A direct thrust control method for permanent magnet linear synchronous motor

Abstract

The present invention relates to a control method for direct thrust of a permanent magnet linear synchronous motor, belonging to the technical field of motors, the method comprising: obtaining an equivalent current and an equivalent voltage in an α-β coordinate system from the three-phase current and three-phase voltage of the permanent magnet linear synchronous motor; obtaining the thrust, flux and flux angle of the permanent magnet linear synchronous motor by using the equivalent current and equivalent voltage in the α-β coordinate system; constructing an ADHDP speed controller to obtain a thrust difference, and obtaining a flux difference by subtracting a given flux amplitude from the flux; determining the state of each switch of the inverter by signals output by a sector selection unit from the thrust difference, flux difference and flux angle, and generating a three-phase AC voltage to operate the motor. The direct thrust control method for a permanent magnet linear synchronous motor proposed by the present invention is a data-driven control method, which is optimized to a certain extent compared to the traditional PI controller with slow convergence speed and low error accuracy, thereby improving the control accuracy of the permanent magnet linear synchronous motor.

Description

Technical Field

[0001] The present invention relates to the technical field of motors, and in particular to a method for controlling direct thrust of a permanent magnet linear synchronous motor. Background Art

[0002] With the continuous strengthening of the mutual promotion and interdependence between cities and towns in my country, and the rapid expansion of urban core areas, it is becoming increasingly important to strengthen fast, safe and comfortable transportation connections between regions, thus putting forward more requirements for my country's rail transit industry.

[0003] The traditional rail transit industry uses rotating motors as power devices, relying on the adhesion between wheels and rails to drive trains. After long-term use and development, this driving method has become relatively mature. However, since this method uses the physical adhesion between wheels and rails, the speed, acceleration, and climbing performance of the vehicle will be subject to certain restrictions. In addition, since a rotating motor is used for driving, the rotational torque needs to be converted into the traction of the train through gear transmission. The existence of the transmission system will make the entire power system large and the system structure complex. Permanent magnet linear synchronous motor (PMLSM) has the characteristics of non-adhesion drive, simple structure and reliable performance. Moreover, unlike rotating motors, linear motors eliminate the intermediate transmission link and directly convert electrical energy into the required mechanical skills, so that the structure of the entire power system can be simplified and the volume can be reduced. Therefore, it is increasingly used in the rail transit industry.

[0004] At present, the control systems of permanent magnet linear synchronous motors are basically closed-loop control systems, and the control method basically adopts PI control. However, the control method of permanent magnet linear synchronous motors is easily affected by time-varying parameters and external unknown disturbances. However, most of the control methods in modern control theory are based on ideal mathematical models. When time-varying parameters occur, fluctuations may occur, making the control system unstable. Summary of the invention

[0005] In view of the defects of the prior art, the present invention proposes a method for controlling the direct thrust of a permanent magnet linear synchronous motor, which aims to reduce the technical problem of unstable control system caused by the influence of time-varying parameters on the control method of the permanent magnet linear synchronous motor.

[0006] The technical solution of the present invention is as follows:

[0007] A method for direct thrust control of a permanent magnet linear synchronous motor comprises the following steps:

[0008] S101: Measuring the three-phase current i of the permanent magnet linear synchronous motor a 、i b 、i c , three-phase voltage ua 、u b 、u c , perform Clark transformation to obtain the equivalent current i in the α-β coordinate system α 、i β And the equivalent voltage u α 、u β ;

[0009] S102: Using the equivalent current i in the α-β coordinate system α 、i β And the equivalent voltage u α 、u β Calculate the thrust fe and flux ψ of the permanent magnet linear synchronous motor s and the flux linkage angle θ s ;

[0010] S103: construct an ADHDP speed controller, and the thrust reference value fe output by the ADHDP speed controller * The thrust difference Δfe is obtained by subtracting the calculated thrust fe, and the given magnetic flux amplitude ψ s * The flux linkage ψ calculated in step S102 s Subtract and get Δψ s ;

[0011] S104: The thrust difference Δfe and the flux difference Δψ in step S103 are combined. s The signal obtained after the thrust and flux hysteresis regulator is related to the flux angle θ s The signals output by the sector selection unit are input to the voltage vector switch selection unit, which then selects the state of each switch of the inverter, so that the inverter generates a three-phase AC voltage to run the motor.

[0012] Furthermore, the construction method of the ADHDP speed controller in step S103 is as follows:

[0013] Two BP neural networks, namely the execution network and the evaluation network, are constructed, and the ADHDP speed controller is constructed using the execution network and the evaluation network.

[0014] Furthermore, both the execution network and the evaluation network in the ADHDP speed controller contain two hidden layers.

[0015] Furthermore, the hidden layers of the execution network and the evaluation network adopt bipolar sigmoidal functions, and the output layer adopts the linear function purelin.

[0016] Furthermore, the structure of the execution network is 3-12-12-1.

[0017] Furthermore, the structure of the evaluation network is 4-10-10-1.

[0018] Furthermore, the speed error e(k) and the variables e(k-1) and e(k-2) at the two previous moments of the speed error are used as the inputs of the execution network, and the thrust reference value fe * is the output of the execution network.

[0019] Furthermore, the velocity error e(k) and the variables e(k-1), e(k-2) of the velocity error two moments before and the thrust reference value fe output by the execution network are used. * As the input of the evaluation network, the estimated value of the system performance indicator function is the output.

[0020] Furthermore, the speed error e(k) is obtained by using the magnetic scale to obtain the motor speed v and comparing it with the expected speed v ref Subtract and get the velocity error e(k).

[0021] Furthermore, the learning rate l of the evaluation network and the execution network is a,c (k) are adaptive learning rates, as shown in the following formula:

[0022]

[0023] Among them, α and β are learning rate gains; l a,c The initial value of (k) is 0.1; E a,c (k) is the objective function of the execution network and the evaluation network.

[0024] Compared with the prior art, the present invention has the following advantages:

[0025] 1. The direct thrust control method of the permanent magnet linear synchronous motor proposed in the present invention is a data-driven control method. Compared with the traditional PI controller with slow convergence speed and low error accuracy, it is optimized to a certain extent and improves the control accuracy of the permanent magnet linear synchronous motor.

[0026] 2. The evaluation network and execution network in the direct thrust control method of the permanent magnet linear synchronous motor proposed in the present invention respectively adopt BP neural network, and design two hidden layers to improve the accuracy of approximation, and use adaptive learning rate to improve the convergence speed of the neural network and avoid the BP neural network falling into the local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the direct thrust control method of the permanent magnet linear synchronous motor of the present invention;

[0028] Figure 2 It is a block diagram of the direct thrust control system of the permanent magnet linear synchronous motor of the present invention;

[0029] Figure 3 This is the internal logic structure diagram of ADHDP;

[0030] Figure 4 To evaluate the network structure of the network;

[0031] Figure 5 The network structure for executing the network;

[0032] Figure 6 Tracking curve for the speed controller based on adaptive dynamic programming. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below with reference to the accompanying drawings.

[0034] Figures 1 to 6 As shown, the present invention provides a method for direct thrust control of a permanent magnet linear synchronous motor, which is specifically as follows.

[0035] A method for direct thrust control of a permanent magnet linear synchronous motor comprises the following steps:

[0036] Step 1: Measure the three-phase current i of the permanent magnet linear synchronous motor a 、i b 、i c Three-phase voltage u a 、u b 、u c , perform Clark transformation to obtain the equivalent current i in the α-β coordinate system α 、i β And the equivalent voltage u α 、u β ;

[0037] Step 2: Use the equivalent current i in the α-β coordinate system in step 1 α 、i β And the equivalent voltage u α 、u β Calculate the thrust fe and flux ψ of the permanent magnet linear synchronous motor s , and then through the magnetic flux component ψ in the α-β coordinate system α , β Perform inverse trigonometric tangent calculation to obtain the magnetic flux angle θ s , the specific expression is,

[0038]

[0039]

[0040] where p n is the number of motor pole pairs, τ is the pole pitch of the motor;

[0041] Step 3: Construct two BP neural networks, namely the execution network and the evaluation network to form the ADHDP speed controller. The control algorithm of the ADHDP speed controller is composed of the evaluation network updating the control strategy and the execution network adjusting the speed error based on the idea of ​​reinforcement learning. The thrust reference value fe output by the ADHDP speed controller * The thrust difference Δfe is obtained by subtracting the calculated thrust fe; at the same time, the given magnetic flux amplitude ψ s * The flux ψ calculated in step 2 s The difference in flux is Δψ s .

[0042] Step 4: Substitute the thrust difference Δfe and flux difference Δψ in step 3 s The signal obtained after the thrust and flux hysteresis regulator is input into the voltage vector switch selection unit together with the signal output by the sector selection unit of the flux angle, and then the state of each switch of the inverter is selected. The inverter generates a three-phase AC voltage to make the motor run stably.

[0043] The specific construction process of constructing the ADHDP speed controller in step 3 above is as follows:

[0044] Use the magnetic scale to obtain the motor speed v and compare it with the expected speed v ref The speed error e(k) is obtained by subtraction, and the values ​​of the speed error at the previous two moments e(k-1), e(k-2) are selected, and e(k), e(k-1), e(k-2) are used as the state variables of the ADHDP speed controller, and the thrust reference value fe output by the ADHDP speed controller is * as a control quantity.

[0045] The execution network and evaluation network in the ADHDP speed controller both contain two hidden layers. The specific description of the evaluation network and the execution network is as follows.

[0046] The evaluation network structure is 4-10-10-1, which selects the speed error and the variables of the speed error at the first two moments and the thrust reference value fe output by the speed controller. * As the input of the evaluation network, the estimated value of the system performance indicator function For output, the hidden layer of the evaluation network uses a bipolar sigmoidal function, and the output layer uses a linear function purelin.

[0047] Define the input vector of the evaluation network as:

[0048] x c (k)=[e(k),e(k-1),e(k-2),fe * (k)]

[0049] The process of forward calculation of the evaluation network is as follows:

[0050]

[0051] Where i, j represent the number of rows and columns of each matrix, j′ represents the number of columns of each neural network input vector, 1j (k), ch 2j (k), ch 3j (k), ch 4j (k) are ch 1 (k), ch 2 (k), ch 3 (k), ch 4 (k) Elements in ch 1j (k), ch 2j (k), ch 3j (k), ch 4j (k) are the input and output of the jth node in the two hidden layers. c1ij (k), W c2ij (k), W c3i (k) are W c1 (k), W c2 (k), W c3 (k) The elements in W. c1ij (k), W c2ij (k), W c3i (k) are the weights from the input layer to the hidden layer, the weights between two hidden layers, and the weights from the hidden layer to the output layer.

[0052] The utility function of the system is defined as:

[0053] U(k)=e(k)Ae(k) T +fe * (k)Bfe * (k) T

[0054] Where A and B are positive definite diagonal matrices.

[0055] Define the objective function E of the evaluation network c (k) is:

[0056]

[0057]

[0058] where e c (k) is the error function of the evaluation network, where γ is the discount factor.

[0059] In the formula and They are implemented by two evaluation networks, and the inputs of the two evaluation networks are respectively,

[0060] x c (t) = [e(k), e(k-1), e(k-2), fe * (k)],x c (t)=[e(k-1),e(k-2),e(k-3),fe * (k-1)], the two evaluation networks have the same structure except for the input.

[0061] The weight update process of the evaluation network is shown as follows

[0062] The weight change from the hidden layer to the output layer is defined as:

[0063]

[0064] The updated weight matrix is:

[0065] W c3 (k+1)=W c3 (k)+ΔW c3 (k)

[0066] The weight change from hidden layer to hidden layer is defined as:

[0067]

[0068] The updated weight matrix is:

[0069] W c2 (k+1)=W c2 (k)+ΔW c2 (k)

[0070] The weight change from the input layer to the hidden layer is defined as:

[0071]

[0072] The updated weight matrix is:

[0073] W c1 (k+1)=W c1 (k)+ΔW c1 (k)

[0074] Among them l c (k) is the learning rate of the evaluation network, W c3 (k+1), W c2 (k+1), W c1 (k+1) is the updated weight matrix.

[0075] The structure of the execution network is 3-12-12-1. The speed error and the variables at the two moments before the speed error are selected as the input of the execution network. The thrust reference value fe output by the speed controller is used as the input of the execution network. * For output, the hidden layer of the execution network uses a bipolar sigmoidal function, and the output layer uses a linear function purelin.

[0076] Define the input vector of the execution network as:

[0077] x a (t) = [e(k), e(k-1), e(k-2)]

[0078] The following is the forward calculation process of executing the network:

[0079]

[0080] Among them ah 1j (k), ah 2j (k), ah 3j (k), ah 4j (k) are ah 1 (k), ah 2 (k), ah 3 (k), ah 4 (k) Elements within. ah 1j (k), ah 2j (k), ah 3j (k), ah 4j (k) are the input and output of the jth node in the two hidden layers. a1ij (k), W a2ij (k), W a3i (k) are W a1 (k), W a2 (k), W a3 (k) The elements in W. a1ij (k), W a2ij (k), W a3i (k) are the weights from the input layer to the hidden layer, the weights from the hidden layer to the hidden layer, and the weights from the hidden layer to the output layer. The gradient descent method is used to update the above weights, and the weights are minimized. To achieve the goal, update the weights of the execution network, that is, execute the objective function of the network

[0081] The weight update process of executing the network is shown as follows:

[0082] The weight change from the hidden layer to the output layer is defined as:

[0083]

[0084] The updated weight matrix is:

[0085] W a3 (k+1)=W a3 (k)+ΔW a3 (k)

[0086] Define the weight change between two hidden layers as:

[0087]

[0088] The updated weight matrix is:

[0089] W a2 (k+1)=W a2 (k)+ΔW a2 (k)

[0090] The weight change from the input layer to the hidden layer is defined as:

[0091]

[0092] The updated weight matrix is:

[0093] W a1 (k+1)=W a1 (k)+ΔW a1 (k)

[0094] Among them l a (k) is the learning rate of the evaluation network, W a3 (k+1), W a2 (k+1), W a1 (k+1) is the updated weight matrix.

[0095] The above weight update for:

[0096]

[0097] In the evaluation network and the execution network, the learning rate l a,c (k) are all adaptive learning rates, as shown in the following formula:

[0098]

[0099] Where α, β are learning rate gains, l a,c The initial value of (k) is 0.1, E a,c (k) is the objective function of the execution network and the evaluation network.

[0100] Figure 6As shown, the direct thrust control method of the permanent magnet linear synchronous motor involved in the present invention is a data-driven control method. The control process only includes the input and output of the control system. When there are conditions in the system where parameters change due to operation, or when a sudden load disturbance occurs, the ADHDP control method of the present invention can be used to feed back accurate speed values ​​and output accurate thrust values. The present invention has the characteristics of good robustness and no overshoot, and can enable the permanent magnet linear synchronous motor to operate stably.

Claims

1. A method for direct thrust control of a permanent magnet linear synchronous motor, comprising the following steps: S101: Measuring the three-phase current i of the permanent magnet linear synchronous motor a 、i b 、i c , three-phase voltage u a 、u b 、u c , perform Clark transformation to obtain the equivalent current i in the α-β coordinate system α 、i β And the equivalent voltage u α 、u β ; S102: Using the equivalent current i in the α-β coordinate system α 、i β And the equivalent voltage u α 、u β Calculate the thrust fe and flux ψ of the permanent magnet linear synchronous motor s and the flux linkage angle θ s ; S103: construct an ADHDP speed controller, and the thrust reference value fe output by the ADHDP speed controller * The thrust difference △fe is obtained by subtracting the calculated thrust fe, and the given magnetic flux amplitude ψ s * The flux linkage ψ calculated in step S102 s The difference in flux is obtained by subtracting △ψ s ; The construction method of the ADHDP speed controller is as follows: construct two BP neural networks, namely, an execution network and an evaluation network, and use the execution network and the evaluation network to form the ADHDP speed controller; the execution network and the evaluation network in the ADHDP speed controller both contain two hidden layers; the learning rates of the evaluation network and the execution network are l a,c (k) are adaptive learning rates, as shown in the following formula: in, α, β are learning rate gains; l a,c The initial value of (k) is 0.1; E a,c (k) is the objective function of the execution network and the evaluation network; S104: The thrust difference △fe and the flux difference △ψ in step S103 are combined. s The signal and flux angle θ obtained after passing through the thrust and flux hysteresis regulators respectively s The signals output by the sector selection unit are input to the voltage vector switch selection unit, which then selects the state of each switch of the inverter to generate a three-phase AC voltage to operate the motor.

2. The method for direct thrust control of a permanent magnet linear synchronous motor according to claim 1, characterized in that: The hidden layers of the execution network and the evaluation network adopt bipolar sigmoidal functions, and the output layer adopts the linear function purelin.

3. The method for direct thrust control of a permanent magnet linear synchronous motor according to claim 2, characterized in that: The structure of the execution network is 3-12-12-1.

4. The method for direct thrust control of a permanent magnet linear synchronous motor according to claim 2, characterized in that: The structure of the evaluation network is 4-10-10-1.

5. The method for direct thrust control of a permanent magnet linear synchronous motor according to claim 3, characterized in that: The speed error e(k) and the variables e(k-1) and e(k-2) at the two moments before the speed error are used as the inputs of the execution network, and the thrust reference value fe * is the output of the execution network.

6. The method for direct thrust control of a permanent magnet linear synchronous motor according to claim 4, characterized in that: The speed error e(k) and the variables e(k-1), e(k-2) at the two moments before the speed error and the thrust reference value fe output by the execution network are used. * As the input of the evaluation network, the estimated value of the system performance indicator function is the output.

7. The method for direct thrust control of a permanent magnet linear synchronous motor according to claim 5 or 6, characterized in that: The speed error e(k) is obtained by using a magnetic scale to obtain the motor speed v and comparing it with the expected speed v ref Subtract and get the velocity error e(k).

Images