Control Method and System for High-Speed ​​Maglev Suspension System Based on Edge Computing and Transformer Prediction

Abstract

This invention provides a control method and system for a high-speed maglev suspension system based on edge computing and Transformer prediction. The method includes: constructing a Transformer prediction model with spatiotemporal attention and autoregressive mechanisms based on acquired local low-latency computing resources and real-time train sensor data; deploying the Transformer prediction model on an onboard edge computing unit; performing short-term, high-precision predictions of gaps, acceleration, and disturbance trends at several future sampling times using the Transformer prediction model to obtain prediction results; utilizing a model predictive controller to handle actuator current saturation and gap safety threshold hard constraints within a finite prediction domain, and combining the prediction results to solve for and optimize the control sequence in real time; superimposing a control barrier function as a safety filter for the model predictive controller to correct the optimized control sequence, obtaining the optimal control sequence for controlling the high-speed maglev suspension system. This invention reduces cloud communication latency and jitter, improving the robustness and safety of the system under uncertain disturbances.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of high-speed maglev suspension technology, and in particular to a high-speed maglev suspension system control method and system based on edge computing and Transformer prediction. BACKGROUND

[0002] As a future-oriented ground transportation mode, the core advantage of high-speed maglev transportation lies in the fact that it is free from wheel-rail adhesion and realizes non-contact operation, thereby avoiding the constraints of mechanical friction and adhesion limits and having the potential to operate at a speed of 500 km / h or above. The suspension system, as the core subsystem of high-speed maglev trains, directly determines whether the train can safely and smoothly hover above the track. However, the suspension system is a strong nonlinear, open-loop unstable electromechanical coupling system, which is extremely sensitive to small disturbances and can be amplified with even slight deviations. As the speed increases to 600 km / h, multi-source disturbances such as aerodynamic lift pulsation, track micro-irregularities, vehicle-track coupling, and sensor noise are significantly enhanced. At the same time, the safety gap margin is continuously compressed due to actuator saturation and communication time delay. Traditional control methods have obvious bottlenecks in response speed, prediction ability, and constraint handling, lack of forward-looking judgment of disturbances and state evolution, and are prone to instability of the suspension system at higher speed conditions. SUMMARY

[0003] Therefore, the present application provides a high-speed maglev suspension system control method and system based on edge computing and Transformer prediction to solve the above problems.

[0004] The present application provides a high-speed maglev suspension system control method based on edge computing and Transformer prediction, comprising: based on the obtained local low-latency computing resources and real-time train sensor data, constructing a Transformer prediction model with a spatio-temporal attention mechanism and an autoregressive mechanism, and deploying the Transformer prediction model on a vehicle-mounted edge computing unit; performing short-term high-precision prediction of the gap, acceleration, and disturbance trend at future sampling time points through the Transformer prediction model to obtain a prediction result; using a model predictive controller to handle actuator current saturation and gap safety threshold hard constraints within a limited prediction domain, and combining the prediction result to solve an optimized control sequence in real time; superimposing a control barrier function as a safety filter of the model predictive controller to modify the optimized control sequence and obtain an optimal control sequence; and using the optimal control sequence to control the high-speed maglev suspension system.

[0005] In another implementation manner of the present application, the multi-head attention result obtained by linear mapping after splicing of the spatio-temporal attention mechanism output is:

[0006]

[0007] wherein, denote query, key, value matrices, respectively, is the projection matrix of the th attention head, is the output projection matrix.

[0008] In another implementation form of the invention, the prediction result is represented as:

[0009]

[0010] wherein, is the predicted system state sequence of future steps, is a nonlinear mapping, denotes the input signal sequence from to time instant, denotes the corresponding control input sequence.

[0011] In another implementation form of the invention, the model predictive controller defines a cost function as:

[0012]

[0013] wherein, is the predicted system state of future k +1 steps, is the reference state of the k +1 step, is the control input of the k step, is the terminal reference state of the Nth step at the end of the prediction horizon, denotes a weighted quadratic form, is a given positive definite weight matrix.

[0014] In another implementation form of the invention, the model predictive controller solves a constrained optimal control problem as:

[0015]

[0016] wherein, is the system state of the k step, is a state matrix, is a control matrix, is a disturbance term, is an influence matrix, and are constraint boundaries for the control and state variables, respectively.

[0017] In another implementation form of the present application, the control barrier function is:

[0018]

[0019] wherein, is a gap deviation, h e is a rated equilibrium gap of the system, is a minimum safety gap.

[0020] In another implementation form of the present application, the optimal control sequence is:

[0021]

[0022] wherein, is a decision variable obtained by QP optimization, is a candidate control input, are Lie derivatives of system drift terms and control terms on respectively, is a like function.

[0023] In another aspect of the present application, a high-speed maglev suspension system control system based on edge computing and Transformer prediction is provided, comprising: a model construction module: based on obtained local low-latency computing resources and real-time train sensing data, a Transformer prediction model with a space-time attention mechanism and an autoregressive mechanism is constructed, and the Transformer prediction model is deployed on a vehicle-mounted edge computing unit; a parameter prediction module: through the Transformer prediction model, short-term high-precision prediction is performed on the gap, acceleration and disturbance trend at future sampling time points, and a prediction result is obtained; a sequence optimization module: using a model predictive controller, the actuator current saturation and the gap safety threshold hard constraint in the limited prediction domain are processed, and the prediction result is combined to solve an optimal control sequence in real time; a safety filtering module: a control barrier function is superimposed as a safety filter of the model predictive controller, the optimized control sequence is corrected to obtain an optimal control sequence; and a control module: using the optimal control sequence, the high-speed maglev suspension system is controlled.

[0024] In another aspect of the present application, an electronic device is provided, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the high-speed maglev suspension system control method based on edge computing and Transformer prediction according to any one of the above.

[0025] In another aspect of the present application, a computer storage medium is provided, and a computer program is stored on the computer storage medium, and the computer program is executed by a processor to implement the steps of the high-speed maglev suspension system control method based on edge computing and Transformer prediction according to any one of the above aspects.

[0026] The high-speed maglev suspension system control method based on edge computing and Transformer prediction of the present application forms a closed loop chain of 'prediction guidance-optimization decision-security guardianship', and the fast response of the edge end significantly reduces the control loop delay and jitter, and reduces the gap fluctuation and overshoot caused by phase lag; the prediction compensation based on deep learning can adaptively compensate the influence of model mismatch and environmental change on control performance, and improves the system robustness and stability under strong disturbance and high uncertainty. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. The advantages and benefits of the solutions will become clear and obvious to those skilled in the art by reading the following detailed description of the embodiments. The drawings are only for the purpose of illustrating the preferred embodiments, and are not considered as limiting the present application. In the drawings:

[0028] Figure 1 The flowchart of the high-speed maglev suspension system control method based on edge computing and Transformer prediction of an embodiment of the present application.

[0029] Figure 2 The overall framework diagram of the suspension control based on Internet of Things edge computing and Transformer prediction of an embodiment of the present application.

[0030] Figure 3 The schematic diagram of the high-speed maglev train suspension system of an embodiment of the present application.

[0031] Figure 4 The schematic diagram of the single-suspension point framework of an embodiment of the present application.

[0032] Figure 5 The schematic diagram of the Transformer-Encoder structure of an embodiment of the present application.

[0033] Figure 6 The schematic diagram of the predictor architecture of an embodiment of the present application.

[0034] Figure 7 The schematic diagram of the control structure of the safety filtering of an embodiment of the present application.

[0035] Figure 8An experimental platform device schematic diagram for an embodiment of the present application.

[0036] Figure 9 A pulse disturbance response schematic diagram for an embodiment of the present application.

[0037] Figure 10 A periodic vibration response schematic diagram for an embodiment of the present application.

[0038] Figure 11 A model mismatch response schematic diagram for an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to enable personnel in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and in detail below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art should belong to the scope of protection of the embodiments of the present application.

[0040] Figure 1 A high-speed maglev suspension system control method flowchart schematic diagram based on edge computing and Transformer prediction provided by the present application is shown in FIG. 1, and the present embodiment mainly includes: Figure 1

[0041] S101, based on the obtained local low-latency computing resources and real-time train sensor data, a Transformer prediction model with a space-time attention mechanism and an autoregressive mechanism is constructed, and the Transformer prediction model is deployed on a vehicle-mounted edge computing unit.

[0042] S102, the gap, acceleration and disturbance trend at future sampling time are short-term high-precision predicted by the Transformer prediction model, and a prediction result is obtained.

[0043] S103, the model predictive controller is used to process the actuator current saturation and the gap safety threshold hard constraint within the limited prediction domain, and the prediction result is combined to solve the optimal control sequence in real time.

[0044] S104, the control barrier function is superimposed as a safety filter of the model predictive controller, the optimized control sequence is corrected, and an optimal control sequence is obtained.

[0045] S105, the optimal control sequence is used to control the high-speed maglev suspension system.

[0046] Exemplarily, as shown in FIG. 1, the present application provides a high-speed maglev suspension system control method based on edge computing and Transformer prediction, which mainly includes: Figure 2 ​To achieve real-time performance of high-speed maglev trains under the operating condition of 600 km / h, a suspension control framework is constructed based on Internet of Things (IoT) and on-board edge computing technology. Through the edge computing nodes deployed locally on the train, the framework realizes real-time acquisition, synchronization and preprocessing of multi-source sensor data. Relying on its low-latency and high-reliability computing capability, the framework runs a lightweight Transformer prediction model and a model predictive controller (MPC) to complete state prediction, optimization solving and safety checking in real time, forming a closed-loop control link of "perception-prediction-decision-execution".

[0047] As shown in Figure 3 , the suspension points of high-speed maglev trains are decoupled mechanically, and there is no interference between the suspension points. Generally, a single suspension point is taken as the modeling object.

[0048] As shown in Figure 4 , a suspension point usually adopts a closed loop composed of an electromagnet, a suspended object and a position signal sensor. The parameters of the system are shown in Table 1.

[0049] Table 1 System parameters and state definitions are as follows:

[0050]

[0051] The system dynamics equation is:

[0052]

[0053] At the equilibrium suspension point , it satisfies , and the equilibrium current is obtained. The control input is defined as the increment of the current to the equilibrium value , and the system is linearized for small perturbations. The approximate local linear model is obtained as follows:

[0054]

[0055] The high-order small quantities are ignored in the above formula. As the prediction model of the MPC controller, the above formula can only be approximately effective in a small range due to the strong nonlinearity and parameter uncertainty of the suspension system; under large deviation, nonlinear prediction and control strategies are needed to ensure performance. Therefore, a time series predictor is integrated into the traditional model predictive control to estimate and compensate the future evolution trend of the system.

[0056] The high-speed maglev suspension system control method based on edge computing and Transformer prediction of the present invention forms a closed-loop chain of "predictive guidance - optimized decision-making - safety protection". The fast response at the edge end significantly reduces control loop delay and jitter, and reduces gap fluctuation and overshoot caused by phase lag. The prediction compensation based on deep learning can adaptively compensate for the impact of model mismatch and environmental changes on control performance, and improve the robustness and stability of the system under strong disturbance and high uncertainty conditions.

[0057] In another implementation of the present invention, the spliced ​​outputs of the spatiotemporal attention mechanism, after being linearly mapped, yield the following multi-head attention result:

[0058]

[0059] in, These represent the query, key, and value matrices, respectively. For the first The projection matrix of each attention head. This is for outputting the projection matrix.

[0060] For example, such as Figure 5 As shown, to improve prediction accuracy, a lightweight temporal prediction model with a Transformer-Encoder structure was designed and deployed on an edge computing unit. The prediction model utilizes historical sensor data and real-time streaming data stored locally on the edge side, constructing an input sequence through a sliding window mechanism to achieve high-precision prediction of the system state in multiple future steps. Edge deployment not only avoids the communication overhead of uploading data to the cloud but also enables efficient model inference through hardware acceleration, meeting the real-time requirements of the control system. Furthermore, compared to conventional RNNs or LSTMs, the Transformer, based on a self-attention mechanism, can extract complex temporal dependencies over longer time windows without being easily forgotten.

[0061] The prediction model employs an encoder-based multi-layer Transformer block structure, performing spatial and temporal dual-path attention modeling on the input temporal features. First, the input multidimensional feature sequence undergoes linear projection to obtain a high-dimensional embedding representation, and positional encoding is added to preserve temporal order information. Subsequently, the embedding sequence is fed in parallel into the spatial and temporal attention branches: the spatial branch obtains the query, key, and value matrix through linear mapping. The scaling dot product attention method captures the correlation between features from different sensors; its calculation formula is as follows:

[0062]

[0063] in, These represent the query, key, and value matrices, respectively. is the scaling factor (key vector dimension). With parallel multi-head attention, the model can focus on the feature relationship of different positions and different dimensions in the sequence. After concatenation, the output is linearly mapped to obtain the multi-head attention result:

[0064]

[0065] wherein, is the projection matrix of the th attention head, is the output projection matrix. The multi-head attention output of the spatial branch is connected in residual and normalized, and then enters the forward feedback network (FFN) to extract higher layer nonlinear features. Similarly, the time branch reuses the query of the spatial branch, and performs cross-attention calculation with another set of keys and values representing the time sequence pattern to extract the dynamic dependence of the sensor data over time. The calculation form of the time attention is similar to that of the spatial attention, except that the keys and values come from the time sequence features. After the double-path attention mechanism, the output of the spatial branch and the time branch is connected in residual and normalized, and then sent to the FFN layer to further extract features. The two output features are concatenated and fused as the input of the next layer of Transformer encoding block. Through this serial spatial-time double-path attention structure, the correlation between multiple sensing channels and the change pattern of single-channel signals over time can be modeled at the same time.

[0066] Considering the limited resources of vehicle edge computing, the Transformer prediction model is designed to be small, such as reducing the number of network layers and the dimension of hidden units, and using knowledge distillation to compress model parameters. The final model can complete inference at an extremely fast speed, meeting the real-time requirements. With the help of vehicle GPU / FPGA hardware acceleration, the prediction model can calculate the system output for the next steps in each control period, providing forward-looking information for MPC. The model is trained using offline historical data, and the mean square error (MSE) between the predicted output and the true output is selected as the loss function, and a regularization term is added to prevent overfitting. After training, the model is deployed on the vehicle edge, and can be updated online regularly using newly collected data to maintain sensitivity to system changes and ensure that the predictor continues to provide accurate rolling predictions.

[0067] The Transformer prediction model uses offline historical data for supervised learning during the training phase, and uses knowledge distillation techniques to compress the model size to adapt to the limited computing resources of edge devices. After the model is deployed, the edge computing platform can implement dynamic updating and online learning, gradually adapting to system dynamics and further improving prediction accuracy.

[0068] In another implementation form of the present application, the prediction result is represented as:

[0069]

[0070] wherein, is the future step system state prediction sequence, is a nonlinear mapping, represents an input signal sequence from to time instant, represents the corresponding control input sequence.

[0071] Exemplarily, the multi-channel sensor features of the most recent time instants are selected as the prediction model input sequence, including the electromagnet current increment, levitation gap deviation, and other sensor measurement values , and the control input at the corresponding time instant . The prediction model output is the future step system state prediction sequence ， For example, the prediction of the suspension acceleration or vibration amplitude in the next period of time. By the method of sliding time window, the long time sequence data is divided into multiple groups of training samples to fully extract the local time sequence pattern. The nonlinear mapping represented by the Transformer prediction model is denoted as , and there is:

[0072]

[0073] wherein, represents an input signal sequence from to time instant, represents the corresponding control input sequence, is the future first step output prediction value. In order to improve the multi-step prediction performance, the prediction model adopts an autoregressive mechanism: when predicting the future sequence step by step, the prediction output at the previous time instant (for example, the acceleration prediction at the last time instant) is fed back as one of the inputs at the subsequent time instant, together with the newly collected observation to form a new input sequence. Through output feedback, the model can utilize its own prediction results to continuously predict in the inference stage, thereby improving the continuity and stability of the prediction sequence and reducing the phenomenon of error amplification over time in multi-step prediction.

[0074] In another implementation form of the present application, the model predictive controller defines a cost function as:

[0075]

[0076] where, is the future k step system predicted state, is the k step reference state, is the k step control input, is the terminal reference state at the end of the prediction horizon (Nth step), denotes a weighted quadratic form, is a given positive definite weight matrix.

[0077] Exemplarily, model predictive control (MPC) with finite prediction horizon is employed to generate control commands for the maglev system. MPC solves an optimal control problem with the current state as the initial value online at each sampling time, obtaining the optimal control sequence for the future steps , and actually applying the first step control Figure 6 to the system, then rolling the time horizon forward by one period and repeating the above optimization. The MPC controller uses the state-space model obtained by linearization as the prediction model, and incorporates the future disturbance estimated by the Transformer prediction model into the optimization to improve the decision-making of the control system, and its closed-loop architecture is shown in .

[0078] The edge computing platform provides a hardware basis for the real-time optimization of MPC. An efficient quadratic programming (QP) solver is used, and the parallel computing capability of the edge device is utilized to quickly solve the constrained optimization problem within each control period. By localizing the prediction model and the objective function on the edge side, the dependence on the central control unit is reduced, and the autonomy and response speed of the system control are improved.

[0079] The optimization formula of MPC is described as follows, where denotes the current system state, and the reference trajectory is (the steady-state suspension control can take , which means that the suspension gap deviation is zero). The trajectory error and control energy consumption are minimized within the time horizon to , and thus the cost function is defined as:

[0080]

[0081] where, denotes a weighted quadratic form, is a given positive definite weight matrix (or scalar), which is used to balance the state error, control increment, and terminal state deviation, etc.

[0082] In another implementation of the present application, the optimal control problem with constraints solved by the model predictive controller is:

[0083]

[0084] where, is the system state at the k th step, is the state matrix, is the control matrix, is the disturbance term, is the influence matrix, and are the constraint boundaries for control and state variables, respectively.

[0085] Exemplarily, based on the cost function described above, the MPC needs to solve the following optimal control problem with constraints at each time instant:

[0086]

[0087] where, represents the current state, the state equation constraint utilizes the state matrix of the linearized model, the control matrix , and adds the disturbance term and its influence matrix ; and represent the constraint boundaries for the actuator current, the levitation gap, and other control and state variables. By solving the above convex quadratic programming problem, the optimal control sequence that satisfies the constraints can be obtained. The MPC optimization can obtain the solution in real time using an efficient QP solver numerically, and since the prediction of future disturbances is introduced , the MPC can prospectively adjust the control strategy to offset the impact of the upcoming disturbance. For example, when it is predicted that the levitation gap will decrease due to the disturbance at the next time instant (i.e. causes to drop), the optimization will tend to increase the current control input (elevate the electromagnetic force) in advance to pre-inhibit the gap narrowing, thereby playing a feed-forward compensation role.

[0088] At each sampling period, the first-step control obtained by optimization is applied And the above process is repeated in the cycle optimization window. Since MPC considers input and state constraints when optimizing, it can ensure that the optimal solution will not make the system enter a dangerous state. However, in actual operation, model inaccuracy or prediction errors may still cause the system to approach the constraint boundary. Therefore, a layer of safety constraint control is added outside the MPC controller as a safety filter to further improve the safety of system operation. In actual implementation, the MPC control algorithm runs on the vehicle-mounted edge computing platform, and the QP solver is used to ensure that the optimal solution is solved within milliseconds of the control cycle, meeting the real-time requirements.

[0089] A light-weight Transformer network deployed on the vehicle-mounted edge computer is used to make short-term predictions of the future key states and disturbances of the maglev system, and the prediction results are integrated into the optimization process of the MPC to compensate for the shortcomings of traditional model prediction in nonlinear and uncertain scenarios.

[0090] In another implementation of the present application, the control barrier function is:

[0091]

[0092] wherein, is the gap deviation, h e is the rated equilibrium gap of the system, is the minimum safety gap.

[0093] Illustratively, to further improve the safety of system operation, a control barrier function (CBF) is introduced on the edge computing platform as a safety filter for the MPC. Similarly, it is deployed on the edge computing platform, which can greatly reduce the delay and ensure the real-time performance of the safety system. The CBF method constructs a function to describe the safe set of system states, and forces not to decrease, so as to ensure that the state always stays within the safe region. For the maglev system, a natural safe set is that the suspension gap is not lower than a threshold to avoid collision between the suspended object and the electromagnet. For example, if the minimum safety gap is set to , the barrier function can be defined as and the safety requirement is for all times.

[0094] According to the CBF theory, if there exists a continuously differentiable barrier function at the dangerous boundary , then as long as the control input satisfies:

[0095]

[0096] For the evolution of the closed-loop system, the set will remain unchanged. Where are Lie derivatives of the system drift term and control term with respect to , is a class function (typically a linear function ). The above inequality describes the linear constraint that the control input needs to satisfy. When is close to 0 (the state is close to the safety boundary), the control needs to be chosen such that the rate of change of is positive, i.e., to drive the state away from the dangerous boundary.

[0097] The control barrier function (CBF) constraint is added in the control framework to correct the control command given by the MPC in real time, so that the state of the levitation system satisfies the safety constraint requirement such as the levitation gap, and ensures that the system still operates safely and stably under disturbance.

[0098] In another implementation of the present application, the optimal control sequence is:

[0099]

[0100] where is the decision variable obtained by QP optimization, is the candidate control input, are Lie derivatives of the system drift term and control term with respect to , is a class function.

[0101] Exemplarily, the CBF constraint is combined with the MPC control. At each sampling time, after the MPC gives the candidate control input , a QP module is added to perform safety filtering to solve the control input closest to that satisfies the safety condition:

[0102]

[0103] Since the QP has only one linear inequality constraint, it can be solved in each cycle, the calculation speed is guaranteed, and the obtained satisfies the safety condition. If itself already satisfies the safety constraint, then , the original control is not changed, otherwise the control parameters will be adjusted appropriately to avoid touching the safety boundary. It should be noted that when the MPC and the CBF conflict (i.e., the MPC tries to sacrifice safety to optimize performance), the CBF constraint will become the last line of defense, i.e., it is better to lose part of the performance to ensure safety.

[0104] Figure 7 The control structure diagram with CBF safety filter is shown. It can be seen that the CBF module monitors the key state of the system, and when the control instruction given by the MPC may cause a violation of the safety boundary, the CBF will correct the control input online. By adjusting the value of the parameter in the CBF, the safety conservatism and control performance can be balanced to some extent: When the CBF is larger, it will be more aggressive to move away from the constraint boundary, but it may slightly reduce the tracking speed; When the CBF is smaller, it is the opposite. In the experimental verification, the appropriate ensures that sufficient margin is maintained near the minimum suspension gap while not affecting the dynamic performance.

[0105] In summary, the suspension control framework deeply integrates IoT sensing, edge computing and intelligent predictive control technology, building a closed-loop system that integrates data collection, local processing, real-time decision-making and security protection. By deploying the Transformer prediction model, MPC optimizer and CBF safety filter on the vehicle edge platform, the system can maintain excellent control performance and safety level under high-speed and high-disturbance conditions.

[0106] The present application deploys a predictive control algorithm on a vehicle edge computing unit, uses local low-latency computing resources and real-time train sensor data to build a Transformer prediction model with spatio-temporal attention and autoregressive mechanism, and performs short-term high-precision prediction on the gap / acceleration and disturbance trend at future sampling times to provide feedforward compensation for the controller. According to the prediction result, the model predictive controller (MPC) processes the actuator current saturation, gap safety threshold and other hard constraints within a limited prediction domain, and solves the optimal control sequence in real time to ensure that the tracking performance and energy efficiency are guaranteed while ensuring that the constraints are met. And superimposed control barrier function (CBF) as an outer safety filter, to ensure that the state is always within the feasible safety domain, forming a closed-loop chain of "prediction guidance - optimization decision - safety guardianship". With the processing power of vehicle edge computing, this architecture has the following advantages: First, the fast response of the edge end significantly reduces the control loop delay and jitter, reducing the gap fluctuations and overshoot caused by phase lag; second, the prediction compensation based on deep learning can adaptively compensate for the impact of model mismatch and environmental changes on control performance, improving the robustness and stability of the system under strong disturbance and high uncertainty.

[0107] Embodiment 1

[0108] To verify the performance of the proposed design, a high-speed maglev vehicle-track coupled vibration experimental system based on the Internet of Things (IoT) architecture and vehicle edge computing platform is built. The system consists of a high-speed maglev vehicle-track coupled vibration platform, a dSPACE real-time simulation platform as a vehicle edge computing unit, and a suspension controller.

[0109] IoT perception layer: The vibration platform simulates track excitations at different speeds by the shaker, while the multi-source IoT sensor network (including gap sensors, current sensors, and accelerometers) collects key state data of the system in real time.

[0110] Edge computing layer: The dSPACE platform serves as the on-board edge computing core, interconnected with sensors and controls through high-speed communication buses. It is responsible for running the lightweight Transformer prediction model, MPC optimization solver, and CBF safety filtering algorithm proposed earlier, completing the complete process of "perception-prediction-decision-execution" within one control cycle.

[0111] Execution layer: The suspension controller receives control instructions from the edge computing unit and outputs current-driven electromagnets.

[0112] The verification platform can test the integrated IoT edge intelligent control process of "perception-computation-control". Since it is deployed locally, it can ensure that the control instructions generated by the edge computing unit can be sent to the actuator with extremely low delay, effectively verifying the real-time performance of edge computing in high-speed maglev suspension systems. The experimental platform device is shown in Figure 8 .

[0113] The Transformer prediction model inputs past step sensor measurements and control inputs and outputs position estimates for future steps. The MPC prediction horizon is matched with the predictor length; the weight matrix is taken . The constraint condition is that the suspension rated gap is 10 mm, the minimum suspension gap is 5 mm, and less than 5 mm is considered unsafe, and the CBF parameters are selected . The performance of the MPC controller is verified by comparing it with the traditional PID control under three typical working conditions.

[0114] Pulse disturbance performance verification: First, verify the response of the suspension system to sudden disturbances. At , a pulse impact is applied to simulate a sudden change in the track. Figure 9The response curve of the suspension gap under different control strategies is given. The traditional PID controller has a instantaneous drop of about 5 mm in the gap after the disturbance occurs, and then the system adjusts for about 4 s to recover to normal. While the method of the application can reduce the control current in advance to compensate due to the advance prediction of the disturbance by the Transformer prediction model, so that the gap only fluctuates by 2 mm, and returns to the normal state in a very short time, and the current and acceleration are basically normal. Plus the safety filtering effect of CBF, the gap is kept within the safety threshold range. The numerical results show that the MPC controller reduces the maximum deviation by more than 50% and shortens the recovery time by about 60%. The performance of the MPC based on the transformer network prediction compensation in the anti-pulse is verified.

[0115] Disturbance performance verification of track long-wave irregularity: The disturbance of track long-wave irregularity can be regarded as a periodic vibration disturbance of the track to the vehicle. A vehicle-track coupling vibration table is used to simulate the disturbance of track long-wave irregularity, and a sinusoidal disturbance with a vibration frequency of 3 Hz and an amplitude of 3 mm is set. Figure 10 The gap, acceleration and current curves under the traditional PID controller and the MPC controller are shown. It can be found that the control current of the traditional PID controller lags behind the disturbance by about 0.3 s in phase, which leads to the output not being aligned with the reference. While the MPC controller can give a pre-gain in advance to change the control current appropriately before the periodic disturbance comes, so that the suspension gap is significantly smaller than the fluctuation of the traditional controller, and there is no phase lag, and the fluctuation of the current is also smaller than that of the traditional controller, which shows that the prediction compensation avoids excessive drastic control action.

[0116] Performance verification under model mismatch: One pole module of the actual system is removed, and at the same time a part of the load, which is the mass in the suspension system, is increased and The real system is mismatched with the mathematical model. Figure 11 It can be seen from the figure that the gap, current and acceleration responses of the traditional PID controller and the MPC controller. The results show that the traditional PID controller takes nearly 4 s from levitation to stable, while the MPC controller takes only 2 s to reach the stable state, and the amplitude of the gap oscillation is smaller than that of the traditional PID control. This is because the predictor learns the actual dynamic characteristics of the system from real-time data, which to some extent makes up for the model mismatch. In addition, CBF appropriately increases the control effort when it detects that there is a slight safety risk in the initial oscillation, and quickly pulls the system back to the safety domain. Overall, the method of the application shows a certain robustness under model mismatch.

[0117] In summary, the proposed MPC+CBF control framework based on vehicle-mounted edge Transformer prediction enhancement performs better than traditional methods in various typical working conditions, ensuring the stability of the suspension system and the reliability of the operation through safety constraints.

[0118] The present application is aimed at the stability control problem of the magnetic levitation suspension system, and proposes a real-time closed-loop control method integrating vehicle-mounted IoT edge intelligence. By deploying a Transformer time series predictor on the edge computing platform, the future state changes of the system are perceived in advance, and combined with model predictive control, fast and accurate adjustment of the suspension system is realized. At the same time, a control barrier function is introduced to safely filter the control input, ensuring that the system always works within a safe range. A hardware system including the dynamic model of the suspension system, the predictor, the optimization controller and the safety filter is established and experimentally verified. The results show that the present application improves the dynamic response speed, and the method can effectively resist transient and periodic disturbances, significantly improve the suspension stability and tracking performance, and has a certain robustness. Compared with the traditional PID controller, the present application reduces the overshoot and phase lag of the suspension gap, ensures the stability of the suspension, effectively improves the response speed, and is significantly better than the baseline control scheme without combining prediction compensation.

[0119] In another aspect of the present application, a high-speed magnetic levitation suspension system control system based on edge computing and Transformer prediction is provided, comprising:

[0120] A model construction module: based on the obtained local low-latency computing resources and real-time train sensor data, a Transformer prediction model with spatiotemporal attention mechanism and autoregressive mechanism is constructed, and the Transformer prediction model is deployed on the vehicle-mounted edge computing unit.

[0121] A parameter prediction module: the Transformer prediction model is used to perform short-term high-precision prediction on the gap, acceleration and disturbance trend at future sampling time points, and the prediction result is obtained.

[0122] A sequence optimization module: a model predictive controller is used to handle actuator current saturation and gap safety threshold hard constraints within a limited prediction domain, and the prediction result is used to solve an optimization control sequence in real time.

[0123] A safety filtering module: a control barrier function is superimposed as a safety filter of the model predictive controller to modify the optimized control sequence and obtain an optimal control sequence.

[0124] A control module: the optimal control sequence is used to control the high-speed magnetic levitation suspension system.

[0125] The high-speed maglev suspension system control system based on edge computing and Transformer prediction of the application forms a closed loop chain of "prediction guidance-optimization decision-security guardianship", and the rapid response of the edge end significantly reduces the control loop delay and jitter, reduces the gap fluctuation and overshoot caused by phase lag; the prediction compensation based on deep learning can adaptively compensate the influence of model mismatch and environmental change on control performance, and improve the system robustness and stability under strong disturbance and high uncertainty.

[0126] In another aspect of the application, the electronic device comprises a processor, a memory, and a communication bus, a communication interface.

[0127] Wherein:

[0128] The processor, the memory and the communication interface complete the communication with each other through the communication bus.

[0129] The communication interface is used for communication with other electronic devices or servers.

[0130] The processor is used for executing a program, and specifically can execute the steps of any one of the high-speed maglev suspension system control methods based on edge computing and Transformer prediction in the above embodiments.

[0131] Specifically, the program can include program code comprising computer operation instructions.

[0132] The processor can be a central processing unit CPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement one or more embodiments of the application. The one or more processors included in the smart device can be the same type of processor, such as one or more CPUs; or can be different types of processors, such as one or more CPUs and one or more ASICs.

[0133] The memory is used for storing the program. The memory can include a high-speed RAM memory, and can also include a non-volatile memory, such as at least one disk memory.

[0134] The program can be specifically used for enabling the processor to execute to realize the steps of any one of the high-speed maglev suspension system control methods based on edge computing and Transformer prediction described in the embodiments. The specific implementation of each step in the program can refer to the corresponding description in the steps and units executed by any one of the high-speed maglev suspension system control methods based on edge computing and Transformer prediction described above, and details are not described here. It can be clearly understood by those skilled in the art that, for the convenience and brevity of description, the specific working processes of the devices and modules described above can refer to the corresponding process description in the foregoing method embodiments.

[0135] The exemplary embodiments of the present application also provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the method of the embodiments of the present application.

[0136] The method according to the embodiments of the present application described above can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium such as a CD ROM, a RAM, a floppy disk, a hard disk or a magneto-optical disk, or be downloaded through a network originally stored in a remote recording medium or a non-transitory machine-readable medium and then stored in a local recording medium, so that the method described herein can be processed by such software using a general-purpose computer, a special-purpose processor, or programmable or special-purpose hardware (such as ASIC or FPGA). It can be understood that the computer, processor, microprocessor controller or programmable hardware includes a storage component (for example, RAM, ROM, flash memory, etc.) that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code will convert the general-purpose computer into a special-purpose computer for executing the method shown herein.

[0137] So far, specific embodiments of the present application have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. In addition, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results.

[0138] It should be noted that all directional indications, such as upper, lower, left, right, back, etc., in the embodiments of the present application are only used to explain the relative positional relationship between components, etc. in a certain order (as shown in the drawings), and if the certain order changes, the directional indications also change accordingly.

[0139] In the description of the present application, the terms "first", "second" are only used for the convenience of describing different components or names, and cannot be understood as indicating or implying the order relationship, relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can be explicitly or implicitly included at least one of the features.

[0140] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0141] It should be noted that, although the specific embodiments of the present application are described in detail with reference to the accompanying drawings, it should not be understood as limiting the scope of protection of the present application. Various modifications and variations made by those skilled in the art within the scope described in the claims are still within the scope of protection of the present application.

[0142] The examples of the embodiments of the present application are intended to simply illustrate the technical features of the embodiments of the present application, so that those skilled in the art can intuitively understand the technical features of the embodiments of the present application, and are not improper limitations of the embodiments of the present application.

[0143] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A high-speed maglev suspension system control method based on edge computing and Transformer prediction, characterized in that, The method comprises the following steps: Based on the obtained local low-latency computing resources and real-time train sensor data, a Transformer prediction model with a spatio-temporal attention mechanism and an autoregressive mechanism is constructed, and the Transformer prediction model is deployed on a vehicle-mounted edge computing unit; The multi-head attention result obtained by linear mapping after splicing of the spatio-temporal attention mechanism output is: wherein, respectively represent a query, a key, a value matrix, is the projection matrix of the th attention head, is the output projection matrix; The gap, acceleration and disturbance trend at a plurality of future sampling time points are short-term high-precision predicted by the Transformer prediction model to obtain a prediction result; The prediction result is represented as: wherein for future steps of the system state prediction sequence, is a nonlinear mapping, denotes the input signal sequence from to time instant, denotes the corresponding control input sequence; An optimal control sequence is solved in real time by a model predictive controller in a limited prediction domain to process actuator current saturation and gap safety threshold hard constraints in combination with the prediction result; The definition cost function of the model predictive controller is: wherein, is the future k +1 step system predicted state, is the k +1 step reference state, is the k step control input, is the terminal predicted state at the end of the prediction horizon at step N, is the terminal reference state at the end of the prediction horizon at step N, denotes a weighted quadratic form, is a given positive definite weight matrix; The optimal control problem with constraints solved by the model predictive controller is: wherein, is the state of the system at the kth k step, is the state matrix, is the control matrix, is the disturbance term, is the influence matrix, and are the constraint boundaries for the control and state variables, respectively. A control barrier function is superimposed as a safety filter of the model predictive controller to correct the optimized control sequence to obtain an optimal control sequence; The control barrier function is: wherein is a gap deviation, h e is a rated balance gap of the system, is a minimum safety gap; The optimal control sequence is: wherein is the decision variable obtained by QP optimization, is the candidate control input, are Lie derivatives of the system drift term and control term pair respectively, is a similar function; The high-speed maglev suspension system is controlled by using the optimal control sequence.

2. A high-speed maglev suspension system control system based on edge computing and Transformer prediction, characterized in that, The method comprises the following steps: A model construction module: based on the obtained local low-latency computing resources and real-time train sensor data, a Transformer prediction model with a spatio-temporal attention mechanism and an autoregressive mechanism is constructed, and the Transformer prediction model is deployed on a vehicle-mounted edge computing unit; the multi-head attention result obtained by linear mapping after splicing of the spatio-temporal attention mechanism output is: wherein, respectively represent a query, a key, a value matrix, is a projection matrix of the th attention head, is an output projection matrix; A parameter prediction module: the gap, acceleration and disturbance trend at a plurality of future sampling time points are short-term high-precision predicted by the Transformer prediction model to obtain a prediction result; the prediction result is represented as: wherein for future steps of the system state prediction sequence, is a nonlinear mapping, denotes the input signal sequence from to time instant, denotes the corresponding control input sequence; A sequence optimization module: an optimal control sequence is solved in real time by a model predictive controller in a limited prediction domain to process actuator current saturation and gap safety threshold hard constraints in combination with the prediction result; the definition cost function of the model predictive controller is: wherein, is the future k +1 step system predicted state, is the k +1 step reference state, is the k step control input, is the terminal predicted state at the end of the prediction horizon at step N, is the terminal reference state at the end of the prediction horizon at step N, denotes a weighted quadratic form, is a given positive definite weight matrix; The optimal control problem with constraints solved by the model predictive controller is: wherein, is the system state at the kth k step, is the state matrix, is the control matrix, is the disturbance term, is the influence matrix, and are the constraint boundaries for the control and state variables, respectively. A safety filtering module: a control barrier function is superimposed as a safety filter of the model predictive controller to correct the optimized control sequence to obtain an optimal control sequence; the control barrier function is: wherein, is a gap deviation, h e is a rated balance gap of the system, is a minimum safety gap; The optimal control sequence is: wherein, is the decision variable obtained by QP optimization, is the candidate control input, are Lie derivatives of the system drift term and control term pair respectively, is a similar function; A control module: the high-speed maglev suspension system is controlled by using the optimal control sequence.

3. An electronic device, comprising: The method comprises the following steps: A memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the high-speed maglev suspension system control method based on edge computing and Transformer prediction according to claim 1 when executing the computer program.

4. A computer storage medium, characterized in that, The computer storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in the high-speed maglev suspension system control method based on edge computing and Transformer prediction according to claim 1.

Images