CN101938246B - Fuzzy fusion identification method of rotating speed of sensorless motor - Google Patents

Abstract

The invention discloses a fuzzy fusion identification method for the rotational speed of a speed sensorless motor, which uses a neural network MRAS speed identification method and a slip frequency direct speed identification method to simultaneously identify the motor rotational speed; adopts a fuzzy fusion method to identify the neural network MRAS speed The method is fused with the slip frequency direct speed identification method to obtain the reliable value of the motor speed. The beneficial technical effects of the present invention are: the two technical indexes of the dynamic and steady-state performance of the motor are optimized; during the starting and state switching process of the motor, the speed identification is fast, the dynamic tracking performance is strong, and the speed identification accuracy is high and robust. Strong performance; for the speed sensorless motor vector control system, it realizes the real intelligent cross comprehensive online speed identification, and obtains the best speed identification effect, so as to realize the effective control of the motor.

Description

Fuzzy Fusion Identification Method of Speed Sensorless Motor Speed

technical field

The present invention relates to a motor control technology, in particular to a fuzzy fusion identification method for a motor speed without a speed sensor.

Background technique

In the speed control system of AC asynchronous motor (hereinafter referred to as "motor" or "motor"), the lowest-end method is to use speed sensors to detect motor speed feedback signals; these speed sensors are installed on the motor shaft, not only need to install , maintenance, and increase the cost of the control system, it is not suitable for working in harsh environments and reduces the reliability of the system. If the speed sensor is not used, only the motor speed is obtained according to the voltage and current signals output by the frequency converter for closed-loop control, the speed sensor can be omitted, and the requirements of simplicity, cheapness and reliability of the motor speed control can be met. mainstream research directions.

Scholars at home and abroad have done a lot of work in this area, and proposed open-loop direct identification, MRAS identification based on rotor flux linkage, MRAS identification based on back EMF, full-order flux observer, extended Kalman filter, and high-frequency injection method. and other speed identification methods. The aforementioned methods are all traditional sensorless speed identification. In order to improve the performance of the control system, many scholars have introduced intelligent control technologies such as fuzzy and neural networks into the speed identification of the motor. This is the research hotspot and development direction of the speed sensorless motor control.

The most direct way to identify the motor speed with an intelligent method is to use the easy-to-detect motor stator voltage and current to design a BP neural network to identify the speed, but this method has the problem that it is difficult to determine the hidden layer and the number of nodes in the network. At present, there is no good way to determine the specific network structure, and it is still based on experience. At the same time, the convergence speed of the learning algorithm is slow, and the convergence speed is related to the selection of the initial weight.

Aiming at the problem of poor real-time speed identification using BP neural network, some scholars have proposed using diagonal recursive neural network to identify speed; such as: Yang Junyou, Chen Daming, "Sensorless Control of Permanent Magnet Synchronous Motor with Diagonal Recurrent Neural Network", [J]. Journal of Shenyang University of Technology, 2008, 30(1): 24-27, the method recorded in the literature is to estimate the actual measured voltage and current with a diagonal recursive neural network observer after coordinate transformation For current and angular velocity, the difference between the estimated value and the actual value is used to adjust the connection weight of the neural network observer until the prediction error reaches the set value; but the problem with this method is that while predicting the angular velocity, the stator current must be For prediction, two neural network observers are used. It makes the learning algorithm more complicated and difficult to adjust.

Therefore, in the speed identification of the motor, the neural network MRAS (Model Reference Adaptive System) speed identification method is introduced, such as: Chen Bing, Dong Lei, etc., "Research on Sensorless Vector Control of Asynchronous Motors at Low Speed", [J ]. Journal of Northwest University (Natural Science Edition), 2007, 37(1): 45-47, the method recorded in the literature is that the neuron model is derived from the current model as an adjustable model of MARS, and its weight contains motor For speed information, the neuron weight δ learning rule derived from the gradient method replaces the PI adaptive law, which makes the speed identification method simple in structure, strong in self-learning ability, and has good system stability. However, this method has the problems of large fluctuations in the starting speed of the motor, even negative fluctuations, many oscillation times, and long adjustment time. Not only that, but it is further found in the research that when switching from one steady state to another, the speed estimation converges to a wrong value, which causes the actual speed to fail to converge to the given value (which is in the control The links in the system refer to the mark 2 in Figure 1, and the identification effect is shown in Figure 10). Therefore, the neural network MRAS speed identification method cannot be practically applied in the prior art.

Contents of the invention

In order to solve the problems existing in the background technology, the present invention proposes a fuzzy fusion identification method of the speed sensorless motor speed, which uses the neural network MRAS speed identification method and the slip frequency direct speed identification method to simultaneously identify the motor speed; The network MRAS speed identification method and the slip frequency direct speed identification method assign their respective action strength values, and use the following formula to calculate the sure value of the motor speed:

为转差频率直接速度辨识方法所识别出的电机转速值；

为神经网络MRAS速度辨识方法所识别出的电机转速值；

为最终识别出的电机运行时的转速确信值；

为神经网络MRAS速度辨识方法的作用强度值；

为转差频率直接速度辨识方法的作用强度值。

is the motor speed value identified by the slip frequency direct speed identification method;

The motor speed value identified by the neural network MRAS speed identification method;

is the final recognized value of the motor speed when it is running;

is the action strength value of the neural network MRAS speed identification method;

is the action intensity value of slip frequency direct speed identification method.

The aforementioned "assigning respective action strength values to the neural network MRAS speed identification method and the slip frequency direct speed identification method" includes:

和

，同一误差阈值范围内的

和

满足

，形成作用强度分布表； 1) Set the numerical continuous error threshold range according to the value, and set the corresponding error threshold range for each error threshold range

and

, within the same error threshold

and

satisfy

, to form the action intensity distribution table;

； 2) Real-time calculation of the error between the identification value of the neural network MRAS speed identification method and the expected value of the motor speed

;

处于哪个误差阈值范围内，则将该误差阈值范围对应的

和

代入公式

。 3) Judging the error in step 2)

Which error threshold range is in, then the corresponding error threshold range

and

Into the formula

.

Step 1), including:

1] Mark the 10 error threshold ranges with continuous values as ten levels from 0 to 9; set the error threshold range corresponding to the 9th level, the effect strength of the neural network MRAS speed identification method is 0, and the slip frequency directly identifies the speed The action strength of the method is 1; corresponding to the 0th grade error threshold range, the action strength of the neural network MRAS speed identification method is 1, and the action strength of the slip frequency direct speed identification method is 0;

2] Calculate the action intensity distribution of the neural network MRAS speed identification method and the slip frequency direct speed identification method in the error threshold range of the 1st to 8th grades using a symmetrical function;

Make a table of action intensity distribution.

函数，计算第1至8等级时，神经网络MRAS速度辨识方法和转差频率直接速度辨识方法的作用强度分布： Step 2], including: according to the symmetrical

Function to calculate the action intensity distribution of the neural network MRAS speed identification method and the slip frequency direct speed identification method for grades 1 to 8:

，

,

为误差等级的级数；

为

函数的参数；

分别取0.1、0.2、0.3、0.4、0.5、0.6、0.7、0.8或0.9，

的每个取值，分别对应一组互相匹配的神经网络MRAS速度辨识方法和转差频率直接速度辨识方法的作用强度取值。 in,

is the series of error levels;

for

function parameters;

Take 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8 or 0.9 respectively,

Each value of , corresponds to a group of matching neural network MRAS speed identification method and slip frequency direct speed identification method respectively.

的方法，包括： In step 2), calculate

methods, including:

为电机转速期望值；

为神经网络MRAS速度辨识方法的辨识值。 in,

is the expected value of motor speed;

is the identification value of the neural network MRAS speed identification method.

The beneficial technical effects of the present invention are: at the same time, the two technical indicators of the dynamic and steady-state performance of the motor are optimized when the speed identification method is adopted by an intelligent method; And the problem that the speed identification cannot converge to the given value correctly when switching from one steady state to another steady state; the speed identification is fast and the dynamic tracking performance is strong during the starting and state switching process of the motor ;In the steady state process, the speed identification accuracy is high and the robustness is strong; for the speed sensorless motor vector control system, the real sense of the intelligent cross comprehensive online speed identification is achieved, and the best speed identification effect is obtained, so as to realize the effective control of the motor .

Description of drawings

Fig. 1, the speed identification block diagram of motor vector control system in the prior art;

Fig. 2, adopt a kind of motor vector control system speed identification block diagram of the scheme of the present invention;

Fig. 3. Schematic diagram of the control system structure of the neural network MRAS speed identification method;

Figure 4. Block diagram of the improved voltage model module;

Fig. 5, block diagram of neural network current model module principle;

Figure 6. Schematic diagram of the control system structure of the direct speed identification method of slip frequency;

Figure 7. Schematic diagram of fuzzy fusion principle of neural network MRAS speed identification method and slip frequency direct speed identification method;

Figure 8. Flow chart of speed identification during motor starting and state switching;

Figure 9, Schematic diagram of the relationship between the value and the change of the action intensity;

Figure 10, the speed identification effect diagram of the motor vector control system in Figure 1;

Fig. 11 is a speed identification effect diagram of the motor vector control system according to the solution of the present invention.

Detailed ways

Proceeding from the problem of "background technology", the inventor, after painstaking research, found that the neural network MRAS speed identification method has the following characteristics: the convergence speed is slow, and the speed fluctuates greatly in dynamic stages such as motor starting and speed change, and even negative fluctuations occur. The number of oscillations is large, the real-time performance is poor, and even the speed estimation converges to an error value; while the neural network needs to continuously learn and adjust the weights, and has a strong self-regulation ability. It is relatively stable in the steady-state stage with small errors. High precision and strong robustness.

At the same time, the inventor also found that the existing slip frequency direct speed identification method is a simple, direct and effective method with less computation, no delay, good dynamics, and can greatly improve the rapidity of speed identification. The disadvantage of this method is that the calculation results are not corrected, the speed identification effect is heavily dependent on the accuracy of the motor parameters, and the anti-interference performance is poor.

Based on the aforementioned analysis, the inventor considers to combine the advantages of the two identification methods; one way of thinking is to use the speed identification of variable structure switching, that is, to use the neural network MRAS speed identification method to identify the speed in the steady state stage, and to identify the speed in the dynamic stage. Switch to the speed identification method using slip frequency direct speed identification; but after more in-depth research, the inventor found that there are two problems in this variable-structure switching speed identification: first, switching between different speed identification methods, switching The selection of points is blind, which requires repeated experimental debugging and comparison; second, the speed identification of variable structure switching is an accurate choice of either 0 or 1, and switching between different speed identification methods is prone to sudden changes in the identification amount , the sudden change of the identification value will cause speed jitter, generate noise, and directly affect the control effect.

In the above-mentioned variable structure switching speed identification, in the dynamic stage, if a single speed identification method is absolutely selected, the useful information of other speed identification methods will be lost. In fact, different speed identification methods have certain effects in different error domains, and the difference is only in the identification performance. Although the neural network MRAS speed identification method has defects in the dynamic process, it still provides some useful feature information, but the useful information is not much, and there are still some differences in the useful information in different error stages.

Based on the above factors, the inventor proposes the following scheme: in the dynamic process of a large change in the given rotational speed, fuzzy fusion is performed on the identification values of the two rotational speed identification methods, so that the useful information of the two rotational speed identification methods can play a role: In the stage where the error domain is large, the information of the slip frequency direct speed identification method (that is, the identified rotational speed) plays a leading role in order to speed up the convergence speed of the system and achieve fast speed tracking; while when the error domain is small, it plays a leading role The source of information is converted to the neural network MRAS speed identification method; in the steady state of the motor (that is, an extreme case, the action strength of the neural network MRAS speed identification method is 1, and the action strength of the slip frequency direct speed identification method is 0; Correspondingly, there is also an extreme situation in the dynamic stage, that is, the action strength of the neural network MRAS speed identification method is 0, and the action strength of the slip frequency direct speed identification method is 1. At this time, it is equivalent to only slip The frequency direct speed identification method works), only the neural network MRAS speed identification method is used to identify the motor speed, and its principle is shown in Figure 7.

The basic steps of neural network MRAS speed identification, slip frequency direct speed identification and fuzzy fusion method of the two are introduced below in conjunction with the accompanying drawings:

(1) Neural network MRAS speed identification method

In the present invention, the model reference adaptive speed identification method based on the neural network is adopted for the speed identification of the motor steady-state section and dynamic section (the link in the control system is shown at 50 in Figure 2), and the modules involved are detailed The structure is shown in the dotted box 52 in Fig. 3, which consists of an improved voltage model module 54, a neural network current model module 60, a learning algorithm (also called minimum variance method, LMS algorithm) module 82 and so on.

与神经网络电流模型的估计磁链

，二者在80处求误差：当辨识的速度与实际速度转速一致时，二者辨识的磁链相等；与实际值不符时，磁链就会产生误差。误差通过82处LMS算法模块反向调节网络的权值，即改变速度的辨识结果。通过不断地调整使辨识的速度能够跟踪实际速度的变化。 The principle of this method is as follows: There are two flux linkage identification units in the virtual frame 52 in Fig. 3, one uses the improved voltage model as the expected value unit of the rotor flux linkage at 54, and the other uses the neural network as the rotor flux identification unit at 60 Estimated value unit for flux linkage. The rotor flux linkage obtained by the improved voltage model

Estimating Flux Linkages with Neural Network Current Models

, the two calculate the error at 80: when the identified speed is consistent with the actual speed, the flux linkage identified by the two is equal; when it does not match the actual value, the flux linkage will produce an error. The error reversely adjusts the weight of the network through the 82 LMS algorithm modules, that is, the identification result of changing the speed. By constantly adjusting, the identified speed can track the change of the actual speed.

The three modules comprising the improved voltage model module 54 , the neural network current model module 60 and the LMS algorithm module 82 will be introduced in detail below.

(1) Improved voltage model module

坐标系下，由定子电压方程和磁链方程推导出磁链的电压模型： exist -

In the coordinate system, the voltage model of the flux linkage is derived from the stator voltage equation and the flux linkage equation:

,

、

为

轴和

轴的转子磁链；

、

和

、

分别为

轴和

轴定子电压与电流在两相静止坐标系下的

、

轴分量； In the above formula,

,

for

axis and

The rotor flux linkage of the shaft;

,

and

,

respectively

axis and

Shaft stator voltage and current in two-phase stationary coordinate system

,

axis component;

为漏感系数，

为互感系数，

为定子电感，

为转子电感，为微分算子，

为定子电阻。

is the leakage inductance coefficient,

is the mutual inductance coefficient,

is the stator inductance,

is the rotor inductance, is a differential operator,

is the stator resistance.

Turn the above formula into vector form (at 56 in Figure 4):

在59处滤波后所得信号补偿，并令滞后环节的时间常数等于转子时间常数

，得到改进的电压方程： The pure integral link in the vector form expression is replaced by the first-order inertial filter link at 58 in Fig. 4, and the state estimation phase lag generated by the inertial link is determined by the reference rotor flux linkage

The signal obtained after filtering at 59 is compensated, and the time constant of the hysteresis link is equal to the rotor time constant

, get the improved voltage equation:

的纯积分环节，不存在积分漂移的问题，克服了电机参数偏差经积分的累积产生漂移，影响到系统的调速精度和稳定性的问题。 The advantage of this method is that the common voltage model method is canceled about the back electromotive force

The pure integral link does not have the problem of integral drift, which overcomes the problem that the motor parameter deviation will drift through the accumulation of integral, which will affect the speed regulation accuracy and stability of the system.

(2) Neural network current model module

The rotor flux current model can be obtained from the rotor flux equation of the asynchronous motor:

，  

,

、

为

轴和

轴电流模型估计的转子磁链；、分别为

轴和

轴定子电流在两相静止坐标系下的

、轴分量；

为互感系数，

为转子时间常数，

为微分算子；

为转子电气角速度。 In the above formula,

,

for

axis and

Rotor flux linkage estimated by the shaft current model; , respectively

axis and

Shaft stator current in two-phase stationary coordinate system

, axis component;

is the mutual inductance coefficient,

is the rotor time constant,

is a differential operator;

is the rotor electrical angular velocity.

Using the backward difference method to discretize the above current model, the discretized state equation of the current model is obtained:

为步数； In the above formula, is the sampling period,

is the number of steps;

make:

，

，，

,

, ,

，

，

，

,

,

,

Then the discretized state equation can be simplified as:

、64处

和66处

，68处

、70处

和72处为网络的权值，在这个神经网络模型中，各权值都有确定的物理意义。当模型用来辨识速度时，

和

已知，而

未知，在线调整，即可得到速度的辨识值。在74处的

表示一步滞后（时延）环节。 It can be seen from the above formula that the current model of the rotor flux observer can be replaced by the two-layer linear neural network model shown in Figure 5. In this neural network model, the input is 62 places

, 64 places

and 66 places

, 68 places

, 70 places

and 72 places is the weight of the network, in this neural network model, each weight has a definite physical meaning. When the model is used to identify velocity,

and

known, while

Unknown, adjust online , the identification value of velocity can be obtained. at 74

Indicates a one-step lag (time delay) link.

(3) Learning algorithm (LMS algorithm) module

。此误差通过LMS学习算法来对电机的估计转速

进行调整。LMS算法也称为最小方差法，即是

（全称为Widrow-Hoff ）规则。 When the speed used for calculation in the neural network is not equal to the actual speed, there will be an error between the voltage model and the current model

. This error is used by the LMS learning algorithm to estimate the speed of the motor

Make adjustments. The LMS algorithm is also called the minimum variance method, which is

(full name Widrow-Hoff )rule.

Let the error equation be

The energy function is defined as:

，

两个权值不变，都为定值，而3个权重中仅包含 In the above neural network model, after the sampling period is determined, it is assumed that the motor parameters remain unchanged, that is,

,

The two weights remain unchanged, both of which are fixed values, and among the three weights only Include

项，只需要调整。用于权值调整的自适应学习算法为： speed to be identified

item, only need to adjust . The adaptive learning algorithm for weight adjustment is:

The expression of the weight during training is:

为学习系数，

为

的转置矩阵，最后推出速度辨识公式： In the formula,

is the learning coefficient,

for

The transposition matrix, and finally the speed identification formula is introduced:

This method has a simple structure and does not require offline training. It can learn online after the motor is running, and it is not affected by the motor load, speed, parameters, etc. The correct identification result can be obtained through simple and fast operation.

Combining Figure 3, Figure 4 and Figure 5, the steps of the aforementioned process can be summarized as follows:

1] Set the initial weight coefficient for the current model neural network flux linkage identification unit:

，

，

,

,

、

和

为神经网络的权值；为采样周期，

为转子时间常数，为互感系数； In the formula,

,

and

is the weight of the neural network; is the sampling period,

is the rotor time constant, is the mutual inductance coefficient;

；改进电压模型磁链辨识方程为： 2] According to the voltage and current values in the two-phase static coordinate system, the expected value of the rotor flux linkage is calculated with the improved voltage model flux linkage identification unit

; The flux linkage identification equation of the improved voltage model is:

、

分别为定子电压与电流； In the formula,

,

are stator voltage and current respectively;

为漏感系数，

为互感系数，

为定子电感，

为转子电感；

is the leakage inductance coefficient,

is the mutual inductance coefficient,

is the stator inductance,

is the rotor inductance;

为微分算子，

为定子电阻，

转子时间常数，为参考转子磁链；

is a differential operator,

is the stator resistance,

rotor time constant, is the reference rotor flux linkage;

组采样输入下转子磁链的输出值

为： 3] Calculate the objective function of the network: neurons in

The output value of the rotor flux linkage under the group sampling input

for:

为神经网络模型的输入；

=1、2、3； In the formula,

is the input of the neural network model;

=1, 2, 3;

和

相减，得误差方程： 4] Will

and

Subtract to get the error equation:

；

;

5] Define the energy function:

;

调整的自适应学习算法为： 6] LMS rules: for weight

The adjusted adaptive learning algorithm is:

The expression of the weight during training is:

为

的转置矩阵，

为学习系数，

为步数； In the formula,

for

The transpose matrix of

is the learning coefficient,

is the number of steps;

中包含了待辨识的速度

项，速度由下式推算： exist

contains the velocity to be identified

Item, the speed is calculated by the following formula:

跟踪上改进电压模型磁链辨识

为止，辨识的速度就能跟踪上实际速度变化。本发明的改进就是将其计算结果与转差频率直接速度辨识方法的计算结果（图3的84处）一起在模糊融合模块（图3中的86处）中进行模糊融合。 7] Repeat steps 2] to 6] until the current model neural network flux linkage identification

Improving Flux Linkage Identification of Voltage Model on Tracking

So far, the identified speed can track the actual speed change. The improvement of the present invention is to carry out fuzzy fusion in the fuzzy fusion module (86 place in Fig. 3) together with the calculation result of the slip frequency direct speed identification method (84 place in Fig. 3).

(2) Slip frequency direct speed identification method

In Fig. 2, the direct speed identification method of slip frequency is composed of two modules; the first module is to observe the rotor flux linkage at 32; the second module is based on the rotor flux observation, at 40 Calculate the synchronous angular velocity and slip angular velocity of the rotor flux linkage, and then use the difference between the two to identify the rotor angular velocity.

The starting point of this method is that in the rotating coordinate system, the direct field oriented control of the current model has better flux linkage estimation accuracy. Therefore, the flux linkage value is obtained by using the observation model of the rotor flux linkage in the rotating coordinate system. Since the input is the stator current signal in the two-phase static coordinate system, it is necessary to carry out the Park transformation of the motor, and use the observation model of the rotor flux linkage in the rotating coordinate system to obtain the flux linkage value, and then perform the Parker inverse ( Park ^-1 ) transformation, and finally obtain the rotor flux value in the two-phase static coordinate system; on this basis, first calculate the synchronous angular velocity of the rotor flux, and filter it; at the same time, use the obtained two-phase static The slip angular velocity is calculated from the rotor flux linkage and current value in the coordinate system. The angular velocity of the motor rotor can be identified by subtracting the slip angular velocity from the synchronous angular velocity of the rotor flux linkage.

The advantage of using this speed estimation method is that it is a simple, direct and effective method with less computation. It has no delay in theory and has good dynamic performance. It has been practically applied in some products.

、

经过Park变换为两相旋转坐标系下的电流

、

。转子闭环磁链、

的估计采用了36处的电流模型（

、

上标中的

表示由电流模型得到）。将在两相旋转坐标系下观测到的磁链

、

经38处的Park逆变换为两相静止坐标系下的转子磁链

、。以上便完成了转子磁链的估计，这将为下一步的转差频率的直接速度辨识作好了准备。将得到的两相静止坐标系下转子磁链、

在42处可计算出转子磁链角的正余弦

、

值，并将之用于如图2中28处的旋转/固定坐标变换。同时，计算出同步角速度

，由于该角速度是由微分信号产生的，它会放大噪声信号，因此需在44处用一个一阶低通滤波器，将放大的噪声信号滤除得到

。而转差角速度

的值可由电动机电流检测值和两相静止坐标系下磁链值

、

在46处计算得到。这样就可在48处，根据滤波后的同步角速度值

与转差角速度

值，计算出当前的转速值。 Referring to Fig. 6, the electric current under the two-phase static coordinate system is put at 34 in the figure

,

After Park transformation, it is the current in the two-phase rotating coordinate system

,

. Rotor closed-loop flux linkage ,

is estimated using a current model at 36 (

,

in superscript

means obtained from the current model). The flux linkage observed in the two-phase rotating coordinate system

,

The rotor flux linkage in the two-phase stationary coordinate system is transformed by Park inverse transformation at 38

, . The estimation of the rotor flux linkage is completed above, which will be ready for the next step of direct speed identification of slip frequency. The obtained rotor flux linkage in the two-phase stationary coordinate system ,

The sine and cosine of the rotor flux linkage angle can be calculated at 42

,

value, and use it for the rotation/fixed coordinate transformation at 28 in Figure 2. At the same time, calculate the synchronous angular velocity

, since the angular velocity is generated by the differential signal, it will amplify the noise signal, so a first-order low-pass filter needs to be used at 44 to filter the amplified noise signal to obtain

. while the slip angular velocity

The value of can be determined by the motor current detection value and the flux linkage value in the two-phase static coordinate system

,

Calculated at 46. In this way, at 48, according to the filtered synchronous angular velocity value

and slip angular velocity

value, calculate the current speed value .

The foregoing speed identification method can be summarized into the following steps:

、经过Park变换为两相旋转坐标系下的定子电流

、

； 1] The stator current in the two-phase stationary coordinate system

, The stator current in the two-phase rotating coordinate system after Park transformation

,

;

2] Estimate the rotor flux linkage according to the following formula:

，

,

为互感，

为微分算子，转子时间常数

； In the formula,

for mutual inductance,

is the differential operator, the rotor time constant

;

、，上标中表示由电流模型得到； Obtain the rotor flux linkage value in the two-phase rotating coordinate system

, , in the superscript means obtained from the current model;

、

进行Park逆变换，得两相静止坐标系下转子磁链

、

； 3] Yes

,

Perform Park inverse transformation to obtain the rotor flux linkage in the two-phase stationary coordinate system

,

;

、

，计算出转子磁链角正余弦

、

值以及同步角速度

： 4] According to

,

, calculate the sine and cosine of the rotor flux linkage angle

,

value and synchronous angular velocity

:

一阶低通滤波得到

，滤波按下式进行： 5] For the obtained synchronous angular velocity

First-order low-pass filtering to get

, the filtering is performed as follows:

为微分算子，

，

是滤波器的截止频率。 In the formula,

is a differential operator,

,

is the cutoff frequency of the filter.

、和

、

计算出转差角速度

： 6] Meanwhile, according to

, and

,

Calculate the slip angular velocity

:

）

)

和

计算出电机当前转速值

： 7] According to

and

Calculate the current speed value of the motor

:

In this way, the direct speed identification of the slip frequency of the motor can be realized.

(3) The specific combination of the neural network MRAS speed identification method and the slip frequency direct speed identification method in the present invention:

。其中，转差频率直接速度辨识法的作用强度值

随着误差阈值的减小而减小，而神经网络MRAS速度辨识方法的作用强度值

则与之相反，参见图7中的箭头方向。当

减为0时，

则增为1，速度辨识进入稳态区域，此时仅由神经网络MRAS速度辨识方法对电机速度进行辨识。 Combined with Figure 7, the principle of the fuzzy fusion speed identification method is: in the dynamic stage of the motor (such as motor starting, state switching, etc.), the neural network MRAS speed identification method and the slip frequency direct speed identification method are used to simultaneously identify the motor speed , so that the valuable information in the two speed values participates in the calculation of the speed, and the sure value of the motor speed is obtained: set the numerical continuous error threshold range according to the value, and in different error threshold ranges, the neural network MRAS speed identification method and Slip Frequency Direct Velocity Identification Method Assigns Respective Action Intensity Values

. Among them, the action strength value of the slip frequency direct speed identification method

Decreases with the decrease of the error threshold, while the action intensity value of the neural network MRAS speed identification method

On the contrary, see the direction of the arrow in FIG. 7 . when

When reduced to 0,

If it is increased to 1, the speed identification enters the steady-state region. At this time, only the neural network MRAS speed identification method is used to identify the motor speed.

The program block diagram of this method is shown in Figure 8, and its specific steps are:

；计算

的方法，包括： 1) Real-time calculation of the error between the identification value of the neural network MRAS speed identification method and the expected value of the motor speed

;calculate

methods, including:

为电机转速期望值；

为神经网络MRAS速度辨识方法的辨识值。 in,

is the expected value of motor speed;

is the identification value of the neural network MRAS speed identification method.

和

，同一误差阈值范围内的

和

满足

，形成作用强度分布表； 2) Set the numerical continuous error threshold range according to the value, and set the corresponding error threshold range for each error threshold range

and

, within the same error threshold

and

satisfy

, to form the action intensity distribution table;

处于哪个误差阈值范围内，也即查找作用强度分布表。采用该误差阈值范围对应的作用强度分布值，对神经网络MRAS速度辨识方法和转差频率直接速度辨识方法所识别出的两个电机转速值进行模糊融合，计算出电机运行时的转速确信值；  3) Judging the error in step 1)

In which error threshold range, that is, look up the action intensity distribution table. Using the action intensity distribution value corresponding to the error threshold range, fuzzy fusion is performed on the two motor speed values identified by the neural network MRAS speed identification method and the slip frequency direct speed identification method to calculate the speed confidence value when the motor is running;

，动态过程结束，电机进入稳态运行阶段。 4) Repeat step 3) to identify the motor speed in real time. When the speed reaches the command given value

, the dynamic process ends, and the motor enters the steady-state operation stage.

In step 2), the action intensity distribution table can be drawn according to the experimental data, but this method has a certain degree of blindness, increases the workload, and makes the error of the final result larger. In this regard, the inventor proposes the following method for making the action intensity distribution table:

1] Mark the 10 error threshold ranges with continuous values as ten levels from 0 to 9; set the error threshold range corresponding to the 9th level, the effect strength of the neural network MRAS speed identification method is 0, and the slip frequency directly identifies the speed The action strength of the method is 1; corresponding to the 0th grade error threshold range, the action strength of the neural network MRAS speed identification method is 1, and the action strength of the slip frequency direct speed identification method is 0;

2] Calculate the action intensity distribution of the neural network MRAS speed identification method and the slip frequency direct speed identification method in the error threshold range of the 1st to 8th grades using a symmetrical function;

Make a table of action intensity distribution.

The concept of "action strength" is assigned the speed identification function, which is consistent with the meaning of fuzzy membership degree. When making action intensity distribution table, piecewise linear function can be used for calculation, such as:

However, the slope k of the piecewise linear function can only take a fixed value, and the obtained action intensity distribution table is only a one-dimensional table. The value of action intensity is not easy to adjust, and the flexibility is not good.

Functions of the following expressions can also be considered:

 

为系统的设定值，

、

为待定参数。通过设置合理的、

和

值，作用强度

可以根据

值的变化灵活地调整，但此函数参数多达三个，不易设置和调整。 in, is the series of error levels,

is the set value of the system,

,

is an undetermined parameter. by setting a reasonable ,

and

value, strength

can be based on

The change of the value can be adjusted flexibly, but there are as many as three parameters in this function, which is not easy to set and adjust.

函数来计算作用强度；对称型

函数也称正切型

型函数，其表达式如下： and It is a nonlinear function, and its characteristic is that the function itself and its derivatives are continuous, which is more convenient in processing. Therefore, a preferred scheme selected by the present invention is: adopt symmetrical

function to calculate the action strength; symmetric type

function also called tangent type

type function, its expression is as follows:

，

,

According to the above expression, the action strength distribution of the neural network MRAS speed identification method and the slip frequency direct speed identification method can be calculated for grades 1 to 8;

为误差等级的级数；

为

函数的参数；为避免此曲线出现饱和的状态，

取小于1的值，即分别取0.1、0.2、0.3、0.4、0.5、0.6、0.7、0.8、0.9。

的每个取值，分别对应一组互相匹配的神经网络MRAS速度辨识方法和转差频率直接速度辨识方法的作用强度取值。 in,

is the series of error levels;

for

parameter of the function; to avoid saturation of this curve,

Take a value less than 1, that is, take 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 respectively.

Each value of , corresponds to a group of matching neural network MRAS speed identification method and slip frequency direct speed identification method respectively.

，可改变

函数的曲线形状，也就对两种速度辨识的作用强度进行了调整，从而获得不同的辨识效果。对表1（见后文）分析可看出，当

在0.1~0.9的范围内： By adjusting parameters

, can be changed

The shape of the curve of the function also adjusts the strength of the two speed identifications, so as to obtain different identification effects. From the analysis of Table 1 (see later), it can be seen that when

In the range of 0.1~0.9:

值较小，则在融合区域的前期，随着误差减小，

变化较大，曲线的陡度大，即转差频率直接速度辨识方法的作用强度值迅速减小。在融合区域的中、后期，

的变化较小，曲线较为平缓，即转差频率直接速度辨识方法的作用强度值缓慢减小； 1) if

If the value is small, then in the early stage of the fusion area, as the error decreases,

The change is large, and the steepness of the curve is large, that is, the action intensity value of the slip frequency direct speed identification method decreases rapidly. In the middle and late stages of the fusion region,

The change of is small, and the curve is gentler, that is, the action strength value of the slip frequency direct speed identification method decreases slowly;

值较大，在融合区域的前期和中期，随着误差减小，

变化不大，曲线较为平缓，即转差频率直接速度辨识方法的作用强度缓慢减小。在融合区域的后期时，变化较大，曲线的陡度大，即转差频率直接速度辨识方法的作用强度值迅速减小。前面描述的

与作用强度变化的关系见图9。 2) if

The value is larger, in the early and middle stages of the fusion area, as the error decreases,

The change is not large, and the curve is relatively gentle, that is, the effect strength of the slip frequency direct speed identification method decreases slowly. At the later stage of the fusion region, The change is large, and the steepness of the curve is large, that is, the action intensity value of the slip frequency direct speed identification method decreases rapidly. previously described

The relationship with the change of action intensity is shown in Figure 9.

值可调整两种速度辨识的作用强度的分析，可得到

的选取依据： Different from the above selection

The value can be adjusted for the analysis of the strength of the two speed identifications, which can be obtained

is selected based on:

较大时，应赋予转差频率直接速度辨识方法较大的作用强度，以尽快消除偏差，提高响应速度，此时

应取较大的值。 1) The absolute value of the error between the expected speed value and the identification value when the motor is running

When the value is large, the slip frequency direct speed identification method should be assigned a greater action strength to eliminate the deviation as soon as possible and improve the response speed.

A larger value should be taken.

较小时，应赋予转差频率直接速度辨识方法较小的作用强度，而神经网络MRAS速度辨识方法给予较大的作用强度，以尽快地进入稳态，使速度辨识稳定，此时

应取较小的值。 2) The absolute value of the error between the expected speed value and the identification value when the motor is running

When the slip frequency is small, the slip frequency direct speed identification method should be given a smaller action strength, while the neural network MRAS speed identification method should be given a larger action strength, so as to enter the steady state as soon as possible and make the speed identification stable.

Should take a smaller value.

Table 1 and Table 2 are the calculated action intensity distribution tables:

Table 1. The action intensity distribution table of slip frequency direct speed identification method

Table 2. The action strength distribution table of the neural network MRAS speed identification method

In the table: E represents the grade of error threshold range. The error strength values in Table 1 and Table 2 correspond to each other.

的数值，按

数值查找对应的作用强度分布，确定在不同误差等级时的作用强度。  In specific applications, it is necessary to determine

value, press

Numerically find the corresponding action intensity distribution, and determine The strength of action at different error levels.

After finding the corresponding action intensity values of the neural network MRAS speed identification method and the slip frequency direct speed identification method, the two motor speeds identified by the neural network MRAS speed identification method and the slip frequency direct speed identification method are as follows: Values are fuzzy blended:

为转差频率直接速度辨识方法所识别出的电机转速值；

为神经网络MRAS速度辨识方法所识别出的电机转速值；

为最终识别出的电机动态运行时的转速确信值；

为查找作用强度分布表得到的转差频率直接速度辨识方法的作用强度值；

为查找作用强度分布表得到的神经网络MRAS速度辨识方法的作用强度值。 in,

is the motor speed value identified by the slip frequency direct speed identification method;

The motor speed value identified by the neural network MRAS speed identification method;

is the final identified speed assurance value of the motor during dynamic operation;

The action intensity value of the slip frequency direct speed identification method obtained by looking up the action intensity distribution table;

It is the action intensity value obtained by searching the action intensity distribution table of the neural network MRAS speed identification method.

The scheme of the present invention can be realized by using a digital control system with DSP as the core. For example, the computing speed of TMS320F2812 is as high as 150MIPS, and it can be used as speed identification, vector control algorithm and SVPWM wave generation at the same time.

After adopting the solution of the present invention, the speed identification effect of the control system is shown in FIG. 11 . Using the existing single neural network MRAS speed identification method, it can be seen from Figure 10 that the speed fluctuates greatly when the motor starts, and even negative fluctuations occur, the number of oscillations is large, and the adjustment time is long. At the same time, the velocity estimate converges to an erroneous value when switching from one steady state to another by a large amount. However, after adopting the fuzzy fusion speed identification method proposed by the present invention, it can be seen from Figure 11 that the speed identified by the motor at startup has almost no oscillation and fluctuation except for a few glitches, and the dynamic tracking performance of the speed command is relatively strong; At the same time, when switching from one steady state to another steady state, although there are a few glitches in the speed identification, it can correctly converge to the command given value, and the steady state accuracy is high, so that the motor can be realized Effective control.

Claims (3)

1. A fuzzy fusion identification method for the rotating speed of a motor without a speed sensor is characterized in that: identifying the rotating speed of the motor simultaneously by adopting a neural network MRAS speed identification method and a slip frequency direct speed identification method; distributing respective action intensity values for a neural network MRAS speed identification method and a slip frequency direct speed identification method, and calculating a confident value of the motor rotating speed by adopting the following formula:

the motor rotating speed value is identified by a slip frequency direct speed identification method;

the motor rotating speed value identified by the neural network MRAS speed identification method;

a rotation speed confident value for the finally identified motor operation;

the action strength value of the neural network MRAS speed identification method is obtained;

the value is the action intensity value of the slip frequency direct speed identification method;

distributing respective action intensity values for a neural network MRAS speed identification method and a slip frequency direct speed identification method, wherein the action intensity values comprise the following steps:

1) setting error threshold value ranges with continuous numerical values according to numerical values, and setting corresponding error threshold value ranges for each error threshold value range

Andwithin the same error thresholdAnd

satisfy the requirement of

Forming an action intensity distribution table;

2) real-time calculation of error between identification value of neural network MRAS speed identification method and motor rotating speed expected value

；

3) Judging the error in the step 2)

Within which error threshold range, the error threshold range is corresponding to

And

substitution formula

；

Step 1), comprising:

(1) marking 10 error threshold value ranges with continuous numerical values as ten grades of 0-9; setting the action intensity of a neural network MRAS speed identification method to be 0 and the action intensity of a slip frequency direct speed identification method to be 1 corresponding to the 9 th grade error threshold range; corresponding to the 0 th level error threshold range, the action intensity of the neural network MRAS speed identification method is 1, and the action intensity of the slip frequency direct speed identification method is 0;

(2) calculating the action intensity distribution of a neural network MRAS speed identification method and a slip frequency direct speed identification method within the error threshold range of 1 st to 8 th grades by adopting a symmetrical function;

and (5) making an action intensity distribution table.

2. The method of claim 1The fuzzy fusion identification method of the rotating speed of the motor without the speed sensor is characterized in that: step (2), comprising: according to the symmetrical type

And (3) function, when the levels 1 to 8 are calculated, the action intensity distribution of the neural network MRAS speed identification method and the slip frequency direct speed identification method is as follows:

，

wherein,the number of stages corresponding to the error threshold range;

is composed of

Parameters of the function;

respectively taking 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8 or 0.9,

each value of (1) is respectively corresponding to the action strength value of a group of mutually matched neural network MRAS speed identification method and slip frequency direct speed identification method.

3. The fuzzy fusion identification method of the rotating speed of the motor without the speed sensor according to claim 1, characterized in that: in step 2), calculating

The method of (1), comprising:

wherein,

the expected value of the rotating speed of the motor is obtained;

is the identification value of the neural network MRAS speed identification method.

Images