Asynchronous motor thermal robust parameter identification device and method based on working condition anchor optimization

Abstract

The application belongs to the field of asynchronous motor variable frequency speed regulation, parameter identification and intelligent optimization control, discloses an asynchronous motor thermal robust parameter identification device and method based on working condition anchoring optimization, and is suitable for the parameter mismatch working condition caused by rotor resistance thermal drift. The device comprises a data acquisition unit, a working condition characteristic database unit, a characteristic extraction unit, a torque deviation monitoring unit, a trigger judgment unit, an anchoring convergence optimization unit and a parameter correction unit. The rotor resistance reference initial value and the calibrated torque reference value are obtained by matching the current working condition, the limited convergence optimization is triggered when the torque deviation continuously exceeds the threshold value, the rotor resistance identification value is determined, and the rotor time constant, the slip frequency and the flux linkage observer parameters are updated. The application can improve the parameter identification stability and torque control robustness under thermal drift working condition.

Description

Technical Field

[0001] This invention belongs to the field of asynchronous motor variable frequency speed regulation, parameter identification and intelligent optimization control, specifically involving an asynchronous motor thermal robust parameter identification device and method based on working condition anchoring optimization. Background Technology

[0002] Asynchronous motors, with their outstanding advantages of simple structure, low cost, reliable operation, and convenient maintenance, have been widely used in industrial drives, electric vehicles, rail transit, and elevator systems. The high-efficiency and high-precision operation of these motors relies on vector control, and the performance of vector control is limited by the accuracy of motor parameters. Among these, the real-time state of the rotor resistance is crucial to torque accuracy and efficiency. Therefore, real-time and accurate online identification of the rotor resistance is a core prerequisite for ensuring the high-performance operation of asynchronous motors.

[0003] Existing online parameter identification methods for asynchronous motors typically employ filtering, adaptive observation, or global optimization algorithms to estimate rotor resistance online. While these methods can update parameters to some extent, they still have significant shortcomings under thermal transient conditions: Firstly, rotor resistance changes continuously with temperature, making it difficult for fixed-parameter models or conventional filtering methods to track thermal drift in a timely manner, easily leading to flux linkage observation errors and torque output deviations. Secondly, global search methods such as standard particle swarm optimization usually lack prior constraints related to the current operating conditions, requiring the optimization process to be performed within a large search space, which can easily result in invalid searches, decreased convergence speed, and parameter oscillations. Specifically, frequent initiation of global identification not only increases the computational burden on the controller but also makes it susceptible to transient disturbances and measurement noise, leading to abrupt changes in identification results; conversely, failure to initiate parameter correction in a timely manner can leave the control model in a state of mismatch for extended periods. Therefore, how to utilize historical operating condition knowledge to provide reasonable initial values ​​and search boundaries for online identification, and how to initiate local optimization only when a persistent abnormal torque deviation is detected, are key issues for improving the real-time performance, stability, and robustness of parameter identification under thermal drift conditions in asynchronous motors. Summary of the Invention

[0004] To address the problem that the rotor resistance of an asynchronous motor drifts with temperature changes during operation, leading to mismatch in control model parameters, increased flux linkage observation errors, and deviation of electromagnetic torque output from the target value, this invention provides a device and method for identifying thermally robust parameters of an asynchronous motor based on operating condition anchoring optimization.

[0005] To achieve the above objectives, the present invention provides an asynchronous motor thermal robust parameter identification device based on working condition anchoring optimization, comprising a data acquisition unit, a working condition feature database unit, a feature extraction unit, a torque deviation monitoring unit, a trigger judgment unit, an anchoring convergence optimization unit, and a parameter correction unit.

[0006] The data acquisition unit is used to acquire the operating status parameters of the asynchronous motor, which include at least stator voltage, stator current and rotor mechanical angular velocity;

[0007] The operating condition feature database unit is used to store the mapping relationship between operating condition features and reference parameter set, the reference parameter set including at least the initial reference value of rotor resistance and the reference value of calibration torque;

[0008] The feature extraction unit is used to generate current operating condition features based on current operating status parameters, and match the current operating condition features with the database sample operating condition features in the operating condition feature database unit to obtain the rotor resistance reference initial value and calibration torque reference value corresponding to the current operating condition.

[0009] The torque deviation monitoring unit is used to calculate the torque deviation based on the real-time torque observation value and the calibrated torque reference value.

[0010] The triggering judgment unit is used to generate an optimized trigger signal when the torque deviation exceeds a preset threshold and the number of continuous cycles reaches a preset value.

[0011] The anchored convergence optimization unit is used to respond to the optimization trigger signal, construct a restricted search space with the rotor resistance reference initial value as the search anchor point, generate candidate rotor resistances in the restricted search space, and update the candidate rotor resistances according to the iterative optimization rules with convergence constraints; the anchored convergence optimization unit is also used to determine the rotor resistance identification value with the smallest composite fitness function value from the candidate rotor resistances according to the composite fitness function containing the torque tracking error term and the rotor resistance deviation from the reference initial value penalty term;

[0012] The parameter correction unit is used to update the rotor time constant, slip frequency, and flux observer parameters in the asynchronous motor control model according to the rotor resistance identification value. The calibrated torque reference value in the reference parameter set is used for torque deviation monitoring and optimized trigger judgment before triggering. The rotor resistance reference initial value is used for the construction of the restricted search space, generation of candidate rotor resistances, and setting of the rotor resistance deviation penalty term in the composite fitness function during triggering. The rotor resistance identification value is used for the correction of rotor time constant, slip frequency, and flux observer parameters after triggering, so that the prior parameter information obtained by the working condition matching is used throughout the entire rotor resistance identification process before, during, and after triggering.

[0013] The current operating condition features are selected from at least two of the following: speed features, current features, voltage features, and load-related features. The load-related features include directly measured load features or load-related features indirectly characterized by at least one of stator current, stator voltage, and rotor mechanical angular velocity. Before calculating the distance between the current operating condition features and the database sample operating condition features, the feature extraction unit normalizes or standardizes the current operating condition features and the database sample operating condition features, and calculates the matching distance between them using a weighted squared Euclidean distance. The reference parameter set corresponding to the database sample with the smallest matching distance is taken as the matching result of the current operating condition. The torque deviation monitoring unit calculates the real-time torque observation value based on the asynchronous motor current model or flux linkage observation model, and calculates the difference or absolute difference between the real-time torque observation value and the calibrated torque reference value as the torque deviation. The trigger judgment unit uses a sliding window method to count the number of consecutive cycles in which the torque deviation exceeds a preset threshold. When the number of consecutive cycles reaches a preset value, an optimized trigger signal is generated.

[0014] The anchored convergence optimization unit, centered on the initial rotor resistance value, determines a restricted search space based on the allowable range of rotor resistance variation, offline calibration error, or preset boundary coefficients. Within this restricted search space, multiple candidate rotor resistance values ​​are generated, each corresponding to the position variable of an optimization individual. These candidate rotor resistance values ​​are randomly or uniformly generated within the restricted search space, with the initial rotor resistance value as the initial center. During iteration, the velocity and position of the optimization individual are updated based on the convergence factor, learning factor, damping coefficient, and fitness evaluation results, and the updated position is restricted to the restricted search space. The anchored convergence optimization unit also updates the individual optimal position and global optimal position of the optimization individual based on the composite fitness function value corresponding to the candidate rotor resistance. This composite fitness function is weighted by a torque tracking error term and a rotor resistance deviation penalty term. The torque tracking error term characterizes the deviation between the electromagnetic torque corresponding to the candidate rotor resistance and the calibrated torque reference value, while the rotor resistance deviation penalty term limits the degree of deviation of the candidate rotor resistance from the initial rotor resistance value.

[0015] The parameter correction unit updates the rotor time constant in the asynchronous motor control model based on the rotor resistance identification value, and updates the slip frequency and flux linkage observer parameters based on the updated rotor time constant. After completing one parameter correction, the trigger judgment unit resumes torque deviation monitoring, and restarts parameter optimization when the torque deviation exceeds the preset threshold again and the number of continuous cycles reaches the preset value.

[0016] This invention also provides a method for identifying thermal robustness parameters of asynchronous motors based on operating condition anchoring optimization, comprising:

[0017] Establish a mapping database between operating condition characteristics and a set of reference parameters, wherein the set of reference parameters includes at least the initial reference value of rotor resistance and the reference value of calibration torque;

[0018] Real-time acquisition of asynchronous motor operating status parameters, and generation of current operating condition characteristics based on the operating status parameters;

[0019] The current operating condition features are matched with the database sample operating condition features in the mapping database to obtain the initial value of rotor resistance and the reference value of calibration torque corresponding to the current operating condition.

[0020] The torque deviation is calculated based on the real-time torque observation value and the calibrated torque reference value.

[0021] When the torque deviation exceeds a preset threshold and the number of continuous cycles reaches a preset value, a limited search space is constructed using the rotor resistance reference initial value as the search anchor point.

[0022] Candidate rotor resistors are generated within the restricted search space, updated according to the iterative optimization rule with convergence constraints, and the rotor resistor identification value is determined according to the composite fitness function weighted by the torque tracking error term and the rotor resistor deviation from the reference initial value penalty term.

[0023] The rotor time constant in the asynchronous motor control model is updated based on the rotor resistance identification value, and the slip frequency and flux linkage observer parameters are updated based on the updated rotor time constant. After parameter correction, torque deviation monitoring is restored. The calibrated torque reference value in the reference parameter set is used for torque deviation monitoring and optimized trigger judgment before triggering. The initial rotor resistance reference value is used for the construction of the restricted search space, generation of candidate rotor resistances, and setting of the rotor resistance deviation penalty term in the composite fitness function during triggering. The rotor resistance identification value is used for the correction of rotor time constant, slip frequency, and flux linkage observer parameters after triggering, so that the prior parameter information obtained by the operating condition matching is used throughout the entire rotor resistance identification process before, during, and after triggering.

[0024] The mapping database covers the rotor resistance thermal drift range using temperature conditions during the offline calibration phase. During the online operation phase, it generates current operating condition characteristics based on at least the stator voltage, stator current, and rotor mechanical angular velocity collected in real time. The initial value of rotor resistance and the benchmark value of calibration torque are obtained by matching the current operating condition characteristics with the operating condition characteristics of the database samples.

[0025] The restricted search space is centered on the initial value of the rotor resistance reference. The upper and lower boundaries of the candidate rotor resistances are determined according to the allowable variation ratio of the rotor resistance, the offline calibration error, or the preset boundary coefficient, and the candidate rotor resistances are restricted between the upper and lower boundaries. The iterative optimization rule with convergence constraints includes: updating the velocity and position of the optimization individual according to the convergence factor determined by the learning factor and the damping coefficient, the individual optimal position and the global optimal position of the optimization individual, and imposing boundary constraints on the updated position; updating the individual optimal position and the global optimal position according to the composite fitness function value corresponding to the candidate rotor resistance.

[0026] After completing one parameter calibration, the torque deviation between the real-time torque observation value and the calibrated torque reference value continues to be monitored. When subsequent changes in operating conditions or temperature cause the torque deviation to exceed the preset threshold again and the number of consecutive cycles reaches the preset value, the current operating condition feature matching, restricted search space construction, anchoring convergence optimization and parameter calibration process is re-executed.

[0027] The aforementioned working-condition-anchored and convergent optimization-coordinated thermal robust parameter identification method can be called Working-condition Anchored Event-triggered Optimization (WA-EO). This method revolves around the process of "offline database construction - online matching - torque deviation triggering - constrained convergent optimization - control model correction," achieving efficient online updating of the asynchronous motor rotor resistance through a coordinated mechanism of working-condition prior constraints, event-triggered optimization, and local convergent identification.

[0028] Through the above technical solution, the present invention can obtain the initial rotor resistance reference value and the calibration torque reference value that match the current operating conditions using the operating condition feature database; construct an event triggering mechanism using the deviation between the real-time torque observation value and the calibration torque reference value, and initiate parameter optimization only when the torque deviation continuously exceeds the threshold; and construct a restricted search space with the initial rotor resistance reference value as the anchor point, and perform anchoring convergence optimization through a composite fitness function that includes a torque tracking error term and a rotor resistance deviation from the initial reference value penalty term, thereby forming a collaborative identification mechanism of "operating condition prior anchoring - torque deviation triggering - restricted convergence optimization - control model correction", which improves the parameter identification stability and torque control robustness of asynchronous motors under thermal drift conditions.

[0029] The beneficial effects of this invention are as follows:

[0030] (1) In the device provided by the present invention, the data acquisition unit, the working condition feature database unit, the feature extraction unit, the torque deviation monitoring unit, the trigger judgment unit, the anchoring convergence optimization unit, and the parameter correction unit cooperate with each other to form an integrated architecture of "offline database construction - online matching - trigger optimization - closed-loop correction". The online identification process of this device does not require an additional temperature sensor as a necessary input. The temperature information can be used for offline working condition calibration or experimental working condition monitoring. During the online operation phase, measurable parameters such as stator voltage, stator current, and rotor mechanical angular velocity can be used to match the current working condition and complete the online identification of rotor resistance. The structure is simple and easy to integrate into existing motor controllers.

[0031] (2) This invention pre-establishes a working condition-parameter mapping database through Working-condition Anchored Feature Extraction (WA-FE) and uses weighted squared Euclidean distance to quickly match the current working condition, which can provide a more accurate initial value of rotor resistance for anchoring convergence optimization. This mechanism can reduce the number of iterations in online optimization, which is beneficial to complete parameter correction within the calculation time allowed by the controller and meets the requirements of real-time control for calculation speed.

[0032] (3) The anchored convergence optimization algorithm used in this invention introduces a convergence factor during the speed update process and restricts the candidate rotor resistance to a limited search space centered on the initial value of the rotor resistance reference. This method can maintain the necessary search capability while enabling the optimization individuals to gradually converge to the neighborhood of the reference parameter corresponding to the current operating condition, avoiding invalid iterations, premature convergence, or parameter oscillations caused by an excessively large global search range, thereby improving the stability of rotor resistance identification under thermal transient conditions.

[0033] (4) The composite fitness function constructed in this invention includes both a torque tracking error term and a penalty term for rotor resistance deviation from the initial reference value. The former reflects the deviation of the control objective, while the latter constrains the identification result within the neighborhood of the initial reference value of rotor resistance obtained from the database matching. This multi-objective optimization strategy helps to suppress parameter jumps while ensuring torque accuracy, making the identification result smoother and more reliable.

[0034] (5) Compared with existing global online identification methods, this invention utilizes operating condition anchoring to obtain parameter priors that match the current operating state, and ensures that these parameter priors are integrated throughout the torque deviation triggering, constrained search space construction, candidate rotor resistance generation, composite fitness evaluation, and parameter correction processes. On the one hand, the event-triggered approach avoids invalid identification processes and reduces the computational burden on the system; on the other hand, by continuously anchoring the online optimization process using the initial rotor resistance reference value, online correction can be performed on rotor time constant mismatch, slip frequency calculation deviation, and flux linkage observation error caused by rotor resistance thermal drift. This suppresses the error accumulation process from parameter thermal drift to electromagnetic torque output, thereby improving the torque control stability and operational robustness of the asynchronous motor under thermal transient conditions.

[0035] (6) In the following specific implementation, the running results show that under the thermal condition of 100℃, the WA-EO algorithm of the present invention has a better parameter correction effect than the comparison method that only uses the operating condition anchoring feature extraction. Specifically, the steady-state phase current amplitude is reduced, the peak value of the stator current vector amplitude obtained by synthesizing the d-axis stator current and the q-axis stator current is reduced, and the torque deviation can be suppressed within the preset threshold. Attached Figure Description

[0036] For ease of explanation, the present invention will be further described in detail with reference to the following specific embodiments and accompanying drawings:

[0037] Figure 1 Overall structural block diagram of the device of the present invention;

[0038] Figure 2 Overall flowchart of the algorithm of this invention;

[0039] Figure 3 Flowchart of the iterative optimization process of the anchored convergence optimization algorithm described in this invention;

[0040] Figure 4 This is a schematic diagram of the torque deviation response waveform of the method of the present invention;

[0041] Figure 5 This is a comparison of the fitness convergence curves of the method of this invention, the optimization method without working condition anchoring constraints, and the standard particle swarm optimization algorithm.

[0042] Figure 6These are comparison diagrams of phase current waveforms under different operating conditions using the WA-EO algorithm described in this invention and the WA-FE algorithm using only the operating condition anchoring feature extraction algorithm; where (a) WA-FE - 500 r / min, 10 N·m; (b) WA-EO - 500 r / min, 10 N·m; (c) WA-FE - 500 r / min, 20 N·m; (d) WA-EO - 500 r / min, 20 N·m; (e) WA-FE - 1000 r / min, 10 N·m; (f) WA-EO - 1000 r / min, 10 N·m; (g) WA-FE - 1000 r / min, 20 N·m; (h) WA-EO - 1000 r / min, 20 N·m.

[0043] Figure 7 The diagrams show the comparison of the rotational speed feedback, the composite load current, and the phase current waveforms obtained by combining the d-axis stator current and the q-axis stator current under dynamic speed command tracking using the WA-FE and WA-EO algorithms of this invention; where (a) is the waveform of the WA-FE algorithm under sinusoidal speed command tracking (10 N·m); (b) is the waveform of the WA-EO algorithm under sinusoidal speed command tracking (10 N·m); (c) is the waveform of the WA-FE algorithm under triangular speed command tracking (10 N·m); and (d) is the waveform of the WA-EO algorithm under triangular speed command tracking (10 N·m). Detailed Implementation

[0044] This invention primarily focuses on identifying rotor resistance. Stator resistance can be processed using nominal values, offline calibration values, or conventional compensation methods; its influence can also be ignored when the stator resistance variation is small. Within a single optimization window, motor parameters other than rotor resistance are processed using nominal values, offline calibration values, or conventional compensation values. The invention will be further described in detail below with reference to the accompanying drawings, providing a clear and complete description of the apparatus and methods in the embodiments of the invention. The advantages and features of the invention will become clearer from the following description.

[0045] I. Device Structure

[0046] like Figure 1 As shown, the asynchronous motor thermal robust parameter identification device based on operating condition anchoring optimization provided by the present invention includes: a data acquisition unit, an operating condition feature database unit, a feature extraction unit, a torque deviation monitoring unit, a trigger judgment unit, an anchoring convergence optimization unit, and a parameter correction unit. Each unit sequentially completes the following processes: operating data acquisition, operating condition matching, torque deviation triggering, anchoring convergence optimization, and parameter correction, and re-enters the torque deviation monitoring process after parameter correction.

[0047] Data acquisition unit: Real-time acquisition of stator voltage of asynchronous motor through voltage sensor, current sensor and encoder. , Stator current , and rotor mechanical angular velocity The data is then output to the feature extraction unit and the torque deviation monitoring unit. When the speed acquired by the data acquisition unit is the rotor mechanical angular velocity, the controller converts the rotor mechanical angular velocity into the rotor electrical angular velocity according to the number of motor pole pairs, and uses the rotor electrical angular velocity for the calculation of slip frequency and synchronous angular velocity.

[0048] Operating Condition Characteristic Database Unit: Stores a pre-built offline operating condition-parameter mapping database. This database contains multiple operating condition characteristic vectors and corresponding reference parameter sets (rotor resistance reference initial value). Calibration torque reference value wait).

[0049] Feature extraction unit: This unit extracts prior parameters from the operating condition feature database unit based on the current operating conditions and uses these prior parameters to constrain the subsequent online optimization process. The feature extraction unit receives the current operating status parameters output by the data acquisition unit, generates current operating condition features, reads sample operating condition features from the operating condition feature database unit, calculates the weighted squared Euclidean distance between the current operating condition features and the sample operating condition features in the database, selects the database sample with the smallest distance as the matching sample, and outputs the initial reference value of the rotor resistance corresponding to the matching sample. and the calibration torque reference value .

[0050] Torque deviation monitoring unit: based on real-time torque observation values Compared with the rated torque reference value Calculate torque deviation The torque deviation is then sent to the trigger judgment unit. The real-time torque observation value can be calculated from the current flux linkage observation value and the q-axis stator current according to the electromagnetic torque relationship shown in equation (6) below. Since rotor resistance thermal drift will cause the rotor time constant in the control model to be inconsistent with the actual motor state, the torque deviation can be used as the basis for judging whether the rotor resistance needs to be re-identified.

[0051] Triggering judgment unit: used to judge the torque deviation. Does it exceed the preset threshold? The system counts the number of consecutive cycles in which the torque deviation exceeds a preset threshold. When the number of consecutive cycles reaches a preset value... At that time, the trigger judgment unit outputs an optimization trigger signal to the anchoring convergence optimization unit. For example, the preset threshold can be determined offline based on the rated torque, sensor noise level, and flux linkage observation error; the preset value can be taken as 5 control cycles.

[0052] Anchored Convergence Optimization Unit: The restricted search space is constructed centered on the initial rotor resistance reference value obtained from database matching. Its boundaries are determined based on the allowable variation range of rotor resistance, offline calibration error, or preset boundary coefficients. For example, the initial rotor resistance reference value can be recorded as the search center, and the allowable variation ratio or calibration error range can be used as the search radius to obtain the upper and lower boundaries of the candidate rotor resistances. The initial optimization group is generated within the restricted search space, limiting the optimization process to the initial rotor resistance reference value corresponding to the current operating condition. The optimization is performed within the neighborhood. Upon receiving the optimization trigger signal, the anchored convergent optimization unit uses the rotor resistance reference initial value as the initial center and iteratively solves for the rotor resistance identification value that minimizes the composite fitness function value. Specifically, this includes:

[0053] The parameter initialization module is used to set the group size. Maximum number of iterations Learning factors , and damping coefficient ;

[0054] The convergence factor calculation module is used to calculate the convergence factor based on the learning factor and the damping coefficient. ;

[0055] The iterative update module is used to update the velocity and position of the optimized individual and apply boundary constraints.

[0056] The fitness calculation module is used to calculate the composite fitness function value corresponding to each candidate rotor resistance.

[0057] The convergence determination module is used to determine the convergence threshold when the maximum number of iterations is reached. Or, the change in fitness during consecutive iterations is less than a preset convergence threshold. At that time, the rotor resistance identification value with the smallest composite fitness function value is output. .

[0058] Compared to searching for rotor resistance globally, constructing a restricted search space centered on the initial rotor resistance reference value obtained from database matching can reduce the invalid search range and lower the risk of identification results jumping due to noise or transient disturbances.

[0059] Parameter correction unit: This unit reads the rotor resistance identification value. Substitute into the vector control system and update the rotor time constant of the flux linkage observer. The slip frequency is then recalculated according to equation (17) described later. After the update is completed, the system re-enters the real-time torque deviation monitoring state. When subsequent changes in operating conditions or continued temperature changes cause the torque deviation to meet the triggering conditions again, the operating condition matching, constrained optimization, and parameter correction processes are re-executed, thereby forming an online closed-loop identification. The flux linkage observer parameters include at least the rotor time constant used for rotor flux linkage estimation, and if necessary, also slip frequency or synchronous angular velocity related parameters recalculated based on the updated rotor time constant.

[0060] In this embodiment, the aforementioned functional units can be implemented by a processor, a memory, and a control program stored therein in the motor vector controller. The processor can be a DSP, MCU, FPGA, or a combination of the above processors. Data acquisition, operating condition matching, trigger judgment, anchoring convergence optimization, and parameter correction can be performed on the existing motor controller hardware platform, and no additional temperature sensor or dedicated parameter identification circuit is required during the online phase.

[0061] II. Method and Flow

[0062] This invention also provides a method for identifying thermal robust parameters of asynchronous motors based on operating condition anchoring optimization. Through a collaborative process of offline database construction, online matching, trigger optimization, and closed-loop correction, it achieves real-time, high-precision identification of rotor resistance. The overall algorithm flow of this invention is as follows: Figure 2 As shown, the iterative process of the anchored convergence optimization algorithm is as follows: Figure 3 As shown.

[0063] like Figure 2 As shown, the method of this invention includes an offline database construction section and an online identification section. The offline section collects operating data of the asynchronous motor under different speeds, loads, and temperatures, constructing a mapping database between operating condition characteristics and a set of reference parameters. The online section collects the current operating status parameters of the asynchronous motor in real time, obtains the initial reference value of rotor resistance and the reference value of calibrated torque corresponding to the current operating condition through operating condition feature matching, and calculates the torque deviation based on the real-time torque observation value and the reference value of calibrated torque. When the torque deviation exceeds a preset threshold and the number of consecutive cycles reaches a preset value, anchoring convergence optimization is initiated to determine the rotor resistance identification value within a limited search space, and updates the rotor time constant, slip frequency, and flux linkage observer parameters accordingly; when the triggering condition is not met, the current rotor resistance parameters remain unchanged.

[0064] like Figure 3As shown, the anchored convergence optimization algorithm is executed after the optimization trigger signal is generated. First, the optimization group is initialized with the rotor resistance reference initial value obtained from the working condition matching as the initial center, so that the candidate rotor resistances are distributed within a reasonable neighborhood corresponding to the current working condition. Then, during the iteration process, the velocity and position of the optimization individuals are updated, and the composite fitness function value corresponding to each candidate rotor resistance is calculated. Then, the individual optimal solution and the global optimal solution are updated according to the composite fitness function value. When the maximum number of iterations is reached, or the fitness change is less than the preset convergence threshold, the iteration stops, and the rotor resistance identification value with the smallest composite fitness function value is output.

[0065] Step 1: Model and Influence Mechanism Explanation. A mathematical model of the synchronous rotating dq-axis of the asynchronous motor is established, clarifying the influence mechanism of rotor resistance thermal drift. Asynchronous motor vector control is based on the synchronous rotating dq-axis model, describing its electromagnetic characteristics. In this coordinate system, the stator d and q-axis voltage equations are:

[0066] (1);

[0067] in For differential operators, , This refers to the synchronous electric angular velocity. Since the stator and rotor are analyzed separately, all stator parameters are distinguished by adding "s" to their subscripts, and rotor parameters are distinguished by adding "r". The magnetic flux linkage corresponding to the d-axis of the stator. The flux linkage corresponding to the q-axis of the stator. Let be the stator current along the d-axis. Let be the stator current along the q-axis. The rotor current is the d-axis current. Let be the rotor current along the q-axis. Stator self-sensing, Rotor self-inductance, mutual inductance between stator and rotor Rotor resistance, Stator resistance. The stator flux linkage equation is:

[0068] (2);

[0069] The rotor flux linkage is a vector with two components in the synchronously rotating dq coordinate system. The magnitude of the rotor flux linkage at this point... The formula for electromagnetic torque is:

[0070] (3);

[0071] (4);

[0072] in , Let be the flux linkages along the d and q axes of the rotor. The dynamic equation for the rotor flux linkage is:

[0073] (5);

[0074] in For slip angular velocity, The rotor's mechanical angular velocity, Let be the time constant. Under the rotor field-oriented control strategy, let be... At this time, the rotor flux Determined only by the d-axis component, the equations for electromagnetic torque and rotor flux linkage can be simplified to:

[0075] (6);

[0076] in The number of pole pairs is given. Furthermore, the rotor resistance exhibits a significant temperature dependence, with its value changing approximately linearly with winding temperature. The thermal drift model can be expressed as:

[0077] (7);

[0078] in, For the motor at room temperature The nominal value of the rotor resistance is below. The rotor resistance temperature coefficient, This represents the current equivalent temperature of the winding. In actual operation, temperature is difficult to measure directly online, but the drift in rotor resistance will affect control performance through the following paths: First, the rotor time constant... Changes occur, causing the flux linkage observer to detect the flux linkage along the rotor's d-axis. The estimation is biased. Secondly, the slip frequency... Inaccurate calculations cause the magnetic field orientation angle to deviate from the true direction, ultimately resulting in an increased deviation between the electromagnetic torque output and the commanded value.

[0079] From equation (6), we can see that The dynamic response is affected by the rotor time constant Impact. When the actual motor temperature rises, it increases the rotor resistance. Increase the size, while the controller still uses the original one. When the value is true, the rotor flux estimated by the flux observer will deviate from the actual rotor flux.

[0080] Rotor resistance Thermal drift affects the dynamic response of rotor flux linkage and the accuracy of magnetic field orientation, altering the coupling relationship between rotor current and flux linkage. This directly impacts flux linkage observation accuracy and electromagnetic torque output, leading to a decrease in vector control performance. Therefore, it is necessary to address this issue. Online identification is performed. During online identification, an additional temperature sensor is not required as a necessary input. Instead, real-time measurable electrical quantities are used to form real-time feedback. Combined with the relatively accurate initial reference value provided by database matching in subsequent steps and the online fine-tuning of anchored convergence optimization, the rotor resistance is corrected in real time. This actively compensates for model mismatch caused by thermal drift from the control level, thereby achieving the purpose of thermally robust identification.

[0081] Step Two: Construct a working condition-parameter mapping database and obtain the baseline parameter set for each working condition offline (the offline part of WA-FE). First, collect the operating data of the asynchronous motor under different speeds, loads, and temperatures. This refers to the d-axis stator voltage; This is the q-axis stator voltage; This refers to the d-axis stator current. This refers to the q-axis stator current. This represents the rotor's mechanical angular velocity. Each database entry is represented as:

[0082] (8);

[0083] When establishing a condition-parameter mapping database, features that can reflect the operating and thermal states of asynchronous motors can be selected as input items for database samples.

[0084] During the offline calibration phase, the operating condition characteristics may include rotational speed, load, stator voltage, stator current, winding temperature, or indirect thermal state characteristics calculated from voltage, current, and rotational speed. The winding temperature is used to cover the rotor resistance thermal drift range during the offline calibration phase and is not a necessary input for the online identification process. For each offline calibration operating point, the corresponding initial rotor resistance reference value and calibration torque reference value are recorded, and these two values ​​are used as the reference parameter set for that operating point.

[0085] During online operation, the feature extraction unit generates current operating condition features based on real-time collected stator voltage, stator current, and rotor mechanical angular velocity, and matches them with the operating condition features of the database samples using a normalized or standardized weighted squared Euclidean distance. To facilitate rapid online matching, the database is divided into a subset of operating condition features and a subset of reference parameters.

[0086] The operating condition feature subset is used to characterize the current operating state, preferably using online measurable parameters such as voltage, current, and speed; the reference parameter subset is used to store the initial reference value of rotor resistance and the reference value of calibrated torque under the corresponding operating condition. The operating condition-parameter mapping database is reusable among asynchronous motors of the same model. If there are batch differences in motor parameters, compensation can be made online through subsequent torque deviation triggering mechanisms combined with the anchoring convergence optimization module in WA-EO.

[0087] The annotation rules for database samples are as follows:

[0088] (9);

[0089] in, For the first A subset of operating condition features from a database sample is used for online matching of the current operating condition. Matching only requires quantities that reflect the operating point, namely current, voltage, and speed. These quantities are measurable in real time during motor operation and do not depend on parameter calibration. For the first Each database sample corresponds to a set of reference parameters, which stores the reference parameters for that operating condition, including stator resistance, rotor resistance, inductance, mutual inductance, and reference values ​​for rated torque. Although torque can be measured in real time, in the database it represents a "calibrated torque baseline value," belonging to the target performance index for this operating condition, rather than a "fingerprint" used to identify the operating condition. Therefore, it is categorized into the baseline parameter subset. , The first d-axis and q-axis stator currents at various operating points , The first d-axis and q-axis stator voltages at various operating points For the first Rotor mechanical angular velocity at each operating point; , , and The first The stator resistance, rotor resistance, stator inductance, stator-rotor mutual inductance, and rated torque reference values ​​corresponding to each operating point.

[0090] For each operating condition feature vector With torque error and current tracking error To optimize the objective, among which, , The d-axis and q-axis current command values ​​are respectively used. An offline optimization algorithm (such as a genetic algorithm or multi-starting-point local search) is employed to find the parameter combination that minimizes torque error and current tracking error under stable operating conditions. This parameter combination is then written into the baseline parameter subset of the corresponding database entry. After offline calibration, each set of operating condition feature vectors... Its corresponding set of reference parameters This constitutes a database sample, and multiple database samples together form a mapping relationship between operating condition characteristics and the benchmark parameter set; the database is established in total. A sample of samples is provided to cover the typical operating range of the motor, such as zero speed to rated speed, zero load to rated load, and temperature range of 20 to 120°C.

[0091] During the offline calibration phase of the motor, the motor is mounted on a test bench, and different load torques are applied to the motor through a load test. Different speed commands are given through the frequency converter. The temperature of the motor windings is controlled by a heating device. It should be noted that the temperature variable is only used to cover the rotor resistance thermal drift range during the offline calibration stage and is not a necessary input for the online identification process. During online operation, only real-time measurable operating parameters such as stator voltage, stator current, and rotor mechanical angular velocity need to be collected to complete operating condition matching, trigger optimization, and parameter correction. Within the allowable operating range of the motor, the following discrete operating points are selected:

[0092] (1) From the lowest stable speed to the rated speed, take a point at regular speed intervals (e.g., 100 rpm), for a total of One point.

[0093] (2) From no-load to rated load, take a point at certain torque intervals (e.g., 10% of rated torque), for a total of One point.

[0094] (3) Take a point at regular intervals (e.g., 10℃) from room temperature to the maximum allowable temperature of the motor (e.g., 120℃). Points. Total number of samples. For example, take , , ,but Each operating condition point.

[0095] At each stable operating point (speed, load, and temperature are maintained in a steady state for at least 1 second), the following electrical quantities are collected:

[0096] (1) The three-phase current is sampled by a current sensor and obtained by Clarke / Park transformation. , .

[0097] (2) The bus voltage and inverter switching state are reconstructed by sampling the voltage sensor to obtain the signal. , .

[0098] (3) Obtain the rotor mechanical angular velocity through an encoder or tachometer. .

[0099] When calibrating using a physical experimental platform, high-precision offline measuring instruments (such as LCR bridges, DC resistance testers, etc.) can be used to measure the stator resistance and inductance parameters at that temperature, and the rotor resistance calibration value can be obtained through offline parameter identification methods. Simultaneously, the actual output torque under the corresponding operating conditions is measured using a torque sensor and used as the calibration torque reference value. When using high-precision simulation, the initial rotor resistance reference value and calibration torque reference value under different speed, load, and temperature conditions can also be generated based on the calibrated asynchronous motor model. The database construction method of this invention is not limited to a physical experimental platform.

[0100] Step 3: Based on the operating condition anchoring feature extraction algorithm, quickly obtain the initial reference value of rotor resistance and the reference value of calibration torque (online part of WA-FE). It should be noted that the temperature variable is used to cover the rotor resistance thermal drift range during the offline calibration stage, and it is not required that the winding temperature be measured in real time during the online operation stage; during the online operation stage, real-time measurable operating state parameters such as stator voltage, stator current, and rotor mechanical angular velocity can be matched with the operating conditions in the database sample to obtain a reference parameter set that is close to the current thermal state.

[0101] Real-time acquisition of current operating status data of asynchronous motors and the working condition characteristics of the database samples A comparison is then made. The similarity between the current operating condition and the operating condition in the database sample can be characterized by the distance between their feature vectors in the five-dimensional feature space. The smaller the distance, the closer the current operating condition is to the corresponding database sample operating condition. Considering the differences in the dimensions, value ranges, and noise levels of different sensors or physical quantities, the features of each operating condition can be normalized or standardized, and matching weights can be set according to the reliability of the features. This is to reduce the impact of noisy features on the matching results.

[0102] During normal motor operation, the following operations are performed in each sampling cycle:

[0103] (1) Sample the three-phase stator current of the asynchronous motor using a current sensor , , ,in , , Let a, b, and c represent the stator currents respectively; the Clarke transform is used to obtain the current components in the two-phase stationary coordinate system. , Then, based on the current rotor magnetic field orientation angle, a Park transformation is performed to obtain the synchronous rotating coordinate system. , Simultaneously, voltage reconstruction is obtained. , Obtained through the encoder .

[0104] (2) Construct a vector from the above five quantities, and label it as follows:

[0105] (10);

[0106] (3) Since the database only stores a limited number of discrete operating conditions, the motor may be in any intermediate state between adjacent discrete operating conditions during actual operation. Therefore, the initial value of rotor resistance and the reference value of calibrated torque obtained by matching mainly reflect the approximate parameters under the nearest offline sample operating condition, and are not necessarily accurately applicable to the current continuously changing temperature and load conditions. In particular, the rotor resistance changes continuously with temperature, and the database is difficult to cover all intermediate temperature points. Even if the matched operating condition is similar to the current operating condition, parameter deviations may still occur due to differences in thermal state, which in turn leads to a continuous deviation between the real-time torque observation value and the reference value of calibrated torque. Therefore, the subsequent steps will determine whether to start the anchoring convergence optimization module in WA-EO based on the torque deviation. After triggering, the anchoring convergence optimization module uses the initial value of rotor resistance as the search anchor point, performs a local search in its neighborhood, and fine-tunes the rotor resistance through a composite fitness function that includes a torque tracking error term and a rotor resistance deviation penalty term to obtain a rotor resistance identification value that can reduce torque deviation and does not deviate from the reasonable range of the current operating condition.

[0107] For each sample in the database, calculate the weighted squared Euclidean distance. Furthermore, before calculating the weighted squared Euclidean distance, it can be normalized or standardized according to the dimensions and value range of each working condition to avoid adverse effects of differences in the dimensions of different physical quantities on the matching results. , , , , ; For the first in the database The corresponding features of each sample. Weight coefficients. This value can be set empirically, with a default of 0.2, to adjust the weighting of different operating conditions' characteristics on the matching results. If some sensors have high noise, their weights can be reduced. However, it must be maintained... .

[0108] (4) Find out Minimum index According to the index Retrieve the corresponding initial value of rotor resistance from the database. Compared with the rated torque reference value .

[0109] (11);

[0110] (5) Obtain the reference parameter set that best matches the current operating condition from the database based on the minimum distance index. The reference parameter set includes the initial reference value of rotor resistance. and the calibration torque reference value The initial value of the rotor resistance reference. Used to determine the initial distribution center of the subsequent online optimization group and to construct the constrained search space; the calibration torque reference value This is used to calculate the torque deviation from real-time torque observations during the online phase. In this invention, the initial rotor resistance reference value obtained through database matching is not merely used as a general initial value for the optimization algorithm, but is simultaneously used for the construction of the constrained search space, the generation of candidate rotor resistances, and the setting of the rotor resistance deviation penalty term in the composite fitness function. Therefore, the database matching results can directly participate in the online optimization process, reducing invalid searches and suppressing unreasonable jumps in candidate rotor resistances.

[0111] Step 4: Perform the WA-EO anchoring convergence optimization process

[0112] Because the online fitness function involves a nonlinear torque observation model, flux recalculation, and measurement noise, the relationship between candidate rotor resistance and torque deviation is nonlinear. Directly using gradient information for optimization is easily affected by noise and model errors, causing fluctuations in the identification results. This invention adopts an anchored convergence optimization algorithm, utilizing a group parallel search mechanism in conjunction with a convergence factor, which helps improve the anti-interference and convergence stability of the online identification process. Furthermore, since the optimization variable is a single variable (rotor resistance), the computational load is low. Centering on the initial rotor resistance reference value obtained in step three, a preset neighborhood range is set according to the allowable variation range of rotor resistance, offline calibration error, or pre-stored fixed boundaries. This neighborhood range serves as the restricted search space for candidate rotor resistances. The position of each optimization individual is randomly or uniformly generated within this restricted search space, and the initial velocity is generated within a preset velocity range. During real-time operation, the system continuously calculates the torque deviation in each control cycle according to the calibration torque reference value obtained in step three. and using a length of Sliding window records recent Torque deviation within a control cycle. When the torque deviation within the sliding window exceeds a preset threshold... The number of consecutive cycles reaches the preset value At that time, that is continued In each control cycle, if the rotor resistance parameter mismatch persists, an optimization trigger signal is generated to initiate anchoring convergence optimization; if If the current torque deviation does not exceed the trigger threshold, the current rotor resistance parameter will remain unchanged, and optimization will not be initiated; if there is a partial cycle However, the number of consecutive cycles did not reach the target. If so, continue monitoring the sliding window.

[0113] Once the triggering condition is met, the following subroutine will be executed:

[0114] Initialize the population of individuals for optimization, and set the population size. (For example, 20 can be selected, which can be adjusted according to the processor's computing power, ranging from 10 to 50):

[0115] (1) The position of each optimization individual The distribution is random and uniform within the interval, but the total population must not exceed [a certain value]. initial velocity exist Uniformly distributed within the interior. Upper speed limit. It can be obtained based on the limited search space width, the maximum number of iterations, or empirical calibration.

[0116] (2) Initialize the individual optimal of each individual to its current position, and initialize the global optimal to the position with the minimum fitness among all individuals (the fitness is calculated according to the formula in step five).

[0117] Iterative updates:

[0118] (1) For each individual being searched, generate two random numbers in the interval [0,1]. and .

[0119] (2) Calculate the convergence factor based on the learning factor. (A typical value of 0.8 can also be used) to control the decay of the speed of the individual being optimized, to ensure the convergence of the algorithm, and to avoid speed divergence or premature convergence.

[0120] (12);

[0121] The convergence factor is calculated jointly by the learning factor and the damping coefficient. It is used to adjust the inertia term, individual cognitive term, and social learning term during the optimization process, enabling the optimization individual to gradually converge towards the individual optimal position and the global optimal position while maintaining its search capability. The learning factor... , Can be preferred to satisfy The specific value can be determined through thermal transient sensitivity analysis or experimental tuning.

[0122] When constructing the anchoring convergence optimization framework, the rotor resistance to be identified is used as the optimization variable. Algorithm parameters and iteration rules are set, and the anchoring convergence optimization module in WA-EO is initialized. Among these, the first... The positional variable of each optimization individual is: The velocity variable is Damping coefficient This is used to balance recognition speed and stability, and is usually taken as 0.6 to 0.9.

[0123] (3) Perform optimization of individual velocity and position updates. The iterative update formula for individual velocity and position is:

[0124] (13);

[0125] (14);

[0126] During the velocity update process, the convergence factor simultaneously acts on the velocity term, individual cognition term, and social learning term from the previous time step to suppress velocity divergence and improve convergence stability. At the same time, boundary constraints are applied to the position and velocity of the optimizing individual, keeping the candidate rotor resistance within a confined search space and a preset velocity range. For the first The speed of the next optimal individual at the next moment; For the current optimal individual position; The velocity at the previous moment; The inertial component maintains its original tendency to move. For individual cognitive items, it guides individuals seeking optimization to move closer to their historical best position. As a social learning activity, it guides individuals seeking excellence to move closer to the optimal position within the group. , A random number in the interval [0,1]. For the first The first individual seeking optimization The optimal position of an individual in the next iteration, i.e., the optimal solution for that individual. For the first The global optimal position in the next iteration is the global optimal solution.

[0127] (4) Apply boundary constraints to the position and velocity of the individual being optimized:

[0128] (15);

[0129] The optimal individual position constraint is The speed constraint is This is to keep the candidate rotor resistance within a reasonable range of variation and to avoid divergence in the optimization speed. Among these... , This represents the reasonable range of variation for the rotor resistance. This represents the upper limit of the velocity. Under the aforementioned boundary constraints, since the initial positions of each optimization individual are randomly or uniformly distributed within the restricted search space, a velocity update formula is used to guide the optimization individuals to gradually approach their historical optimal positions and the global optimal positions. Simultaneously, a convergence factor stabilizes this approach process and suppresses oscillations. Ultimately, each optimization individual converges to a region near the same location, which corresponds to the rotor resistance identification value that minimizes or reduces the composite fitness function.

[0130] (5) Calculate the fitness of the new position using the fitness function described in step five. If the new fitness is less than... Then update and the corresponding position.

[0131] (6) After all the optimal individuals have been updated, find the optimal individual with the minimum fitness. If it is smaller than the current fitness, then... Then update .

[0132] Convergence criterion: If the number of iterations... (For example, (Approximately 20 times), or (For example If the value is taken as one ten-thousandth, then the iteration terminates and the output is given. .

[0133] It should be noted that the anchored convergence optimization described in this invention is not a simple parameter adjustment of the standard particle swarm optimization algorithm. Instead, it constructs a constrained search space centered on the rotor resistance baseline initial value obtained from the anchored features under operating conditions, and limits the search range of individual optimizations through convergence factors and boundary constraints. To further ensure that the optimization results simultaneously meet the requirements of torque tracking accuracy and parameter stability, subsequent steps construct a composite fitness function that includes a torque tracking error term and a rotor resistance deviation penalty term from the baseline initial value to evaluate and screen candidate rotor resistances.

[0134] Step 5: Construct a composite fitness function and calculate the fitness to identify rotor resistance under thermal drift conditions. After optimization is triggered, a composite fitness function is constructed, weighted by a torque tracking error term and a rotor resistance deviation penalty term, to evaluate each candidate rotor resistance. The recognition effect is shown in formula (16):

[0135] (16);

[0136] in, The electromagnetic torque is calculated based on the current rotor resistance. The initial value of the rotor resistance reference obtained from step three is denoted as . , These are weighting coefficients. , ,and Its specific value is determined based on the motor's rated parameters or pre-calibrated through offline testing, or it can be set to a fixed empirical value (typically, a value of...). , ).

[0137] The composite fitness function includes a torque tracking error term and a rotor resistance deviation penalty term from the initial reference value. The first term... This is the torque tracking error term, used to characterize the deviation between the electromagnetic torque corresponding to the candidate rotor resistance and the calibrated torque reference value; the smaller this term, the closer the electromagnetic torque corresponding to the candidate rotor resistance is to the calibrated torque reference value. In specific calculations, the currently sampled... , , With the operating parameters constant, based on the candidate rotor resistance Calculate the rotor time constant The rotor d-axis flux linkage was recalculated using a current model flux linkage observer. Then, substituting into the electromagnetic torque formula, we get... This allows us to obtain the absolute value of the torque deviation.

[0138] Second item This is a penalty term for rotor resistance deviation from the initial reference value. It limits the deviation of candidate rotor resistances from the initial reference value obtained from the database matching, ensuring that the candidate rotor resistances converge within the neighborhood of the initial reference value and preventing resistance jumps due to noise or transient disturbances. It also includes a parameter proximity term in the real-time calculation of the fitness function. At that time, the absolute difference between the current optimal individual position and the initial rotor resistance reference value obtained in step three is directly calculated. The weighting coefficients are adjusted according to the operating conditions. , In the initial stage of a temperature step change, the weighting coefficient of the torque tracking error term can be temporarily increased. This suppresses torque deviation; once the system reaches steady state, it can be restored to its original value. When calculating the composite fitness function, the torque tracking error term and the rotor resistance deviation penalty term can be normalized, or error terms of different dimensions can be converted using weighting coefficients to ensure the comparability of the composite fitness function evaluation results.

[0139] Step 6: Iteratively optimize the rotor resistance identification value that minimizes the composite fitness function value to complete the online parameter correction.

[0140] After the iteration, we get The controller applies this value in the following manner:

[0141] (1) Update flux linkage observer parameters: Modify rotor time constant .

[0142] (2) Update slip frequency:

[0143] (17);

[0144] in, The slip angular velocity is used to synthesize the synchronous electric angular velocity. According to the formula Coordinate transformation is performed to achieve accurate rotor magnetic field orientation. In each PWM cycle ( In ), according to the current , , Recalculate the slip angular velocity according to equation (17). And use it for coordinate transformation, where, Indicates the first One PWM cycle.

[0145] (3) Reset Monitoring: After completing one parameter calibration, the system clears or updates the torque deviation counter and sliding window record, and resumes real-time torque deviation monitoring. If the temperature continues to rise or fall, causing the torque deviation to exceed the preset threshold again, the system will reset the monitoring function. And the number of consecutive cycles reaches the preset value. (For example, if the torque deviation exceeds the preset threshold and continues for 5 control cycles), the operating condition matching, anchoring convergence optimization and parameter correction process will be re-executed to achieve continuous online identification and control correction of rotor resistance, thereby improving the system's thermal robustness and control accuracy.

[0146] Experimental verification:

[0147] The following describes the implementation and technical effects of the WA-EO algorithm described in this invention, based on experimental verification. This embodiment is based on a 5kW asynchronous motor experimental platform. Under the same controller hardware, sampling period, operating conditions, and motor temperature rise, the comparison method using only WA-FE and the WA-EO algorithm of this invention were run respectively. Except for whether the anchoring convergence optimization module in WA-EO was enabled for online fine-tuning, all other control parameters remained consistent. The experimental platform consists of the motor under test, the load motor, the load host computer, the load power supply, the main power supply, the controller, and a multi-channel oscilloscope. The main parameters of the experimental motor and control system are shown in Table 1.

[0148] Table 1 Main parameters of the tested asynchronous motor

[0149] Rated power (kW) 5 Rated voltage (V) 60 Rated speed (rpm) 3000 Rated current (A) 98 Rated torque (N·m) 15.8 Extreme logarithm 2

[0150] The parameters listed in Table 1 are only used to illustrate the initialization method of the experimental platform and control system in this embodiment, and do not constitute a limitation on the applicable motor type, power level, rated speed, voltage level, or load type of this invention. The experiments cover steady-state performance comparison, dynamic speed command response testing, and evaluation of torque deviation suppression capability under thermal transients. In the steady-state performance comparison and dynamic speed command response testing, the motor windings are heated to 100°C and maintained at a constant temperature to simulate thermal conditions. In the evaluation of torque deviation suppression capability under thermal transients, the motor windings are heated from room temperature to 100°C to simulate the temperature change process. This embodiment uses the 5kW asynchronous motor shown in Table 1 as the test object. Although the rated torque of this motor is 15.8 N·m, in order to examine the thermal robustness parameter identification capability of the method of this invention under higher load disturbances, two load torque conditions of 10 N·m and 20 N·m were set in the experiment, with the 20 N·m condition serving as a short-time loading verification condition.

[0151] It should be noted that the WA-FE method can be considered a comparative method that only uses operating condition anchoring feature extraction and database matching, and it does not include the anchoring convergence optimization process after triggering. The WA-EO method of this invention, based on the initial rotor resistance reference value and calibration torque reference value obtained by WA-FE, further introduces torque deviation triggering, a constrained search space, a convergence factor, and a composite fitness function to perform local fine-tuning of the rotor resistance. Therefore, the comparison between the WA-FE method and the WA-EO method can reflect the contribution of the trigger-based anchoring convergence optimization module to the thermal drift compensation effect.

[0152] (1) Comparison of steady-state phase currents:

[0153] The motor windings were heated to 100℃ and maintained at a constant temperature. Four typical operating conditions were established, with speeds of 500 r / min and 1000 r / min, and load torques of 10 N·m and 20 N·m. Under each condition, the comparative method using only the Condition Anchoring Feature Extraction (WA-FE) algorithm and the WA-EO algorithm of this invention were run, and the phase current waveforms corresponding to the WA-FE method and the WA-EO method of this invention were recorded. The results are as follows: Figure 6 As shown.

[0154] Experimental results show that, under the same speed and load conditions, when using only the WA-FE method, the initial value of rotor resistance obtained by database matching is difficult to fully compensate for the continuous thermal drift under 100℃ thermal conditions, and the motor requires a large phase current to maintain torque output. After adopting the WA-EO method of this invention, the anchoring convergence optimization can locally fine-tune the rotor resistance in the neighborhood of the initial value of rotor resistance reference, so that the phase current amplitude is significantly reduced.

[0155] Specifically, under operating conditions of 500 r / min and 10 N·m, the steady-state phase current amplitude corresponding to the WA-FE method is approximately 113 A, while the WA-EO method of this invention reduces it to approximately 102 A; under operating conditions of 500 r / min and 20 N·m, the steady-state phase current amplitude corresponding to the WA-FE method is approximately 175 A, while the WA-EO method of this invention reduces it to approximately 155 A; under operating conditions of 1000 r / min and 10 N·m, the steady-state phase current amplitude corresponding to the WA-FE method is approximately 117 A, while the WA-EO method of this invention reduces it to approximately 105 A; under operating conditions of 1000 r / min and 20 N·m, the steady-state phase current amplitude corresponding to the WA-FE method is approximately 177 A, while the WA-EO method of this invention reduces it to approximately 155 A.

[0156] The above results demonstrate that the WA-EO method of this invention can reduce the phase current amplitude under thermal conditions under different speeds and loads. Furthermore, the magnified waveforms show that the phase current waveforms corresponding to the WA-EO method of this invention are more stable, with reduced distortion and spikes. This indicates that by locally fine-tuning the rotor resistance through a limited search space, convergence factor, and composite fitness function, the discretization error caused by relying solely on the operating condition database matching can be compensated, thereby improving the torque output stability and current regulation robustness of the asynchronous motor under thermal drift conditions.

[0157] (2) Response results of dynamic speed command:

[0158] Furthermore, to verify the control response performance of the method of the present invention under dynamic speed commands, under a constant load of 10 N·m, sinusoidal speed commands (-500 r / min to 500 r / min, period 2 s) and triangular speed commands (-500 r / min to 500 r / min, period 4 s) were applied respectively to examine the algorithm's ability to track variable speed commands. Figure 7 As shown, during the test, speed feedback and composite load current were simultaneously collected. The phase current waveform and sinusoidal speed tracking results are as follows: Figure 7 As shown in (a)(b), the triangular rotational speed tracking results are as follows: Figure 7 As shown in (c) and (d). It should be noted that the composite load current in this embodiment refers to the magnitude of the stator current vector obtained by combining the d-axis stator current and the q-axis stator current, which is used to characterize the motor current demand during dynamic tracking.

[0159] Sinusoidal speed tracking: During sinusoidal speed command tracking, the maximum combined load current corresponding to the WA-FE method is approximately 265A, while the maximum combined load current corresponding to the WA-EO method of this invention is approximately 240A. Simultaneously, the phase current waveform corresponding to the WA-EO method of this invention is more stable during acceleration and deceleration, and the current fluctuation amplitude in the locally magnified waveform is reduced, indicating that the rotor resistance parameters after anchoring convergence optimization can improve the current regulation performance during dynamic speed tracking.

[0160] Triangular speed tracking: During triangular speed command tracking, the speed reference exhibits significant slope changes and inflection points, placing higher demands on the controller's dynamic response capabilities. Figure 7 (c) and (d) show that the maximum combined load current corresponding to the WA-FE method is approximately 280A, while the maximum combined load current corresponding to the WA-EO method of this invention is approximately 255A. In the speed command transition region, the phase current waveform and combined load current waveform of the WA-EO method of this invention are more stable, and no obvious divergence or continuous oscillation is observed.

[0161] The above results demonstrate that, under dynamic speed command tracking conditions, the WA-EO method of this invention, compared to the WA-FE method alone, can reduce the peak value of the synthesized load current, decrease current fluctuations, and improve the dynamic current regulation stability under thermal drift conditions. Since the experimental load torque remains consistent, the lower synthesized load current requirement indicates that the method of this invention can improve current utilization efficiency while maintaining torque output capability, thereby enhancing the dynamic response performance and operational robustness of the asynchronous motor vector control system under thermal conditions.

[0162] (3) Torque deviation suppression capability:

[0163] To simulate rotor resistance drift caused by sudden temperature changes in the motor during actual operation, the motor windings were heated from room temperature to 100℃ while maintaining a constant speed of 500 r / min and a load of 10 N·m. Torque observations based on the current model were calculated in real time and compared with the calibrated torque reference value. The torque deviation response results are as follows: Figure 4 As shown. By Figure 4 It can be seen that during the process of rotor resistance change caused by temperature rise, the WA-EO algorithm of the present invention can correct according to the torque deviation trigger parameter, so that the torque deviation is suppressed and stabilized within the preset threshold range after short-term fluctuation, indicating that the method of the present invention can reduce the torque observation deviation caused by thermal drift.

[0164] Furthermore, under the same temperature rise process and the same initial population size, the fitness convergence of the WA-EO algorithm of this invention, the optimization method without operating condition anchoring constraints, and the standard particle swarm optimization algorithm were compared. The results are as follows: Figure 5As shown. The optimization method without operating condition anchoring constraints refers to a comparative optimization method that does not use the rotor resistance baseline initial value obtained from operating condition matching to construct a restricted search space, and does not set a penalty term for rotor resistance deviation from the baseline initial value in the fitness function. From Figure 5 It can be seen that the standard particle swarm optimization algorithm and the optimization method without working condition anchoring constraints fluctuate greatly in the early stage of convergence, and there are still sudden increases or oscillations in fitness values ​​within the local amplification interval. The fitness function value of the WA-EO algorithm of this invention decreases faster and maintains a low fluctuation level after convergence, indicating that the limited search space, the candidate rotor resistance generation constraint, and the rotor resistance deviation from the baseline initial value penalty term can jointly suppress invalid search and parameter jumps.

[0165] The results of this embodiment are as follows: Under the conditions of this embodiment, the three sets of results show that the WA-EO algorithm of this invention is improved in terms of steady-state accuracy, dynamic response, and robustness. Compared with the comparison method that only uses the working condition anchoring feature extraction (WA-FE), the algorithm of this invention reduces the phase current amplitude in steady state, reduces current fluctuation during dynamic tracking, decreases the peak value of the synthesized load current, and can suppress and stabilize the torque deviation within 0.5 N·m.

[0166] The experimental results above indicate that the reduction in phase current amplitude and the maximum value of the combined load current is not solely due to the matching of the operating condition database. Rather, it is because the initial value of the rotor resistance reference simultaneously participates in the construction of the constrained search space, the generation of candidate rotor resistances, and the setting of the penalty term in the composite fitness function. This ensures that the online optimization results are continuously constrained within the reasonable rotor resistance neighborhood corresponding to the current operating condition, thereby reducing rotor time constant mismatch, slip frequency deviation, and flux linkage observation error caused by thermal drift, ultimately improving current regulation stability and torque output stability. Therefore, the asynchronous motor thermal robust parameter identification method based on operating condition anchoring optimization proposed in this invention does not rely solely on offline operating condition matching results. Instead, when torque deviation remains abnormal, it uses anchoring convergence optimization to locally fine-tune the rotor resistance, thereby suppressing the control performance degradation caused by temperature drift and improving the stability of parameter identification, control robustness, and operating efficiency of the asynchronous motor vector control system under thermal drift conditions.

[0167] Through the above specific embodiments, those skilled in the art can implement the thermal robust parameter identification method proposed in this invention in various asynchronous motor drive systems. The online identification process of this invention does not require an additional temperature sensor as a necessary input; temperature information can be used for offline operating condition calibration or experimental operating condition monitoring. The operating condition matching results in the online phase are used to provide an approximate initial reference value for the rotor resistance, and it is not required to directly and accurately determine the current temperature state solely through operating condition characteristics. When changes in the thermal state cause the real-time torque observation value to continuously deviate from the calibrated torque reference value, the rotor resistance is then locally fine-tuned by anchoring convergence optimization triggered by the torque deviation. In the online operation phase, real-time measurable parameters such as stator voltage, stator current, and rotor mechanical angular velocity can be used to complete the current operating condition matching and online identification of the rotor resistance. Experimental results show that, under the conditions of this embodiment, the method of this invention can reduce the phase current amplitude and the peak value of the synthesized load current, and stabilize the torque deviation within a preset threshold.

[0168] In summary, the asynchronous motor thermal robustness parameter identification device and method based on operating condition anchoring optimization proposed in this invention can suppress rotor resistance drift caused by temperature and improve the thermal robustness and operating efficiency of the asynchronous motor vector control system. The basic principles and main features of this invention have been shown and described above. Those skilled in the art should understand that the above embodiments are only for illustrating this invention and are not intended to limit the scope of protection of this invention; various changes and modifications can be made by those skilled in the art without departing from the spirit and substance of this invention, and all such changes and modifications should fall within the scope of protection of this invention.

Claims

1. A device for identifying thermal robust parameters of an asynchronous motor based on operating condition anchoring optimization, characterized in that, It includes a data acquisition unit, a working condition feature database unit, a feature extraction unit, a torque deviation monitoring unit, a trigger judgment unit, an anchoring convergence optimization unit, and a parameter correction unit; The data acquisition unit is used to acquire the operating status parameters of the asynchronous motor, which include at least stator voltage, stator current and rotor mechanical angular velocity; The operating condition feature database unit is used to store the mapping relationship between operating condition features and reference parameter set, the reference parameter set including at least the initial reference value of rotor resistance and the reference value of calibration torque; The feature extraction unit is used to generate current operating condition features based on current operating status parameters, and match the current operating condition features with the database sample operating condition features in the operating condition feature database unit to obtain the rotor resistance reference initial value and calibration torque reference value corresponding to the current operating condition. The torque deviation monitoring unit is used to calculate the torque deviation based on the real-time torque observation value and the calibrated torque reference value. The triggering judgment unit is used to generate an optimized trigger signal when the torque deviation exceeds a preset threshold and the number of continuous cycles reaches a preset value. The anchored convergence optimization unit is used to respond to the optimization trigger signal, construct a restricted search space with the rotor resistance reference initial value as the search anchor point, generate candidate rotor resistances in the restricted search space, and update the candidate rotor resistances according to the iterative optimization rules with convergence constraints; the anchored convergence optimization unit is also used to determine the rotor resistance identification value with the smallest composite fitness function value from the candidate rotor resistances according to the composite fitness function containing the torque tracking error term and the rotor resistance deviation from the reference initial value penalty term; The parameter correction unit is used to update the rotor time constant, slip frequency, and flux observer parameters in the asynchronous motor control model according to the rotor resistance identification value. The calibrated torque reference value in the reference parameter set is used for torque deviation monitoring and optimized trigger judgment before triggering. The rotor resistance reference initial value is used for the construction of the restricted search space, generation of candidate rotor resistances, and setting of the rotor resistance deviation penalty term in the composite fitness function during triggering. The rotor resistance identification value is used for the correction of rotor time constant, slip frequency, and flux observer parameters after triggering, so that the prior parameter information obtained by the working condition matching is used throughout the entire rotor resistance identification process before, during, and after triggering.

2. The asynchronous motor thermal robustness parameter identification device based on operating condition anchoring optimization according to claim 1, characterized in that, The current operating condition characteristics are selected from at least two of the following: speed characteristics, current characteristics, voltage characteristics, and load-related characteristics. The load-related characteristics include directly measured load characteristics or load-related characteristics indirectly characterized by at least one of stator current, stator voltage, and rotor mechanical angular velocity. Before calculating the distance between the current operating condition features and the database sample operating condition features, the feature extraction unit normalizes or standardizes the current operating condition features and the database sample operating condition features, and uses a weighted squared Euclidean distance to calculate the matching distance between them. The reference parameter set corresponding to the database sample with the smallest matching distance is taken as the matching result of the current operating condition. The torque deviation monitoring unit calculates the real-time torque observation value according to the asynchronous motor current model or flux linkage observation model, and calculates the difference or absolute difference between the real-time torque observation value and the calibrated torque reference value as the torque deviation. The trigger judgment unit uses a sliding window method to count the number of consecutive cycles in which the torque deviation exceeds a preset threshold. When the number of consecutive cycles reaches the preset value, an optimized trigger signal is generated.

3. The asynchronous motor thermal robustness parameter identification device based on operating condition anchoring optimization according to claim 1, characterized in that, The anchoring convergence optimization unit takes the rotor resistance reference initial value as the center and determines the limited search space according to the allowable variation range of rotor resistance, offline calibration error or preset boundary coefficient. Multiple candidate rotor resistance values ​​are generated within the restricted search space. Each candidate rotor resistance value corresponds to the position variable of an optimization individual. The multiple candidate rotor resistance values ​​are randomly or uniformly generated within the restricted search space with the rotor resistance reference initial value as the initial center. During the iteration process, the velocity and position of the optimization individual are updated based on the convergence factor, learning factor, damping coefficient, and fitness evaluation results of the optimization individual, and the updated position is restricted within the confined search space. The anchored convergence optimization unit also updates the individual optimal position and global optimal position of the optimization individual based on the composite fitness function value corresponding to the candidate rotor resistance. The composite fitness function is composed of a weighted sum of a torque tracking error term and a rotor resistance deviation from the initial reference value penalty term. The torque tracking error term is used to characterize the deviation between the electromagnetic torque corresponding to the candidate rotor resistance and the calibrated torque reference value, and the rotor resistance deviation from the initial reference value penalty term is used to limit the degree of deviation of the candidate rotor resistance from the initial rotor resistance reference value.

4. The asynchronous motor thermal robustness parameter identification device based on operating condition anchoring optimization according to claim 1, characterized in that, The parameter correction unit updates the rotor time constant in the asynchronous motor control model based on the rotor resistance identification value, and updates the slip frequency and flux linkage observer parameters based on the updated rotor time constant. After completing one parameter correction, the trigger judgment unit resumes torque deviation monitoring, and restarts parameter optimization when the torque deviation exceeds the preset threshold again and the number of continuous cycles reaches the preset value.

5. A method for identifying thermal robust parameters of an asynchronous motor based on operating condition anchoring optimization, characterized in that, include: Establish a mapping database between operating condition characteristics and a set of reference parameters, wherein the set of reference parameters includes at least the initial reference value of rotor resistance and the reference value of calibration torque; Real-time acquisition of asynchronous motor operating status parameters, and generation of current operating condition characteristics based on the operating status parameters; The current operating condition features are matched with the database sample operating condition features in the mapping database to obtain the initial value of rotor resistance and the reference value of calibration torque corresponding to the current operating condition. The torque deviation is calculated based on the real-time torque observation value and the calibrated torque reference value. When the torque deviation exceeds a preset threshold and the number of continuous cycles reaches a preset value, a limited search space is constructed using the rotor resistance reference initial value as the search anchor point. Candidate rotor resistors are generated within the restricted search space, updated according to the iterative optimization rule with convergence constraints, and the rotor resistor identification value is determined according to the composite fitness function weighted by the torque tracking error term and the rotor resistor deviation from the reference initial value penalty term. The rotor time constant in the asynchronous motor control model is updated based on the rotor resistance identification value, and the slip frequency and flux linkage observer parameters are updated based on the updated rotor time constant. After parameter correction, torque deviation monitoring is restored. The calibrated torque reference value in the reference parameter set is used for torque deviation monitoring and optimized trigger judgment before triggering. The initial rotor resistance reference value is used for the construction of the restricted search space, generation of candidate rotor resistances, and setting of the rotor resistance deviation penalty term in the composite fitness function during triggering. The rotor resistance identification value is used for the correction of rotor time constant, slip frequency, and flux linkage observer parameters after triggering, so that the prior parameter information obtained by the operating condition matching is used throughout the entire rotor resistance identification process before, during, and after triggering.

6. The method for identifying thermal robust parameters of asynchronous motors based on operating condition anchoring optimization according to claim 5, characterized in that, The mapping database covers the rotor resistance thermal drift range using temperature conditions during the offline calibration phase. During the online operation phase, it generates current operating condition characteristics based on at least the stator voltage, stator current, and rotor mechanical angular velocity collected in real time. The initial value of rotor resistance and the benchmark value of calibration torque are obtained by matching the current operating condition characteristics with the operating condition characteristics of the database samples.

7. The method for identifying thermal robust parameters of asynchronous motors based on operating condition anchoring optimization according to claim 5, characterized in that, The restricted search space is centered on the initial value of the rotor resistance reference. The upper and lower boundaries of the candidate rotor resistance are determined according to the allowable change ratio of the rotor resistance, the offline calibration error, or the preset boundary coefficient, and the candidate rotor resistance is restricted between the upper and lower boundaries. The iterative optimization rule with convergence constraints includes: updating the velocity and position of the optimization individual based on the convergence factor determined by the learning factor and damping coefficient, the individual optimal position and the global optimal position of the optimization individual, and imposing boundary constraints on the updated position; updating the individual optimal position and the global optimal position based on the composite fitness function value corresponding to the candidate rotor resistance.

8. The method for identifying thermal robust parameters of asynchronous motors based on operating condition anchoring optimization according to claim 5, characterized in that, After completing one parameter calibration, the torque deviation between the real-time torque observation value and the calibrated torque reference value continues to be monitored. When subsequent changes in operating conditions or temperature cause the torque deviation to exceed the preset threshold again and the number of consecutive cycles reaches the preset value, the current operating condition feature matching, restricted search space construction, anchoring convergence optimization and parameter calibration process is re-executed.

Images