Self-adaptive FOC double-loop control method and system of permanent magnet synchronous traction machine for elevator

Abstract

The invention discloses a self-adaptive FOC double-loop control method and system for a permanent magnet synchronous traction machine for an elevator, and the method comprises the steps: completing the initialization configuration of hardware peripherals and clustering algorithm parameters after an elevator controller is powered on, collecting the operation state parameters of the elevator, and carrying out the filtering processing; then, according to the current working condition characteristics, a clustering center is iterated through a clustering algorithm, the membership degree of each working condition is obtained, speed ring parameters are dynamically adjusted in real time, and dynamic adaptation is achieved; and finally, through a mechanical vibration model and a self-adaptive frequency band-pass filter, extracting a vibration component from a motor rotating speed signal, generating a vibration suppression current instruction, superposing the vibration suppression current instruction with an output q-axis current instruction of a speed loop, and outputting the superposed vibration suppression current instruction as a reference current into a current loop controller together with the reference current. Based on multi-working-condition classification, self-adaptive adjustment of speed ring parameters is achieved, the dynamic response time is shortened, and the overshoot is reduced; elastic vibration feedforward control is introduced into a current loop, vibration of the lift car is restrained, and operation stability is improved.

Description

Technical Field

[0001] This invention relates to the field of elevator traction machine control technology, specifically to an adaptive field-oriented control (FOC) dual-loop system driven by elevator-customized fuzzy C-means clustering (EC-FCM). It is particularly designed for permanent magnet synchronous traction machines (PMSM) in the specific operating conditions of elevators, such as "large-range load fluctuation (0~1000kg) + dynamic switching of operation phases (start / cruise / deceleration) + mechanical elastic coupling (wire rope vibration)". It achieves precise adaptation of speed loop PI parameters and active suppression of current loop vibration, belonging to the intersection of industrial control and elevator safety technology. Background Technology

[0002] With the acceleration of urbanization, elevators, as the core equipment for vertical transportation in high-rise buildings, have seen their operational safety, comfort, and energy efficiency become the focus of industry attention. Permanent Magnet Synchronous Motors (PMSMs), due to their advantages such as high efficiency, energy saving, and low noise, have gradually replaced traditional asynchronous traction machines as the mainstream choice. FOC control technology, by decoupling the stator current into excitation current (id) and torque current (iq), achieves high-precision torque and speed control of PMSMs, and is currently the mainstream solution for traction machine control.

[0003] However, elevator operation is significantly complex, with conditions such as large load fluctuations (from no-load to rated load, typically 0~1000kg), dynamic switching between operating phases (start, acceleration, cruise, deceleration, leveling), and elastic coupling in the mechanical system (car vibration caused by wire rope deformation). The torque demand difference between no-load (0kg) and rated load (e.g., 1000kg) can be more than 10 times. Rapid response is required during start / acceleration, steady-state precision during cruise, and no overshoot during deceleration / leveling. Traditional FOC dual-loop control (speed + current loop) uses a fixed-parameter PI controller, which is ill-suited to these complex conditions, presenting a technical bottleneck. For example, insufficient speed loop parameter adaptability leads to response lag and overshoot during sudden load changes or operating state switching (during heavy-load start-up, a fixed Kp will cause response lag (prolonged start-up time), and a fixed Ki will cause integral saturation; while during light-load deceleration, parameter mismatch will lead to speed overshoot (decreased leveling accuracy)), affecting the smoothness and comfort of elevator operation. Meanwhile, fixed-parameter PI controllers struggle to balance dynamic performance and steady-state accuracy under different operating conditions, further limiting the overall control effect of the system. In existing technologies, clustering algorithms are mostly applied to industrial motors and robots (e.g., general-purpose FCMs are used for motor speed classification), but are not designed specifically for elevator characteristics. Feature vectors are not bound to core elevator requirements (general-purpose algorithms often select current and voltage, failing to incorporate the key elevator load indicator of "car load"); cluster centers do not match elevator operating stages (general-purpose algorithms have random cluster numbers, failing to classify multiple operating conditions according to the elevator); and the iterative logic does not meet the real-time requirements of elevators (general-purpose FCMs have long iteration cycles, unable to adapt to the millisecond-level switching between elevator start-up, cruising, and deceleration).

[0004] Secondly, the current loop does not consider the mechanical characteristics of the elevator. As the inner loop, the control accuracy of the current loop directly affects the stability of the torque output. However, traditional current loops are designed only based on the mathematical model of the motor and do not incorporate the unique load and mechanical vibration factors of the elevator. The steel wire rope of the traction system can be regarded as an elastic body, which will generate periodic vibrations during the starting and braking phases, which are transmitted to the car and cause perceptible bumps (vibration acceleration > 0.15 m / s²). 2 (This can affect comfort).

[0005] In the existing technology, the relevant improvement solutions have limitations. Patent CN114826072A proposes speed loop optimization based on phase-locked loop, which improves speed stability, but does not involve load adaptation and vibration control. Patent CN113992099B suppresses field weakness runaway by limiting current loop amplitude, but its core is to protect the motor and does not solve the problem of working condition adaptation. For the problem of PI controller parameter tuning, the most widely used method is fuzzy control, but it does not take into account the segmented characteristics of elevator load, and its adaptability is limited.

[0006] Therefore, developing a customized EC-FCM clustering algorithm combined with mechanical vibration coupling current compensation scheme for elevator scenarios, tailored to the characteristics of elevator operating conditions, is of great significance for improving elevator operating performance. Summary of the Invention

[0007] This invention aims to address the aforementioned shortcomings of traditional FOC dual-loop control in elevator applications, and provides an adaptive FOC dual-loop control method and system for permanent magnet synchronous traction machines in elevators, achieving the following objectives: First, based on the multi-condition classification of elevator heavy load and light load, real-time adaptive adjustment of the speed loop PI parameters is achieved, shortening the dynamic response time and reducing overshoot; second, elastic vibration feedforward control is introduced into the current loop to suppress car vibration and improve operational stability.

[0008] The objective of this invention is achieved through the following technical solution: an adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators, the method comprising the following steps: S1: After the elevator controller is powered on, it completes the initialization configuration of hardware peripherals and clustering algorithm parameters; S2: Collect elevator operating status parameters and filter them; S3: Based on the characteristics of the current working conditions, the clustering center is iterated through a clustering algorithm to obtain the membership degree of each working condition, and the speed loop PI parameter is dynamically adjusted in real time to achieve dynamic adaptation; S4: By using a mechanical vibration model and an adaptive frequency bandpass filter, the vibration component is extracted from the motor speed signal to generate a vibration suppression current command.

[0009] S5: The generated vibration suppression command is superimposed with the output q-axis current command of the speed loop and output together as a reference current into the current loop PI controller.

[0010] Furthermore, the hardware peripheral initialization mentioned in S1 includes configuring the peripheral configuration of the DSP chip. The initialization configuration of the algorithm parameters is to load pre-configured parameters from the Flash memory, including EC-FCM clustering parameters, PI parameter table, and motor parameters. Among them, the EC-FCM clustering parameters include the initial clustering center matrix of 6 working conditions obtained by coupling the heavy load / light load level with the start / cruise / deceleration operation stage.

[0011] EC-FCM clustering parameters include an initial cluster center matrix for 6 operating conditions:

[0012] In the formula: The rotational speed of the traction machine corresponding to the kth cluster center, in rad / s; The rotational acceleration corresponding to the kth cluster center, in rad / s. 2 ; Let represent the total load of the car corresponding to the k-th cluster center, in kg. The cluster centers corresponding to these six operating conditions, and the optimal PI parameters for each condition, were determined through offline experiments.

[0013] Furthermore, the elevator operating status parameters described in S2 are collected by acquiring current values ​​through a three-phase current sensor and simultaneously obtaining the motor's mechanical speed through an encoder. and rotor position angle A scenario-based filtering strategy is adopted for... A sliding average filter is used, and the data W from the pressure sensor at the bottom of the car is read and filtered using a sliding window.

[0014] Furthermore, in S3, based on the current operating condition characteristics, the speed loop PI parameters are adjusted in real time through EC-FCM mean clustering to achieve dynamic adaptation and ensure that each feature is bound to the control requirements. The feature vector is as follows:

[0015] In the formula: The normalized rotational speed of the motor at time i, in rad / s; and Here, represents the normalized motor speed and acceleration at time i, and the elevator car load, respectively, in rad / s and kg.

[0016] Furthermore, the membership degree u of the current sample for the six identified cluster centers is calculated for the six working conditions. ik The formula is as follows:

[0017] In the formula: The function represents the membership degree of the current elevator operating state to the k-th type of operating condition; x(i) represents the elevator feature vector at the current sampling time; C k This represents the cluster center of the k-th working condition; m is a fuzzy factor that controls the degree of fuzziness in the membership degree. The parameters of the PI controller are dynamically adjusted based on the membership degree of the parameters during elevator operation to achieve a smooth transition of parameters. The current PI parameters are calculated by weighting the membership degree in real time.

[0018] Furthermore, in S4, the motor speed is measured in real time. A bandpass filter is used to extract the pure vibration component from the motor speed signal to filter out low-frequency operating commands and high-frequency noise. The natural frequency of the wire rope is automatically tracked as the car position and load change. Based on this, a virtual impedance model is constructed, and a q-axis current compensation amount that is proportional to the vibration displacement and velocity but opposite in direction is continuously generated. This compensation is then executed immediately through the current loop with the fastest response.

[0019] Furthermore, the center frequency f of the bandpass filtercenter Dynamically configured by an independent frequency identification unit, which performs real-time spectrum analysis on the motor speed signal acquired by the encoder to calculate the dominant frequency f of the current vibration energy. d This value is used as the center frequency f of the bandpass filter. center The set value; secondly, the bandpass filter receives the center frequency configuration command f. center The filter internally adjusts according to the currently configured center frequency f. center With bandwidth BW, the coefficients of its digital filter are calculated in real time through a coefficient update algorithm, thereby dynamically adjusting its frequency response characteristics to ensure that its passband center is always locked at the current dominant vibration frequency.

[0020] Furthermore, the calculation of the center frequency is based on a two-stage identification architecture of "coarse positioning-fine calculation". The first stage, coarse positioning, uses short-time Fourier transform (STFT) to perform spectral analysis on the current loop vibration signal within a set time window, quickly scanning the vibration energy distribution of the elevator wire rope to obtain candidate frequencies and calculate their corresponding energy values ​​to ensure dynamic response speed. The second stage, fine calculation, is the fine positioning stage, which uses an improved Goertzel algorithm to accurately identify candidate frequencies and introduces an energy weighting coefficient to optimize the allocation of computational resources, with higher energy having a greater weight.

[0021] Furthermore, a three-element linkage model of "center frequency-load-bandwidth" is established to dynamically adjust the filter bandwidth. The bandwidth calculation formula is as follows:

[0022] In the formula: BW is the bandwidth at time t; BW0 is the preset base bandwidth; k f k represents the frequency weighting coefficient. W W is the load weighting factor; W(t) is the real-time load of the car; W max This is the rated load of the car.

[0023] On the other hand, the present invention also provides an adaptive FOC dual-loop control system for a permanent magnet synchronous traction machine for elevators, the system comprising: The configuration initialization module is used to complete the initialization configuration of hardware peripherals and clustering algorithm parameters after the elevator controller is powered on. The parameter filtering module is used to collect elevator operating status parameters and then filter them. The parameter dynamic adaptation module is used to obtain the membership degree of each working condition by iteratively clustering the cluster centers through a clustering algorithm based on the characteristics of the current working condition, and dynamically adjust the speed loop PI parameter in real time to achieve dynamic adaptation. The vibration suppression current generation module is used to extract vibration components from the motor speed signal through a mechanical vibration model and an adaptive frequency bandpass filter, and generate vibration suppression current commands.

[0024] The reference current output module is used to superimpose the generated vibration suppression command with the output q-axis current command of the velocity loop, and output them together as a reference current into the current loop PI controller.

[0025] The beneficial effects of this invention are: (1) Significantly improved dynamic performance: The EC-FCM clustering algorithm significantly improves the working condition recognition rate by using elevator-specific feature vectors, solving the problem of "PI mismatch caused by misjudgment of working conditions" in traditional algorithms. At the same time, it adaptively adjusts the speed loop PI controller parameters, effectively solving the problem of insufficient dynamic response performance of fixed parameter PI controllers under different working conditions such as heavy load and light load, start-up and braking, greatly shortening the dynamic response time and reducing overshoot.

[0026] (2) Improved vibration suppression: Vibration feedforward compensation based on a virtual impedance model is introduced into the current loop. The vibration frequency of the wire rope is accurately extracted through an improved adaptive bandpass filter, and a reverse compensation current is generated. This scheme can effectively suppress the car vibration caused by the elastic deformation of the traction system, which is significantly better than the traditional control method and meets the comfort standards of advanced elevators.

[0027] (3) Strong system adaptability: The “speed loop parameter adaptive + current loop multi-dimensional compensation” architecture constructed by this invention can automatically adapt to the large range of changes in elevator load from no load to rated load, as well as the switching of the entire operation stage from start-up, acceleration, cruise to deceleration, demonstrating good working condition robustness.

[0028] (4) Strong engineering practicality: The algorithm of this invention has a moderate amount of computation, does not require additional hardware costs, and can be directly implemented on existing elevator control systems (such as DSP processors) through software upgrades. It has good prospects for promotion and application and economic value. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a block diagram of the adaptive FOC dual-loop control system provided in an embodiment of the present invention; Figure 2 This is a flowchart of the adaptive FOC dual-loop control system provided in an embodiment of the present invention; Figure 3 This is a flowchart of the EC-FCM clustering algorithm provided in an embodiment of the present invention; Figure 4 This is a flowchart of vibration suppression current compensation calculation provided in an embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are merely illustrative and not intended to limit the invention.

[0032] See attached document Figure 1 In this embodiment of the invention, the control system core uses the TI 32-bit DSP chip TMS320F28335 as the main controller. The inverter uses an Infineon IMZ120R045M1H IPM module (rated voltage 1200V, rated current 45A) with built-in IGBTs and freewheeling diodes. The pre-drive circuit uses an IR2110 half-bridge driver chip to provide gate drive signals and support overcurrent and overvoltage protection. The communication module uses a CAN bus interface to achieve communication with the elevator control cabinet. After the main controller is powered on, the initial configuration of the system hardware peripherals is completed first. For example, the ADC module is configured in synchronous sampling mode with a sampling frequency of 20kHz to collect the analog signals output by the three-phase current sensor; the ePWM module is configured as a centrally symmetrical counting module with a carrier frequency of 10kHz to generate the SVPWM signal to drive the inverter; the encoder interface is configured with a QEP unit and connected to a 2048-line incremental photoelectric encoder to obtain the motor rotor position and mechanical speed.

[0033] The core of this invention is to construct a dual-loop architecture of "EC-FCM clustering-driven speed loop parameter adaptation + elevator mechanical characteristic coupled current loop vibration compensation", as shown in the appendix. Figure 2 This invention achieves precise control for elevators under specific operating conditions. The specific technical solution includes the following steps: S1: After the controller powers on, it completes the initialization configuration of hardware peripherals and algorithm parameters, providing a foundation for real-time control. The initialization of hardware peripherals includes configuring the core peripherals of the DSP chip, such as the ADC module, ePWM module, encoder interface, and communication interface. The initialization configuration of algorithm parameters mentioned in S1 involves loading pre-configured parameters from the Flash memory. These include EC-FCM clustering parameters, PI parameter table, and motor parameters. Further EC-FCM clustering parameters include initial cluster center matrices for six operating conditions: (1) In the formula: The rotational speed of the traction machine corresponding to the kth cluster center, in rad / s; The rotational acceleration corresponding to the kth cluster center, in rad / s. 2 ; This represents the total load capacity of the elevator car corresponding to the kth cluster center, in kg.

[0034] This invention is the first to couple and classify "load level (heavy load / light load)" with "operation phase (start / cruise / deceleration)," setting the number of clusters K=6. The multi-dimensional feature differences and scenario pain points (problems that traditional fixed PI parameters cannot adapt to) of the 6 types of operating conditions are as follows: Heavy load start (W≥50%) >0, <0.8 max Full load requires rapid start-up, but traditional fixed Kp causes response lag and prolongs start-up time; heavy load cruise (W≥50%), ≈0, max Full-load constant speed operation, traditional fixed Ki easily accumulates errors, aggravating speed fluctuations); heavy-load deceleration (W≥50%), <0, <0.2 max ; rapid deceleration under full load, traditional fixed parameters are prone to overshoot, resulting in large leveling errors); light load start-up (W < 50%), >0, <0.8 max Light load torque response is fast, but traditional fixed Kp easily leads to acceleration overshoot; light load cruise (W < 50%). ≈0, ≈ max Under light load and constant speed, traditional fixed parameters increase current harmonics, leading to reduced efficiency; under light load and deceleration (W < 50%), <0, >0.2 max (Light load deceleration, traditional fixed parameters prolong deceleration time). Furthermore, offline experiments were conducted to determine the cluster centers corresponding to these six operating conditions and the optimal PI parameters for each condition. The "offline experiment + online clustering adaptive" mechanism overcomes the limitations of traditional fixed PI parameters, which cannot adapt to the coupling of multiple operating conditions in elevators under "load level + operating phase". Furthermore, the motor parameters, namely the pre-stored motor nameplate and identification results, including the number of pole pairs... Permanent magnet flux d / q axis inductance Furthermore, the state variables are initialized, and the velocity loop integral term, current loop error cumulative value, and EC-FCM membership matrix U are cleared to zero.

[0035] S2: Collect elevator operating status parameters, filter them, and use them for subsequent control. The collection of elevator operating status parameters is achieved through a three-phase current sensor. Simultaneously, the mechanical speed of the motor is obtained through the encoder. and rotor position angle Furthermore, to address the "dynamic fluctuations" in elevator sensor data (such as sudden changes in speed during startup and changes in load with passenger load), a scenario-based filtering strategy is adopted. Additionally, the pressure sensor data W at the bottom of the car is read, and optionally, a sliding window filter is applied to W.

[0036] S3: Based on the current operating condition characteristics, the speed loop PI parameters are adjusted in real time through EC-FCM clustering to achieve dynamic adaptation. This is achieved by adjusting the speed loop PI parameters in real time through EC-FCM mean clustering, unlike the "current-voltage" characteristics of general FCMs. A 3D feature vector is constructed by selecting key dimensions of elevator control to ensure that each feature is strongly bound to control requirements. The 3D feature vector is as follows: (2) In the formula: The normalized rotational speed of the motor at time i, in rad / s; and Here, represents the normalized motor speed acceleration and elevator car load at time i, in rad / s and kg, respectively. If the feature vector formed by filtering the sensor data is directly used for membership calculation, features with large values ​​will dominate the distance calculation, causing the clustering results to over-rely on these features and ignore the influence of other features. Therefore, it is necessary to normalize the values ​​in the three-dimensional feature vector, mapping them to the 0-1 range to ensure fair weighting of each feature in the clustering.

[0037] Through full feature normalization, EC-FCM clustering can more accurately capture the combination of "load + speed + acceleration" operating conditions, providing a reliable basis for adaptive adjustment of PI parameters. The membership degree u of the current sample for the six operating conditions is calculated based on the six cluster centers determined in offline experiments. ik The formula is as follows: (3) In the formula: represents the membership degree of the current elevator operating state to the k-th operating condition; x(i) represents the elevator feature vector (fusion of speed, acceleration, and load dimensions) at the current sampling time; C kLet represent the cluster center of the k-th working condition; m is a fuzzy factor that controls the degree of fuzziness in the membership degree; the detailed calculation method for the Euclidean distance between the current feature and the center of the k-th class is as follows: (4) In the formula: x(i) represents the feature vector at the current time; C k Represents the cluster center of the k-th type of working condition; Regarding the cluster centers for the six operating conditions, initial cluster centers need to be manually set during motor startup due to insufficient sample size. These initial cluster centers are obtained through offline experiments. During elevator operation, the initial cluster centers need to be iterated based on real-time collected sample data to reach the optimal cluster centers. The iteration formula is as follows: (5) The termination condition for cluster center iteration is: (6) The PI controller parameters are dynamically adjusted based on the membership degree of parameters during elevator operation to achieve smooth parameter transitions and avoid torque shocks caused by abrupt changes. The optimal PI parameters determined by offline testing are used to calculate the current PI parameters through real-time membership degree weighting, as shown in the following formula: (7) (8) In the formula: u ik Let K be the membership degree of the data collected at time i to the cluster center of the k-th class; pk The optimal K determined offline for the k-th type of working condition p Parameter; K ik The optimal K determined offline for the k-th type of working condition i parameter.

[0038] S4: A vibration suppression current command is generated using a mechanical vibration model and an adaptive frequency bandpass filter. The current compensation utilizes a high-resolution encoder to measure the motor speed in real time. A bandpass filter extracts the pure vibration component from the motor speed signal without delay, filtering out low-frequency operating commands and high-frequency noise. Based on this, a virtual impedance model is constructed to continuously generate a q-axis current compensation amount that is proportional to the vibration displacement and velocity but in the opposite direction. This compensation is immediately executed through the fastest-responding current loop. The formula is as follows: (9) Where: K t_t_motor_constant_, in Nm / A, is an inherent parameter of the motor, converting torque commands into current commands executable by the current loop; _R_traction_sheave_radius_, in meters; _v_vibration_linear_velocity_, in meters / s, is obtained by processing the motor speed measured by the encoder through a bandpass filter (BPF) with a center frequency equal to the natural frequency of the wire rope, resulting in a pure vibration_angular_velocity, multiplied by the traction_sheave_radius_, directly reflecting the vibration trend and speed; _x_vibration_displacement_, in meters, is obtained by integrating the pure vibration_velocity signal. _x_reflects the magnitude of the vibration amplitude, i.e., the tension or compression of the wire rope relative to its equilibrium position (a bandpass filter or high-pass filter must be used to handle the DC drift of the integrator to ensure the accuracy of the displacement signal); K_ stiff K is the wire rope stiffness coefficient, expressed in N / m. This coefficient is one of the core optimization parameters, defining the system's ability to resist elastic deformation. Increasing K... stiff This means the virtual spring is "stiffer," the system is more willing to correct any positional deviation x, and can effectively reduce the steady-state amplitude. K damp K is the wire rope damping coefficient, expressed in N / (m / s). This parameter is another core tuning parameter; it defines the system's ability to dissipate vibration energy. Increasing K... damp This means the system has a stronger ability to suppress vibration velocity v and can calm oscillations more quickly. The negative sign ensures that the compensating force generated by the motor is always opposite to the vibration motion. When the wire rope is stretched (x > 0, v > 0), the compensating force is negative, reducing the motor torque; when the wire rope is stretched (x < 0, v < 0), the compensating force is positive, increasing the motor torque. In this way, the elastic vibration of the wire rope in the vertical direction can be actively suppressed.

[0039] The aforementioned bandpass filter (BPF) employs an adaptive intermediate frequency bandpass filter to accurately separate the vibration component of the wire rope from the motor rotation signal. This invention addresses the technical shortcomings of existing adaptive bandpass filters in suppressing the vibration of the current loop in elevator permanent magnet synchronous traction machines, such as insufficient frequency identification accuracy, fixed bandwidth unable to adapt to dynamic working conditions, weak anti-interference capability, and reliance on manual calibration. It proposes a multi-dimensional collaborative adaptive frequency bandpass filter design scheme. The design and implementation of this adaptive bandpass filter is as follows: First, the center frequency f of the bandpass filter... center It is dynamically configured by an independent frequency identification unit. This unit performs real-time spectrum analysis on the motor speed signal acquired by the encoder to calculate the dominant frequency f of the current vibration energy. d This value is used as the center frequency f of the bandpass filter. center The set value. Secondly, the bandpass filter receives the aforementioned center frequency configuration command f. center The filter internally adjusts the current center frequency f based on the configured settings. centerWith bandwidth BW, the coefficients of its digital filter are calculated in real time through a coefficient update algorithm, thereby dynamically adjusting its frequency response characteristics to ensure that its passband center is always locked at the current dominant vibration frequency.

[0040] To address the issue of "lagging or misjudgment in frequency identification" in traditional single algorithms (such as the Goertzel algorithm) under dynamic conditions such as elevator start / braking and sudden load changes, a two-stage identification architecture of "coarse positioning-fine calculation" is constructed. The first stage, coarse positioning, employs Short-Time Fourier Transform (STFT) to perform spectral analysis on the current loop vibration signal within a set time window, rapidly scanning the vibration energy distribution within the 1-5Hz range (the natural frequency range of the elevator wire rope). This selects three candidate frequencies with the highest energy (denoted as (f1, f2, f3)) and calculates their corresponding energy values ​​(E1, E2, E3), with a time consumption of ≤20ms, ensuring rapid dynamic response. The second stage, fine calculation, or fine positioning, uses an improved Goertzel algorithm to accurately identify the candidate frequencies, introducing energy weighting coefficients to optimize computational resource allocation. The weighting formula for the i-th candidate frequency is as follows: (10) In the formula: Ei represents the energy of the i-th candidate frequency. The higher the energy, the greater the weight. Furthermore, the number of iterations is allocated, with a total of N iterations. total =1000 (adapted to DSP computing power), then the number of iterations for the i-th frequency is: (11) Higher-weighted frequencies are allocated more iterations to improve accuracy. More iterations allow for a longer "observation time" of the signal, resulting in higher resolution and accuracy in frequency calculation. Allocating more iterations to high-energy candidate frequencies (more likely true vibration frequencies) selected by STFT significantly improves their identification accuracy. Lower-energy candidate frequencies (more likely interference) are allocated fewer iterations to reduce unnecessary calculations while ensuring basic judgment, thus keeping the total time within real-time requirements (elevator control has extremely strict real-time requirements; delays exceeding 20ms can cause vibration suppression failure).

[0041] The spectral width of elevator vibration is positively correlated with the center frequency. High-frequency vibrations (such as 3~5Hz, commonly seen in lightly loaded or high-tension steel wire rope conditions) have richer harmonic components (including 2nd and 3rd harmonics, etc.) and a wider spectral distribution. If the bandwidth is too narrow, useful harmonic signals will be filtered out, resulting in incomplete vibration feature extraction. Low-frequency vibrations (such as 1~2Hz, commonly seen in heavy-load or slack steel wire rope conditions) have a more concentrated spectrum (low harmonic energy). If the bandwidth is too wide, more low-frequency noise (such as car swaying interference) will be introduced. At the same time, the bandwidth also needs to be dynamically adjusted according to the load. Under heavy loads (e.g., 800~1000kg), the pressure of the car on the wire rope increases, leading to increased wire rope deflection. This introduces more low-frequency disturbances (0.5~1Hz) caused by "load imbalance" into the vibration, which are crucial components for vibration suppression. A narrow bandwidth will filter out these effective disturbance signals. Under light loads (e.g., 100~300kg), the wire rope tension is more uniform, resulting in fewer low-frequency disturbances, but more pronounced high-frequency vibrations (e.g., car resonance). In this case, the bandwidth can be appropriately narrowed to reduce noise. To address the issue of insufficient filtering accuracy with a fixed bandwidth under different vibration frequencies and car loads, a three-element linkage model of "center frequency-load-bandwidth" is established to dynamically adjust the filter bandwidth. The bandwidth calculation formula is as follows: (12) In the formula: BW is the bandwidth at time t; BW0 is the preset base bandwidth; k f k represents the frequency weighting coefficient. W W is the load weighting factor; W(t) is the real-time load of the car; W max The rated load of the car is determined; data is collected in real time, substituted into the formula for calculation, and the bandwidth is updated in a short time through the DSP hardware module to ensure that the bandwidth always matches the spectral characteristics of the vibration signal.

[0042] Through the above design, the bandpass filter can automatically track the change of the wire rope's natural frequency with the car position and load, thereby always extracting the pure vibration velocity component v from the motor speed signal with high fidelity.

[0043] S5: The generated vibration suppression command is superimposed on the output q-axis current command of the velocity loop and output together into the current loop PI controller. The formula is as follows: (13) In the formula: iq ref0 ∆iq is the output reference current of the speed loop PI controller. ref For vibration compensation current; Iq ref This is the input reference current for the q-axis current loop pi controller.

[0044] Example: Refer to Appendix Figure 3 After initialization, pre-configured algorithm parameters are loaded from the DSP's on-chip Flash memory, including FCM clustering parameters, i.e., the initial cluster center matrix and fuzzy factor (preferably m=2.0 in this embodiment) for six operating conditions (a combination of load level (heavy load / light load) and running phase (start / cruise / deceleration)); and a PI parameter table, which stores the optimal speed loop PI parameters corresponding to the six operating conditions. And related parameters of the motor, including the number of pole pairs P=4 and the flux linkage of the permanent magnet. d / q axis inductance Stator resistance R s =0.5Ω; and vibration compensation parameters: virtual stiffness coefficient Kstiff=500N / m, virtual damping coefficient Kdamp=50N / (m / s), and bandpass filter bandwidth BW=2Hz.

[0045] The initial cluster centers for the above six operating conditions were obtained using an offline experimental method. This invention defines the feature vector value ranges for the six operating conditions, and through offline experiments, forces the elevator to operate under each condition (e.g., when W≥50%). >0, <0.8 max Under the heavy-load start-up condition, a large number of samples (speed, acceleration, load) are collected for this condition, and the mean of these samples is calculated as the initial cluster center for this type of condition. This makes subsequent iterations more efficient. More accurate clustering of real-time conditions results in clusters that better match the actual needs of the "load-running phase," thus allowing for reasonable adaptive adjustment of the PI parameter.

[0046] The optimal parameters for the speed loop PI controller under six operating conditions were also determined through offline experiments. Initial PI parameters were set based on motor rated parameters and experience, and then K was fixed. i =0, send the corresponding speed command for the operating condition, and gradually increase K. p (Increase by 0.1 each time), record the speed response curve, and select the optimal K. p Value. Then, fix K. p The optimal value is achieved by gradually increasing K. i Record the steady-state speed error and select the optimal value based on comprehensive analysis.

[0047] The sensors collect data, which is then filtered and uploaded to the main controller for further calculations. This includes data such as the speed and angle values ​​output by the encoder, and the current load value output by the weight sensor inside the elevator car. The filtered motor speed is then used to calculate the elevator car's speed and acceleration. This data filtering process affects the motor's mechanical speed. A 5-point moving average filter (optional) is used, as shown in the following formula: (14) In the formula: Let k be the rotational speed at time k after filtering. Since this filtering formula requires at least 5 historical data points, when k≤3, a "previous value compensation" strategy is usually adopted in engineering. That is, if the data is insufficient, the missing historical values ​​are repeatedly filled in with existing data. If k=1, only the existing data is considered. , Then, use existing data to repeatedly fill in the missing historical values. For example, , , Use them Similarly, when k=2 or 3, it is also used. This is used to fill in the missing information. Although the initial accuracy is slightly lower, it ensures that filtering is introduced during the startup phase to avoid the impact of sudden data changes on control. Once k≥4, it automatically switches to a complete 5-point moving average to ensure the filtering effect after stable operation.

[0048] Optionally, the output W of the elevator weight sensor can be preprocessed using a sliding window filter. The filtering formula is as follows: (15) In the formula, For the input weight data; n is the window length; n is the index of the current sampling point. This is the filtered output value at time n.

[0049] Window length N The choice of window length has a significant impact on the smoothness of the data. In this embodiment, the preferred window length is 4.

[0050] The above data is normalized to prevent features with large numerical values ​​from dominating distance calculations, thus avoiding over-reliance on these features in the clustering results and neglecting the influence of other features. The normalization formulas for velocity, acceleration, and load weight are as follows: (16) In the formula: This is the minimum speed of the motor, since motors always start running from 0 speed. Take 0 rad / s; The point angular velocity corresponding to the rated speed of the motor (needs to be calculated based on motor parameters), in rad / s. 2 ; This is the real-time motor speed after filtering the data collected by the sensor, in rad / s.

[0051] (17) In the formula: The motor speed acceleration is obtained from real-time sampling by the sensor, in rad / s. 2 ; This is the maximum deceleration acceleration of the motor, measured in rad / s. 2 ; This is the maximum acceleration of the motor, expressed in rad / s. 2 ; (18) In the formula: W n The elevator car load weight, measured in kg, is obtained in real-time from sensor data and filtered. min For no-load conditions, generally take 0 kg; W max This is the elevator's rated load, expressed in kg.

[0052] Construct feature vectors using the normalized parameters. The data is input into the FCM clustering module for clustering. The membership values ​​of the 6 working conditions are output according to equation (3). The parameters of the speed loop PI controller are adjusted in a timely manner according to the membership values. At the same time, the cluster centers are updated in real time according to equation (5) until the conditions of equation (6) are met.

[0053] See attached document Figure 4 The system uses a high-resolution encoder to measure the motor speed in real time. It uses an adaptive intermediate frequency bandpass filter to extract the pure vibration component from the motor speed signal without delay, filtering out low-frequency operating commands and high-frequency noise. Based on this, a virtual impedance model is constructed, and a q-axis current compensation amount that is proportional to the vibration displacement and velocity and opposite in direction is continuously generated. This compensation is then executed immediately through the current loop with the fastest response.

[0054] The aforementioned adaptive intermediate frequency bandpass filter is implemented through the following specific steps: its core lies in the ability of the filter coefficients to be updated and adjusted in real time and online according to the identified dominant vibration frequency and bandwidth. The controller processes the motor speed signal ω acquired by the encoder and downsampled (preferably downsampling frequency f=100Hz). md Perform real-time spectrum analysis to determine the dominant vibration frequency f. d In this embodiment, a two-stage architecture of "STFT coarse positioning + improved Goertzel fine calculation" is used to achieve this function.

[0055] The first stage, the coarse positioning stage, involves creating a circular buffer of length N (preferably 400 in this embodiment) to store the ω values ​​of the N nearest points. md The signal, the buffer forms a signal of length T window=N / f time window. Based on the possible range of the natural frequency of the wire rope, the target frequency band to be scanned is set to [f_min, f_max] (preferably 1Hz to 5Hz), and the frequency band is discretized into M frequency points with a preset frequency resolution Δf (preferably 0.25Hz). For each frequency point, a short-time Fourier transform (STFT) is performed, traversing the N data points in the buffer. The three candidate frequencies with the highest energy are selected (denoted as (f1, f2, f3)), and their corresponding energy values ​​(E1, E2, E3) are calculated. The time taken is ≤20ms to ensure dynamic response speed. The second stage, the fine calculation stage, involves accurately identifying candidate frequencies based on the improved Goertzel algorithm and introducing energy weighting coefficients to optimize computational resource allocation. Ultimately, the frequency point corresponding to the maximum energy value is determined as the current dominant vibrational frequency f. d To avoid false noise judgments, a power threshold can be set. f is only updated when the maximum power exceeds this threshold. d Furthermore, the bandwidth is calculated in real time according to the filter bandwidth calculation formula. In this embodiment, BW0 is preferably 1Hz (to ensure low-frequency noise immunity), k f The preferred value is 0.2 (calibrated through 100 sets of experiments to characterize the contribution of the center frequency to the bandwidth), k W The optimal value is 0.5 (calibrated through 50 sets of heavy / light load comparison experiments to characterize the contribution of load percentage to bandwidth). After obtaining the current dominant vibration frequency and bandwidth, the dominant vibration frequency is used as the new center frequency f of the bandpass filter. center Secondly, the bandpass filter (preferably an infinite impulse response IIR type) receives the aforementioned center frequency configuration command f. center The filter internally adjusts the current center frequency f based on the configured settings. center The frequency response characteristics of the bandpass filter are dynamically adjusted based on the real-time calculated bandwidth BW, ensuring that its passband center is always locked at the current dominant vibration frequency. The passband center of the bandpass filter dynamically tracks the changes in the natural frequency of the wire rope, thus ensuring accurate extraction of vibration signals throughout the entire elevator operation. Vibration displacement is obtained by integrating the pure vibration velocity signal. Finally, the extracted vibration velocity and displacement are input to the compensation module to generate a compensation current based on the input reference current calculation formula.

[0056] The vibration compensation parameter, namely the virtual stiffness coefficient K stiff and virtual damping coefficient K damp The shaft parameters need to be finalized on-site to ensure optimal vibration damping. This embodiment uses the following progressive parameter tuning method: First, based on the basic mechanical parameters of the elevator system, initial estimates of the parameters can be calculated. The damping coefficient is then set, and the virtual stiffness coefficient is temporarily set to 0 to mask its influence. The elevator is then run between target floors at rated speed, especially during start-up and braking, to observe the car vibration caused by the wire rope. Starting from 0, K is gradually increased... damp The value of K is determined by observing the vibration decay rate. The goal is to find a critical value that allows vibration to be quickly suppressed without introducing high-frequency noise. At this point, the system should exhibit good overdamping characteristics, meaning the vibration is quickly suppressed without significant overshoot. This value is then determined as the optimal K under the current operating conditions. damp value.

[0057] Setting stiffness coefficient K stiff (Performance optimization) Maintain the optimal K determined in the first step. damp The value remains unchanged. Starting from 0, K gradually increases. stiff The value of K. Observe the steady-state amplitude of the vibration. Increase K. stiff This enhances the system's resistance to elastic deformation, thereby effectively reducing the amplitude of vibration. The goal is to find a point that minimizes the steady-state amplitude without causing resonance or new oscillations in the system. This value is determined as the optimal K under the current operating conditions. stiff value.

[0058] By following the steps above, a set of parameters suitable for this specific elevator system can be obtained, enabling the virtual impedance model to achieve the best vibration suppression effect. This value is then determined as the optimal K under the current operating conditions. stiff value.

[0059] Corresponding to the aforementioned embodiment of the adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators, this invention also provides an embodiment of an adaptive FOC dual-loop control system for a permanent magnet synchronous traction machine for elevators. This system includes a configuration initialization module, a parameter filtering processing module, a parameter dynamic adaptation module, a vibration suppression current generation module, and a reference current output module; for the implementation of each module, please refer to the specific implementation steps of the aforementioned embodiment of the adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators.

[0060] The configuration initialization module is used to complete the initialization configuration of hardware peripherals and clustering algorithm parameters after the elevator controller is powered on; The parameter filtering module is used to collect elevator operating status parameters and then filter them. The parameter dynamic adaptation module is used to obtain the membership degree of each working condition by iteratively clustering the cluster centers through a clustering algorithm based on the characteristics of the current working condition, and dynamically adjust the speed loop PI parameter in real time to achieve dynamic adaptation. The vibration suppression current generation module is used to extract vibration components from the motor speed signal through a mechanical vibration model and an adaptive frequency bandpass filter, and generate a vibration suppression current command.

[0061] The reference current output module is used to superimpose the generated vibration suppression command with the output q-axis current command of the velocity loop, and output them together as a reference current into the current loop PI controller.

[0062] The above embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. An adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators, characterized in that, The method includes the following steps: S1: After the elevator controller is powered on, it completes the initialization configuration of hardware peripherals and clustering algorithm parameters; S2: Collect elevator operating status parameters and filter them; S3: Based on the characteristics of the current working conditions, the clustering center is iterated through a clustering algorithm to obtain the membership degree of each working condition, and the speed loop PI parameter is dynamically adjusted in real time to achieve dynamic adaptation; S4: By using a mechanical vibration model and an adaptive frequency bandpass filter, the vibration component is extracted from the motor speed signal to generate a vibration suppression current command; S5: The generated vibration suppression command is superimposed with the output q-axis current command of the speed loop and output together as a reference current into the current loop PI controller.

2. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 1, characterized in that, The hardware peripheral initialization described in S1 includes configuring the peripheral configuration of the DSP chip. The initialization configuration of the algorithm parameters is to load pre-configured parameters from the Flash memory, including EC-FCM clustering parameters, PI parameter table, and motor parameters. Among them, the EC-FCM clustering parameters include the initial clustering center matrix of 6 working conditions obtained by coupling the heavy load / light load level with the start / cruise / deceleration operation phase. EC-FCM clustering parameters include an initial cluster center matrix for 6 operating conditions: In the formula: The rotational speed of the traction machine corresponding to the kth cluster center, in rad / s; The rotational acceleration corresponding to the kth cluster center, in rad / s. 2 ; The total load of the car corresponding to the kth cluster center is expressed in kg. The cluster centers corresponding to these 6 working conditions and the optimal PI parameters for the corresponding working conditions are determined through offline experiments.

3. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 1, characterized in that, The acquisition of elevator operating status parameters, as described in S2, involves collecting current values ​​via a three-phase current sensor and simultaneously obtaining the motor's mechanical speed via an encoder. and rotor position angle ; Employing a scenario-based filtering strategy, for A sliding average filter is used, and the data W from the pressure sensor at the bottom of the car is read and filtered using a sliding window.

4. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 1, characterized in that, In S3, the speed loop PI parameters are adjusted in real time based on the current operating conditions using EC-FCM mean clustering to achieve dynamic adaptation and ensure that each feature is bound to the control requirements. The feature vector is as follows: In the formula: The normalized rotational speed of the motor at time i, in rad / s; and Here, represents the normalized motor speed and acceleration at time i, and the elevator car load, respectively, in rad / s and kg.

5. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 4, characterized in that, The membership degree u of the current sample for the six working conditions is calculated using the six identified cluster centers. ik The formula is as follows: In the formula: The function represents the membership degree of the current elevator operating state to the k-th type of operating condition; x(i) represents the elevator feature vector at the current sampling time; C k This represents the cluster center of the k-th working condition; m is a fuzzy factor that controls the degree of fuzziness in the membership degree. The parameters of the PI controller are dynamically adjusted based on the membership degree of the parameters during elevator operation to achieve a smooth transition of parameters. The current PI parameters are calculated by weighting the membership degree in real time.

6. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 1, characterized in that, In S4, the motor speed is measured in real time. A bandpass filter is used to extract the pure vibration component from the motor speed signal to filter out low-frequency operating commands and high-frequency noise. The natural frequency of the wire rope is automatically tracked as the car position and load change. Based on this, a virtual impedance model is constructed, and a q-axis current compensation amount that is proportional to the vibration displacement and velocity but opposite in direction is continuously generated. This compensation is then executed immediately through the current loop with the fastest response.

7. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 6, characterized in that, The center frequency f of the bandpass filter center Dynamically configured by an independent frequency identification unit, which performs real-time spectrum analysis on the motor speed signal acquired by the encoder to calculate the dominant frequency f of the current vibration energy. d This value is used as the center frequency f of the bandpass filter. center The set value; secondly, the bandpass filter receives the center frequency configuration command f. center The filter internally adjusts according to the currently configured center frequency f. center With bandwidth BW, the coefficients of its digital filter are calculated in real time through a coefficient update algorithm, thereby dynamically adjusting its frequency response characteristics to ensure that its passband center is always locked at the current dominant vibration frequency.

8. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 7, characterized in that, The center frequency is calculated based on a two-stage identification architecture of "coarse positioning-fine calculation". The first stage, coarse positioning, uses short-time Fourier transform (STFT) to perform spectral analysis on the current loop vibration signal within a set time window. This quickly scans the vibration energy distribution of the elevator wire rope to obtain candidate frequencies and calculates their corresponding energy values ​​to ensure dynamic response speed. The second stage, fine calculation, is the fine positioning stage. Based on the improved Goertzel algorithm, candidate frequencies are accurately identified. An energy weighting coefficient is introduced to optimize the allocation of computational resources, with higher energy having a greater weight.

9. The adaptive FOC dual-loop control method for a permanent magnet synchronous traction machine for elevators according to claim 7, characterized in that, A three-element linkage model of "center frequency-load-bandwidth" is established to dynamically adjust the filter bandwidth. The bandwidth calculation formula is as follows: In the formula: BW is the bandwidth at time t; BW0 is the preset base bandwidth; k f k represents the frequency weighting coefficient. W W is the load weighting factor; W(t) is the real-time load of the car; W max This is the rated load of the car.

10. An adaptive FOC dual-loop control system for a permanent magnet synchronous traction machine for elevators implementing the method of any one of claims 1-9, characterized in that, The system includes: The configuration initialization module is used to complete the initialization configuration of hardware peripherals and clustering algorithm parameters after the elevator controller is powered on. The parameter filtering module is used to collect elevator operating status parameters and then filter them. The parameter dynamic adaptation module is used to obtain the membership degree of each working condition by iteratively clustering the cluster centers through a clustering algorithm based on the characteristics of the current working condition, and dynamically adjust the speed loop PI parameter in real time to achieve dynamic adaptation. The vibration suppression current generation module is used to extract vibration components from the motor speed signal through a mechanical vibration model and an adaptive frequency bandpass filter, and generate a vibration suppression current command. The reference current output module is used to superimpose the generated vibration suppression command with the output q-axis current command of the velocity loop, and output them together as a reference current into the current loop PI controller.

Images