AC induction motor prediction control method based on gradient descent

Abstract

The invention discloses an AC induction motor prediction control method based on gradient descent, and the method comprises the following steps: building a mathematical model of an AC induction motor, employing a sliding model reference adaptive MRAS observer, employing a gradient descent iteration method, employing the gradient descent iteration method as an adaptive law, and carrying out the iteration in the reverse direction of the gradient of a quantization error, the method comprises the following steps: searching a minimum value of a quantization error, performing model prediction torque control, obtaining a switching state of an upper bridge arm switching tube of each phase in an inverter in a driving system of the alternating current induction motor, selecting a switching state which minimizes a cost function, and outputting the switching state to the inverter, so as to accurately and quickly minimize the error, and improve the reliability of the inverter. And current tracking performance similar to carrier waves is realized.

Description

Technical Field

[0001] The present invention relates to the field of motor control, and more particularly to a predictive control method for an AC induction motor based on gradient descent. Background Art

[0002] An AC induction motor is a device that converts alternating current into mechanical energy and is widely used in industries, household appliances, transportation, and other fields.

[0003] The control method of an AC induction motor can adopt sensorless control, such as MRAS model control or sliding mode observer control. Its working principle is to estimate the speed / position through a motor model and algorithm, eliminating the encoder, which has the advantages of low cost and high reliability. Among them, the core of MRAS model control is to dynamically adjust system parameters through the error between the reference model and the adjustable model, so that the actual system output approaches the ideal reference model. However, it is difficult to accurately minimize the error between MRAS models in the existing MARS model control.

[0004] In view of this, the present application conducts in-depth research on this basis, and thus this case is generated. Summary of the Invention

[0005] The object of the present invention is to provide a predictive control method for an AC induction motor based on gradient descent, which can quickly and accurately minimize the error between models.

[0006] To achieve the above object, the solution of the present invention is: A predictive control method for an AC induction motor based on gradient descent. The drive system of the AC induction motor includes an inverter. The inverter has three-phase bridge arms, and the three-phase bridge arms correspond to the a-phase, b-phase, and c-phase respectively. Each phase is provided with two switching tubes, and the two switching tubes in each phase bridge arm are respectively located in the upper bridge arm and the lower bridge arm. The method includes the following steps: Step S1: Establish a mathematical model of the AC induction motor, perform mathematical modeling on the stator and rotor fluxes, stator and rotor currents, stator and rotor inductances, and rotor parameters of the AC induction motor to obtain the mathematical model of the AC induction motor; Step S2: Adopt a sliding model reference adaptive (MRAS) observer, obtain the adjustable model and the reference model in the MRAS observer according to the mathematical model in Step S1, and correct the error between the adjustable model and the reference model through the sliding surface in the sliding model; Step S3: Gradient descent iteration method, adopt the gradient descent iteration method as the adaptation law, perform iteration along the opposite direction of the gradient of the quantization error to find the minimum value of the quantization error; Step S4, Model Predictive Torque Control: The predicted speed value is output by the MRAS observer. The error between the actual speed value output by the AC induction motor and the predicted speed value is calculated through a PI regulator to obtain the electromagnetic torque reference value, and the cost function is calculated by predicting the tracking deviation and reference value between the stator flux and torque. Step S5, Obtain the switching state of the upper-bridge arm switch tube of each phase in the inverter, select the switching state that minimizes the cost function, and output it to the inverter.

[0007] In step S1, the established mathematical model of the AC induction motor is as follows: (1) (2) (3) (4), In the formula, represents the stator voltage vector, , respectively represent the stator flux and rotor flux, , respectively represent the stator current and rotor current, and respectively represent the stator resistance and rotor resistance, represents the speed, , and respectively represent the stator inductance, rotor inductance and mutual inductance.

[0008] In step S2, the designed MRAS observer model is as follows: (5), (6), (7), (8), In the formula, represents the predicted value of the rotor voltage flux, represents the predicted value of the rotor current flux, represents the output voltage of the inverter, is the predicted speed value, , , , respectively represent and In , The component on the axis, j represents the imaginary number, σ is the magnetic leakage coefficient of the motor, and s represents the error between the reference model and the adjustable model.

[0009] In the fixed reference frame, select the voltage model in formula (5) as the reference model, and select the current model in formula (6) as the adjustable model; In formula (6), As the sliding mode surface, Represents a continuous function, and the formula is (9), where α Represents the sliding mode parameter, x represents the independent variable of the function, Represents the coefficient that makes up the function; In formula (5), Represents the stator resistance compensation, and the calculation formula is (10), where K p 、K i Represent the proportional coefficient and the integral coefficient respectively, Represents the stator resistance error, and the calculation formula of the stator resistance error is (11), in formula (11) 、 Represent the components of the stator current on the α and β axes respectively.

[0010] In step S3, the gradient descent iteration method is as follows, Step S3-1: Establish the objective function of gradient iteration as follows: J = (12), where s is obtained through formula (7); Step S3-2: Obtain the objective function J For the predicted value of the rotational speed Gradient , the calculation formula is as follows, (13), (14), (15), Where Represents the predicted value of the rotor current flux linkage Gradient of the predicted value of the rotational speed , Represents the error s between the reference model and the adjustable model for the predicted value of the rotational speed Gradient; And Represent the component gradients of the rotor flux linkage on the α and β axes respectively; Step S3-3: Adopt The iterative formula is used to obtain the rotational speed prediction value, and the iterative formula is (16), where is the observer gain greater than 0; Iteration is performed using formula (16) until the value of, so that the error s converges, and the rotational speed prediction value is obtained.

[0011] In step S4, model predictive torque control is as follows Step S4-1: The error between the rotational speed prediction value obtained in step S3-3 and the actual rotational speed value ω output by the AC induction motor is calculated by a PI regulator to obtain the electromagnetic torque reference value ; Step S4-2: The calculation formulas for predicting the stator flux linkage, predicted current, and predicted torque are as follows (17), (18), (19), where represents the stator flux linkage, represents the rotor flux linkage, T represents the electromagnetic torque, represents the stator flux linkage at time k + 1, represents the predicted value of the stator flux linkage at time k + 1, v s represents the stator voltage, represents the stator voltage at time k, represents the stator current at time k, represents the predicted value of the stator current at time k + 1, represents the predicted rotational speed value at time k, k r represents the rotor coupling coefficient, represents the rotor time constant, represents the predicted value of the rotor flux linkage at time k; represents the sampling period, p represents the number of pole pairs, represents the predicted value of the electromagnetic torque, represents the predicted value of the electromagnetic torque at time k + 1, represents the imaginary part of the complex number; Among them, , R σ represents the equivalent resistance; , , stator transient time ; Step S4-3: Establish the cost function as follows (20), Represents the value of the cost function in the case of switch state j; in the formula, Represents the reference value of the stator flux linkage, which is directly input by the AC induction motor; Is the weighting factor for modifying the torque and flux linkage values according to the working conditions; Represents the predicted torque value at time k + 2, Represents the predicted stator flux linkage value at time k + 2.

[0012] In step S5, select the cost function that minimizes the cost function , and obtain the cost function The corresponding switch state j, and transmit it to the inverter. The inverter switches the switch state of the upper bridge arm switch of each phase according to the switch state.

[0013] Step S4-4, parameter optimization, compare the predicted torque value and the predicted flux linkage value in the k + 2 cycles with the reference values in the cost function for optimization. The optimization formula is, (21), in the formula, Represents the maximum current limit term at time k + 2, γ represents a value much greater than zero, i max Represents the set maximum current value.

[0014] The Clarke transformation is used to obtain the voltage vector , and the formula is as follows, (22), In the formula, Is the DC voltage, Represents the switch state, , And Respectively correspond to the switch states of the upper bridge arm switches of phases a, b, and c in the two-level three-phase inverter, = 0, = 0 and = 0 respectively represent that the upper bridge arm switches of phases a, b, and c in the inverter are turned on, = 1, = 1 and = 1 respectively represent that the upper bridge arm switches of phases a, b, and c in the inverter are turned off.

[0015] According to the switch states of the upper bridge arm switches of the three phases in the inverter, a three-bit binary number is obtained, and the control input j is used as the input variable, where j represents the existing state of the switch state.

[0016] After adopting the above method, the present invention has the following beneficial effects: 1. The present invention processes the error between the reference model and the adjustable model in the MRAS observer through the gradient descent iteration method, selects an appropriate MRAS observer gain, and makes it perform a descent iteration in the opposite direction of the gradient, so as to accurately and quickly minimize the error, make the control process more accurate, and obtain an ideal parameter result.

[0017] 2. The present invention adopts a sliding mode surface reference adaptive MRAS observer of the sliding mode model, and corrects it according to the error between the adjustable model and the reference model. Compared with the traditional MRAS observer, the MRAS designed by the present invention has stronger robustness and easier tunability.

[0018] 3. In the model predictive torque control of the present invention, k the predicted torque values and flux linkage predicted values in the +2 cycles are respectively compared with the reference values in the cost function for optimization, so as to compensate for the time delay in the model predictive torque control. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is the circuit topology diagram of the AC induction motor and the two-level inverter in the present invention.

[0020] Figure 2 is the control principle block diagram of the MRAS observer in the present invention.

[0021] Figure 3 is the control principle block diagram of the model predictive torque control in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] In order to further explain the technical solution of the present invention, the present invention will be elaborated in detail through specific embodiments below.

[0023] A predictive control method for an AC induction motor based on gradient descent, which is based on a common drive system for AC induction motor or permanent magnet motor control, such as Figures 1-3 shown, the drive system includes an inverter and a conventional two-level inverter topology. The inverter adopts a conventional two-level three-phase inverter. The inverter includes three-phase bridge arms, and the three-phase bridge arms respectively correspond to the following a-phase, b-phase, and c-phase, and each phase is provided with two switching tubes. The two switching tubes in each phase bridge arm are respectively located in the upper bridge arm and the lower bridge arm.

[0024] In addition, the above drive system also includes a flux linkage estimation module, a torque and flux linkage prediction module, and a PI regulator, and all its input quantities can be obtained through conventional means in the art or the following methods, so no further description will be given.

[0025] In this embodiment, the predictive control method includes the following steps.

[0026] Step S1: Establish the mathematical model of an AC induction motor: Mathematical modeling is performed on the stator and rotor magnetic fluxes, stator and rotor currents, stator and rotor inductances, and rotor parameters of the AC induction motor to obtain the mathematical model of the AC induction motor.

[0027] Specifically, the established mathematical model of the AC induction motor (i.e., the IM model) is as follows. (1) (2) (3) (4), wherein, represents the stator voltage vector, , respectively represent the stator magnetic flux and the rotor magnetic flux, , respectively represent the stator current and the rotor current, and respectively represent the stator resistance and the rotor resistance, represents the rotational speed, , and respectively represent the stator inductance, the rotor inductance, and the mutual inductance.

[0028] Step S2: Adopt a sliding model (SM) reference adaptive MRAS observer: Based on the mathematical model in Step S1, the adjustable model and the reference model in the MRAS observer are obtained. Among them, the voltage model is adopted as the reference model, and the current model is adopted as the adjustable model. Then, the error between the adjustable model and the reference model is corrected through the sliding surface in the sliding model (SM).

[0029] Specifically, as Figure 2 shown, the designed MRAS observer model is as follows. (5), (6), (7), (8), wherein, represents the predicted value of the rotor voltage magnetic flux, represents the predicted value of the rotor current magnetic flux, represents the output voltage of the inverter, is the predicted value of the rotational speed, , respectively represent at , The components on the axis and respectively represent the components on the and axis. Here, j represents the imaginary number, σ is the magnetic leakage coefficient of the motor, and s represents the error between the reference model and the adjustable model. That is, the error between the reference model and the adjustable model can be obtained through formula (7).

[0030] Furthermore, in this embodiment, in the fixed reference frame, the voltage model in formula (5) is selected as the reference model in the MRAS observer, and the current model in formula (6) is selected as the adjustable model in the MRAS observer.

[0031] Furthermore, to mitigate the chattering problem, in the above formula (6) is used as the sliding surface in the sliding mode, represents a continuous function, and its calculation formula is (9). In formula (9), α represents the sliding mode parameter, x represents the independent variable of the function, represents the coefficient constituting the function.

[0032] In the above formula (5), represents the stator resistance compensation, and its calculation formula is (10). In the formula, K p and K i respectively represent the proportional coefficient and the integral coefficient. Among them, represents the stator resistance error, and this stator resistance error is calculated through the following formula. The calculation formula is (11). In formula (11), and respectively represent the components of the stator current on the α and β axes. It should be noted that the stator resistance error obtains the stator resistance compensation through the feedback of a conventional PI regulator .

[0033] In this way, the above MRAS observer is designed as an adaptive mechanism, and the error term of the MRAS observer is used as the sliding surface to improve the accuracy of flux linkage prediction.

[0034] Step S3: Adopt the gradient descent iteration method. To minimize the prediction deviation, the gradient descent iteration method is adopted to replace the traditional PI regulator as the adaptive law, and iteration is carried out along the opposite direction of the gradient of the quantization error to find the minimum value of the quantization error. The specific gradient descent iteration method is as follows.

[0035] As Figure 2 shown, step S3-1: Establish the objective function for gradient iteration. The objective function is as follows:J = (12). In Equation (12), s is the error between the reference model and the adjustable model of the MRAS observer in step S2. This error s can be obtained through Equation (7), that is, the error s is incorporated into the objective function J of the gradient iteration.

[0036] Step S-2: Obtain the objective function J For the predicted rotational speed of the gradient : The predicted rotational speed as a parameter, respectively calculate the predicted values of the rotor current flux linkage For the predicted rotational speed of the gradient and the error s between the reference model and the adjustable model in the MRAS observer for the predicted rotational speed of the gradient , thus calculating the gradient . The specific calculation formula is as follows.

[0037] (13). (14). (15). In the formula, represents the predicted value of the rotor current flux linkage for the predicted rotational speed of the gradient, represents the error s between the reference model and the adjustable model for the predicted rotational speed of the gradient, and respectively represent the component gradients of the rotor flux linkage on the α and β axes.

[0038] Step S3-3: Use the iteration formula to obtain the predicted rotational speed. The iteration formula is (16). In the formula, is the observer gain greater than 0; use Equation (16) for iteration until the value of makes the error s converge, and obtain the predicted rotational speed .

[0039] For example, first determine an initial value, and then continuously calculate according to Equation (16). After each calculation, replace the original value with the newly calculated value. Stop calculating when the newly calculated value is equal to the original value; among them, in Equation (16), when the gradient is equal to 0, it is the final result of the iteration.

[0040] It should be noted that the observer gain being too large may cause oscillations, and being too small may result in a slow convergence rate. Therefore the value of still needs to be accurately selected, that is, the selection of

[0041] needs to be tested according to the actual situation of the experimental platform.

[0042] As Figure 3 shown, Step S4, Model Predictive Torque Control (PTC): Select torque and stator flux linkage as the control objects, output the predicted speed value through the MRAS observer, calculate the electromagnetic torque reference value through the error between the actual speed value output by the AC induction motor and the predicted speed value, and calculate the cost function through the tracking deviation and reference value between the predicted stator flux linkage and torque. The specific control is as described below. The error between the predicted speed value obtained in Step S3-3 and the actual speed value ω output by the AC induction motor is calculated through the PI regulator to obtain the electromagnetic torque reference value

[0043] Step S4-2, The calculation formulas for predicting stator flux linkage, predicted current, and predicted torque are as follows.

[0044] (17), (18), (19), In the formula, represents the stator flux linkage, represents the rotor flux linkage, T represents the electromagnetic torque, represents the stator flux linkage at the (k + 1)th moment, represents the predicted value of the stator flux linkage at the (k + 1)th moment, represents the predicted value of the rotor flux linkage at the kth moment, v s represents the stator voltage, represents the stator voltage at the kth moment, represents the stator current at the kth moment, represents the predicted value of the stator current at the (k + 1)th moment, represents the predicted speed value at the kth moment; k r represents the rotor coupling coefficient, that is, it represents the ratio of the mutual inductance and the rotor inductance; represents the sampling period, p represents the number of pole pairs, represents the predicted value of the electromagnetic torque, represents the predicted value of the electromagnetic torque at the (k + 1)th moment, represents the imaginary part of the complex number; represents the rotor time constant.

[0045] Furthermore, in formula (18), R σ represents the equivalent resistance, which is usually related to the resistances of the stator and rotor, and its calculation formula is .

[0046] Furthermore, in formula (18), , , the stator transient time .

[0047] Step S4-3: Establish a cost function, which is calculated based on the tracking deviation between the predicted stator flux linkage and torque, its stator flux linkage reference value, and torque reference value. The calculation formula of the cost function is as follows: (20), where represents the value of the cost function in the case of switching state j; represents the stator flux linkage reference value, which is directly input by the AC induction motor; is a weighting factor for modifying the torque and flux linkage values according to the working conditions; represents the predicted torque value at time k + 2, represents the predicted stator flux linkage value at time k + 2.

[0048] As a preferred method, add time-delay compensation after step S4-3. Specifically, compare the predicted torque value in the k + 2 cycle with the torque reference value of the cost function and the predicted flux linkage value in the k + 2 cycle with the flux linkage reference value of the cost function respectively for optimization, and output the switching state j at this time. Then, calculate the voltage vector according to the output of the inverter in this switching state, that is, the following formula (22); among them, the optimization formula is (21), where represents the maximum limit term of the current at time k + 2, γ represents a value much greater than zero, i max represents the set maximum current value.

[0049] Step S5: Obtain the switching state of the upper-bridge-arm switch tube in each phase of the inverter: Select the switching state j that minimizes the cost function in step S4 and input it into the inverter, and the inverter controls the conduction or cutoff of the upper-bridge-arm switch tube.

[0050] The calculation of the above voltage vector is as follows. The voltage vector is obtained by using the Clarke transformation .

[0051] (22), where is the DC voltage, represents the switching state, , and respectively correspond to the switching states of the upper-bridge-arm switching transistors of phase a, phase b, and phase c in a two-level three-phase inverter. = 0, = 0, and = 0 respectively indicate that the upper-bridge-arm switching transistors of phase a, phase b, and phase c in the inverter are conducting. = 1, = 1, and = 1 respectively indicate that the upper-bridge-arm switching transistors of phase a, phase b, and phase c in the inverter are turned off.

[0052] Furthermore, according to the switching states of the upper-bridge-arm switching transistors of the three phases in the above inverter, a three-bit binary number is obtained. Using the control input j as the input variable, this j is the j in , and j represents the existing state of the switching state, that is, it represents the j-th switching state; in this embodiment, there are 8 switching states, j = 1, 2, 3, 4, 5, 6, 7, or 8, and the value of S abc respectively corresponds to j at this time. For example, when j = 4, at this time S abc = [0, 1, 1]; where represents the value of the cost function in the case of the switching state j, which means selecting the with the smallest value, and the corresponding j is the optimal switching state. Similarly, the voltage vector v is calculated by formula (22) and is also a quantity that changes with j.

[0053] The above is only the preferred embodiment of this embodiment. All equivalent changes and modifications made to the scope of the claims of the present invention shall fall within the scope of the claims of the present invention.

Claims

1. A predictive control method for an AC induction motor based on gradient descent. The drive system of the AC induction motor includes an inverter. The inverter has three-phase bridge arms, and the three-phase bridge arms correspond to the a-phase, b-phase, and c-phase respectively. Each phase is provided with two switching tubes, and the two switching tubes in each phase bridge arm are respectively located in the upper bridge arm and the lower bridge arm. It is characterized in that, The steps are as follows: Step S1: Establish a mathematical model of an AC induction motor, perform mathematical modeling on the stator and rotor fluxes, stator and rotor currents, stator and rotor inductances, and rotor parameters of the AC induction motor to obtain the mathematical model of the AC induction motor; Step S2: Adopt a sliding-mode model reference adaptive MRAS observer, obtain the adjustable model and the reference model in the MRAS observer according to the mathematical model in Step S1, and correct the error between the adjustable model and the reference model through the sliding surface in the sliding mode; Step S3: Gradient descent iterative method, use the gradient descent iterative method as the adaptation law, iterate along the opposite direction of the gradient of the quantization error to find the minimum value of the quantization error; Step S4: Model predictive torque control, output the predicted speed value through the MRAS observer, calculate the electromagnetic torque reference value through the error between the actual speed value output by the AC induction motor and the predicted speed value by a PI regulator, and calculate the cost function through the tracking deviation and reference value between the predicted stator flux and torque; Step S5: Obtain the switching states of the upper-bridge-arm switching tubes of each phase in the inverter, select the switching state that minimizes the cost function, and output it to the inverter.

2. The predictive control method for an AC induction motor based on gradient descent according to claim 1, wherein: In Step S1, the established mathematical model of the AC induction motor is as follows, （1） （2） （3） （4）， In the formula, represents the stator voltage vector, , represent the stator flux linkage and the rotor flux linkage respectively, , represent the stator current and the rotor current respectively, and represent the stator resistance and the rotor resistance respectively, represents the rotational speed, , and represent the stator inductance, the rotor inductance and the mutual inductance respectively.

3. The predictive control method for an AC induction motor based on gradient descent according to claim 2, characterized in that: In Step S2, the designed MRAS observer model is as follows, （5）， （6）， （7）， （8）， In the formula, represents the predicted value of the rotor voltage flux linkage, represents the predicted value of the rotor current flux linkage, represents the output voltage of the inverter, is the predicted value of the rotational speed, 、 、 、 respectively represent and on the 、 axis components, j represents the imaginary number, σ is the magnetic leakage coefficient of the motor, and s represents the error between the reference model and the adjustable model.

4. The predictive control method for an AC induction motor based on gradient descent according to claim 3, wherein: In the fixed reference frame, select the voltage model in Equation (5) as the reference model, and select the current model in Equation (6) as the adjustable model; In formula (6), As the sliding mode surface, represents a continuous function, and the formula is (9), where α represents the sliding mode parameter, x represents the independent variable of the function, represents the coefficient that constitutes the function; In Formula (5), represents stator resistance compensation, and its calculation formula is (10), where K p and K i represent the proportional coefficient and integral coefficient respectively, represents the stator resistance error, and the calculation formula of the stator resistance error is (11). In formula (11), and represent the components of the stator current on the α-axis and β-axis respectively.

5. The predictive control method for an AC induction motor based on gradient descent according to claim 3 or 4, characterized in that: The gradient descent iterative method in Step S3 is as follows, Step S3-1: Establish the objective function for gradient iteration as follows: J = (12), where s is obtained through formula (7); Step S3-2, obtain the objective function J For the rotational speed prediction value Gradient of , the calculation formula is as follows （13）， （14）， （15）， In the formula, represents the predicted value of the rotor current flux linkage is the gradient of the predicted value of the rotational speed ; represents the gradient of the error s between the reference model and the adjustable model with respect to the predicted value of the rotational speed ; and respectively represent the component gradients of the rotor flux linkage on the α and β axes; Step S3-3: Use 's iterative formula to obtain the rotational speed prediction value. The iterative formula is (16), where is the observer gain greater than 0; Iterate using formula (16) until 's value makes the error s converge, and obtain the rotational speed prediction value .

6. The predictive control method for an AC induction motor based on gradient descent according to claim 5, characterized in that: In Step S4, the model predictive torque control is as follows, Step S4-1: Calculate the electromagnetic torque reference value through a PI regulator based on the error between the rotational speed prediction value obtained in step S3-3 and the actual rotational speed value ω output by the AC induction motor ; In Step S4-2, the calculation formulas for the predicted stator flux, predicted current, and predicted torque are as follows, （17）， （18）， （19）， In the formula, represents the stator flux linkage, represents the rotor flux linkage, T represents the electromagnetic torque, represents the stator flux linkage at the (k + 1)th moment, represents the predicted value of the stator flux linkage at the (k + 1)th moment, v s represents the stator voltage, represents the stator voltage at the kth moment, represents the stator current at the kth moment, represents the predicted value of the stator current at the (k + 1)th moment, represents the predicted value of the rotational speed at the kth moment, k r represents the rotor coupling coefficient, represents the rotor time constant, represents the predicted value of the rotor flux linkage at the kth moment; denotes the sampling period, p denotes the number of pole pairs, denotes the predicted value of the electromagnetic torque, denotes the predicted value of the electromagnetic torque at time k + 1, denotes the imaginary part of a complex number; where, , R σ denotes the equivalent resistance; , , the stator transient time ; Step S4-3: Establish the cost function as follows, (20), which represents the value of the cost function in the case of switch state j; in the formula, represents the reference value of the stator flux, which is directly input by the AC induction motor; is the weight factor for modifying the torque and flux values according to the working conditions; represents the predicted torque value at time k + 2, represents the predicted stator flux value at time k + 2.

7. The predictive control method for an AC induction motor based on gradient descent according to claim 6, characterized in that: In step S5, select the cost function that minimizes the cost function , and obtain the cost function corresponding switching state j, and transmit it to the inverter. The inverter switches the switching state of the upper-bridge-arm switching tubes of each phase according to the switching state.

8. The predictive control method for an AC induction motor based on gradient descent according to claim 6, characterized in that: Step S4-4, parameter optimization, compare the torque prediction value and the flux linkage prediction value in the k + 2 cycles with the reference values in the cost function for optimization. The optimization formula is (21), where represents the maximum limit term of the current at time k+2, γ represents a value much larger than zero, and i max represents the set maximum current value.

9. The predictive control method for an AC induction motor based on gradient descent according to claim 7, wherein: The voltage vector is obtained by using the Clarke transformation , and the formula is as follows （22）， Wherein, is the DC voltage, represents the switch state, , and respectively correspond to the switch states of the upper-arm switching transistors of phase a, phase b, and phase c in the two-level three-phase inverter. = 0, = 0, and = 0 respectively represent that the upper-arm switching transistors of phase a, phase b, and phase c in the inverter are turned on. = 1, = 1, and = 1 respectively represent that the upper-arm switching transistors of phase a, phase b, and phase c in the inverter are turned off.

10. The predictive control method for an AC induction motor based on gradient descent according to claim 9, characterized in that: Obtain a three - bit binary number according to the switching states of the upper - bridge - arm switching transistors of the three phases in the inverter , and use the control input j as the input variable, where j represents the existing state of the switching state.

Images