A time-domain adaptive multi-step predictive control method for permanent magnet synchronous motor

Abstract

The present invention provides a time-domain adaptive multi-step predictive control method for a permanent magnet synchronous motor, comprising: constructing a current loop control algorithm based on predicted current, judging whether the current sampling period is the first sampling period in the control time domain, and if so, using a hysteresis loop to judge the current working state, if the current system is working in a transient state, updating the predicted time domain and the control time domain, if the system enters a steady state, the current time domain maintains the previous state, and the predicted time domain and the control time domain remain unchanged; performing multi-step model prediction and cost function calculation, and selecting the optimal vector; otherwise, the control vector keeps the optimal vector selected in the first period unchanged, and determines the weight factor online according to the determined control vector. The present invention realizes full parameter control, solves the problem of difficult setting of the weight factor of the permanent magnet synchronous motor model predictive control strategy, and also improves the control performance of the permanent magnet synchronous motor system during actual operation.

Description

Technical Field

[0001] The present invention relates to the technical field of permanent magnet synchronous motors, and in particular to a time-domain adaptive multi-step predictive control method for permanent magnet synchronous motors. Background Art

[0002] Permanent magnet synchronous motor (PMSM) is widely used in drive devices due to its many advantages such as high efficiency and high power density. Common algorithms used to control permanent magnet synchronous motors mainly include fuzzy control, adaptive control, predictive control, etc., and generally only achieve automatic adjustment by controlling weight parameters. Among them, predictive control (MPC) uses the precise mathematical model of the controlled object to predict the voltage vector that should be applied in the next control cycle, so that theoretically after one cycle of the voltage vector, the motor current can accurately follow the command current value. This model predictive control can make the motor current obtain good dynamic and steady-state response within a certain range, but there are also certain problems. For example, after a period of use, its stability will also decrease. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a time-domain adaptive multi-step predictive control method for a permanent magnet synchronous motor, solve the difficulty in adjusting the weight factors of the permanent magnet synchronous motor model predictive control strategy, and also improve the control performance of the permanent magnet synchronous motor system during actual operation.

[0004] The present invention is implemented as follows: a time-domain adaptive multi-step predictive control method for a permanent magnet synchronous motor comprises the following steps:

[0005] Step S1, construct a current loop control algorithm based on predicted current, determine which sampling cycle is currently in the control time domain, if it is the first sampling cycle in the control time domain, proceed to step S2, otherwise proceed to step S4;

[0006] Step S2: Use the hysteresis loop to judge the current working state. If the current system is working in a transient state, the prediction time domain and the control time domain are updated. If the system enters a steady state, the current time domain maintains the previous state, and the prediction time domain and the control time domain remain unchanged.

[0007] Step S3, perform multi-step model prediction and cost function calculation, and select the optimal vector;

[0008] Step S4, the control vector keeps the optimal vector selected in the first cycle unchanged, and according to the determined optimal vector, gradually selects weight factors in other sampling cycles in the control time domain until the last sampling cycle in the control time domain, and updates the weight factors in the model prediction process, thereby affecting the selection of the optimal vector.

[0009] Further, "judging the current working state by using a hysteresis loop" in step S2 specifically includes: when the rotational speed error rises and is greater than the high threshold, it is determined that the current system is working in a transient state; when the rotational speed error drops and is less than the low threshold, it is determined that the system enters a steady state.

[0010] Further, the prediction horizon and control horizon in step S2 are specifically obtained by using the following formulas:

[0011]

[0012]

[0013] where PH(k) and CH(k) are the prediction horizon step length and control horizon step length at the k-th sampling period respectively, ΔPH is the prediction horizon change step length, ΔCH is the control horizon change step length; steady-state is the steady state, transient-state#1 is the transient state 1, transient-state#2 is the transient state 2; among them, when the rotational speed error rises and is greater than the high threshold, it is the transient state 1, when the rotational speed error drops and is greater than the low threshold, it is the transient state 2, and in other cases it is the steady state;

[0014] The prediction horizon is obtained by multiplying the time T by the prediction horizon step length PH(k);

[0015] The control horizon is equal to the time T multiplied by the control horizon step length CH(k).

[0016] Further, the prediction horizon change step length ΔPH and the control horizon change step length ΔCH are specifically obtained in the following manner:

[0017]

[0018] ΔCH = C1ΔPH;

[0019] where C and C1 are adjustment coefficients, satisfying the condition 0 < C1 ≤ 1, C is a positive integer, Eisq and Eisd respectively represent the errors of the q-axis component and d-axis component of the stator current, sgn represents the sign function, and n is a positive integer greater than 1.

[0020] Further, in step S3, it is specifically as follows:

[0021] Step S31: Set the discrete-time model of the permanent magnet synchronous motor as:

[0022]

[0023] where, u sd and u sq are the voltages in the d-q coordinate axes respectively; i sd 、isq are the currents in the dq coordinate axes respectively; L sd , L sq are the stator inductance parameters under the dq coordinate axis; R s is the stator resistance; ψ f is the permanent magnet flux; ω r is the electrical angular velocity, T s is the sampling time;

[0024] Step S32: Design the cost function as follows:

[0025]

[0026] Among them, J is the cost function value, Q j is a set of weights that converge to zero, λ is the optimal weight, is the reference value of isq;

[0027] The first and second terms in the brackets in the cost function are the given values ​​corresponding to the tracking of the q-axis and d-axis stator current components, respectively;

[0028] Step S33, according to the prediction time domain step PH(k), the discrete time model calculation and cost function calculation are executed PH(k) times in a loop, and the optimal weights calculated in the previous control time domain are brought in to calculate the cost function value J of each vector respectively. The vector with the smallest cost function value is taken as the optimal vector. The switch state corresponding to the optimal vector will act on the motor driver to adjust the working state of the motor.

[0029] Furthermore, in step S4, "updating weight factors in the model prediction process" is specifically as follows: using online bisection in each sampling period, based on the optimal vector determined in the first sampling period in the control time domain, the weight factors are respectively introduced into the cost function to determine the optimal value, firstly the maximum and minimum values ​​in the test range are introduced for comparison, and the weight with a smaller cost function value is compared with the cost function value substituted with the intermediate value to determine the smaller value of the two, and so on, and the optimal weight with high precision is selected as much as possible under limited processor resources.

[0030] Furthermore, the optimal vector is an optimal voltage vector.

[0031] The present invention has the following advantages: in the design of model predictive control strategy, it fully utilizes processor resources, increases the control time domain and weight factors without increasing the computational burden, realizes the full control parameter adaptive tuning control strategy for the first time, and adopts the online dichotomy method to design the weight factors, and determines the local optimal value under limited computational burden. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0033] Figure 1 The present invention is a flowchart of the execution of a time-domain adaptive multi-step predictive control method for a permanent magnet synchronous motor.

[0034] Figure 2 A schematic diagram of the main control algorithm principle of a time-domain adaptive multi-step predictive control method for a permanent magnet synchronous motor according to the present invention;

[0035] Figure 3 Schematic diagram of the principle of time flow in an embodiment of the present invention;

[0036] Figure 4 Schematic diagram of the structure of the online dichotomy method in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The overall idea of ​​the embodiment of the present invention is as follows: first, the current working state is determined based on a finite control set predictive current control strategy. If in a transient control process, the degree of system deviation from the steady state is quantified and the prediction and control time domains are designed to improve system stability and convergence radius; secondly, in the process of maintaining a state where the control time domain is greater than 1, an online dichotomy is designed to determine the weight factor based on the optimal voltage vector, and the local optimal weight is obtained while avoiding the problem of generating a large amount of computational burden.

[0038] like Figures 1 to 4 As shown, the present invention provides a time-domain adaptive multi-step predictive control method for a permanent magnet synchronous motor, comprising the following steps:

[0039] Step S1: construct a current loop control algorithm based on predicted current, such as Figure 1 As shown, the variable x is used to determine which sampling period the current control time domain is in. If it is the first sampling period in the control time domain, the process proceeds to step S2, otherwise, the process proceeds to step S4;

[0040] Step S2: Use the hysteresis loop to judge the current working state. If the current system is working in a transient state, the prediction time domain and the control time domain are updated. If the system enters a steady state, the current time domain maintains the previous state, and the prediction time domain and the control time domain remain unchanged.

[0041] Step S3, performing multi-step model prediction and cost function calculation, and selecting an optimal vector; the optimal vector is an optimal voltage vector;

[0042] Step S4, the control vector keeps the optimal vector selected in the first cycle unchanged, and according to the determined optimal vector, gradually selects weight factors in other sampling cycles in the control time domain until the last sampling cycle in the control time domain, updates the weight factors in the model prediction process, thereby affecting the selection of the optimal vector, and then controls the motor according to the switching state corresponding to the output voltage vector.

[0043] In a preferred embodiment, "judging the current working state by using a hysteresis loop" in the step S2 is specifically as follows: when the speed error rises and is greater than the high threshold, it is determined that the current system is working in a transient state; when the speed error drops and is less than the low threshold, it is determined that the system enters a steady state.

[0044] In a preferred embodiment, the prediction horizon and the control horizon in the step S2 are specifically obtained by the following formulas:

[0045]

[0046]

[0047] wherein, PH(k) and CH(k) are respectively the prediction horizon step length and the control horizon step length of the k-th sampling period, ΔPH is the prediction horizon change step length, ΔCH is the control horizon change step length, and the change step length can be set as a fixed value or a variable; steady-state is the steady state, transient-state#1 is the transient state 1, transient-state#2 is the transient state 2; wherein, when the speed error rises and is greater than the high threshold, it is the transient state 1, when the speed error drops and is greater than the low threshold, it is the transient state 2, and the rest of the cases are all in the steady state (including when the error rises below the high threshold and when the error drops below the low threshold).

[0048] The prediction horizon is obtained by multiplying the time T by the prediction horizon step length PH(k);

[0049] The control horizon is equal to the time T multiplied by the control horizon step length CH(k).

[0050] In a preferred embodiment, the prediction horizon change step length ΔPH and the control horizon change step length ΔCH are specifically obtained in the following manner:

[0051]

[0052] ΔCH = C1ΔPH;

[0053] wherein, C and C1 are adjustment coefficients, satisfying the condition 0 < C1 ≤ 1, C is a positive integer, E isq 、E isd respectively represent the errors of the q-axis component and the d-axis component of the stator current, sgn represents the sign function, and n is a positive integer greater than 1.

[0054] In a preferred embodiment, the step S3 is specifically as follows:

[0055] Step S31, set the discrete-time model of the permanent magnet synchronous motor as:

[0056]

[0057] Among them, u sd and u sq are the voltages on the dq coordinate axes respectively; i sd 、i sq are the currents in the dq coordinate axes respectively; L sd , L sq are the stator inductance parameters under the dq coordinate axis; R s is the stator resistance; ψ f is the permanent magnet flux; ω r is the electrical angular velocity, T s is the sampling time;

[0058] Step S32: Design the cost function as follows:

[0059]

[0060] Among them, J is the cost function value, Q j is a set of weights that converge to zero (known weights, generally 1 / (1+j)), λ is the optimal weight (calculated online), is the reference value of isq;

[0061] The first and second terms in the brackets in the cost function are the given values ​​corresponding to the tracking of the q-axis and d-axis stator current components, respectively;

[0062] Step S33, according to the prediction time domain step PH(k), the discrete time model calculation and cost function calculation are executed PH(k) times in a loop, and the optimal weights calculated in the previous control time domain are brought in to calculate the cost function value J of each vector respectively. The vector with the smallest cost function value is taken as the optimal vector. The switch state corresponding to the optimal vector will act on the motor driver to adjust the working state of the motor.

[0063] In a preferred embodiment, if Figure 4 As shown, the "update weight factors in the model prediction process" in step S4 is specifically as follows: using online binary search in each sampling period, based on the optimal vector determined in the first sampling period in the control time domain, respectively bring the weight factors into the cost function to determine the optimal value, first bring in the maximum and minimum values ​​in the test range for comparison, compare the weight with the smaller cost function value with the cost function value substituted with the intermediate value, and determine the smaller value of the two, and so on, select the optimal weight with high precision as much as possible under limited processor resources. Figure 4 , for Layer1, λ min and λ max is the optimal value of the boundary, and the result (λ min +λ max) / 2 is the local optimal value, and so on, a Layer operation is performed in each sampling period after the second sampling period in the control time domain (including the second sampling period), and the accuracy is gradually improved. The local optimal value obtained in the last sampling period is the optimal weight.

[0064] The technical solution provided in the embodiments of the present application has at least the following technical effects or advantages:

[0065] The control strategy design formula of the present invention quantifies the degree of deviation of the system from the steady state, and adjusts the system stability and convergence domain online according to the degree, so that the current working state does not exceed the convergence radius and enter an unstable state. In addition, the system can make full use of the maintenance time to determine the weight factor and realize the full control parameter setting under limited computing burden. Specifically, when the system enters the steady state, the prediction time domain is reduced and the control time domain is increased to improve the control accuracy and reduce the computing burden; when the system enters the transient state, the prediction time domain is increased and the control time domain is reduced to improve the system stability and convergence domain. When the control time domain is greater than 1, there is a cross-time domain maintenance phenomenon in which one or more sampling cycles are idle. The time period can be used to design the weight coefficient online based on the selected optimal vector to realize the cross-time domain weight factor setting. The present invention realizes the simultaneous online adaptive setting of all control parameters of the predictive control strategy for the first time, makes full use of the limited computing power of the processor, solves the problem of limited control performance of the control strategy under fixed control parameters, and improves the stability and control accuracy of the control strategy; at the same time, an online dichotomy is designed to determine the local optimal value of the weight factor online to improve the system control performance.

[0066] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A time-domain adaptive multi-step predictive control method for a permanent magnet synchronous motor, characterized in that: The steps include: Step S1, construct a current loop control algorithm based on predicted current, determine which sampling cycle is currently in the control time domain, if it is the first sampling cycle in the control time domain, proceed to step S2, otherwise proceed to step S4; Step S2: Use the hysteresis loop to judge the current working state. If the current system is working in a transient state, the prediction time domain and the control time domain are updated. If the system enters a steady state, the current time domain maintains the previous state, and the prediction time domain and the control time domain remain unchanged. Step S3, perform multi-step model prediction and cost function calculation, and select the optimal vector; Step S4: the control vector keeps the optimal vector selected in the first cycle unchanged, and according to the determined optimal vector, gradually selects weight factors in other sampling cycles in the control time domain until the last sampling cycle in the control time domain, and updates the weight factors in the model prediction process, thereby affecting the selection of the optimal vector; The prediction time domain and control time domain in step S2 are specifically obtained by using the following formula: ； ； Among them, PH(k) and CH(k) are the prediction time domain step size and control time domain step size of the kth sampling period, respectively, ΔPH is the prediction time domain change step size, and ΔCH is the control time domain change step size; For steady state, is transient 1, It is transient state 2; when the speed error rises and is greater than the high threshold, it is transient state 1, when the speed error drops and is greater than the low threshold, it is transient state 2, and the rest are steady state; The prediction time domain is obtained by multiplying the time T by the prediction time domain step length PH(k); The control time domain is equal to time T multiplied by the control time domain step length CH(k); The predicted time domain change step size ΔPH and the controlled time domain change step size ΔCH are specifically obtained in the following manner: ； ； Where C and C1 are adjustment coefficients, satisfying the condition 0 < C1 ≤ 1, C is a positive integer, E isq 、E isd They represent the errors of the q-axis component and the d-axis component of the stator current respectively, sgn represents the sign function, and n is a positive integer greater than 1; The specific details of step S3 are as follows: Step S31, setting the discrete time model of the permanent magnet synchronous motor to: ； Among them, u sd and u sq are the voltages on the dq coordinate axes respectively; i sd 、i sq are the currents in the dq coordinate axes respectively; L sd , L sq are the stator inductance parameters under the dq coordinate axis; R s is the stator resistance; ψ f is the permanent magnet flux; ω r is the electrical angular velocity, T s is the sampling time; Step S32: Design the cost function as follows: ； Among them, J is the cost function value, is a set of weights that converge to zero, is the optimal weight, is the reference value of isq; The first and second terms in the brackets in the cost function are the given values ​​corresponding to the tracking of the q-axis and d-axis stator current components, respectively; Step S33: According to the prediction time domain step length PH(k), the discrete time model calculation and cost function calculation are executed PH(k) times in a loop, and the optimal weight calculated in the previous control time domain is brought in to calculate the cost function value J of each vector respectively. The vector with the smallest cost function value is taken as the optimal vector. The switch state corresponding to the optimal vector will act on the motor driver to adjust the working state of the motor. The optimal vector is an optimal voltage vector.

2. The method according to claim 1, characterized in that: In step S2, "using the hysteresis loop to judge the current working state" is specifically: when the speed error rises and is greater than the high threshold, it is determined that the current system is working in a transient state; when the speed error drops and is less than the low threshold, it is determined that the system enters a steady state.

3. The method according to claim 1, characterized in that: In step S4, "updating the weight factors in the model prediction process" is specifically as follows: using online binary search within each sampling period, based on the optimal vector determined in the first sampling period in the control time domain, the weight factors are respectively introduced into the cost function to determine the optimal value, firstly the maximum value and the minimum value in the test range are introduced for comparison, the weight with the smaller cost function value is compared with the cost function value substituted with the intermediate value, and the smaller value of the two is determined, and so on, the optimal weight with high precision is selected as much as possible under limited processor resources.

Images