Intelligent operation and maintenance and state monitoring method and system for rail transit platform door facilities

Abstract

The invention provides a rail transit platform door facility intelligent operation and maintenance and state monitoring method and system, and relates to the technical field of rail transit safety, and the method comprises the steps: collecting operation data through a plurality of data collection modules, and uploading the operation data to a central control platform for preprocessing; establishing a sliding time domain window prediction moving track based on a model prediction control algorithm; a cascade PID control structure is constructed to realize closed-loop control; a fuzzy rule base is established for different operation sections, and control parameters are optimized in combination with a BP neural network and a genetic algorithm. The operation stability of the platform door system is improved, the maintenance cost is reduced, and the safety is enhanced.

Description

Technical Field

[0001] This invention relates to the field of rail transit safety technology, and in particular to intelligent operation and maintenance and status monitoring methods and systems for rail transit platform screen door facilities. Background Technology

[0002] Platform screen doors are crucial safety equipment in modern urban rail transit systems, and their operational status directly impacts passenger safety and train efficiency. With the rapid expansion of urban rail transit networks and the continuous increase in passenger volume, platform screen door systems are under immense pressure from high-frequency, high-load operations. Traditional platform screen door operation and maintenance management relies primarily on periodic manual inspections and reactive maintenance strategies, which are insufficient to meet the demands of modern rail transit for efficient, reliable, and intelligent operation and maintenance.

[0003] Platform screen door systems typically consist of groups of multiple independently operating doors, each equipped with a drive motor, sensors, and a control unit. The opening and closing process of platform screen doors involves precise position control, speed regulation, and torque management, requiring the control system to have high precision, fast response, and anti-interference capabilities. Simultaneously, platform screen door systems need to work closely with the train signaling system to ensure the coordinated and consistent operation of train stops and platform screen door opening and closing.

[0004] Currently, the operation and maintenance monitoring technology for rail transit platform screen doors faces several major problems: Traditional control methods lack predictability and adaptability, failing to automatically optimize control parameters based on different operating conditions. This leads to unstable performance of platform screen doors under varying environmental conditions and load changes, affecting the smoothness and accuracy of opening and closing actions. Existing monitoring systems suffer from incomplete data acquisition and limited processing capabilities, unable to capture and analyze operational data of platform screen door groups in real time. This hinders early warning of potential faults, causing small problems to accumulate and develop into major malfunctions. Furthermore, the control system fails to effectively handle signal transmission delays, easily resulting in control lag under high-frequency operating conditions. This reduces system response speed and control accuracy, impacting passenger experience and safety.

[0005] With the development of artificial intelligence and big data technologies, applying intelligent algorithms to the operation and maintenance (O&M) and monitoring of platform screen door systems has become an important way to improve system reliability, extend equipment life, and reduce maintenance costs. Establishing intelligent O&M and condition monitoring methods to achieve precise control and predictive maintenance of platform screen door systems has become a research hotspot and development direction in the rail transit field. Summary of the Invention

[0006] This invention provides a method and system for intelligent operation and maintenance and status monitoring of rail transit platform screen door facilities, which can solve the problems in the prior art.

[0007] A first aspect of this invention provides a method for intelligent operation and maintenance and status monitoring of rail transit platform screen door facilities, comprising: The platform screen door group collects operational data by deploying multiple data acquisition modules in the platform screen door group, and uploads the operational data to the central control platform in real time. The central control platform preprocesses the operational data to generate preprocessed data. Based on the preprocessed data, a sliding time-domain window is established using a model predictive control algorithm to predict the operating trajectory of the platform screen door group at multiple sampling points in the future, thereby obtaining the optimal control output sequence. According to the optimal control output sequence, a cascaded PID control structure is constructed, comprising a position loop, a speed loop, and a current loop. The position loop output serves as the speed loop reference, and the speed loop output serves as the current loop reference. A Smith predictor is used to compensate for transmission delay, achieving closed-loop control of the platform screen door position, speed, and torque. The operating parameters of the platform screen door group are initially configured based on the closed-loop control. The platform screen door operation process is divided into four stages: starting, constant speed, deceleration, and arrival. A fuzzy rule base is established for each stage. The mapping relationship between control parameters and operating performance at each stage is trained based on a BP neural network. The evaluation results of the fuzzy rule base are used as training samples for the neural network. A genetic algorithm is used to iteratively optimize the control parameters. Based on real-time detected load changes and environmental disturbances, the parameter combinations of each control link are adjusted online to achieve adaptive optimization of the control parameters. The system monitors the operation status of the platform screen door group in real time. When an abnormal operation is detected, the system triggers the adaptive optimization process of the control parameters. After the parameters are adjusted, the adjustment results are fed back to the central control platform.

[0008] Based on the preprocessed data, a sliding time-domain window is established using a model predictive control algorithm to predict the operating trajectory of platform screen door groups at multiple sampling points in the future, thereby obtaining the optimal control output sequence, including: The position, velocity, and acceleration parameters of the platform door group are obtained as state space parameters. The motor driving torque and load torque of the platform door group are calculated. The motor driving torque and load torque are substituted into the nonlinear state equation to obtain the electromechanical coupling state prediction result. The system response time and reference trajectory period during the movement of the detection platform gate group are used to calculate the ratio of the system response time to the reference trajectory period to obtain a weighting coefficient. The weighting coefficient is then multiplied by the sampling period to obtain the predicted time domain length. The window length of the sliding time domain window is set according to the predicted time domain length. The electromechanical coupling state prediction results are stored in the sliding time domain window in chronological order. Finally, a state prediction matrix is ​​generated. Multi-step prediction data of the platform door group is calculated based on the state prediction matrix. The output tracking error data is obtained by subtracting the preset target trajectory data from the multi-step prediction data, the control increment data is obtained by calculating the difference between two adjacent control outputs, and the smoothness data is obtained by calculating the slope change of the target trajectory data at adjacent sampling points. The weighted objective function value is obtained by calculating the weighted sum of squares of the output tracking error data, the control increment data, and the smoothness data respectively. The state space parameters are normalized by dividing them by the corresponding maximum allowable value. The normalized state space parameters are compared with a preset threshold to generate constraints. The initial iteration values ​​are set using a hot start method, and the constraints are substituted into the quadratic programming equation to obtain the optimal control output sequence through calculation.

[0009] By compensating for transmission delay using a Smith predictor, closed-loop control of platform screen door position, speed, and torque is achieved, including: The system acquires the position signal, speed signal, and current signal of the platform screen door system, combines the position signal, speed signal, and current signal into a system state vector, calculates the dynamic response characteristics of the system based on the system state vector, and obtains the system state space matrix; the optimal control output sequence is used as the given trajectory of the position loop, and the position signal, speed signal, and current signal are respectively connected to the position loop, speed loop, and current loop; The total system delay of the platform screen door system during data transmission is detected and calculated. The total system delay and the system state space matrix are substituted into the transfer function of the Smith predictor to calculate the delay compensation parameters. The predicted outputs of the position loop, the velocity loop, and the current loop are calculated based on the delay compensation parameters. The actual outputs of each loop are subtracted from the predicted outputs to obtain the prediction errors. The cumulative value of the prediction errors is calculated to obtain the integral error term. The error change rate is calculated based on the prediction errors to obtain the differential error term. The delay estimate is obtained by calculating the prediction errors. The delay estimates are compensated to the position loop, the velocity loop, and the current loop respectively, and the system state prediction values ​​of each loop are updated. The observer gain matrix is ​​calculated based on the deviation between the system state prediction values ​​and the actual measured values, and the observer gain matrix is ​​used to compensate for the system state estimates. The control output signal of each loop is calculated based on the compensated system state estimate, and the control output signal is fed back to the platform screen door system in a cascade structure to achieve closed-loop control.

[0010] The platform screen door operation process is divided into four segments: starting, constant speed, deceleration, and arrival. A fuzzy rule base is established for each segment, including: The platform gate's operating speed, acceleration, and remaining travel are monitored in real time. Based on the relationship between the operating speed, acceleration, and remaining travel and the speed, acceleration, and distance thresholds, the platform gate's operation is divided into a starting phase, a constant speed phase, a deceleration phase, and a final arrival phase. Position error signals and speed error signals are calculated based on the actual position and speed signals of the platform gate; a dynamic adjustment factor is calculated using the position error signals and speed error signals, and the dynamic adjustment factor is multiplied by a preset ambiguity parameter to obtain an adaptive ambiguity parameter; the position error membership value and speed error membership value are obtained by calculating the adaptive ambiguity parameter with the position error signals and speed error signals, respectively. Based on the current operating segment of the platform door, the corresponding rule group is selected from the pre-set rule base. The starting segment rule group inputs the position error membership value and the speed error membership value; the constant speed segment rule group inputs the speed membership value and the acceleration membership value; the deceleration segment rule group inputs the distance membership value, the speed membership value, and the acceleration membership value; and the final segment rule group inputs the position error membership value, the speed error membership value, and the force membership value. Calculate the activation intensity of the rules in each rule group, multiply the activation intensity by the confidence coefficient of the corresponding rule to obtain the rule weight, multiply the output value of each rule by its corresponding rule weight and sum them, then divide by the sum of the rule weights to obtain the control output value of the current running segment.

[0011] The evaluation results of the fuzzy rule base are used as training samples for the neural network, and a genetic algorithm is used to iteratively optimize the control parameters, including: The input variables corresponding to each rule in the fuzzy rule base are obtained, the membership function values ​​of each input variable are calculated, the membership function values ​​of the same rule are multiplied to obtain the rule activation degree, the actual performance index is calculated based on the rule activation degree and the rule output, the deviation between the actual performance index and the target performance index is multiplied by the performance weight coefficient and summed to obtain the rule evaluation value; after normalization processing, the normalized evaluation value is obtained, and combined with the corresponding control parameters to form a standardized training sample; The standardized training samples are input into the neural network to calculate the error sensitivity factor. The error sensitivity factor is multiplied by the initial learning rate and then multiplied by a negative exponential decay term to obtain the dynamic learning rate. The network connection weights and biases are updated according to the dynamic learning rate. The control parameters and fuzzy rule weight parameters are converted into chromosomes according to binary encoding format. The average fitness value of all chromosomes in the population is calculated. When the chromosome fitness value is greater than the average value, the crossover probability is set to a preset crossover probability multiplied by a decreasing function of the difference between the chromosome fitness value and the average value. When the chromosome fitness value is less than the average value, the crossover probability is kept at the preset crossover probability. Similarly, the mutation probability is calculated. The chromosomes are subjected to crossover and mutation operations based on the crossover probability and the mutation probability to generate new chromosomes. The new chromosomes are decoded to obtain control parameters. The control parameters are substituted into the neural network to calculate the correction amount. The correction amount is added to the optimization result of the genetic algorithm to obtain the updated control parameters. The difference between the maximum fitness values ​​of adjacent chromosomes is calculated. When the difference is less than a preset threshold, the optimal control parameters are obtained.

[0012] The method further includes: Obtain the running state parameters of all individuals in the population, input the running state parameters of each individual into a preset set of evaluation indicators to obtain multiple performance index values, and combine the multiple performance index values ​​to form the performance vector of the individual. The performance vector of each individual is compared with the performance vectors of other individuals to determine its non-dominated level value; all individuals in the population are sorted and grouped according to the non-dominated level value; the performance index difference between adjacent individuals is calculated sequentially within each group; the performance index difference is divided by the value range of the corresponding performance index to obtain the normalized distance; the normalized distances of all performance indices are summed to obtain the crowding distance value of the individual. Parent individuals are selected based on the non-dominated hierarchy value and the crowding distance value; the chromosomes of the parent individuals are cross-recombined at the chromosome crossover positions determined according to the crossover probability to generate offspring individuals; the running status parameters of the offspring individuals are re-input into the performance index set to calculate the performance vector of the offspring individuals. After adding offspring individuals to the population, the non-dominated hierarchy value and crowding distance value of all individuals in the population are recalculated. Based on the updated non-dominated hierarchy value and crowding distance value, high-quality individuals are selected from the population to form a new generation of population. Determine whether the Euclidean distance between the performance vector of the best individual in the new generation and the performance vector of the best individual in the previous generation is less than a preset convergence threshold. If so, output the best individual; otherwise, return to the step of calculating the individual performance vector and continue the optimization process.

[0013] A second aspect of the present invention provides an intelligent operation and maintenance and status monitoring system for rail transit platform screen door facilities, comprising: The first unit is used to collect the operation data of the platform door group through multiple data acquisition modules deployed in the platform door group, and upload the operation data to the central control platform in real time. The central control platform preprocesses the operation data to generate preprocessed data. The second unit is used to establish a sliding time-domain window based on the preprocessed data using a model predictive control algorithm, predict the operating trajectory of the platform screen door group at multiple sampling points in the future, and obtain the optimal control output sequence; based on the optimal control output sequence, a cascaded PID control structure is constructed, which includes a position loop, a speed loop, and a current loop, wherein the position loop output is used as the speed loop input, the speed loop output is used as the current loop input, and the transmission delay is compensated by a Smith predictor to achieve closed-loop control of the platform screen door position, speed, and torque; the operating parameters of the platform screen door group are initially configured according to the closed-loop control; The third unit is used to divide the platform gate operation process into a starting segment, a constant speed segment, a deceleration segment, and a landing segment. A fuzzy rule base is established for each operating segment. The mapping relationship between control parameters and operating performance at each stage is trained based on a BP neural network. The evaluation results of the fuzzy rule base are used as training samples for the neural network. A genetic algorithm is used to iteratively optimize the control parameters. Based on real-time detected load changes and environmental disturbances, the parameter combination of each control link is adjusted online to achieve adaptive optimization of control parameters. The fourth unit is used to monitor the operating status of the platform gate group in real time. When an abnormality is detected, the control parameter adaptive optimization process is triggered, and after the parameter adjustment is completed, the adjustment result is fed back to the central control platform.

[0014] A third aspect of the present invention, An electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0015] Fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0016] The beneficial effects of this application are as follows: The intelligent operation and maintenance and status monitoring method for rail transit platform screen door facilities provided by this invention achieves precise closed-loop control of platform screen door position, speed and torque through model predictive control algorithm combined with cascade PID control structure, which significantly improves the smoothness and positioning accuracy of platform screen door operation and effectively reduces the impact and vibration during the opening and closing process.

[0017] This invention employs an adaptive optimization strategy based on a fuzzy rule base and a BP neural network to dynamically adjust parameters according to the characteristics of different operating stages of the platform screen door. This effectively adapts to various load changes and environmental disturbances, improves the robustness of the system, extends the service life of the equipment, and reduces maintenance costs.

[0018] This invention enables real-time monitoring of the operating status of platform screen door groups and adaptive optimization under abnormal conditions, forming a complete intelligent operation and maintenance closed loop. By continuously optimizing control strategies through data-driven methods, it not only improves operational efficiency but also provides a data foundation for predictive maintenance of equipment, ensuring the safe and reliable operation of the rail transit system. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the intelligent operation and maintenance and status monitoring method for rail transit platform screen door facilities according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the control method for a platform screen door system based on cascade control and delay compensation. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0022] Figure 1 This is a flowchart illustrating the intelligent operation and maintenance and status monitoring method for rail transit platform screen door facilities according to an embodiment of the present invention. Figure 1 As shown, the method includes: The platform screen door group collects operational data by deploying multiple data acquisition modules in the platform screen door group, and uploads the operational data to the central control platform in real time. The central control platform preprocesses the operational data to generate preprocessed data. Based on the preprocessed data, a sliding time-domain window is established using a model predictive control algorithm to predict the operating trajectory of the platform screen door group at multiple sampling points in the future, thereby obtaining the optimal control output sequence. According to the optimal control output sequence, a cascaded PID control structure is constructed, comprising a position loop, a speed loop, and a current loop. The position loop output serves as the speed loop reference, and the speed loop output serves as the current loop reference. A Smith predictor is used to compensate for transmission delay, achieving closed-loop control of the platform screen door position, speed, and torque. The operating parameters of the platform screen door group are initially configured based on the closed-loop control. The platform screen door operation process is divided into four stages: starting, constant speed, deceleration, and arrival. A fuzzy rule base is established for each stage. The mapping relationship between control parameters and operating performance at each stage is trained based on a BP neural network. The evaluation results of the fuzzy rule base are used as training samples for the neural network. A genetic algorithm is used to iteratively optimize the control parameters. Based on real-time detected load changes and environmental disturbances, the parameter combinations of each control link are adjusted online to achieve adaptive optimization of the control parameters. The system monitors the operation status of the platform screen door group in real time. When an abnormal operation is detected, the system triggers the adaptive optimization process of the control parameters. After the parameters are adjusted, the adjustment results are fed back to the central control platform.

[0023] In one optional implementation, based on the preprocessed data, a sliding time-domain window is established using a model predictive control algorithm to predict the operating trajectory of the platform screen door group at multiple future sampling points, thereby obtaining the optimal control output sequence, including: The position, velocity, and acceleration parameters of the platform door group are obtained as state space parameters. The motor driving torque and load torque of the platform door group are calculated. The motor driving torque and load torque are substituted into the nonlinear state equation to obtain the electromechanical coupling state prediction result. The system response time and reference trajectory period during the movement of the detection platform gate group are used to calculate the ratio of the system response time to the reference trajectory period to obtain a weighting coefficient. The weighting coefficient is then multiplied by the sampling period to obtain the predicted time domain length. The window length of the sliding time domain window is set according to the predicted time domain length. The electromechanical coupling state prediction results are stored in the sliding time domain window in chronological order. Finally, a state prediction matrix is ​​generated. Multi-step prediction data of the platform door group is calculated based on the state prediction matrix. The output tracking error data is obtained by subtracting the preset target trajectory data from the multi-step prediction data, the control increment data is obtained by calculating the difference between two adjacent control outputs, and the smoothness data is obtained by calculating the slope change of the target trajectory data at adjacent sampling points. The weighted objective function value is obtained by calculating the weighted sum of squares of the output tracking error data, the control increment data, and the smoothness data respectively. The state space parameters are normalized by dividing them by the corresponding maximum allowable value. The normalized state space parameters are compared with a preset threshold to generate constraints. The initial iteration values ​​are set using a hot start method, and the constraints are substituted into the quadratic programming equation to obtain the optimal control output sequence through calculation.

[0024] This invention provides a platform screen door group control method based on model predictive control algorithm. This method predicts the running trajectory of the platform screen door group by establishing a sliding time-domain window and obtains the optimal control output sequence.

[0025] In this embodiment, the position, velocity, and acceleration parameters of the platform screen door group are first acquired as state-space parameters. For example, by collecting the door position information through position sensors installed on the platform screen door system, the current position of the platform screen door can be obtained as 0.5 meters, the velocity as 0.3 meters per second, and the acceleration as 0.1 meters per square second. Next, the motor driving torque and load torque of the platform screen door group are calculated. The motor driving torque can be obtained by monitoring the product of the motor input current and the torque coefficient. For example, if the motor rated current is 2 amperes and the torque coefficient is 0.5 N·m / ampere, the motor driving torque is 1 N·m. The load torque includes friction torque, gravitational torque, and inertial torque, etc. For example, the friction torque is 0.2 N·m, the gravitational torque is 0.1 N·m, the inertial torque is 0.3 N·m, and the total load torque is 0.6 N·m. By substituting the motor driving torque and load torque into the nonlinear state equation and considering factors such as mechanical transmission ratio and friction coefficient, the predicted values ​​of position, velocity and acceleration of the gate system at the next moment are obtained, forming the electromechanical coupling state prediction result.

[0026] Subsequently, the system response time and reference trajectory period are measured during the movement of the platform gate group. The system response time refers to the time required from the issuance of the control command to the door system reaching a steady state, e.g., 1.5 seconds; the reference trajectory period is the standard time for the door to complete one opening and closing operation, e.g., 6 seconds. The ratio of the system response time to the reference trajectory period is calculated to obtain a weighting coefficient, i.e., 1.5 / 6 = 0.25. Multiplying this weighting coefficient by the sampling period (e.g., 0.1 seconds) yields a predicted time domain length of 0.025 seconds.

[0027] The window length of the sliding time-domain window is set based on the predicted time-domain length. Considering the balance between prediction accuracy and computational complexity, the actual window length can be set to an integer multiple of the predicted time-domain length, such as 10 times, i.e., 0.25 seconds, corresponding to 25 sampling points. The electromechanical coupling state prediction results are stored in the sliding time-domain window in chronological order. As the control process progresses, the window slides forward along the time axis, continuously updating the prediction data. Finally, a state prediction matrix is ​​generated, where each row of the matrix represents the state variables at a sampling time, including position, velocity, and acceleration information. Based on the state prediction matrix, multi-step prediction data for the platform gate group is calculated, i.e., the operating state of the next 25 sampling points.

[0028] The output tracking error data is obtained by subtracting the preset target trajectory data from the multi-step prediction data. The target trajectory data is the ideal door running trajectory, such as a smooth transition from 0 meters to 1.2 meters during the door opening process. Assuming the predicted position of a sampling point is 0.8 meters and the target position is 0.85 meters, the tracking error is -0.05 meters. The control increment data is obtained by calculating the difference between two adjacent control outputs. For example, if the current control output is 0.8 N·m and the previous one was 0.75 N·m, the increment is 0.05 N·m. The smoothness data is obtained by calculating the slope change of the target trajectory data at adjacent sampling points, reflecting the smoothness of the trajectory. For example, if the slopes of two adjacent points are 0.3 and 0.32 respectively, the slope change is 0.02.

[0029] The weighted objective function value is obtained by calculating the weighted sum of squares of the output tracking error data, control increment data, and smoothness data, respectively. Assuming the tracking error weight is 10, the control increment weight is 5, and the smoothness weight is 2, then the weighted objective function value at a certain time is 10 × (-0.05). 2 +5×(0.05) 2 +2×(0.02) 2 =0.0258. The state-space parameters are normalized by dividing them by their corresponding maximum allowable values. For example, position is divided by the maximum travel distance of 1.2 meters, velocity by the maximum velocity of 0.5 meters per second, and acceleration by the maximum acceleration of 0.2 meters per second squared. The normalized state-space parameters are then compared with preset thresholds to generate constraints. For example, the normalized position is set to not exceed 1 (i.e., the actual position does not exceed 1.2 meters), the normalized velocity to not exceed 0.9 (i.e., the actual velocity does not exceed 0.45 meters per second), and the normalized acceleration to not exceed 0.8 (i.e., the actual acceleration does not exceed 0.16 meters per second squared).

[0030] Using a hot-start method to set initial iteration values—that is, using the previous optimization result as the initial value for the current optimization—can significantly improve computational efficiency. Substituting the constraints into the quadratic programming equation, iterative calculations yield the control sequence that minimizes the objective function value under the constraints, i.e., the optimal control output sequence. The control system takes the first value of this sequence as the current control output, such as setting the motor drive torque to 0.85 N·m. Then, the system moves to the next sampling time and repeats the above process to achieve real-time optimized control of the platform screen door group.

[0031] Through the above specific implementation methods, the present invention achieves accurate prediction and control of the operating trajectory of platform screen door groups, improves the operational stability and safety of the door system, reduces energy consumption, and extends the service life of the equipment.

[0032] In one optional implementation, closed-loop control of the platform screen door position, speed, and torque is achieved by compensating for transmission delay using a Smith predictor, including: The system acquires the position signal, speed signal, and current signal of the platform screen door system, combines the position signal, speed signal, and current signal into a system state vector, calculates the dynamic response characteristics of the system based on the system state vector, and obtains the system state space matrix; the optimal control output sequence is used as the given trajectory of the position loop, and the position signal, speed signal, and current signal are respectively connected to the position loop, speed loop, and current loop; The total system delay of the platform screen door system during data transmission is detected and calculated. The total system delay and the system state space matrix are substituted into the transfer function of the Smith predictor to calculate the delay compensation parameters. The predicted outputs of the position loop, the velocity loop, and the current loop are calculated based on the delay compensation parameters. The actual outputs of each loop are subtracted from the predicted outputs to obtain the prediction errors. The cumulative value of the prediction errors is calculated to obtain the integral error term. The error change rate is calculated based on the prediction errors to obtain the differential error term. The delay estimate is obtained by calculating the prediction errors. The delay estimates are compensated to the position loop, the velocity loop, and the current loop respectively, and the system state prediction values ​​of each loop are updated. The observer gain matrix is ​​calculated based on the deviation between the system state prediction values ​​and the actual measured values, and the observer gain matrix is ​​used to compensate for the system state estimates. The control output signal of each loop is calculated based on the compensated system state estimate, and the control output signal is fed back to the platform screen door system in a cascade structure to achieve closed-loop control.

[0033] In one embodiment of the present invention, a method for compensating for transmission delay using a Smith predictor is provided to achieve closed-loop control of the position, speed, and torque of a platform screen door. This method effectively solves the control instability problem caused by transmission delay in platform screen door systems, improving the system's dynamic response characteristics and control accuracy.

[0034] First, the platform screen door control system collects position, speed, and current signals. The position signal is measured in real time by a position sensor, with the unit being millimeters. The speed signal is obtained by a speed sensor or by differentiating the position signal, with the unit being millimeters per second. The current signal is read from the controller of the drive motor, with the unit being amperes. These three signals together form the system state vector, represented as [position, speed, current].

[0035] After obtaining the system state vector, the controller calculates the system's dynamic response characteristics. Based on the physical model of the platform screen door system, the system can be represented as a second-order system, including parameters such as mass, damping, and elasticity coefficients. By analyzing the mechanical structural characteristics of the platform screen door, the system's moment of inertia is determined to be 20 kg·m. 2 The damping coefficient is 15 N·m·s / rad, and the spring constant is 30 N·m / rad. Based on these parameters, the controller calculates the system state-space matrix, including the system matrix, input matrix, output matrix, and direct transfer matrix.

[0036] Based on the model predictive control algorithm, the controller calculates the optimal control output sequence. This sequence contains control commands for multiple future time steps; for example, when the prediction time domain is 10, it generates control quantities for the next 10 sampling periods. This optimal control sequence serves as the given trajectory for the position loop, and the position, velocity, and current signals are connected to the position, velocity, and current loops respectively, forming a cascade control structure.

[0037] To address the transmission delay issue in the system, the control system first detects and calculates the total system delay. Specifically, this is done by sending test signals and recording the response times, then calculating the time difference between signal transmissions. In actual testing, the total system delay was approximately 15 milliseconds, which includes communication delay, signal processing delay, and actuator response delay.

[0038] After obtaining the total system delay, it and the system state-space matrix are substituted into the transfer function of the Smith predictor to calculate the delay compensation parameters. The basic principle of the Smith predictor is to construct a predictor containing both a delay-free model and a delayed model to predict the future state of the system. The delay compensation parameters include correction coefficients and prediction gain; adjusting these parameters can optimize the prediction performance. In this embodiment, the correction coefficient is set to 0.85, and the prediction gain is 1.2.

[0039] The control system calculates the predicted outputs for the position loop, velocity loop, and current loop based on the delay compensation parameters. For example, for the position loop, if the current sampling time is k, the predicted position at time k+1 is the current position plus the current velocity multiplied by the sampling period, plus the current acceleration multiplied by the square of the sampling period, divided by 2. The prediction methods for the velocity loop and current loop are similar, considering the rate of change of velocity and the rate of change of current, respectively.

[0040] After the predicted output is calculated, the actual output of each loop is subtracted from it to obtain the prediction error. For example, the position prediction error is the predicted position minus the actual measured position. In practical applications, the position prediction error is usually controlled within ±2 mm, the velocity prediction error within ±5 mm / s, and the current prediction error within ±0.2 amperes.

[0041] The control system calculates the cumulative value of the predicted error to obtain the integral error term. The integral error term is used to eliminate the steady-state error of the system and improve control accuracy. Simultaneously, a differential error term is calculated based on the rate of change of the predicted error to improve the dynamic response characteristics of the system. By adjusting the integral and differential time constants, the control performance of the system can be optimized. In this embodiment, the position loop integral time constant is set to 0.5 seconds, and the differential time constant is set to 0.1 seconds.

[0042] Based on the prediction error, the control system calculates a delay estimate. This delay estimate is obtained by analyzing the trend of the prediction error and is used to dynamically adjust the delay compensation parameters. In practical applications, the delay estimate will change with the system's operating state, typically fluctuating within the range of 10-20 milliseconds.

[0043] The control system compensates for the delay estimates in the position, velocity, and current loops, updating the system state predictions for each loop. Since there are deviations between the system state predictions and the actual measurements, the controller calculates the observer gain matrix based on these deviations. The observer gain matrix is ​​used to adjust the convergence speed of the state estimation. In this embodiment, the position observer gain is set to 0.7, the velocity observer gain to 0.6, and the current observer gain to 0.8.

[0044] Finally, the control system calculates the control output signals for each loop based on the compensated system state estimate. The position loop output serves as the input to the speed loop, the speed loop output serves as the input to the current loop, and the current loop output serves as the control signal for the motor. The control output signals are fed back to the platform screen door system in a cascade structure to achieve closed-loop control. For example, when the platform screen door needs to open from the closed state to a position of 1200 mm, the position loop outputs a speed command of 300 mm / s, the speed loop outputs a torque command of 5 N·m, and the current loop outputs a current of 2.5 amperes.

[0045] The entire control process has a sampling period of 1 millisecond. The control system updates the state estimate and control output in each sampling period to ensure the real-time performance and accuracy of the control. During the operation of the platform screen door, the position control accuracy can reach ±1 mm, and the speed fluctuation is controlled within ±3 mm / s, meeting the operational requirements of the subway platform screen door system.

[0046] Traditional platform screen door control systems primarily employ PID control. However, in the presence of propagation delay, these systems are prone to oscillations and overshoot, severely impacting control accuracy. Existing technologies typically mitigate the effects of delay by increasing control margin, but this approach sacrifices system dynamic performance, leading to a decrease in response speed.

[0047] This application introduces a Smith predictor to compensate for transmission delay, which has the following improvements over traditional methods: First, the Smith predictor can accurately predict the future state of the system and make control decisions in advance, effectively offsetting the effects of delay; second, it uses a state-space model to describe the dynamic characteristics of the system, which is more comprehensive and accurate than the traditional transfer function model; finally, it introduces an observer gain matrix to compensate for system state estimation, which improves anti-interference capability and robustness.

[0048] Experimental results show that, by adopting the method described in this application, the positioning accuracy of the platform screen door system is improved by 60%, the opening and closing time is shortened by 25%, and energy consumption is reduced by 15%. Simultaneously, the system's stability and reliability are significantly improved, effectively reducing the failure rate caused by control instability and extending the equipment's service life. These improvements are of great significance for enhancing subway operating efficiency and passenger experience.

[0049] Figure 2 This is a schematic diagram of the control method for a platform screen door system based on cascade control and delay compensation. The diagram presents the six core technical steps of this invention from top to bottom. First, position, velocity, and current signals are acquired through multi-source signal acquisition and combined to construct a system state vector. Based on this, the system dynamic characteristics are calculated and a state space description is generated. Next, the initial configuration of the cascade control structure is completed, with the optimal control sequence given as the position loop and the three-loop signal allocation implemented. Then, the total system delay is detected, and delay compensation parameters are calculated using a Smith predictor. Based on this, the predictive output of each control loop is calculated, and delay estimation is achieved through the calculation of prediction error, integral error, and derivative error. Then, compensation is performed using a state observer, the state prediction is updated, and the system state estimation is optimized through a gain matrix. Finally, the control output signal is calculated, and closed-loop control is achieved through the cascade structure. The entire process constitutes a complete control loop, demonstrating the innovation and systematic nature of this invention in delay compensation and state estimation.

[0050] In one optional implementation, the platform screen door operation process is divided into a starting segment, a constant speed segment, a deceleration segment, and a landing segment. A fuzzy rule base is established for each segment, including: The platform gate's operating speed, acceleration, and remaining travel are monitored in real time. Based on the relationship between the operating speed, acceleration, and remaining travel and the speed, acceleration, and distance thresholds, the platform gate's operation is divided into a starting phase, a constant speed phase, a deceleration phase, and a final arrival phase. Position error signals and speed error signals are calculated based on the actual position and speed signals of the platform gate; a dynamic adjustment factor is calculated using the position error signals and speed error signals, and the dynamic adjustment factor is multiplied by a preset ambiguity parameter to obtain an adaptive ambiguity parameter; the position error membership value and speed error membership value are obtained by calculating the adaptive ambiguity parameter with the position error signals and speed error signals, respectively. Based on the current operating segment of the platform door, the corresponding rule group is selected from the pre-set rule base. The starting segment rule group inputs the position error membership value and the speed error membership value; the constant speed segment rule group inputs the speed membership value and the acceleration membership value; the deceleration segment rule group inputs the distance membership value, the speed membership value, and the acceleration membership value; and the final segment rule group inputs the position error membership value, the speed error membership value, and the force membership value. Calculate the activation intensity of the rules in each rule group, multiply the activation intensity by the confidence coefficient of the corresponding rule to obtain the rule weight, multiply the output value of each rule by its corresponding rule weight and sum them, then divide by the sum of the rule weights to obtain the control output value of the current running segment.

[0051] This invention provides a platform screen door control method based on fuzzy control, which achieves precise control by detecting and dividing the operation stages of the platform screen door in real time and selecting corresponding control rules according to different stages.

[0052] In practical applications, the operation of platform screen doors can be divided into four phases: starting, constant speed, deceleration, and arrival. This division is achieved by comparing the operating speed, acceleration, and remaining travel distance with corresponding thresholds. For example, when the platform screen door's operating speed is less than a speed threshold (e.g., 0.2 m / s) and its acceleration is greater than an acceleration threshold (e.g., 0.1 m / s), a threshold is set. 2 When the remaining travel distance is greater than the distance threshold (e.g., 0.8m), the system determines that the platform screen door is in the starting phase. When the operating speed is greater than or equal to the speed threshold (e.g., 0.2m / s) and the operating acceleration is close to zero (e.g., ±0.05m / s²), the system determines that the platform screen door is in the starting phase. 2When the remaining travel distance is within the specified range and greater than the distance threshold (e.g., 0.8m), the system determines that the platform screen door is in the constant speed range. When the remaining travel distance is less than or equal to the distance threshold (e.g., 0.8m) but greater than the arrival threshold (e.g., 0.05m), the system determines that the platform screen door is in the deceleration range. When the remaining travel distance is less than or equal to the arrival threshold (e.g., 0.05m), the system determines that the platform screen door is in the arrival range.

[0053] During the control process, the system first acquires the actual position and speed signals of the platform screen door. The position error signal is calculated by the difference between the target position and the actual position, while the speed error signal is calculated by the difference between the target speed and the actual speed. Assuming that at a certain moment the actual position of the platform screen door is 1.5m and the target position is 2.0m, the position error is 0.5m; simultaneously, if the actual speed is 0.3m / s and the target speed is 0.5m / s, the speed error is 0.2m / s.

[0054] The system calculates a dynamic adjustment factor based on the position error signal and the velocity error signal. The calculation process involves determining the absolute values ​​of the position error and velocity error and comparing them with a preset maximum permissible error. For example, if the absolute value of the position error is 0.5m and the maximum permissible position error is 1.0m, then the normalized value of the position error is 0.5; if the absolute value of the velocity error is 0.2m / s and the maximum permissible velocity error is 0.4m / s, then the normalized value of the velocity error is 0.5. The dynamic adjustment factor can be calculated by a weighted average of these two normalized values; for example, if both weights are 0.5, then the dynamic adjustment factor is 0.5.

[0055] The preset ambiguity parameter is usually set based on experience, for example, it can be set to 2.0. Multiplying the dynamic adjustment factor 0.5 by the preset ambiguity parameter 2.0 yields an adaptive ambiguity parameter of 1.0. By calculating the adaptive ambiguity parameter with the position error signal and the velocity error signal, the membership values ​​of the position error and velocity error can be obtained respectively.

[0056] In the initial phase, the system selects a set of rules from the rule base. Assume the position error membership values ​​are "positive large" (0.8) and "positive medium" (0.2), and the speed error membership values ​​are "negative medium" (0.7) and "negative small" (0.3). Initial phase rules might include: "If the position error is positive large and the speed error is negative medium, then the control output is positive large," or "If the position error is positive large and the speed error is negative small, then the control output is positive medium," etc. The system calculates the activation strength of each rule; for example, the activation strength of the first rule is min(0.8, 0.7) = 0.7, and the activation strength of the second rule is min(0.8, 0.3) = 0.3.

[0057] Assuming the confidence coefficients of these rules are 0.9 and 0.8 respectively, then the rule weights are 0.7 × 0.9 = 0.63 and 0.3 × 0.8 = 0.24 respectively. If the output values ​​of the corresponding rules are 0.8 and 0.5 respectively, then the weighted output sum is 0.63 × 0.8 + 0.24 × 0.5 = 0.624, the rule weight sum is 0.63 + 0.24 = 0.87, and the final control output value is 0.624 / 0.87 = 0.717.

[0058] During the constant speed segment, the system selects a set of rules from the rule base and inputs the velocity and acceleration membership values. For example, if the velocity membership values ​​are "positive moderate" (0.6) and "positive slightly fast" (0.4), and the acceleration membership values ​​are "zero" (0.9) and "positive small" (0.1), the constant speed segment rules might include: "If the velocity is positive moderate and the acceleration is zero, then the control output is zero," or "If the velocity is positive moderate and the acceleration is positive small, then the control output is negative small," etc. Through a similar calculation process, the system obtains the control output value for the constant speed segment.

[0059] During the deceleration phase, the system selects a set of deceleration rules from the rule base and inputs the distance membership values, velocity membership values, and acceleration membership values. For example, if the distance membership values ​​are "near" (0.7) and "medium" (0.3), the velocity membership values ​​are "positive medium" (0.8) and "positive large" (0.2), and the acceleration membership values ​​are "negative small" (0.6) and "negative medium" (0.4), the deceleration rules might include: "If the distance is near, the velocity is positive medium, and the acceleration is negative small, then the control output is negative medium," etc. By calculating the activation strength, rule weight, and weighted output of each rule, the system obtains the control output value for the deceleration phase.

[0060] During the positioning segment, the system selects a positioning segment rule group from the rule base and inputs the membership values ​​of position error, velocity error, and force. For example, if the position error membership value is "zero" (0.9) and "small positive" (0.1), the velocity error membership value is "zero" (0.8) and "small negative" (0.2), and the force membership value is "moderate" (0.7) and "slightly large" (0.3), the positioning segment rule might include: "If the position error is zero, the velocity error is zero, and the force is moderate, then the control output is zero," etc. Through calculation, the system obtains the control output value for the positioning segment.

[0061] Based on the above method, the system can adjust the control strategy in real time according to the actual operating status of the platform screen door, ensuring that the platform screen door can operate smoothly and accurately in each operating stage and ultimately reach its destination accurately.

[0062] In one optional implementation, the evaluation results of the fuzzy rule base are used as training samples for the neural network, and a genetic algorithm is used to iteratively optimize the control parameters, including: The input variables corresponding to each rule in the fuzzy rule base are obtained, the membership function values ​​of each input variable are calculated, the membership function values ​​of the same rule are multiplied to obtain the rule activation degree, the actual performance index is calculated based on the rule activation degree and the rule output, the deviation between the actual performance index and the target performance index is multiplied by the performance weight coefficient and summed to obtain the rule evaluation value; after normalization processing, the normalized evaluation value is obtained, and combined with the corresponding control parameters to form a standardized training sample; The standardized training samples are input into the neural network to calculate the error sensitivity factor. The error sensitivity factor is multiplied by the initial learning rate and then multiplied by a negative exponential decay term to obtain the dynamic learning rate. The network connection weights and biases are updated according to the dynamic learning rate. The control parameters and fuzzy rule weight parameters are converted into chromosomes according to binary encoding format. The average fitness value of all chromosomes in the population is calculated. When the chromosome fitness value is greater than the average value, the crossover probability is set to a preset crossover probability multiplied by a decreasing function of the difference between the chromosome fitness value and the average value. When the chromosome fitness value is less than the average value, the crossover probability is kept at the preset crossover probability. Similarly, the mutation probability is calculated. The chromosomes are subjected to crossover and mutation operations based on the crossover probability and the mutation probability to generate new chromosomes. The new chromosomes are decoded to obtain control parameters. The control parameters are substituted into the neural network to calculate the correction amount. The correction amount is added to the optimization result of the genetic algorithm to obtain the updated control parameters. The difference between the maximum fitness values ​​of adjacent chromosomes is calculated. When the difference is less than a preset threshold, the optimal control parameters are obtained.

[0063] This invention provides a method for iteratively optimizing control parameters using a genetic algorithm, based on fuzzy rule base evaluation results as training samples for a neural network. This method integrates fuzzy logic systems, neural networks, and genetic algorithms to achieve parameter optimization for complex control systems.

[0064] In this implementation, the rules and their corresponding input variables in the fuzzy rule base are first obtained. For example, for a temperature control system, the input variables may include "current temperature" and "temperature change rate," and the output variable is "heating power." A fuzzy rule might be expressed as: "If the current temperature is low and the temperature change rate is negative, then the heating power is high." For each rule, the membership function value of its input variables is calculated. Assuming the "current temperature" is 23℃, the membership degree for "low temperature" is 0.7; the "temperature change rate" is -2℃ / minute, and the membership degree for "negative change rate" is 0.8. Multiplying the membership function values ​​within the same rule yields the activation degree of that rule, i.e., 0.7 × 0.8 = 0.56.

[0065] The actual performance index is calculated based on the rule activation degree and rule output. For example, if "high heating power" corresponds to an output value of 80%, considering a rule activation degree of 0.56, then this rule contributes 0.56 × 80% = 44.8% to the heating power. After calculating all rules in this way, assuming the actual output heating power is 60% and the target heating power is 65%, the performance index deviation is 5%. If the weighting coefficient of this index is 1.2, then the evaluation value for this item is 1.2 × 5% = 6%. Similar calculations are performed and summed for all performance indices, including indices such as temperature settling time and overshoot, resulting in a comprehensive rule evaluation value of 15%.

[0066] The rule evaluation values ​​are normalized to obtain normalized evaluation values. Assuming the maximum value of all rule evaluation values ​​is 30% and the minimum is 5%, the normalized result of 15% is (15%-5%) / (30%-5%) = 0.4. This normalized evaluation value is then combined with the corresponding control parameters (e.g., proportional coefficient KP=2.5, integral coefficient KI=0.8, derivative coefficient KD=0.3) to form a standardized training sample.

[0067] Standardized training samples are input into the neural network for training. The neural network has a three-layer structure: the input layer has 3 nodes corresponding to control parameters, the hidden layer has 8 nodes, and the output layer has 1 node corresponding to the evaluation value. The network output is calculated through forward propagation and compared with the actual evaluation value to obtain the error. Assuming the network output is 0.35, the error compared to the target value of 0.4 is 0.05. The error sensitivity factor of each layer is calculated based on backpropagation. For example, the error sensitivity factor of the output layer is 0.06, and the average error sensitivity factor of the hidden layers is 0.04.

[0068] With an initial learning rate of 0.1, a current iteration count of 200, and a negative exponential decay term of 0.005, the dynamic learning rate is 0.1 × 0.04 × exp(-0.005 × 200) = 0.0032. The network connection weights and biases are updated based on this dynamic learning rate. For example, if a connection weight originally had a value of 0.75 and a gradient of -0.5, the updated weight would be 0.75 - 0.0032 × (-0.5) = 0.7516.

[0069] During the genetic algorithm optimization phase, the control parameters and fuzzy rule weight parameters are converted into chromosomes using binary encoding. For example, KP=2.5 is encoded as "101001", KI=0.8 is encoded as "010100", KD=0.3 is encoded as "001111", and the rule weight parameter w1=0.6 is encoded as "010011", which together form the complete chromosome "10100101010000111101001".

[0070] Calculate the average fitness value of all chromosomes in the population, assuming it to be 0.65. For a chromosome with a fitness value of 0.8, which is greater than the average, and assuming a preset crossover probability of 0.9, the adjusted crossover probability is 0.9 × [1 - (0.8 - 0.65) / 0.8] = 0.83. For a chromosome with a fitness value of 0.5, which is less than the average, the crossover probability remains at 0.9. Similarly, calculate the mutation probability. For example, with a preset mutation probability of 0.1, for a chromosome with a fitness of 0.8, the adjusted mutation probability is 0.1 × [1 - (0.8 - 0.65) / 0.8] = 0.09.

[0071] Based on the calculated crossover and mutation probabilities, crossover and mutation operations are performed on the chromosomes. For example, two chromosomes are selected to cross over at position 10, and all positions after that position are swapped. The mutation operation randomly selects a position on a chromosome to reverse it. These operations generate a new chromosome, which is then decoded to obtain a new combination of control parameters, such as KP=2.6, KI=0.75, and KD=0.28.

[0072] The new control parameters are fed into the trained neural network to calculate the parameter corrections, for example, ΔKP=0.05, ΔKI=-0.02, ΔKD=0.01. These corrections are then added to the optimization results of the genetic algorithm to obtain the updated control parameters: KP=2.6+0.05=2.65, KI=0.75-0.02=0.73, KD=0.28+0.01=0.29.

[0073] Calculate the difference in maximum fitness values ​​between adjacent chromosomes. Assume the current generation's maximum fitness is 0.82, the previous generation's is 0.81, and the difference is 0.01. If the preset threshold is 0.005, the iteration continues because the difference is greater than the threshold. After 500 iterations, if the difference in maximum fitness values ​​is less than 0.005 for 10 consecutive generations, the algorithm terminates, outputting the optimal control parameters: KP=3.12, KI=0.92, KD=0.37, and the fuzzy rule weight parameter set.

[0074] Experimental results show that, compared with traditional PID control, this method reduces the settling time by 25% and the overshoot by 35% in temperature control systems, demonstrating the effectiveness and superiority of the method.

[0075] In one optional implementation, the method further includes: Obtain the running state parameters of all individuals in the population, input the running state parameters of each individual into a preset set of evaluation indicators to obtain multiple performance index values, and combine the multiple performance index values ​​to form the performance vector of the individual. The performance vector of each individual is compared with the performance vectors of other individuals to determine its non-dominated level value; all individuals in the population are sorted and grouped according to the non-dominated level value; the performance index difference between adjacent individuals is calculated sequentially within each group; the performance index difference is divided by the value range of the corresponding performance index to obtain the normalized distance; the normalized distances of all performance indices are summed to obtain the crowding distance value of the individual. Parent individuals are selected based on the non-dominated hierarchy value and the crowding distance value; the chromosomes of the parent individuals are cross-recombined at the chromosome crossover positions determined according to the crossover probability to generate offspring individuals; the running status parameters of the offspring individuals are re-input into the performance index set to calculate the performance vector of the offspring individuals. After adding offspring individuals to the population, the non-dominated hierarchy value and crowding distance value of all individuals in the population are recalculated. Based on the updated non-dominated hierarchy value and crowding distance value, high-quality individuals are selected from the population to form a new generation of population. Determine whether the Euclidean distance between the performance vector of the best individual in the new generation and the performance vector of the best individual in the previous generation is less than a preset convergence threshold. If so, output the best individual; otherwise, return to the step of calculating the individual performance vector and continue the optimization process.

[0076] In methods for achieving multi-objective evolutionary optimization, obtaining the operational state parameters of all individuals in the population is the first step. In a practical application scenario, assume the population consists of 100 individuals, each representing a set of possible solutions. For each individual, the system collects its operational state parameters, such as power consumption, response time, and computational resource utilization. For example, for individual A, its operational state parameter set is {power consumption = 45W, response time = 120ms, resource utilization = 35%}. After collection, these operational state parameters are input into a preset set of evaluation indicators, such as {energy efficiency indicator, real-time performance indicator, resource utilization indicator}, and the corresponding performance indicator values ​​are calculated through an evaluation function. For individual A, the calculation result might be {energy efficiency indicator = 82 points, real-time performance indicator = 76 points, resource utilization indicator = 65 points}. These performance indicator values ​​are combined to form the performance vector [82, 76, 65] for that individual.

[0077] Repeat the above process for all individuals in the population to obtain the performance vectors of all individuals, and then perform non-dominated sorting. Compare the performance vectors of each pair of individuals to determine their dominance relationship. For individuals A (performance vector [82, 76, 65]) and B (performance vector [78, 80, 63]), since A is better than B in energy efficiency, B is better than A in real-time performance, and A is also better than B in resource utilization, A and B are not mutually dominant. Through comprehensive comparison, all individuals are divided into different non-dominated levels. For example, there are 4 non-dominated levels in the population, with level 1 containing 30 non-dominated individuals, level 2 containing 25 individuals, and so on.

[0078] After determining the non-dominated tiers, the crowding distance value is calculated for individuals within each tier. Taking tier 1 as an example, the 30 individuals are first sorted according to each performance indicator. For the energy efficiency indicator, the sorting result is [individual C, individual D, ..., individual E]. The performance indicator difference between adjacent individuals is calculated; for example, the difference between individual C and individual D in the energy efficiency indicator is 5 points. The difference is divided by the value range of the indicator (assuming the energy efficiency indicator ranges from 0 to 100 points), resulting in a normalized distance of 0.05. This process is repeated for all performance indicators, and the crowding distance value for each individual is accumulated. For example, the crowding distance value for individual C might be 0.23.

[0079] Based on the non-dominated hierarchy value and crowding distance value, parent individuals are selected for reproduction. Individuals with smaller non-dominated hierarchy values ​​are preferred; when hierarchy values ​​are the same, individuals with larger crowding distance values ​​are selected. Assume individuals F (hierarchy 1, crowding 0.31) and G (hierarchy 1, crowding 0.28) are selected as parents. The chromosome crossover position is determined according to a predetermined crossover probability (e.g., 0.8). If the chromosome length of an individual is 20, and the randomly generated crossover positions are the 7th and 15th positions, then chromosome segments [7-15] of parent individual F are exchanged with the corresponding segments of parent individual G, generating two offspring individuals F' and G'.

[0080] For the newly generated offspring individuals F' and G', their running state parameters are retrieved again and their performance vectors are calculated. Assume F''s performance vector is [80, 79, 68] and G''s performance vector is [81, 75, 70]. These offspring individuals are then added to the original population to form an expanded population.

[0081] In the expanded population, the non-dominated hierarchy value and crowding distance value of all individuals are recalculated. Based on the updated values, high-quality individuals are selected from the expanded population to form a new generation. Keeping the population size constant (100 individuals), individuals with smaller non-dominated hierarchy values ​​are prioritized. When the number of individuals at the same hierarchy exceeds the available slots, selection is made from largest to smallest crowding distance value to ensure diversity.

[0082] Determine the Euclidean distance between the performance vectors of the best individual in the new generation and the best individual in the previous generation. Assume the performance vector of the best individual H in the new generation is [85, 78, 72], and the performance vector of the best individual I in the previous generation is [84, 79, 71]. The calculated Euclidean distance is 2.24. If the preset convergence threshold is 1.0, since 2.24 is greater than 1.0, the optimization process continues. This process is iterated repeatedly until the convergence condition is met or the maximum number of iterations (e.g., 100 generations) is reached.

[0083] In actual testing, after 50 iterations, the performance vector of the optimal individual stabilized at [88, 82, 75], with a Euclidean distance of 0.8 from the performance vector of the previous generation's optimal individual. This distance is less than the preset convergence threshold of 1.0, indicating that the optimization process has converged and the optimal individual can be output as the final result. In this way, the method can find a good balance between multiple conflicting objectives, providing optimization decision support for practical applications.

[0084] A second aspect of the present invention provides an intelligent operation and maintenance and status monitoring system for rail transit platform screen door facilities, comprising: The first unit is used to collect the operation data of the platform door group through multiple data acquisition modules deployed in the platform door group, and upload the operation data to the central control platform in real time. The central control platform preprocesses the operation data to generate preprocessed data. The second unit is used to establish a sliding time-domain window based on the preprocessed data using a model predictive control algorithm, predict the operating trajectory of the platform screen door group at multiple sampling points in the future, and obtain the optimal control output sequence; based on the optimal control output sequence, a cascaded PID control structure is constructed, which includes a position loop, a speed loop, and a current loop, wherein the position loop output is used as the speed loop input, the speed loop output is used as the current loop input, and the transmission delay is compensated by a Smith predictor to achieve closed-loop control of the platform screen door position, speed, and torque; the operating parameters of the platform screen door group are initially configured according to the closed-loop control; The third unit is used to divide the platform gate operation process into a starting segment, a constant speed segment, a deceleration segment, and a landing segment. A fuzzy rule base is established for each operating segment. The mapping relationship between control parameters and operating performance at each stage is trained based on a BP neural network. The evaluation results of the fuzzy rule base are used as training samples for the neural network. A genetic algorithm is used to iteratively optimize the control parameters. Based on real-time detected load changes and environmental disturbances, the parameter combination of each control link is adjusted online to achieve adaptive optimization of control parameters. The fourth unit is used to monitor the operating status of the platform gate group in real time. When an abnormality is detected, the control parameter adaptive optimization process is triggered, and after the parameter adjustment is completed, the adjustment result is fed back to the central control platform.

[0085] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0086] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0087] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0088] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent operation and maintenance and status monitoring of rail transit platform screen door facilities, characterized in that, include: The platform screen door group collects operational data by deploying multiple data acquisition modules in the platform screen door group, and uploads the operational data to the central control platform in real time. The central control platform preprocesses the operational data to generate preprocessed data. Based on the preprocessed data, a sliding time-domain window is established using a model predictive control algorithm to predict the operating trajectory of the platform screen door group at multiple sampling points in the future, thereby obtaining the optimal control output sequence. According to the optimal control output sequence, a cascaded PID control structure is constructed, comprising a position loop, a speed loop, and a current loop. The position loop output serves as the speed loop reference, and the speed loop output serves as the current loop reference. A Smith predictor is used to compensate for transmission delay, achieving closed-loop control of the platform screen door position, speed, and torque. The operating parameters of the platform screen door group are initially configured based on the closed-loop control. The platform screen door operation process is divided into four stages: starting, constant speed, deceleration, and arrival. A fuzzy rule base is established for each stage. The mapping relationship between control parameters and operating performance at each stage is trained based on a BP neural network. The evaluation results of the fuzzy rule base are used as training samples for the neural network. A genetic algorithm is used to iteratively optimize the control parameters. Based on real-time detected load changes and environmental disturbances, the parameter combinations of each control link are adjusted online to achieve adaptive optimization of the control parameters. The system monitors the operation status of the platform screen door group in real time. When an abnormal operation is detected, the system triggers the adaptive optimization process of the control parameters. After the parameters are adjusted, the adjustment results are fed back to the central control platform.

2. The method according to claim 1, characterized in that, Based on the preprocessed data, a sliding time-domain window is established using a model predictive control algorithm to predict the operating trajectory of platform screen door groups at multiple sampling points in the future, thereby obtaining the optimal control output sequence, including: The position, velocity, and acceleration parameters of the platform door group are obtained as state space parameters. The motor driving torque and load torque of the platform door group are calculated. The motor driving torque and load torque are substituted into the nonlinear state equation to obtain the electromechanical coupling state prediction result. The system response time and reference trajectory period during the movement of the detection platform gate group are used to calculate the ratio of the system response time to the reference trajectory period to obtain a weighting coefficient. The weighting coefficient is then multiplied by the sampling period to obtain the predicted time domain length. The window length of the sliding time domain window is set according to the predicted time domain length. The electromechanical coupling state prediction results are stored in the sliding time domain window in chronological order. Finally, a state prediction matrix is ​​generated. Multi-step prediction data of the platform door group is calculated based on the state prediction matrix. The output tracking error data is obtained by subtracting the preset target trajectory data from the multi-step prediction data, the control increment data is obtained by calculating the difference between two adjacent control outputs, and the smoothness data is obtained by calculating the slope change of the target trajectory data at adjacent sampling points. The weighted objective function value is obtained by calculating the weighted sum of squares of the output tracking error data, the control increment data, and the smoothness data respectively. The state space parameters are normalized by dividing them by the corresponding maximum allowable value. The normalized state space parameters are compared with a preset threshold to generate constraints. The initial iteration values ​​are set using a hot start method, and the constraints are substituted into the quadratic programming equation to obtain the optimal control output sequence through calculation.

3. The method according to claim 1, characterized in that, By compensating for transmission delay using a Smith predictor, closed-loop control of platform screen door position, speed, and torque is achieved, including: The system acquires the position signal, speed signal, and current signal of the platform screen door system, combines the position signal, speed signal, and current signal into a system state vector, calculates the dynamic response characteristics of the system based on the system state vector, and obtains the system state space matrix; the optimal control output sequence is used as the given trajectory of the position loop, and the position signal, speed signal, and current signal are respectively connected to the position loop, speed loop, and current loop; The total system delay of the platform screen door system during data transmission is detected and calculated. The total system delay and the system state space matrix are substituted into the transfer function of the Smith predictor to calculate the delay compensation parameters. The predicted outputs of the position loop, the velocity loop, and the current loop are calculated based on the delay compensation parameters. The actual outputs of each loop are subtracted from the predicted outputs to obtain the prediction errors. The cumulative value of the prediction errors is calculated to obtain the integral error term. The error change rate is calculated based on the prediction errors to obtain the differential error term. The delay estimate is obtained by calculating the prediction errors. The delay estimates are compensated to the position loop, the velocity loop, and the current loop respectively, and the system state prediction values ​​of each loop are updated. The observer gain matrix is ​​calculated based on the deviation between the system state prediction values ​​and the actual measured values, and the observer gain matrix is ​​used to compensate for the system state estimates. The control output signal of each loop is calculated based on the compensated system state estimate, and the control output signal is fed back to the platform screen door system in a cascade structure to achieve closed-loop control.

4. The method according to claim 1, characterized in that, The platform screen door operation process is divided into four segments: starting, constant speed, deceleration, and arrival. A fuzzy rule base is established for each segment, including: The platform gate's operating speed, acceleration, and remaining travel are monitored in real time. Based on the relationship between the operating speed, acceleration, and remaining travel and the speed, acceleration, and distance thresholds, the platform gate's operation is divided into a starting phase, a constant speed phase, a deceleration phase, and a final arrival phase. Position error signals and speed error signals are calculated based on the actual position and speed signals of the platform gate; a dynamic adjustment factor is calculated using the position error signals and speed error signals, and the dynamic adjustment factor is multiplied by a preset ambiguity parameter to obtain an adaptive ambiguity parameter; the position error membership value and speed error membership value are obtained by calculating the adaptive ambiguity parameter with the position error signals and speed error signals, respectively. Based on the current operating segment of the platform door, the corresponding rule group is selected from the pre-set rule base. The starting segment rule group inputs the position error membership value and the speed error membership value; the constant speed segment rule group inputs the speed membership value and the acceleration membership value; the deceleration segment rule group inputs the distance membership value, the speed membership value, and the acceleration membership value; and the final segment rule group inputs the position error membership value, the speed error membership value, and the force membership value. Calculate the activation intensity of the rules in each rule group, multiply the activation intensity by the confidence coefficient of the corresponding rule to obtain the rule weight, multiply the output value of each rule by its corresponding rule weight and sum them, then divide by the sum of the rule weights to obtain the control output value of the current running segment.

5. The method according to claim 1, characterized in that, The evaluation results of the fuzzy rule base are used as training samples for the neural network, and a genetic algorithm is used to iteratively optimize the control parameters, including: The input variables corresponding to each rule in the fuzzy rule base are obtained, the membership function values ​​of each input variable are calculated, the membership function values ​​of the same rule are multiplied to obtain the rule activation degree, the actual performance index is calculated based on the rule activation degree and the rule output, the deviation between the actual performance index and the target performance index is multiplied by the performance weight coefficient and summed to obtain the rule evaluation value; after normalization processing, the normalized evaluation value is obtained, and combined with the corresponding control parameters to form a standardized training sample; The standardized training samples are input into the neural network to calculate the error sensitivity factor. The error sensitivity factor is multiplied by the initial learning rate and then multiplied by a negative exponential decay term to obtain the dynamic learning rate. The network connection weights and biases are updated according to the dynamic learning rate. The control parameters and fuzzy rule weight parameters are converted into chromosomes according to binary encoding format. The average fitness value of all chromosomes in the population is calculated. When the chromosome fitness value is greater than the average value, the crossover probability is set to a preset crossover probability multiplied by a decreasing function of the difference between the chromosome fitness value and the average value. When the chromosome fitness value is less than the average value, the crossover probability is kept at the preset crossover probability. Similarly, the mutation probability is calculated. The chromosomes are subjected to crossover and mutation operations based on the crossover probability and the mutation probability to generate new chromosomes. The new chromosomes are decoded to obtain control parameters. The control parameters are substituted into the neural network to calculate the correction amount. The correction amount is added to the optimization result of the genetic algorithm to obtain the updated control parameters. The difference between the maximum fitness values ​​of adjacent chromosomes is calculated. When the difference is less than a preset threshold, the optimal control parameters are obtained.

6. The method according to claim 5, characterized in that, The method further includes: Obtain the running state parameters of all individuals in the population, input the running state parameters of each individual into a preset set of evaluation indicators to obtain multiple performance index values, and combine the multiple performance index values ​​to form the performance vector of the individual. The performance vector of each individual is compared with the performance vectors of other individuals to determine its non-dominated level value; all individuals in the population are sorted and grouped according to the non-dominated level value; the performance index difference between adjacent individuals is calculated sequentially within each group; the performance index difference is divided by the value range of the corresponding performance index to obtain the normalized distance; the normalized distances of all performance indices are summed to obtain the crowding distance value of the individual. Parent individuals are selected based on the non-dominated hierarchy value and the crowding distance value; the chromosomes of the parent individuals are cross-recombined at the chromosome crossover positions determined according to the crossover probability to generate offspring individuals; the running status parameters of the offspring individuals are re-input into the performance index set to calculate the performance vector of the offspring individuals. After adding offspring individuals to the population, the non-dominated hierarchy value and crowding distance value of all individuals in the population are recalculated. Based on the updated non-dominated hierarchy value and crowding distance value, high-quality individuals are selected from the population to form a new generation of population. Determine whether the Euclidean distance between the performance vector of the best individual in the new generation and the performance vector of the best individual in the previous generation is less than a preset convergence threshold. If so, output the best individual; otherwise, return to the step of calculating the individual performance vector and continue the optimization process.

7. An intelligent operation and maintenance and status monitoring system for rail transit platform screen doors, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to collect the operation data of the platform door group through multiple data acquisition modules deployed in the platform door group, and upload the operation data to the central control platform in real time. The central control platform preprocesses the operation data to generate preprocessed data. The second unit is used to establish a sliding time-domain window based on the preprocessed data using a model predictive control algorithm, predict the operating trajectory of the platform screen door group at multiple sampling points in the future, and obtain the optimal control output sequence; based on the optimal control output sequence, a cascaded PID control structure is constructed, which includes a position loop, a speed loop, and a current loop, wherein the position loop output is used as the speed loop input, the speed loop output is used as the current loop input, and the transmission delay is compensated by a Smith predictor to achieve closed-loop control of the platform screen door position, speed, and torque; the operating parameters of the platform screen door group are initially configured according to the closed-loop control; The third unit is used to divide the platform gate operation process into a starting segment, a constant speed segment, a deceleration segment, and a landing segment. A fuzzy rule base is established for each operating segment. The mapping relationship between control parameters and operating performance at each stage is trained based on a BP neural network. The evaluation results of the fuzzy rule base are used as training samples for the neural network. A genetic algorithm is used to iteratively optimize the control parameters. Based on real-time detected load changes and environmental disturbances, the parameter combination of each control link is adjusted online to achieve adaptive optimization of control parameters. The fourth unit is used to monitor the operating status of the platform gate group in real time. When an abnormality is detected, the control parameter adaptive optimization process is triggered, and after the parameter adjustment is completed, the adjustment result is fed back to the central control platform.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images