A metro pulse passenger flow-oriented escalator control method, device and medium

Abstract

The application discloses a metro pulse passenger flow-oriented escalator control method, equipment and medium, the method comprises the following steps: according to metro train operation data, construct the pulse passenger flow trigger sequence based on discrete events, calculate the continuous passenger flow injection curve arriving at the entrance of the escalator;The future dynamic load torque of the escalator is predicted by using PINN network;In combination with the predicted dynamic load torque, the minimum system operation energy consumption, mechanical wear and queuing congestion penalty are taken as the comprehensive target, the multi-objective nonlinear model predictive control optimization problem is constructed and solved, the optimal speed control instruction is generated and issued to the escalator frequency converter for execution.The application can accurately reconstruct the real-time quality distribution on the escalator, and calculate the future dynamic load torque accordingly, and the controller adjusts the motor operation accordingly, so that it always works at the optimal slip frequency point, reduces the energy consumption of the escalator, reduces the wear of parts, prolongs the service life.

Description

Technical Field

[0001] This application relates to the field of escalator technology, and in particular to an escalator control method, equipment and medium for subway pulse passenger flow. Background Technology

[0002] With the rapid development of urban rail transit, escalators, as core passenger transport equipment connecting platforms and concourses in subway stations, are experiencing increasingly large numbers and high energy consumption. The energy consumption of subway escalators accounts for a significant proportion of the total energy consumption of a station; therefore, implementing energy-saving control of escalators has become a crucial aspect of green rail transit construction. However, unlike the stable passenger flow in places like shopping malls or airports, subway passenger flow has a highly pronounced "pulsating" characteristic: when a train is not yet arriving at the station, the escalator is empty or lightly loaded; but when the train arrives and the doors open, a large number of passengers surge onto the escalator in a very short time, forming a momentary high-density pulse of passenger flow. This extreme fluctuation in passenger flow presents a significant challenge to the energy-saving control and safe operation of escalators.

[0003] Currently, the energy-saving solutions commonly used in the industry for escalators mainly rely on infrared sensors, photoelectric switches, or radar probes installed at the escalator entrance. Their basic control logic is "passive triggering": when no one is within the detection range, the escalator runs at low idle speed or stops; when a passenger is detected entering the sensing area, the frequency converter controls the motor to accelerate rapidly to the rated speed. However, this passive response mechanism has an inherent lag. Since the sensing area is usually only 1 to 2 meters away from the steps, when faced with a sudden surge of passengers from subway trains, the escalator often only begins to accelerate hastily after passengers have already stepped onto the steps. This abrupt start not only easily causes passengers to lose their balance but also exacerbates the wear and tear on mechanical transmission components such as the drive chain and reducer gears, shortening the equipment's lifespan.

[0004] Furthermore, existing variable frequency energy-saving control systems have safety blind spots and energy efficiency bottlenecks when dealing with large, transient passenger flows in subways. On one hand, existing sensors can only identify whether someone is present or not, and cannot predict the scale of the upcoming passenger flow. When a massive influx of passengers arrives instantly from the train, if the escalator is still operating at low speed and cannot accelerate sufficiently, a bottleneck effect can easily form at the entrance, leading to large crowds and even stampedes. On the other hand, because the controller cannot know the specific number, distribution, and actual weight of passengers on the escalator, it cannot calculate the accurate mechanical load torque. To prevent escalator slippage due to insufficient power, existing motor control strategies typically reserve a large safety margin, causing the motor to operate in an inefficient "overpowered" state for extended periods, significantly reducing the actual overall energy-saving effect.

[0005] In summary, existing subway escalator control technologies face the challenge of balancing energy efficiency and safety when dealing with pulsating passenger flows. Therefore, there is an urgent need in this field for an intelligent, energy-saving control method that can predict passenger flow in advance, accurately map loads, and proactively address congestion risks. Summary of the Invention

[0006] This application provides a method, device, and medium for controlling escalators in subway pulsed passenger flow. Its advantages are that it can accurately reconstruct the real-time mass distribution on the escalator path and calculate the future dynamic load torque accordingly. The controller adjusts the motor operation accordingly to keep it working at the optimal slip frequency point, thereby reducing escalator energy consumption, reducing component wear, and extending service life.

[0007] The technical solution of this application is as follows:

[0008] On the one hand, this application provides an escalator control method for subway pulse passenger flow, including the following steps:

[0009] S1: Construct a pulse-based passenger flow trigger sequence based on discrete events, using subway train operation data;

[0010] S2: Based on the passenger flow trigger sequence and the crowd space transmission delay model, calculate the continuous passenger flow injection curve and queuing status at the escalator entrance;

[0011] S3: The passenger flow injection curve is used as a boundary condition input to the physical information neural network PINN to solve the crowd dynamics equation in order to predict the future dynamic load torque of the escalator.

[0012] S4: Based on the predicted dynamic load torque, establish an electromechanical coupling state-space model of the escalator for calculating the system energy consumption and speed evolution; with the comprehensive objective of minimizing system operating energy consumption, mechanical wear and queuing congestion penalty, construct a multi-objective nonlinear model predictive control optimization problem;

[0013] S5: Solve the optimization problem, generate the optimal speed control command, and send it to the escalator frequency converter for execution.

[0014] Furthermore, in step S1, firstly, a pulse-type subway passenger flow evolution model is constructed:

[0015] Let the operating time domain of the subway station be t∈[0,T], and the scheduled arrival time of the i-th train be τ. i The discrete event of a train entering the station is transformed into a trigger function E(t) on the time axis:

[0016]

[0017] In the formula, δ(⋅) is the Dirac function, and N trainTotal number of train services throughout the day;

[0018] Secondly, construct an asymmetric passenger flow generation rate model:

[0019] By introducing an asymmetric Gaussian kernel function to convolve the function of this event, the instantaneous passenger flow generation rate λ is obtained. in (t):

[0020]

[0021] In the formula, Q load (τ) represents the passenger capacity of the corresponding train, μ lag σ represents the average delay time of passengers exiting the train doors. t A time-scale parameter characterizing the degree of population dispersion;

[0022] Instantaneous passenger flow generation rate λ in (t) constitutes the passenger flow trigger sequence.

[0023] Furthermore, in step S2, a crowd transmission delay model is established between the platform and the escalator entrance, assuming the distance from the platform center to the escalator entrance is L. path The average moving speed of the crowd is v walk (ρ); Defines the predicted flow rate q to the escalator entrance. arrive (t):

[0024]

[0025] In the formula, For smooth travel, Let ρ be the traffic density at time t. jam Blocking density;

[0026] Predicted flow q arrive (t) forms a continuous passenger flow inflow curve reaching the escalator entrance;

[0027] Predict the queuing status at the entrance, and define the number of people in the queue as N. queue The differential equation for (t):

[0028]

[0029] In the formula, C esc Based on the theoretical conveying capacity of the escalator, η is the escalator conveying speed. enter This is the entrance passage efficiency factor.

[0030] Furthermore, in step S3, the crowd is treated as a compressible fluid, and the partial differential equations of macroscopic crowd dynamics are solved using PINN.

[0031] First, define the governing equations for crowd dynamics, and define the crowd density field ρ(x,t) and velocity field u(x,t) as obeying the following conservation of mass and momentum:

[0032]

[0033] In the formula, c0 is the speed of sound for information propagation, and τ relax For relaxation time, U eq (ρ) represents the equilibrium velocity;

[0034] Secondly, construct the PINN network architecture. An adaptive gradient scaling layer and a physical residual loss function are introduced:

[0035]

[0036]

[0037] in, For the density field predicted by the network, For activation function, For adaptive gradient scaling / input scaling layers, Let be the weight matrix of the nth layer of the neural network; For physical residual loss, The total number of residual sampling points. The residuals of the mass conservation equation, The residuals of the momentum conservation equation are... The square of the L2 norm, Weighting coefficients for momentum residuals;

[0038] Perform event-driven boundary condition injection at the escalator entrance, where x=0, using the calculated q. arrive (t) Forced constraint on network boundaries: By introducing the indicator function Ⅱ(⋅), the boundary constraints are strengthened during the pulse arrival window, thereby improving the convergence speed.

[0039]

[0040] in, For boundary condition loss, This represents the total number of boundary sampling points. The entry point and time point predicted by the model pedestrian density, The entry point and time point predicted by the model pedestrian speed, For the pulse arrival window period;

[0041] Load reconfiguration based on PINN prediction utilizes the density distribution of PINN output. Calculate the real-time load mass distribution M on the escalator path. load (t):

[0042]

[0043] In the formula, For the equivalent mass on a single rung, The total length of the escalator. These are the coordinates of the position on the escalator;

[0044] Load resistance torque T L (t):

[0045]

[0046] θ is the tilt angle, g is the acceleration due to gravity, μ is the coefficient of friction, and D sprocket The diameter of the drive sprocket.

[0047] Furthermore, in step S4, the state-space model of the escalator electromechanical coupling system is established as follows:

[0048] Induction motor d q The dynamic equations of the shaft, in the synchronous rotating coordinate system, are as follows: (Equations for motor voltage and electromagnetic torque are missing from the original text.)

[0049]

[0050]

[0051] in, , Let represent the d-axis and q-axis components of the stator voltage in a synchronous rotating coordinate system. For stator winding resistance, , Let represent the components of the stator current along the d and q axes. , Let be the components of the stator flux linkage along the d and q axes. ω is the angular velocity of the synchronous rotating coordinate system; p is the number of pole pairs of the motor, L m L r For the mutual inductance between the stator and rotor, and the self-inductance of the rotor winding, , Let be the components of the rotor flux along the d and q axes;

[0052] Total moment of inertia J of the motor shaft sys The formula is as follows:

[0053]

[0054] in, Let ω be the angular velocity of the motor, and B be the coefficient of viscous friction of the system. This is the braking torque;

[0055] Instantaneous power P of escalator system elec (t) Includes copper losses, iron losses, and mechanical power output:

[0056]

[0057] In the formula, Where is the copper loss factor, and I is the stator current of the motor. This is the iron loss coefficient.

[0058] Furthermore, in step S4, the future load T predicted by PINN is used. L (t) is used as a feedforward disturbance. An MPC controller is designed to solve for the optimal speed curve.

[0059] Discretize the prediction model to construct state variables. Control variables :

[0060]

[0061] in, This represents the angular displacement of the escalator motor. Let be the angular velocity of the escalator motor. Let A be the reference electromagnetic torque of the motor, B be the system state matrix, and D be the system input matrix. for The estimated value;

[0062] Define a multi-objective optimization functional in the prediction time domain N. p Within, minimize the cost function J:

[0063]

[0064] Among them, the weight coefficients of the three sub-objectives α, β, and γ;

[0065] Energy consumption sub-target J energy By utilizing the characteristics of variable frequency speed control, the load on the escalator at low speeds can be reduced.

[0066]

[0067] in, The coefficients of energy consumption models c1 and c2 are the sampling period;

[0068] Introducing the mechanical wear sub-target J wear and J jerk Acceleration penalty is used to suppress rapid acceleration and frequent start-stop cycles.

[0069]

[0070]

[0071] Where Q1 and Q2 are weight matrices;

[0072] Introducing a queuing congestion penalty J queue Using an exponential penalty function, when the number of people in the queue approaches the safe threshold N... max At that time, the escalator speed is forcibly increased:

[0073]

[0074] In the formula, This is the penalty coefficient.

[0075] Furthermore, in step S5, when solving the optimization problem, the safety speed constraint, comfort acceleration constraint, and motor capacity constraint are retained:

[0076]

[0077] In the formula, T is the rated angular velocity of the motor. max This is the torque limiting curve;

[0078] The augmented Lagrangian function is constructed using the ADMM solver.

[0079]

[0080] In the formula, The Lagrange multiplier corresponding to the i-th constraint For the i-th equality constraint, ρ is the augmented term penalty coefficient;

[0081] The iterative update strategy and the hard constraints on processing speed using the soft threshold operator are as follows:

[0082]

[0083]

[0084] In the formula, Let A and b be the control variables for the (k+1)th iteration, and let A and b be the linear matrices and vectors of the constraints, corresponding to the aforementioned safety speed constraints, comfort acceleration constraints, and motor capacity constraints. For the Lagrange multiplier in the k-th iteration, The final output motor angular velocity, The unconstrained rotational speed obtained in the previous update is given by S, which is the interval projection operator that restricts the output value to the interval [0, ..., ... ].

[0085] In another aspect, this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and when the computer program is invoked and executed by the processor, it implements the steps in the method described above.

[0086] In another aspect, this application provides a computer-readable medium storing a computer program that, when executed by a computer, implements the steps in the method described above.

[0087] In summary, compared with the prior art, the beneficial effects of this application are as follows:

[0088] 1. It solves the problems of "response lag" and "abrupt start" in traditional infrared sensing control and realizes soft start.

[0089] Existing technical limitations: Traditional infrared sensing or photoelectric switch control schemes are passively triggered; the escalator only accelerates rapidly from idle speed to rated speed when a passenger has already stepped into the sensing area. This delayed response not only causes passengers to lose their balance but also generates huge surge currents and mechanical stress on the motor and chain.

[0090] Advantages of this invention: This invention introduces train timetables and passenger information as prior information, and combines them with a crowd transmission delay model, enabling the frequency converter to start smooth acceleration within a preset time window before the pulse passenger flow arrives at the escalator entrance.

[0091] Beneficial effects: It eliminates the jerky feeling when passengers board the elevator, improving the comfort of riding the elevator; at the same time, it suppresses the inrush current at the moment of start-up, extending the life of the motor.

[0092] 2. Improve the ability of subway escalators to cope with congestion and trampling.

[0093] Current technological challenges: Existing variable frequency energy-saving escalators typically switch operating speeds only based on whether there are passengers or not. When faced with a sudden surge of passengers released by a subway train arriving at the station, if the escalator is still operating at a low speed or cannot accelerate sufficiently, crowds can easily accumulate at the entrance, potentially leading to stampedes.

[0094] Advantages of this invention: This invention introduces a congestion penalty term into the MPC optimization objective. When it is predicted that the upcoming passenger flow density may exceed the entrance passage threshold, the control system will forcibly ignore the energy-saving objective and raise the escalator to high-capacity mode in advance for rapid evacuation.

[0095] Beneficial effects: Intelligent decision-making that ensures safety while taking energy conservation into account reduces the risk of passenger congestion during peak hours in the subway caused by energy-saving strategies.

[0096] 3. By using PINN to map passenger flow data to escalator load, the energy efficiency ratio of escalator operation is improved.

[0097] Existing technical pain points: Ordinary control methods cannot predict the specific weight and distribution of passengers, and the motor is often in an inefficient "overpowered" state, resulting in low energy efficiency.

[0098] Advantages of this invention: This invention utilizes a Physical Information Neural Network (PINN) to solve the crowd dynamics equations, enabling precise reconstruction of the real-time mass distribution along the escalator path and calculation of future dynamic load torque. The controller then adjusts the motor's operation accordingly, ensuring it always operates at the optimal slip frequency.

[0099] Beneficial effects: Compared with the traditional constant voltage-frequency ratio (V / f) control, this solution effectively improves the energy efficiency ratio of escalators while ensuring power. Attached Figure Description

[0100] Figure 1 This is a flowchart illustrating an embodiment of the escalator control method for subway pulsed passenger flow according to the present invention.

[0101] Figure 2 This is a schematic diagram of the PINN neural network principle in one embodiment of the present invention;

[0102] Figure 3 This is a schematic diagram of the MPC rolling optimization control logic in one embodiment of the present invention;

[0103] Figure 4 This is a comparison chart of the speed curves of the present invention and the prior art;

[0104] Figure 5 This is a comparison chart of energy consumption between the present invention and existing technologies;

[0105] Figure 6 This is a schematic diagram of the architecture of an escalator control system for subway pulse-type passenger flow in one embodiment of the present invention. Detailed Implementation

[0106] The specific embodiments of this application are described in detail below with reference to the accompanying drawings.

[0107] One specific embodiment of this application provides an escalator control method for pulsed passenger flow in subways. It replaces complex black-box model prediction with an improved Physical Information Neural Network (PINN), thereby enabling the field deployment of Model Predictive Control (MPC) and improving the practical application capability of this technology. The main problem addressed by this invention is the contradiction between "high-speed idle operation" and "sudden passenger congestion" in dealing with pulsed passenger flow.

[0108] refer to Figure 1 The method includes the following steps:

[0109] S0: Obtain subway arrival times and passenger flow data, which are obtained from the subway stations.

[0110] S1: Based on subway train operation data, construct a pulse-type passenger flow trigger sequence based on discrete events.

[0111] Traditional escalator operation typically relies solely on real-time passenger volume control triggered by infrared sensors, neglecting the fluctuating passenger flow characteristics of subways, particularly during peak hours. To address this issue, this invention first constructs a continuous spatiotemporal evolution process to describe passenger flow changes.

[0112] First, construct a pulse-based subway passenger flow evolution model:

[0113] Define a train arrival event function, assuming the subway station's operating time domain is t∈[0,T], and the scheduled arrival time of the i-th train is τ. i The discrete event of a train entering the station is transformed into a trigger function E(t) on the time axis:

[0114]

[0115] In the formula, δ(⋅) is the Dirac function, and N train Total number of train services throughout the day;

[0116] Secondly, construct an asymmetric passenger flow generation rate model:

[0117] Considering that the process of passengers alighting and exiting the station has a characteristic of being fast at first and then slowing down, simply treating it as an instantaneous pulse would lead to problems such as the escalator response being too sensitive. Therefore, an asymmetric Gaussian kernel function is introduced to convolve the function of this event to obtain the instantaneous passenger flow generation rate λ. in (t):

[0118]

[0119] In the formula, Q load (τ) represents the passenger capacity of the corresponding train, μ lag σ represents the average delay time of passengers exiting the train doors. t A time-scale parameter characterizing the degree of population dispersion;

[0120] Instantaneous passenger flow generation rate λ in (t) constitutes the passenger flow trigger sequence.

[0121] S2: Based on the passenger flow trigger sequence and the crowd space transmission delay model, calculate the continuous passenger flow injection curve and queuing status at the escalator entrance.

[0122] Establish a crowd transmission delay model between the platform and the escalator entrance, assuming the distance from the platform center to the escalator entrance is L. path The average moving speed of the crowd is v walk (ρ); Defines the predicted flow rate q to the escalator entrance. arrive (t):

[0123]

[0124] In the formula, For smooth travel, Let ρ be the traffic density at time t. jam Blocking density;

[0125] Predicted flow q arrive (t) forms a continuous passenger flow inflow curve reaching the escalator entrance;

[0126] Predict the queuing status at the entrance, and define the number of people in the queue as N. queue The differential equation for (t):

[0127]

[0128] In the formula, C esc Based on the theoretical conveying capacity of the escalator, η is the escalator conveying speed. enter This is the entrance passage efficiency factor.

[0129] S3: The passenger flow injection curve is used as the boundary condition input to the physical information neural network PINN to solve the crowd dynamics equation to predict the future dynamic load torque of the escalator.

[0130] To predict how passenger flow information translates into mechanical load on escalators, crowds are treated as compressible fluids, and PINN is used to solve the partial differential equations of macroscopic crowd dynamics.

[0131] First, define the governing equations for crowd dynamics, and define the crowd density field ρ(x,t) and velocity field u(x,t) as obeying the following conservation of mass and momentum:

[0132]

[0133] In the formula, c0 is the speed of sound for information propagation, and τ relax For relaxation time, U eq (ρ) represents the equilibrium velocity;

[0134] Secondly, construct the PINN network architecture. To mitigate the abrupt change problem caused by pulse boundaries, an adaptive gradient scaling layer and a physical residual loss function are introduced:

[0135]

[0136]

[0137] in, For the density field predicted by the network, For activation function, For adaptive gradient scaling / input scaling layers, Let be the weight matrix of the nth layer of the neural network; For physical residual loss, The total number of residual sampling points. The residuals of the mass conservation equation, The residuals of the momentum conservation equation are... The square of the L2 norm, Weighting coefficients for momentum residuals;

[0138] Perform event-driven boundary condition injection at the escalator entrance, where x=0, using the calculated q. arrive (t) Forced constraint on network boundaries: By introducing the indicator function Ⅱ(⋅), the boundary constraints are strengthened during the pulse arrival window, thereby improving the convergence speed.

[0139]

[0140] in, For boundary condition loss, This represents the total number of boundary sampling points. The entry point and time point predicted by the model pedestrian density, The entry point and time point predicted by the model pedestrian speed, For the pulse arrival window period;

[0141] Load reconfiguration based on PINN prediction utilizes the density distribution of PINN output. Calculate the real-time load mass distribution M on the escalator path. load (t):

[0142]

[0143] In the formula, For the equivalent mass on a single rung, The total length of the escalator. These are the coordinates of the position on the escalator;

[0144] Load resistance torque T L (t):

[0145]

[0146] θ is the tilt angle, g is the acceleration due to gravity, μ is the coefficient of friction, and D sprocket The diameter of the drive sprocket.

[0147] S4: Based on the predicted dynamic load torque, establish an electromechanical coupling state-space model of the escalator for calculating the system energy consumption and speed evolution; with the comprehensive objective of minimizing system operating energy consumption, mechanical wear and queuing congestion penalty, construct a multi-objective nonlinear model predictive control optimization problem.

[0148] To achieve accurate speed tracking and energy consumption calculation, a dynamic model incorporating the characteristics of the induction motor and the mechanical transmission chain is established.

[0149] Induction motor d q The dynamic equations of the shaft, in the synchronous rotating coordinate system, are as follows: (Equations for motor voltage and electromagnetic torque are missing from the original text.)

[0150]

[0151]

[0152] in, , Let represent the d-axis and q-axis components of the stator voltage in a synchronous rotating coordinate system. For stator winding resistance, , Let represent the components of the stator current along the d and q axes. , Let be the components of the stator flux linkage along the d and q axes. ω is the angular velocity of the synchronous rotating coordinate system; p is the number of pole pairs of the motor, L m L r For the mutual inductance between the stator and rotor, and the self-inductance of the rotor winding, , Let be the components of the rotor flux along the d and q axes;

[0153] Total moment of inertia J of the motor shaft sys The formula is as follows:

[0154]

[0155] in, Let ω be the angular velocity of the motor, and B be the coefficient of viscous friction of the system. This is the braking torque;

[0156] The state-space model of the escalator electromechanical coupling system is established as follows:

[0157] Instantaneous power P of escalator system elec (t) Includes copper losses, iron losses, and mechanical power output:

[0158]

[0159] In the formula, Where is the copper loss factor, and I is the stator current of the motor. This is the iron loss coefficient.

[0160] Using PINN to predict future load T L (t) is used as a feedforward disturbance. An MPC controller is designed to solve for the optimal speed curve.

[0161] Discretize the prediction model to construct state variables. Control variables :

[0162]

[0163] in, This represents the angular displacement of the escalator motor. Let be the angular velocity of the escalator motor. Let A be the reference electromagnetic torque of the motor, B be the system state matrix, and D be the system input matrix. for The estimated value;

[0164] Define a multi-objective optimization functional in the prediction time domain N. p Within, minimize the cost function J:

[0165]

[0166] Among them, the weight coefficients of the three sub-objectives α, β, and γ;

[0167] Energy consumption sub-target J energy By utilizing the characteristics of variable frequency speed control, the load on the escalator at low speeds can be reduced.

[0168]

[0169] in, The coefficients of energy consumption models c1 and c2 are the sampling period;

[0170] Introducing the mechanical wear sub-target J wear and J jerk Acceleration penalty is used to suppress rapid acceleration and frequent start-stop cycles.

[0171]

[0172]

[0173] Where Q1 and Q2 are weight matrices;

[0174] Introducing a queuing congestion penalty J queue Using an exponential penalty function, when the number of people in the queue approaches the safe threshold N... max At that time, the escalator speed is forcibly increased:

[0175]

[0176] In the formula, This is the penalty coefficient.

[0177] After obtaining continuous passenger arrival rates, the core process of load torque prediction begins. For example... Figure 2 As shown, to overcome the shortcomings of traditional control methods that cannot accurately determine mechanical load due to a lack of information, this embodiment employs an observation architecture based on a Physical Information Neural Network (PINN). This network uses the current time and escalator spatial location as input layer data, and through the nonlinear mapping of a deep neural network, outputs predicted crowd density and movement speed. More importantly, this architecture embeds a physical information constraint module at the bottom layer, which can automatically calculate the partial derivatives of the output with respect to time and space, and substitute them into the mass conservation equation and momentum conservation equation. The system extracts the computational residuals of these physical equations and incorporates them as penalty terms into the optimization process of the total loss function. This mechanism forces the output of the neural network to strictly follow objective physical laws, enabling the system to obtain the dynamic mechanical load torque of the escalator at various future times based solely on the predicted density distribution.

[0178] After accurately acquiring the predicted dynamic load torque value, the controller immediately starts as follows: Figure 3 The multi-objective model predictive control (MPC) rolling optimization logic is shown. In this control logic, the system constructs a comprehensive cost function, which includes not only an energy consumption term aimed at minimizing the motor input power, but also a wear term to suppress drastic changes in acceleration. To ensure passenger safety, an exponential congestion penalty term based on the entrance queue length is also embedded in the cost function. When the system predicts that the number of people queuing at the escalator entrance is about to reach a safe threshold, the value of this penalty term increases explosively, thereby forcibly maintaining high capacity on the escalator. The controller uses the alternating direction multiplier method (ADMM) to solve this cost function in a rolling manner within a set prediction time domain.

[0179] S5: Solve the optimization problem, generate the optimal speed control command, and send it to the escalator frequency converter for execution.

[0180] To ensure a good user experience, safety speed constraints, comfort acceleration constraints, and motor capacity constraints were retained when solving the optimization problem.

[0181]

[0182] In the formula, T is the rated angular velocity of the motor. max This is the torque limiting curve;

[0183] The augmented Lagrangian function is constructed using the ADMM solver.

[0184]

[0185] In the formula, The Lagrange multiplier corresponding to the i-th constraint For the i-th equality constraint, ρ is the augmented term penalty coefficient;

[0186] The iterative update strategy and the hard constraints on processing speed using the soft threshold operator are as follows:

[0187]

[0188]

[0189] In the formula, Let A and b be the control variables for the (k+1)th iteration, and let A and b be the linear matrices and vectors of the constraints, corresponding to the aforementioned safety speed constraints, comfort acceleration constraints, and motor capacity constraints. For the Lagrange multiplier in the k-th iteration, The final output motor angular velocity, The unconstrained rotational speed obtained in the previous update is given by S, which is the interval projection operator that restricts the output value to the interval [0, ..., ... ].

[0190] The controller sends the first optimal frequency command obtained from the optimization solution to the frequency converter in real time, driving the motor to run smoothly. The control strategy adopted in this embodiment has demonstrated significant technical advantages in actual operation. Figure 4 As shown in the speed curve comparison chart, traditional passive infrared triggering technology suffers from severe response lag, often only being passively triggered when a large passenger flow has already arrived at the escalator entrance. This causes the escalator to accelerate rapidly from idle speed to rated speed, easily causing passengers to lose their balance and creating a dangerous congestion bottleneck at the entrance. In contrast, the control system of this invention benefits from prior knowledge of the timetable and the PINN prediction algorithm, enabling it to predict the train's arrival time in advance and control the escalator to slowly pre-accelerate along a smooth S-shaped curve 10 to 15 seconds before the actual arrival of passengers. When the first batch of passengers step onto the steps, the escalator smoothly transitions to its rated operating speed.

[0191] Furthermore, this embodiment has also made progress in energy conservation and consumption reduction. For example... Figure 5As shown in the bar chart comparing energy consumption and power, this embodiment significantly reduces the peak power demand of the escalator system from the original 45kW to 32kW, substantially reducing the instantaneous power required by the equipment. Simultaneously, the real-time precise matching of dynamic load and motor energy efficiency achieved through the MPC algorithm ensures that the traction motor operates within its optimal efficiency range most of the time, avoiding ineffective excitation losses. Actual measurement data shows that after adopting the control system of this invention, the average daily total energy consumption of the escalator has significantly decreased from the traditional 120kWh to 85kWh, with a comprehensive energy saving rate as high as 29.2%. In summary, this embodiment, through the deep integration of physical information neural networks and model predictive control technology, fills the technical gap where energy saving and safety are difficult to balance, and has high engineering application value.

[0192] In another embodiment, the method described in this application is deployed in escalators within subway stations to form an intelligent energy-saving control system for escalators designed for the pulse-like passenger flow of subways. Figure 6 As shown, the system's hardware architecture is mainly divided into a perception and data layer, a decision and control layer, and an execution and physical layer. The perception and data layer acquires real-time train arrival time data from the subway train automatic monitoring system via a communication interface, and combines this with real-time passenger flow correction data collected by sensors deployed at escalator entrances and subway station exits. The core of the decision and control layer is an edge industrial controller integrating a Physical Information Neural Network (PINN) and a Model Predictive Control (MPC) optimizer. This controller performs complex calculations based on the acquired data, and then sends the generated control commands to the variable frequency drive in the execution and physical layer, thereby adjusting the motor speed and receiving real-time feedback on motor current and speed to form a closed-loop control.

[0193] Another specific embodiment of this application provides an electronic device, including a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the steps in the method described above.

[0194] Another specific embodiment of this application provides a computer-readable medium storing a computer program, which, when executed by a computer, implements the steps in the method described above.

[0195] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the inventive concept of this application, and these all fall within the protection scope of this application.

Claims

1. A method for controlling escalators in response to pulsed passenger flow in subway systems, characterized in that: Includes the following steps: S1: Construct a pulse-based passenger flow trigger sequence based on discrete events, using subway train operation data; S2: Based on the passenger flow trigger sequence and the crowd space transmission delay model, calculate the continuous passenger flow injection curve and queuing status at the escalator entrance; S3: The passenger flow injection curve is used as a boundary condition input to the physical information neural network PINN to solve the crowd dynamics equation in order to predict the future dynamic load torque of the escalator. S4: Based on the predicted dynamic load torque, establish an electromechanical coupling state-space model of the escalator for calculating the system's energy consumption and speed evolution; S5: Construct a multi-objective nonlinear model predictive control optimization problem with the comprehensive objective of minimizing system operating energy consumption, mechanical wear and queuing congestion penalties; S6: Solve the optimization problem, generate the optimal speed control command, and send it to the escalator frequency converter for execution.

2. The escalator control method for subway pulse-type passenger flow according to claim 1, characterized in that, In step S1, firstly, a pulse-type subway passenger flow evolution model is constructed: Let the operating time domain of the subway station be t∈[0,T], and the scheduled arrival time of the i-th train be τ. i The discrete event of a train entering the station is transformed into a trigger function E(t) on the time axis: In the formula, δ(⋅) is the Dirac function, and N train Total number of train services throughout the day; Secondly, construct an asymmetric passenger flow generation rate model: By introducing an asymmetric Gaussian kernel function to convolve the function of this event, the instantaneous passenger flow generation rate λ is obtained. in (t): In the formula, Q load (τ) represents the passenger capacity of the corresponding train, μ lag σ represents the average delay time of passengers exiting the train doors. t A time-scale parameter characterizing the degree of population dispersion; Instantaneous passenger flow generation rate λ in (t) constitutes the passenger flow trigger sequence.

3. The escalator control method for subway pulse-type passenger flow according to claim 2, characterized in that, In step S2, a crowd transmission delay model is established between the platform and the escalator entrance, assuming the distance from the platform center to the escalator entrance is L. path The average moving speed of the crowd is v walk (ρ); Defines the predicted flow rate q to the escalator entrance. arrive (t): In the formula, For smooth travel, Let ρ be the traffic density at time t. jam Blocking density; Predicted flow q arrive (t) forms a continuous passenger flow inflow curve reaching the escalator entrance; Predict the queuing status at the entrance, and define the number of people in the queue as N. queue The differential equation for (t): In the formula, C esc Based on the theoretical conveying capacity of the escalator, η is the escalator conveying speed. enter This is the entrance passage efficiency factor.

4. The escalator control method for subway pulsed passenger flow according to claim 3, characterized in that, In step S3, the crowd is treated as a compressible fluid, and the partial differential equations of macroscopic crowd dynamics are solved using PINN. First, define the governing equations for crowd dynamics, and define the crowd density field ρ(x,t) and velocity field u(x,t) as obeying the following conservation of mass and momentum: In the formula, c0 is the speed of sound for information propagation, and τ relax For relaxation time, U eq (ρ) represents the equilibrium velocity; Secondly, construct the PINN network architecture. An adaptive gradient scaling layer and a physical residual loss function are introduced: in, For the density field predicted by the network, For activation function, For adaptive gradient scaling / input scaling layers, Let be the weight matrix of the nth layer of the neural network; For physical residual loss, The total number of residual sampling points. The residuals of the mass conservation equation, The residuals of the momentum conservation equation are... The square of the L2 norm, Weighting coefficients for momentum residuals; Perform event-driven boundary condition injection at the escalator entrance, where x=0, using the calculated q. arrive (t) Forced constraint on network boundaries: By introducing the indicator function Ⅱ(⋅), the boundary constraints are strengthened during the pulse arrival window, thereby improving the convergence speed. in, For boundary condition loss, This represents the total number of boundary sampling points. The entry point and time point predicted by the model pedestrian density, The entry point and time point predicted by the model pedestrian speed, For the pulse arrival window period; Load reconfiguration based on PINN prediction utilizes the density distribution of PINN output. Calculate the real-time load mass distribution M on the escalator path. load (t): In the formula, For the equivalent mass on a single rung, The total length of the escalator. These are the coordinates of the position on the escalator; Load resistance torque T L (t): θ is the tilt angle, g is the acceleration due to gravity, μ is the coefficient of friction, and D sprocket The diameter of the drive sprocket.

5. The escalator control method for subway pulsed passenger flow according to claim 4, characterized in that, In step S4, the state-space model of the escalator electromechanical coupling system is established as follows: Induction motor d q The dynamic equations of the shaft, in the synchronous rotating coordinate system, are as follows: (Equations for motor voltage and electromagnetic torque are missing from the original text.) in, , Let represent the d-axis and q-axis components of the stator voltage in a synchronous rotating coordinate system. For stator winding resistance, , Let represent the components of the stator current along the d and q axes. , Let be the components of the stator flux linkage along the d and q axes. ω is the angular velocity of the synchronous rotating coordinate system; p is the number of pole pairs of the motor, L m L r For the mutual inductance between the stator and rotor, and the self-inductance of the rotor winding, , Let be the components of the rotor flux along the d and q axes; Total moment of inertia J of the motor shaft sys The formula is as follows: in, Let ω be the angular velocity of the motor, and B be the coefficient of viscous friction of the system. This is the braking torque; Instantaneous power P of escalator system elec (t) Includes copper losses, iron losses, and mechanical power output: In the formula, Where is the copper loss factor, and I is the stator current of the motor. This is the iron loss coefficient.

6. The escalator control method for subway pulse-type passenger flow according to claim 5, characterized in that, In step S5, the future load T predicted by PINN is used. L (t) is used as a feedforward disturbance. An MPC controller is designed to solve for the optimal speed curve. Discretize the prediction model to construct state variables. Control variables : in, This represents the angular displacement of the escalator motor. Let be the angular velocity of the escalator motor. Let A be the reference electromagnetic torque of the motor, B be the system state matrix, and D be the system input matrix. for The estimated value; Define a multi-objective optimization functional in the prediction time domain N. p Within, minimize the cost function J: Among them, the weight coefficients of the three sub-objectives α, β, and γ; Energy consumption sub-target J energy By utilizing the characteristics of variable frequency speed control, the load on the escalator at low speeds can be reduced. in, The coefficients of energy consumption models c1 and c2 are the sampling period; Introducing the mechanical wear sub-target J wear and J jerk Acceleration penalty is used to suppress rapid acceleration and frequent start-stop cycles. Where Q1 and Q2 are weight matrices; Introducing a queuing congestion penalty J queue Using an exponential penalty function, when the number of people in the queue approaches the safe threshold N... max At that time, the escalator speed is forcibly increased: In the formula, This is the penalty coefficient.

7. The escalator control method for subway pulse-type passenger flow according to claim 6, characterized in that, In step S6, when solving the optimization problem, the safety speed constraint, comfort acceleration constraint, and motor capacity constraint are retained: In the formula, T is the rated angular velocity of the motor. max This is the torque limiting curve; The augmented Lagrangian function is constructed using the ADMM solver. In the formula, The Lagrange multiplier corresponding to the i-th constraint For the i-th equality constraint, ρ is the augmented term penalty coefficient; The iterative update strategy and the hard constraints on processing speed using the soft threshold operator are as follows: In the formula, Let A and b be the control variables for the (k+1)th iteration, and let A and b be the linear matrices and vectors of the constraints, corresponding to the aforementioned safety speed constraints, comfort acceleration constraints, and motor capacity constraints. For the Lagrange multiplier in the k-th iteration, The final output motor angular velocity, The unconstrained rotational speed obtained in the previous update is given by S, which is the interval projection operator that restricts the output value to the interval [0, ..., ... ].

8. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is invoked and executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

9. A computer-readable medium, characterized in that, The computer-readable medium stores a computer program that, when executed by a computer, implements the steps of the method as described in any one of claims 1-7.

Images