3. Based on the standardized fusion dataset output by S2, the ORC_SOM algorithm is used to predict subsequent path features and calculate path complexity coefficients. An adaptive UKF-TDC collaborative estimation model is used to couple kinematic and dynamic dual models to estimate driving speed, center of gravity sideslip angle, rollover-side slip critical speed, energy efficiency parameters and battery health state (SOH), and output the core parameter set of path-state-energy efficiency coupling. S4, based on the coupled core parameter set output by S3, combined with the six-category fault classification system constructed by Ishikawa diagram, dynamically calibrates the warning threshold through the LGBM risk prediction model, and adopts an edge-cloud layered architecture to realize fault identification and root cause diagnosis, triggering multi-level offline voice warnings and generating handling suggestions.
2. The method for monitoring the operating status of a sightseeing vehicle based on multi-dimensional sensing data according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21. Data synchronization and anomaly removal optimization; among which, data synchronization adopts a GPS time synchronization mechanism to uniformly calibrate the timestamps of all sensors to the Beidou system time. Anomaly removal combines the fault classification logic of the Ishikawa diagram, categorizing abnormal data into three types: sensor faults, environmental interference, and operational anomalies, and employing different removal strategies for each. Sensor failure: cross-validation using data from multiple sensors; Environmental interference: Transient interference was eliminated using the moving average method; Operational anomalies: Retain operational anomaly data for subsequent early warning analysis; S22, Multi-dimensional Feature Fusion: Spatial feature fusion: The latitude and longitude of the sightseeing vehicle's positioning, path pre-sensing data, and road surface friction coefficient are fused into a path feature vector. ; Dynamic feature fusion: This involves fusing triaxial acceleration, triaxial angular velocity, steering angle data, roll angle data, vehicle speed, and wheel rotation pulse signals into a dynamic vector. ; Energy Feature Fusion: This involves fusing battery voltage, current, and energy consumption data from battery state data into an energy vector. ; S23. Calculate the optimal radius coefficient: ； In the formula, Indicates the first The optimal radius coefficient of the eigenvector, and These correspond to the path feature vector, dynamics vector, and energy vector, respectively. Represents the current input vector With feature center Euclidean distance, Indicates the first The current input vector of the class feature. Indicates the first Feature center vector of class features; Indicates the adaptive radius threshold, and , This represents the total number of iterations of the ORC_SOM algorithm. This indicates the current iteration number of the ORC_SOM algorithm; Indicates the initial radius threshold; S24, Based on the optimal radius coefficient Dynamic weight allocation: ； in, ； ； ； ； In the formula, This represents the final feature vector after dynamic weight fusion; Indicates the first The fusion weights of feature vectors; Indicates the first The correction coefficients of the eigenvectors; , These represent the road curvature correction factor and the road slope correction factor, respectively. and These represent real-time road curvature and road slope, respectively. This represents the roll angle correction factor; Indicates the vehicle body roll angle; Indicates the centroid sideslip angle; This represents the sideslip angle correction factor; This represents the correction factor for the battery current variation coefficient; Indicates the coefficient of variation of battery current; This indicates the battery health correction factor; This indicates an estimated battery health value, and , , , , , , This indicates the current actual effective remaining capacity of the battery. Indicates the battery's rated capacity. Indicates the basic remaining capacity of the battery. The table shows the temperature correction factor. This represents the current ratio correction factor. This represents the correction factor for the number of iterations. Indicates the real-time temperature of the battery. Indicates the real-time battery current. Indicates the battery's rated discharge current. This represents the current value corresponding to 0.2 times the battery's rated capacity. Represents the cyclic decay fitting coefficients. This indicates the cumulative number of charge-discharge cycles of the battery. Indicates the rated number of charge-discharge cycles of the battery. Indicates battery charge / discharge efficiency. Represents the time variable of integration. Indicates the remaining capacity compensation item. and These represent the start and end times of the integration process, respectively. S25 and SRG-TSCGM modules enhance noise reduction; S251, Feature Input and Validity Verification: Receive the final feature vector after dynamic weight fusion. The integrity of the feature data is verified using the CRC32 checksum algorithm, and invalid data that fails the checksum is removed. At the same time, extreme outliers of single-dimensional features are removed using the 3σ criterion, resulting in an effective fused feature vector. ; S252, Temporal Feature Reconstruction and Dimensional Normalization: Effectively fusing feature vectors Reconstructing the temporal feature matrix, with a temporal window length of 5 consecutive time steps, yields... 3D time series feature matrix Then, Min-Max standardization is used to eliminate the dimensional differences between the feature dimensions, completing the feature dimension normalization process and obtaining the normalized time series feature matrix. ; S253. The normalized time series feature matrix Perceptual feature attention weighting is performed, followed by 1D lightweight temporal convolution to achieve temporal smoothing within the feature dimension and cross-dimensional feature interaction. Finally, batch normalization is applied to eliminate feature distribution offset, resulting in the SRG pre-enhanced feature matrix. ; S254. Level 3 Temporal Context Feature Extraction: Using SRG Pre-enhanced Feature Matrix As input, local, surrounding, and global temporal context features are extracted through 1D convolution, 1D dilated convolution with a dilation rate of 2, and global average pooling, respectively, to obtain a three-level context feature set. ; S255, Dynamic Weighted Fusion and Lightweight Dimensionality Reduction: Calculate the fluctuation coefficient of the current feature data, dynamically allocate the fusion weights of the three-level context features based on the fluctuation coefficient, and after completing the weighted fusion of the three-level features, reduce the dimensionality of the temporal feature matrix to a static feature vector through a 1×1 1D convolution. Then, after 8-bit fixed-point quantization, obtain the enhanced feature vector after denoising and enhancement by the SRG-TSCGM module. ; S26. Use Z-score normalization to enhance feature vectors. Normalization is performed to obtain standardized feature vectors. ; S27. Using the LGBM-Boruta feature selection algorithm, with the core objective of sightseeing vehicle operation status monitoring as the evaluation criterion, calculate the standardized feature vector. The importance scores of each feature dimension are calculated, redundant features with importance scores below a threshold are removed, and the core features that contribute the most to the monitoring of operational status are retained to obtain the final preprocessed feature vector. Among them, the core features include driving speed, center of gravity sideslip angle, yaw rate, effective steering angle, longitudinal acceleration, and lateral acceleration.
3. The method for monitoring the operating status of a sightseeing vehicle based on multi-dimensional sensing data according to claim 2, characterized in that: In step S255, the feature vector is enhanced. The calculation formula is as follows: ； in, ； ； ； ； ； ； ； In the formula, This represents the rounding operator; This represents the time-series feature matrix after dimensionality reduction; Indicates the use of 1 Dimensionality reduction is achieved using 1D convolutions with a size of 1, stride of 1, and no padding. This represents the temporal feature matrix after three-level context fusion; , and These represent the fusion weights for local, surrounding, and global features, respectively. This represents the fluctuation coefficient of the SRG pre-enhancement feature matrix; Represents the SRG pre-enhanced feature matrix Standard deviation; Represents the SRG pre-enhanced feature matrix The mean.
4. The method for monitoring the operating status of a sightseeing vehicle based on multi-dimensional sensing data according to claim 3, characterized in that: Step S3 specifically includes the following steps: S31. Path feature prediction based on ORC_SOM; S311. Use LiDAR to scan the road ahead in real time and output raw point cloud data; then use a statistical filtering algorithm to remove outliers, and then use the RANSAC algorithm to fit the ground plane, segment the point cloud of the road area, and extract the point sets of the left and right boundaries of the road. and ; S312. For each frame of point cloud, calculate the midpoint of the corresponding points on the left and right boundaries to form the point set of the road centerline; S313. Fit the equation of the circular arc to three consecutive adjacent points on the centerline, and solve for the center of the circle using the least squares method. and radius ; Simultaneously, a continuous pair of points along the road centerline direction of vehicle travel is selected, and the horizontal distance between the two points is calculated. and vertical height ; S314, Calculate road curvature and road slope : ； ； S315, based on road curvature and road slope The scenic route is divided into flat straight road, flat gentle curve road, flat sharp curve road, up gentle slope straight road, up sharp slope straight road, down gentle slope straight road, down sharp slope straight road, up gentle slope turning into gentle curve, up sharp slope turning into gentle curve, up gentle slope turning into sharp curve, up sharp slope turning into sharp curve, down gentle slope turning into gentle curve, down sharp slope turning into gentle curve, down gentle slope turning into gentle curve, down gentle slope turning into sharp curve, and down sharp slope turning into sharp curve. in, when At that time, it was divided into flat roads and straight sections; when At that time, it was divided into flat roads and gentle curves; when At that time, it was divided into flat roads and sharp curves; when At that time, it was divided into a gentle uphill straight road; when At that time, it was divided into a straight road with a steep uphill slope; when At that time, it was divided into a gentle downhill straight road; when At that time, it was divided into a straight road with a steep downhill slope; when At that time, it is divided into a gentle uphill slope turning into a gentle curve; when At that time, it was divided into a steep uphill turn followed by a gentle curve; when At that time, it was divided into a gentle uphill slope turning into a sharp curve; when At that time, it was divided into a steep uphill section followed by a sharp curve; when At that time, it is divided into a gentle downhill slope turning into a gentle curve; when At that time, it was divided into a steep downhill slope turning into a gentle curve; when At that time, it was divided into a gentle downhill slope turning into a sharp curve; when At that time, it was divided into a steep downhill section followed by a sharp curve; S316. Convert the pre-built scenic area path feature library into a structured feature matrix that can be recognized by the ORC_SOM algorithm. , Indicates the first The median of the road curvature threshold for path-like routes. Indicates the first The median of the slope threshold for similar paths. Indicates the first Classpath type, Indicates the first Baseline value for classpath transition segment length. Indicates the first Class path complexity coefficient; S317. Extract the latitude and longitude sequence of the vehicle's historical travel from the historical data cached by BeiDou, convert it to plane coordinates through Gaussian projection, and calculate the curvature sequence and slope sequence of the historical BeiDou path. ; S318, Current data from lidar Beidou historical path sequence and path feature library median Perform Min-Max standardization and output the standardized data road curvature. and road slope ; S319. Using the path feature library as a template benchmark, the competitive learning process of the ORC_SOM algorithm is integrated to achieve three-source data matching of current lidar data, BeiDou historical data, and path feature library template. S3191. Based on the 15 types of paths in the path feature library, construct an ORC_SOM neural network, with 2 neurons corresponding to each type of path, and set the total number of neurons. Initialize the weight vector for each neuron. , , , Indicates the first The weight vector of each neuron; and They represent the first The standardized curvature and slope weights corresponding to each neuron; Indicates the transpose operation; Indicates the first The path feature library type of each neuron mapping; and These represent the initial weight values being the median of the standardized road curvature and road slope from the path feature library, respectively. S3192, Use the standardized current lidar data as the input feature vector of ORC_SOM. Calculate the input vector Euclidean distance to the weights of each neuron And calculate the adaptive radius threshold by combining the number of iterations. Finally, the optimal radius coefficient for each neuron is output: ； In the formula, Indicates the first During the nth iteration The optimal radius coefficient of each neuron; S3193. Selecting the optimal matching neuron through competitive learning. And the optimal neuron Mapping back to the path feature library yields the most similar path feature library template, and simultaneously matching the best-matching BeiDou historical path segment from the historical BeiDou paths. , This represents the calibrated historical road curvature sequence and road slope sequence. This represents a fragment of BeiDou's historical path. The start and end timestamps; in, ; S3194. Update neuron weights using gradient descent based on the optimal matching result: ； in, ； In the formula, Indicates the updated number The weight vector of each neuron; Indicates the adaptive learning rate; Indicates the initial learning rate; S3110. Based on the template constraints of the path feature library, boundary constraints are imposed on the short-term (0-50m) and long-term (51m-500m) prediction results. Among them, short-term prediction: based on the best-matched historical path segments of BeiDou. Using a path feature library template, extract path features within that range and output the short-term predicted road curvature. and road slope And match the corresponding type in the path feature library to ensure that the short-term predicted value does not exceed the threshold range of the corresponding type in the path feature library; Medium- to long-term forecasts: Combining the matching results of historical BeiDou paths, a linear interpolation algorithm is used to expand path features: ； ； In the formula, This represents the rate of curvature change within the historical matching segment from 50m to 500m. This represents the rate of change of slope from 50m to 500m within the historical matching segment; and These represent the standardized road curvature and road slope at 500m in the historical matching segment, respectively. and These represent the standardized road curvature and road slope at 50m in the historical matching segment, respectively; The calculated and The rate of change thresholds of the path feature library, respectively and To make a comparison, if or Then take , Otherwise, retain the calculation. and ; Denormalize the standardized predictions to the actual physical values, and output the curvature for medium to long term: ； ； In the formula, and These represent distances from vehicles within 50m-500m. The actual road curvature and road slope at the location; and These represent the maximum and minimum values of road curvature in the path feature database, respectively. and These represent the maximum and minimum road gradient values in the path feature database, respectively. S3111, For each prediction point within 50m-500m of and The path feature library is rematched, and the corresponding transition segment length and path complexity coefficient calculation rules are extracted to output medium- and long-term prediction features in the range of 50m to 500m. S3112, Predicted curvature for 0-500m and Perform boundary correction and smoothing correction; Boundary correction: If the predicted value exceeds the threshold range of the path feature library, it is corrected to the maximum or minimum value of the path feature library. Smoothing correction: The moving average method is used to smooth the predicted sequence and eliminate abrupt changes caused by interpolation; S3113. Based on the complexity coefficient calculation rules for each type of path in the path feature library, combined with the predicted curvature... and Calculate the path complexity coefficient for each prediction point within the range of 0-500m. : ； S3114. Integrate all features of the 0-50m short-term prediction and the 50-500m medium- and long-term prediction to output a set of path features that fully cover the 0-500m ahead of the vehicle. S32. Adaptive UKF-TDC Cooperative Parameter Estimation: Based on the path feature set output from S31, multi-dimensional perception fusion data, and the inherent parameters of the sightseeing vehicle, the time delay compensation mechanism of the time delay control (TDC) is integrated into the full iterative process of the unscented Kalman filter (UKF), and based on the path complexity coefficient... To achieve adaptive adjustment of the path between UKF process noise covariance and TDC compensation gain; S321. Construct a joint state model of sightseeing vehicle dynamics and kinematics that couples path characteristics. Adapt the low-speed driving characteristics of the sightseeing vehicle to a two-wheeled bicycle model, incorporating the path gradient output from S31. coefficient of friction with road surface Construct a nonlinear state equation and a linear observation equation based on multidimensional sensing fusion data; S322. Design an adaptive TDC time delay compensation term. To address the time delay deviation caused by sensor acquisition and model calculation, a 6-dimensional TDC time delay compensation term is designed, based on the path complexity coefficient output in S31. The TDC compensation gain is dynamically adjusted, and the compensation term is integrated into the UKF state prediction stage to offset the impact of time delay on parameter estimation. S323. Adaptive UKF algorithm initialization: Based on the static or low-speed start-up characteristics of the sightseeing vehicle, the initial state values, initial error covariance matrix, and unscented transformation of the UKF algorithm are initialized, and the initial path complexity coefficient output from S31 is used. Initialize the noise covariance matrix of the path adaptation process : ； In the formula, Represents the basic process noise covariance matrix; S324. Based on the state estimate and error covariance matrix from the previous time step, generate the following through UT transformation: A set of Sigma points, covering the probability distribution of the state vector; S325, Sigma point prediction with TDC compensation in one step, substitutes the Sigma point set into the nonlinear state equation of S321 for further prediction, and then incorporates the TDC time delay compensation term designed in S322 to complete the Sigma point correction. S326. Based on the predicted Sigma point set with fused TDC compensation, the mean and covariance weights of the predicted state are calculated by combining the mean weights and covariance weights of the UT transform, and the real-time path complexity coefficient output from S31 is used as the basis. To achieve dynamic updating of the process noise covariance matrix; S327. Based on the observation vector of multi-dimensional sensing fusion, the predicted value is updated, the Kalman gain is calculated, and a TDC observation residual compensation term is designed to offset the residual bias caused by the observation time lag, finally obtaining... The state estimate and error covariance matrix at time t; S328, Yes The state estimates at each time point are smoothed using a moving average to eliminate parameter abrupt changes caused by nonlinear models and sensor noise. The core operating state parameters of the sightseeing vehicle are then extracted from the smoothed state estimates. ； in, These represent the estimated values of driving speed, center of gravity sideslip angle, yaw rate, effective steering angle, longitudinal acceleration, lateral acceleration, error covariance matrix, and trace of the error covariance matrix, respectively. S33. Calculate the dynamic critical velocity; By incorporating the curvature and slope of path prediction, the critical speed for sideslip is corrected: ； In the formula, This represents the critical speed at which the sightseeing vehicle can sideslip under the current path conditions; Indicates the predicted slope; Indicates the predicted road curvature; Indicates the center distance of the main sales point of the sightseeing vehicle; Indicates the wheelbase of the sightseeing vehicle; This represents the sideslip angle correction factor; Indicates the wheelbase of the sightseeing vehicle; Represents gravitational acceleration; Indicates the vehicle body roll angle; Incorporating path complexity coefficient Correcting the critical rollover speed: ； in, ； In the formula, This indicates the critical speed at which the sightseeing vehicle will overturn under the current path conditions; Indicates the wheelbase correction term; This represents the roll angle correction factor; This indicates the dynamic center of gravity height of the sightseeing vehicle; Indicates the height of the base center of gravity; S34. Dynamic threshold adjustment: Based on the path prediction results, the critical speed threshold is adjusted 500ms in advance. The threshold is reduced by 10%-15% on sharp bends or steep slopes, and the threshold is restored to the basic value on straight roads. S35, Energy Efficiency-Battery State Coupling Estimation: ； ； in, ； ； In the formula, This indicates the energy efficiency rating of the sightseeing vehicle; This refers to the mechanical work output by the wheels of the sightseeing vehicle during its operation. This indicates the energy consumption of the sightseeing vehicle during operation; Indicates the battery's health status; This represents the current battery capacity estimated by EKF; Indicates the battery's rated capacity; , , and These represent rolling resistance, air resistance, the component of gravity, and inertial force, respectively. This indicates the real-time battery voltage.
5. The method for monitoring the operating status of a sightseeing vehicle based on multi-dimensional sensing data according to claim 4, characterized in that: Step S321 specifically includes the following steps: S3211, Define the state vector Input vector and observation vector ,in, , , , , and They represent At any given moment, the vehicle's speed, sideslip angle, yaw rate, effective steering angle, longitudinal acceleration, and lateral acceleration are measured. and They represent The steering angle and wheel angular velocity after instantaneous integration; , , and They represent The observed velocity, yaw rate, longitudinal acceleration, and lateral acceleration after time-mapping; S3212, Combining the dynamics of sightseeing vehicles with the slope of the path. Road surface friction coefficient Construct a nonlinear state equation that couples path features: ； in, ； ； In the formula, Represents a nonlinear state transition function; express The speed at any given moment; and They represent The longitudinal and lateral accelerations at time t; Represents gravitational acceleration; Indicates the calculation step size; express The centroid sideslip angle at any given moment; express The yaw rate at any given moment; This indicates the distance from the front axle of the sightseeing vehicle to its center of gravity. and These represent the slip angles of the front and rear tires, respectively. Indicates the effective slip angle of the front tire; This indicates the distance from the rear axle of the sightseeing vehicle to its center of gravity. Indicates the effective slip angle of the rear tire; This represents the moment of inertia of the sightseeing vehicle about the z-axis; express Steering angle after moment fusion; This represents the path complexity turning angle correction factor; Indicates the torque of the drive wheels; Indicates the rolling resistance coefficient; Indicates the overall vehicle weight; Indicates the road surface friction correction coefficient; express Effective steering angle at any given time; S3213, Constructing the linear observation equation , This represents the observation matrix.
6. The method for monitoring the operating status of a sightseeing vehicle based on multi-dimensional sensing data according to claim 5, characterized in that: Step S322 specifically includes the following steps: S3221. Based on the model-free adaptive compensation principle of TDC, and taking the deviation between the control quantity at the previous moment and the state at the current moment as a basis, design a time delay compensation term for the motion state of the sightseeing car. : ； in, ； In the formula, express Time-based TDC compensation control quantity; express Time-based TDC compensation control quantity; Let represent the pseudo-input matrix, and ; express The rate of change of state at time t, and , Indicates the sampling interval. and They represent Time and The state vector at any given time; Indicates tracking reference items, and , This represents the rate of change of the ideal reference value. Represents the proportional gain matrix; This represents the path adaptive TDC compensation gain matrix, and , Represents the TDC basic compensation gain matrix; Indicates a status reference value; S3222, Add the 6-dimensional TDC time delay compensation term Additive fusion with UKF state predictions: ； In the formula, This represents the state prediction value after incorporating TDC compensation; This represents the UKF original state prediction value.
7. A method for monitoring the operating status of a sightseeing vehicle based on multi-dimensional sensing data according to claim 6, characterized in that: Step S4 Specifically, the following steps are included: S41. Drawing on the causal analysis method of Ishikawa's illustration, abnormal sightseeing bus operations are categorized into 6 main types: Human factors: operational errors, improper maintenance; Machine-related: Braking system failure, steering system failure, battery system failure, sensor failure; Environmental factors: low road surface adhesion, bumpy road surface, extreme temperatures; Route-related offenses: speeding on sharp bends, overloading on steep slopes; Process-related issues: failure to follow designated routes, failure to conduct regular maintenance; Materials-related: Battery aging, brake pad wear; S42. Dynamically adjust the early warning threshold based on risk level and path characteristics: ； in, ； ； ； In the formula, This indicates the dynamically adjusted warning threshold; The base value representing the warning threshold. Pick or ; This indicates the risk level output by LGBM; These represent the skid risk level and the rollover risk level, respectively. S43. Based on the TinyML CNN model, the fault is quickly identified and offline voice warning is triggered. The vehicle terminal uploads the abnormal data and preliminary warning information to the cloud. The cloud uses the BiLSTM-XGBoost hybrid model to locate the root cause of the fault and generate targeted suggestions.