Decoupling control method for inverse system of bearingless composite rotor cage type asynchronous motor

Abstract

The invention relates to a decoupling control method for an inverse system of a bearingless composite rotor cage type asynchronous motor. The bearing-free composite rotor cage type asynchronous motor inverse system decoupling control method comprises the following steps: S1, establishing an ISA-GA optimized SVM inverse system; s101, establishing an SVM (Support Vector Machine) model: on the reversible basis of the CCR-BIM system, constructing the SVM model to identify and fit an inverse system of the CCR-BIM system; s102, carrying out optimization design on an SVM kernel function based on ISA-GA; s2, an SVM inverse system decoupling control system based on ISA-GA optimization; according to the bearing-free composite rotor cage type asynchronous motor inverse system decoupling control method, through selection operator crossing and mutation operation in an improved SA-GA algorithm, the search efficiency and the local optimization capacity are improved, the inverse system and an original system are connected in series to form a pseudo-linear system, CCR-BIM linearization and decoupling control are achieved, and the decoupling precision is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of electric transmission control equipment, and in particular relates to a decoupling control method for an inverse system of a bearingless composite rotor cage-type asynchronous motor. Background Art

[0002] Asynchronous motors (IM) have the advantages of simple structure, reliable operation, good starting performance, high efficiency, large starting torque, and convenient speed regulation. They are widely used in medical equipment, aerospace, transportation, wind power generation, mining machinery, flywheel energy storage, etc. Traditional asynchronous motors support the shaft through mechanical bearings. As the speed increases, it will cause severe friction and wear, which will reduce its own life and reduce the reliability of the motor system.

[0003] In order to solve this problem, the existing literature Lu C, Yang Z, Sun X, et al. Design and Multi-Objective Optimization of a Composite Cage Rotor Bearingless Induction Motor [J]. Electronics, 2023, 12 (3): 775. and the literature Lu C, Yang Z, Sun X, et al. Vibration compensation control strategy of composite cage rotor bearingless induction motor based on fuzzy coefficient adaptive-linear-neuron method [J]. ISA transactions, 2024. proposed a bearingless composite rotor cage asynchronous motor. Compared with the traditional asynchronous motor, the bearingless composite rotor cage asynchronous motor has the advantages of no friction and wear, no need for lubrication and long service life. Therefore, it has broad application prospects in special fields such as clean, easy to corrode, and vacuum. The bearingless composite rotor cage asynchronous motor (CCR-BIM) is a multivariable, nonlinear and strongly coupled system. During the operation of the motor, there is a strong coupling relationship between the rotor displacement, speed and air gap flux. The key to achieve precise control of CC-RBIM is to linearize its nonlinear system, so that the CCR-BIM system uses a linear controller to empty the box and realize the dynamic decoupling control of the system. Inverse system linearization is an effective method of nonlinear conversion linearization, which can decouple the original system into multiple pseudo-linear subsystems. Support vector machine (SVM) has good nonlinear modeling and generalization capabilities on small samples. By using support vector machine (SVM) to approximate the inverse system of CCR-BIM motor system, the controlled object can be compensated as a linear and decoupled pseudo-linear system. However, the control algorithm proposed in the existing technical literature does not optimize the support vector machine parameters, and the decoupling accuracy still has room for further improvement. Summary of the invention

[0004] The purpose of the present invention is to provide a bearingless composite rotor cage asynchronous motor inverse system decoupling control method with simple structure and reasonable design in order to solve the above problems.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] A method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor comprises the following steps:

[0007] S1, ISA-GA optimized SVM inverse system establishment;

[0008] S101, SVM model establishment: Based on the reversibility of the CCR-BIM system, a SVM model is constructed to identify and fit the inverse system of the CCR-BIM system;

[0009] S102. Optimization design of SVM kernel function based on ISA-GA;

[0010] S2. Decoupling control system of SVM inverse system based on ISA-GA optimization: obtain the output signal of the CCR-BIM system, input the output signal of the CCR-BIM system into the SVM inverse system optimized by ISA-GA, obtain the control signal processed by the SVM inverse system, and connect the control signal in series with the input signal of the original system to form a pseudo-linear system.

[0011] Furthermore, in step S101, it specifically includes using a convex quadratic programming problem to construct an optimal classification hyperplane and construct a Lagrange function with n training samples:

[0012]

[0013] ω is the normal vector, b is the bias term, α i is the Lagrange multiplier, v i is the feature vector of the i-th training sample, z i is the label of the i-th training sample, ||ω|| 2 is the square of the norm of the normal vector ω.

[0014] Limit the parameter to α i ≥0, the objective function expression is:

[0015]

[0016] There is a saddle point in the Lagrange function, and it satisfies the stability condition, feasibility condition and complementary relaxation condition in the KKT condition. The objective function expression is equivalent to solving a set of dual problems. The specific expression is as follows:

[0017]

[0018] Fix the parameter α and take partial derivatives with respect to ω and b to find the minimum value of the Lagrange function with respect to ω and b. The mathematical expression is:

[0019]

[0020] After calculating the Lagrange function, we can get:

[0021]

[0022] Then the objective function will be transformed into:

[0023]

[0024] stα i ≥0,i=1,2,...,n

[0025]

[0026] The hand-pushed Sequential Minimal Optimization (SMO) method is used to recursively calculate the Lagrange multiplier α, and from this, ω and b are calculated, and the maximum margin hyperplane can be determined.

[0027] Furthermore, in step S102, the following steps are specifically included:

[0028] S1021. Initialize the population: randomly generate the initial population, each individual represents a solution;

[0029] S1022. Genetic operations: Evolve the population and update the fitness of individuals through selection, adaptive crossover and mutation operations;

[0030] S1023, simulated annealing operation: performing a simulated annealing operation on each individual to locally search for a better solution;

[0031] S1024, determining a stop condition: determining whether a condition for stopping iteration is met according to a preset stop condition;

[0032] S1025. Return the optimal solution: return the optimal individual as the result.

[0033] Furthermore, in step S1022, the specific expression of the selection operation is:

[0034]

[0035] Set the adaptive coefficient to k c The adaptive crossover operator value expression is:

[0036]

[0037] Where p is a settable fractional value ranging from 0 to 1, and f avg is the average fitness value, f max is the maximum fitness value, It is used to measure the current population dispersion;

[0038] Set the adaptive coefficient to k m The mutation operator value expression is:

[0039] p is a settable fractional value ranging from 0 to 1.

[0040] Further, in step S1023, the simulated annealing algorithm will compare the new individual fitness value obtained by the genetic algorithm with the old individual fitness value to decide whether to accept the inferior solution. The specific calculation formula is as follows:

[0041]

[0042] Where fit(x') is the fitness value of the new individual, fit(x) is the fitness value of the old individual, and T is the annealing temperature.

[0043] Further, in step S2, the second-order integral link, the first-order integral link and the first-order integral link are used in the displacement part, the speed regulation part and the flux regulation part respectively; and the displacement subsystem, the speed subsystem and the flux subsystem in the x and y directions are constructed, and the input variables are: And output variable: Z = [i 2sd ,i 2sq ,i 1rd ,i 1rq ] T Sampling is performed, and dynamic decoupling is achieved using the designed ISA-GA support vector machine inverse system decoupling control strategy.

[0044] Furthermore, the sampling process of input data and output data is as follows: the sampling time is set to 10s, the sampling interval is 1e-2, and 80% of the sample data is divided into the training sample set V train and the corresponding label set Z train , and the remaining 20% ​​is used as the test set V test and the corresponding label set Z test The training sample set is used to build a regression model, and the test set is used to evaluate the fit of the regression model to the data.

[0045] Furthermore, the input parameters and output parameters are normalized as follows:

[0046]

[0047] In the formula, Var(Z train (i))、 Var(Z train (i)) are the mean and variance of the i-th column of the input and output matrices of the training sample set, respectively; n is the number of samples in the training set, and m is the number of samples in the test set;

[0048] After the input vector is normalized, it is sent to the kernel function for calculation. After the calculation is completed, it is multiplied with the Lagrange multiplier, and the multiplication results are accumulated one by one. After the accumulation is completed, the bias constant is superimposed to obtain the corresponding output variable;

[0049] Then, four SVM learning machines are constructed, and their initial regression model functions are:

[0050]

[0051] In the formula

[0052] Finally, post-process the regression results output by the test set:

[0053]

[0054] The genetic generation, crossover rate and mutation rate in ISA-GA are set, and the fitness of the system is calculated based on these initial parameters. The calculation formula of the fitness function is as follows:

[0055]

[0056] According to the calculation formula of the fitness function, the fitness value corresponding to the system is calculated. The selection and adaptive crossover mutation algorithm used in the ISA-GA algorithm can obtain the corresponding offspring set and iterate the above process.

[0057] Beneficial effect: The present invention improves the search efficiency and local optimization ability through the crossover and mutation operations of the selection operator in the improved SA-GA algorithm. The optimized SVM constructs an inverse system and is connected in series with the original system to form a pseudo-linear system, thereby realizing the linearization and decoupling control of CCR-BIM and improving the decoupling accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flow chart of the steps of the present invention;

[0059] Figure 2 is an analysis process diagram of the nonlinear sample processing of the present invention;

[0060] Figure 3 is a flow chart of the ISA-GA algorithm of the present invention;

[0061] Figure 4 is a schematic diagram of the topology of the support vector machine optimized by ISA-GA of the present invention;

[0062] Figure 5 It is the inverse system structure diagram of the support vector machine optimized by ISA-GA of the present invention;

[0063] Figure 6It is the ISA-GA support vector machine inverse system decoupling control block diagram of the present invention;

[0064] Figure 7 is the x-axis and y-axis radial displacement response diagram of the present invention;

[0065] Figure 8 is a speed response curve diagram of the present invention;

[0066] Fig. 9 is a magnetic flux response curve diagram of the present invention;

[0067] Fig.10 is a current response curve diagram of the suspension force winding of the present invention to a sudden load;

[0068] Fig.11 is a torque winding current response curve diagram of the present invention under sudden load;

[0069] Fig.12 is an x-axis displacement response curve diagram of the present invention;

[0070] Fig.13 is a y-axis displacement response curve diagram of the present invention;

[0071] Fig.14 is a speed response curve diagram of the present invention;

[0072] Fig.15 is a torque winding current waveform diagram of the present invention;

[0073] Fig.16 It is the suspension force winding current waveform of the present invention. DETAILED DESCRIPTION

[0074] The present application is further described in detail below in conjunction with the accompanying drawings. It is necessary to point out here that the following specific implementation methods are only used to further illustrate the present application and cannot be understood as limiting the scope of protection of the present application. Technical personnel in this field can make some non-essential improvements and adjustments to the present application based on the above application content.

[0075] A method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor. Figure 1 As shown, the following steps are included:

[0076] S1, ISA-GA optimized SVM inverse system establishment;

[0077] S101, SVM model establishment: Based on the reversibility of the CCR-BIM system, a SVM model is constructed to identify and fit the inverse system of the CCR-BIM system;

[0078] S102. Optimization design of SVM kernel function based on ISA-GA;

[0079] S2. Decoupling control system of SVM inverse system based on ISA-GA optimization: The output signal of the CCR-BIM system is obtained, and the output signal of the CCR-BIM system is input into the SVM inverse system optimized by ISA-GA to obtain the control signal processed by the SVM inverse system. This signal is connected in series with the input signal of the original system to form a pseudo-linear system. Through the construction of the pseudo-linear system, the coupling between the subsystems in the CCR-BIM system is eliminated, and the linearization and decoupling control of the system are realized.

[0080] It should be noted that compared with the traditional asynchronous motor, the stator of CCR-BIM has two sets of windings: torque winding and suspension force winding. The pole pairs of the torque winding and the suspension winding are recorded as P1 and P2 respectively. If the two satisfy P1=P2±1 and the electrical angular frequency of the two windings dj1=dj2, the interaction of the magnetic fields generated by the windings with different pole pairs can generate a controllable radial force to achieve stable suspension of the CCR-BIM rotor. The topological design of CCR-BIM follows the structure of the traditional asynchronous motor, which is mainly composed of a stator, a rotor and a shaft. The rotor of CCR-BIM is a specially designed composite cage rotor structure, which consists of an outer rotor and an inner rotor. The outer rotor is a solid layer made of high magnetic permeability silicon steel material. The solid structure made of high magnetic permeability silicon steel is affected by the skin effect, and the current will be concentrated on the surface of the silicon steel. The composite cage rotor can obtain a larger starting torque, so that the motor can overcome greater resistance during the starting process, improving the starting performance of CCR-BIM. The inner rotor is composed of a rotor core, a specially designed bar set and a rotating shaft. In the special bar set structure of the inner rotor squirrel cage structure, a single bar set consists of two symmetrical bars and two end rings at the head and tail, forming a unique bar closed loop structure. Since the inner rotor core has 20 slots, and a set of bar sets is installed for every two symmetrical slots, a total of 10 sets of bar sets are required to fill the inner rotor core slots. The bar sets are installed one by one according to the rules;

[0081] The induction mechanism of the special squirrel cage structure is as follows: In the motor magnetic field constructed by the CCR-BIM three-phase symmetrical winding, the magnetomotive force is used to describe the distribution of the magnetic field in time and space. The magnetomotive force f(x, t) at any circumferential position x of the cage rotor can be expressed as:

[0082] f(x2,t)=Fcos(djt-Px) Formula 1

[0083] Where F is the fundamental magnetomotive force amplitude, dj is the angular velocity of magnetic field rotation, and P is the number of winding pole pairs.

[0084] The distribution of the conductor current is analyzed separately when only the torque winding and the suspension force winding are energized. Randomly select any point x1 on the cage circumference, and mark its symmetrical position on the circumference as x2. The number of pole pairs of the torque winding P1 = 1, the number of pole pairs of the suspension force winding P2 = 2, and the mathematical expression of the magnetomotive force can be expressed as:

[0085]

[0086] In the torque winding magnetic field, the induced electromotive force at points x1 and x2 are in opposite directions, resulting in a closed current loop in a specific single bar set, and the torque winding can induce the special cage rotor. In the suspension force winding magnetic field, the induced electromotive force at points x1 and x2 are in the same direction, resulting in the currents in the closed single bar set canceling each other out, and the total current is zero, so the suspension force winding cannot induce the special cage rotor;

[0087] This indicates that only the induced current generated by the torque winding exists in the specially designed cage-type conductor, and the induced current generated by the suspension force winding is offset. The specially designed cage-type rotor only senses the torque winding, and only the torque winding needs to be considered in the torque mathematical model. The induction of the suspension force winding to the special cage-type rotor is shielded, which also improves the control accuracy of the torque winding.

[0088] According to the existing literature Lu C, Yang Z, Sun X, et al. Design and Multi-Objective Optimization of a Composite Ca ge Rotor Bearingless Induction Motor [J]. Electronics, 2023, 12 (3): 775., the CCR-BIM suspension force mathematical model is expressed as:

[0089]

[0090] In the formula, F x 、F y are the resultant force components of the radial force in the x-axis and y-axis directions respectively; B 1M , B 2M are the electromagnetic induction of the torque winding and the suspension force winding respectively; r is the rotor radius; l is the effective length of the rotor; μ0 is the air magnetic permeability, and λ is the initial phase angle difference between the torque winding and the suspension force winding.

[0091] T e =L 1m (i 1dr i 1qs -i 1ds i 1qr )=i 1qs ψ 1d -i 1ds ψ1q Formula 4

[0092] In the formula, ψ 1d and ψ 1q They represent the air gap flux components of the d-axis and q-axis of the torque winding, L1 is the self-inductance of the torque winding, and i 1dr and i 1qr They represent the rotor current components in the d-axis and q-axis directions of the torque winding, respectively, i 1ds and i 1qs They represent the stator current components in the d-axis and q-axis directions of the torque winding, respectively, and L 1m is the self-inductance of the torque winding.

[0093] Torque motion equation expression:

[0094]

[0095] Where, T L is the load torque; J is the moment of inertia.

[0096] Before the inverse system is established, the reversibility analysis of the system should be carried out. According to the inverse system principle, the output variables, state variables and input variables of the CCR-BIM system are defined as follows: The output variable is defined as: Z = [x, y, ω r ,ψ], the state variable is defined as: The input variable is defined as: U = [i 2sd ,i 2sq ,i 1rd ,i 1rq ] T , the state equation of CCR-BIM is:

[0097]

[0098] The state equation of CCR-BIM obtains the partial derivative of the input variable U and the Jacobian matrix:

[0099]

[0100] The determinant of the Jacobian matrix A is detA≠0, and the relative order of the system is α=[α1,α2,α3,α4]=[2,2,1,1], satisfying α1+α2+α3+α4≤6. Therefore, the system meets the reversibility judgment condition of the inverse system.

[0101] The inverse system of the state equation of the CCR-BIM system is expressed as:

[0102]

[0103] In actual control system applications, the complexity of the CCR-BIM state equation is too high, and conventional analytical methods cannot accurately obtain the regression model of CCR-BIM. In order to realize the inverse system decoupling control of CCR-BIM, it is necessary to use support vector machine (SVM) to build a more accurate and effective CCR-BIM inverse system model.

[0104] In a specific implementation manner, in n-dimensional space, the geometric interval d between a point and a surface is expressed as:

[0105]

[0106] In the formula, y i (<ω,v>+b)≥1(i=1,2,3,···,n).

[0107] According to the above analysis, the scaling of ω and b has little effect on the geometric interval. Therefore, d min = 1. In addition, solve The process is converted into calculation Therefore, the process of finding the minimum value is as follows:

[0108]

[0109] The above process converts the original problem into a convex quadratic programming problem under linear constraints. To solve this problem, a Lagrange function with n training samples is constructed:

[0110]

[0111] ω is the normal vector, b is the bias term, α i is the Lagrange multiplier, v i is the feature vector of the i-th training sample, z i is the label of the i-th training sample, ||ω|| 2 is the square of the norm of the normal vector ω.

[0112] Limit the parameter to α i ≥0, the objective function expression is:

[0113]

[0114] There is a saddle point in Formula 11, and the stability condition, feasibility condition and complementary relaxation condition in the KKT condition are satisfied. Formula 12 is equivalent to solving a set of dual problems. The specific expression is as follows:

[0115]

[0116] Fix the parameter α and take partial derivatives with respect to ω and b to find the minimum value of ω and b in equation (10). The mathematical expression is:

[0117]

[0118] Substituting the equation corresponding to Formula 14 into the Lagrange function shown in Formula 11 yields:

[0119]

[0120] Then the objective function in the original formula 13 will be converted to:

[0121]

[0122] The Lagrange multiplier α is calculated recursively by hand-pushing Sequential Minimal Optimization (SMO), and ω and b are calculated from it. The maximum margin hyperplane can be determined, and the classification function can be expressed as:

[0123]

[0124] like Figure 2 As shown: (a) the sample points are inseparable; (b) they are inseparable in the sample space; (c) the sample points are separable;

[0125] When the outlier is located Figure 2 (a) shows the state, the sample points will be linearly inseparable. Data measurement errors or noise will introduce outliers, which will inevitably affect the interval of the maximum interval hyperplane. This situation can be effectively improved by introducing a loss function, which is expressed as follows:

[0126]

[0127] In order to make the maximum margin hyperplane meet the constraints at any position, the deviation needs to be constrained accordingly. Here, the deviation is squared and then added to the objective function in an accumulated form. The optimization goal after the objective function in Formula 10 is adjusted is:

[0128]

[0129] stz i (<ω,v i >+b)+e i =1,i=1,2,...,n

[0130] In the formula, γ is the parameter that controls the weight.

[0131] Formula 18 performs square sum operation on the deviations, and the Lagrange function is adjusted accordingly to become:

[0132]

[0133] The corresponding objective function is expressed as follows:

[0134]

[0135] Formula 20 is the optimized Lagrange function. It can be observed that the original problem has been transformed into a problem of solving a system of linear equations after processing.

[0136] Figure 2 (b) shows the situation where the two types of samples cannot be separated in the sample space, that is, the sample points themselves are in a nonlinear state. The solution is to map the sample to a higher-dimensional space, such as Figure 2 (c) As shown in Figure 2, the nonlinear sample classification process mainly includes two stages: first, the sample is mapped to a higher-dimensional feature space, and then a linear learning machine is used in this high-dimensional space for classification.

[0137] The specific process is: In two-dimensional space, the coordinates can be expressed as (v1, v2), then the equation of any curve can be written as:

[0138]

[0139] Taking five-dimensional space as an example, suppose w3=v1v2,w4=v1,w5=v2,then the curve equation in two-dimensional space can be expressed as:

[0140]

[0141] Therefore, in the newly constructed coordinate system, by introducing the hyperplane equation, the sample points that were originally linearly inseparable in the low-dimensional space can be linearly separated in the high-dimensional space. The classification function can be restated as:

[0142]

[0143] When the number of SVM input features is 2, the hyperplane appears as a straight line. When the number of input features increases to 3, the hyperplane expands to a two-dimensional plane. However, when the number of features exceeds 3, the dimension of the mapping will increase significantly, resulting in a dimensional explosion. In order to avoid this situation, the kernel function k is introduced. The purpose of introducing the kernel function is to compress the sample in a low-dimensional space, but the calculation result can be equivalent to the high-dimensional mapping calculation result, thereby effectively avoiding the dimensional explosion problem.

[0144] The specific expression of the kernel function is as follows:

[0145] k(v1,v2)=χ<v1,v2> =<λ(v1),λ(v2)<Formula 25

[0146] After kernel function dimensionality reduction processing, the classification function corresponding to formula 24 can be rewritten as:

[0147]

[0148] From the above analysis, we can see that support vector machines have the ability to fit any nonlinear function.

[0149] The following function fitting results can be obtained:

[0150]

[0151] The constraints in the above formula are:

[0152] z i =ω T λ(v i )+b+e i ,i=1,...,nFormula 28

[0153] In the formula, z i is the actual output corresponding to vi.

[0154] The optimization objective is transformed into:

[0155]

[0156] stz i =ω T λ(v i )+b+e i ,i=1,...,nFormula 29

[0157] According to Formula 27 and Formula 28, the Lagrange equation is constructed as follows:

[0158]

[0159] According to the KKT condition, the partial differential equation is solved for the above formula. The specific expression is as follows:

[0160]

[0161] Formula 31 can be solved to obtain:

[0162]

[0163] Where Ω is the kernel function diagonal matrix, and the diagonal elements are k(v i,v),i=1,2,...,n;z=[z1,...,z n ] T ;I n is the identity matrix.

[0164] Select a suitable kernel function and obtain the support vector machine regression equation:

[0165]

[0166] Formula 33 The support vector machine regression equation fitting process is heavily dependent on the kernel function of the support vector machine, so the selection of the kernel width of the kernel function is crucial. The appropriate kernel width can effectively improve the prediction accuracy of the model. Therefore, the ISA-GA algorithm is selected and designed to find the optimal value of the kernel width of the kernel function. The improved SA-GA algorithm flow chart is as follows: Figure 3 shown.

[0167] In a specific implementation manner, in step S103, the specific steps are as follows:

[0168] S1031. Initialize the population: randomly generate an initial population, where each individual represents a solution.

[0169] S1032. Genetic operations: Evolve the population and update the fitness of individuals through selection, adaptive crossover and mutation operations.

[0170] S1033, simulated annealing operation: perform a simulated annealing operation on each individual to locally search for a better solution.

[0171] S1034, determine the stop condition: according to the preset stop condition, determine whether the condition for stopping iteration is met.

[0172] S1035. Return the optimal solution: return the optimal individual as a result.

[0173] It should be noted that the traditional SA-GA uses the global search capability of the genetic algorithm and the local search capability of the simulated annealing algorithm to show good performance in solving complex optimization problems. The SA-GA algorithm uses two basic operations: genetic operation and simulated annealing operation. The genetic operation is the global search part of the genetic simulated annealing algorithm. However, the genetic algorithm has limitations in local search, which often causes the algorithm to fall into the local optimal solution prematurely. Traditional genetic algorithms usually use fixed crossover and mutation operators. Although these parameters are adjusted through multiple experiments, it is still difficult to ensure that they can reach the optimal state every time. Therefore, this affects the optimization efficiency and accuracy of the algorithm to a certain extent. In view of this, it is necessary to improve the crossover and mutation operators of the genetic algorithm to enhance its global search capability.

[0174] In a specific implementation, in step S1032, the specific expression for selecting the operation is:

[0175]

[0176] Set the adaptive coefficient to k c The adaptive crossover operator value expression is:

[0177]

[0178] Where p is a settable fractional value ranging from 0 to 1, and f avg is the average fitness value, f max is the maximum fitness value, It is used to measure the current population dispersion;

[0179] Set the adaptive coefficient to k m The mutation operator value expression is:

[0180]

[0181] In a specific implementation, in step S1033, before the simulated annealing operation, the genetic algorithm has obtained new individuals through a series of selection, crossover, mutation and other operations. The simulated annealing algorithm will compare the fitness value of the new individual obtained by the genetic algorithm with the fitness value of the old individual to decide whether to accept the inferior solution. The specific calculation formula is as follows:

[0182]

[0183] Where fit(x') is the fitness value of the new individual, fit(x) is the fitness value of the old individual, and T is the annealing temperature.

[0184] In a specific implementation, in step S2, the second-order integral link, the first-order integral link and the first-order integral link are used in the displacement part, the speed adjustment part and the flux adjustment part respectively; and the displacement subsystem, the speed subsystem and the flux subsystem in the x and y directions are constructed. By inputting the variables: And output variable: Z = [i 2sd ,i 2sq ,i 1rd ,i 1rq ] T Sampling is performed, and dynamic decoupling is achieved using the designed ISA-GA support vector machine inverse system decoupling control strategy;

[0185] The sampling process of input data and output data is as follows: the sampling time is set to 10s, and the sampling interval is 1e-2. Therefore, within the sampling time of 10s, 1000 sets of input and output variables can be obtained. 80% of the sample data is divided into the training sample set Vtrain and the corresponding label set Z train , and the remaining 20% ​​is used as the test set V test and the corresponding label set Z test , the training sample set is used to build a regression model, and the test set is used to evaluate the fit of the regression model to the data;

[0186] In order to eliminate the influence between parameters of different orders of magnitude, the input parameters and output parameters are normalized. The process is as follows:

[0187]

[0188] In the formula, Var(Z train (i))、 Var(Z train (i)) are the mean and variance of the data in the i-th column of the input and output matrices of the training sample set, respectively; n is the number of samples in the training set, and m is the number of samples in the test set.

[0189] like Figure 4 As shown in the figure, after the input vector is normalized, it is sent to the kernel function for calculation. After the calculation is completed, it is multiplied with the Lagrange multiplier, and the multiplication results are accumulated one by one. After the accumulation is completed, the bias constant is superimposed to obtain the corresponding output variable;

[0190] The output variable consists of four variables, so four SVM learning machines need to be built, and their initial regression model function is:

[0191]

[0192] In the formula

[0193] Finally, post-process the regression results output by the test set:

[0194]

[0195] The genetic generation, crossover rate and mutation rate in ISA-GA are set. Based on these initial parameters, the fitness of the system is calculated. The calculation formula of the fitness function is as follows:

[0196]

[0197] According to Formula 41, the corresponding fitness value of the system can be calculated. The selection and adaptive crossover mutation algorithm used in the ISA-GA algorithm can obtain the corresponding offspring set. The above process is iterated until the target requirements are met or the maximum number of iterations is reached;

[0198] The structural control block diagram of the support vector machine inverse system optimized by ISA-GA is shown in Figure 5 As shown. In order to meet the performance requirements of the CCR-BIM control system, a linear closed-loop controller is designed and connected in series with the ISA-GA optimized support vector machine inverse system to form a closed-loop control of speed and displacement. The linear closed-loop controller is mainly composed of a position controller, a speed controller and a flux controller.

[0199] The current regulation adopts the hysteresis tracking pulse width modulation method. In the feedback signal, the displacement signal is collected by the eddy current sensor. The speed signal is collected by the photoelectric encoder, and the magnetic flux signal is calculated using the observation equation shown below:

[0200]

[0201] The prototype parameters of CCR-BIM are shown in Table 1:

[0202] Table 1

[0203]

[0204] The block diagram of the support vector machine inverse system decoupling control optimized by ISA-GA is shown in Figure 6 As shown, the control system is divided into two sub-parts: suspension control and rotation control. The suspension part includes x-axis and y-axis displacement control, and the rotation control includes speed control and magnetic flux control.

[0205] In order to verify the effectiveness of the proposed decoupling control strategy, the displacement, speed and flux response curves under the traditional inverse system decoupling control and the proposed ISA-GA optimized support vector machine inverse decoupling control are compared and analyzed to compare the effects of the two decoupling controls:

[0206] like Figure 7 As shown, (a) is the x-axis radial displacement response diagram, and (b) is the y-axis radial displacement response diagram; it shows the x-axis and y-axis radial displacement responses under inverse system decoupling control and ISA-GA support vector machine inverse system decoupling control, given the x-axis and y-axis radial displacements of 0 μm, Figure 7As can be seen from (a), compared with the inverse system decoupling control, when the motor starts, the maximum offset of the proposed ISA-GA support vector machine inverse decoupling control on the x-axis is significantly reduced, and the starting displacement in the x-axis direction is reduced by 670μm. When the load interference signal is suddenly added at t=300ms, the proposed ISA-GA support vector machine inverse decoupling control takes 329ms to reach the set value state, while the inverse system takes 341ms, and the proposed ISA-GA support vector machine inverse decoupling control takes less time. When t=300ms, the set speed suddenly changes from 3000r / min to 6000r / min. Under the inverse system decoupling control, the x-axis offset is 202μm, and the recovery time is longer. The radial offset of the ISA-GA support vector machine inverse system decoupling control is 109μm, and the vibration is significantly smaller than the inverse system decoupling control. Figure 7 (b) is the y-axis displacement waveform. The change process is similar to the x-axis radial displacement. The ISA-GA support vector machine inverse system decoupling control reduces the starting displacement in the y-axis direction by 230μm. When the speed suddenly changes, the radial displacement of the ISA-GA support vector machine inverse system decoupling control is significantly smaller than that of the inverse system decoupling control, and it takes less time to recover to the set displacement value. This shows that the rotor can maintain stable suspension under the ISA-GA support vector machine inverse system decoupling control and has a faster response speed and better anti-disturbance performance;

[0207] When the inverse system decoupling control strategy and the ISA-GA support vector machine inverse decoupling control strategy are used, the speed response is as follows: Figure 8As shown in the figure: given the speed signals of 3000r / min and 6000r / min respectively, under the inverse system decoupling control, the speed reaches 3000r / min in 52ms; when the ISA-GA support vector machine inverse decoupling control is adopted, the speed reaches 3000r / min in 41ms. At t=300ms, the set speed rises from 3000r / min to 6000r / min. When the inverse system decoupling control is adopted, the speed reaches 6000r / min after 97ms, while when the ISA-GA support vector machine inverse system decoupling control is adopted, the speed reaches 6000r / min after 64ms. The proposed ISA-GA support vector machine inverse system decoupling control not only shortens the response time, but also improves the initial performance of CCR-BIM. In order to more comprehensively evaluate the performance of inverse system decoupling control and ISA-GA support vector machine inverse decoupling control, this paper observes and analyzes the speed response characteristics of the system under sudden load while observing the speed response characteristics of the motor during normal operation. When the load is suddenly added at t = 150ms and t = 450ms, the speed of CCR-BIM fluctuates significantly under the inverse system decoupling control, and recovers to the given speed after t = 180ms and t = 490ms, respectively. However, the speed under the proposed ISA-GA support vector machine inverse system method only fluctuates slightly when the load is suddenly added, and recovers to the given speed after t = 160ms and t = 460ms. The proposed ISA-GA support vector machine inverse system decoupling control greatly reduces the speed recovery time and has a faster decoupling control response.

[0208] like Fig. 9 As shown in the figure, the flux response in the two control strategies of inverse system decoupling control and ISA-GA support vector machine inverse decoupling control, the initial flux is 0.54Wb. CCR-BIM starts at t=0ms. During the startup process, the flux response output by the inverse decoupling control of the ISA-GA support vector machine is more sensitive. When the load is suddenly added at t=300ms, the flux drops from 0.54Wb to 0.48Wb. Under the inverse decoupling control strategy of the ISA-GA support vector machine, the fluctuation of the flux curve is small, and the recovery speed is faster than that of the inverse decoupling control strategy. During operation, the inverse decoupling control of the ISA-GA support vector machine outputs a smoother flux waveform.

[0209] Taking the A-phase suspension winding current as an example, Fig.10It can be seen that the torque winding induces current on the rotor bars, forming a magnetic field that interferes with the suspension winding current. The sine characteristics of the suspension winding current under the two decoupling control strategies are weaker than the torque winding current. The suspension winding current value in the stable state is 0.5A. When the motor is suddenly loaded at 300ms, the inverse decoupling control current waveform is significantly larger than the current under the inverse decoupling control of the ISA-GA support vector machine, and it takes longer to return to normal current. After the sudden load, the stable value of the suspension winding current value has hardly changed, indicating that the increase in load has almost no effect on the suspension winding current.

[0210] Taking the torque winding current of phase A as an example, when the load is suddenly added, the torque winding current waveform is as follows Fig.11 As shown in the figure, before and after the sudden load is added to the motor, the current waveforms of the two decoupling control methods are both sinusoidal waveforms, but the inverse decoupling control torque winding current waveform of the ISA-GA support vector machine is smoother than the inverse system decoupling control waveform, and the current finally stabilizes at 3A. When the load is suddenly added for 300ms, the torque current changes suddenly, and the current waveform of the inverse decoupling control of the ISA-GA support vector machine responds more sensitively, reaches stability in a shorter time, and the current value finally stabilizes at 3.5A. The current under the inverse decoupling control of the ISA-GA support vector machine has a faster dynamic response performance, and the sinusoidality of the waveform is better than that of the current waveform of the inverse system decoupling control, which also proves that the inverse decoupling control of the ISA-GA support vector machine is more sensitive to the decoupling control of the decoupling parameters.

[0211] In summary, from Figure 7 , Figure 8 and Fig. 9 It can be seen that the proposed ISA-GA support vector machine inverse decoupling control has good x-axis and y-axis radial displacement, velocity response and flux decoupling effects, and has a fast response speed. Fig.10 and Fig.11 The ISA-GA support vector machine inverse decoupling control makes the torque winding current and suspension force winding current waveforms smoother and the response time shorter, which proves the decoupling control effect of the above-mentioned electrical quantities and also proves the superiority of the proposed ISA-GA support vector machine inverse decoupling control.

[0212] In order to further verify the effectiveness of the control strategy proposed in this paper, an experimental verification was carried out on a self-designed CCR-BIM experimental prototype. The experimental prototype platform is mainly composed of the CCR-BIM prototype and its peripheral equipment, including control and drive components such as the DSP main control board, power drive circuit board and interface circuit board. In order to realize the closed-loop feedback control of speed and displacement, the platform also integrates monitoring equipment such as photoelectric encoders and eddy current sensors. In addition, in order to measure and collect experimental data, it is also equipped with measuring equipment such as a host computer, an oscilloscope, a torque meter and a dynamometer. The DSP main control board mainly controls the torque and suspension parts. The power drive board drives the two sets of windings, and the interface circuit board realizes the boost and signal amplification between the DSP main control board and the power drive board. The photoelectric encoder and eddy current sensor respectively collect and modulate the CCR-BIM speed and shaft displacement. The oscilloscope, torque meter and dynamometer are mainly used to collect or monitor the corresponding parameters and transmit them to the host computer (PC). The position value is given as 0μm in the experimental system. Fig.12 As shown in the figure, under the inverse system decoupling control, the maximum radial offset of the rotor center in the x-direction is 42 μm. However, under the ISA-GA inverse system decoupling control strategy, the maximum radial offset of the rotor center in the x-direction is only 31 μm.

[0213] like Fig.13 As shown in the figure, under the inverse system decoupling strategy, the maximum radial offset of the rotor center in the y direction is 43μm. However, under the ISA-GA inverse system decoupling control, the maximum radial offset of the rotor center in the y direction is only 30μm. At the same time, in order to verify the anti-disturbance performance, a disturbance experiment was carried out in the y-axis direction. Under the same disturbance load, the disturbance offset under the ISA-GA inverse system decoupling control is 60μm, and the recovery time is shorter. The disturbance offset under the inverse system decoupling control is 160μm, and it takes longer to recover to the set value. It shows that the ISA-GA support vector machine inverse decoupling control has a better decoupling effect and can handle external disturbances at a faster speed. From the comparison, it can be seen that under the proposed ISA-GA support vector machine inverse decoupling control strategy, CCR-BIM can achieve displacement decoupling control and have better anti-disturbance performance.

[0214] like Fig.14As shown in the figure, when t = 0ms, under the inverse decoupling control of the ISA-GA support vector machine, the starting speed reaches the stable value faster. When t = 150ms, the speed increases from 3000r / min to 6000r / min. Under the inverse decoupling control of the ISA-GA support vector machine, the stable speed is reached faster, and the speed fluctuation is smaller and the speed is more stable. Under the inverse system, the speed response curve has obvious jitter. During the stable operation with the speed set at 3000r / min, at t = 90ms, the CCR-BIM suddenly adds a load of 12N·m. Under the inverse decoupling control of the ISA-GA support vector machine, the speed mutation value is 4r / min, and the set speed can be restored in a shorter time. Under the inverse decoupling control, the speed mutation value is 8r / min. The speed is set to 6000r / min and the CCR-BIM is running stably. At t=240ms, a load of 12N·m is suddenly added to the CCR-BIM. Under the inverse decoupling control of the ISA-GA support vector machine, the speed mutation value is 7r / min, and the set speed can be restored in a shorter time. Under the inverse decoupling control, the speed mutation value is 19r / min. This shows that the inverse decoupling control of the ISA-GA support vector machine is faster and more resistant to disturbances.

[0215] Fig.15 The torque winding current waveform under the inverse decoupling control of the inverse system and ISA-GA support vector machine when there is a sudden load, taking phase A as an example. As can be seen from the figure, the sine performance under the inverse decoupling control of the ISA-GA support vector machine is better, the waveform curve is smooth and fluent, there are no abrupt peaks, and the current waveform fluctuates less. When the load current is suddenly added at 300ms, the current waveform of the inverse decoupling control of the ISA-GA support vector machine changes faster and the waveform distortion is small. There are multiple sharp peaks and thorn-like protrusions in the current waveform of the inverse system decoupling control, which makes the waveform appear rough and not smooth, which is easy to cause torque fluctuations.

[0216] Fig.16 The current waveform of the suspension winding under inverse decoupling control and ISA-GA support vector machine inverse decoupling control. Fig.16 It can be seen that the sine property of the suspension force winding current under the two decoupling control strategies is better and the waveform is smoother. The suspension force winding current value in the stable state is 0.5A. When the motor is suddenly loaded at 300ms, the current fluctuation under the inverse decoupling control of the ISA-GA support vector machine is smaller, and the time to return to normal current is shorter, which proves that the inverse decoupling control of the ISA-GA support vector machine has a faster decoupling calculation speed for the radial displacement of the x-axis and y-axis, better effect, and better suspension performance.

[0217] According to the above experimental results, it can be proved that the ISA-GA support vector machine inverse decoupling control provided by the present invention can realize the decoupling control of speed, x-axis and y-axis radial displacement and flux linkage, with fast decoupling speed and strong anti-disturbance. The superiority of the proposed ISA-GA support vector machine inverse decoupling control is verified.

[0218] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor, characterized in that: The following steps are involved: S1, ISA-GA optimized SVM inverse system establishment; S101, SVM model establishment: Based on the reversibility of the CCR-BIM system, a SVM model is constructed to identify and fit the inverse system of the CCR-BIM system; S102. Optimization design of SVM kernel function based on ISA-GA; S2. Decoupling control system of SVM inverse system based on ISA-GA optimization: obtain the output signal of the CCR-BIM system, input the output signal of the CCR-BIM system into the SVM inverse system optimized by ISA-GA, obtain the control signal processed by the SVM inverse system, and connect the control signal in series with the input signal of the original system to form a pseudo-linear system.

2. The method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor according to claim 1, characterized in that: In step S101, it specifically includes using a convex quadratic programming problem to construct an optimal classification hyperplane and construct a Lagrange function with n training samples: Among them, ω is the normal vector, b is the bias term, α i is the Lagrange multiplier, v i is the feature vector of the i-th training sample, z i is the label of the i-th training sample, ||ω|| 2 is the square of the norm of the normal vector ω; Limit the parameter to α i ≥0, the objective function expression is: There is a saddle point in the Lagrange function, and it satisfies the stability condition, feasibility condition and complementary relaxation condition in the KKT condition. The objective function expression is equivalent to solving a set of dual problems. The specific expression is as follows: Fix the parameter α and take partial derivatives with respect to ω and b to find the minimum value of the Lagrange function with respect to ω and b. The mathematical expression is: After calculating the Lagrange function, we can get: Then the objective function will be transformed into: s.t.α i ≥0,i=1,2,...,n The hand-pushed Sequential Minimal Optimization (SMO) method is used to recursively calculate the Lagrange multiplier α, and then ω and b are calculated to determine the maximum margin hyperplane.

3. The method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor according to claim 1, characterized in that: In step S102, the following steps are specifically included: S1021. Initialize the population: randomly generate the initial population, each individual represents a solution; S1022. Genetic operations: Evolve the population and update the fitness of individuals through selection, adaptive crossover and mutation operations; S1023, simulated annealing operation: performing a simulated annealing operation on each individual to locally search for a better solution; S1024, determining a stop condition: determining whether a condition for stopping iteration is met according to a preset stop condition; S1025. Return the optimal solution: return the optimal individual as the result.

4. The method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor according to claim 3, characterized in that: In step S1022, the specific expression of the selection operation is: Set the adaptive coefficient to k c The adaptive crossover operator value expression is: Where p is a settable fractional value ranging from 0 to 1, and f avg is the average fitness value, f max is the maximum fitness value, It is used to measure the current population dispersion; Set the adaptive coefficient to k m The mutation operator value expression is:

5. The method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor according to claim 4, characterized in that: In step S1023, the simulated annealing algorithm compares the new individual fitness value obtained by the genetic algorithm with the old individual fitness value to decide whether to accept the inferior solution. The specific calculation formula is as follows: Where fit(x') is the fitness value of the new individual, fit(x) is the fitness value of the old individual, and T is the annealing temperature.

6. The method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor according to claim 5, characterized in that: In step S2, the second-order integral link, the first-order integral link and the first-order integral link are used in the displacement part, the speed regulation part and the flux regulation part respectively; and the displacement subsystem, the speed subsystem and the flux subsystem in the x and y directions are constructed, and the input variables are: And output variable: Z = [i 2sd ,i 2sq ,i 1rd ,i 1rq ] T Sampling is performed, and dynamic decoupling is achieved using the designed ISA-GA support vector machine inverse system decoupling control strategy.

7. The method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor according to claim 6, characterized in that: The sampling process of input data and output data is as follows: the sampling time is set to 10s, the sampling interval is 1e-2, and 80% of the sample data is divided into the training sample set V train and the corresponding label set Z train , and the remaining 20% ​​is used as the test set V test and the corresponding label set Z test The training sample set is used to build a regression model, and the test set is used to evaluate the fit of the regression model to the data.

8. The method for inverse system decoupling control of a bearingless composite rotor cage asynchronous motor according to claim 7, characterized in that: The input and output parameters are normalized as follows: In the formula, Var(Z train (i))、 Var(Z train (i)) are the mean and variance of the i-th column of the input and output matrices of the training sample set, respectively; n is the number of samples in the training set, and m is the number of samples in the test set; After the input vector is normalized, it is sent to the kernel function for calculation. After the calculation is completed, it is multiplied with the Lagrange multiplier, and the multiplication results are accumulated one by one. After the accumulation is completed, the bias constant is superimposed to obtain the corresponding output variable; Then, four SVM learning machines are constructed, and their initial regression model functions are: In the formula Finally, post-process the regression results output by the test set: The genetic generation, crossover rate and mutation rate in ISA-GA are set, and the fitness of the system is calculated based on these initial parameters. The calculation formula of the fitness function is as follows: According to the calculation formula of the fitness function, the fitness value corresponding to the system is calculated. The selection and adaptive crossover mutation algorithm used in the ISA-GA algorithm can obtain the corresponding offspring set and iterate the above process.

Images