阅读说明：本技术 一种轴向线圈励磁辅助双凸极无刷直流电机的控制方法 (A kind of control method of axial direction coil magnetization auxiliary doubly-salient brushless DC generator ) 是由 刘爱民 王新宇 娄家川 任达 孟繁贵 于 2019-08-29 设计创作，主要内容包括：本发明提供一种轴向线圈励磁辅助双凸极无刷直流电机的控制方法,涉及电机的控制及调节技术领域。方法如下：步骤1：设置控制中央励磁线圈的初始权值、控制定子绕组的初始权值、双神经元时变学习率、RBF神经网络参数；步骤2：获取数据并计算双神经网络的输入；步骤3：根据输入量得到k时刻控制中央励磁线圈的控制量u<Sub>1</Sub>(k)和控制定子绕组控制量u<Sub>2</Sub>(k)；步骤4：采用自适应梯度下降法对RBF神经网络的中心向量与连接权值进行修正,根据f(k+1)和实际输出y(k+1)产生的偏差采用最速梯度下降学习算法对双神经元控制器连接权值进行修正；步骤5：令k＝k+1,返回步骤2。本方法实现电机定子绕组与中央励磁线圈最优联合控制,更好更快地调节电机转速和转矩。(The present invention provides a kind of control method of axial coil magnetization auxiliary doubly-salient brushless DC generator, is related to control and the regulation technology field of motor.Method is as follows: step 1: learning rate changing, RBF neural parameter when setting controls the initial weight of central magnet exciting coil, the initial weight for controlling stator winding, double neuron；Step 2: obtaining data and calculate the input of amphineura network；Step 3: the control amount u that the k moment controls central magnet exciting coil is obtained according to input quantity 1 (k) and control stator winding control amount u 2 (k)；Step 4: being modified using center vector of the self-adaption gradient descent method to RBF neural with connection weight, the deviation generated according to f (k+1) and reality output y (k+1) is modified double neuron controller connection weight using steepest descent learning algorithm；Step 5: enabling k=k+1, return step 2.This method realizes that motor stator winding and central magnet exciting coil optimal joint control, and faster and better adjusts motor speed and torque.)

1. a kind of control method of axial direction coil magnetization auxiliary doubly-salient brushless DC generator, it is characterised in that: including walking as follows It is rapid:

Step 1: setting controls the initial weight v of central magnet exciting coil_i1, control stator winding initial weight v_i2, double neuron When learning rate changing η₁、η₂And the parameter of RBF neural, the parameter include RBF neural weight, Basis Function Center to Amount and width；

Step 2: obtaining data, the data include the specified rate r (k) and actual motor output speeds y of k moment motor speed (k), the specified rate r (k-1) and actual motor output speeds y (k-1) of k-1 moment motor speed, k-2 moment motor speed Specified rate r (k-2) and actual motor output speeds y (k-2) calculates the input x of amphineura network according to the data of acquisition₁ (k)、x₂(k)、x₃(k)；

Wherein, e (k) represents the error of r (k) Yu y (k)；E (k-1) represents the error of r (k-1) Yu y (k-1), and e (k-2) represents r (k-2) with the error of y (k-2)；

Step 3: the input quantity x obtained according to step 2_i(k) the control center magnet exciting coil at the k moment that PID controller provides is obtained Control amount u₁(k) and control stator winding control amount u₂(k), by u₁(k) and u₂(k) export respectively to RBF neural with And the driving circuit of central exciting coil drive circuit and control stator winding；Central exciting coil drive circuit and control stator The driving circuit of winding is for controlling central magnet exciting coil and motor；

Wherein, i=1,2,3；u₁(k-1) control amount of central magnet exciting coil is controlled for the k-1 moment；u₂(k-1) it is controlled for the k-1 moment The control amount of stator winding processed；v_i1、v_i2For the connection weight in double neuron controller；

Step 4: calculating the output f (k+1) of RBF neural, wherein f (*) is the Speed Identification of RBF neural output；It adopts It is modified with center vector of the self-adaption gradient descent method to RBF neural with connection weight, according to f (k+1) and reality The deviation that output y (k+1) generates is modified double neuron controller connection weight using steepest descent learning algorithm；

Step 5: enabling k=k+1, return step 2.

2. a kind of control method of axial coil magnetization auxiliary doubly-salient brushless DC generator according to claim 1, Be characterized in that: specific step is as follows for the step 4:

Step 4.1: being learnt according to the deviation that given output r (k+1) and reality output y (k+1) generate using steepest descent Algorithm is modified double neuron controller connection weight；

Quadratic performance index J is found out according to quadratic performance index function_c, formula is as follows:

K+1 moment double neuron controller connection weight v_iθ(k+1) be modified to along J_cTo v_iθ(k) negative gradient direction search, Wherein θ=[1,2], formula are as follows:

Step 4.2: determining that the weight of double neuron is modified according to revised RBF neural；

Double neuron modified weight formula is obtained according to the output of the RBF neural after amendment weight are as follows:

Wherein η₁, η₂For the when learning rate changing of double neuron；

RBF neural is made of input layer, hidden layer and output layer, the input of RBF neural by the real-time revolving speed y of motor, u₁、u₂Output composition；Hidden layer chooses gaussian kernel function, and k-th implies the output of layer unit are as follows:

Wherein, X_iFor i-th of input quantity；c_kFor the midpoint of k-th of hidden layer node；b_kFor the width of k-th of hidden layer node；

The output equation of RBF neural are as follows:

Wherein ω_kFor the connection weight of output layer and k-th hidden layer node；N is hidden layer node number.

Technical field

The present invention relates to the control of motor and regulation technology fields more particularly to a kind of axial coil magnetization to assist double-salient-pole The control method of brshless DC motor.

Background technique

Switched reluctance machines have many advantages, such as that structure is simple, reliable performance, and good application is all obtained in every field, Become the research hotspot of scholar.As electric machines control technology, power electronics integrated technology are constantly progressive, chip, function are controlled The rapid development of the device materials such as rate device, doubly-salient brushless DC generator become in recent years on the basis of switched reluctance machines A kind of new special motor to grow up, application field will be widened further.

Doubly-salient brushless DC generator, due to the shutdown of every phase current and open-minded, leads to the torque of motor in commutation process Pulsation is big, simultaneously because the introducing of central magnet exciting coil, electromagnetic property become more complicated, also become more to be stranded in control The control effect of difficulty, traditional linear control method is undesirable.

Summary of the invention

It is auxiliary the technical problem to be solved by the present invention is in view of the above shortcomings of the prior art, provide a kind of axial coil magnetization The control method of doubly-salient brushless DC generator is helped, this method is using double neuron PID controller and RBF neural identification system System achievees the purpose that on-line identification and On-line Control, realizes that motor stator winding and central magnet exciting coil optimal joint control, more It is good quickly to adjust motor speed and torque.

In order to solve the above technical problems, the technical solution used in the present invention is:

The present invention provides a kind of control method of axial coil magnetization auxiliary doubly-salient brushless DC generator, including walks as follows It is rapid:

Step 1: setting controls the initial weight v of central magnet exciting coil_i1, control stator winding initial weight v_i2, double minds When learning rate changing η through member₁、η₂And the parameter of RBF neural, the parameter include RBF neural weight, in basic function Heart vector and width；

Step 2: obtaining data, the data include specified rate r (k) and the output of actual motor of k moment motor speed Revolving speed y (k), the specified rate r (k-1) of k-1 moment motor speed and actual motor output speeds y (k-1), k-2 moment motor The specified rate r (k-2) and actual motor output speeds y (k-2) of revolving speed calculate the defeated of amphineura network according to the data of acquisition Enter x₁(k)、x₂(k)、x₃(k)；

Wherein, e (k) represents the error of r (k) Yu y (k)；E (k-1) represents the error of r (k-1) Yu y (k-1), e (k-2) generation The error of table r (k-2) and y (k-2)；

Step 3: the input quantity x obtained according to step 2_i(k) it encourages in the control center for obtaining the k moment that PID controller provides The control amount u of magnetic coil₁(k) and control stator winding control amount u₂(k), by u₁(k) and u₂(k) it exports respectively to RBF nerve The driving circuit of network and central exciting coil drive circuit and control stator winding；Central exciting coil drive circuit and control The driving circuit of stator winding processed is for controlling central magnet exciting coil and motor；

Wherein, i=1,2,3；u₁(k-1) control amount of central magnet exciting coil is controlled for the k-1 moment；u₂(k-1) be k-1 when Carve the control amount of control stator winding；v_i1、v_i2For the connection weight in double neuron controller；

Step 4: calculating the output f (k+1) of RBF neural, wherein f (*) is that the identification of RBF neural output turns Speed；Be modified using center vector of the self-adaption gradient descent method to RBF neural with connection weight, according to f (k+1) and The deviation that reality output y (k+1) is generated carries out double neuron controller connection weight using steepest descent learning algorithm Amendment；

Step 5: enabling k=k+1, return step 2.

Specific step is as follows for the step 4:

Step 4.1: steepest descent is used according to the deviation that given output r (k+1) and reality output y (k+1) generate Learning algorithm is modified double neuron controller connection weight；

Quadratic performance index J is found out according to quadratic performance index function_c, formula is as follows:

K+1 moment double neuron controller connection weight v_iθ(k+1) be modified to along J_cTo v_iθ(k) negative gradient direction is searched Rope, wherein θ=[1,2], formula are as follows:

Step 4.2: determining that the weight of double neuron is modified according to revised RBF neural；

Double neuron modified weight formula is obtained according to the output of the RBF neural after amendment weight are as follows:

Wherein η₁, η₂For the when learning rate changing of double neuron.

RBF neural is made of input layer, hidden layer and output layer, and the input of RBF neural is turned in real time by motor Fast y, u₁、u₂Output composition；Hidden layer chooses gaussian kernel function, and k-th implies the output of layer unit are as follows:

Wherein, X_iFor i-th of input quantity；c_kFor the midpoint of k-th of hidden layer node；b_kFor the width of k-th of hidden layer node Degree.

The output equation of RBF neural are as follows:

Wherein ω_kFor the connection weight of output layer and k-th hidden layer node；N is hidden layer node number.

The beneficial effects of adopting the technical scheme are that a kind of axial coil magnetization auxiliary provided by the invention The control method of doubly-salient brushless DC generator constitutes doubly-salient brushless using the single neuron with self study adaptive ability Direct current generator double neuron adaptive controller, can adapt to the variation of environment, have preferable robustness.RBF network simultaneously Online identification is carried out, real-time online updates the gradient information of double neuron controller, then by the defeated of double neuron controller It controls stator winding and central magnet exciting coil respectively out, to reduce torque pulsation, preferably controls motor speed；By right Central magnet exciting coil control, sets up magnetic circuit quickly, revolving speed quickly rises, while small electromotor produces during commutation Raw torque pulsation achievees the effect that torque pulsation inhibited.

Detailed description of the invention

Fig. 1 is the doubly-salient brushless DC generator structure chart provided in an embodiment of the present invention that central magnet exciting coil is added；

Fig. 2 is the doubly-salient brushless DC generator magnetic circuit trend provided in an embodiment of the present invention that central magnet exciting coil is added；

Fig. 3 is provided in an embodiment of the present invention based on RBF network on-line identification double neuron PID self-adaptation control method stream Cheng Tu；

Fig. 4 is double neuron structure chart provided in an embodiment of the present invention；

Fig. 5 is RBF neural network structure figure provided in an embodiment of the present invention；

Fig. 6 is the double neuron PID control system figure provided in an embodiment of the present invention based on RBF neural；

Wherein, 1. left Stator and Rotor Windings, 2. right Stator and Rotor Windings, 3. central magnet exciting coils, 4. stator winding magnetic circuits trend, 5. central magnet exciting coil magnetic circuit trend.

Specific embodiment

With reference to the accompanying drawings and examples, specific embodiments of the present invention will be described in further detail.Implement below Example is not intended to limit the scope of the invention for illustrating the present invention.

The method of the present embodiment is as described below.

The present invention is based on a kind of axial coil magnetizations to assist doubly-salient brushless DC generator, in traditional dual protruding pole brushless DC Central magnet exciting coil is added on the basis of motor, electric machine structure figure is as shown in Figure 1, in the two sets of Stator and Rotor Windings (1) (3) in left and right Between central magnet exciting coil (2) is added.The electromagnetic torque of motor is provided by stator winding, and magnetic torque is by the central excitation wire that is added Circle provides, and increases the magnetic circuit trend of motor, and motor magnetic circuit trend is as shown in Figure 2.It is walked in originally complicated stator winding magnetic circuit Central magnet exciting coil magnetic circuit trend (5) is added on the basis of to (4), becomes more complicated the decoupling of motor, for point of motor Analysis becomes more difficult.

Therefore control is carried out to motor using the double neuron pid control algorithm based on RBF neural to reach preferably Control precision.Double neuron PID control system based on RBF neural as shown in fig. 6, speed preset and actual speed mistake Difference and the gradient information of RBF neural output are input to double neuron controller, control central excitation wire respectively to generate The control amount of circle and motor, the two control amounts control central magnet exciting coil and motor by two driving circuits. The actual speed and two control amounts that motor generates are input to RBF neural and generate gradient information.According to this hair of above system Bright to provide a kind of double neuron PID control method based on RBF neural of New-type electric machine, this method is in single neuron control It is improved on the basis of algorithm processed, obtains double neuron control algolithm, be then applied on New-type electric machine.Using RBF nerve net The gradient information of network on-line identification system real-time update double neuron PID controller, the output of double neuron PID controller are led to respectively It crosses three-phase drive circuit control motor stator winding and full-bridge circuit controls central magnet exciting coil.Central magnet exciting coil can be effective Auxiliary rate controlling, reduce the fluctuation of torque, improve speed adjustable range, while can have better anti-interference.

The present invention provides a kind of control method of axial coil magnetization auxiliary doubly-salient brushless DC generator, as shown in figure 3, Include the following steps:

Step 1: setting controls the initial weight v of central magnet exciting coil_i1, control stator winding initial weight v_i2, double minds When learning rate changing η through member₁、η₂And the parameter of RBF neural, the parameter include RBF neural weight, in basic function Heart vector and width generally take the random number between [- 1,1]；

The initial weight { 0.1,0.2,0.3 } of central magnet exciting coil is controlled in the present embodiment；Control the initial of stator winding Weight { 0.15,0.25,0.35 }；Double neuron when learning rate changing { 0.01,0.1 }；

RBF neural weight { 0.11,0.21,0.13,0.14,0.21,0.31 }；

Basis Function Center vector { 0.1,0.4,0.1,0.2,0.3,0.15 }, 0.2,0.3,0.15,0.23,0.23, 0.5},{0.15,0.42,0.11,0.23,0.43,0.15}}；

Width { 0.11,0.21,0.13,0.14,0.21,0.31 }

Step 2: obtaining data, the data include specified rate r (k) and the output of actual motor of k moment motor speed Revolving speed y (k), the specified rate r (k-1) of k-1 moment motor speed and actual motor output speeds y (k-1), k-2 moment motor The specified rate r (k-2) and actual motor output speeds y (k-2) of revolving speed calculate the defeated of amphineura network according to the data of acquisition Enter x₁(k)、x₂(k)、x₃(k), as shown in Figure 4；

Wherein, e (k) represents the error of r (k) Yu y (k)；E (k-1) represents the error of r (k-1) Yu y (k-1), e (k-2) generation The error of table r (k-2) and y (k-2)；

Step 3: the input quantity x obtained according to step 2_i(k) it encourages in the control center for obtaining the k moment that PID controller provides The control amount u of magnetic coil₁(k) and control stator winding control amount u₂(k), by u₁(k) and u₂(k) it exports respectively to RBF nerve The driving circuit of network and central exciting coil drive circuit and control stator winding；Central exciting coil drive circuit and control The driving circuit of stator winding processed is for controlling central magnet exciting coil and motor；

Wherein, i=1,2,3；u₁(k-1) control amount of central magnet exciting coil is controlled for the k-1 moment；u₂(k-1) be k-1 when Carve the control amount of control stator winding；v_i1、v_i2For the connection weight in double neuron controller；Function as traditional PI D control Three control parameters of device processed.u₁, u₂The respectively control output of double neuron controller.In this paper double neuron PID control system In system, the control parameter of conventional delta formula PID is changed into v_i, v is realized by on-line identification system_iReal-time adjusting, to realize Central magnet exciting coil auxiliary control optimal effectiveness.

By x_iIdentical as increment type PID structure as input quantity, the formula in step 2 does not need to add up, and controlling increment is only It is related with nearest 3 sampled values, it is easy to obtain relatively good control effect by weighting processing.

Step 4: calculating the output f (k+1) of RBF neural, wherein f (*) is that the identification of RBF neural output turns Speed；Be modified using center vector of the self-adaption gradient descent method to RBF neural with connection weight, according to f (k+1) and The deviation that reality output y (k+1) is generated carries out double neuron controller connection weight using steepest descent learning algorithm Amendment；Specific step is as follows:

Step 4.1: steepest descent is used according to the deviation that given output r (k+1) and reality output y (k+1) generate Learning algorithm is modified double neuron controller connection weight；

Double neuron PID control system is using self-adaption gradient descent method to the center vector and connection weight of identification system It is modified and steepest descent learning algorithm is modified double neuron controller connection weight.By to weighting Adaptive, self organizing function is realized in the adjustment of coefficient, and learning rules are exactly the algorithm for adjusting weight, it is double neuron PID The core of control system.The quadratic performance index for introducing error originated from input, finds out secondary performance according to quadratic performance index function and refers to Mark J_c, formula is as follows:

For the optimum control for realizing self-adaptive PID, k+1 moment double neuron controller connection weight v_iθ(k+1) amendment For along J_cTo v_iθ(k) negative gradient direction is searched for, wherein θ=[1,2], so that performance indicator is minimum, formula are as follows:

Step 4.2: determining that the weight of double neuron is modified according to revised RBF neural；

After identification network learns by limited times, output gradually approaches the output of object, i.e. f (k+1) ≈ y (k+ 1), so double neuron updates gradient information and has respectively:

Double neuron modified weight formula is obtained according to the output of the RBF neural after amendment weight are as follows:

Wherein f (k+1) is the identification output infinitely approached after identification system limited times learns, η₁, η₂For double neuron when Learning rate changing.

RBF neural is made of input layer, hidden layer and output layer, as shown in Figure 5；The input of RBF neural by The real-time revolving speed y of motor, double neuron controller u₁、u₂Output composition；Hidden layer generally chooses gaussian kernel function, gaussian kernel function Local acknowledgement can be generated to input, k-th implies the output of layer unit are as follows:

Wherein, X_iFor i-th of input quantity；c_kFor the midpoint of k-th of hidden layer node；b_kFor the width of k-th of hidden layer node Degree.

The output equation of RBF neural are as follows:

Wherein ω_kFor the connection weight of output layer and k-th hidden layer node；N is hidden layer node number；

Step 5: enabling k=k+1, return step 2.

Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used To modify to technical solution documented by previous embodiment, or some or all of the technical features are equal Replacement；And these are modified or replaceed, model defined by the claims in the present invention that it does not separate the essence of the corresponding technical solution It encloses.