Switched reluctance motor online self-correction fourier model prediction control method and system

By correcting the phase resistance and residuals of the Fourier model online, a highly adaptable candidate voltage vector selection method is generated, which solves the model mismatch problem of switched reluctance motors and improves control accuracy and safety.

CN123074333APending Publication Date: 2026-11-10SHANDONG UNIV OF TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611558334.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-09
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing Fourier models for switched reluctance motors are ill-suited to adapting to model mismatch during operation, making it difficult to effectively distinguish between phase resistance changes and flux linkage model parameter changes, unreliable residuals can easily lead to erroneous model parameter updates, and model mismatch makes it difficult to directly constrain the selection of candidate voltage vectors.

Method used

By collecting the operating state parameters of the switched reluctance motor, calculating the residual between the predicted response and the measured response, compensating for the phase resistance using winding temperature and correcting the Fourier coefficients online, generating a set of candidate voltage vectors, and constructing an evaluation function based on torque error and model credibility to select the target voltage vector.

Benefits of technology

It improves the Fourier model's adaptability to operating conditions, reduces the inconsistency between predicted and measured phase currents and torque deviation, and enhances the reliability of candidate voltage vector selection and control safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123074333A_ABST
    Figure CN123074333A_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of motor control, and relates to a switching reluctance motor online self-correction Fourier model predictive control method and system. The method comprises the following steps: calculating a predicted response in a current sampling period; obtaining a measured response, calculating a Fourier model residual error, compensating a phase resistance by using winding temperature, online correcting Fourier coefficients by using the Fourier model residual error still existing after the phase resistance compensation, and obtaining an online corrected Fourier model; determining a current sector, generating a motor side candidate voltage vector set of the current sector, predicting phase currents and torques corresponding to each motor side candidate voltage vector at a next sampling time; determining a torque reference value according to a speed, constructing an evaluation function; selecting a motor side candidate voltage vector with the minimum evaluation function value as a motor side target voltage vector, and outputting the motor side target voltage vector to a switching reluctance motor power converter. The application can improve the matching degree between candidate voltage vector selection and an actual running state.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the fields of motor control, power electronics and embedded real-time control technology, and specifically relates to an online self-calibrating Fourier model predictive control method and system for switched reluctance motors. Background Technology

[0002] In existing technologies, model predictive control of switched reluctance motors typically employs offline calibrated Fourier analytical models. These models predict phase current and torque under different candidate voltage vectors based on operating conditions such as phase current, rotor position angle, speed, and DC bus voltage. Control actions are then selected from a limited number of candidate voltage vectors to balance model prediction accuracy and real-time computational efficiency. However, the electromagnetic characteristics of switched reluctance motors change with operating conditions such as winding temperature, load, and magnetic saturation. A fixed offline model cannot consistently adapt to actual operating conditions.

[0003] Existing predictive control methods for switched reluctance motors based on Fourier analytical models typically employ offline parameter calibration. However, under the influence of factors such as winding temperature rise, changes in device on-state voltage drop, migration of the magnetic saturation operating region, and sensor bias, offline calibration models may experience model mismatch. This can lead to inconsistencies between predicted and measured phase currents, predicted torque deviations, and commutation control mismatch, potentially further causing tailing of the output phase current and an increase in negative torque. Therefore, it is necessary to identify model mismatches based on measured responses during operation and to perform constrained online correction of the Fourier model parameters.

[0004] In actual online calibration, the phase resistance change caused by winding temperature variations will also be reflected in the residual between the predicted and measured responses. If this residual is directly used to correct the Fourier coefficients, the error caused by the phase resistance change may be mistakenly attributed to the change in flux linkage model parameters. At the same time, abnormal residuals caused by noise, weak excitation, and sensor anomalies may also lead to unreliable parameter updates. In addition, model mismatch will further affect the prediction of phase current and torque corresponding to each candidate voltage vector. If the selection of candidate voltage vectors does not consider the degree of model mismatch, it may easily lead to a mismatch between the control action and the actual operating state.

[0005] Therefore, existing technologies often suffer from problems such as the inability of the Fourier model of switched reluctance motors to adapt to model mismatch during operation, difficulty in effectively distinguishing between changes in phase resistance and changes in flux linkage model parameters, unreliable residuals easily causing erroneous updates of model parameters, and difficulty in directly constraining the selection of candidate voltage vectors due to model mismatch. These are the shortcomings of existing technologies.

[0006] In view of this, it is very necessary to provide an online self-calibrating Fourier model predictive control method and system for switched reluctance motors. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies, such as the inability of the Fourier model of a switched reluctance motor to adapt to model mismatch during operation, the difficulty in effectively distinguishing between changes in phase resistance and changes in flux linkage model parameters, the tendency of unreliable residuals to cause erroneous updates of model parameters, and the difficulty in directly constraining the selection of candidate voltage vectors due to model mismatch. This invention provides an online self-calibrating Fourier model predictive control method and system for switched reluctance motors to solve the aforementioned technical problems.

[0008] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, this application provides an online self-calibrating Fourier model predictive control method for a switched reluctance motor, comprising: The operating status parameters of the switched reluctance motor are collected, including phase current, rotor position angle, speed, DC bus voltage, winding temperature and power converter switching status. Based on the current Fourier model and the running state variables, calculate the predicted response for the current sampling period; After obtaining the measured response corresponding to the predicted response, the Fourier model residual is calculated, the phase resistance is compensated using the winding temperature, the identification error is obtained by using the Fourier model residual that still exists after the phase resistance compensation, and the Fourier coefficients of the current Fourier model are corrected online to obtain the online corrected Fourier model. The current sector is determined based on the rotor position angle, turn-on angle, turn-off angle and commutation angle. A set of candidate voltage vectors for each motor side in the current sector is generated. The phase current and torque corresponding to each candidate voltage vector at the next sampling time are predicted using an online modified Fourier model. The torque reference value is determined based on the rotational speed. For each candidate voltage vector on the motor side, an evaluation function is constructed based on the torque error, Fourier model residuals or model credibility obtained from Fourier model residuals, and constraint terms. The candidate voltage vector on the motor side with the smallest evaluation function value is selected as the target voltage vector on the motor side, and the control signal corresponding to the target voltage vector on the motor side is output to the switched reluctance motor power converter.

[0009] By adopting the above technical solution, Fourier model mismatch is identified based on the measured response during operation, and the online corrected Fourier model is used to participate in the subsequent prediction of phase current and torque. This allows the Fourier model to adapt to the actual operating state as the operating state changes, thereby reducing the impact of model mismatch on the accuracy of predictive control. At the same time, by introducing the Fourier model residual or model credibility into the evaluation of candidate voltage vectors on the motor side, the degree of model mismatch can be directly reflected in the control decision, improving the matching degree between the candidate voltage vector selection and the actual operating state.

[0010] Specifically, based on the current Fourier model and the calculation of the operating state variables, the predicted response can be obtained by utilizing the analytical calculation characteristics of the Fourier analytical model to obtain the predicted results of phase current and torque. While maintaining the real-time calculation advantage of the Fourier analytical model, it provides a predictive response benchmark for subsequent Fourier model residual calculation.

[0011] By correlating the measured response with the predicted response and calculating the Fourier model residuals, model mismatch information reflecting the difference between the current Fourier model and the actual operating state can be obtained. This allows changes in the operating state of the Fourier model to be characterized through the Fourier model residuals, providing a basis for identification in subsequent online corrections.

[0012] By using winding temperature to compensate for phase resistance and using the Fourier model residuals that still exist after phase resistance compensation to obtain the identification error, the influence of phase resistance changes caused by winding temperature on the Fourier coefficient correction results can be reduced, so that the changes in Fourier coefficients can better reflect the changes in flux linkage model parameters, thereby improving the accuracy and stability of online correction results.

[0013] By obtaining an online corrected Fourier model through online correction of the Fourier coefficients of the current Fourier model, the influence of model mismatch caused by winding temperature rise, device conduction voltage drop changes, magnetic saturation operating region migration, and sensor deviation can be reduced. This allows the model prediction results to be adjusted according to changes in the operation process, thereby reducing the inconsistency between the predicted phase current and the measured phase current, as well as the deviation of the predicted torque.

[0014] By determining the current sector based on the rotor position angle, turn-on angle, turn-off angle, and commutation angle, and generating a set of candidate voltage vectors for each motor side, a targeted candidate control range can be formed within the current operating area. This reduces the impact of an excessively large candidate control range on real-time calculations while maintaining a small number of candidate voltage vectors on the motor side, and makes the candidate voltage vectors adapt to the phase conduction, commutation, and demagnetization states.

[0015] By using an online modified Fourier model to predict the phase current and torque corresponding to the candidate voltage vectors on each motor side, the prediction results of the candidate voltage vectors can be obtained based on the Fourier model under the current operating conditions, reducing the impact of the mismatch of the fixed Fourier model on the prediction results of the candidate voltage vectors, and providing a basis for the effective evaluation of the candidate voltage vectors.

[0016] By constructing an evaluation function based on torque error, Fourier model residuals or model reliability, and constraint terms, the selection of candidate voltage vectors can simultaneously consider torque tracking requirements, the current mismatch degree of the Fourier model, and operational constraints. This avoids selecting candidate voltage vectors solely based on torque error and improves the reliability of control decisions under model mismatch conditions.

[0017] By employing switching action constraints, current peak constraints, and negative torque constraints, it is possible to limit the switching state changes caused by candidate voltage vectors, predict phase current over-limits, and negative torque, thereby reducing the risk of phase current tailing and negative torque increase during commutation and improving the control safety of switched reluctance motors during operation.

[0018] Preferably, the predicted response for the current sampling period is calculated, including: Input the phase current and rotor position angle into the current Fourier model, and calculate the phase flux linkage according to the finite-order cosine Fourier series in the direction of rotor position angle and the functional expression in the direction of phase current. The phase electromagnetic torque is calculated based on the phase flux linkage, and the predicted phase current and predicted torque for the current sampling period are calculated by combining the rotational speed, DC bus voltage, winding temperature and power converter switching state, so as to obtain the predicted response for the current sampling period.

[0019] By adopting the above technical solution, the nonlinear electromagnetic characteristics of the switched reluctance motor can be characterized by phase current and rotor position angle, so that the predicted response can simultaneously reflect the influence of the motor operating state and the switching state of the power converter, thereby improving the ability of the predicted response to characterize the actual operating state.

[0020] As a preferred method, the Fourier model residuals are calculated, including: The measured phase current at the current sampling time is time-aligned with the predicted phase current at the previous sampling time, and the current prediction residual is calculated. The estimated electromagnetic torque at the current sampling time is determined based on the measured phase current, rotor position angle, and pre-calibrated electromagnetic characteristics, and is time-aligned with the predicted torque at the previous sampling time, and the torque prediction residual is calculated. The current prediction residual and the torque prediction residual are normalized or combined to obtain the Fourier model residual.

[0021] By employing the above technical solution, the Fourier model residuals are constructed using the measured and predicted responses corresponding to the time. This ensures that the measured and predicted responses are time-consistent, reducing the impact of time misalignment on the Fourier model residuals. It can also identify the mismatch of the offline Fourier model after temperature rise, parameter drift, and operating condition migration, thereby improving the ability of the Fourier model residuals to reflect changes in operating conditions.

[0022] As a preferred option, the identification error includes: Calculate the temperature-compensated phase resistance based on the winding temperature and the phase resistance at the reference temperature; Substitute the temperature-compensated phase resistance into the phase voltage balance equation, and recalculate the predicted phase current in combination with the phase current, rotor position angle and DC bus voltage. Then, calculate the current prediction residual based on the recalculated predicted phase current and the measured phase current. The current prediction residual is mapped to the Fourier model residual. Based on the Fourier model residual that still exists after phase resistance compensation, the identification error for the flux linkage model parameter correction is determined, and the identification error used for online correction of Fourier coefficients is obtained.

[0023] By adopting the above technical solution, the phase resistance is first compensated by the winding temperature, and then the flux linkage model is corrected by the residual. This can distinguish the effect of the phase resistance change caused by the winding temperature change from the Fourier model parameter change, which helps to avoid the resistance change being mistakenly absorbed into the Fourier coefficients, and makes the online correction result more accurately reflect the flux linkage model parameter change.

[0024] As a preferred option, the online modified Fourier model is obtained, including: The regression vector is constructed based on the phase current power term and the rotor position angle harmonic term, and the Fourier coefficients to be corrected in the current Fourier model are arranged into a parameter vector. Calculate the recursive gain based on the regression vector, identification error, forgetting factor, and covariance matrix; The parameter vector and covariance matrix are updated using recursive gain, and the updated parameter vector is restored to the Fourier coefficient matrix to obtain the online modified Fourier model.

[0025] By adopting the above technical solution, the Fourier coefficients can be continuously adjusted according to the identification error obtained during operation, limiting the impact of abnormal samples on the update of Fourier coefficients, enabling the Fourier model to have parameter adaptation ability during operation, while using the forgetting factor to maintain historical effective information, so that online correction takes into account both parameter change response capability and correction stability.

[0026] Preferably, the parameter vector is updated, including: The update amount limit is determined based on the single update amount of the parameter vector, and the cumulative offset envelope is determined based on the cumulative offset of the parameter vector relative to the most recent reliable Fourier coefficient matrix. The Fourier model residuals are compared with the update threshold and the protection threshold respectively to obtain the first residual interval, the second residual interval and the third residual interval; In the first residual interval, the Fourier coefficients are kept unchanged. In the second residual interval, the parameter vector is updated using the recursive least squares method with a forgetting factor based on the update amount limit and the cumulative offset envelope. In the third residual interval, the Fourier coefficients are frozen and the most recent reliable Fourier coefficient matrix is ​​called to obtain the constrained Fourier coefficient update result, thus completing the parameter vector update.

[0027] By adopting the above technical solution, it is possible to distinguish between normal fluctuations, correctable mismatches and abnormal mismatches based on the different degrees of Fourier model residuals, avoid unnecessary parameter changes caused by small residuals, and limit the range of single parameter updates and cumulative offsets; in the case of abnormal mismatches, the most recent reliable Fourier coefficient matrix is ​​maintained, thereby reducing the risk of model parameter distortion caused by noise and abnormal residuals.

[0028] Preferably, the frozen Fourier coefficients include: Fourier coefficient updates are stopped and Fourier coefficients are frozen under any of the following conditions: phase current is in the zero current neighborhood, speed is below the reliable identification threshold, winding temperature is abnormal, DC bus is under-voltage or over-voltage, power converter switching state is abnormal, parameter correction exceeds the allowable upper limit, covariance matrix is ​​non-positive definite or numerically divergent, and regression vector does not meet the sufficient excitation conditions.

[0029] By adopting the above technical solution, the Fourier coefficients can be limited from continuing to change when reliable parameter correction conditions are not available. This reduces the possibility of unreliable parameter updates under zero-current neighborhood, low-speed, abnormal operating conditions, and weak excitation conditions, thereby improving the stability of the online correction process.

[0030] Preferably, the updated parameter vector is restored to a Fourier coefficient matrix, including: The freeze state is lifted after the sensor status recovers, the DC bus voltage recovers to the allowable range, the phase current enters the effective identification range, the Fourier model residual is lower than the recovery threshold for a continuous predetermined period, the parameter update meets the limit, and the power converter exits the protection or degradation state. Call the most recent trusted Fourier coefficient matrix and restore the parameter update, and restore the updated parameter vector to the Fourier coefficient matrix.

[0031] By adopting the above technical solution, it is possible to avoid resuming Fourier coefficient updates before the abnormal state has been eliminated, so that the online correction has controlled recovery conditions, and the most recent credible Fourier coefficient matrix is ​​used as the recovery basis, thereby reducing the impact of repeated abnormal states on the Fourier model.

[0032] Preferably, a set of candidate voltage vectors for each motor side of the current sector is generated, including: The current sector is determined based on the relationship between the rotor position angle and the turn-on angle, turn-off angle, and commutation angle. Based on the current phase conduction, commutation, and demagnetization states of the current sector, and combined with the switching states of the power converter, the torque increase vector, torque decrease vector, leading phase excitation vector, outgoing phase freewheeling vector, and outgoing phase demagnetization vector are generated to obtain the set of candidate voltage vectors for each motor side of the current sector.

[0033] By adopting the above technical solution, the candidate voltage vector on the motor side can be matched with the current sector and phase conduction, commutation and demagnetization states. While reducing the number of candidate voltage vectors, the control capability of torque regulation and demagnetization process is retained, and the relevance of the candidate voltage vector set is improved.

[0034] Preferably, the phase current and torque corresponding to the candidate voltage vectors on each motor side at the next sampling time are predicted, including: Input the running state variables into the online corrected Fourier model; The predicted phase current at the next sampling moment is calculated based on the phase terminal voltage, temperature-compensated phase resistance, and phase current corresponding to the candidate voltage vectors on each motor side. The predicted phase torque is calculated based on the predicted phase current and rotor position angle at the next sampling time, and the predicted phase torques of each phase are superimposed to obtain the predicted phase current and torque corresponding to the candidate voltage vector on each motor side.

[0035] By adopting the above technical solution, the phase current and torque response corresponding to different candidate voltage vectors on the motor side can be obtained, so that there is a comparable predictive basis between the candidate voltage vectors, and the degree of matching between the control action and the actual motor response at the next sampling moment can be improved.

[0036] Preferably, the steps for constructing the evaluation function include: The torque error is calculated based on the torque reference value determined by the rotational speed and the predicted torque corresponding to the candidate voltage vector on each motor side. The model reliability is determined based on the Fourier model residuals, parameter update amounts, and runtime state quantities. The constraint terms are determined based on the change between the switching state of the power converter and the corresponding switching state of each motor-side candidate voltage vector, the predicted phase current over-limit, and the predicted negative torque. The evaluation weights corresponding to the torque error term and the constraint term are determined based on the model credibility, and the torque error and constraint terms are weighted and summed to obtain the evaluation function value corresponding to the candidate voltage vector on each motor side; among them, when the model credibility decreases, the evaluation weights corresponding to the peak current constraint and the negative torque constraint are increased.

[0037] By adopting the above technical solution, the Fourier model residual or model credibility is used in the evaluation of candidate voltage vectors on the motor side. This can strengthen the current peak constraint and negative torque constraint when the model credibility decreases, reduce continuous aggressive switching actions, and enable the evaluation function to simultaneously reflect the torque tracking requirement, the current credibility of the Fourier model, and the operating constraints corresponding to the candidate voltage vector. This allows the model mismatch information to directly affect the candidate voltage vector evaluation process, thereby improving the reliability of the selection of the target voltage vector on the motor side under the model mismatch state.

[0038] Preferably, the constraints include switching action constraints, peak current constraints, and negative torque constraints.

[0039] Preferably, the control signal corresponding to the target voltage vector on the motor side is determined, including: The evaluation function values ​​corresponding to the candidate voltage vectors on each motor side are compared, and the candidate voltage vectors on each motor side are constrained and screened based on the predicted phase current and predicted torque. Remove candidate voltage vectors on the motor side whose predicted phase current exceeds the current peak constraint or whose predicted torque is negative, and select the candidate voltage vector with the smallest evaluation function value from the remaining candidate voltage vectors on the motor side; The control signal is generated based on the switching state of the power converter corresponding to the selected candidate voltage vector, and the control signal corresponding to the target voltage vector on the motor side is obtained.

[0040] By adopting the above technical solution, candidate voltage vectors with predicted phase current exceeding the current peak constraint or predicted torque being negative can be preferentially excluded during the candidate voltage vector selection process. This ensures that the target voltage vector on the motor side simultaneously meets the operational safety requirements and torque control needs, thereby improving the safety and stability of the predictive control process.

[0041] Secondly, this application also provides an online self-calibrating Fourier model predictive control system for a switched reluctance motor, comprising: The status acquisition unit is used to acquire the operating status data of the switched reluctance motor and output the phase current, rotor position angle, speed, DC bus voltage, winding temperature and power converter switching status. The predictive response calculation unit is used to call the current Fourier model to process the running state variables and calculate the predictive response for the current sampling period. The online model correction unit is used to obtain the measured response corresponding to the predicted response, calculate the Fourier model residual, compensate the phase resistance according to the winding temperature, and correct the Fourier coefficients using the Fourier model residual that still exists after the phase resistance compensation, and output the online corrected Fourier model. The candidate vector processing unit is used to determine the current sector based on the rotor position angle, turn-on angle, turn-off angle and commutation angle, generate a set of candidate voltage vectors for each motor side of the current sector, and call the online modified Fourier model to calculate the phase current and torque corresponding to each candidate voltage vector at the next sampling time. The evaluation decision unit is used to determine the torque reference value based on the rotational speed, and to calculate the evaluation function value based on the torque error corresponding to the candidate voltage vector on each motor side, the Fourier model residual or the model credibility obtained from the Fourier model residual, and the constraint terms. The drive output unit is used to determine the candidate voltage vector on the motor side with the smallest evaluation function value as the target voltage vector on the motor side based on the evaluation function value, and output the control signal corresponding to the target voltage vector on the motor side to the switched reluctance motor power converter.

[0042] Preferably, the online model correction unit includes: The temperature compensation module is used to calculate the temperature-compensated phase resistance based on the winding temperature and the phase resistance at the reference temperature, and then substitute the temperature-compensated phase resistance into the phase voltage balance equation to recalculate the predicted phase current. The error identification module is used to calculate the current prediction residual based on the recalculated predicted phase current and the measured phase current, and to determine the identification error of the flux linkage model parameter correction based on the Fourier model residual that still exists after phase resistance compensation. The coefficient update module is used to construct a regression vector based on the phase current power term and the rotor position angle harmonic term, arrange the Fourier coefficients to be corrected into a parameter vector, and update the parameter vector using the recursive least squares method based on the identification error, forgetting factor and covariance matrix, and restore the updated parameter vector into the Fourier coefficient matrix to obtain the online corrected Fourier model.

[0043] As a preferred option, the evaluation decision-making unit includes: The credibility determination module is used to calculate the model credibility based on the Fourier model residuals, parameter update amounts, and effective states of operating state quantities within a predetermined statistical window, and to adjust the weights of the torque error term, switching action constraint term, current peak constraint term, and negative torque constraint term based on the model credibility. The candidate filtering module is used to determine the peak current constraint and negative torque constraint based on the predicted phase current and predicted torque corresponding to each motor-side candidate voltage vector, and to remove motor-side candidate voltage vectors whose predicted phase current exceeds the peak current constraint value or whose predicted torque is negative, and to determine the motor-side candidate voltage vector with the smallest evaluation function value from the remaining motor-side candidate voltage vectors.

[0044] As can be seen from the above technical solutions, the present invention has the following advantages: This application provides an online self-calibrating Fourier model predictive control method and system for switched reluctance motors. It identifies Fourier model mismatch by utilizing measured responses during operation and corrects the Fourier coefficients online after phase resistance compensation. This allows the Fourier model to adapt to changes in winding temperature and operating conditions, reducing inconsistencies between predicted and measured phase currents and deviations in predicted torque caused by a fixed Fourier model. By distinguishing phase resistance changes from flux linkage model parameter changes and setting update thresholds, protection thresholds, update amount limits, cumulative offset envelopes, and freezing conditions for Fourier coefficient updates, unreliable parameter updates caused by noise, weak excitation, and abnormal operating conditions are reduced, improving the stability of the online correction process. The online corrected Fourier model predicts the phase current and torque corresponding to each candidate voltage vector on the motor side, and the Fourier model residuals or model reliability are introduced into the evaluation function. This allows the degree of model mismatch to directly affect the selection of candidate voltage vectors on the motor side. Combined with switching action constraints, current peak constraints, and negative torque constraints, the adverse effects of model mismatch on commutation and demagnetization control are reduced, improving the reliability of target voltage vector selection and operational safety on the motor side. Attached Figure Description

[0045] Figure 1 This is a flowchart of an online self-calibrating Fourier model predictive control method for a switched reluctance motor provided by the present invention; Figure 2 This is a block diagram of the principle of an online self-calibrating Fourier model predictive control system for a switched reluctance motor provided by the present invention; Figure 3 The curves showing the online convergence of Fourier coefficients and the variation of Fourier model residuals in the numerical simulation experiment are shown. Figure 4 Comparison curves of phase current prediction under parameter drift and load step conditions; Figure 5 The torque response and model reliability curves under sudden load changes are shown. Figure 6 The curves show the comparison between the output phase current and the negative torque under high-speed demagnetization conditions.

[0046] The system comprises: 1. State acquisition unit; 2. Predictive response calculation unit; 3. Online model correction unit; 4. Candidate vector processing unit; 5. Evaluation and decision-making unit; and 6. Drive output unit. Detailed Implementation

[0047] It should be noted in advance that, in order to facilitate a clear and accurate description of the technical solutions in the embodiments of this application, the following is a brief explanation of some terms and related technologies involved in the embodiments of this application: 1. Fourier analytical model: This model uses Fourier series to analytically represent the periodic characteristics of the flux linkage of a switched reluctance motor as the rotor position angle changes, and combines the functional expression of the phase current direction to describe the flux linkage characteristics. It is used to calculate the phase flux linkage and phase electromagnetic torque based on the phase current and rotor position angle, and provides a model basis for the prediction of phase current and torque.

[0048] 2. Model Predictive Control: This is a control method that predicts the future operating response corresponding to different control actions based on the motor model, and selects the control action according to the deviation between the prediction result and the control target.

[0049] 3. Candidate voltage vector: This is the candidate control voltage state that the power converter can apply to the switched reluctance motor in the current control sector.

[0050] To address the problems of offline calibrated Fourier models in the predictive control of switched reluctance motors (SRMs) being unable to adapt to changes in operating conditions, the difficulty in effectively distinguishing between changes in phase resistance and flux linkage model parameters, and the tendency of unreliable residuals to cause erroneous model parameter updates, this application discloses an online self-calibrating Fourier model predictive control method and system for SRMs. By utilizing Fourier model residuals and combining them with winding temperature to compensate for phase resistance, the Fourier coefficients are corrected online, and parameter updates under abnormal conditions are constrained, thereby improving the adaptability of the Fourier model to changes in operating conditions. Simultaneously, the online corrected Fourier model is used to predict candidate voltage vectors on the motor side, and the reliability of the model is evaluated, improving the reliability of the selection of the target voltage vector on the motor side.

[0051] like Figure 1 As shown in the figure, this embodiment provides an online self-calibrating Fourier model predictive control method for a switched reluctance motor, characterized by comprising: Step S1: Collect the operating status data of the switched reluctance motor, including phase current, rotor position angle, speed, DC bus voltage, winding temperature and power converter switching status; Step S2: Based on the current Fourier model and the running state variables, calculate the predicted response for the current sampling period; Step S3: After obtaining the measured response corresponding to the predicted response, calculate the Fourier model residual, compensate the phase resistance using the winding temperature, obtain the identification error using the Fourier model residual that still exists after phase resistance compensation, and correct the Fourier coefficients of the current Fourier model online to obtain the online corrected Fourier model. Step S4: Determine the current sector based on the rotor position angle, turn-on angle, turn-off angle and commutation angle, generate a set of candidate voltage vectors for each motor side in the current sector, and use the online modified Fourier model to predict the phase current and torque corresponding to each candidate voltage vector at the next sampling time. Step S5: Determine the torque reference value based on the rotational speed. For each candidate voltage vector on the motor side, construct an evaluation function based on the torque error, Fourier model residuals or the model credibility obtained from the Fourier model residuals, and constraint terms. Step S6: Select the candidate voltage vector on the motor side with the smallest evaluation function value as the target voltage vector on the motor side, and output the control signal corresponding to the target voltage vector on the motor side to the switched reluctance motor power converter.

[0052] Hereinafter, steps S1 to S6 will be specifically described according to embodiments of this application.

[0053] In step S1, the core task is to collect the operating state quantities during the operation of the switched reluctance motor, and to synchronously organize the collected operating state quantities to form state data that can be used by the current Fourier model for predictive response calculation and subsequent online correction of the Fourier model.

[0054] In this embodiment, operating state quantities of the switched reluctance motor are collected during operation. These quantities include phase current, rotor position angle, rotational speed, DC bus voltage, winding temperature, and power converter switching status. Specifically, phase current characterizes the current state of each phase winding of the switched reluctance motor; rotor position angle characterizes the spatial position of the rotor relative to the stator; rotational speed reflects the current mechanical operating state of the switched reluctance motor; DC bus voltage characterizes the power supply status on the input side of the power converter; winding temperature reflects the temperature rise of the switched reluctance motor windings; and power converter switching status characterizes the current output voltage status of the power converter.

[0055] Specifically, when collecting operating status data, the phase current and DC bus voltage are synchronously acquired through the same sampling trigger signal, and the position sensor output, power converter switching status and protection flag are acquired at the corresponding sampling time; the rotor position angle is calculated based on the position sensor output, and then the rotor position angle is correlated with the phase current, DC bus voltage and power converter switching status according to the sampling time.

[0056] Each set of acquired operating status quantities is appended with a sampling sequence number and a timestamp to ensure the time correspondence between different operating status quantities. Range checks, rate of change checks, phase sequence checks, and sensor validity checks are then performed. Specifically, the range check determines whether the operating status quantity is within the effective range of the corresponding sensor or acquisition channel; the rate of change check determines whether the change in operating status quantity between adjacent sampling times meets the acquisition continuity requirements; the phase sequence check confirms the correspondence between the phase current sampling sequence and the phase sequence of the switched reluctance motor; and the sensor validity check confirms the validity of the source of the operating status quantity.

[0057] Furthermore, the winding temperature is collected according to the second sampling period; when forming the operating status frame, the timestamp of the most recent valid winding temperature sampling is retained to maintain the correspondence between the winding temperature and other operating status quantities. The operating status quantities containing the sampling sequence number and timestamp are uniformly organized into an operating status frame for subsequent steps to call.

[0058] In this embodiment, model predictive control is executed according to a first sampling period, and online correction of Fourier coefficients is executed according to a second sampling period, which is an integer multiple of the first sampling period. Online correction of Fourier coefficients can also be triggered by events such as the Fourier model residual exceeding the update threshold, winding temperature changes, or sudden load changes. If online correction of Fourier coefficients is not completed within the second sampling period, model predictive control still uses the current reliable Fourier coefficients to perform predictive response calculations without waiting for online correction to complete.

[0059] Step S1 completes the synchronous acquisition, validity check, and frame organization of the operating status quantities of the switched reluctance motor. This enables subsequent steps to call the current Fourier model to calculate the predicted response based on phase current, rotor position angle, speed, DC bus voltage, winding temperature, and power converter switching status under a unified time reference. It also provides operating status data for Fourier model residual calculation and online correction of Fourier coefficients.

[0060] In step S2, the core task is to calculate the predicted response corresponding to the current sampling period based on the current Fourier model and the running state variables, so as to provide a predictive data basis for subsequent Fourier model residual calculation, online correction of Fourier coefficients and candidate voltage vector prediction.

[0061] In this embodiment, the current Fourier model is input using the phase current and rotor position angle. The phase flux linkage is calculated according to the finite-order cosine Fourier series in the rotor position angle direction and the functional expression in the phase current direction. The phase electromagnetic torque is calculated based on the phase flux linkage. The predicted phase current and predicted torque for the current sampling period are calculated by combining the rotational speed, DC bus voltage, winding temperature and power converter switching state, and the predicted response for the current sampling period is obtained.

[0062] Specifically, the phase current and rotor position angle are input into the current Fourier model, and the phase flux linkage is calculated according to the finite-order cosine Fourier series in the direction of the rotor position angle and the functional expression in the direction of the phase current. The corresponding Fourier phase flux linkage model is expressed as: ; in, for Phase flux linkage, unit ; for Phase current, unit ; for Relative rotor position angle, unit ; This is the current sampling time; The highest order of harmonics to be retained; This is the conversion factor from the mechanical position angle to the model position period; These are the Fourier coefficients that vary with phase current and sampling time. It can be represented by a phase current polynomial, spline function, or piecewise function, and its coefficients form the Fourier coefficient matrix to be corrected online. This serves as the parameter object for subsequent online correction of the Fourier coefficients.

[0063] Furthermore, the phase electromagnetic torque is calculated based on the calculated phase flux linkage. The partial derivative with respect to the position angle is obtained from the relationship between the magnetic co-energy and the rotor position angle. Phase electromagnetic torque , represented as: ; in, Represents the electromagnetic torque of phase p, in units of ; This represents the integral current variable, with the unit being A.

[0064] After obtaining the phase flux linkage and phase electromagnetic torque, the operating response of the switched reluctance motor within the current sampling period is calculated by combining the rotational speed, DC bus voltage, winding temperature, and power converter switching state, resulting in the predicted phase current and predicted torque for the current sampling period. Specifically, the DC bus voltage characterizes the current input voltage state of the power converter, the power converter switching state determines the current output state, and the winding temperature reflects the current winding operating state.

[0065] In this embodiment, the calculated predicted response is recorded with corresponding relationships, including the currently used Fourier coefficient version number, temperature-compensated phase resistance, sampling sequence number, and timestamp, ensuring a correspondence between the predicted response and the subsequently acquired measured response. The prediction results include the predicted phase current and predicted torque corresponding to the current sampling period, used for subsequent calculation of the Fourier model residuals.

[0066] Furthermore, during the predicted response calculation, the initial offline calibration values, the currently active Fourier coefficients, and the most recent reliable Fourier coefficients are saved. The phase flux linkage, flux linkage partial derivatives, predicted phase current, and predicted phase torque are calculated based on the current Fourier model. When subsequent online correction of the Fourier coefficients is performed, the current Fourier model is updated based on the online corrected Fourier coefficients and used for the predicted response calculation in subsequent sampling periods.

[0067] Step S2 completes the prediction response calculation based on the current Fourier model and operating state variables, obtaining the predicted phase current and predicted torque that match the sampling sequence number and Fourier coefficient version. This provides prediction response data for subsequent steps to calculate the Fourier model residual, determine the identification error, and correct the Fourier coefficients of the current Fourier model online.

[0068] In step S3, the core task is to perform a corresponding analysis based on the predicted response and the measured response during actual operation to determine the degree of model mismatch between the current Fourier model and the actual operating state of the switched reluctance motor. The phase resistance is compensated by the winding temperature to eliminate the influence of phase resistance temperature drift on the identification of model parameters. The identification error is then determined by using the Fourier model residuals that still exist after phase resistance compensation. The Fourier coefficients in the current Fourier model are corrected online to obtain the online corrected Fourier model.

[0069] In this embodiment, the predicted response and the measured response are time-aligned according to the sampling sequence number, timestamp, and phase, so that the measured phase current at the current sampling moment corresponds to the predicted phase current obtained based on the current control state at the previous sampling moment. Specifically, the collected measured phase current undergoes zero-bias correction and outlier checking. When a single pulse spike exceeds the amplitude threshold or rate of change threshold, the corresponding sample is marked as invalid to prevent abnormal sampling data from affecting the subsequent online correction process of the Fourier coefficients.

[0070] The measured phase current and the predicted phase current after time alignment are compared, and the current prediction residual is calculated: ; in express time Phase current prediction residual, This represents the measured phase current at time k. This represents the predicted phase current formed based on the actual output state at the previous sampling time.

[0071] Furthermore, the estimated electromagnetic torque is determined based on the measured phase current, rotor position angle, and pre-calibrated electromagnetic characteristics at the current sampling time. The estimated electromagnetic torque is then time-aligned with the predicted torque at the previous sampling time to obtain the torque prediction residual. The current prediction residual and the torque prediction residual are normalized or combined to form the Fourier model residual, which is used to characterize the prediction error of the current Fourier model.

[0072] After completing the Fourier model residual calculation, the phase resistance is compensated based on the winding temperature. Specifically, the temperature-compensated phase resistance is calculated based on the winding temperature and the phase resistance at the reference temperature. ; in, For winding temperature Below Phase resistance, unit: Ω; Reference temperature The phase resistance is expressed in Ω. This represents the temperature coefficient of resistance of copper conductors, expressed in units of 1 / ℃.

[0073] Substitute the temperature-compensated phase resistance into the phase voltage balance equation, and recalculate the predicted phase current by combining the phase current, rotor position angle and DC bus voltage. Then, recalculate the current prediction residual based on the recalculated predicted phase current and the measured phase current.

[0074] The above processing allows for the differentiation between prediction errors caused by changes in phase resistance temperature and those caused by changes in Fourier model parameters.

[0075] Specifically, the temperature-compensated phase resistance is used to update the resistance parameters in the current Fourier model, and the predicted phase current is recalculated: ; in, To predict phase current, unit ; For the first sampling period, the unit ; The phase terminal voltage corresponding to the candidate voltage vector on the motor side, in units of ; For the equivalent voltage drop of power devices, unit ; Mechanical angular velocity, unit: rad / s Substituting the temperature-compensated phase resistance into the phase voltage balance equation, the predicted phase current is recalculated, and the residual current prediction after temperature compensation is determined based on the recalculated predicted phase current and the measured phase current. This residual current prediction is used as error information for correcting the flux linkage model parameters, and combined with the regression vector composed of the phase current power term and the rotor position angle harmonic term, the identification error used for the recursive update of the Fourier coefficients is determined. This ensures that the identification error primarily reflects the flux linkage model mismatch that still exists after temperature-compensated phase resistance changes, avoiding misidentification of phase resistance changes caused by winding temperature variations as changes in Fourier coefficients.

[0076] Furthermore, the Fourier coefficients of the current Fourier model are corrected online based on the identification error. Specifically, a regression vector is constructed based on the phase current power term and the rotor position angle harmonic term, and the Fourier coefficients to be corrected in the current Fourier model are arranged in a fixed order as a parameter vector.

[0077] Calculate the recursive gain based on the regression vector, identification error, forgetting factor, and covariance matrix: ; in, This represents the regression vector consisting of the phase current function term and the rotor position angle harmonic term. Represents the covariance matrix. The forgetting factor is greater than 0 and not greater than 1. This represents the recursive gain.

[0078] Based on the recursive gain, the parameter vector is updated using the recursive least squares method with a forgetting factor. Covariance Matrix ; in, This is the parameter vector obtained by expanding the Fourier coefficient matrix in a fixed order. This is the current identification error. This is the identity matrix. The plus sign indicates the value after this recursion. Each quantity is updated when the second sampling period arrives. During the first sampling period, the model predictive control continues to use the current reliable Fourier coefficients without waiting for the Fourier coefficients to be updated.

[0079] Furthermore, during the parameter vector update process, the update amount limit is determined based on the single update amount of the parameter vector, and the cumulative offset envelope is determined based on the cumulative offset of the parameter vector relative to the most recent reliable Fourier coefficient matrix, so that the Fourier coefficient update process simultaneously satisfies the instantaneous change limit and the long-term change range limit.

[0080] The update limit for the parameter vector is: ; in, express The parameter vector updated at each time step. This represents the parameter vector from the previous time step. This represents the correction amount of the original parameters obtained through recursion. This indicates the single allowable correction amount for the corresponding parameter, and sat represents the element-wise saturation function.

[0081] In addition to limiting the amount of each update, the cumulative offset envelope is determined based on the cumulative offset of the parameter vector relative to the most recent reliable Fourier coefficient matrix or relative to the offline initial value. When the cumulative parameter offset exceeds the allowable range, the parameter update process is restricted to avoid excessive offset of the Fourier coefficients caused by abnormal residuals, sensor disturbances, or insufficient excitation conditions.

[0082] Furthermore, the Fourier model residuals are compared with the update threshold and the protection threshold respectively to determine the first residual interval, the second residual interval, and the third residual interval. The relationship for determining the residual intervals is as follows: ; in, This represents the normalized current prediction residual or the Fourier model residual obtained by combining the current prediction residual and the torque prediction residual. This indicates that the threshold needs to be updated. Indicates the protection threshold, and Greater than . This indicates the residual dead zone, which is the first residual interval. This indicates the residual correction zone, i.e., the second residual interval. This indicates the outlier region, which is the third residual interval.

[0083] When the Fourier model residuals are in the first residual interval, the Fourier coefficients remain unchanged; when the Fourier model residuals are in the second residual interval, the parameter vector is updated using the recursive least squares method with a forgetting factor based on the update amount limit and the cumulative offset envelope; when the Fourier model residuals are in the third residual interval, the Fourier coefficients are frozen, and the most recent reliable Fourier coefficient matrix is ​​called to obtain the constrained Fourier coefficient update result.

[0084] In the abnormal region, when the phase current is in the zero current neighborhood, the speed is lower than the reliable identification threshold, the winding temperature is abnormal, the DC bus is undervoltage or overvoltage, the power converter switching state is abnormal, the parameter correction exceeds the allowable upper limit, the covariance matrix is ​​not positive definite or numerically divergent, or the regression vector does not meet any of the sufficient excitation conditions, the Fourier coefficient update is stopped and the Fourier coefficients are frozen.

[0085] During the process of unfreezing, when the sensor status recovers, the DC bus voltage recovers to the allowable range, the phase current enters the effective identification range, the Fourier model residual is continuously lower than the recovery threshold for a predetermined period, the parameter update amount meets the limit, and the power converter exits the protection or degradation state, the freeze state is lifted, the most recent reliable Fourier coefficient matrix is ​​called, and the parameter update is restored.

[0086] Finally, the updated parameter vector is restored to the Fourier coefficient matrix, forming an online corrected Fourier model. The online corrected Fourier model, the updated Fourier coefficient version, and the model parameter update status are output, providing a corrected model basis for subsequent prediction of candidate voltage vectors on the motor side.

[0087] It should be noted that, in addition to the recursive least squares method with a forgetting factor, other online identification methods can also be used to correct Fourier coefficients online.

[0088] In some embodiments of this application, online coefficient updates can employ the sliding window least squares method. The sliding window least squares method updates parameters by retaining data samples within a fixed-length window and explicitly deleting old samples outside the window. This method is suitable for controllers with ample storage resources and slow workspace switching. When using the sliding window least squares method, parameter updates are still performed after temperature compensation and are subject to the same residual interval, update amount limit, freeze condition, and recovery condition constraints.

[0089] In some embodiments of this application, online coefficient updates may also employ extended Kalman filtering or unscented Kalman filtering. Extended Kalman filtering or unscented Kalman filtering can simultaneously estimate Fourier parameters and unmeasured states, making it suitable for scenarios requiring joint estimation of load torque or device voltage drop. When using this type of filtering, the state covariance and parameter covariance are checked for finiteness and positive definiteness; large residuals in the outlier region are not used to directly increase the parameter update step size.

[0090] In some embodiments of this application, online coefficient updates can also employ a gradient update method. The gradient update method is suitable for controllers with limited computational resources, adjusting the direction of Fourier coefficient updates by identifying errors.

[0091] In some embodiments of this application, online coefficient updates can also employ a piecewise parameter identification method. This method is suitable for high-current regions with strong magnetic saturation. Parameters corresponding to low-current regions are used to describe the unsaturated region, while parameters corresponding to high-current regions are used to describe the saturated region. Different regions are connected through continuity constraints. Online updates only apply to the current region and adjacent transition regions, avoiding changes in the local operating conditions that could lead to overall changes in the Fourier model of unobserved regions.

[0092] Regardless of whether the recursive least squares method with forgetting factor, sliding window least squares method, extended Kalman filter, unscented Kalman filter, gradient update method, or piecewise parameter identification method is used, the following constraints are maintained: the measured response and the predicted response form time-aligned residuals; temperature-compensated phase resistance is executed first; the online update process is constrained by residual dead zone, update amount limit, cumulative offset envelope, freezing condition, and recovery condition; the corrected Fourier model participates in subsequent candidate voltage vector prediction; the Fourier model residual or the model credibility obtained from the Fourier model residual participates in the evaluation of the target voltage vector on the motor side.

[0093] Thus far, step S3 has completed the calculation of the Fourier model residuals based on the predicted response and the measured response, the compensation of phase resistance caused by winding temperature, the extraction and identification of errors, and the online correction of Fourier coefficients. This enables the current Fourier model to self-correct as the operating state of the switched reluctance motor changes, thereby improving the consistency between the model's predicted response and the actual operating response.

[0094] In step S4, the core task is to determine the current control sector based on the rotor position angle, and generate a set of candidate voltage vectors on the motor side within the current sector by combining the phase conduction state, commutation state, and demagnetization state corresponding to the turn-on angle, turn-off angle, and commutation angle. At the same time, the obtained online modified Fourier model is used to predict the response at the next sampling time after each candidate voltage vector is applied.

[0095] In this embodiment, the current sector is determined based on the relationship between the rotor position angle, turn-on angle, turn-off angle, and commutation angle. Specifically, the rotor position angle is converted to a position cycle, and the position cycle boundary effect caused by angle counting rewind is eliminated through modulo operation.

[0096] For a three-phase 12 / 8-pole switched reluctance motor, one position cycle is divided into six control sectors, A1 to A6, according to the commutation relationship between adjacent phases. Among them, A1, A3 and A5 are commutation-related sectors, while A2, A4 and A6 include single-phase conduction sections and Z demagnetization sections adjacent to the single-phase conduction sections.

[0097] The turn-on angle determines the position where the phase winding begins excitation, the turn-off angle determines the position where the phase winding ends excitation, and the commutation angle or sector width determines the commutation boundary between adjacent phases. These angle parameters can be determined using existing calculation relationships or through offline calibration mapping. Based on the correspondence between the current rotor position angle range and the turn-on, turn-off, and commutation angles, the current sector is determined.

[0098] In this embodiment of the application, after determining the current sector, the current sector is given a set of candidate voltage vectors on the motor side based on the phase conduction state, commutation state and demagnetization state corresponding to the current sector, and combined with the power converter switching state. This set of vectors includes torque increase vector, torque decrease vector, leading phase excitation vector, outgoing phase freewheeling vector and outgoing phase demagnetization vector.

[0099] For an asymmetrical half-bridge power converter, each corresponding equivalent voltage state includes a positive bus excitation state, a zero-voltage freewheeling state, and a negative bus demagnetization state. The polarity and ideal amplitude of the corresponding phase terminal voltage are determined based on the DC bus voltage, the conduction direction of the power devices, and the diode freewheeling path. The on-state voltage drop of the power devices and the freewheeling voltage drop of the diodes are included as independent equivalent voltage drop terms in the switched reluctance motor prediction model, and the actual switching state of the power converter is used as the input for the prediction calculation at the next sampling time.

[0100] Specifically, in the single-phase conduction region, the candidate voltage vector set includes at least an increasing torque vector and a decreasing torque vector. The increasing torque vector increases the rate of change of the phase current by applying an excitation voltage to the current operating phase; the decreasing torque vector reduces the phase current by using zero-voltage freewheeling, soft chopping, or demagnetizing to ensure that the output torque meets the control requirements.

[0101] When the reliability of the online corrected Fourier model decreases, the number of times the torque-increasing vector is selected is limited, and the torque-decreasing vector with a smaller current rise rate is retained first, so as to avoid the rapid change of phase current affecting control stability under model mismatch.

[0102] In the commutation region, a set of candidate voltage vectors is generated based on the current commutation state, simultaneously considering the states of the leading and trailing phase windings. The leading phase winding includes both excitation and freewheeling states, while the trailing phase winding includes both freewheeling and demagnetizing states. Candidate combinations are filtered based on the current torque direction and commutation sequence, removing those involving faulty phase excitation, non-permissible arm conduction, or failure to meet the trailing phase current reduction requirement. Each candidate combination is then labeled as an increasing torque vector or a decreasing torque vector based on its effect on phase current and torque at the current rotor position.

[0103] In the Z demagnetization region, the candidate voltage vector set preferentially includes the outgoing phase demagnetization vector that connects the outgoing phase winding to the negative bus. The outgoing phase current and phase torque after the action of the outgoing phase demagnetization vector are predicted using an online modified Fourier model. If the predicted outgoing phase current still exceeds the allowable range when entering the undesirable negative torque region, the evaluation weight of the negative torque constraint and current peak constraint corresponding to the candidate voltage vector is increased, or the candidate voltage vector is directly marked as unselectable.

[0104] Furthermore, to avoid frequent switching between adjacent sectors due to rotor position quantization errors, an angle hysteresis or minimum dwell time is set near the sector boundary to keep the current sector judgment stable. When the position sensor signal is invalid or the speed calculation is unreliable, the Fourier coefficient update is frozen; if the switched reluctance motor is still allowed to operate at a degraded level, the most recent reliable sector boundary and the limited candidate voltage vector set are used for control.

[0105] Furthermore, for four-phase or multi-phase switched reluctance motors, the number of sectors is adjusted according to the number of phases, the number of rotor poles, and the commutation overlap relationship. At the same time, the number of candidate voltage vectors is adjusted according to the power converter topology changes. However, the candidate voltage vector set does not unconditionally enumerate all power converter switching states, but retains the valid candidate voltage vectors related to the current operating state.

[0106] After obtaining the set of candidate voltage vectors for each motor side in the current sector, the acquired operating state quantities are input into the online modified Fourier model to predict the phase current and torque corresponding to each candidate voltage vector at the next sampling time.

[0107] For any candidate voltage vector, the predicted phase current at the next sampling moment is calculated based on the phase terminal voltage, temperature-compensated phase resistance, and current phase current corresponding to the candidate voltage vector. ; in, Indicates the first The next sampling time corresponding to each candidate voltage vector Predicted phase current of phase, in units ; For the first sampling period, the unit ; The phase terminal voltage corresponding to the candidate voltage vector on the motor side, in units of ; For the equivalent voltage drop of power devices, unit ; Mechanical angular velocity, unit: rad / s; This represents the phase flux calculated using the online corrected Fourier model.

[0108] Furthermore, the predicted phase torque is calculated based on the predicted phase current and predicted rotor position angle at the next sampling time. The predicted phase torque is determined based on the magnetic co-energy relation: ; in, Indicates the predicted phase torque; Represents the integral current variable; This represents the phase flux linkage output of the online corrected Fourier model.

[0109] For a multiphase switched reluctance motor, the predicted phase torques of each corresponding phase are superimposed to obtain the predicted total torque corresponding to the candidate voltage vector. The predicted phase current and predicted torque of each candidate voltage vector are calculated in the same way, forming a predicted response result corresponding to the candidate voltage vector.

[0110] Furthermore, the Fourier model residuals and model confidence in the online modified Fourier model are used to determine the validity of the candidate voltage vector prediction results. When the online modified Fourier model is frozen or the model confidence decreases, the most recent confident Fourier coefficient matrix is ​​used for prediction to avoid abnormal parameter updates affecting the candidate voltage vector evaluation results.

[0111] At this point, step S4 has completed the determination of the current sector and generated a set of candidate voltage vectors on the motor side based on the current sector's operating status. At the same time, the online modified Fourier model is used to calculate the predicted phase current and predicted phase torque at the next sampling time corresponding to each candidate voltage vector, providing a predictive response basis for subsequent candidate voltage vector evaluation and target voltage vector selection.

[0112] In step S5, the core task is to determine the torque reference value based on the rotational speed, and for each candidate voltage vector on the motor side, calculate the torque error between the predicted torque and the torque reference value corresponding to each candidate voltage vector on the motor side. At the same time, the model reliability is determined based on the Fourier model residual, parameter update amount, and operating state amount. Constraint terms are constructed by combining the switching state change amount, predicted phase current over-limit amount, and predicted negative torque amount corresponding to the candidate voltage vector. The evaluation function value corresponding to each candidate voltage vector on the motor side is obtained by weighted summation.

[0113] In this embodiment, a torque reference value is determined based on the speed loop output. This torque reference value is then compared with the predicted total torque corresponding to each motor-side candidate voltage vector, and the torque error corresponding to each motor-side candidate voltage vector is calculated. For the j-th motor-side candidate voltage vector, its torque error can be expressed as: ; in, This represents the torque reference value determined based on the rotational speed, in units of... ; This represents the predicted total torque corresponding to the j-th motor-side candidate voltage vector, in units of... ; This represents the torque error corresponding to the j-th candidate voltage vector on the motor side.

[0114] Furthermore, the model reliability is determined based on the Fourier model residuals, parameter update amounts, and operational status variables. Model reliability characterizes the reliability of the current prediction results of the Fourier model when corrected online. When the Fourier model residuals increase, the parameter change rate increases, or the operational status is abnormal, the model reliability is reduced; when the Fourier model residuals are small, the parameter updates are stable, and the operational status is normal, the model reliability is increased.

[0115] The formula for calculating model credibility is: ; in , represents the model's confidence level, and its value ranges from 0 to 1; Indicates the effective factor of sensor and protection status; This represents the root mean square of the current prediction residuals within the statistical window. This represents the normalized benchmark for current prediction residuals; Indicates the parameter correction amount; This represents the normalized baseline for parameter variation; and These are non-negative weights. When the Fourier model residuals increase, the parameters change more rapidly, or the sensor fails... reduce.

[0116] In some embodiments of this application, the weights of each evaluation item in the evaluation function are dynamically adjusted based on the model credibility. When the model credibility decreases, the evaluation weights corresponding to the peak current constraint, negative torque constraint, and switching action constraint are increased, while the evaluation weights corresponding to the torque error item are decreased as needed, thereby reducing the impact of the uncertainty of the model prediction results on the selection of candidate voltage vectors on the motor side; when the model credibility recovers, the weights of each evaluation item are gradually restored to normal values.

[0117] Furthermore, constraints are determined based on the changes between the switching states of the power converter and the corresponding switching states of the candidate voltage vectors on each motor side, the predicted phase current over-limit, and the predicted negative torque. These constraints include switching action constraints, peak current constraints, and negative torque constraints.

[0118] The switching action constraint is used to limit the amplitude of the power converter switching state change between adjacent sampling periods. For the j-th motor-side candidate voltage vector, the switching action change can be determined based on the difference between the candidate switching state and the actual switching state in the previous period, denoted as . ;in, This indicates the switching state corresponding to the j-th motor-side candidate voltage vector; This indicates the actual switching state in the previous sampling period.

[0119] The predicted phase current over-limit constraint is used to limit candidate voltage vectors whose predicted phase current exceeds the allowable range at the next sampling time. The predicted negative torque constraint is used to limit the predicted torque from entering the undesirable negative torque region.

[0120] Candidate voltage vectors on the motor side that exceed the peak current constraint or generate negative torque are marked as unselectable and do not participate in the subsequent selection of target voltage vectors.

[0121] After obtaining the torque error, model confidence level, and various constraint terms, the torque error and constraint terms are normalized. The weights of each evaluation term are determined based on the model confidence level, and a weighted sum is performed to obtain the evaluation function value corresponding to each motor-side candidate voltage vector. The evaluation function corresponding to the j-th motor-side candidate voltage vector can be expressed as: ; in, This represents the evaluation function value corresponding to the j-th candidate voltage vector on the motor side; Indicates the torque error term; Indicates the switch action constraint terms; This indicates the penalty for predicting phase current exceeding the limit; This indicates the predicted negative torque penalty term; Indicates the model's credibility; , , and This represents the non-negative weights determined based on the model's credibility.

[0122] When model credibility decreases, improve and And improve as needed or reduce When the model's credibility is restored, each evaluation weight gradually returns to its normal value according to the preset rate of change.

[0123] Torque tracking performance is ensured by using a torque error term, the influence of Fourier model prediction results is corrected online by using model credibility constraints, and the safe operation of the power converter is ensured by switching action constraints, current peak constraints, and negative torque constraints.

[0124] Thus far, step S5 has completed the determination of the torque reference value based on the rotational speed, and constructed an evaluation function based on the predicted response, torque error, Fourier model residual or model credibility, and constraint terms corresponding to each candidate voltage vector on the motor side, thereby realizing a comprehensive evaluation of each candidate voltage vector on the motor side and providing an evaluation basis for the subsequent selection of the target voltage vector.

[0125] In step S6, the core task is to select the target voltage vector on the motor side within the candidate range that satisfies the peak current constraint and the negative torque constraint based on the evaluation function value corresponding to each candidate voltage vector on the motor side, and generate a control signal based on the switching state of the power converter corresponding to the target voltage vector on the motor side, and output the control signal to the switched reluctance motor power converter to complete the motor side predictive control of the current sampling period.

[0126] Specifically, after obtaining the evaluation function values ​​corresponding to each motor-side candidate voltage vector, the candidate voltage vectors are constrained and screened based on the predicted phase current and predicted torque. For motor-side candidate voltage vectors whose predicted phase current exceeds the peak current constraint, or whose predicted torque is negative, they are marked as unselectable candidate voltage vectors; for motor-side candidate voltage vectors that satisfy both the peak current constraint and the negative torque constraint, they are retained as selectable candidate voltage vectors.

[0127] Furthermore, the evaluation function values ​​corresponding to each motor-side candidate voltage vector are compared from the remaining selectable candidate voltage vectors, and the motor-side candidate voltage vector with the smallest evaluation function value is selected as the motor-side target voltage vector. When the difference between the evaluation function values ​​corresponding to two motor-side candidate voltage vectors is less than the control resolution threshold, further comparisons are made based on the number of switching actions, current safety margin, or predicted negative torque, and the motor-side candidate voltage vector with fewer switching actions, larger current safety margin, or smaller predicted negative torque is preferentially selected as the motor-side target voltage vector.

[0128] After determining the target voltage vector on the motor side, a control signal is generated based on the switching state of the power converter corresponding to the target voltage vector. This control signal undergoes switch dead-time processing, gate interlocking, and isolation drive before being output to the switched reluctance motor power converter, enabling the converter to perform power conversion control according to the switching state corresponding to the target voltage vector on the motor side.

[0129] In some embodiments of this application, after the target voltage vector on the motor side is output, the actual output state is fed back to the next sampling period. Specifically, the actual output voltage vector, active Fourier coefficient version, operating state quantity, prior prediction result at the next sampling time, and online correction state are written into the circular cache, and the Fourier model residual statistics window, model confidence, freeze and resume consecutive count, and event flag are updated.

[0130] The actual switching state of the switched reluctance motor power converter serves as the input for the prediction process in the next sampling period, participating in subsequent prediction of the motor-side candidate voltage vector, evaluation of the motor-side candidate voltage vector, and calculation of the Fourier model residual. When a hardware protection signal is detected, the control signal output can be asynchronously blocked between sampling periods to prevent the power converter from continuing to drive under abnormal conditions.

[0131] Thus far, step S6 has completed the selection of the target voltage vector on the motor side based on the evaluation function value, the generation of the control signal, and the drive output of the switched reluctance motor power converter. It also updates the model state and control state required for the next sampling period through feedback of the actual output state, forming a closed-loop predictive control.

[0132] like Figure 2 As shown, the following is an embodiment of an online self-calibrated Fourier model predictive control system for a switched reluctance motor provided by the present invention. This online self-calibrated Fourier model predictive control system for a switched reluctance motor belongs to the same inventive concept as the online self-calibrated Fourier model predictive control methods for switched reluctance motors in the above embodiments. For details not described in detail in the embodiments of the online self-calibrated Fourier model predictive control system for a switched reluctance motor, please refer to the embodiments of the online self-calibrated Fourier model predictive control methods for switched reluctance motors described above.

[0133] Based on the same concept, another embodiment of this application provides an online self-calibrating Fourier model predictive control system for a switched reluctance motor, comprising: Status acquisition unit 1 is used to acquire the operating status data of the switched reluctance motor and output the phase current, rotor position angle, speed, DC bus voltage, winding temperature and power converter switching status. Predictive response calculation unit 2 is used to call the current Fourier model to process the running state variables and calculate the predicted response for the current sampling period; The online model correction unit 3 is used to obtain the measured response corresponding to the predicted response, calculate the Fourier model residual, compensate the phase resistance according to the winding temperature, and use the Fourier model residual that still exists after the phase resistance compensation to correct the Fourier coefficients and output the online corrected Fourier model. Candidate vector processing unit 4 is used to determine the current sector based on rotor position angle, turn-on angle, turn-off angle and commutation angle, generate a set of candidate voltage vectors for each motor side of the current sector, and call the online modified Fourier model to calculate the phase current and torque corresponding to each candidate voltage vector at the next sampling time. Evaluation and decision unit 5 is used to determine the torque reference value based on the rotational speed, and to calculate the evaluation function value based on the torque error corresponding to the candidate voltage vector on each motor side, the Fourier model residual or the model credibility obtained from the Fourier model residual, and the constraint terms. The drive output unit 6 is used to determine the candidate voltage vector on the motor side with the smallest evaluation function value as the target voltage vector on the motor side based on the evaluation function value, and output the control signal corresponding to the target voltage vector on the motor side to the switched reluctance motor power converter.

[0134] In some embodiments of this application, the online model correction unit 3 includes: The temperature compensation module is used to calculate the temperature-compensated phase resistance based on the winding temperature and the phase resistance at the reference temperature, and then substitute the temperature-compensated phase resistance into the phase voltage balance equation to recalculate the predicted phase current. The error identification module is used to calculate the current prediction residual based on the recalculated predicted phase current and the measured phase current, and to determine the identification error of the flux linkage model parameter correction based on the Fourier model residual that still exists after phase resistance compensation. The coefficient update module is used to construct a regression vector based on the phase current power term and the rotor position angle harmonic term, arrange the Fourier coefficients to be corrected into a parameter vector, and update the parameter vector using the recursive least squares method based on the identification error, forgetting factor and covariance matrix, and restore the updated parameter vector into the Fourier coefficient matrix to obtain the online corrected Fourier model.

[0135] In some embodiments of this application, the evaluation decision unit 5 includes: The credibility determination module is used to calculate the model credibility based on the Fourier model residuals, parameter update amounts, and effective states of operating state quantities within a predetermined statistical window, and to adjust the weights of the torque error term, switching action constraint term, current peak constraint term, and negative torque constraint term based on the model credibility. The candidate filtering module is used to determine the peak current constraint and negative torque constraint based on the predicted phase current and predicted torque corresponding to each motor-side candidate voltage vector, and to remove motor-side candidate voltage vectors whose predicted phase current exceeds the peak current constraint value or whose predicted torque is negative, and to determine the motor-side candidate voltage vector with the smallest evaluation function value from the remaining motor-side candidate voltage vectors.

[0136] In summary, the online self-calibrating Fourier model predictive control system for switched reluctance motors provided in this embodiment achieves integrated processing of switched reluctance motor operating state perception, online Fourier model correction, and predictive control decision-making through the coordinated operation of the state acquisition unit 1, predictive response calculation unit 2, online model correction unit 3, candidate vector processing unit 4, evaluation and decision-making unit 5, and drive output unit 6. Specifically, the system identifies model mismatch by utilizing the Fourier model residual between the predicted response and the measured response, and compensates for phase resistance by combining winding temperature, thus distinguishing parameter offsets caused by temperature factors from model structural errors. It then uses the temperature-compensated residual Fourier model to online correct the Fourier coefficients, improving the adaptability of the flux linkage model under different operating conditions. Simultaneously, by adjusting the evaluation function weights based on model credibility and filtering candidate voltage vectors by combining predicted phase current and predicted torque constraints, it reduces the impact of abnormal predictive responses on control decisions, achieving a synergistic improvement in the stability of switched reluctance motor torque output and model prediction accuracy, meeting the high-precision predictive control requirements of switched reluctance motors under complex operating conditions.

[0137] To further verify the effectiveness of the method proposed in this invention, this embodiment employs numerical simulation for verification and provides the configuration of a verification platform and verification steps that can be used for physical experiments. The verification content includes online correction of Fourier coefficients, temperature compensation, model confidence adjustment, evaluation of candidate voltage vectors on the motor side, and freezing and recovery under abnormal conditions. For ease of description, the control method that always calls the offline initial Fourier coefficient matrix without online correction is referred to as fixed model control, and the control method that calls the online corrected Fourier model is referred to as online calibration model control; the above abbreviations are only used for... Figures 3 to 6 And a comparative explanation of the corresponding verification results.

[0138] The online self-calibrating Fourier model predictive control system for the switched reluctance motor in this embodiment can be implemented by a DSP or ARM alone, or by a combination of DSP and FPGA. In the DSP and FPGA combined implementation, the DSP or ARM mainly performs speed loop control, temperature compensation calculation, Fourier model residual interval judgment, parameter vector recursive update, Fourier coefficient freezing and recovery, and selection of target voltage vector on the motor side; the FPGA is used to calculate the predicted phase current and predicted torque corresponding to each candidate voltage vector on the motor side in parallel to meet the calculation time limit within the first sampling period.

[0139] The online correction of Fourier coefficients employs a double-buffered storage structure. First, recursive calculations, update limiting, covariance matrix checks, and magnetic flux linkage model physical consistency checks are performed in the shadow parameter area. Once all checks are satisfied, the active parameter area is switched at the boundary of the first sampling period, and the Fourier coefficient version number is updated synchronously. The FPGA consistently reads the same version of the Fourier coefficient matrix within the same candidate prediction batch, avoiding inconsistencies in the prediction model caused by partial parameter updates and non-updated parameters during online updates.

[0140] The isolated drive circuit in the drive section of the switched reluctance motor is used to perform level conversion, power device gate drive, and hardware interlocking. The protection circuit independently detects overcurrent, DC bus overvoltage, DC bus undervoltage, drive undervoltage, and power device faults. After the drive output unit 6 generates the control signal corresponding to the target voltage vector on the motor side, it performs a validity check through an allowable state table. Only control signals that meet the safety constraints of the power converter are written to the PWM register or the direct switching control register.

[0141] In the experimental platform, current sensors are preferentially configured independently for each phase. Bus voltage sensors are used to detect the DC bus voltage at the input of the power converter. Position sensors can be incremental encoders, absolute encoders, or resolvers. Temperature sensors are located at the winding ends or near the stator core to acquire information on winding temperature changes. Each sensor interface is equipped with independent range checks and validity flags. When any sensor data is invalid, this status flag directly participates in the model reliability calculation process.

[0142] The DSP and FPGA transmit timestamped data via dual-port RAM, high-speed serial bus, or on-chip interconnect. The DSP only receives the FPGA prediction result corresponding to the current sampling sequence number; when the FPGA fails to complete the candidate prediction calculation before the current control cycle deadline, the controller continues to operate using the previous cycle's safety vector or preset degraded vector without waiting for the delayed prediction result, thus ensuring the determinism of the control cycle.

[0143] In this embodiment, the speed loop sampling frequency is set to 500Hz, and the sampling frequencies for current, bus voltage, and position are set to 10kHz. The online update frequency of the Fourier coefficients is lower than or equal to the fast control frequency. These sampling frequencies are used to illustrate the dual-timescale relationship between the fast predictive control task and the slow parameter identification task. In practical applications, these frequencies can be adjusted according to the speed range of the switched reluctance motor, the switching frequency of the power devices, and the computing power of the controller.

[0144] To verify the effectiveness of the control system during actual operation, the physical test platform can consist of a three-phase 12 / 8-pole switched reluctance motor, an asymmetrical half-bridge power converter, a DC power supply, a magnetic powder brake or a loaded motor, a torque sensor, an incremental encoder, a current sensor, a bus voltage sensor, a winding temperature sensor, and a DSP, FPGA, or dSPACE real-time controller. During the online calibration function operation, actual response data, model predicted response data, online updates of model parameters, control commands, and protection status information are recorded synchronously for subsequent result analysis.

[0145] During the verification of the effects of temperature rise, the switched reluctance motor was operated at multiple stable winding temperature operating points while maintaining the same speed and load conditions. Three states were tested: no phase resistance compensation, phase resistance compensation only with temperature adjustment, and simultaneous temperature compensation and online flux linkage model correction. This experiment verifies whether the main current prediction error caused by temperature changes can be eliminated first through phase resistance compensation, and whether the remaining Fourier model residuals after compensation can drive the online update of the Fourier coefficients.

[0146] During load mutation verification, the applied torque was varied within the safe current range, and the following data were recorded: torque reference value, speed, three-phase current, bus voltage, actual switch status, predicted phase current, predicted torque, Fourier model residual, model reliability, and parameter version information. For both the fixed model control method and the online correction model control method, the same candidate voltage vector set and the same initial evaluation function weights were used; only the online correction of the Fourier coefficients was changed to ensure that the comparison between the two methods only reflected the effect of online parameter correction.

[0147] During the high-speed demagnetization verification process, the changes in output phase current, rotor position angle, and phase torque near the turn-off angle were recorded at multiple operating speeds. The degree of improvement of the online correction model on the demagnetization constraint effect was evaluated based on the residual current, minimum phase torque, and total torque ripple when entering the negative torque region. Before the high-speed test began, the overcurrent protection, bus overvoltage protection, and power converter protection functions were verified. After the protection functions were confirmed to be normal, the operating speed and load were gradually increased.

[0148] For the verification of the freeze and recovery functions, Fourier coefficient freeze was triggered by manually setting conditions such as sensor disconnection, range exceeding limits, DC bus undervoltage or overvoltage, weak excitation window, and covariance anomaly. This experiment focuses on verifying whether the parameters maintain the most recent reliable value under abnormal conditions, whether the controller can operate according to the preset degradation strategy, and whether the parameter correction process is restarted with limited update gain after the fault is recovered and the recovery conditions are met.

[0149] For each set of experiments, the original time-series data, sampling frequency, filtering parameters, control thresholds, evaluation function weights, controller version information, and ambient temperature information for both the fixed model and the online calibration model were saved. All quantitative evaluation indicators were calculated based on the saved original data, and the corresponding operating conditions, comparison objects, and statistical time windows were clearly defined. Among them, the steady-state performance improvement index was calculated using steady-state statistical intervals, and single transient peak values ​​were not used as evaluation criteria.

[0150] After the experimental platform is built, in order to ensure that the online corrected Fourier model has a reliable initial parameter foundation, this embodiment further explains the initialization, offline calibration, and online debugging process: When deploying an online self-calibrating Fourier model predictive control system for a switched reluctance motor for the first time, flux linkage-current relationships at several characteristic rotor positions are obtained through torque balance tests, locked-rotor tests, or finite element analysis data. Based on the flux linkage-current data collected at different rotor positions, finite-order periodic basis functions are used to reconstruct the system in the rotor position angle direction, and polynomial functions, spline functions, or piecewise functions are used to fit the Fourier coefficients of each order in the phase current direction, thereby obtaining the offline initial model parameters.

[0151] Before writing the offline model into the controller, the obtained offline initial model parameter matrix undergoes an integrity check, including flux linkage fitting error, flux linkage value under zero current conditions, incremental inductance variation range, position period consistency, phase-to-phase relationship, and torque curve variation trend. When high-order harmonics cause non-physical oscillations in local areas, corrections are made by reducing the Fourier expansion order or adding smoothing constraints. The Fourier coefficient matrix, reference phase resistance, reference temperature, model order, and checksum, after offline verification, together constitute the initial parameter version.

[0152] After powering on, the controller reads the initial parameter version and verifies the parameter matrix dimension, data range, and checksum. If the parameter version meets the validity requirements, it is copied to the active parameter area and the most recently trusted parameter area. If the parameter version verification fails, the online update function is disabled, and an abnormal parameter status is output. The online calibration process is always based on offline-verified model parameters, avoiding searching for the magnetolink model using arbitrary initial parameters, and preventing the online identification process from entering non-physical parameter areas.

[0153] Before going online, the zero bias of the current sensor, the proportional coefficient of the DC bus voltage, the rotor zero position, and the phase sequence relationship were calibrated. Since the zero position error affects both the Fourier basis function calculation and the commutation sector division, online updates of the Fourier coefficients were prohibited if the position calibration failed. The temperature difference relationship between the temperature sensor installation location and the winding hot spot, as well as the thermal response time constant, were determined through temperature rise tests, providing a basis for calculating the phase resistance of temperature compensation.

[0154] In the first stage of the online debugging process, parameter writing is disabled, and only the current model prediction, Fourier model residual calculation, and model credibility calculation processes are run to check the time correspondence between the actual output voltage vector and the prior model prediction results. When the residual sign of the same phase winding is inconsistent with the expected change direction in the excitation region, freewheeling region, and demagnetization region, the phase terminal voltage sign, rotor zero position, and power switch status readback information are checked first, rather than directly performing model parameter correction.

[0155] The second phase initiates small-gain parameter updates within the residual dead zone and online correction region, while maintaining the abnormal state freeze logic and protection functions. The controller records the sample window used for each recursive calculation, the regression vector, the original parameter update amount, the update amount after limiting, the covariance matrix status, and parameter version information. When the direction of parameter change is inconsistent with the physical change trend of the flux linkage model, the parameter submission is rejected, and the regression vector arrangement and data validity are rechecked.

[0156] In the third stage, the freezing and recovery process of Fourier coefficients was verified through scenarios such as current sensor overshoot, DC bus voltage anomaly, position data loss, temperature signal overshoot, overcurrent protection triggering, power converter anomaly, covariance matrix anomaly, and weak excitation state. During the experiment, it was confirmed that the Fourier coefficients maintained the values ​​corresponding to the most recent reliable Fourier coefficient matrix during the anomaly period. The controller operated according to the preset degradation strategy, and after the fault was cleared, the recovery observation period, trial update amount, and model reliability change trend were checked to determine whether to re-enter the parameter update state.

[0157] The fourth stage involves activating the model confidence term to verify the changes in the set of candidate voltage vectors on the motor side and the weights of the evaluation function when the model confidence is in the high confidence interval, the transitional confidence interval, and the low confidence interval. The three model confidence intervals are defined based on the model confidence value, unlike the first, second, and third residual intervals defined based on the Fourier model residuals. When model confidence decreases, the number of consecutive selections of torque-increasing vectors is reduced, and current peak constraints and negative torque constraints are strengthened. When model confidence recovers, the weights of the evaluation function gradually return to normal values ​​to avoid repeated switching of the target voltage vector on the motor side near the boundaries of the model confidence intervals.

[0158] After the controller completes initialization and online debugging, the consistency of the recorded and output prediction data is constrained. Specifically, the online correction of the Fourier coefficient matrix, predicted phase current, predicted phase torque, predicted total torque, predicted bus current or predicted bus power, current sector, turn-on angle, turn-off angle, commutation angle, Fourier model residual, model confidence, set of candidate voltage vectors on the motor side, and target voltage vector on the motor side are all appended with timestamps under a unified controller time base. When external data recording devices read the above data, they also read the corresponding Fourier coefficient version number to avoid combining prediction results from different Fourier coefficient versions.

[0159] Furthermore, the phase current prediction data and candidate voltage vector data can be output in real time according to the first sampling period, or they can be cached internally in the controller and downsampled according to the communication bandwidth. The downsampling process only affects the data interface transmission and does not change the actual execution frequency of the model predictive control on the motor side. When parameter freeze, recovery, protection action, or parameter version switching events occur, the corresponding event data frames are not downsampled and discarded to ensure complete traceability of abnormal processes.

[0160] The predicted bus current or predicted bus power is calculated based on the current DC bus voltage, the phase state corresponding to the candidate voltage vector on the motor side, and the predicted phase current. This prediction result is provided as read-only data to an external data recording device. It does not participate in the selection of the target voltage vector on the motor side, nor does it change the functional boundary of this control system, which only outputs the control signal corresponding to the target voltage vector on the motor side to the switched reluctance motor power converter.

[0161] The internal data of the control system and the external communication interface are uniformly described using the terms "motor-side candidate voltage vector" and "motor-side target voltage vector". While external data recording devices can perform state analysis using predicted data, this control system still generates the motor-side target voltage vector solely based on the prediction results of the current sampling period and outputs the corresponding control signal to the switched reluctance motor power converter.

[0162] The aforementioned data consistency constraints ensure that the online-corrected Fourier coefficients, the predicted candidate voltage vectors on the motor side, and the actual drive output correspond to each other within the same sampling period. This enables the calculation of Fourier model residuals, the determination of Fourier coefficient freezing, the calculation of model credibility, and... Figures 3 to 6 The simulation results shown have a consistent data foundation.

[0163] Based on the experimental platform and data recording conditions described above, this embodiment further uses discrete-time numerical simulation to verify the proposed method. Figures 3 to 6The simulation results shown are all generated based on a unified simulation environment. The sampling frequency for voltage, current, and position data is set to 10kHz, the operating speed is set to 1500 r / min, the motor's rated power is 5.5kW, the rated current is 11.3A, the rated load is 35 N·m, and the stator phase resistance reference value is set to 1.472Ω. During the simulation, the same online threshold, normalization coefficients, and simplified object parameters were used to construct a fixed model and an online correction model to compare the operating conditions.

[0164] The numerical simulation first generates pulsating phase currents based on the position-period relationship of a three-phase 12 / 8-pole switched reluctance motor, and then applies normalized Fourier principal coefficient drift within the 40ms to 60ms interval. The response of the controlled object is generated from the drifted Fourier coefficients. The fixed model always maintains the offline initial coefficients, while the online calibration model is recursively updated based on the updated parameter vector. Measurement noise is generated using a fixed random seed to ensure consistent and repeatable simulation results across different experimental processes.

[0165] Numerical simulation results are as follows Figures 3 to 6 As shown. Among them, Figure 3 This was used to verify the effectiveness of the recursive least squares method with a forgetting factor for the online correction process of Fourier coefficients and the freezing and recovery mechanism of Fourier coefficients under abnormal conditions. Figure 3 The upper curve records the preset true Fourier coefficients and the online corrected Fourier coefficients at each sampling time. The lower curve calculates the root mean square of the Fourier model residuals using 100 sampling points from the first sampling period as a statistical window. Within 78ms to 86ms, the sensor validity flag is set to invalid, and the online model correction unit 3 saves the most recent reliable Fourier coefficient matrix and stops recursive updates. After sensor recovery, recursive updates resume only when the Fourier model residuals and trial update amounts meet the recovery conditions.

[0166] Specifically, to verify the online coefficient update relationship, a three-phase 12 / 8-pole switched reluctance motor was used as the simulation object. The rated power of the motor was set to 5.5kW, the rated speed was set to 1500r / min, the voltage and current sampling frequency was set to 10kHz, and the stator phase resistance reference value was set to 1.472Ω.

[0167] During the simulation, a Fourier coefficient that significantly affects the prediction error of the phase current was selected as the online identification object, and the coefficient was normalized. Within a time range of 40ms to 60ms, the normalized Fourier coefficient was gradually changed from 1 to 0.86 to simulate the flux linkage model mismatch caused by changes in the magnetic saturation region, operating point migration, or changes in model parameters during the long-term operation of the switched reluctance motor.

[0168] The online correction process for Fourier coefficients is performed according to step S3. In this embodiment, the forgetting factor is set to 0.995, and the single normalization correction amount of the parameter vector is limited to 0.0025. These values ​​are used to generate... Figure 3 The results shown can be adjusted in practical applications based on the target motor parameters, sensor measurement accuracy, and controller calculation range. To verify the Fourier coefficient update protection mechanism under abnormal conditions, the sensor validity flag is set to invalid within 78ms to 86ms, the recursive update of Fourier coefficients is stopped, and the most recent reliable Fourier coefficient matrix is ​​maintained. After the sensor state recovers, Fourier coefficient updates are resumed only when the Fourier model residual and the trial update amount meet the recovery conditions.

[0169] Figure 3 The upper curve is used to record the changes in the set true Fourier coefficients and the online identified Fourier coefficients. Simulation results show that after model parameters drift, the online identified coefficients gradually track the changes in the true coefficients using a constrained update method; after entering the sensor's abnormal freeze zone, the identified coefficients maintain the reliable parameter values ​​before freezing; once the recovery conditions are met, the online identified coefficients continue to converge under constrained conditions. Figure 3 The lower curve is used to compare the changes in the root mean square of the Fourier model residuals of the fixed model and the online calibration model. In the fixed model, the residuals gradually increase after parameter drift because the initial Fourier coefficients are continuously used. In the online calibration model, the model mismatch is reduced by recursive updates, so the root mean square of the residuals gradually decreases.

[0170] In this embodiment, the online update is not performed simultaneously on all Fourier coefficients, but only on a subset of parameters with sufficient identification sensitivity within the current current-position operating region. When the range of phase current amplitude variation, rotor position angle coverage, or minimum eigenvalue of the regression matrix within the regression window does not meet the requirements, the controller keeps the current parameters unchanged, but continues to execute the model predictive control process to avoid invalid samples under low excitation conditions changing the overall flux linkage model.

[0171] To verify the stability of the parameter update process at different time scales, the second sampling period was set to K times the first sampling period, where K is a positive integer. Within one second sampling period, the sample buffer stores K sets of valid data for each phase. Valid data includes at least the phase current, position angle, equivalent phase terminal voltage, temperature-compensated phase resistance, Fourier model residual, sensor validity flag, and model version number. At the end of the second sampling period, only samples with the same model version and valid sensor status are selected to form the recursive regression window.

[0172] The regression vector in the recursive process includes the phase current function term, the position angle cosine harmonic term, and their product. The parameter vectors are arranged in the same order as the offline Fourier coefficient matrix to avoid shifts in the correspondence between parameters and physical model positions after online updates. To reduce the order-of-magnitude differences between parameters of different orders, normalization can be performed based on the rated current and reference flux linkage before online recursion, and the actual physical dimensions are restored after the recursion is completed.

[0173] During parameter updates, a forgetting factor closer to 1 indicates a longer retention period of historical samples in the current parameter estimation, resulting in a relatively slower model update speed. A smaller forgetting factor allows the model to respond more quickly to changes in operating conditions, but it also increases the impact of measurement noise on the parameter update results. Therefore, in actual operation, it is necessary to combine the second sampling period, residual dead zone, and update amount limit for tuning. Event-triggered updates are only used to initiate the parameter identification process once in advance and do not change the update amount limit or freeze determination conditions.

[0174] To ensure the numerical stability of the recursive process, the finiteness, symmetry, and positive definiteness of the covariance matrix are checked after each parameter update. When using a fixed-point controller, symmetric storage or square root recursion can be employed to reduce numerical errors. If the covariance matrix contains non-numerical elements, the condition number exceeds the allowable range, or the minimum eigenvalue is abnormal, the current parameter version is rejected, and the most recent reliable covariance matrix is ​​restored.

[0175] After completing the parameter recursion, a physical consistency check of the flux linkage model is performed, including: the flux linkage value near zero current should be close to zero; the incremental inductance within the effective current range should remain positive or within the preset physical boundary range; the phase flux linkage should satisfy the position period relationship and the phase relationship between phases; and the direction of change of the predicted torque near the aligned and unaligned positions should conform to the structural characteristics of the target switched reluctance motor. If any physical boundary condition is not met, the current updated parameter version is not submitted.

[0176] The updated Fourier coefficients are first written to the shadow parameter area. After passing the update amount limit, covariance check, and magnetic flux model physical consistency check, they are atomically switched at the boundary of the first sampling period to become active Fourier coefficients, and the Fourier coefficient version number is updated simultaneously.

[0177] Figure 4 Adopted and Figure 3The same simulation object and Fourier coefficient drift trajectory are used. At 60ms, the phase current amplitude is switched from a low load level to a high load level, and the phase current at the next sampling time is calculated using both the offline initial Fourier coefficients and the online corrected Fourier coefficients. The root mean square error of the phase current prediction is calculated as the square root of the average of the squares of the difference between the predicted phase current and the simulated actual phase current. Since changes in winding temperature will cause changes in phase resistance and result in current prediction errors, the temperature-compensated phase resistance is calculated first, and then it is determined whether the Fourier model residuals still existing after compensation enter the second residual interval.

[0178] The winding temperature sensor is located at the end of the stator winding or near the winding hot spot. The controller calculates the phase resistance compensation value based on the temperature measurement results, taking into account the thermal inertia between the temperature sensor installation location and the winding hot spot. When the temperature rise causes the predicted phase current to be too high, the controller adjusts the phase resistance parameter and re-predicts. If the Fourier model residual after temperature compensation enters the first residual interval, the Fourier coefficient update is not performed. If the Fourier model residual is still in the second residual interval, only the Fourier coefficients in the corresponding current interval and position harmonic range are updated.

[0179] like Figure 4 As shown, the simulation process includes: first, running at a low load operating point and recording the actual phase current and the model-predicted phase current within three position cycles; then, changing the load-corresponding current level at 60ms; finally, fixing the model and keeping the initial Fourier coefficients unchanged, and continuing to use the online calibration model. Figure 3 The Fourier coefficients after online correction are shown. Both prediction processes use the same measured bus voltage, temperature-compensated phase resistance, and corresponding sampling time, thus ensuring that the difference between the two prediction curves only stems from whether the flux linkage model parameters are updated online.

[0180] Figure 4 Simulation data from 54ms to 73ms was captured, and the phase current amplitude was switched from a low load level to a high load level at 60ms. The simulated actual phase current in the figure is calculated from the controlled object model after Fourier coefficient drift. The fixed model predicts the phase current using the offline initial Fourier coefficients, while the online correction model predicts the phase current using... Figure 3 The Fourier coefficients are shown after online correction. Both control methods use the same DC bus voltage, temperature-compensated phase resistance, and sampling time; therefore, the difference in prediction results only reflects the effectiveness of the online correction of the Fourier coefficients.

[0181] Under simulated conditions, the prediction errors after a load step change were statistically analyzed. The root mean square error (RMS) of the phase current prediction for the fixed model was 0.299 A, while that for the online correction model was 0.050 A. The results indicate that after temperature compensation eliminates the influence of the main resistance variation, the remaining prediction error caused by flux linkage model mismatch can be reduced through online correction using Fourier coefficients.

[0182] exist Figure 4 The online verification process for temperature compensation and Fourier model correction further considers the handling of temperature sensor anomalies. When a temperature sensor malfunctions, the recursive update of Fourier coefficients is stopped and the model reliability is reduced. If the winding temperature can be obtained online through phase resistance estimation, the estimated temperature is used as a degraded input to continue calculating the temperature-compensated phase resistance to maintain model predictive control. When both the winding temperature and phase resistance estimation results are unavailable, the most recent reliable Fourier coefficient matrix is ​​called, and the weight of the current peak constraint term is increased to reduce the impact of model uncertainty on the selection of candidate voltage vectors on the motor side.

[0183] When calculating the predicted phase current corresponding to the candidate voltage vector on the motor side, in addition to determining the phase terminal voltage based on the DC bus voltage and the switching state of the power converter, the on-state voltage drops of the upper and lower power devices and the freewheeling diodes are also included as independent equivalent voltage drop terms in the prediction model. The on-state voltage drops of the power devices can be determined by looking up tables based on the conduction direction, phase current, and heat sink temperature. When device temperature measurements or on-state voltage drop calibration data are lacking, offline nominal on-state voltage drops are used, and the remaining error is included in the model reliability evaluation. Not all phase terminal voltage errors are attributed to changes in the Fourier coefficients.

[0184] For a three-phase switched reluctance motor with only one winding temperature measurement point, this measurement point can be used as a common temperature reference, and the interphase temperature difference can be estimated based on the duty cycle of each phase winding and the most recent thermal state. The thermal time constant in the temperature estimation model is determined through temperature rise experiments. When the temperature estimation process is not yet stable, the online flux linkage parameter update uses a small update gain to avoid coupling between the transient changes in the temperature model and the flux linkage parameter correction process.

[0185] After completing the temperature compensation verification, this implementation method further combines... Figure 5 The effects of event-triggered updates, model confidence adjustments, and online correction control under load mutation conditions were verified.

[0186] During operation, the controller determines load abrupt events based on the torque reference change rate, speed change rate, phase current change rate, and bus voltage change rate. When a load abrupt event is detected, the original parameter update limiting conditions are not changed. Instead, the next second sampling cycle parameter identification process is triggered in advance, and the regression samples before and after the abrupt event are segmented and saved to avoid the old steady-state samples occupying too high a proportion in the sliding window and weakening the information of the new operating condition.

[0187] Figure 5During the simulation, the torque reference value was increased from 15 N·m to 25 N·m at 60 ms. Fixed model control and online calibration model control used the same controlled object model, the same control sampling period, and the same torque reference; only the Fourier model parameters were changed to determine whether online updates were performed. Model reliability was calculated from the root mean square of the residuals, parameter changes, and the effective sensor flags. Figure 5 The upper curve is used to represent the changes in electromagnetic torque response under the two control methods. Figure 5 The lower curve represents the change in model reliability calculated based on the Fourier model residuals, parameter variations, and sensor status.

[0188] When using a fixed model control, the offline initial Fourier coefficients continuously mismatch, resulting in significant steady-state torque deviation and torque ripple after load changes. When using an online correction model control, the Fourier coefficients gradually converge, the predicted flux linkage becomes closer to that of the controlled object, and the electromagnetic torque gradually tracks the torque reference value. Model reliability initially decreases due to increased Fourier model residuals during the initial stage of Fourier coefficient drift, but gradually recovers as online correction reduces the residuals. During the sensor inactivity period from 78ms to 86ms, model reliability further decreases, and the Fourier coefficients remain frozen.

[0189] The peak-to-peak value of torque ripple is calculated as follows: First, the slow-varying component of the sliding window is subtracted from the electromagnetic torque. Then, the difference between the maximum and minimum values ​​of the remaining fast torque component is calculated within the specified steady-state statistical interval. This separates the slow-varying response caused by load step and the fast torque fluctuation caused by position period, avoiding the inclusion of the amplitude of the load step itself in the torque ripple evaluation.

[0190] Based on the above statistical methods, the peak-to-peak torque ripple corresponding to the fixed model control is 3.644 N·m, while the peak-to-peak torque ripple corresponding to the online correction model control is 1.376 N·m. This result indicates that, under the same simulation conditions, load variations, and control cycles, the online correction Fourier model can reduce torque fluctuations caused by model mismatch.

[0191] During load abrupt changes, when the load suddenly increases, online calibration prioritizes samples with high sensitivity to flux linkage prediction within the leading phase excitation region; when the load suddenly decreases, it prioritizes checking the current residual and negative torque prediction in the outgoing phase demagnetization region. Samples before and after the abrupt change are saved separately, allowing the parameter update process to adapt to the new operating region more quickly and avoiding the old steady-state samples in the sliding window from occupying too high a weight. The event triggering mechanism only changes the parameter identification task start time, without changing the real-time control cutoff sequence of the first sampling period.

[0192] It should be noted that model reliability does not directly replace the Fourier model residual protection threshold. The protection threshold is mainly used to identify single or short-term anomalies, while model reliability is used to comprehensively evaluate prediction consistency, parameter change rate, covariance status, and sensor effectiveness within the statistical window. Therefore, even if the instantaneous Fourier model residual recovers to the residual dead zone, when the parameters have just undergone a significant change or the sensor has just recovered, the model reliability still gradually recovers according to the restricted change slope, avoiding frequent switching of the target voltage vector on the motor side caused by sudden changes in control weights.

[0193] The evaluation function employs different weighting strategies depending on the model's confidence level. In the high-confidence region, the evaluation function primarily selects the target voltage vector based on torque tracking error and the number of switching actions. In the transitional-confidence region, penalties for model uncertainty, current peak, and switching changes are increased, while the number of consecutive selections of torque-increasing vectors is limited. In the low-confidence region, only candidate voltage vectors satisfying current and negative torque constraints are retained, and the torque reference rise rate is limited as needed. Simultaneously, a maximum limit is set on the change in evaluation weights between adjacent sampling periods to reduce repeated switching of the target voltage vector near the confidence boundary.

[0194] After completion Figure 5 After verifying the load mutation shown, this embodiment further combines Figure 6 The effectiveness of online correction model-assisted high-speed demagnetization control and negative torque constraint under high-speed operation conditions was verified.

[0195] Under high-speed operation, the effective demagnetization time of a switched reluctance motor within one position cycle is shortened. If the fixed model underestimates the flux linkage or the decay rate of the outgoing phase current, it may lead to a delay in the turn-off time, resulting in a large residual current even after the outgoing phase current enters the inductance reduction region, thus generating negative torque. Therefore, during high-speed demagnetization, the controller uses an online-corrected flux linkage model to predict the outgoing phase current decline trajectory in the Z demagnetization region and imposes a large penalty or directly eliminates candidate voltage vectors that cannot be reduced to a safe range before the negative torque region.

[0196] Figure 6 The simulation conditions shown are a rotational speed of 1500 r / min and an initial output phase current of 11.3 A at the turn-off time. The demagnetization process is calculated under fixed model control and online correction model control, respectively. Figure 6 The upper curve represents the trajectory of the output phase current decrease under the two control modes, where the vertical line indicates the position of entering the negative torque region; Figure 6 The lower curve represents the change process of the negative torque of the outgoing phase calculated based on the remaining outgoing phase current. Both control methods use the same initial outgoing phase current, speed, DC bus voltage, and outgoing phase negative torque calculation relationship.

[0197] Depend on Figure 6 It can be seen that the online correction model control adjusts the demagnetization process based on the current flux linkage state obtained from the online corrected Fourier model, causing the outgoing phase current to drop to a lower level before entering the negative torque region. Compared with the fixed model control, the remaining outgoing phase current when entering the negative torque region is smaller under the online correction model control, and the corresponding outgoing phase negative torque amplitude is lower.

[0198] Under this simulation condition, the minimum phase torque corresponding to the fixed model control is -0.414 N·m, and the minimum phase torque corresponding to the online correction model control is -0.094 N·m. The curves show that the online correction flux linkage model, demagnetizing current prediction, and negative torque constraint evaluation term work synergistically to reduce the risk of negative torque caused by model mismatch under high-speed operation.

[0199] When the model reliability decreases, the controller does not compensate for the unknown model error by simply increasing the excitation voltage. Instead, it reduces the operational risk by limiting the number of times the continuous torque-increasing voltage vector is selected, increasing the priority of the demagnetizing candidate voltage vector, and limiting the torque reference change rate. This avoids the tailing of the out-phase current caused by model mismatch and improves the control stability during high-speed operation.

[0200] The phase inductance and inductance position bias required for high-speed demagnetization prediction are calculated using an online modified Fourier model. At turn-off, the controller saves the outgoing phase current, rotor position angle, DC bus voltage, and the current version of the Fourier coefficients, and predicts the outgoing phase current trajectory for multiple subsequent sampling cycles for different demagnetization candidate voltage vectors. When the prediction result indicates that a certain motor-side candidate voltage vector cannot reduce the outgoing phase current to a safe range before entering the negative torque region, the evaluation function value of that motor-side candidate voltage vector is increased or it is marked as unselectable.

[0201] Figure 6 The fixed model control and the online correction model control shown use the same initial outgoing phase current, speed, DC bus voltage, and outgoing phase negative torque calculation relationships. The fixed model control keeps the offline initial Fourier coefficients unchanged, and its predicted effective demagnetization start time is relatively delayed; the online correction model control uses the online corrected Fourier coefficients to calculate the predicted demagnetization current trajectory, thus forming different demagnetization processes and outgoing phase negative torque responses.

[0202] For switched reluctance motors with different power ratings or different stator and rotor pole numbers, the angular position corresponding to entering the negative torque region, the allowable residual current range, and the set of candidate demagnetizing voltage vectors all need to be recalibrated according to the specific motor structure. However, as long as the controller uses the online-corrected flux linkage model to predict the demagnetizing current and uses the predicted negative torque as an evaluation index or elimination condition for candidate voltage vectors, it falls under the technical solution described in this embodiment.

[0203] Based on online parameter updates, temperature compensation, load mutation, and high-speed demagnetization verification, this implementation further verifies the freezing, degradation, and recovery mechanisms during the online calibration process.

[0204] Except when the Fourier model residuals reach the protection threshold, the controller freezes the Fourier coefficient update process when any of the following conditions are met: the phase current is in the zero current neighborhood; the rotational speed is below the reliable identification threshold; the current, voltage, position, or temperature sensor is abnormal; overcurrent protection occurs; the DC bus is under-voltage or over-voltage; the power converter fails; the parameter correction exceeds the allowable range; the covariance matrix is ​​non-positive definite, numerically divergent, or exceeds the allowable range; the current sample window lacks sufficient excitation conditions.

[0205] The freeze process only halts the recursive update of Fourier coefficients, not the model predictive control process. The controller continues to use the most recently validated parameters to perform candidate voltage vector prediction; if the most recently validated parameters are unavailable, it reverts to the initial parameter version that has been validated offline. Simultaneously, the evaluation function enters a degraded weighting mode, increasing the weights of peak current constraints, negative torque constraints, and switching change penalties. When overcurrent, bus anomalies, or power converter faults occur and drive shutdown is required, hardware protection circuitry takes priority in executing protection actions, rather than using parameter freeze to replace hardware protection.

[0206] To lift the Fourier coefficient freeze, at least the following conditions must be met simultaneously: the current sensor, voltage sensor, position sensor, and temperature sensor return to normal; the DC bus voltage returns to the allowable range; the phase current re-enters the effective identification range; the Fourier model residual is below the recovery threshold for multiple consecutive second sampling periods; the trial update meets the limiting condition; the covariance matrix remains valid; and the power converter exits the protection state or the control system exits the degraded state.

[0207] The online calibration status includes initialization status, hold status, calibration status, freeze status, degradation status, and recovery observation status. After the controller is powered on, it verifies the dimension and check code of the offline parameter matrix and checks the physical boundaries of the Fourier coefficients; it enters the hold status when the Fourier model residual is in the first residual interval; it enters the calibration status when the Fourier model residual is in the second residual interval and all update conditions are met; it enters the freeze status when hardware abnormalities, sensor abnormalities, or parameter abnormalities occur.

[0208] When a freeze event occurs, the triggering reason, corresponding phase, sampling time, current sector position, Fourier model residual, phase current, bus voltage, and model version information are recorded for subsequent fault analysis and operational traceability. After the fault is cleared, the system enters a recovery observation state, calculating only the prior model prediction results and Fourier model residuals without immediately submitting a new parameter version. When the recovery conditions are met consecutively, the Fourier coefficient correction process is restarted with a limited update gain, and the normal evaluation function weights are gradually restored based on changes in model confidence. If any condition fails again during the recovery period, the recovery count is reset to zero, and the system re-enters the freeze state.

[0209] It's important to note that degraded state and parameter freeze state are not the same. Parameter freeze only indicates that Fourier coefficient updates have stopped, while the motor-side model predictive control can still continue to operate using the most recent reliable parameters. Degraded state further restricts control behavior, such as limiting the rate of change of torque reference, narrowing the allowable candidate voltage vector range, or using a fixed safety angle. When a position sensor failure or power converter failure causes the predictive control itself to be unable to operate reliably, the protection circuit directly blocks the drive output, rather than replacing hardware protection with parameter freeze logic.

[0210] To further illustrate the adaptability of the method of the present invention to switched reluctance motors with different structures, this embodiment also provides extension methods for four-phase and multi-phase switched reluctance motors.

[0211] For four-phase or multi-phase switched reluctance motors, each phase can be built with an independent Fourier model, or a unified model can be built by combining common basic Fourier coefficients with phase corrections. The position period, phase difference, and number of commutation sectors in the model are configured according to the number of stator and rotor poles and the phase overlap relationship. For each local sector, only a small number of candidate voltage vectors related to the current leading phase state, leading phase state, and torque direction are generated to maintain the computational complexity of predictive control.

[0212] In the multiphase extension scenario, the calculation of Fourier model residuals, recursive update of parameter vectors, and freezing and recovery of Fourier coefficients for each phase are all performed separately. If a phase does not meet the sufficient excitation condition, only the corresponding Fourier coefficients are frozen; if a common DC bus fault or position sensor fault occurs, the Fourier coefficients of all phases are frozen. This extension method does not change the closed-loop relationship of the online corrected Fourier model participating in the prediction of candidate voltage vectors on the motor side in the next sampling period.

[0213] Furthermore, the controller can be configured with a read-only data output interface to output online corrected Fourier coefficient matrix, predicted phase current, predicted phase torque, predicted total torque, predicted bus current or predicted bus power, current sector, turn-on angle, turn-off angle, commutation angle, Fourier model residual, model confidence, set of candidate voltage vectors on the motor side, and target voltage vector on the motor side to external data recording devices according to a unified timestamp and Fourier coefficient version number. These data are used for verification result recording and operational status analysis.

[0214] The output data described above can be read by external data analysis tools during the design phase, or by external data recording devices during the operation phase. The data output interface only provides the online corrected Fourier coefficient matrix and prediction results; the controller still generates the target voltage vector on the motor side based on the prediction results of the current sampling period and outputs the corresponding control signal to the switched reluctance motor power converter.

[0215] Numerical simulations use a fixed random seed, a determined sampling frequency, and preset parameter drift trajectories, which can be repeatedly obtained using the same algorithm. Figures 3 to 6 The results shown illustrate the interaction between parameter drift, online correction of Fourier coefficients, model confidence adjustment, candidate voltage vector constraints, and the final control response.

[0216] Through the above verification platform configuration, parameter initialization, data consistency management, numerical simulation, and extended verification process, it can be verified that the online self-tuning Fourier model predictive control method proposed in this invention can maintain effective operation under conditions of model parameter changes, temperature disturbances, load mutations, high-speed demagnetization, and abnormal states, thereby improving the accuracy of the predictive model and control stability of the switched reluctance motor.

Claims

1. A method for online self-calibrating Fourier model predictive control of a switched reluctance motor, characterized in that, include: The operating status parameters of the switched reluctance motor are collected, including phase current, rotor position angle, speed, DC bus voltage, winding temperature and power converter switching status. Based on the current Fourier model and the running state variables, calculate the predicted response for the current sampling period; After obtaining the measured response corresponding to the predicted response, the Fourier model residual is calculated, the phase resistance is compensated using the winding temperature, the identification error is obtained by using the Fourier model residual that still exists after the phase resistance compensation, and the Fourier coefficients of the current Fourier model are corrected online to obtain the online corrected Fourier model. The current sector is determined based on the rotor position angle, turn-on angle, turn-off angle and commutation angle. A set of candidate voltage vectors for each motor side in the current sector is generated. The phase current and torque corresponding to each candidate voltage vector at the next sampling time are predicted using an online modified Fourier model. The torque reference value is determined based on the rotational speed. For each candidate voltage vector on the motor side, an evaluation function is constructed based on the torque error, Fourier model residuals or model credibility obtained from Fourier model residuals, and constraint terms. The candidate voltage vector on the motor side with the smallest evaluation function value is selected as the target voltage vector on the motor side, and the control signal corresponding to the target voltage vector on the motor side is output to the switched reluctance motor power converter.

2. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, Calculate the predicted response for the current sampling period, including: Input the phase current and rotor position angle into the current Fourier model, and calculate the phase flux linkage according to the finite-order cosine Fourier series in the direction of rotor position angle and the functional expression in the direction of phase current. The phase electromagnetic torque is calculated based on the phase flux linkage, and the predicted phase current and predicted torque for the current sampling period are calculated by combining the rotational speed, DC bus voltage, winding temperature and power converter switching state, so as to obtain the predicted response for the current sampling period.

3. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, Calculating the Fourier model residuals includes: The measured phase current at the current sampling time is time-aligned with the predicted phase current at the previous sampling time, and the current prediction residual is calculated. The estimated electromagnetic torque at the current sampling time is determined based on the measured phase current, rotor position angle, and pre-calibrated electromagnetic characteristics, and is time-aligned with the predicted torque at the previous sampling time, and the torque prediction residual is calculated. The current prediction residual and the torque prediction residual are normalized or combined to obtain the Fourier model residual.

4. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, The identification error includes: Calculate the temperature-compensated phase resistance based on the winding temperature and the phase resistance at the reference temperature; Substitute the temperature-compensated phase resistance into the phase voltage balance equation, and recalculate the predicted phase current in combination with the phase current, rotor position angle and DC bus voltage. Then, calculate the current prediction residual based on the recalculated predicted phase current and the measured phase current. The current prediction residual is mapped to the Fourier model residual. Based on the Fourier model residual that still exists after phase resistance compensation, the identification error for the flux linkage model parameter correction is determined, and the identification error used for online correction of Fourier coefficients is obtained.

5. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, The online corrected Fourier model is obtained, including: The regression vector is constructed based on the phase current power term and the rotor position angle harmonic term, and the Fourier coefficients to be corrected in the current Fourier model are arranged into a parameter vector. Calculate the recursive gain based on the regression vector, identification error, forgetting factor, and covariance matrix; The parameter vector and covariance matrix are updated using recursive gain, and the updated parameter vector is restored to the Fourier coefficient matrix to obtain the online modified Fourier model.

6. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 5, characterized in that, Update the parameter vector, including: The update amount limit is determined based on the single update amount of the parameter vector, and the cumulative offset envelope is determined based on the cumulative offset of the parameter vector relative to the most recent reliable Fourier coefficient matrix. The Fourier model residuals are compared with the update threshold and the protection threshold respectively to obtain the first residual interval, the second residual interval and the third residual interval; In the first residual interval, the Fourier coefficients are kept unchanged. In the second residual interval, the parameter vector is updated using the recursive least squares method with a forgetting factor based on the update amount limit and the cumulative offset envelope. In the third residual interval, the Fourier coefficients are frozen and the most recent reliable Fourier coefficient matrix is ​​called to obtain the constrained Fourier coefficient update result, thus completing the parameter vector update.

7. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 6, characterized in that, Freezing Fourier coefficients includes: Fourier coefficient updates are stopped and Fourier coefficients are frozen under any of the following conditions: phase current is in the zero current neighborhood, speed is below the reliable identification threshold, winding temperature is abnormal, DC bus is under-voltage or over-voltage, power converter switching state is abnormal, parameter correction exceeds the allowable upper limit, covariance matrix is ​​non-positive definite or numerically divergent, and regression vector does not meet the sufficient excitation conditions.

8. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 5, characterized in that, The updated parameter vector is restored to a Fourier coefficient matrix, including: The freeze state is lifted after the sensor status recovers, the DC bus voltage recovers to the allowable range, the phase current enters the effective identification range, the Fourier model residual is lower than the recovery threshold for a continuous predetermined period, the parameter update meets the limit, and the power converter exits the protection or degradation state. Call the most recent trusted Fourier coefficient matrix and restore the parameter update, and restore the updated parameter vector to the Fourier coefficient matrix.

9. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, Generate a set of candidate voltage vectors for each motor side of the current sector, including: The current sector is determined based on the relationship between the rotor position angle and the turn-on angle, turn-off angle, and commutation angle. Based on the current phase conduction, commutation, and demagnetization states of the current sector, and combined with the switching states of the power converter, the torque increase vector, torque decrease vector, leading phase excitation vector, outgoing phase freewheeling vector, and outgoing phase demagnetization vector are generated to obtain the set of candidate voltage vectors for each motor side of the current sector.

10. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, Predict the phase current and torque corresponding to the candidate voltage vectors on each motor side at the next sampling time, including: Input the running state variables into the online corrected Fourier model; The predicted phase current at the next sampling moment is calculated based on the phase terminal voltage, temperature-compensated phase resistance, and phase current corresponding to the candidate voltage vectors on each motor side. The predicted phase torque is calculated based on the predicted phase current and rotor position angle at the next sampling time, and the predicted phase torques of each phase are superimposed to obtain the predicted phase current and torque corresponding to the candidate voltage vector on each motor side.

11. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, The steps to construct the evaluation function include: The torque error is calculated based on the torque reference value determined by the rotational speed and the predicted torque corresponding to the candidate voltage vector on each motor side. The model reliability is determined based on the Fourier model residuals, parameter update amounts, and runtime state quantities. The constraint terms are determined based on the change between the switching state of the power converter and the corresponding switching state of each motor-side candidate voltage vector, the predicted phase current over-limit, and the predicted negative torque. The evaluation weights corresponding to the torque error term and the constraint term are determined based on the model credibility, and the torque error and constraint terms are weighted and summed to obtain the evaluation function value corresponding to the candidate voltage vector on each motor side; among them, when the model credibility decreases, the evaluation weights corresponding to the peak current constraint and the negative torque constraint are increased.

12. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 11, characterized in that, The constraints include switching action constraints, peak current constraints, and negative torque constraints.

13. The online self-calibrating Fourier model predictive control method for switched reluctance motors as described in claim 1, characterized in that, Determine the control signal corresponding to the target voltage vector on the motor side, including: The evaluation function values ​​corresponding to the candidate voltage vectors on each motor side are compared, and the candidate voltage vectors on each motor side are constrained and screened based on the predicted phase current and predicted torque. Remove candidate voltage vectors on the motor side whose predicted phase current exceeds the current peak constraint or whose predicted torque is negative, and select the candidate voltage vector with the smallest evaluation function value from the remaining candidate voltage vectors on the motor side; The control signal is generated based on the switching state of the power converter corresponding to the selected candidate voltage vector, and the control signal corresponding to the target voltage vector on the motor side is obtained.

14. A self-calibrating Fourier model predictive control system for a switched reluctance motor, characterized in that, include: The status acquisition unit is used to acquire the operating status data of the switched reluctance motor and output the phase current, rotor position angle, speed, DC bus voltage, winding temperature and power converter switching status. The predictive response calculation unit is used to call the current Fourier model to process the running state variables and calculate the predictive response for the current sampling period. The online model correction unit is used to obtain the measured response corresponding to the predicted response, calculate the Fourier model residual, compensate the phase resistance according to the winding temperature, and correct the Fourier coefficients using the Fourier model residual that still exists after the phase resistance compensation, and output the online corrected Fourier model. The candidate vector processing unit is used to determine the current sector based on the rotor position angle, turn-on angle, turn-off angle and commutation angle, generate a set of candidate voltage vectors for each motor side of the current sector, and call the online modified Fourier model to calculate the phase current and torque corresponding to each candidate voltage vector at the next sampling time. The evaluation decision unit is used to determine the torque reference value based on the rotational speed, and to calculate the evaluation function value based on the torque error corresponding to the candidate voltage vector on each motor side, the Fourier model residual or the model credibility obtained from the Fourier model residual, and the constraint terms. The drive output unit is used to determine the candidate voltage vector on the motor side with the smallest evaluation function value as the target voltage vector on the motor side based on the evaluation function value, and output the control signal corresponding to the target voltage vector on the motor side to the switched reluctance motor power converter.

15. The online self-calibrating Fourier model predictive control system for switched reluctance motors as described in claim 14, characterized in that, The online model correction unit includes: The temperature compensation module is used to calculate the temperature-compensated phase resistance based on the winding temperature and the phase resistance at the reference temperature, and then substitute the temperature-compensated phase resistance into the phase voltage balance equation to recalculate the predicted phase current. The error identification module is used to calculate the current prediction residual based on the recalculated predicted phase current and the measured phase current, and to determine the identification error of the flux linkage model parameter correction based on the Fourier model residual that still exists after phase resistance compensation. The coefficient update module is used to construct a regression vector based on the phase current power term and the rotor position angle harmonic term, arrange the Fourier coefficients to be corrected into a parameter vector, and update the parameter vector using the recursive least squares method based on the identification error, forgetting factor and covariance matrix, and restore the updated parameter vector into the Fourier coefficient matrix to obtain the online corrected Fourier model.

16. The online self-calibrating Fourier model predictive control system for switched reluctance motors as described in claim 14, characterized in that, The evaluation and decision-making units include: The credibility determination module is used to calculate the model credibility based on the Fourier model residuals, parameter update amounts, and effective states of operating state quantities within a predetermined statistical window, and to adjust the weights of the torque error term, switching action constraint term, current peak constraint term, and negative torque constraint term based on the model credibility. The candidate filtering module is used to determine the peak current constraint and negative torque constraint based on the predicted phase current and predicted torque corresponding to each motor-side candidate voltage vector, and to remove motor-side candidate voltage vectors whose predicted phase current exceeds the peak current constraint value or whose predicted torque is negative, and to determine the motor-side candidate voltage vector with the smallest evaluation function value from the remaining motor-side candidate voltage vectors.