Improved sliding mode control method and system for permanent magnet synchronous motor based on extreme learning machine

Abstract

The application relates to the technical field of permanent magnet synchronous motor control, and discloses an improved sliding mode control method and system of a permanent magnet synchronous motor based on a limit learning machine. According to a reference rotating speed and an actual rotating speed, the method determines an operation state feature, and the limit learning machine is pre-trained to map target control parameters of a sliding mode surface parameter, a reaching law parameter and an anti-interference adjustment parameter; according to the operation state feature, a performance evaluation index is determined, a parameter update rate is determined in combination with a state range, and the sliding mode control parameters of a current period are updated; a sliding mode surface is constructed, an adaptive anti-interference item gain is updated, an improved reaching law is constructed to determine a quadrature axis reference current, and the permanent magnet synchronous motor is subjected to closed-loop control. The application solves the problem that the fixed sliding mode control parameters are difficult to adapt to operation state changes, the target values of the control parameters are adaptively adjusted according to the operation state, the adaptive anti-interference item gain is adaptively adjusted according to the sliding mode surface to participate in the determination of the quadrature axis reference current, and the adaptability of the control parameters to operation state changes is improved.

Description

Technical Field

[0001] This invention relates to the field of permanent magnet synchronous motor control technology, and in particular to an improved sliding mode control method and system for permanent magnet synchronous motors based on extreme learning machine. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) are characterized by high power density, high operating efficiency, compact structure, and fast dynamic response, and are widely used in industrial robots, CNC machine tools, electric vehicles, and high-performance servo drives. In actual operation, the reference speed, load torque, and operating state of a PMSM may continuously change and may be affected by parameter perturbations and external load disturbances. Therefore, high requirements are placed on the dynamic response, steady-state accuracy, and disturbance rejection performance of speed control.

[0003] Current speed control of permanent magnet synchronous motors typically employs proportional-integral control (PIC) and sliding mode control. Among these, sliding mode control exhibits strong robustness and is suitable for motor control scenarios involving parameter perturbations and external disturbances. The control performance of sliding mode control is closely related to the settings of relevant control parameters. Fixed-parameter methods are usually tuned according to preset operating conditions. Alternatively, fuzzy control, neural networks, and other methods can be used to adjust relevant control parameters based on the motor's operating state to improve control performance under different operating conditions.

[0004] However, when the reference speed, load condition, or operating condition of a permanent magnet synchronous motor changes, the requirements for sliding mode control parameters differ at different operating stages. Therefore, improving the adaptability of the sliding mode control parameters of a permanent magnet synchronous motor to different operating conditions is a technical problem that needs to be solved in the existing technology. Summary of the Invention

[0005] This invention provides an improved sliding mode control method and system for permanent magnet synchronous motors based on extreme learning machines, which can solve the problem of insufficient adaptability of sliding mode control parameters when the operating state of a permanent magnet synchronous motor changes.

[0006] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: In a first aspect, an improved sliding mode control method for a permanent magnet synchronous motor based on an extreme learning machine is provided. The method includes: acquiring a reference speed and an actual speed of the permanent magnet synchronous motor; determining operating state characteristics based on the reference speed and the actual speed; inputting the operating state characteristics into a pre-trained extreme learning machine for parameter mapping to obtain target control parameters for the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters; determining system performance evaluation indicators based on the operating state characteristics, and determining a parameter update rate based on the system performance evaluation indicators and the state range of the operating state characteristics; and based on the parameter update rate, performing parameter updates according to the sliding surface parameters and their target control parameters, the reaching law parameters and their target control parameters, and the disturbance rejection adjustment parameters and their target control parameters from the previous control cycle. The parameters are updated to obtain the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters for the current control cycle. These parameters constitute the sliding control parameters for the current control cycle. A sliding surface is constructed based on the operating state characteristics and the sliding surface parameters for the current control cycle. The adaptive disturbance rejection term gain is updated based on the sliding surface and the disturbance rejection adjustment parameters for the current control cycle. An improved reaching law is constructed based on the sliding surface, the reaching law parameters for the current control cycle, and the updated adaptive disturbance rejection term gain. The quadrature-axis reference current is determined based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved reaching law. Closed-loop control of the permanent magnet synchronous motor is performed based on the quadrature-axis reference current to obtain the updated actual speed.

[0007] The beneficial effects of this invention are as follows: By inputting the operating state characteristics of the permanent magnet synchronous motor into a pre-trained extreme learning machine, the target control parameters for the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters are obtained respectively. The parameter update rate is determined based on the system performance evaluation indicators and the state range of the operating state characteristics. This ensures that the sliding surface parameters of the previous control cycle are updated to the target control parameters of the sliding surface parameters according to the parameter update rate, the reaching law parameters of the previous control cycle are updated to the target control parameters of the reaching law parameters according to the parameter update rate, and the disturbance rejection adjustment parameters of the previous control cycle are updated according to the parameter update rate. The parameter update rate updates the target control parameter of the disturbance rejection adjustment parameter instead of directly replacing the parameter of the previous control cycle with the output of the extreme learning machine, thereby reducing the direct jump of sliding mode control parameters between adjacent control cycles. At the same time, a sliding surface is constructed according to the sliding surface parameters of the current control cycle. The adaptive disturbance rejection term gain is updated using the sliding surface and the disturbance rejection adjustment parameter of the current control cycle. An improved approach law is constructed in combination with the approach law parameter of the current control cycle to determine the quadrature axis reference current. This allows the sliding mode control parameter and the adaptive disturbance rejection term gain to be adjusted according to the operating state of the permanent magnet synchronous motor and used for the closed-loop control of the permanent magnet synchronous motor.

[0008] Based on the above technical solution, the present invention can be further improved as follows.

[0009] In one possible implementation, the step of inputting the operating state features into a pre-trained extreme learning machine for parameter mapping to obtain target control parameters for the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters includes: keeping the network parameters of the pre-trained extreme learning machine unchanged, inputting the operating state features into the pre-trained extreme learning machine for forward computation to obtain parameter output; determining the target parameter values ​​for the sliding surface integral gain, the exponential reaching term gain, the constant velocity reaching term gain, the boundary layer thickness, and the adaptive term coefficient based on the parameter output; determining the target parameter value of the sliding surface integral gain as the target control parameter for the sliding surface parameters, determining the target parameter values ​​of the exponential reaching term gain, the constant velocity reaching term gain, and the boundary layer thickness as the target control parameters for the reaching law parameters, and determining the target parameter value of the adaptive term coefficient as the target control parameter for the disturbance rejection adjustment parameters.

[0010] The method provided by this invention maintains the network parameters of the pre-trained Extreme Learning Machine (ELM) unchanged during online control and performs forward computation based on the operating state characteristics, thus eliminating the need for iterative updates to the ELM network parameters during online control. Simultaneously, the parameter outputs are defined as target parameter values ​​for the sliding mode surface integral gain, the exponential reaching term gain, the constant velocity reaching term gain, the boundary layer thickness, and the adaptive term coefficients, respectively. Furthermore, target control parameters are formed for the sliding mode surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters, thereby providing clear parameter adjustment targets for the subsequent updates of these three types of parameters according to the parameter update rate.

[0011] In one possible implementation, the operating state characteristics include: real-time speed error, real-time speed error change rate, and reference speed change rate; determining the system performance evaluation index based on the operating state characteristics includes: extracting the real-time speed error, real-time speed error change rate, and reference speed change rate from the operating state characteristics; and constructing the system performance evaluation index based on the real-time speed error, the real-time speed error change rate, and the reference speed change rate.

[0012] The method provided by this invention uses actual rotational speed, reference rotational speed, rotational speed error, rotational speed error rate of change, and reference rotational speed rate of change as operating state characteristics, and uses the rotational speed error, rotational speed error rate of change, and reference rotational speed rate of change to construct system performance evaluation indicators. The system performance evaluation indicators simultaneously reflect the degree of deviation of the actual rotational speed from the reference rotational speed, the change state of the rotational speed error, and the change state of the reference rotational speed itself, thereby providing a state evaluation basis for determining the state range of the operating state characteristics and the subsequent parameter update rate.

[0013] In one possible implementation, determining the parameter update rate based on the system performance evaluation index and the state range of the operating state characteristics includes: The real-time speed error is compared with preset dynamic state conditions and preset stable state conditions for speed error, and the real-time speed error change rate is compared with preset dynamic state conditions and preset stable state conditions for speed error change rate, respectively, to obtain a state range determination result. Based on the state range determination result, the update rate gain coefficient of the previous control cycle is increased, decreased, or maintained to obtain the update rate gain coefficient of the current control cycle. Based on the update rate gain coefficient of the current control cycle and the system performance evaluation index, the parameter update rate before limiting is determined, and the parameter update rate before limiting is range-constrained to obtain the parameter update rate.

[0014] The method provided by this invention obtains a state range determination result by comparing the speed error with preset dynamic state conditions and stable state conditions of the speed error, as well as the speed error change rate with preset dynamic state conditions and stable state conditions of the speed error change rate. Based on this, the update rate gain coefficient of the previous control cycle is increased, decreased, or maintained, so that the update rate gain coefficient of the current control cycle can change with the state range of the operating state characteristics. Then, the parameter update rate before the limit is determined and range constraint is applied in combination with the system performance evaluation index, so that the parameter update rate can adapt to the dynamic state of the permanent magnet synchronous motor, while avoiding the parameter update rate from exceeding the preset adjustment range.

[0015] In one possible implementation, the step of updating parameters based on the parameter update rate, according to the sliding surface parameters and their target control parameters of the previous control cycle, the reaching law parameters and their target control parameters of the previous control cycle, and the disturbance rejection adjustment parameters and their target control parameters of the previous control cycle, to obtain the sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters of the current control cycle, includes: determining the sliding surface parameter difference between the sliding surface parameters and their target control parameters of the previous control cycle, determining the reaching law parameter difference between the reaching law parameters and their target control parameters of the previous control cycle, and determining the disturbance rejection adjustment parameter difference between the disturbance rejection adjustment parameters and their target control parameters of the previous control cycle; according to the... The parameter update rate is adjusted to the difference in sliding surface parameters, the difference in reaching law parameters, and the difference in disturbance rejection adjustment parameters, respectively, to obtain the changes in sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters. The sliding surface parameters of the previous control cycle are updated based on the changes in sliding surface parameters, the reaching law parameters of the previous control cycle are updated based on the changes in reaching law parameters, and the disturbance rejection adjustment parameters of the previous control cycle are updated based on the changes in disturbance rejection adjustment parameters. The updated sliding surface parameters, updated reaching law parameters, and updated disturbance rejection adjustment parameters are subject to parameter range constraints to obtain the sliding mode control parameters for the current control cycle, which include the sliding surface integral gain, the exponential reaching term gain, the constant velocity reaching term gain, the boundary layer thickness, and the adaptive term coefficient.

[0016] The method provided by this invention determines the difference between the target control parameter of the sliding surface parameter and the sliding surface parameter of the previous control cycle, the difference between the target control parameter of the reaching law parameter and the reaching law parameter of the previous control cycle, and the difference between the target control parameter of the disturbance rejection parameter and the disturbance rejection parameter of the previous control cycle. It then uses the parameter update rate to adjust these three types of parameter differences to form three types of parameter changes. This ensures that the sliding surface parameter, reaching law parameter, and disturbance rejection parameter are updated according to the change range controlled by the parameter update rate, rather than directly replacing the parameters of the previous control cycle with the target control parameter. Finally, parameter range constraints are applied to the updated three types of parameters to keep the sliding surface control parameter of the current control cycle within a preset adjustment range.

[0017] In one possible implementation, the operating state characteristics include: real-time speed error, real-time speed error rate of change, and real-time speed error integral value; the step of constructing a sliding surface based on the operating state characteristics and the sliding surface parameters of the current control cycle includes: extracting the real-time speed error, real-time speed error rate of change, and real-time speed error integral value from the operating state characteristics; and constructing the sliding surface based on the real-time speed error, the real-time speed error rate of change, the real-time speed error integral value, and the sliding surface integral gain of the current control cycle.

[0018] The method provided by this invention introduces the speed error, the rate of change of speed error, and the integral value of speed error into the sliding surface, and uses the sliding surface integral gain of the current control cycle to participate in the construction of the sliding surface. This allows the sliding surface to simultaneously reflect the current speed deviation, the speed deviation change state, and the speed deviation accumulation state. Furthermore, the integral effect of the sliding surface is adjusted by the sliding surface integral gain of the current control cycle, providing a state quantity basis for subsequently updating the adaptive disturbance rejection term gain based on the sliding surface state and constructing an improved reaching law.

[0019] In one possible implementation, updating the adaptive anti-disturbance term gain based on the sliding surface and the anti-disturbance adjustment parameters of the current control cycle includes: determining a sliding surface region determination result based on the positional relationship between the sliding surface and a preset sliding surface region; wherein the sliding surface region determination result includes an outside-region determination result and an inside-region determination result, the outside-region determination result indicating that the sliding surface is located outside the preset sliding surface region, and the inside-region determination result indicating that the sliding surface is located within the preset sliding surface region; when the sliding surface region determination result is the outside-region determination result, determining the anti-disturbance gain change based on the adaptive term coefficient of the current control cycle and the sliding surface, and updating the adaptive anti-disturbance term gain of the previous control cycle based on the anti-disturbance gain change to obtain the updated adaptive anti-disturbance term gain; when the sliding surface region determination result is the inside-region determination result, attenuating the adaptive anti-disturbance term gain of the previous control cycle to obtain the updated adaptive anti-disturbance term gain.

[0020] The method provided by this invention selects the update mode of the adaptive disturbance rejection term gain based on the positional relationship between the sliding surface and the preset sliding surface region. When the sliding surface is outside the preset sliding surface region, the change in the disturbance rejection gain is determined by the adaptive term coefficient of the current control cycle and the sliding surface, and the adaptive disturbance rejection term gain of the previous control cycle is updated to enhance the disturbance rejection effect when the sliding surface deviates from the preset region. When the sliding surface is within the preset sliding surface region, the adaptive disturbance rejection term gain of the previous control cycle is attenuated to reduce the control quantity fluctuation caused by continuously using a large disturbance rejection gain during the stable phase. Thus, the adaptive disturbance rejection term gain is switched and adjusted according to the sliding surface region determination result.

[0021] In one possible implementation, the step of constructing an improved reaching law based on the sliding surface, the reaching law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain, and determining the quadrature-axis reference current based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved reaching law, includes: determining an exponential reaching term based on the sliding surface and the exponential reaching term gain of the current control cycle; determining a continuous saturation function output based on the sliding surface and the boundary layer thickness of the current control cycle, and determining a constant velocity reaching term based on the constant velocity reaching term gain of the current control cycle; determining an adaptive disturbance rejection term based on the sliding surface and the updated adaptive disturbance rejection term gain; combining the exponential reaching term, the constant velocity reaching term, and the adaptive disturbance rejection term to obtain the improved reaching law; and solving for the quadrature-axis reference current based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved reaching law.

[0022] The method provided by this invention determines the constant velocity approach term by using a continuous saturation function output and combining it with the constant velocity approach term gain of the current control cycle. This allows the constant velocity approach term to change continuously with the sliding surface when the sliding surface is within the boundary layer range, reducing the control quantity changes caused by discontinuous switching near the sliding surface. Simultaneously, the updated adaptive disturbance rejection term gain is introduced into the adaptive disturbance rejection term and combined with the exponential approach term and the constant velocity approach term to form an improved approach law. This allows the approach control action and the disturbance rejection control action that changes with the sliding surface state to jointly participate in the determination of the quadrature axis reference current. Thus, the approach law parameters of the current control cycle and the updated adaptive disturbance rejection term gain are applied to the closed-loop control of the permanent magnet synchronous motor.

[0023] In one possible implementation, the step of constraining the parameter ranges of the updated sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters to obtain sliding mode control parameters for the current control cycle, including sliding surface integral gain, exponential reaching term gain, constant velocity reaching term gain, boundary layer thickness, and adaptive term coefficients, includes: obtaining the upper bound of the sliding surface disturbance and a preset upper limit of the fluctuation of the constant velocity reaching term; determining the lower limit of the boundary layer thickness based on the constant velocity reaching term gain of the current control cycle, the upper bound of the sliding surface disturbance, and the preset upper limit of the fluctuation; constraining the updated boundary layer thickness based on the lower limit of the boundary layer thickness, and constraining the parameter ranges of the updated sliding surface integral gain, the updated exponential reaching term gain, the updated constant velocity reaching term gain, and the updated adaptive term coefficients to obtain the sliding mode control parameters for the current control cycle.

[0024] The method provided by this invention determines the lower limit of the boundary layer thickness by utilizing the gain of the constant velocity approach term, the upper limit of the sliding surface disturbance, and the preset upper limit of the fluctuation of the constant velocity approach term in the current control cycle. The lower limit of the boundary layer thickness is then used to constrain the updated boundary layer thickness, establishing a constraint relationship between the lower limit of the boundary layer thickness and the fluctuation of the constant velocity approach term that may be caused by the sliding surface disturbance. This limits the increase in the sensitivity of the constant velocity approach term response caused by an excessively small boundary layer thickness. Simultaneously, parameter range constraints are applied to the updated sliding surface integral gain, exponential approach term gain, constant velocity approach term gain, and adaptive term coefficient, respectively, so that the five sliding mode control parameters in the current control cycle are within the preset adjustment range.

[0025] Secondly, this invention provides an improved sliding mode control system for permanent magnet synchronous motors based on extreme learning machines, the system comprising: The operating status determination module is used to obtain the reference speed and actual speed of the permanent magnet synchronous motor, and determine the operating status characteristics based on the reference speed and the actual speed. The target parameter generation module is used to input the running state features into a pre-trained extreme learning machine for parameter mapping to obtain the target control parameters of the sliding surface parameters, the target control parameters of the reaching law parameters, and the target control parameters of the disturbance rejection adjustment parameters. The update rate determination module is used to determine the system performance evaluation index based on the operating state characteristics, and to determine the parameter update rate based on the system performance evaluation index and the state range of the operating state characteristics. The parameter update module is used to update parameters based on the parameter update rate, according to the sliding surface parameters and their target control parameters of the previous control cycle, the reaching law parameters and their target control parameters of the previous control cycle, and the disturbance rejection adjustment parameters and their target control parameters of the previous control cycle, respectively, to obtain the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters of the current control cycle; wherein, the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters of the current control cycle constitute the sliding control parameters of the current control cycle; The sliding mode state processing module is used to construct a sliding surface based on the operating state characteristics and the sliding surface parameters of the current control cycle, and to update the adaptive disturbance rejection term gain based on the sliding surface and the disturbance rejection adjustment parameters of the current control cycle. The control quantity determination module is used to construct an improved approach law based on the sliding surface, the approach law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain, and to determine the quadrature axis reference current based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved approach law. The closed-loop control module is used to perform closed-loop control on the permanent magnet synchronous motor based on the quadrature axis reference current to obtain the updated actual speed.

[0026] The beneficial effects of the second aspect of the present invention can be referred to in the beneficial effects of the first aspect and any possible implementation thereof, and will not be repeated here. Attached Figure Description

[0027] Figure 1 This is a structural diagram of the dual closed-loop control system for permanent magnet synchronous motors of the present invention.

[0028] Figure 2 This is a network structure diagram of the Extreme Learning Machine of this invention.

[0029] Figure 3 This is a flowchart of the improved sliding mode control method for permanent magnet synchronous motors according to the present invention.

[0030] Figure 4 This is a schematic diagram of the permanent magnet synchronous motor test platform structure of the present invention.

[0031] Figure 5 This is the electrical structure diagram of the dual-motor counter-support control test platform for permanent magnet synchronous motors of the present invention.

[0032] Figure 6 This is a graph showing the output evaluation and closed-loop speed tracking evaluation of the extreme learning machine of this invention; wherein... Figure 6 (a) is a regression analysis graph showing the predicted output versus the target value. Figure 6 (b) is a diagram showing the relative error distribution of closed-loop speed dynamic tracking.

[0033] Figure 7 This is the sliding mode surface integral gain of the current sampling period in this invention. A graph showing how the graph changes over time.

[0034] Figure 8 This is a comparison diagram of the quadrature-axis current response of the control method described in this invention with other control methods.

[0035] Figure 9 This is a curve showing the change of the sliding surface during the control process of this invention.

[0036] Figure 10 This is a comparison diagram of the speed response of the control method described in this invention with other control methods.

[0037] Figure 11 This is a comparison chart of the speed error between the control method described in this invention and other control methods.

[0038] Figure 12 This is a speed-current operating trajectory diagram of the permanent magnet synchronous motor of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. In the description of the present invention, unless otherwise stated, " / " indicates that the objects before and after are in an "or" relationship. For example, A / B can represent A or B. The "or" in the present invention is merely a description of the relationship between the related objects, indicating that three relationships can exist. For example, A or B can represent: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Furthermore, in the description of the present invention, unless otherwise stated, "multiple" refers to two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items.

[0040] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0041] In this embodiment of the invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this embodiment of the invention should not be construed as superior or more advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner for ease of understanding.

[0042] Permanent magnet synchronous motors (PMSMs) are characterized by high power density, high operating efficiency, compact structure, and fast dynamic response, making them suitable for applications in industrial robots, CNC machine tools, electric vehicles, and high-performance servo drive systems. During actual operation, the reference speed, load torque, and operating state of a PMSM may change, and it may also be affected by factors such as parameter perturbations and external load disturbances. Therefore, the speed outer loop needs to continuously provide a quadrature-axis reference current that matches the current state under different operating conditions.

[0043] To meet the speed control requirements of permanent magnet synchronous motors, proportional-integral control or sliding mode control can be used. Sliding mode control can form the outer loop control quantity of speed using the sliding surface and the reaching law. Its control performance is related to the settings of sliding surface integral gain, reaching law gain, boundary layer thickness, and disturbance rejection parameters. Fixed parameters are usually tuned according to preset operating conditions, but fuzzy control or neural networks can also be used to adjust the control parameters.

[0044] However, when the reference speed changes continuously or the load condition changes, the speed error, error change state, and disturbance state differ at different operating stages. If the actual control parameters cannot be adjusted accordingly with the operating state, the same set of parameters will be difficult to continuously meet the control requirements of different operating stages.

[0045] Therefore, embodiments of the present invention provide an improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machines. The method obtains the reference speed and actual speed of the permanent magnet synchronous motor (PMSM), and determines the operating state characteristics based on the reference speed and actual speed. The operating state characteristics are input into a pre-trained extreme learning machine for parameter mapping to obtain the target control parameters for the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters. System performance evaluation indicators are determined based on the operating state characteristics, and the parameter update rate is determined based on the system performance evaluation indicators and the state range of the operating state characteristics. The sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters of the previous control cycle are updated based on the parameter update rate to obtain the sliding control parameters for the current control cycle. A sliding surface is constructed based on the operating state characteristics and the sliding surface parameters of the current control cycle, and the adaptive disturbance rejection term gain is updated based on the sliding surface and the disturbance rejection adjustment parameters of the current control cycle. An improved reaching law is constructed based on the sliding surface, the reaching law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain. The quadrature-axis reference current is determined based on the mechanical motion relationship of the PMSM, the sliding surface, and the improved reaching law, and closed-loop control of the PMSM is performed based on the quadrature-axis reference current to obtain the updated actual speed.

[0046] The method provided by this invention generates target control parameters corresponding to sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters based on the characteristics of the operating state using a pre-trained extreme learning machine. It then uses the parameter update rate associated with the current operating state to gradually adjust the three types of parameters from the previous control cycle to the corresponding target control parameters, avoiding the extreme learning machine output directly replacing the parameters from the previous control cycle. The sliding surface parameters, disturbance rejection adjustment parameters, and reaching law parameters of the current control cycle are then applied to the sliding surface construction, adaptive disturbance rejection term gain update, and improved reaching law construction, respectively, so that the generation of target control parameters, the update of actual parameters, the determination of control quantities, and the motor closed-loop control form a continuous data processing chain.

[0047] In some embodiments, the improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine provided in this invention can be executed by an improved sliding mode control system for permanent magnet synchronous motors based on extreme learning machine (hereinafter referred to as the control system).

[0048] As an example, the control system may include a controller, an encoder, a current sensor, a three-phase inverter, and a permanent magnet synchronous motor. The controller includes a processing unit for performing state calculations, extreme learning machine forward operations, parameter updates, sliding mode control, current loop regulation, and space vector pulse width modulation, and a storage unit for storing the control program, extreme learning machine model parameters, input scale parameters, and operating data from the previous control cycle. The encoder provides the actual rotational speed and rotor position angle, the current sensor provides the three-phase stator current, and the three-phase inverter outputs three-phase AC voltage to the permanent magnet synchronous motor based on the switching drive signal output by the controller.

[0049] Figure 1 The structure of a dual-closed-loop control system for a permanent magnet synchronous motor, as illustrated in an embodiment of the present invention, is shown. This control system includes an outer speed loop, a direct-axis current inner loop, a quadrature-axis current inner loop, a coordinate transformation section, a space vector pulse width modulation section, a three-phase inverter, and a permanent magnet synchronous motor. The outer speed loop employs the improved sliding mode control method described in this invention, while the direct-axis and quadrature-axis current inner loops utilize proportional-integral controllers.

[0050] Figure 2 The diagram illustrates the Extreme Learning Machine (ELM) network structure used in an embodiment of the present invention. The ELM includes an input layer, a single hidden layer, and an output layer. The input layer receives seven operational state features, and the output layer outputs five specific target parameters. These five specific target parameters are respectively categorized into the target control parameters for the sliding surface parameters, the target control parameters for the reaching law parameters, and the target control parameters for the disturbance rejection adjustment parameters.

[0051] It is understandable that speed state calculation, limit learning machine forward computation, parameter updates, and sliding mode control in the control system can be performed by the same processing unit, or the functions can be divided according to the controller hardware resources. The above functional division does not require corresponding to independent physical devices.

[0052] The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machines, provided in the embodiments of the present invention, will be described below with reference to the accompanying drawings. Figure 3 A flowchart illustrating an improved sliding mode control method provided in an embodiment of the present invention. The method includes the following steps: In this embodiment, the operating state characteristics are used to characterize the speed state, reference speed change state, and speed tracking error state of the permanent magnet synchronous motor. One specific form includes actual speed, actual speed change rate, reference speed, reference speed change rate, speed error, speed error change rate, and speed error integral value. Both the control cycle and the control sampling cycle refer to the same update cycle in which the controller completes one state calculation, parameter update, and control quantity calculation. The sliding surface parameter specifically corresponds to the sliding surface integral gain c; the reaching law parameter specifically corresponds to the exponential reaching term gain ε, the constant velocity reaching term gain β, and the boundary layer thickness φ; the disturbance rejection adjustment parameter specifically corresponds to the adaptive term coefficient δ; the adaptive disturbance rejection term gain corresponds to k, and the adaptive term coefficient δ is used to update the adaptive disturbance rejection term gain k; these are different control parameters.

[0053] S1. Obtain the reference speed and actual speed of the permanent magnet synchronous motor, and determine the operating state characteristics based on the reference speed and the actual speed.

[0054] Specifically, within each control cycle, the reference speed is provided by the host controller or speed setting unit, while the actual speed is obtained by encoder detection. Based on the reference speed and the actual speed, operating state characteristics that characterize the motor speed state, the reference speed change state, and the speed tracking error state are determined. The reference speed and the actual speed are the basic inputs for subsequent operating state characteristic calculations, target control parameter generation, and sliding mode control calculations.

[0055] Let the reference speed of the permanent magnet synchronous motor be denoted as... The actual rotational speed detected by the encoder is recorded as . The speed error is then expressed as: ; In one possible implementation, determining the operating state characteristics based on the reference speed and the actual speed includes: performing differential processing on the reference speed and the actual speed to obtain the reference speed change rate and the actual speed change rate; determining the speed error based on the difference between the reference speed and the actual speed; performing differential processing and integral processing on the speed error to obtain the speed error change rate and the speed error integral value; and combining the actual speed, the actual speed change rate, the reference speed, the reference speed change rate, the speed error, the speed error change rate, and the speed error integral value to obtain the operating state characteristics.

[0056] Among them, the The rate of change of the reference rotational speed over each sampling period can be expressed as: ; ; ; In the formula, To control the sampling period index, Indicates the previous control cycle; To control the sampling period, the unit is seconds; and The first The reference speed and actual speed for each sampling period are expressed in rpm in the state calculation and test record; , and These are the reference speed change rate, the actual speed change rate, and the speed error change rate, respectively. When speed is expressed in rpm, the unit is rpm / s. For the first The speed error over each sampling period. The integral value of the speed error is obtained by integrating the speed error over the control sampling period, and is used to characterize the cumulative state of the speed tracking deviation.

[0057] In one specific implementation, the control sampling frequency is 5000Hz, corresponding to a control sampling period of 0.0002s. To reduce the impact of encoder detection noise and discrete difference operations on the rate of change signal, low-pass filtering or moving average methods can be used to process the rate of change signal, as described in this embodiment. The same data processing method and parameters are used in both the offline training phase and the online operation phase.

[0058] It should be noted that the rotational speeds in the experimental records and inputs to the Extreme Learning Machine can be expressed in rpm; when the rotational speed is incorporated into the mechanical motion equations of the permanent magnet synchronous motor, the reference speed and the actual speed use the same unit of mechanical angular velocity. The offline training phase and the online operation phase employ consistent unit and scale processing methods.

[0059] The method provided by this invention combines actual rotational speed, actual rotational speed change rate, reference rotational speed, reference rotational speed change rate, rotational speed error, rotational speed error change rate, and rotational speed error integral value to form operating state features. These operating state features simultaneously include the motor rotational speed state, the reference rotational speed change state, and the current, changing, and accumulated states of the tracking error. These operating state features continue to serve as common inputs for the extreme learning machine parameter mapping, parameter update process, and sliding mode control process.

[0060] S2. Input the operating state features into a pre-trained extreme learning machine for parameter mapping to obtain the target control parameters of the sliding surface parameters, the target control parameters of the reaching law parameters, and the target control parameters of the disturbance rejection adjustment parameters.

[0061] Specifically, such as Figure 2 As shown, the Extreme Learning Machine (ELM) is trained before the permanent magnet synchronous motor (PMSM) is put into online operation, thus forming a pre-trained ELM. During online control, network parameters such as input weights, hidden layer biases, and output weights remain unchanged, and forward computation is performed only based on the operating state characteristics of the current control cycle.

[0062] In one possible implementation, the operating state features are scaled to obtain the extreme learning machine input vector; the extreme learning machine input vector is input into the hidden layer for nonlinear mapping to obtain the hidden layer output vector; the hidden layer output vector is input into the output layer for parameter mapping to obtain the parameter output; the normalized target results corresponding to the sliding mode surface integral gain, exponential reaching term gain, constant velocity reaching term gain, boundary layer thickness, and adaptive term coefficient are determined based on the parameter output, and mapped to five specific target parameters according to the corresponding parameter adjustment range.

[0063] In one specific implementation, the Extreme Learning Machine (ELM) employs a 7-20-5 single-hidden-layer network structure. The seven input nodes correspond sequentially to the actual rotational speed, the rate of change of the actual rotational speed, the reference rotational speed, the rate of change of the reference rotational speed, the rotational speed error, the rate of change of the rotational speed error, and the integral value of the rotational speed error. The twenty hidden-layer nodes are used to perform nonlinear mapping. The five output nodes correspond to the sliding mode surface integral gain, the exponential approach term gain, the constant velocity approach term gain, the boundary layer thickness, and the adaptive term coefficients, respectively.

[0064] In one possible implementation, the training data for the pre-trained extreme learning machine is obtained through the following process: S21. Obtain the reference speed and actual speed of the permanent magnet synchronous motor under different operating conditions, and form training state samples according to the same processing method as the online operation stage.

[0065] Specifically, the training conditions cover constant speed, step speed, trapezoidal speed, sinusoidal speed, and forward / reverse switching; the load conditions cover no-load, constant load, sudden load increase, sudden load decrease, and variable load. Reference speed, actual speed, three-phase stator current, rotor position angle, and load condition are recorded synchronously, and seven training state characteristics are formed based on the reference speed and actual speed.

[0066] No. The group training state samples can be represented as: ; In the formula, For the first Group training state samples, For training sample index; , , , , , and The numbers represent the order of the numbers. The sample includes the actual rotational speed, actual rotational speed change rate, reference rotational speed, reference rotational speed change rate, rotational speed error, rotational speed error change rate, and rotational speed error integral value. Each rotational speed state variable uses the same physical units and data processing methods as during online operation.

[0067] Data that is interrupted by communication, lacks data, has sensor output saturation, fails to enter closed-loop operation normally, or significantly exceeds the preset physical range will not be considered valid training samples. Isolated abnormal spikes can be processed using median filtering, moving average, or anomaly thresholding as described in this embodiment. During stable operation, samples can be extracted at equal intervals, while dynamic samples from startup, reference speed changes, load abrupt changes, and direction switching phases will be retained.

[0068] S22. Perform offline parameter tuning on the five sliding mode control parameters to obtain the target control parameters corresponding to each training state sample.

[0069] Specifically, offline parameter tuning is performed within the allowable ranges of sliding mode surface integral gain, exponential approach term gain, constant velocity approach term gain, boundary layer thickness, and adaptive term coefficients. This embodiment employs a grid search to test candidate parameter combinations and comprehensively considers rotational speed tracking error, dynamic recovery time, cross-axis current variation amplitude, sliding mode surface amplitude, and steady-state rotational speed oscillation degree to select the target control parameter corresponding to the training state sample from the effective candidate combinations.

[0070] ; In the formula, For the first The target control parameter vector corresponding to the group of training state samples; , , , and These are the target values ​​for the sliding mode area integral gain, the exponential reaching term gain, the constant velocity reaching term gain, the boundary layer thickness, and the adaptive term coefficient, respectively. All five parameters are internal control parameters of the controller, and their values ​​are used according to the parameter range and project calibration scale described later; they are not physical quantities directly measured by the sensor.

[0071] S23. Combine the training state samples with the corresponding target control parameters to form extreme learning machine training data, and divide the training data into training set, validation set and test set.

[0072] Specifically, the training set is used to determine the output weights, the validation set is used to compare different random initialization results and corresponding network parameters, and the test set is used for independent evaluation after training. Highly adjacent samples in consecutive time are avoided from being assigned to different datasets simultaneously.

[0073] The method provided by this invention generates training state samples under different rotational speeds, load conditions, and operating directions, and determines the corresponding five target control parameters offline for each training state sample. This enables the training data to establish a correspondence between the operating state and multiple sliding mode control target parameters, providing a data foundation for the offline learning of the Extreme Learning Machine.

[0074] In some embodiments, the training state samples and target control parameters are scaled, the input weights and hidden layer biases of the extreme learning machine are initialized, the training input data is input into the hidden layer for nonlinear mapping, the output weights are determined according to the training output of the hidden layer and the training target data, and the input weights, hidden layer biases and output weights are stored in the controller.

[0075] Specifically, the seven input features have different dimensions and numerical ranges. In this embodiment, 750, 95, 750, 95, 18, 95, and 100 are used as the scale parameters for the corresponding state features, and a hyperbolic tangent function is used for scaling. Following the order of the seven inputs, the online input vector can be represented as: ; In the formula, For the first The extreme learning machine input vector after scaling for each sampling period; For the first The integral value of the rotational speed error over one sampling period; The hyperbolic tangent function can be expressed as: , where exp(·) represents the natural exponential function. The values ​​750, 95, 750, 95, 18, 95 and 100 are the scale parameters of the seven input features, and the components after the scale transformation are dimensionless values.

[0076] The scale parameters mentioned above are derived from the training data range and control experiment settings of this embodiment, and belong to the project calibration parameters. The offline training phase and the online operation phase use the same input order, scale parameters, and scale processing method.

[0077] In one specific implementation, the elements in the input weight matrix independently follow a normal distribution with a mean of 0 and a standard deviation of 0.6, and the elements in the hidden layer bias matrix independently follow a normal distribution with a mean of 0 and a standard deviation of 0.4. The random seed is set to 42. The hidden layer uses the hyperbolic tangent function, and the output layer uses the sigmoid function. The hidden layer gain coefficient is set to 1.15, and the output layer gain coefficient is set to 1.3.

[0078] Input the training input matrix into the hidden layer to obtain the hidden layer training output: ; In the formula, The hidden layer training output matrix has a dimension of . ; For the reason The training input matrix, formed by arranging the group-scaled training inputs row-wise, has a dimension of [missing information]. ; The input weight matrix has dimensions of . ; for A column vector of all 1s; for 3D hidden layer bias vector; This represents the hidden layer gain coefficient. This represents the number of training samples. Perform a hyperbolic tangent nonlinear mapping on each element of the matrix. , , and These are all numerical quantities within the model and do not correspond to independent sensor physical quantity units.

[0079] The five target control parameters are normalized according to their respective lower limits and adjustable ranges to obtain the first... Normalized target output vector corresponding to the group of training state samples: ; In the formula, For the first The normalized target output vector corresponding to the group training state samples; , , , and These are the lower limits of the five control parameters; , , , and These are the adjustable ranges for the five control parameters. Group Arranged in rows to form the training target matrix Its dimensions are The matrix elements are all dimensionless normalized results.

[0080] Since the output layer uses the Sigmoid function, the training target matrix is ​​processed before performing the inverse Sigmoid transformation. Boundary constraints are applied to each element in the data: ; In the formula, For training target matrix The Middle Line number Column elements, , ; The target element after boundary constraints; Let be the target boundary constraint constant, satisfying In one specific implementation, it is taken as 0.001; and These represent taking the minimum value and taking the maximum value, respectively.

[0081] Perform an inverse Sigmoid transform on the normalized objective after boundary constraints to obtain the linear objective matrix of the output layer. : ; In the formula, The linear target matrix of the output layer The Middle Line number Column elements, The dimension is ; This represents the natural logarithm function. Because... It is obtained by taking the natural logarithm of the ratio of the normalized objective value, therefore This is the dimensionless numerical matrix inside the model.

[0082] The output matrix is ​​trained based on the hidden layer. and the linear target matrix of the output layer The output weights are determined using Moore-Penrose generalized inverse analysis. The output weight relationship in this embodiment is as follows: ; In the formula, This is the output weight matrix between the hidden layer and the output layer, with dimension 1. ; Train the output matrix for the hidden layer Moore-Penrose generalized inverse; The output layer is a linear target matrix; This represents the output layer gain coefficient. In one specific implementation, Set to 1.3. and These are all numerical parameters within the model and do not correspond to independent sensor physical quantity units. After training is complete, , and It remains unchanged during online operation.

[0083] When running online, the first Scaled input vector for each sampling period Input the hidden layer, and obtain the hidden layer output vector: ; In the formula, For the first The hidden layer output vector for each sampling period has a dimension of . ; Scale the input vector transpose; , and The input weight matrix, hidden layer bias vector, and hidden layer gain coefficients stored during the offline training phase are reused.

[0084] Input the hidden layer output vector into the output layer to obtain five normalized target results: ; The Sigmoid function is defined as follows: ; The output vector is represented as: ; In the formula, For the first The output vector of the extreme learning machine for each sampling period, with dimension [missing information]. ; , , , and Corresponding to the sliding mode area integral gain Exponential approach term gain Gain of constant velocity approaching term Boundary layer thickness and adaptive term coefficients The normalized objective results are all dimensionless values; For the input variables of the Sigmoid function, This represents the natural exponential function.

[0085] For the The control parameters, based on their corresponding lower limits and adjustable ranges, map the normalized target results to the actual target parameters: ; In the formula, For the first The sampling period The target value of the control parameter, ; For the first The lower limit of the control parameter. For the first Adjustable range of the control parameter. The nth vector in the output vector of the extreme learning machine The results of the normalized objective are as follows. The five parameters are calculated according to... Corresponding to 5 in sequence , , , and ; The target parameter uses the same numerical scale as the corresponding actual control parameter. It serves as the adjustment target for subsequent actual control parameters and does not directly override the control parameter of the previous control cycle.

[0086] The target parameter values ​​of the sliding mode surface integral gain are determined as the target control parameters of the sliding mode surface parameters. The target parameter values ​​of the exponential reaching term gain, the constant velocity reaching term gain, and the boundary layer thickness are jointly determined as the target control parameters of the reaching law parameters. The target parameter values ​​of the adaptive term coefficients are determined as the target control parameters of the disturbance rejection adjustment parameters. Thus, the pre-trained extreme learning machine maintains unchanged network parameters and performs only forward computation during the online phase. Simultaneously, the five specific target parameters are categorized into three types of target control parameters, providing corresponding adjustment targets for subsequent updates of the three types of parameters according to the parameter update rate.

[0087] S3. Determine the system performance evaluation index and parameter update rate based on the operating state characteristics, and update the sliding surface parameters, reaching law parameters and disturbance rejection adjustment parameters based on the parameter update rate to obtain the sliding control parameters for the current control cycle.

[0088] Specifically, real-time speed error, real-time speed error change rate, and reference speed change rate are extracted from the operating state characteristics. These three state variables are then scaled and combined to form a system performance evaluation index. This system performance evaluation index reflects the current speed tracking deviation, the deviation change status, and the reference speed change status, and serves as the basis for determining the parameter update rate.

[0089] ; In the formula, For the first The performance evaluation index for each sampling period is a dimensionless value. For speed error, The rate of change of rotational speed error. The reference rotational speed change rate; , and These are the characteristic scale values ​​for the rotational speed error, the rate of change of the rotational speed error, and the rate of change of the reference rotational speed, respectively. All state variables and their corresponding characteristic scales use the same unit; therefore, all three ratios are dimensionless values.

[0090] In one specific implementation, based on typical operating data such as start-up, acceleration, deceleration, constant speed operation, trapezoidal speed tracking, sinusoidal speed tracking, sudden load increase, sudden load decrease, and forward / reverse switching, and combined with control test adjustments, the characteristic scale value of the speed error is set to 25 rpm, the characteristic scale value of the speed error change rate is set to 120 rpm / s, and the characteristic scale value of the reference speed change rate is set to 1000 rpm / s. These values ​​are project calibration parameters.

[0091] In one possible implementation, the real-time speed error is compared with preset dynamic state conditions and preset stable state conditions for speed error, respectively, and the real-time speed error change rate is compared with preset dynamic state conditions and preset stable state conditions for speed error change rate, respectively, to obtain a state range determination result; based on the state range determination result, the update rate gain coefficient of the previous control cycle is increased, decreased, or maintained. The update rate gain relationship before limiting corresponding to this embodiment is as follows: ; In the formula, For the first Update rate gain before clipping in each sampling period, This is the update rate gain after the previous control cycle's limiting. Represents logical OR, This indicates a logical AND operation. When rpm is used to represent rotational speed, the unit for the speed error thresholds 15 and 3 is rpm, and the unit for the speed error change rate thresholds 80 and 20 is rpm / s. The update rate gain is increased when the system is in the dynamic range, decreased when it simultaneously enters the steady-state range, and otherwise maintained based on the update rate gain from the previous control cycle.

[0092] By applying upper and lower bound constraints to the update rate gain before limiting, the actual update rate gain used in the current control cycle can be obtained: ; In the formula, For the first Update rate gain after limiting for each sampling period; and These are the lower and upper limits of the update rate gain, respectively. and These represent taking the minimum value and taking the maximum value, respectively.

[0093] In one specific implementation, the initial value of the update rate gain is set to 1.5, the lower limit is set to 0.7, and the upper limit is set to 3.0. The dynamic operating condition gain ratio is 1.1, and the steady-state operating condition attenuation ratio is 0.97.

[0094] The parameter update rate before limiting is determined based on the update rate gain coefficient of the current control cycle and the system performance evaluation index. This embodiment uses the basic parameter update rate as the calibration benchmark and employs a dynamic adjustment coefficient to allow the system performance evaluation index to participate in the determination of the parameter update rate before limiting under dynamic conditions. ; In the formula, For the first The dynamic adjustment coefficient for each sampling period; As a performance evaluation metric; This indicates that 1 and 2 are taken when the dynamic condition is met. The larger value in; This indicates a logical OR.

[0095] ; In the formula, For the first Parameter update rate before amplitude limiting in each sampling period This is the update rate gain after limiting. Based on the update rate of basic parameters, This is a dynamic adjustment coefficient. In one specific implementation, Set to 0.018. The parameter update rate is a dimensionless proportionality coefficient representing the shift of the control parameters towards the target control parameters.

[0096] To avoid abrupt changes in control parameters due to an excessively high parameter update rate, and to prevent the actual control parameters from losing their necessary online adjustment capability due to an excessively low parameter update rate, upper and lower limits are imposed on the parameter update rate before limiting: ; In the formula, For the first The actual parameter update rate used in each sampling period; and These are the lower and upper limits of the parameter update rate, respectively, and satisfy the following conditions: ; This is the parameter update rate before amplitude limiting.

[0097] The upper and lower limits of the parameter update rate are calibrated based on the allowable changes in five specific control parameters per control cycle, the changes in the quadrature shaft reference current, and control tests for start-up, reference speed changes, sudden load increases, sudden load decreases, and direction of travel switching. This embodiment can also gradually adjust the upper and lower limits of the update rate and the upper and lower limits of the update rate gain by 5% to 20% of the current parameter values, and repeatedly execute control tests to determine the parameters suitable for the corresponding motor and operating range.

[0098] In one possible implementation, the sliding surface parameter difference between the sliding surface parameter and its target control parameter in the previous control cycle is determined; the reaching law parameter difference between the reaching law parameter and its target control parameter in the previous control cycle is determined; and the disturbance rejection adjustment parameter difference between the disturbance rejection adjustment parameter and its target control parameter in the previous control cycle is determined. The three types of parameter differences are adjusted according to the parameter update rate to obtain the corresponding parameter changes, and these parameter changes are used to update the three types of parameters in the previous control cycle. For any specific control parameter among the sliding surface integral gain, exponential reaching term gain, constant velocity reaching term gain, boundary layer thickness, and adaptive term coefficient, it can be expressed as: ; In the formula, For the first The sampling period Actual control parameters, These are the actual control parameters corresponding to the previous control cycle. The target control parameters are for the current control cycle. The update rate of the parameters for the current control cycle. The five specific parameters correspond to the following in sequence. , , , and .

[0099] The updated sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters are subject to parameter range constraints. Specifically, the updated sliding surface integral gain, exponential reaching term gain, constant velocity reaching term gain, boundary layer thickness, and adaptive term coefficient are each limited to their respective preset parameter ranges. In this embodiment, the parameter ranges are 50–150, 5–25, 2–20, 0.15–0.70, and 0.1–0.8, respectively. When the control system starts, the five specific parameters are initialized to 90, 15, 8, 0.7, and 0.45, respectively. The sliding surface integral gain after parameter range constraints constitutes the sliding surface parameters for the current control cycle. The exponential reaching term gain, constant velocity reaching term gain, and boundary layer thickness after parameter range constraints constitute the reaching law parameters for the current control cycle. The adaptive term coefficient after parameter range constraints constitutes the disturbance rejection adjustment parameters for the current control cycle. These three types of parameters together constitute the sliding mode control parameters for the current control cycle.

[0100] Furthermore, to impose a lower limit constraint on the updated boundary layer thickness that is compatible with the sliding surface disturbance, the upper limit of the sliding surface disturbance Δsmax and the preset upper limit of the fluctuation of the constant velocity approach term Δuβ,max are obtained. Δsmax can be calibrated based on the sensor resolution, discrete difference error, and the range of sliding surface disturbance in the steady-state operation sample, while Δuβ,max is preset as the fluctuation limit of the constant velocity approach term allowed by the controller.

[0101] Based on the local relationship of the continuous saturated isotropic approach term within the boundary layer, its change in sliding surface disturbance satisfies Δuβ≤βΔsmax / φ. Therefore, the lower limit of boundary layer thickness φmin=βΔsmax / Δuβ,max is determined according to the isotropic approach term gain β of the current control cycle, the upper limit of sliding surface disturbance Δsmax, and the preset upper limit of fluctuation Δuβ,max.

[0102] The boundary layer thickness obtained according to the target control parameters and parameter update rate and constrained by the original parameter range is denoted as φr. The preset upper limit of the boundary layer thickness is denoted as φmax. Then, the updated boundary layer thickness is further constrained according to the lower limit of the boundary layer thickness. The actual boundary layer thickness used in the current control cycle is determined according to φ=max[φr,min(φmax,φmin)], and together with the sliding mode surface integral gain, exponential approach term gain, constant velocity approach term gain and adaptive term coefficient constrained by the parameter range, it forms the sliding mode control parameters of the current control cycle.

[0103] The method provided by this invention determines the system performance evaluation index and state range determination result by analyzing the operating state characteristics, and then determines the parameter update rate. Based on the parameter update rate and the difference between the three types of parameters and the corresponding target control parameters, the parameter change is formed, so that the sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters of the current control cycle are simultaneously constrained by the operating state, target control parameters, and parameters of the previous control cycle. At the same time, a lower limit constraint based on the upper limit of the sliding surface disturbance and the preset upper limit of the fluctuation is added to the boundary layer thickness, so that the sliding control parameters of the current control cycle entering the subsequent sliding control calculation complete the range constraint during the parameter update stage.

[0104] S4. Construct a sliding surface based on the operating state characteristics and the sliding surface parameters of the current control cycle, and update the adaptive disturbance rejection term gain based on the sliding surface and the disturbance rejection adjustment parameters of the current control cycle.

[0105] Specifically, the real-time speed error, the rate of change of the real-time speed error, and the integral value of the real-time speed error are extracted from the operating state characteristics. A sliding surface is constructed based on the three state variables and the sliding surface integral gain of the current control cycle. The sliding surface integral gain of the current control cycle is the sliding surface parameter of the current control cycle.

[0106] ; In the formula, It is a sliding surface; This refers to the rotational speed error; This represents the rate of change of rotational speed error. This is the integral of the rotational speed error from the initial moment to the current moment; This is the sliding mode surface integral gain actually used in the current control cycle; The values ​​are integral variables. The speed error, the rate of change of speed error, and the integral value all use the same numerical units and sampling methods as the aforementioned operating state characteristics.

[0107] The speed error term reflects the current speed tracking deviation, the integral term reflects the cumulative state of the speed error, and the speed error change rate term reflects the changing trend of the speed error.

[0108] Taking the derivative with respect to the sliding surface, and neglecting the second derivative term of the rotational speed error, which is easily affected by detection noise, in this embodiment of the controller design, the rate of change of the sliding surface is approximately expressed as: ; In the formula, Let be the first-order rate of change of the sliding mode surface over time; and These are the first-order rates of change of the reference speed and the actual speed, respectively; and The aforementioned definitions are retained. It should be noted that in this embodiment, the rotational speed quantities involved in the calculation of rotational speed state and mechanical motion relationship adopt the same physical quantity definition and derivative order, so as to maintain the consistency of the dimensions and data meanings in the sliding surface change relationship, mechanical motion relationship, and cross-axis reference current calculation.

[0109] In one possible implementation, the sliding surface region determination result is determined based on the positional relationship between the sliding surface and a preset sliding surface region. The sliding surface region determination result includes both an out-of-region determination result and an in-region determination result. When the sliding surface is outside the preset sliding surface region, the adaptive disturbance rejection term gain change is determined based on the adaptive term coefficients of the current control cycle and the sliding surface, and the adaptive disturbance rejection term gain of the previous control cycle is updated based on the change. When the sliding surface is within the preset sliding surface region, the adaptive disturbance rejection term gain of the previous control cycle is attenuated. The discrete update relationship corresponding to this embodiment is: ; In the formula, For the first The adaptive disturbance rejection term gain for each sampling period, This is the gain of the adaptive disturbance rejection term from the previous control cycle; and These are the lower and upper limits of the adaptive disturbance rejection term gain, respectively. For the first The adaptive term coefficients for each sampling period; The current sliding surface; To control the sampling period, the unit is seconds; The threshold value for the sliding surface region; and These represent taking the minimum value and taking the maximum value, respectively. , and All of these are internal control gains for improving the reaching law, used according to the controller parameter scale, and are not independent measured physical quantities. In one specific implementation, Set it to 0.02, and the attenuation coefficient to 0.995.

[0110] The upper and lower limits of the adaptive disturbance rejection term gain can be calibrated based on offline operating data before the controller is put into operation. Specifically, the equivalent disturbance amplitude is statistically analyzed in effective operating samples where the sliding surface enters a preset adjacent region to determine the lower limit of the adaptive disturbance rejection term gain; the equivalent disturbance amplitude is statistically analyzed in effective design conditions such as start-up, speed change, sudden load increase, sudden load decrease, and direction change to determine the upper limit of the adaptive disturbance rejection term gain. Corresponding disturbance rejection margins are reserved for both the upper and lower limits. This calibration process provides an engineering range for the adaptive disturbance rejection term gain.

[0111] During the offline calibration phase, the equivalent disturbance can be determined based on the recorded quadrature-axis current and rate of change of mechanical angular velocity, according to... The calculation shows that this calculation is only used for offline parameter calibration and does not constitute an online independent disturbance observer.

[0112] S5. Construct an improved approach law based on the sliding surface, the approach law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain. Determine the quadrature axis reference current based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved approach law. Perform closed-loop control on the permanent magnet synchronous motor based on the quadrature axis reference current to obtain the updated actual speed.

[0113] In one possible implementation, an exponential reaching term is determined based on the sliding surface and the exponential reaching term gain of the current control cycle; the continuous saturation function output is determined based on the sliding surface and the boundary layer thickness of the current control cycle, and a constant velocity reaching term is determined by combining it with the constant velocity reaching term gain of the current control cycle; an adaptive disturbance rejection term is determined based on the sliding surface and the updated adaptive disturbance rejection term gain; and the exponential reaching term, constant velocity reaching term, and adaptive disturbance rejection term are combined to obtain the improved reaching law. The improved reaching law corresponding to this embodiment is: ; The saturation function is: ; In the formula, The rate of change of the sliding surface. This is the gain of the exponential approach term of the current control period. This represents the gain of the iso-rhythmic approach term in the current control cycle. The boundary layer thickness for the current control cycle and , This represents the current adaptive disturbance rejection term gain. It is a continuously saturated function, and its input is ; A symbolic function used to represent sliding surfaces. The sign of s. The sign function sign(s) is defined as: 1 when s is greater than 0, 0 when s equals 0, and -1 when s is less than 0. When the saturation function takes a linear continuous form; when or When the saturation function takes the values ​​1 or -1, the saturation function takes the values ​​1 or -1 respectively

[0114] When the sliding surface is located within the boundary layer, the continuously saturated isotropic reaching term changes continuously with the sliding surface. This embodiment further demonstrates the correspondence between the boundary layer thickness and the sensitivity of the isotropic reaching term to small perturbations of the sliding surface. When the gain of the isotropic reaching term remains constant, this local sensitivity is inversely proportional to the boundary layer thickness.

[0115] ; In the formula, This represents the local sensitivity coefficient of the continuous saturated isotropic approach term within the boundary layer to changes in the sliding surface. For the gain of the isotropic approach term, This represents the boundary layer thickness.

[0116] Therefore, as the boundary layer thickness increases, the sensitivity of the continuous saturated constant velocity approaching term to small changes in the sliding surface decreases; conversely, as the boundary layer thickness decreases, this sensitivity increases. This embodiment provides an adjustment range of 0.15 to 0.70 for the boundary layer thickness. The above relationship represents the theoretical relationship of the approaching term and does not imply that the actual chattering amplitude of the entire motor closed-loop system changes proportionally.

[0117] In digital control implementation, after updating the five specific control parameters and the adaptive disturbance rejection term gain at a single sampling time, this set of parameters is used for sliding mode control calculations within the current control cycle, and then recalculated and updated at the next sampling time. Therefore, the current control parameters can be considered constant within a single control cycle.

[0118] When determining the quadrature-axis reference current based on the improved approach law, the solution is derived by combining the mechanical motion relationship of the permanent magnet synchronous motor. In one possible implementation, the controlled object is a surface-mounted permanent magnet synchronous motor, with the direct-axis inductance equal to the quadrature-axis inductance, and field-oriented control with zero direct-axis current is employed. In this case, the electromagnetic torque is: ; The mechanical motion relationship of the permanent magnet synchronous motor is as follows: ; Define the equivalent total disturbance as the combined effect of load torque, viscous damping, motor parameter variations, and unmodeled dynamics in the mechanical motion equations, then: ; In the formula, Electromagnetic torque, in N·m; This represents the number of pole pairs of the motor. The flux linkage of a permanent magnet is measured in Wb. This is the quadrature-axis current, measured in amperes (A). This is the equivalent moment of inertia of the motor and load, expressed in kg·m². The total disturbance torque is expressed in N·m. This is the equivalent total disturbance acting on the speed ring. The time variable. In mechanical motion relationships. The mechanical angular velocity is used, with the unit being rad / s, and its first rate of change is in the unit of rad / s².

[0119] This embodiment uses Lyapunov functions to analyze sliding mode dynamics: ; In the formula, The Lyapunov function is used to analyze the sliding mode dynamics within a single control sampling period. It is a sliding surface.

[0120] Considering that the five specific control parameters and the adaptive disturbance rejection term gain in this embodiment are all updated discretely according to the control period, the sliding surface approach is analyzed under the condition that the current parameters remain unchanged within a single control sampling period. When the equivalent disturbance is bounded and the current adaptive disturbance rejection term gain can cover the corresponding disturbance effect, the exponential approach term, the continuous saturated constant velocity approach term, and the adaptive disturbance rejection term together enable the sliding surface to approach the preset region; after the sliding surface enters the preset region, the adaptive disturbance rejection term gain gradually decays according to the second branch and is constrained by the lower limit.

[0121] The method provided by this invention involves the sliding surface integral gain of the current control cycle in determining the sliding surface, and uses the current sliding surface and adaptive term coefficients to update the adaptive disturbance rejection term gain, so that the model parameter update results can further affect the disturbance rejection control; then the exponential reaching term gain, constant velocity reaching term gain, boundary layer thickness and adaptive disturbance rejection term gain are respectively applied to the corresponding reaching terms, so that different control parameters are implemented in the corresponding control links of the improved reaching law.

[0122] Specifically, the sliding surface variation relationship is combined with the improved reaching law to obtain the desired mechanical angular velocity variation relationship that satisfies the improved reaching law; then, the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved reaching law are combined for solution. In the actual controller, no independent disturbance observer is set up to explicitly estimate and compensate for the equivalent total disturbance. The quadrature-axis reference current obtained in this embodiment is: ; In the formula, This is the quadrature-axis reference current, in amperes (A). , and These are the equivalent moment of inertia, number of pole pairs, and permanent magnet flux linkage, respectively. For reference, the rate of change of mechanical angular velocity; , , , , , and All of these are state variables or control parameters corresponding to the current control cycle. Simultaneously, a field-oriented control strategy with zero direct-axis current is adopted, setting the direct-axis reference current to: ; In the formula, The current is the direct-axis reference current, measured in amperes (A). This embodiment employs a field-oriented control strategy with a zero direct-axis reference current.

[0123] In one possible implementation, the permanent magnet synchronous motor is controlled in a closed loop based on the quadrature axis reference current: the current inner loop is adjusted based on the quadrature axis reference current to obtain the voltage control quantity; space vector pulse width modulation is performed based on the voltage control quantity to obtain the inverter switch drive signal, and the permanent magnet synchronous motor is controlled to run based on the inverter switch drive signal, so that the actual speed of the permanent magnet synchronous motor is updated; the updated actual speed is used to redetermine the operating state characteristics in the next control cycle.

[0124] Specifically, current sensors collect three-phase stator currents, and encoders provide rotor position angles. First, a Clarke transform is performed on the three-phase stator currents to obtain two-phase stationary coordinate system currents: ; Then, based on the rotor electrical angle, perform the Park transformation to obtain the actual current of the direct axis and the actual current of the quadrature axis: ; In the formula, , and For three-phase stator current, and For two-phase stationary coordinate system currents, and The above currents refer to the actual currents along the direct axis and the actual currents along the quadrature axis in a synchronously rotating coordinate system. The unit of the above current quantities is A. The rotor electrical angle is expressed in rad and is used in trigonometric function calculations.

[0125] Calculate the direct-axis current deviation and quadrature-axis current deviation separately: ; In the formula, and These are the direct-axis current deviation and the quadrature-axis current deviation, respectively, in A; and These are the direct-axis reference current and the quadrature-axis reference current, respectively. and These correspond to the actual currents.

[0126] In one specific implementation, both the direct-axis current inner loop and the quadrature-axis current inner loop employ proportional-integral controllers to obtain the direct-axis reference voltage and the quadrature-axis reference voltage, respectively: ; ; In the formula, and These are the direct-axis reference voltage and the quadrature-axis reference voltage, respectively, in V; , These are the proportional and integral coefficients of the direct-axis current loop, respectively. , These are the proportional coefficient and integral coefficient of the quadrature-axis current loop, respectively. and To correspond to the current deviation, The time variable is used. The proportional and integral coefficients are tuned in the current loop according to the current and voltage scales used by the controller.

[0127] The direct-axis reference voltage and quadrature-axis reference voltage are subjected to inverse Park transformation to obtain the reference voltage in the two-phase stationary coordinate system: ; In the formula, and These are the reference voltages in the two-phase stationary coordinate system, in units of V; , and The aforementioned definition will be used.

[0128] Subsequently, the two-phase stationary coordinate system reference voltage is input to the space vector pulse width modulation (SPWM) module. Based on the sector containing the reference voltage vector and the duration of action of adjacent basic voltage vectors and the zero vector, the SPWM module obtains the duty cycle of the three-phase bridge arms and generates the drive signals for the three-phase inverter power switching devices via the controller's PWM output interface. The three-phase inverter converts the DC bus voltage into three-phase AC voltage based on the switching drive signals and applies it to the stator windings of the permanent magnet synchronous motor.

[0129] The actual direct-axis current and quadrature-axis current track the corresponding reference current under the action of the inner current loop. The quadrature-axis current changes the electromagnetic torque, which in turn changes the actual speed of the permanent magnet synchronous motor, resulting in an updated actual speed. In the next control cycle, the updated actual speed, rotor position angle, and three-phase stator current are reacquired, and S1-S5 are executed again, thus forming a closed-loop control.

[0130] The method provided by this invention implements the sliding mode control output as a quadrature axis reference current, and applies it to the motor through the current inner loop, coordinate inverse transformation, space vector pulse width modulation and three-phase inverter, so that the aforementioned target control parameter generation and actual parameter update results can be transmitted to the physical execution process of the permanent magnet synchronous motor.

[0131] As described in S1-S5 above, this embodiment of the invention first determines the operating state characteristics based on the reference speed and the actual speed; then, the operating state characteristics are input into a pre-trained extreme learning machine for parameter mapping to obtain the target control parameters corresponding to the sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters; next, the system performance evaluation index and parameter update rate are determined based on the operating state characteristics, and the three types of parameters are updated according to the parameter update rate to complete the parameter range constraints, including the lower limit constraint of the boundary layer thickness, and obtain the sliding control parameters for the current control cycle; subsequently, a sliding surface is constructed based on the operating state characteristics and the sliding surface parameters for the current control cycle, and the adaptive disturbance rejection term gain is updated based on the sliding surface and the disturbance rejection adjustment parameters for the current control cycle; finally, an improved reaching law is constructed based on the sliding surface, the reaching law parameters for the current control cycle, and the updated adaptive disturbance rejection term gain, and the quadrature axis reference current is determined in combination with the mechanical motion relationship of the permanent magnet synchronous motor, and closed-loop control is performed based on the quadrature axis reference current to obtain the updated actual speed. Thus, a continuous data processing chain is formed between operating state determination, target control parameter generation, parameter updating, sliding state processing, control quantity determination, and motor closed-loop execution.

[0132] In some embodiments, to ensure that the online operation process can distinguish between normal adaptive control and abnormal data states, the validity of the input data and the training range of the model can be checked.

[0133] In one possible implementation, when a complete set of seven operating state characteristics cannot be reliably generated in a given control cycle, but the encoder, current sampler, and inverter are still in effective operation, the updating of sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters using the current limit learning machine output is paused, and the sliding mode control parameters from the previous effective control cycle are maintained. Once the operating state characteristics become effective again, the generation and updating of target control parameters are resumed. If the system has just started and the parameters from the previous effective control cycle do not yet exist, the aforementioned initial control parameters are used.

[0134] In another possible implementation, after offline training is completed, the effective ranges corresponding to the seven training inputs are saved together with the model parameters. When the online running state features significantly exceed the coverage of the training data, the model output of the control cycle is not used to change the current sliding mode control parameters, but the sliding mode control parameters of the previous effective control cycle are maintained; when the running state features re-enter the coverage of the training data, online parameter updates are resumed. This processing is only used as boundary protection for model application and does not change the aforementioned closed-loop data processing chain.

[0135] It should be noted that in the online operation phase of this embodiment, model predictive control optimization, nonlinear programming, or other online optimization processes that require iterative search for feasible solutions are not performed. The online phase of the Extreme Learning Machine uses fixed-weight forward mapping, and the updates of the five control parameters are deterministic operations. Therefore, there are no processing branches where the online optimization solver cannot obtain a feasible solution.

[0136] In the event of equipment failure, when the controller receives an existing fault status from the encoder, current sampling circuit, or inverter, or when key feedback quantities cannot be effectively obtained, it will not continue to use invalid feedback to perform extreme learning machine parameter updates and speed closed-loop calculations, and the motor drive system will execute existing protection procedures. This embodiment does not construct separate algorithms for encoder fault diagnosis, current sensor fault diagnosis, or power device fault identification.

[0137] For execution deviations within the normal operating range, such as tracking deviations between the actual quadrature axis current and the quadrature axis reference current, parameter perturbations, and unmodeled dynamics, their impact can be incorporated into the bounded equivalent disturbance along with load disturbances according to this embodiment, and handled by the adaptive disturbance rejection term in the improved approach law; if the execution deviation exceeds the normal operating range of the drive equipment and triggers hardware protection, it will be handled as a device fault.

[0138] To facilitate understanding of this solution, the following detailed explanation of the method provided in this embodiment of the invention will be provided with reference to a specific example.

[0139] In this example, a test platform for a permanent magnet synchronous motor is built. Figure 4 The structure of the test platform is shown. The test platform includes a host computer, a controller, a driver, a DC power supply, a controlled permanent magnet synchronous motor, and a load motor. The host computer is used to send reference speed and test condition commands and record operating data; the controller is used to execute the above S1-S5; the load motor is coaxially connected to the controlled permanent magnet synchronous motor to form operating conditions such as constant load, sudden load increase, sudden load decrease, and variable load.

[0140] Figure 5 The dual closed-loop control structure of the test platform is shown. The outer velocity loop employs an improved sliding mode control using an extreme learning machine, as described in this embodiment of the invention, while the direct-axis current inner loop and the quadrature-axis current inner loop employ proportional-integral control.

[0141] Table 1. Main parameters of the permanent magnet synchronous motor used in the experiment. The rated parameters, resistance, inductance, flux linkage, and moment of inertia in Table 1 are parameters of the experimental permanent magnet synchronous motor and are not part of the model prediction results.

[0142] Table 2. Main parameter settings for the improved sliding mode controller and extreme learning machine. The network structure, activation gain, input scale, initial control parameters, and update rate values ​​in Table 2 are derived from the control experiments and model training settings of this embodiment and are illustrative settings or project calibration parameters. The extreme learning machine output is the model output, while the actual rotational speed and stator current collected by the sensors are measured data, both of which are recorded separately in this embodiment.

[0143] Table 3 Examples of training conditions for the Extreme Learning Machine The operating conditions and sample sizes in Table 3 are the training data settings for this embodiment. The training samples are used to obtain model parameters and should not be mixed with performance data from subsequent closed-loop tests.

[0144] Specifically, after completing the offline training of the Extreme Learning Machine, the fixed network parameters obtained from the training are written into the controller. When the control system starts, the five actual control parameters are set to the initial values ​​in Table 2. During each control sampling cycle, the aforementioned S1-S5 are executed, and the five control parameters are adjusted according to the current model target value and parameter update rate.

[0145] Figure 6 (a) shows the regression analysis results between the Extreme Learning Machine output and the corresponding target parameters. Within the training and validation data range used in this embodiment, the goodness of fit of the model output regression is 0.99999997. This value is used to evaluate the consistency between the Extreme Learning Machine model output and the target parameters, and belongs to the model evaluation results.

[0146] Figure 6 (b) shows the relative tracking error of the permanent magnet synchronous motor during closed-loop operation. Of the 1,400,000 sampled data points recorded in the closed-loop test of this embodiment, 99.45% of the sampled points had an absolute value of relative tracking error less than 2%, and 99.76% had an absolute value less than 5%. This data belongs to the closed-loop motor test results and is not the parameter prediction error of the extreme learning machine.

[0147] Figure 7 The variation of the sliding mode surface integral gain during operation is shown. The curve illustrates that after the target parameters of the extreme learning machine undergo a parameter update process, the actual control parameters can be adjusted according to the reference speed, actual speed, and load condition. The remaining four control parameters are processed according to the same parameter update chain.

[0148] Figure 8 The quadrature-axis current response under different speed outer loop control methods is shown. When the reference speed or load condition changes, the quadrature-axis current changes with the speed outer loop control requirements. Figure 8 The curves in the figure represent the experimental results.

[0149] Figure 9 The changes in the sliding surface are shown under continuously varying reference speed and load disturbance conditions. When the reference speed or load condition changes, the sliding surface deviates and re-enters the region near the sliding surface under subsequent control actions.

[0150] Figure 10 The speed response under different outer-loop control methods is shown. Under the corresponding test conditions, the speed overshoot of the control method in this embodiment is less than 0.2 rpm during the start-up phase, the speed error after entering the steady state is less than 0.01 rpm, and the speed oscillation amplitude is less than 0.1 rpm. The above values ​​are all measured results under the corresponding test conditions.

[0151] Figure 11 The speed error under different outer-loop control methods is shown. During dynamic moments such as startup, sudden load increase, sudden load decrease, and speed direction switching, the speed error changes briefly and then returns to near the zero error line in subsequent control processes.

[0152] To quantitatively evaluate the speed tracking performance of different control methods, the root mean square error of the speed is used in this embodiment: ; In the formula, This is the root mean square error of the rotational speed, using the same units as the rotational speed. The number of sampling points involved in the calculation; For sampling point index, , here This only represents the sampling point index in the root mean square error calculation, and is different from the disturbance rejection gain in the aforementioned improved reaching law. Different meanings; and The first The reference and actual rotational speeds at each sampling point are all from the same set of control test records. In the test results of this embodiment, rotational speed is expressed in rpm. Therefore... The unit is rpm. The corresponding comparison results are as follows: Table 4 Comparison of root mean square error of rotational speed for different control methods The values ​​in Table 4 are all derived from the torsion control test in this embodiment and are not theoretical calculation values ​​or model prediction values.

[0153] Table 5 Comparison of load disturbance suppression performance of different control methods The maximum speed deviation, overshoot, and recovery time in Table 5 are all measured results under the given motor, reference speed, load conditions, and parameters in this embodiment, and do not necessarily mean that other motors or other operating conditions will obtain the same values.

[0154] Figure 12 The diagram illustrates the speed-current trajectory of a permanent magnet synchronous motor, composed of its actual speed and quadrature-axis current. In this embodiment, since the direction of the load torque applied by the load motor remains unchanged, the trajectory is primarily located in the first and second quadrants. This figure is used to illustrate the correspondence between quadrature-axis current and actual speed during forward operation, reverse operation, and direction switching.

[0155] Combination Figures 6 to 12 As shown in Tables 4 and 5, this embodiment records the operating results from the perspectives of model output, closed-loop speed tracking, control parameter changes, quadrature-axis current, sliding surface, speed error, and operating quadrant. The sources of the above measured results and model evaluation results are indicated. The model output evaluation indicators are not interpreted as the actual control accuracy of the motor, nor are the theoretical sensitivity relationships interpreted as the proportion of actual performance improvement.

[0156] The above mainly describes the solution of the embodiments of the present invention from a methodological perspective. It is understood that, in order to achieve the above functions, the control system includes at least one of the hardware structures and software modules corresponding to each function. In conjunction with the processing steps described in the above method embodiments, each function can be executed by the processing unit in the controller, or it can be jointly implemented by the processing unit and the encoder, current sensor, interface circuit, and three-phase inverter.

[0157] In this embodiment of the invention, the control system can be functionally divided according to the above method example. The control system may include an operating state determination module, a target parameter generation module, an update rate determination module, a parameter update module, a sliding mode state processing module, a control quantity determination module, and a closed-loop control module. The above modules are logical functional divisions; two or more modules can be implemented by the same processing unit, or they can be implemented separately according to the controller hardware resources.

[0158] For example, the operating state determination module is used to acquire the reference speed and the actual speed and determine the operating state characteristics; the target parameter generation module is used to input the operating state characteristics into a pre-trained extreme learning machine to obtain the target control parameters corresponding to the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters; the update rate determination module is used to determine the system performance evaluation index and the parameter update rate based on the operating state characteristics; the parameter update module is used to obtain the sliding mode control parameters for the current control cycle based on the parameter update rate, the sliding mode control parameters of the previous control cycle, and the target control parameters; the sliding mode state processing module is used to construct the sliding surface and update the adaptive disturbance rejection term gain; the control quantity determination module is used to construct the improved reaching law and determine the quadrature axis reference current; and the closed-loop control module is used to control the permanent magnet synchronous motor to operate based on the quadrature axis reference current and obtain the updated actual speed.

[0159] Specifically, the inputs to the operating state determination module are the reference rotational speed and the actual rotational speed output by the encoder. The operating state determination module calculates the rate of change, error, and error integral through the controller processing unit and the operating memory, outputs the operating state characteristics, and provides these characteristics to the target parameter generation module, the update rate determination module, and the sliding mode state processing module, respectively. In physical implementation, the operating state determination module can correspond to the reference rotational speed input interface, the encoder interface, the processing unit, and the operating memory.

[0160] The target parameter generation module takes runtime features as input. It calls the input scale parameters, input weights, hidden layer biases, output weights, and control parameter adjustment ranges stored in the storage unit to perform an extreme learning machine forward mapping, outputting target control parameters, which are then provided to the parameter update module. In physical implementation, the target parameter generation module primarily corresponds to the processing unit and non-volatile memory used to store fixed model parameters.

[0161] The update rate determination module takes operating state characteristics as its input. Based on the speed error, the speed error change rate, and the reference speed change rate, the update rate determination module generates system performance evaluation indicators, updates the update rate gain according to the state range, determines the parameter update rate, and provides the parameter update rate to the parameter update module. The update rate gain of the previous control cycle is stored in the operating memory.

[0162] The parameter update module receives inputs including the parameter update rate, the target control parameters, and the sliding mode control parameters from the previous control cycle. Based on the difference between the target control parameters and the corresponding sliding mode control parameters from the previous control cycle, and the parameter update rate, the parameter update module obtains the sliding mode control parameters for the current control cycle and provides them to the sliding mode state processing module and the control quantity determination module. The sliding mode control parameters from the previous control cycle are stored in the runtime memory.

[0163] The sliding mode state processing module constructs a sliding surface based on the operating state characteristics and the sliding surface parameters of the current control cycle, and updates the adaptive disturbance rejection term gain based on the sliding surface and the disturbance rejection adjustment parameters of the current control cycle. The control quantity determination module constructs an improved reaching law based on the sliding surface, the reaching law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain, and determines the quadrature axis reference current in conjunction with the mechanical motion relationship of the permanent magnet synchronous motor. The closed-loop control module performs closed-loop control of the permanent magnet synchronous motor through the current inner loop, coordinate transformation, space vector pulse width modulation, and three-phase inverter, so that the actual speed of the motor after operation is fed back to the next control cycle.

Claims

1. An improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machines, characterized in that, The method includes: The reference speed and actual speed of the permanent magnet synchronous motor are obtained, and the operating state characteristics are determined based on the reference speed and the actual speed. The operating state features are input into a pre-trained extreme learning machine for parameter mapping to obtain the target control parameters of the sliding surface parameters, the target control parameters of the reaching law parameters, and the target control parameters of the disturbance rejection adjustment parameters. The system performance evaluation index is determined based on the operating state characteristics, and the parameter update rate is determined based on the system performance evaluation index and the state range in which the operating state characteristics are located. Based on the parameter update rate, parameters are updated according to the sliding surface parameters and their target control parameters of the previous control cycle, the reaching law parameters and their target control parameters of the previous control cycle, and the disturbance rejection adjustment parameters and their target control parameters of the previous control cycle, respectively, to obtain the sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters of the current control cycle; wherein, the sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters of the current control cycle constitute the sliding control parameters of the current control cycle; A sliding surface is constructed based on the operating state characteristics and the sliding surface parameters of the current control cycle, and the adaptive disturbance rejection term gain is updated based on the sliding surface and the disturbance rejection adjustment parameters of the current control cycle. An improved approach law is constructed based on the sliding surface, the approach law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain. The quadrature axis reference current is determined based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved approach law. The permanent magnet synchronous motor is subjected to closed-loop control based on the quadrature axis reference current to obtain the updated actual speed.

2. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 1, characterized in that, The step of inputting the operating state features into a pre-trained extreme learning machine for parameter mapping to obtain the target control parameters for the sliding surface parameters, the target control parameters for the reaching law parameters, and the target control parameters for the disturbance rejection adjustment parameters includes: Keeping the network parameters of the pre-trained extreme learning machine unchanged, the running state features are input into the pre-trained extreme learning machine for forward computation to obtain the parameter output; Based on the parameter output, determine the target parameter values ​​for the sliding mode surface integral gain, the exponential reaching term gain, the constant velocity reaching term gain, the boundary layer thickness, and the adaptive term coefficient. The target parameter value of the sliding mode surface integral gain is determined as the target control parameter of the sliding mode surface parameter. The target parameter values ​​of the exponential reaching term gain, the constant velocity reaching term gain, and the boundary layer thickness are determined as the target control parameters of the reaching law parameter. The target parameter value of the adaptive term coefficient is determined as the target control parameter of the disturbance rejection adjustment parameter.

3. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 1, characterized in that, The operating status characteristics include: real-time speed error, real-time speed error change rate, and reference speed change rate; The step of determining system performance evaluation indicators based on the operating state characteristics includes: Extract the real-time speed error, the real-time speed error change rate, and the reference speed change rate from the operating status characteristics; The system performance evaluation index is constructed based on the real-time speed error, the real-time speed error change rate, and the reference speed change rate.

4. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 3, characterized in that, Determining the parameter update rate based on the system performance evaluation indicators and the state range of the operating state characteristics includes: The real-time speed error is compared with the preset dynamic state condition and the preset stable state condition of speed error, and the real-time speed error change rate is compared with the preset dynamic state condition and the preset stable state condition of speed error change rate, so as to obtain the state range determination result. Based on the state range determination result, the update rate gain coefficient of the previous control cycle is increased, decreased, or kept constant to obtain the update rate gain coefficient of the current control cycle. The parameter update rate before limiting is determined based on the update rate gain coefficient of the current control cycle and the system performance evaluation index, and the parameter update rate before limiting is range-constrained to obtain the parameter update rate.

5. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 1, characterized in that, Based on the parameter update rate, parameters are updated according to the sliding surface parameters and their target control parameters of the previous control cycle, the reaching law parameters and their target control parameters of the previous control cycle, and the disturbance rejection adjustment parameters and their target control parameters of the previous control cycle, respectively, to obtain the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters of the current control cycle, respectively. This includes: Determine the difference between the sliding surface parameters of the previous control cycle and its target control parameters, determine the difference between the approach law parameters of the previous control cycle and its target control parameters, and determine the difference between the disturbance rejection adjustment parameters of the previous control cycle and its target control parameters. The sliding surface parameter difference, the reaching law parameter difference, and the disturbance rejection adjustment parameter difference are adjusted according to the parameter update rate to obtain the sliding surface parameter change, the reaching law parameter change, and the disturbance rejection adjustment parameter change. The sliding surface parameters of the previous control cycle are updated according to the change in the sliding surface parameters, the approach law parameters of the previous control cycle are updated according to the change in the approach law parameters, and the disturbance rejection adjustment parameters of the previous control cycle are updated according to the change in the disturbance rejection adjustment parameters. The updated sliding surface parameters, updated reaching law parameters, and updated disturbance rejection adjustment parameters are constrained by parameter range to obtain the sliding control parameters for the current control cycle, which include the sliding surface integral gain, exponential reaching term gain, constant velocity reaching term gain, boundary layer thickness, and adaptive term coefficient.

6. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 5, characterized in that, The operating status characteristics include: real-time speed error, real-time speed error change rate, and real-time speed error integral value; The construction of the sliding surface based on the operating state characteristics and the sliding surface parameters of the current control cycle includes: Extract the real-time speed error, the rate of change of real-time speed error, and the integral value of real-time speed error from the operating status characteristics; The sliding surface is constructed based on the real-time speed error, the rate of change of the real-time speed error, the integral value of the real-time speed error, and the sliding surface integral gain of the current control cycle.

7. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 6, characterized in that, The step of updating the adaptive disturbance rejection term gain based on the sliding surface and the disturbance rejection adjustment parameters of the current control cycle includes: Based on the positional relationship between the sliding surface and the preset sliding surface region, the sliding surface region determination result is determined; wherein, the sliding surface region determination result includes an outside region determination result and an inside region determination result, the outside region determination result indicates that the sliding surface is located outside the preset sliding surface region, and the inside region determination result indicates that the sliding surface is located within the preset sliding surface region; When the sliding surface region determination result is the result outside the region, the change in anti-disturbance gain is determined based on the adaptive term coefficient of the current control cycle and the sliding surface, and the adaptive anti-disturbance term gain of the previous control cycle is updated based on the change in anti-disturbance gain to obtain the updated adaptive anti-disturbance term gain; when the sliding surface region determination result is the result within the region, the adaptive anti-disturbance term gain of the previous control cycle is attenuated to obtain the updated adaptive anti-disturbance term gain.

8. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 7, characterized in that, The step of constructing an improved reaching law based on the sliding surface, the reaching law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain, and determining the quadrature-axis reference current based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved reaching law, includes: The exponential approach term is determined based on the gain of the exponential approach term of the sliding surface and the current control cycle; The continuous saturation function output is determined based on the sliding surface and the boundary layer thickness of the current control cycle, and the constant velocity approach term is determined by combining the gain of the constant velocity approach term of the current control cycle. The adaptive disturbance rejection term is determined based on the sliding surface and the updated adaptive disturbance rejection term gain. The improved approach law is obtained by combining the exponential approach term, the constant velocity approach term, and the adaptive disturbance rejection term. The quadrature-axis reference current is obtained by solving the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved approach law.

9. The improved sliding mode control method for permanent magnet synchronous motors based on extreme learning machine according to claim 5, characterized in that, The updated sliding surface parameters, reaching law parameters, and disturbance rejection adjustment parameters are constrained to obtain sliding mode control parameters for the current control cycle, including sliding surface integral gain, exponential reaching term gain, constant velocity reaching term gain, boundary layer thickness, and adaptive term coefficients. Obtain the upper bound of the sliding surface disturbance and the preset fluctuation limit of the isodynamic approach term; The lower limit of the boundary layer thickness is determined based on the gain of the iso-velocity approach term of the current control cycle, the upper limit of the sliding surface disturbance, and the preset upper limit of the fluctuation. Based on the boundary layer thickness lower limit constraint, the updated boundary layer thickness is constrained, and parameter range constraints are applied to the updated sliding mode surface integral gain, the updated exponential approach term gain, the updated constant velocity approach term gain, and the updated adaptive term coefficient, respectively, to obtain the sliding mode control parameters for the current control cycle.

10. An improved sliding mode control system for permanent magnet synchronous motors based on extreme learning machines, characterized in that, The system includes: The operating status determination module is used to obtain the reference speed and actual speed of the permanent magnet synchronous motor, and determine the operating status characteristics based on the reference speed and the actual speed. The target parameter generation module is used to input the running state features into a pre-trained extreme learning machine for parameter mapping to obtain the target control parameters of the sliding surface parameters, the target control parameters of the reaching law parameters, and the target control parameters of the disturbance rejection adjustment parameters. The update rate determination module is used to determine the system performance evaluation index based on the operating state characteristics, and to determine the parameter update rate based on the system performance evaluation index and the state range of the operating state characteristics. The parameter update module is used to update parameters based on the parameter update rate, according to the sliding surface parameters and their target control parameters of the previous control cycle, the reaching law parameters and their target control parameters of the previous control cycle, and the disturbance rejection adjustment parameters and their target control parameters of the previous control cycle, respectively, to obtain the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters of the current control cycle; wherein, the sliding surface parameters, the reaching law parameters, and the disturbance rejection adjustment parameters of the current control cycle constitute the sliding control parameters of the current control cycle; The sliding mode state processing module is used to construct a sliding surface based on the operating state characteristics and the sliding surface parameters of the current control cycle, and to update the adaptive disturbance rejection term gain based on the sliding surface and the disturbance rejection adjustment parameters of the current control cycle. The control quantity determination module is used to construct an improved approach law based on the sliding surface, the approach law parameters of the current control cycle, and the updated adaptive disturbance rejection term gain, and to determine the quadrature axis reference current based on the mechanical motion relationship of the permanent magnet synchronous motor, the sliding surface, and the improved approach law. The closed-loop control module is used to perform closed-loop control on the permanent magnet synchronous motor based on the quadrature axis reference current to obtain the updated actual speed.

Images