A device and method for monitoring the running deviation posture of a cableway cabin

Abstract

The present application relates to the field of ropeway transportation safety control, in particular to a kind of ropeway hanging cabin running deflection posture monitoring device and control method, the method is: by constructing multimodal sensor network real-time acquisition hanging cabin position, speed, attitude angle and local micro deflection disturbance information;Adaptive sampling mechanism is used to dynamically adjust sensor frequency, combined with dynamic time regularity and least square estimation to complete the spatio-temporal alignment and fusion of heterogeneous data;Based on the attention mechanism of disturbance driving, the key disturbance dimension is strengthened, and the future deflection angle sequence is predicted;Through spatial deflection vector field modeling, regional coupling tensor analysis and dynamic direction alignment matrix, decoupling torque is generated, and control instruction is optimized by fusing power consumption constraint and dynamic environment parameter;Finally drive four corner torque gyro to execute correction, and based on attitude response residual, multidimensional evaluation index is constructed.The present application realizes the adaptive closed-loop adjustment of control strategy, and improves the robustness under wind speed disturbance and load variation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ropeway transportation safety control, and particularly relates to a device for monitoring the running yawing posture of a ropeway cabin and a control method. BACKGROUND

[0002] The existing ropeway cabin is easily affected by strong wind disturbance, load imbalance and ropeway vibration during operation, causing the cabin to yaw out of limits. The traditional scheme relies on fixed threshold alarms and emergency braking mechanisms, which have the problems of passive response lag and lack of predictive correction capability.

[0003] A three-dimensional posture monitoring method for a suspended traction chair based on a multi-intelligent sensor is disclosed in Chinese patent application CN120194686A, which comprises: obtaining and processing original sensor data to obtain sensor data; calculating posture data using the sensor data; constructing and training a multi-element autoregressive integral moving average model to generate a posture data prediction value; calculating dynamic posture data using the sensor data; fusing data in the posture data prediction value and the dynamic posture data to obtain a comprehensive pitch angle, a comprehensive roll angle and a comprehensive yaw angle; further, obtaining a comprehensive attitude value; based on the comprehensive attitude value, determining whether to trigger a warning mechanism, starting an automatic adjustment operation or starting an emergency shutdown program.

[0004] With the development of ropeway systems towards high speed, large capacity and complex environmental scenarios, accurate control of the running posture of the cabin has become a key technical requirement to ensure safety and comfort. The current ropeway control field is experiencing a technical evolution from passive response to active prediction, and multi-modal sensor fusion, nonlinear state prediction and dynamic decoupling optimization have become research hotspots. The yawing problem of the cabin under strong wind disturbance, load mutation and ropeway vibration urgently needs a systematic solution that integrates high-precision spatio-temporal perception, advanced risk prediction and adaptive closed-loop correction to meet the engineering requirements of modern ropeway systems for real-time performance, energy efficiency and long-term operation robustness. SUMMARY

[0005] The present application aims to solve the problems in the background art and proposes a device for monitoring the running yawing posture of a ropeway cabin and a control method.

[0006] The technical solution of the present application is a device for monitoring the running yawing posture of a ropeway cabin and a control method, which comprises the following specific implementation steps:

[0007] S1, collect the position, speed, attitude angle and local micro-yawing disturbance information of the cabin through a multi-source perception network, adaptively adjust the sampling frequency, normalize the data, and then fuse asynchronous data using dynamic time warping and least mean square estimation, outputting a fused state vector sequence;

[0008] S2, define a sensor channel disturbance energy index through local disturbance tensor analysis, screen a set of significant interference dominant factors, and construct a disturbance attention weight mechanism to weight and strengthen key disturbance dimensions to generate an enhanced state vector; input the D-GRU network to predict a future yaw angle sequence, and when the maximum absolute value of the predicted sequence exceeds a safety threshold, trigger an early warning and correction mechanism;

[0009] S3, construct a spatial vector field based on the predicted yaw sequence, analyze the moment interference between the corners through regional coupling tensor analysis, project and decouple combined with a dynamic direction alignment matrix, and optimize the moment command vector by fusing power consumption constraints and dynamic environment parameters, and package it as a standard control frame to synchronously drive the four-corner moment gyro to execute correction;

[0010] S4, divide the control risk level based on the comprehensive score function constructed by the attitude residual sequence, and execute short-period parameter fine-tuning and long-period model evolution update combined with an abnormal event graph attribution mechanism.

[0011] Preferably, the multi-source perception network comprises:

[0012] GPS / RTK unit: provides the position P(t), speed V(t), and heading angle θ(t) of the crane;

[0013] High-precision six-axis IMU unit: provides the pitch angle α(t), roll angle β(t), and yaw angle γ(t) and angular velocity;

[0014] Distributed neuron attitude array: deployed at the four corners of the crane, providing local micro-yaw disturbance information δ i (t), i.e., the local micro-disturbance yaw angle monitored by the i-th neuron attitude array sensor node at time t.

[0015] Preferably, the generation process of the enhanced state vector is as follows:

[0016] Calculate the disturbance change rate average of each sensor channel in the disturbance observation window as the disturbance energy index, select the channels exceeding the set threshold into the current interference dominant factor set, and identify the significant yaw driving trend;

[0017] Calculate the attention weight of each channel based on the interference dominant factor set, strengthen the key disturbance dimension using an exponential amplification mechanism, and weight and reconstruct the state vector to generate a disturbance enhanced state vector.

[0018] Preferably, the prediction process of the D-GRU network to predict the future yaw angle sequence is as follows:

[0019] ;

[0020] wherein, represents the disturbance gating variable at time t; denotes the hidden state at the previous time step; denotes the candidate hidden state; and denote the weight matrices of the perturbation gate for the current input and the previous state, respectively; and denote the input and state weight matrices of the reset gate, respectively; W and U denote the weight matrices of the input and the history state of the candidate state update function, respectively; denotes element-wise multiplication; denotes the perturbation enhanced state vector weighted by the perturbation attention; denotes the hidden state at the current time step;

[0021] the predicted sequence of the current state :

[0022] ;

[0023] ;

[0024] wherein, denotes the predicted sequence of the yaw angle composed of all the predicted yaw angles from time 1 to T; denotes the predicted yaw angle value at the k-th time step in the future; denotes the D-GRU hidden state at the predicted time t+k, which is the internal memory of the system state; denotes the weight matrix of the output mapping layer, which is used to map the hidden state to a specific angle prediction value.

[0025] Preferably, the optimization process of the torque command vector is as follows:

[0026] Based on the predicted yaw sequence, a spatial vector field is constructed to map the four-corner trend torque, and a regional coupling torque tensor is constructed to analyze the interference between the corner points by combining the position vector, the moment of inertia and the direction angle through a coupling coefficient;

[0027] Based on the regional torque distribution and the coupling tensor, the disturbance main direction is calculated by calculating the yaw trend gradient, a dynamic direction alignment matrix is constructed to map the yaw direction to the gyro installation direction, and the decoupling torque control vector is obtained to complete the decoupling calculation;

[0028] Through the dynamic torque optimization function, the physical boundary, the energy constraint and the historical inertia response characteristics are fused, the decoupled torque command is optimized by combining the dynamic parameters and the maximum power limit, and a stable and executable corrected torque vector is generated.

[0029] Preferably, the construction process of the regional coupling torque tensor is as follows:

[0030] Based on the predicted yaw angle sequence, the yaw trend moment vectors of the four corners of the bottom of the car are mapped to a spatial yaw vector field through a nonlinear mapping function, and a vector array is formed.

[0031] The four corners of the bottom of the car include: front left FL, front right FR, rear left BL, and rear right BR.

[0032] The regional coupling tensor is constructed:

[0033] Wherein, represents the coupling moment tensor item between the i and j corners of the car; represents the coupling coefficient between the i and j control corners in the car structure; and represent the yaw trend moment vectors on the i and j control corners, respectively. and represent the position vectors of the control corners i and j relative to the center of mass of the car, respectively. i and j represent the local rotational inertia about the center of mass, respectively. represents the yaw main direction angle between points i and j.

[0034] Preferably, the dynamic moment optimization function is:

[0035]

[0036]

[0037] Wherein, Tv represents the output moment vector of the four moment gyroscopes to be optimized at the current time; represents the system power consumption estimation function required for the current output moment; represents the set weight factor; represents the adaptive control penalty factor of the i control point; represents the basic penalty term; represents the weight coefficient corresponding to each dynamic index; represents the set wind speed reference value; represents the reference upper limit value of the car angular velocity; represents the acceptable upper limit of vertical vibration; represents the maximum safety delay threshold tolerated by the control system; Q(t) represents the theoretically optimal control moment vector after decoupling; represents the moment history output sequence; represents the current car running dynamic parameter set, including wind speed , car angular velocity , vertical vibration acceleration of the car , control delay​​​ ; represents the maximum output power of the moment gyro; represents the dynamic moment optimization function.

[0038] Preferably, the comprehensive score function is composed of the stability, overshoot and response delay three indexes, and the form is:

[0039] ; ;

[0040] ;

[0041] ;

[0042] wherein, represents the modified stability index; represents the weighted residual contribution value at the time point t+k; represents the modified overshoot index; represents the maximum safe swing angle threshold value acceptable for the operation of the gondola; represents the response delay index; represents the swing angle residual tolerance threshold value; represents the comprehensive score of the control score function, i.e. the comprehensive score function; , and represent weight coefficients.

[0043] Preferably, the disturbance energy index adopts a sliding observation window calculation mode, and the window length is a dynamic adjustable parameter, which is adaptively increased or decreased according to the disturbance intensity.

[0044] The technical scheme of the present application: a cableway gondola operation swing posture monitoring device, which is applicable to the above-mentioned cableway gondola operation swing posture monitoring device control method, comprising:

[0045] The integrated operation posture sensing and preprocessing module is used for synchronously collecting multi-dimensional state information of the gondola in the operation process and performing data preprocessing operation; the integrated operation posture sensing and preprocessing module comprises a GPS position and motion sensing unit, a three-axis gyroscope unit and a master control processing unit MCU;

[0046] The master control processing unit MCU integrates data-driven posture recognition and swing prediction algorithms, and performs the whole process from data preprocessing, state modeling, trend prediction to correction strategy generation;

[0047] The GPS position and motion sensing unit is used for collecting position and motion information of the gondola in the operation process in real time;

[0048] A three-axis gyroscope unit is arranged to collect the angular velocity and attitude change of the cabin in three degrees of freedom, capture the micro-sway trend and spatial disturbance response;

[0049] A disturbance-driven sway trend prediction and modeling module is configured to construct a predicted sequence of sway angles in a future time domain, establish a disturbance-gated recurrent unit network (D-GRU) for the attitude change of the cabin, and predict the future sway trend angle sequence.

[0050] A multi-axis coupled torque decoupling and correction strategy generation module is configured to construct a sway vector field according to the predicted sequence, perform coupled tensor modeling, generate a direction alignment matrix, and optimize dynamic power consumption, and generate a control torque instruction for the correction of the gyroscope.

[0051] A control closed-loop performance evaluation and adaptive strategy adjustment module is configured to perform correction effect scoring, abnormality attribution, and strategy updating according to the attitude residual error, and complete system self-evolution regulation.

[0052] Compared with the prior art, the above technical scheme of the present application has the following beneficial technical effects:

[0053] The present application designs a cableway cabin running sway attitude monitoring device and control method, which realizes real-time collection of cabin position, speed, attitude angle and local micro-sway disturbance information by constructing a multi-modal sensing network (GPS / RTK unit, high-precision IMU unit, distributed neuron attitude array), and realizes spatio-temporal alignment and high-precision fusion of heterogeneous data by combining adaptive sampling mechanism (frequency mapping scheduling) and cross-modal synchronous transformation function (DTW dynamic time warping + least mean square estimation), eliminating multi-source asynchronous errors and laying a reliable control data foundation.

[0054] Based on the disturbance attention mechanism, the dominant disturbance dimension is screened and the enhanced state vector is reconstructed, and the self-developed disturbance gated recurrent unit (D-GRU) network is input to predict the future sway sequence, the newly added disturbance gating term dynamically controls the hidden state update, significantly improves the nonlinear trend prediction accuracy, and realizes the early risk warning.

[0055] For the prediction result, the spatial sway vector field modeling and regional coupled tensor analysis are used to calculate the coupled torque interference weight, and the dynamic direction alignment matrix is used to realize torque decoupling and spatial projection optimization, generate a feasible control instruction considering physical constraints (power upper limit, dynamic delay, mechanical inertia), and drive the four-corner torque gyro to cooperate in correction.

[0056] The closed-loop control system constructs a multi-dimensional scoring mechanism (stability, overshoot, and response delay) based on real-time residual sequences. Combined with abnormal event graph attribution and dynamic strategy adjustment (short-cycle parameter fine-tuning and long-cycle model evolution), it significantly improves the system's robustness in dealing with wind speed disturbances, load changes, and long-term operation. Ultimately, it achieves comprehensive benefits such as rapid and stable convergence of the gondola's attitude, reduced emergency braking rate, suppression of ineffective power consumption (dynamic power constraint optimization to reduce gyroscope losses), and mitigation of mechanical fatigue. Attached Figure Description

[0057] Figure 1 This is a flowchart of a method for controlling the swaying attitude monitoring device of a cableway gondola, as proposed in this invention. Detailed Implementation

[0058] Example 1, as Figure 1 As shown, the present invention proposes a control method for a cableway gondola yaw attitude monitoring device, which includes the following specific implementation steps:

[0059] S1. By constructing a multimodal sensor network, an adaptive sampling mechanism, and a cross-modal synchronous transformation function, spatial fusion and temporal alignment of heterogeneous data are achieved, laying a high-quality data foundation for subsequent state evolution prediction and yaw control. The specific implementation process is as follows:

[0060] S11. Construct a multi-source sensing network based on an edge computing architecture, where sensors include, but are not limited to:

[0061] GPS / RTK unit: provides the gondola's position P(t), velocity V(t), and heading angle θ(t);

[0062] High-precision six-axis IMU unit: provides attitude angles (three-axis angles: pitch angle α(t), roll angle β(t), and yaw angle γ(t) and angular velocity);

[0063] Distributed neuron pose array: deployed at the four corners of the gondola, providing local micro-sway perturbation information δ i (t), which is the local perturbation sway angle detected by the i-th neuron attitude array sensing node at time t, reflecting the instantaneous attitude fluctuation of a certain corner of the gondola;

[0064] S12. Set up a frequency mapping scheduling mechanism to adaptively adjust the sampling period according to the disturbance change rate, so that different modal data can form a synchronization cluster under event-driven conditions: ;

[0065] in, This represents the adaptive sampling frequency of the i-th type of sensor at time t; Indicates the preset base sampling frequency; This represents the disturbance measured by the i-th type of sensor at time t; represents the disturbance change rate; k represents the disturbance sensitivity coefficient (empirical setting);

[0066] S13, normalize each sensor data to obtain normalized data, and then construct a state fusion vector S(t): ;

[0067] wherein, represents the normalized vector of the position triplet P(t); and respectively represent the normalized results of the velocity and the heading angle; represents the yaw angle disturbance normalized value of the i-th attitude node;

[0068] S14, an asynchronous interpolation fusion mechanism based on dynamic time warping (DTW) + least squares estimation is introduced to interpolate the optimal state estimation value under a unified time index, specifically:

[0069] establish a target time axis , which is the unified time base for state modeling;

[0070] synchronize all data using a DTW (dynamic time warping) + linear interpolation combination mechanism:

[0071] ;

[0072] Interpolation introduces a cross-modal similarity constraint: ;

[0073] wherein, represents the i-th dimensional data estimation value calculated by interpolation at the target alignment time t; and respectively represent the original values of the i-th dimensional data at time points t k+1 and t k ; t k and t k+1 respectively represent the nearest previous and subsequent sampling time points (less than and greater than t, respectively) from the target time t; represents a time series alignment error cost function, which is used to measure the interpolation consistency of all modalities at the target time, and the smaller the value, the more successful the interpolation alignment between different data streams; represents the similarity weight between the i-th and j-th modalities at time t, i.e., the weight of their physical logic or functional coupling relationship (for example, position and velocity influence each other, with a large weight; yaw angle and height have no strong association, with a small weight);

[0074] S15, output the fusion state vector sequence {S(t i )}.

[0075] S2, convert the high-dimensional time sequence state vector into a future time domain yaw attitude evolution trajectory, model and predict the nonlinear evolution trend of the attitude of the hoist car, and the specific implementation process is as follows:

[0076] S21, a disturbance energy spectrum model based on local disturbance tensor analysis is constructed, and the input fusion state vector sequence is subjected to disturbance significance analysis to determine whether there is a significant yaw driving trend in the current system state, specifically:

[0077] The disturbance energy index is defined as: ;

[0078] Wherein, represents the disturbance energy of the i-th sensor channel; represents the yaw angle data of the i-th node; represents the disturbance observation window length;

[0079] If a certain dimension exceeds the threshold , the channel is included in the current disturbance dominant factor set D t ;

[0080] S22, based on the disturbance dominant channel set D t selected above, a disturbance-driven attention distribution mechanism is constructed to weight and strengthen the key disturbance dimension in the high-dimensional state sequence;

[0081] The following disturbance attention weight function is used: ;

[0082] Wherein, represents the disturbance attention weight of the i-th disturbance dominant channel at time t, and the greater the value, the greater the influence of the channel on the overall yaw trend; represents the disturbance energy amplification coefficient (disturbance sensitivity coefficient), which controls the influence degree of disturbance energy on attention distribution, and the value is generally adjusted between 2-5; represents the natural exponential function, which is used to map the disturbance energy value to a nonlinear attention value; represents the disturbance energy of the j-th sensor channel in the current disturbance dominant factor set D t ;

[0083] At each time t, the disturbance-enhanced state vector is generated by performing disturbance attention weighting on the state vector S(t): ;

[0084] Wherein, represents the disturbance-enhanced state vector after disturbance attention weighting; denotes the normalized state vector component of the i-th channel at time t, extracted from S(t) as a sub-vector;

[0085] S23, the weighted state sequence is input into a Disturbance-Gated Recurrent Unit (D-GRU) network for nonlinear state modeling, which adds a disturbance gating term to the traditional GRU to dynamically control the amplitude and direction of memory state update:

[0086] ;

[0087] where, denotes the disturbance gating variable at time t, which determines whether (and how) the current state information is updated due to disturbance, ranging between [0, 1]; denotes the hidden state (memory vector) at the current time, which is the main state representation inside the D-GRU unit, carrying historical and current pose change information; denotes the hidden state at the previous time, which is used to control whether the current state continues the previous trend; denotes the candidate hidden state, which is the intermediate result of the current input state update (with disturbance control), used for fusion with the previous state; and denote the weight matrices of the disturbance gate corresponding to the current input and the previous state, respectively; and denote the input and state weight matrices of the reset gate, respectively; W and U denote the input and historical state weight matrices of the candidate state update function, respectively; denotes element-wise multiplication;

[0088] S24, the final output of the predicted sequence of the current state :

[0089] ;

[0090] ;

[0091] where, denotes the predicted sequence composed of all predicted yaw angles within a future period of time (1~T time steps); denotes the yaw angle value predicted at the k-th time step in the future; denotes the D-GRU hidden state at the prediction time t+k, which is the internal memory of the system state; denotes the weight matrix of the output mapping layer, which is used to map the hidden state mapping to a specific angle prediction value;

[0092] S25, after predicting the output, the whole trend discrimination:

[0093] If , trigger the early swing correction mechanism, and generate an alarm signal;

[0094] where, represents the safety swing threshold (set as the safety upper limit angle of the car swing), exceeding this value is considered as a potential dangerous state; represents the absolute value of the maximum swing angle in the prediction sequence, which is used to judge whether early correction is needed.

[0095] S3, for the predicted future swing trend of the car, a dynamic adaptive spatial correction strategy generation mechanism is constructed combining the spatial structure characteristics of the car and the torque interference law. This mechanism not only considers the interference and collaborative relationship between multiple points, but also generates real-time executable control instructions by decoupling modeling, direction registration and power optimization, to drive the four high-precision torque gyro modules to work collaboratively. The specific implementation process is as follows:

[0096] S31, based on the future swing angle sequence , a set of nonlinear mapping functions is used to convert the overall swing trend into a spatial swing vector field: ;

[0097] where, respectively represent the swing trend vectors (mapped from the swing prediction value) of the four corners (front left FL, front right FR, rear left BL, rear right BR) of the car at the current time; represents the vector array composed of the four corner swing vectors;

[0098] Construct a regional coupling tensor:

[0099] ;

[0100] where, represents the coupling torque tensor item between the i and j corners of the car; represents the coupling coefficient between the i and j control corners in the car structure, which is used to measure the disturbance transmission strength between the two corners; and respectively represent the swing trend torque vectors on the i, j control corners (such as front left and rear right); and respectively represent the position vectors of control corners i, j relative to the mass center of the car;I i andI jThese represent the local rotational inertia about the center of mass, reflecting the sensitivity of this corner point to overall attitude disturbances. This represents the angle between the principal directions of the sway at points i and j;

[0101] S32. Based on the obtained regional moment distribution and coupling tensor, a dynamic direction alignment matrix is ​​constructed. The spatial projection direction of the control command is automatically adjusted according to the current dominant yaw direction. A dynamic projection mechanism between the dominant disturbance direction and the spatial control normal is adopted to achieve highly robust decoupling calculations. Specifically:

[0102] Calculate the principal direction of the perturbation based on the gradient of the yaw trend function: ;

[0103] Construct an orientation alignment matrix A(t) to describe the spatial mapping relationship between the yaw direction and the gyroscope mounting direction: ;

[0104] Based on this, the decoupling torque control vector is obtained: ;

[0105] in, Indicates the dominant direction vector of the yaw trend; Indicates skewing prediction sequence The gradient; A(t) represents the unit vector of direction of the gyroscopes installed at the four corners of the gondola; A(t) represents the dynamic direction alignment matrix; Q(t) represents the theoretically optimal control torque vector after decoupling.

[0106] S33. Combining physical boundaries, energy constraints, and historical inertial response characteristics, the decoupled torque command is dynamically optimized to ensure that the output command is both executable and meets system stability requirements.

[0107] Specifically, a control optimization function is introduced: ;

[0108] in, This represents the optimized true torque command vector, which is the physical torque value that will ultimately be used to drive the correction actions of each gyroscope. This represents the historical output sequence of torque, used to smooth the difference between the current output and the historical output; This represents the current set of dynamic parameters for the gondola's operation, including but not limited to wind speed. angular velocity of the hoisting car Vertical vibration acceleration of the hoisting car Control delay This is used to adjust the strength of the control strategy and the response delay; This indicates the maximum output power of the torque gyroscope, used to constrain the maximum amplitude of the current control torque and prevent damage to the device due to excessive power. denotes a dynamic torque optimization function;

[0109] Specifically, the projection optimization function used in the embodiment The overall cost function comprehensively considers the consistency of the current correction target and the theoretical expectation, the inertia retention of the historical correction amplitude, the soft constraint penalty of the upper limit of power consumption, and the adjustment penalty term consistent with the current dynamic environment. A physically feasible and dynamically stable correction torque vector is automatically projected by minimizing the overall cost function:

[0110]

[0111]

[0112] wherein Tv represents the output torque vector of the four torque gyroscopes to be optimized at the current time; denotes a system power consumption estimation function required for the current output torque; denotes a set weight factor; denotes an adaptive control penalty factor of the i-th control point (torque gyroscope); denotes a basic penalty term for maintaining the minimum suppression intensity under low dynamic disturbance conditions to prevent unstable correction caused by zero penalty; denotes a weight coefficient corresponding to each dynamic indicator; denotes a set wind speed reference value; denotes a reference upper limit value of the cabin angular velocity; denotes an acceptable upper limit of vertical vibration; denotes a maximum safety delay threshold value that the control system can tolerate;

[0113] S34, after obtaining the actual feasible correction torque, this step is responsible for converting it into a standardized control instruction and issuing it to the set four torque gyroscopes in a synchronous manner, and entering the closed-loop correction control. The operation process includes:

[0114] determines the direction code;

[0115] constructs a packaging control instruction frame

[0116] concurrent transmission to four gyro control chips, real-time monitoring of feedback return (attitude response and angular velocity residual error) as the next cycle feedback input;

[0117] wherein, denotes the direction code (encoding indicates the direction of action, such as clockwise / counterclockwise or axial number) executed by the i-th gyroscope, which is used to accurately control the rotation direction of the gyroscope; ​​​​The duration of the torque action, according to the deflection trend persistence and the dynamic response window setting, affects the correction action duration intensity; The standardized control instruction frame is indicated; The instruction packaging function is indicated, which combines control parameters into a communication frame format recognizable by the device, facilitating MCU distribution and communication module recognition.

[0118] S4, a dynamic adaptive closed-loop control system that integrates real-time response evaluation, feedback anomaly identification, and control strategy reconstruction is constructed, continuously improving the response efficiency and robustness of the trolley deflection correction system, ensuring that the trolley still has stable correction ability after wind speed disturbance, load variation, or long-time operation. The specific implementation process is as follows:

[0119] S41, based on the actual attitude change result of the trolley after the torque correction instruction is issued, the residual sequence between control execution and target is constructed, specifically:

[0120] The actual attitude state of the trolley after the torque action is collected in real time, and the feedback state sequence is constructed:

[0121] ;

[0122] The predicted deflection trend sequence Compare and define the deflection correction residual vector:

[0123] ;

[0124] Among them, The actual observed trolley attitude deflection angle sequence, i.e. the actual deflection response state after torque correction execution; The deflection correction residual vector is used to evaluate the attitude correction error between prediction and actual; T represents the time decay weight coefficient; T represents the control evaluation window length, i.e. the length of the future time sequence that needs to be evaluated;

[0125] S42, after obtaining the complete residual vector, build a multi-dimensional evaluation index to score the current control strategy, and identify the control potential failure or overcompensation risk level;

[0126] The scoring mechanism uses a set of self-defined multi-dimensional evaluation index functions:

[0127] Correction stability index (whether the correction quickly tends to be stable): ;

[0128] Correction overshoot index (whether there is an overcorrection trend): ;

[0129] Response delay index (actual correction start lag time): ;

[0130] Combining the above indicators, a comprehensive scoring function is constructed: ;

[0131] wherein, represents the modified stability index, i.e., the relative accuracy of the average residual, which is used to evaluate whether the state of the gondola after correction is stable and close to the target state; represents the weighted residual contribution value at time point t+k, which measures the influence of the yaw error at that time on the overall correction effect; represents the correction overshoot index, i.e., the ratio between the maximum actual attitude deviation after correction and the safety upper limit, which reflects whether an overshoot phenomenon occurs; represents the maximum safe yaw angle threshold acceptable for the gondola operation; represents the response delay index, i.e., the number of time steps required for the actual yaw angle to first approach the predicted angle, and the smaller the number, the faster the response; represents the yaw angle residual tolerance threshold; represents the comprehensive score of the control scoring function, which combines the weighted evaluation of the overall performance of the current control strategy based on the three indicators; , and represent the weight coefficients;

[0132] Accordingly, based on the comprehensive score of the control scoring function, the risk level is divided as follows:

[0133] A level (excellent control), > ; accurate correction, rapid response, and no obvious overshoot; maintain the current strategy without adjustment;

[0134] B level (suboptimal control), ; there is a slight response delay or yaw rebound; trigger a slight strategy parameter fine-tuning, such as control torque amplitude or duration;

[0135] C level (imbalance or overcompensation), ≤ ; characterized by correction failure, severe overcompensation, or increasing yaw; trigger the adaptive control strategy update process to reconstruct the torque decoupling and execution strategy;

[0136] wherein, and are the set scoring thresholds, which are dynamically adjusted according to different cableway systems;

[0137] S43, based on the residual distribution and the history of strategy execution, the abnormal correction behavior is attributed, the source of yaw abnormality is identified, and it is determined whether the control strategy needs to be adjusted, and a correction abnormal event graph G is constructed e :

[0138] ;

[0139] wherein, G e represents an abnormal response event graph; V represents a set of vertices in the graph, containing key variables related to abnormal events, containing residual , torque command and disturbance direction ;

[0140] S44, on the basis of the evaluation result, the core model (such as yaw decoupling matrix, control command mapping function) of yaw correction strategy is locally fine-tuned or long-term learning is updated, and an evolution control system is formed:

[0141] Short cycle fine-tuning (strategy layer), including but not limited to: weight adjustment of torque distribution matrix Q(t); adjusting control command duration and direction threshold parameters;

[0142] Long-term model self-evolution (model layer), including but not limited to: accumulation of historical response-command data pairs in the sliding window; using lightweight RNN or autoregressive model to model the control effect of "prediction-inversion-correction".

[0143] Embodiment two, the present application proposes a cableway cabin running yaw posture monitoring device, which is suitable for the cableway cabin running yaw posture monitoring device control method proposed in embodiment one, which comprises: an integrated running posture sensing and preprocessing module, a disturbance driven yaw trend prediction and modeling module, a multi-axis coupled torque decoupling and correction strategy generation module, and a control closed loop performance evaluation and adaptive strategy adjustment module.

[0144] The integrated running posture sensing and preprocessing module synchronously collects multi-dimensional state information of the cabin during running, performs data preprocessing operation, and provides high-quality input for subsequent modeling;

[0145] The integrated running posture sensing and preprocessing module includes a GPS position and motion sensing unit, a three-axis gyroscope unit and a master control processing unit MCU;

[0146] The master control processing unit MCU, as the core of the system, integrates data-driven posture recognition and yaw prediction algorithm, and performs the whole process from data preprocessing, state modeling, trend prediction to correction strategy generation;

[0147] The GPS position and motion sensing unit is used for real-time collection of running speed, geographic position, height change and running direction of the cabin, as the sensing entrance of macro running state;

[0148] A three-axis gyroscope unit is used to collect the angular velocity and attitude change of the cage in three degrees of freedom, to capture the micro-swing trend and spatial disturbance response;

[0149] A disturbance-driven swing trend prediction and modeling module is used to construct a swing angle prediction sequence in the future time domain, to establish a time-direction coupling trend model of the cage attitude change, and to output the swing vector at multiple time points in the future;

[0150] A multi-axis coupled torque decoupling and correction strategy generation module is used to construct a swing vector field based on the prediction results, to perform regional coupling tensor modeling, direction alignment matrix solving, and dynamic power constraint optimization strategy, to generate a minimum interference correction vector for controlling the four-corner torque gyro, and to issue the vector to generate a directional counter-torque to maintain the stability of the cage;

[0151] A control closed-loop performance evaluation and adaptive strategy adjustment module is used to collect the cage attitude correction response and prediction residual, to construct a multi-index control effect evaluation system, to judge the current correction effect level, to determine the swing abnormality reason in combination with the residual atlas attribution mechanism, and to perform online adjustment on the decoupling matrix, control instruction, and parameter model, to form a systematic dynamic adaptive correction closed loop.

[0152] The embodiments of the present application are described in detail above in combination with the drawings, but the present application is not limited thereto, and various changes can be made within the knowledge range of those skilled in the art without departing from the purpose of the present application.

Claims

1. A method of controlling a device for monitoring the attitude of a cableway gondola in a running deviation, characterized in that The specific implementation steps include the following: S1, collect the position, speed, attitude angle and local micro-tilt disturbance information of the cage through a multi-source perception network, adaptively adjust the sampling frequency, normalize the data, and then fuse the asynchronous data using dynamic time warping and least mean square estimation, and output the fused state vector sequence; S2, define the sensor channel disturbance energy index through local disturbance tensor analysis, screen the significant disturbance dominant factor set, and construct a disturbance attention weight mechanism to weight and strengthen the key disturbance dimension to generate an enhanced state vector; input the D-GRU network to predict the future yaw angle sequence, and when the maximum yaw angle absolute value of the predicted sequence exceeds the safety threshold, trigger the early warning and correction mechanism; S3, based on the predicted yaw sequence, construct a spatial vector field, analyze the moment disturbance between the corners through regional coupling tensor, project and decouple combined with the dynamic direction alignment matrix, and optimize the moment command vector by fusing the power consumption constraint and the dynamic environment parameter, and package it as a standard control frame to drive the four-corner torque gyro to execute the correction; S4, based on the attitude residual sequence, construct a comprehensive score function to divide the control risk level, and execute short-period parameter fine-tuning and long-period model evolution update combined with the abnormal event graph attribution mechanism.

2. A method of controlling a device for monitoring the attitude of a cableway gondola in operation according to claim 1, characterized in that, The multi-source perception network includes: GPS / RTK unit: provides the position P(t), speed V(t), and heading angle θ(t) of the cage; High-precision six-axis IMU unit: provides the pitch angle α(t), roll angle β(t), and yaw angle γ(t) and angular velocity; Distributed neuron posture array: deployed at the four corners of the car, providing local micro-tilt disturbance information δ i (t), i.e. the local disturbance tilt angle monitored by the i-th neuron posture array sensor node at time t.

3. A method of controlling a device for monitoring the attitude of a cableway gondola in operation according to claim 2, characterized in that The generation process of the enhanced state vector is as follows: Calculate the disturbance change rate average of each sensor channel in the disturbance observation window as the disturbance energy index, select the channels exceeding the set threshold into the current disturbance dominant factor set, and identify the significant yaw driving trend; Based on the disturbance dominant factor set, calculate the attention weight of each channel, use the exponential amplification mechanism to strengthen the key disturbance dimension, and weight the state vector to generate a disturbance enhanced state vector.

4. A method of controlling a device for monitoring the attitude of a cableway gondola in operation according to claim 3, characterized in that, The prediction process of the D-GRU network to predict the future yaw angle sequence is as follows: ； wherein, denotes the perturbation gating variable at time t; denotes the hidden state at the previous time step; denotes the candidate hidden state; and denote the weight matrices for the current input and previous state in the perturbation gating, respectively; and denote the input and state weight matrices for the reset gate, respectively; W and U denote the weight matrices for the input and history state of the candidate state update function, respectively; denotes element-wise multiplication; denotes the perturbation augmented state vector after weighting by the perturbation attention; denotes the hidden state at the current time step; Output a predicted sequence of current states : ； ； wherein, represents the predicted sequence of yaw angles at future time steps from 1 to T; represents the predicted yaw angle value at future time step k; represents the D-GRU hidden state at the predicted time t+k, i.e., the internal memory of the system state; represents the weight matrix of the output mapping layer, which is used to map the hidden state to a specific angle prediction value.

5. A method of controlling a device for monitoring the attitude of a cableway gondola in operation according to claim 4, characterized in that, The optimization process of the moment command vector is as follows: Based on the predicted yaw sequence, construct a spatial vector field to map the four-corner trend moment, construct a regional coupling moment tensor to analyze the disturbance between the corners by coupling coefficients combined with the position vector, moment of inertia, and direction angle; Based on the regional moment distribution and coupling tensor, calculate the disturbance main direction from the yaw trend gradient, construct a dynamic direction alignment matrix to map the yaw direction to the gyro installation direction, obtain the decoupled moment control vector, and complete the decoupling calculation; Through the dynamic moment optimization function, fuse the physical boundary, energy constraint, and historical inertia response characteristics, combine the dynamic parameters and maximum power limit to optimize the decoupled moment command, and generate a stable and executable correction moment vector.

6. A method of controlling a device for monitoring the attitude of a cableway gondola in operation according to claim 5, characterized in that, The construction process of the regional coupling moment tensor is as follows: Based on the predicted yaw angle sequence, convert it into a spatial yaw vector field through a nonlinear mapping function, map the yaw trend moment vector of the four corners at the bottom of the cage, and form a vector array; The four corners at the bottom of the cage include: front left FL, front right FR, back left BL, and back right BR; Construct a regional coupling tensor: ； where, represents the coupling moment tensor term between the i and j control angles of the gondola structure; represents the coupling coefficient between the i and j control angles of the gondola structure; and represent the yawing tendency moment vector on the i, j control angles, respectively; and represent the position vector of the control angles i, j with respect to the gondola mass center, respectively; i and j represent the local moment of inertia about the mass center, respectively; represents the yawing principal direction angle between the points i, j.

7. A method of controlling a device for monitoring the attitude of a cableway gondola in operation according to claim 6, characterized in that The dynamic moment optimization function is as follows: ； ； wherein Tv represents the output torque vector of the four torque gyroscopes to be optimized at the current time instant; represents the system power consumption estimation function of the current output torque; represents the set weight factor; represents the adaptive control penalty factor of the i-th control point; represents the basic penalty term; represents the weight coefficient corresponding to each dynamic index; represents the set wind speed reference value; represents the reference upper limit value of the cabin angular velocity; represents the acceptable upper limit of the vertical vibration; represents the maximum safety delay threshold value that the control system can tolerate; Q(t) represents the theoretically optimal control torque vector after decoupling; represents the torque historical output sequence; represents the current cabin operating dynamic parameter set, including the wind speed , the cabin angular velocity , the vertical vibration acceleration of the cabin , the control delay ; represents the maximum output power of the torque gyroscope; represents the dynamic torque optimization function.

8. A method of controlling a device for monitoring the attitude of a cableway gondola in operation according to claim 7, characterized in that, The comprehensive score function is composed of stability, overshoot and response delay, and its form is: ； ； ； ； wherein, denotes a modified stability index; denotes a weighted residual contribution value at time point t+k; denotes a modified overshoot index; denotes a maximum safe sway angle threshold value acceptable for the operation of the car; denotes a response delay index; denotes a sway angle residual tolerance threshold value; denotes a combined score of the control scoring function, i.e. a combined scoring function; , and denote a weight coefficient.

9. A method of controlling a device for monitoring the attitude of a cableway gondola in operation deviating from the vertical according to claim 3, characterized in that, The disturbance energy index adopts a sliding observation window calculation method, and the window length is a dynamic adjustable parameter, which is adaptively increased or decreased according to the disturbance intensity.

10. A device for monitoring the running skew posture of a cableway cabin, which is suitable for use in a method for monitoring the running skew posture of a cableway cabin according to any one of claims 1 to 9, characterized in that Comprise: The integrated running posture perception and preprocessing module is used for synchronously collecting multi-dimensional state information of the cage during the running process and performing data preprocessing operations. The integrated running posture perception and preprocessing module comprises a GPS position and motion perception unit, a three-axis gyroscope unit and a master control processing unit MCU. The master control processing unit MCU integrates data-driven posture recognition and yawing prediction algorithms, and performs the whole process from data preprocessing, state modeling, trend prediction to correction strategy generation. The GPS position and motion perception unit is used for collecting position and motion information of the cage in the running process in real time. The three-axis gyroscope unit collects angular velocity and attitude change of the cage in three degrees of freedom, and captures micro yawing trend and spatial disturbance response. The disturbance-driven yawing trend prediction and modeling module constructs a yawing angle prediction sequence in the future time domain, establishes a disturbance-gated recurrent unit network D-GRU of the cage attitude change, and predicts the future yawing trend angle sequence. The multi-axis coupled torque decoupling and correction strategy generation module is used for constructing a yawing vector field according to the prediction sequence, performing coupled tensor modeling, generating a direction alignment matrix and dynamic power consumption optimization, and generating a control torque instruction for gyro execution correction. The control closed-loop performance evaluation and adaptive strategy adjustment module is used for performing correction effect scoring, abnormal attribution and strategy updating according to the posture residual error, and completing system self-evolution regulation.

Images