CN110224649A - A method of DTC PREDICTIVE CONTROL is used for based on support vector machines - Google Patents

Abstract

The invention relates to a method for DTC predictive control based on a support vector machine, which first selects existing motor operating parameters as input and output samples; performs normalization processing on the selected input and output samples; normalization processing The final sample is trained to obtain the prediction model based on the support vector machine of the output voltage vector for DTC; the operating parameters of the current motor are normalized and then input into the prediction model, and the basic voltage vector applied is output through the prediction model; According to the basic voltage vector output by the predictive model, the switching state of the inverter is controlled, and the DTC predictive control based on the support vector machine is completed. The invention can reduce related calculation amount, reduce torque ripple, and keep switching frequency constant.

Description

A method for DTC predictive control based on support vector machine

technical field

The invention belongs to the field of motor control, and in particular relates to a method for DTC predictive control based on a support vector machine.

Background technique

Direct torque control (DTC) controls the electromagnetic torque by controlling the magnitude of the stator flux linkage and the load angle. Permanent magnet synchronous motor direct torque control usually uses two hysteresis comparators to control the stator flux linkage and torque deviation respectively. Since it avoids a large number of calculations in the rotating coordinate system, its dynamic performance will be improved and the response will be faster .

In the permanent magnet synchronous motor direct torque control system realized by the traditional switch table, the voltage vector is continuously applied within a sampling period, and the actual torque increase or decrease exceeds the expected requirement, resulting in overshoot pulsation.

In order to solve such problems, predictive control is introduced, cost function is introduced, torque error and stator flux error are considered comprehensively, and controlled, and space vector modulation technology is used to reduce torque ripple. However, there are still some problems in predictive control. Firstly, the scientificity of its cost function is still to be studied. Secondly, in order to make the torque error and flux error on the same dimension, weight coefficients are usually introduced, and the selection of weight coefficients mostly depends on experience. Lack of strong theoretical support. At the same time, the use of predictive control requires the calculation of the cost of six basic voltage vectors at the same time, which requires a huge amount of calculation, which is contrary to the original intention of introducing direct torque control.

Contents of the invention

The purpose of the present invention is to overcome the above disadvantages, and provide a method for DTC predictive control based on support vector machine, which can reduce the related calculation amount, reduce the torque ripple, and the switching frequency is constant.

In order to achieve the above object, the present invention comprises the following steps:

Include the following steps:

Step 1: Select the existing motor operating parameters as input and output samples;

Step 2: normalize the selected input and output samples respectively;

Step 3: Train the normalized samples to obtain a prediction model for DTC based on the support vector machine of the output voltage vector;

Step 4: Normalize the operating parameters of the current motor;

Step 5: Input the normalized operating parameters in step 4 into the prediction model, and output the applied basic voltage vector through the prediction model;

Step 6: Control the switching state of the inverter according to the basic voltage vector output by the predictive model, and complete the DTC predictive control based on the support vector machine.

Further, in step 1, the existing motor operating parameters are the motor operating parameters that use the cost function to select the voltage vector, and the cost function is shown in the following formula (1):

in is the reference value, is the reference torque, is the actual amplitude, T _e is the actual torque.

Furthermore, the number of cost functions is equal to the number of basic voltage vectors, and the basic voltage vector corresponding to the minimum cost function is selected as the applied voltage vector; the stator flux linkage errors ψ _r , The output torque error T _r , load angle δ and torque angle α are used as the input of the training prediction model, and the applied voltage vector is selected as the output.

Further, in step 2, the expressions for normalizing the input and output samples are shown in the following formula (2):

Among them, x _min is the minimum value in the input sample or the output sample, and x _max is the maximum value in the input sample or the output sample.

Further, in step three, training is performed using a support vector machine.

Further, the radial basis function is used as the kernel function of the support vector machine, as shown in formula (3):

k(x, x _i )=exp(-γ‖xx _i ‖ ² ),γ＞0 (3)

Where x is the input, _xi is the support vector, and γ is the kernel parameter.

Further, the input and output samples selected in step 1 are divided into 3 groups, and the subset data of each group is used as a verification set, and the remaining 2 sets of subset data are used as training sets to obtain 3 sub-models, each sub-model The models correspond to a set of penalty parameters c and kernel function parameters g respectively, and the average of the classification accuracy of the verification set in the three sub-models is used as the performance index of the classifier.

Further, if the average of the classification accuracy is greater than or equal to 90%, select the parameters c and g of the group with the highest classification accuracy as the parameters of the prediction model, and if the average of the classification accuracy is less than 90%, retrain.

Further, in step four, normalization processing is performed after obtaining the stator flux error ψ _r at time k, outputting the torque error T _r , load angle δ and torque angle α.

Further, the normalization processing method in step four is the same as that in step two.

Compared with the prior art, the present invention has the following beneficial technical effects:

In the present invention, by obtaining the operating data of the motor using the cost function, using the support vector machine method to train offline to obtain the corresponding prediction model, and then using the trained model to select the voltage vector according to the current operating parameters of the motor, so as to realize the rapid operation of the motor control. Through the above method, the performance of the permanent magnet synchronous motor direct torque control system is improved, the related calculation amount is reduced, the torque ripple is reduced, and the switching frequency is constant. Compared with the direct torque control, the present invention reduces the torque ripple by introducing a good prediction model learned offline, and avoids the introduction of a cost function compared to the predictive control, thereby avoiding the huge amount of calculation of the cost function and achieving The purpose of simplifying the amount of calculation is to optimize the real-time performance of motor control.

Description of drawings

Fig. 1 is based on the model training figure of the present invention;

Fig. 2 is a block diagram of the present invention;

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

Referring to Fig. 1 and Fig. 2, the present invention is used for DTC predictive control method based on support vector machine, and concrete steps are as follows:

Step 1: Determine the appropriate sampling frequency based on the existing motor running parameters using the cost function for voltage vector selection, and select appropriate input and output samples for DTC predictive control based on the support vector machine.

Among them, the expression of the cost function of the motor that uses the cost function to select the voltage vector is shown in (1).

in is the reference value, is the reference torque, is the actual amplitude, T _e is the actual torque.

When the number of alternative basic voltage vectors is 6, 6 cost functions will be obtained, and the basic voltage vector corresponding to the smallest cost function is selected, which is the applied voltage vector. After recording all the parameters in the above motor operation process, the sampling period is ten times of the motor sampling period, and the stator flux error ψ _r , output torque error T _r , load angle δ and torque angle α under all working conditions are selected As input to train the predictive model, the applied voltage vector is chosen as output.

Step 2: Normalize the input and output samples determined in Step 1.

In order to make the input and output at the same magnitude respectively, it is necessary to normalize the input and output samples respectively, that is, the x before normalization is normalized into the value of y according to the following formula, and the expression is as shown in formula (2): Show.

Among them, x _min is the minimum value in the sample, and x _max is the maximum value in the sample; that is, when the input sample is normalized, x _min is the minimum value in the input sample, and x _max is the maximum value in the input sample; for the output When samples are normalized, x _min is the minimum value in the output sample, and x _max is the maximum value in the output sample.

Step 3: use the support vector machine to train the normalized samples, and obtain the prediction model of the output voltage vector based on the support vector machine for DTC.

The normalized samples are trained using a support vector machine. According to the method idea of support vector machine, the radial basis function is selected as the kernel function of support vector machine, as shown in formula (3).

k(x, _xi )=exp(-γ‖xx _i ‖ ² ),γ>0 (3)

Where x is the input, _xi is the support vector, and γ is the kernel parameter.

Randomly scramble the input and output samples selected in step 1 to form a new sequence. In order to ensure the accuracy of the off-line training-based support vector machine used in the DTC prediction model, the sampling cross-validation method uses the samples selected in step 1 It is divided into 3 groups, and each group is used as a verification set, and the remaining 2 sets of subset data are used as training sets, so that 3 sub-models will be obtained, and each sub-model corresponds to a set of penalty parameters c and kernel function parameters g , using the average of the classification accuracy of the validation set in the three sub-models as the performance index of the classifier. If the average of the classification accuracy is greater than or equal to 90%, select the parameters c and g of the group with the highest classification accuracy of the three sub-models as the parameters of the prediction model. If the average of the classification accuracy is less than 90%, retrain. The classification accuracy of its sub-models and the average of classification accuracy are calculated as shown in formula (4).

Among them, η _i is the classification accuracy rate of the i-th sub-model, n _i1 is the number of correct voltage vectors selected in the i-th sub-model, n _i2 is the total number of selected voltage vectors in the i-th sub-model, and η _ave is the classification average accuracy.

Step 4: Normalize the operating parameters of the current motor.

By obtaining the stator flux error ψ _r at time k, the output torque error T _r , the load angle δ and the torque angle α, the normalization process is performed, which is the same as formula (2).

Step 5: Input the normalized parameters in step 4 into the prediction model for DTC based on the support vector machine. The basic voltage vector applied by the motor is predicted by using the trained prediction model based on support vector machine for DTC.

The processed data is input into the trained support vector machine model, and the applied basic voltage vector is output through the prediction model based on the support vector machine for DTC. This procedure replaces the use of a cost function to select the base voltage vector.

Step 6: According to the basic voltage vector output by the prediction model based on the support vector machine for DTC, the switching state of the inverter is controlled, so as to realize the operation of the motor.

According to the principle of direct torque control, the basic voltage vector output by the prediction model based on the support vector machine for DTC determines the switch state of the inverter, thereby controlling the operation of the motor.

Example 1

Firstly, collect the parameters of motor operation using the cost function for voltage vector selection, including stator flux error ψ _r , output torque error T _r , load angle δ, torque angle α and the selected voltage vector, and the sampling frequency is the frequency used by the motor Ten times of , normalize the sampled data, as shown in formula (2). The support vector machine using formula (3) as the kernel function is used for offline training, and the cross-validation method is used to obtain the model with the highest accuracy. The calculation method of accuracy is shown in formula (4), and the model with the highest accuracy is used as the basis A support vector machine is used for the prediction model of DTC.

Secondly, when the motor is running, the stator flux error ψ _r , the output torque error T _r , the load angle δ, and the torque angle α are used as input at a certain moment, and are input into the prediction model through normalization processing to obtain this The basic voltage vector selected by the motor at the moment. The inverter switching state is determined according to the basic voltage vector to control the motor. The relationship between the selected basic voltage vector and the state of the inverter switch is shown in Table 1. For example, when the SVM-based prediction model for DTC selects the voltage vector as the 0 voltage vector (ie u ₁ ), its inverter state is S _A =1, S _B =0, S _C =0.

Table 1 The relationship between the basic voltage vector and the state of the inverter switch

Claims (10)

1. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines, it is characterised in that: the following steps are included:

Step 1: existing motor operating parameter is chosen as input and output sample；

Step 2: the input of selection and output sample are normalized respectively；

Step 3: being trained the sample after normalized, obtains being used for based on support vector machines for output voltage vector The prediction model of DTC；

Step 4: the operating parameter of current motor is normalized；

Step 5: the operating parameter after normalized in step 4 is input in prediction model, prediction model is passed through Export the basic voltage vectors applied；

Step 6: the basic voltage vectors exported according to prediction model control the switch state of inverter, complete based on support to Amount machine is used for DTC PREDICTIVE CONTROL.

2. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 1, it is characterised in that: In step 1, existing motor operating parameter is the motor operating parameter that voltage vector selection is carried out using cost function, cost Shown in function such as following formula (1):

WhereinFor reference amplitude,For torque reference,For actual magnitude, T_eFor actual torque.

3. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 2, it is characterised in that: The number of cost function and the number of basic voltage vectors are equal, choose the arrow of fundamental voltage corresponding to the smallest cost function Amount, the voltage vector as applied；Choose the stator magnetic linkage error ψ under full working scope_r, output torque error T_r, power angle δ and Input of the angle of torsion α as training prediction model chooses the voltage vector of application as output.

4. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 1, it is characterised in that: In step 2, to outputting and inputting, the expression formula such as following formula (2) that sample is normalized respectively is shown:

Wherein, x_minFor minimum value in input sample or output sample, x_maxFor maximum value in input sample or output sample.

5. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 1, it is characterised in that: In step 3, it is trained using support vector machines.

6. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 5, it is characterised in that: Using radial basis function as support vector machines kernel function, as shown in formula (3):

k(x,x_i)=exp (- γ ‖ x-x_i‖²),γ>0 (3)

Wherein x is input, x_iFor supporting vector, γ is nuclear parameter.

7. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 5, it is characterised in that: The sample that outputs and inputs chosen in step 1 is divided into 3 groups, each group of subset data is made into one-time authentication collection respectively, remaining 2 groups of subset datas as training set, obtain 3 submodels, each submodel respectively corresponds one group of penalty parameter c and core letter Number parameter g, uses the performance indicator that the average of the classification accuracy of collection is verified in 3 submodels as classifier.

8. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 7, it is characterised in that: If the average of classification accuracy is more than or equal to 90%, highest one group of classification accuracy of parameter c and g are chosen as prediction The parameter of model, if the average of classification accuracy less than 90%, re -training.

9. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 1, it is characterised in that: In step 4, by the stator magnetic linkage error ψ for obtaining the k moment_r, output torque error T_r, after power angle δ and angle of torsion α, carry out Normalized.

10. a kind of method for being used for DTC PREDICTIVE CONTROL based on support vector machines according to claim 1, it is characterised in that: Normalized mode is identical with step 2 in step 4.