Method and system for realizing model-free predictive current control of synchronous reluctance motor based on neural network interference observer

Abstract

The invention discloses a model-free predictive current control method for a synchronous reluctance motor based on a neural network interference observer, and aims to solve the problems of slow speed response and poor control effect of the synchronous reluctance motor under complex working conditions so as to realize accurate control of the synchronous reluctance motor. Meanwhile, according to the control method, a neural network disturbance observer is provided for observing disturbance caused by factors such as motor parameter mismatch, and the disturbance serves as lumped disturbance to be compensated into a prediction model so as to improve the parameter robustness of the control system. The biggest characteristic of the neural network interference observer is that the thought of pure integration, compensation and transfer function types is adopted, and a super-local model is introduced to construct a motor mathematical model, so that the robustness of the system is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor control, and relates to a method and system for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer. Background Art

[0002] Synchronous reluctance motor (Synchronous Reluctance Motor, referred to as SynRM) has a special rotor structure, no permanent magnets inside (excluding permanent magnet assisted synchronous reluctance motors), high performance, not only can be used in a variety of harsh environments, but also in torque ripple, harmonic content, power density and other aspects of the superiority of the switched reluctance motor, and has attracted more and more attention from researchers. Therefore, in the face of complex working conditions or parameter changes due to high temperature and external disturbances, the control performance of SynRM is put forward higher requirements.

[0003] In recent years, the synchronous reluctance motor control method based on model predictive control (MPC) is regarded as an effective alternative to the traditional control method of SynRM due to its intuitive concept and easy implementation. MPC achieves fast and stable control based on the mathematical model of the controlled object and the cost function composed of control constraints.

[0004] However, the existing MPC-based synchronous reluctance motor control method has some non-negligible defects:

[0005] First, its predictive control relies on the motor body model to construct prediction models for various control targets, which is easily affected by the body parameters. At the same time, the magnetic saturation phenomenon of the synchronous reluctance motor causes nonlinear changes in the motor parameters. Both of them will cause motor parameter mismatch problems, reducing the robustness of the control algorithm.

[0006] Second, MPC applies a single voltage vector within the control cycle and continuously rolls the optimization to make the motor have good dynamic performance, but the steady-state performance of the motor is poor. Summary of the invention

[0007] In response to the above defects or improvement needs of the prior art, the present invention provides a method and system for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer. The purpose is to solve the technical problems that the existing MPC-based synchronous reluctance motor control method is easily affected by the body parameters because the predictive control relies on the motor body model to construct the prediction model of various control targets, and the magnetic saturation phenomenon of the synchronous reluctance motor causes nonlinear changes in the motor parameters, resulting in motor parameter mismatch and thus reducing the robustness of the control algorithm; and the technical problem that the steady-state performance of the motor is poor.

[0008] To achieve the above object, according to one aspect of the present invention, a method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer is provided, comprising the following steps:

[0009] (1) Obtain the ABC three-phase voltage signal u of the synchronous reluctance motor in the kth sampling period through voltage sensor sampling a (k),u b (k),u c (k), the position signal θ(k) of the kth sampling period of the synchronous reluctance motor is obtained by the position sensor, and the three-phase voltage signal u is calculated according to the position signal θ(k). a (k),u b (k) and u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k),u q (k), the d-axis and q-axis voltage signals u of the k-th sampling period are d (k),u q (k) is input into the hyperlocal model for weighted processing, and the processing results are respectively compared with the d-axis and q-axis interference observation values ​​output by the neural network interference observer in the kth sampling period. Add to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and The first-order forward Euler discretization method is used to discretize the d-axis and q-axis current differential signals respectively, and the current obtained by current sensor sampling is combined to obtain the d-axis and q-axis discrete current prediction values ​​of the k-th sampling period output by the super-local model. Wherein, k is a positive integer.

[0010] (2) The predicted values ​​of the d-axis and q-axis discrete currents i of the k+1th sampling period of the hyperlocal model obtained in step (1) d (k+1), i q (k+1) Get the first optimal voltage vector u of the kth sampling period g1d (k),u g1q (k), according to the optimal voltage vector u g1d (k),u g1q (k) Calculate the voltage difference ΔE u1d (k), ΔE u1q (k) Obtain the second and third optimal voltage vectors u of the d and q axes according to the voltage difference table g2d (k) and u g2q (k) and u g2q (k) and u g3q(k), use the duty cycle direct calculation method to obtain the d and q axis duty cycles d of the kth sampling period after the hyperlocal model optimization nd (k), d nq (k), and according to the duty cycle d nd (k) and d nq (k) Control the pulse generator to output a drive signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the drive signal, thereby driving the synchronous reluctance motor to operate, where n∈[1,3].

[0011] (3) The d-axis and q-axis current signals i of the synchronous reluctance motor in the k+1th sampling cycle are obtained by sampling the current sensor. d (k+1), i q (k+1), according to the d and q axis current signals i d (k+1), i q (k+1) respectively obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period d (k+1) and e q (k+1), obtain the d and q axis error output matrix of the k+1th sampling period according to the obtained d and q axis current estimation errors and the estimation of the d and q axis weights for the k+1th sampling period The error output matrix is ​​obtained Estimation with weights Multiply them together to get an estimate of the disturbed observation The synchronous reluctance motor is controlled based on the estimation of the disturbance observation value.

[0012] Preferably, step (1) comprises the following sub-steps:

[0013] (1-1) The ABC three-phase voltage signal u of the kth sampling period of the synchronous reluctance motor is obtained by sampling with a voltage sensor a (k),u b (k),u c (k), the position signal θ(k) of the synchronous reluctance motor in the kth sampling period is obtained through the position sensor, and the three-phase voltage signal u of the kth sampling period is calculated according to the position signal θ(k) of the kth sampling period. a (k) and u b (k),u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k) and u q (k).

[0014] (1-2) The d-axis and q-axis voltage signals u of the k-th sampling period obtained in step (1-1) are d(k) and u q (k) Use the corresponding scaling factor α d , α q They are weighted and then compared with the interference observation value of the kth sampling period output by the neural network interference observer. Add together to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and

[0015] (1-3) Based on the d-axis and q-axis currents of the k-th sampling period and the d-axis and q-axis voltage signals u of the k-th sampling period obtained in step (1-1) d (k) and u q (k), and use the first-order forward Euler discretization method to respectively obtain the d-axis and q-axis current differential signals of the k-th sampling period obtained in (1-2) Discretization is performed to obtain the predicted values ​​of the d-axis and q-axis discrete currents of the k+1th sampling period output by the hyperlocal model.

[0016] Preferably, step (1-2) is performed using the following formula:

[0017]

[0018] where α d =1 / L d and α q =1 / L q represents the weighting factor, α d =1 / L d and α q =1 / L q These are all values ​​obtained through experience debugging;

[0019] Steps (1-3) are performed using the following formula:

[0020]

[0021] Where T s is the system sampling time.

[0022] Preferably, the d-axis and q-axis interference observation values ​​of the k-th sampling period are It is obtained through the following sub-steps:

[0023] (a) Obtain the current a0, a1, ..., a of the traditional model-predicted current control model at different load starting times x , and according to the current a0, a1, ...., a xObtain the actual load disturbance current error set F = [d1, d2, ...., d x ], where x represents the total number of loads in the model-predicted current control model.

[0024] (b) Randomly select multiple load disturbance current errors from the load disturbance current error set F obtained in step (a) to form a data set n times. For each data set p (where p∈[1,3]), obtain all load disturbance current errors in the pth data set The mean of the p-th data set is used as the center of the Gaussian function corresponding to the p-th data set. Where n is a positive integer less than or equal to x, m p represents the total number of load disturbance current errors in the pth data set;

[0025] (c) For each data set p obtained in step (b), obtain the center of the Gaussian function corresponding to each load disturbance current error in the data set p and the pth data set The Euclidean distance between them is calculated, and all the Euclidean distances are summed up and averaged to obtain the Gaussian function center b corresponding to the pth data set. p , and obtain the load disturbance current error set consisting of all load disturbance current errors in the pth data set obtained in step (1-2) Each load disturbance current error in the pth data set corresponds to the center of the Gaussian function The difference between them is calculated, and all the differences are summed up and averaged to obtain the central width b of the neural network. n .

[0026] (d) For each data set p obtained in step (b), the m load disturbance current errors in the pth data set are Input Gaussian function to calculate and get multiple error outputs Construct the initial error output matrix H0;

[0027] (e) Obtain ideal weights based on the initial error output matrix H0 obtained in step (d) and the actual load disturbance current error set F obtained in step (a)

[0028] (f) Based on the output matrix H0 obtained in step (d) and the ideal weights obtained in step (e) Get an estimate of the noise observation for the first sampling period

[0029] (g) Use the estimate of the interference observation obtained in step (f) As the initial value, the neural network disturbance observer is input for multiple iterations to obtain the disturbance observation value of the kth sampling period.

[0030] Preferably, step (a) uses the following formula:

[0031]

[0032] Where a0 is the current predicted by the model when the current control model is unloaded, d x for a x The corresponding load disturbance current error.

[0033] Step (e) is to use the following formula:

[0034]

[0035] in, is the transpose of the initial error output matrix H0.

[0036] Step (f) uses the following formula:

[0037]

[0038] Where ε is the estimation error of the neural network disturbance observer and satisfies ε≤ε n , ε n is the estimation error of the neural network disturbance observer after n iterations.

[0039] Preferably, step (2) comprises the following sub-steps:

[0040] (2-1) The predicted values ​​of the d-axis and q-axis discrete currents of the k+1th sampling period output by the hyperlocal model obtained in step (1) Get the ideal voltage vector of the d and q axes of the kth sampling period

[0041] (2-2) The ideal voltage vector of the d and q axes of the kth sampling period obtained according to step (2-1) And the position signal θ(k) of the kth sampling period, obtain the α and β axis voltage cost functions.

[0042] (2-3) Substitute the effective voltage vector u(k) of the kth sampling period into the α and β axis voltage cost functions obtained in step (2-2) to obtain the values ​​that make g α and g β The minimum voltage vector is taken as the first optimal voltage vector u of the d and q axes in the kth sampling period. g1d (k) and u g1q(k), the ideal voltage vector of the kth sampling period is respectively compared with the first optimal voltage vector u of the d and q axes g1d (k) and u g1q (k) to obtain the d-axis and q-axis voltage difference ΔE of the k-th sampling period respectively. u1d (k), ΔE u1q (k).

[0043] (2-4) Divide the voltage vector distribution diagram of the pulse generator into 6 sectors, and calculate the d-axis and q-axis voltage vector errors ΔE obtained in step (2-3) u1d (k), ΔE u1q (k) The corresponding sectors in the voltage vector distribution diagram, and query the corresponding d-axis and q-axis second voltage vectors u of the k-th sampling period in the pre-established second optimal voltage vector table according to the acquired sectors. g2d (k) and u g2q (k), and query the corresponding d-axis and q-axis third voltage vectors u of the k-th sampling period in the pre-established third optimal voltage vector table according to the acquired sectors. g3d (k) and u g3q (k).

[0044] (2-5) The nth voltage vector u of the d and q axes of the kth sampling period obtained in step (2-4) gnd (k) and u gnd (k), respectively obtain the current prediction value i of the d and q axes corresponding to the k-th sampling period of the n-th voltage vector of the d and q axes dn (k+1) and i qn (k+1).

[0045] (2-6) The d-axis and q-axis reference currents of the k-th sampling period obtained in step (2-1) are and The current prediction value i of the d and q axes in the kth sampling period corresponding to the nth voltage vector of the d and q axes obtained in step (2-5) respectively dn (k+1) and i qn (k+1) to obtain the nth voltage vector u gnd (k) and u gnd (k) The current error Δε of the d and q axes of the k-th sampling period respectively dn (k) and Δε qn (k).

[0046] (2-7) The nth voltage vector u obtained according to step (2-6) gnd (k) and u gnd (k) The d-axis and q-axis current errors Δε of the k-th sampling period respectively dn(k), Δε qn (k) Obtain the nth voltage vector u gnd (k) and u gnd (k) The corresponding duty cycle d nd (k) and d nq (k).

[0047] (2-8) The duty cycle d obtained according to step (2-7) nd (k) and d nq (k) The control pulse generator outputs a driving signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the driving signal, thereby driving the synchronous reluctance motor to operate.

[0048] Preferably, step (2-1) adopts the following formula:

[0049]

[0050] in, is the d-axis and q-axis reference current of the k-th sampling period, and e d (k), e q (k) is the estimated error of the d-axis and q-axis currents in the k-th sampling period, and its initial value is 0; m is the error coefficient, and its value is 19000.

[0051] Step (2-2) uses the following formula:

[0052]

[0053] where u α (k) and u β (k) is the effective voltage vector of the kth sampling period of the α and β axes.

[0054] The calculation of the voltage difference between the α and β axes in the kth sampling period in step (2-3) is performed using the following formula:

[0055]

[0056] Steps (2-5) are calculated using the following formula:

[0057]

[0058] Steps (2-6) are calculated using the following formula:

[0059]

[0060] Steps (2-7) are calculated using the following formula:

[0061]

[0062] where d 1d (k) represents d-axis u g1d (k) The corresponding duty cycle, d 2d (k) represents u g2d (k) The corresponding duty cycle, d 3d (k) represents u g3d (k) The corresponding duty cycle, d 1q (k) represents the q-axis u g1q (k) The corresponding duty cycle, d 2q (k) represents u g2q (k) The corresponding duty cycle, d 3q (k) represents u g3q (k) The corresponding duty cycle.

[0063] Preferably, step (3) comprises the following sub-steps:

[0064] (3-1) Obtain the ABC three-phase current signal i of the k+1th sampling period after the synchronous reluctance motor starts running through the current sensor a (k+1), i b (k+1), i c (k+1), and according to the position signal θ(k+1) of the synchronous reluctance motor in the k+1th sampling period obtained by the position sensor, the d-axis and q-axis current signals i of the k+1th sampling period are obtained. d (k+1) and i q (k+1).

[0065] (3-2) The ideal voltage vector of the d and q axes of the k-th sampling period obtained according to step (2-1) and Get the d-axis and q-axis current estimates of the k-th sampling period respectively and

[0066] (3-3) The d-axis and q-axis currents i obtained in the k+1th sampling period in step (3-1) are d (k+1) and i q (k+1) and the d-axis and q-axis current estimates of the k-th sampling period obtained in step (3-2) and To obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period, d (k+1) and e q (k+1).

[0067] (3-4) Determine the d-axis and q-axis current estimation errors e of the k+1th sampling period obtained in step (3-3) d (k+1), eq (k+1) are all 0. If so, it means that the synchronous reluctance motor runs smoothly and the process ends. Otherwise, it goes to step (3-5);

[0068] (3-5) The d-axis and q-axis current estimation errors e of the k+1th sampling period obtained in step (3-3) are respectively d (k+1) and e q (k+1), and use the Gaussian function to obtain the error output h corresponding to the d and q axes of the k+1th sampling period respectively d (k+1) and h q (k+1), and use the error output h corresponding to the d and q axes d (k+1) and h q (k+1) Get the error output matrix of the k+1th sampling period

[0069] (3-6) Obtain an estimate of the weight of the k+1th sampling period based on the error output matrix H(k+1) of the k+1th sampling period obtained in step (3-5)

[0070] (3-7) Substitute the error output matrix H(k+1) of the k+1th sampling period obtained in step (3-5) with the weight estimate of the k+1th sampling period obtained in step (3-6) Multiply them together to obtain an estimate of the interfering observation value for the k+1th sampling period Then return to step (3-1).

[0071] Preferably, step (3-1) adopts the following formula:

[0072]

[0073] The d-axis and q-axis current estimates in step (3-2) are obtained using the following formula:

[0074]

[0075] Step (3-3) is to calculate the current estimation error using the following formula:

[0076]

[0077] Steps (3-5) use the following formula:

[0078]

[0079] in, are the centers of Gaussian functions of d and q axes respectively; b d 、b qThey are the center widths of the neural network on the d and q axes respectively.

[0080] Steps (3-6) are done using the following formula:

[0081]

[0082] in is the derivative of the estimate of the weight at the k+1th sampling period; is the derivative of the error of the estimated weight of the k+1th sampling period and satisfies W * (k) is the ideal value of the weight for the kth sampling period, is the estimate of the weight of the k+1th sampling period; e T (k+1) is the transpose of e(k+1), γ is a positive constant.

[0083] According to another aspect of the present invention, a system for implementing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer is provided, comprising:

[0084] The first module is used to obtain the ABC three-phase voltage signal u of the kth sampling period of the synchronous reluctance motor through voltage sensor sampling. a (k) and u b (k) and u c (k), the position signal θ(k) of the kth sampling period of the synchronous reluctance motor is obtained by the position sensor, and the three-phase voltage signal u is calculated according to the position signal θ(k). a (k) and u b (k) and u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k) and u q (k), the d-axis and q-axis voltage signals u of the k-th sampling period are d (k) and u q (k) is input into the hyperlocal model for weighted processing, and the processing results are respectively compared with the d-axis and q-axis interference observation values ​​output by the neural network interference observer in the kth sampling period. Add to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and The first-order forward Euler discretization method is used to discretize the d-axis and q-axis current differential signals respectively, and the current obtained by current sensor sampling is combined to obtain the d-axis and q-axis discrete current prediction values ​​of the k-th sampling period output by the super-local model. Wherein, k is a positive integer.

[0085] The second module is used to obtain the d-axis and q-axis discrete current prediction values ​​i of the k+1th sampling period of the super-local model obtained in the first module. d (k+1), i q (k+1) Get the first optimal voltage vector u of the kth sampling period g1d (k) and u g1q (k), according to the optimal voltage vector u g1d (k) and u g1q (k) Calculate the voltage difference ΔE u1d (k), ΔE u1q (k) Obtain the second and third optimal voltage vectors u of the d and q axes according to the voltage difference table g2d (k) and u g2q (k) and u g2q (k) and u g3q (k), use the duty cycle direct calculation method to obtain the d and q axis duty cycles d of the kth sampling period after the hyperlocal model optimization nd (k), d nq (k), and according to the duty cycle d nd (k) and d nq (k) Control the pulse generator to output a drive signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the drive signal, thereby driving the synchronous reluctance motor to operate, where n∈[1,3].

[0086] The third module is used to obtain the d-axis and q-axis current signals i of the synchronous reluctance motor in the k+1th sampling cycle through current sensor sampling. d (k+1), i q (k+1), according to the d and q axis current signals i d (k+1), i q (k+1) respectively obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period d (k+1) and e q (k+1), obtain the d and q axis error output matrix of the k+1th sampling period according to the obtained d and q axis current estimation errors and the estimation of the d and q axis weights for the k+1th sampling period The error output matrix is ​​obtained Estimation with weights Multiply them together to get an estimate of the disturbed observation The synchronous reluctance motor is controlled based on the estimation of the disturbance observation value.

[0087] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0088] (1) The present invention adopts step (1) combined with step (3), and uses a model-free control method to estimate and update the system disturbance in real time through the current estimation error. This method does not need to identify the specific motor parameters of the controlled object, and thus can solve the problem of reduced robustness caused by motor parameter mismatch in the existing MPC-based synchronous reluctance motor control method;

[0089] (2) Since the present invention adopts step (2), the inverter output voltage under the three-vector MPC strategy is closer to the ideal voltage vector, thereby solving the problem of poor motor steady-state performance existing in the existing MPC-based synchronous reluctance motor control method;

[0090] (3) The present invention adopts sub-steps (2-3) to (2-5), applies three voltage vectors under the three-vector MPC strategy to improve steady-state performance, optimizes the selection process of the three-vector combination by designing the voltage cost function and re-dividing the vector sectors, and reduces the complexity of algorithm implementation;

[0091] (4) Since the present invention adopts sub-steps (1-3), if the sampling time is very short, the hyperlocal model can be approximately regarded as a system model updated in real time, thereby ensuring the continuity of system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 It is a control block diagram of a method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer of the present invention;

[0093] Figure 2 A voltage vector distribution diagram obtained in steps (2-4) of the method of the present invention is shown;

[0094] Figure 3 It is a flow chart of a method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to the present invention.

[0095] Figure 4 It is the offline training flow chart of neural network disturbance observer. DETAILED DESCRIPTION

[0096] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0097] The present invention discloses a model-free predictive current control method for a synchronous reluctance motor based on a neural network disturbance observer, the purpose of which is to solve the problems of slow speed response and poor control effect of the synchronous reluctance motor under complex working conditions, so as to achieve accurate control of the synchronous reluctance motor. At the same time, the control method provides a neural network disturbance observer to observe disturbances caused by factors such as motor parameter mismatch, and compensate these disturbances as lumped disturbances into the prediction model to improve the parameter robustness of the control system. The biggest feature of the neural network disturbance observer is that it adopts the ideas of pure integration, compensation and transfer function types, and introduces a super-local model to construct a motor mathematical model to improve the system robustness.

[0098] The present invention provides a method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer, which is applied in Figure 1 In the control system of the synchronous reluctance motor shown in the figure, the control system includes a signal detection circuit, a main circuit and a control circuit; the main circuit includes a three-phase inverter and a synchronous reluctance motor, and the three-phase inverter is mainly used to drive the synchronous reluctance motor; the signal detection circuit includes a current detection circuit and a rotary encoder, which are mainly used to detect the current and rotor position signals of the synchronous reluctance motor; the control circuit includes a coordinate transformation module, a neural network interference observer, a super-local model prediction module, an optimal vector selection module, a duty cycle calculation module and a pulse generator module, which are mainly used to process the signal obtained by the signal detection circuit to obtain a control signal for controlling the main circuit.

[0099] Among them, the rotary encoder detects the rotor position angle θ of the synchronous reluctance motor, and the mechanical angular velocity ω of the motor is obtained by derivation m ; Motor mechanical angular velocity given value And the motor mechanical angular velocity feedback value ω m After the difference is made, the q-axis current given value is obtained through the PI controller The control circuit uses the three-phase current i obtained by the current detection circuit after detecting the synchronous reluctance motor a 、i b 、i c The rotor position angle θ is processed by the coordinate transformation module to obtain i d 、i q ;i d 、i q After being processed by the neural network disturbance observer, the lumped disturbance part F in the hyperlocal model is obtained. d 、F q The observed value of current i d 、i q and the total disturbance F d 、F qAfter being processed by the super-local model prediction module, the predicted value of the current is obtained; the predicted value of the dq axis current and the reference value of the dq axis current are substituted into the optimal vector selection module to obtain the optimal voltage vector u g1 、u g2 ; The duty ratios d1 and d2 of the corresponding voltage vectors are calculated through the duty ratio; finally, the switching sequence of the three-phase inverter is obtained through the pulse generator module to control the synchronous reluctance motor.

[0100] like Figure 3 As shown, the present invention provides a method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer, comprising the following steps:

[0101] (1) Obtain the ABC three-phase voltage signal u of the synchronous reluctance motor in the kth sampling period through voltage sensor sampling a (k) and u b (k) and u c (k), the position signal θ(k) of the kth sampling period of the synchronous reluctance motor is obtained by the position sensor, and the three-phase voltage signal u is calculated according to the position signal θ(k). a (k) and u b (k) and u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k) and u q (k), the d-axis and q-axis voltage signals u of the k-th sampling period are d (k) and u q (k) is input into the hyperlocal model for weighted processing, and the processing results are respectively compared with the d-axis and q-axis interference observation values ​​output by the neural network interference observer in the kth sampling period. Add to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and The first-order forward Euler discretization method is used to discretize the d-axis and q-axis current differential signals respectively, and the current obtained by current sensor sampling is combined to obtain the d-axis and q-axis discrete current prediction values ​​of the k-th sampling period output by the super-local model. Wherein, k is a positive integer.

[0102] Specifically, the hyperlocal model in this step is described in detail in the article "Model-free control-based vector control of synchronous reluctance motor" by Belkacem Selma, Elhadj Bounadja et al., published in the journal "Dynamics and Control" in November 2023.

[0103] This step includes the following sub-steps:

[0104] (1-1) The ABC three-phase voltage signal u of the kth sampling period of the synchronous reluctance motor is obtained by sampling with a voltage sensor a (k) and u b (k) and u c (k), the position signal θ(k) of the synchronous reluctance motor in the kth sampling period is obtained through the position sensor, and the three-phase voltage signal u of the kth sampling period is calculated according to the position signal θ(k) of the kth sampling period. a (k) and u b (k) and u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k) and u q (k).

[0105] (1-2) The d-axis and q-axis voltage signals u of the k-th sampling period obtained in step (1-1) are d (k) and u q (k) Use the corresponding scaling factor α d , α q They are weighted and then compared with the interference observation value of the kth sampling period output by the neural network interference observer. Add together to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and

[0106] Specifically, this step uses the following formula:

[0107]

[0108] In the formula, α d =1 / L d and α q =1 / L q represents the weighting factor, α d =1 / L d and α q =1 / L q These are all values ​​obtained through experience debugging;

[0109] like Figure 4 Specifically, the d and q axis interference observation values ​​of the kth sampling period are It is obtained through the following sub-steps:

[0110] (a) Obtain the current a0, a1, ..., a of the traditional model-predicted current control model at different load starting times x, and according to the current a0, a1, ...., a x Obtain the actual load disturbance current error set F = [d1, d2, ...., d x ], where x represents the total number of loads in the model-predicted current control model.

[0111] Specifically, this step uses the following formula:

[0112]

[0113] Where a0 is the current predicted by the model when the current control model is unloaded, d x for a x The corresponding load disturbance current error.

[0114] (b) Randomly select multiple load disturbance current errors from the load disturbance current error set F obtained in step (a) to form a data set (thus obtaining n data sets), and for each data set p (where p∈[1,3]), obtain all load disturbance current errors in the pth data set The mean of the p-th data set is used as the center of the Gaussian function corresponding to the p-th data set. Where n is a positive integer less than or equal to x, m p represents the total number of load disturbance current errors in the pth data set;

[0115] (c) For each data set p obtained in step (b), obtain the center of the Gaussian function corresponding to each load disturbance current error in the data set p and the pth data set The Euclidean distance between them is calculated, and all the Euclidean distances are summed up and averaged to obtain the Gaussian function center b corresponding to the pth data set. p , and obtain the load disturbance current error set consisting of all load disturbance current errors in the pth data set obtained in step (1-2) Each load disturbance current error in the pth data set corresponds to the center of the Gaussian function The difference between them is calculated, and all the differences are summed up and averaged to obtain the central width b of the neural network. n .

[0116] (d) For each data set p obtained in step (b), the m load disturbance current errors in the pth data set are Input Gaussian function to calculate and get multiple error outputs Construct the initial error output matrix H0;

[0117] (e) Obtain ideal weights based on the initial error output matrix H0 obtained in step (d) and the actual load disturbance current error set F obtained in step (a)

[0118] Specifically, this step uses the following formula:

[0119]

[0120] in, is the transpose of the initial error output matrix H0.

[0121] (f) Based on the output matrix H0 obtained in step (d) and the ideal weights obtained in step (e) Get an estimate of the noise observation for the first sampling period

[0122] Specifically, this step uses the following formula:

[0123]

[0124] Where ε is the estimation error of the neural network disturbance observer and satisfies ε≤ε n , ε n is the estimation error of the neural network disturbance observer after n iterations.

[0125] (g) Use the estimate of the interference observation obtained in step (f) As the initial value, the neural network disturbance observer is input for multiple iterations to obtain the disturbance observation value of the kth sampling period.

[0126] (1-3) Based on the d-axis and q-axis currents of the k-th sampling period and the d-axis and q-axis voltage signals u of the k-th sampling period obtained in step (1-1) d (k) and u q (k), and use the first-order forward Euler discretization method to respectively obtain the d-axis and q-axis current differential signals of the k-th sampling period obtained in (1-2) Discretization is performed to obtain the predicted values ​​of the d-axis and q-axis discrete currents of the k+1th sampling period output by the hyperlocal model.

[0127] Specifically, this step uses the following formula:

[0128]

[0129] Where, T s is the system sampling time.

[0130] The advantage of this sub-step is that if the sampling time is very short, the hyperlocal model can be approximated as a system model that is updated in real time.

[0131] (2) The predicted values ​​of the d-axis and q-axis discrete currents i of the k+1th sampling period of the hyperlocal model obtained in step (1) d (k+1), i q (k+1) Get the first optimal voltage vector u of the kth sampling period g1d (k) and u g1q (k), according to the optimal voltage vector u g1d (k) and u g1q (k) Calculate the voltage difference ΔE u1d (k), ΔE u1q (k) Obtain the second and third optimal voltage vectors u of the d and q axes according to the voltage difference table g2d (k) and u g2q (k) and u g2q (k) and u g3q (k), use the duty cycle direct calculation method to obtain the d and q axis duty cycles d of the kth sampling period after the hyperlocal model optimization nd (k), d nq (k), and according to the duty cycle d nd (k) and d nq (k) Control the pulse generator to output a drive signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the drive signal, thereby driving the synchronous reluctance motor to operate, where n∈[1,3].

[0132] The advantage of this step is that the inverter output voltage under the three-vector MPC strategy is closer to the ideal voltage vector, so it can solve the problem of poor motor steady-state performance existing in the existing MPC-based synchronous reluctance motor control method.

[0133] This step includes the following sub-steps:

[0134] (2-1) The predicted values ​​of the d-axis and q-axis discrete currents of the k+1th sampling period output by the hyperlocal model obtained in step (1) Get the ideal voltage vector of the d and q axes of the kth sampling period

[0135] Specifically, this step uses the following formula:

[0136]

[0137] in, is the d-axis and q-axis reference current of the k-th sampling period, and e d (k), eq (k) is the estimated error of the d-axis and q-axis currents in the k-th sampling period, and its initial value is 0; m is the error coefficient, and its value is 19000.

[0138] (2-2) The ideal voltage vector of the d and q axes of the kth sampling period obtained according to step (2-1) And the position signal θ(k) of the kth sampling period, obtain the α and β axis voltage cost functions.

[0139] Specifically, the obtained α and β axis voltage cost functions are formulated as follows:

[0140]

[0141] where u α (k) and u β (k) is the effective voltage vector of the kth sampling period of the α and β axes.

[0142] (2-3) The eight basic voltage vectors obtained according to the on-off state of the inverter are used as the effective voltage vector u of the α and β axes in the kth sampling period α (k) and u β (k) Substitute the α and β axis voltage cost functions obtained in step (2-2) to calculate and obtain the values ​​that make g α and g β The minimum voltage vector is taken as the first optimal voltage vector u of the d and q axes in the kth sampling period. g1d (k) and u g1q (k), the ideal voltage vector of the kth sampling period is respectively compared with the first optimal voltage vector u of the d and q axes g1d (k) and u g1q (k) to obtain the d-axis and q-axis voltage difference ΔE of the k-th sampling period respectively. u1d (k), ΔE u1q (k).

[0143] Specifically, the voltage difference between the α and β axes in the kth sampling period is calculated using the following formula:

[0144]

[0145] (2-4) Divide the voltage vector distribution diagram of the pulse generator into 6 sectors (such as Figure 2 As shown), the d and q axis voltage vector errors ΔE obtained according to step (2-3) u1d (k), ΔE u1q (k) The corresponding sectors in the voltage vector distribution diagram, and query the corresponding d-axis and q-axis second voltage vectors u of the k-th sampling period in the pre-established second optimal voltage vector table according to the acquired sectors. g2d(k) and u g2q (k), and query the corresponding d-axis and q-axis third voltage vectors u of the k-th sampling period in the pre-established third optimal voltage vector table according to the acquired sectors. g3d (k) and u g3q (k).

[0146] The difference ΔE between the first optimal vector and the ideal voltage vector in the kth sampling period can be calculated u1 (k) The second and third voltage vectors are obtained by looking up the table in the sector where the voltage vector is located.

[0147] Specifically, the second and third voltage vectors are selected as shown in Table 1 below.

[0148] Table 1

[0149]

[0150] The advantage of the above sub-steps (2-3) to (2-5) is that, under the three-vector MPC strategy, by designing the voltage cost function and redividing the vector sectors, the selection process of the three-vector combination is optimized and the complexity of the algorithm implementation is reduced.

[0151] (2-5) The nth voltage vector u of the d and q axes of the kth sampling period obtained in step (2-4) gnd (k) and u g x d (k), respectively obtain the current prediction value i of the d and q axes corresponding to the k-th sampling period of the n-th voltage vector of the d and q axes dn (k+1) and i qn (k+1).

[0152] Specifically, this step is calculated using the following formula:

[0153]

[0154] (2-6) The d-axis and q-axis reference currents of the k-th sampling period obtained in step (2-1) are and The current prediction value i of the d and q axes in the kth sampling period corresponding to the nth voltage vector of the d and q axes obtained in step (2-5) respectively dn (k+1) and i qn (k+1) to obtain the nth voltage vector u gnd (k) and u gnd (k) The current error Δε of the d and q axes of the k-th sampling period respectively dn (k) and Δε qn (k).

[0155] Specifically, this step is calculated using the following formula:

[0156]

[0157] (2-7) The nth voltage vector u obtained according to step (2-6) gnd (k) and u gnd (k) The d-axis and q-axis current errors Δε of the k-th sampling period respectively dn (k), Δε qn (k) Obtain the nth voltage vector u gnd (k) and u gnd (k) The corresponding duty cycle d nd (k) and d nq (k).

[0158] Specifically, this step is calculated using the following formula:

[0159]

[0160] where d 1d (k) represents d-axis u g1d (k) The corresponding duty cycle, d 2d (k) represents u g2d (k) The corresponding duty cycle, d 3d (k) represents u g3d (k) The corresponding duty cycle, d 1q (k) represents the q-axis u g1q (k) The corresponding duty cycle, d 2q (k) represents u g2q (k) The corresponding duty cycle, d 3q (k) represents u g3q (k) The corresponding duty cycle.

[0161] (2-8) The duty cycle d obtained according to step (2-7) nd (k) and d nq (k) The control pulse generator outputs a driving signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the driving signal, thereby driving the synchronous reluctance motor to operate.

[0162] (3) The d-axis and q-axis current signals i of the synchronous reluctance motor in the k+1th sampling cycle are obtained by sampling the current sensor. d (k+1), i q (k+1), and compare it with the d-axis and q-axis current estimates calculated by the neural network disturbance observer obtained in step (1-3) The difference is calculated to obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period. d(k+1) and e q (k+1), obtain the d and q axis error output matrix of the k+1th sampling period according to the obtained d and q axis current estimation errors and the estimation of the d and q axis weights for the k+1th sampling period The error output matrix is ​​obtained Estimation with weights Multiply them together to get an estimate of the disturbed observation The synchronous reluctance motor is controlled based on the estimation of the disturbance observation value.

[0163] The advantages of step (1) to step (3) are that the system disturbance is estimated and updated in real time through the current estimation error. This method does not need to identify the specific motor parameters of the controlled object, so it can solve the problem of reduced robustness caused by motor parameter mismatch in the existing MPC-based synchronous reluctance motor control method.

[0164] Specifically, this step includes the following sub-steps:

[0165] (3-1) Obtain the ABC three-phase current signal i of the k+1th sampling period after the synchronous reluctance motor starts running through the current sensor a (k+1), i b (k+1), i c (k+1), and according to the position signal θ(k+1) of the synchronous reluctance motor in the k+1th sampling period obtained by the position sensor, the d-axis and q-axis current signals i of the k+1th sampling period are obtained. d (k+1) and i q (k+1).

[0166] Specifically, this step uses the following formula:

[0167]

[0168] (3-2) The ideal voltage vector of the d and q axes of the k-th sampling period obtained according to step (2-1) and Get the d-axis and q-axis current estimates of the k-th sampling period respectively and

[0169] Specifically, the d-axis and q-axis current estimation in this step is obtained using the following formula:

[0170]

[0171] (3-3) The d-axis and q-axis currents i obtained in the k+1th sampling period in step (3-1) are d (k+1) and iq (k+1) and the d-axis and q-axis current estimates of the k-th sampling period obtained in step (3-2) and To obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period, d (k+1) and e q (k+1).

[0172] Specifically, this step calculates the current estimation error using the following formula:

[0173]

[0174] (3-4) Determine the d-axis and q-axis current estimation errors e of the k+1th sampling period obtained in step (3-3) d (k+1), e q (k+1) are all 0. If so, it means that the synchronous reluctance motor runs smoothly and the process ends. Otherwise, it goes to step (3-5);

[0175] (3-5) The d-axis and q-axis current estimation errors e of the k+1th sampling period obtained in step (3-3) are respectively d (k+1) and e q (k+1), and use the Gaussian function to obtain the error output h corresponding to the d and q axes of the k+1th sampling period respectively d (k+1) and h q (k+1), and use the error output h corresponding to the d and q axes d (k+1) and h q (k+1) Get the error output matrix of the k+1th sampling period

[0176] Specifically, this step uses the following formula:

[0177]

[0178] in, are the centers of Gaussian functions of d and q axes respectively; b d 、b q They are the center widths of the neural network on the d and q axes respectively.

[0179] (3-6) Obtain an estimate of the weight of the k+1th sampling period based on the error output matrix H(k+1) of the k+1th sampling period obtained in step (3-5)

[0180] Specifically, this step uses the following formula:

[0181]

[0182] In the formula, is the derivative of the estimate of the weight at the k+1th sampling period; is the derivative of the error of the estimated weight of the k+1th sampling period and satisfies W * (k) is the ideal value of the weight for the kth sampling period, is the estimate of the weight of the k+1th sampling period; e T (k+1) is the transpose of e(k+1), γ is a positive constant.

[0183] (3-7) Substitute the error output matrix H(k+1) of the k+1th sampling period obtained in step (3-5) with the weight estimate of the k+1th sampling period obtained in step (3-6) Multiply them together to obtain an estimate of the interfering observation value for the k+1th sampling period Then return to step (3-1).

[0184] In summary, the present invention combines model-free predictive current control with a neural network disturbance observer, introduces a hyperlocal model into the predictive control of the synchronous reluctance motor, and improves it into model-free predictive control; the unknown part in the hyperlocal model is observed by the neural network disturbance observer; the optimal three-vector combination and the corresponding duty cycle are obtained through an improved cost function structure and a duty cycle calculation method based on the current vector error, so that the synchronous reluctance motor can still maintain its stable output under complex working conditions, thereby significantly improving the robustness of the system.

[0185] The present invention solves the problems of low type, poor convergence and low estimation accuracy when using an extended observer to estimate unknown parts in a hyperlocal model. The present invention uses a neural network disturbance observer for estimation. The neural network disturbance observer adopts the ideas of pure integration, compensation and transfer function types, changes the structure of the extended observer, and makes the neural network disturbance observer higher than the extended observer by two types, with high accuracy and strong convergence.

[0186] The present invention aims at the problem that the steady-state performance of the motor is not satisfactory when a single voltage vector is applied in the control cycle of the classic MPC. Therefore, it is proposed to apply three voltage vectors in the control cycle to improve the steady-state performance. By designing the voltage cost function and re-dividing the vector sectors, the selection process of the three-vector combination is optimized, the steady-state performance of the system is improved, and the complexity of the algorithm implementation is reduced. It is also proposed to use the duty cycle calculation of the current vector error, so that the duty cycle calculation is not affected by the perturbation of the motor body parameters, and the system robustness is improved.

[0187] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer, characterized in that: The following steps are involved: (1) Obtain the ABC three-phase voltage signal u of the synchronous reluctance motor in the kth sampling period through voltage sensor sampling a (k) and u b (k) and u c (k), the position signal θ(k) of the kth sampling period of the synchronous reluctance motor is obtained by the position sensor, and the three-phase voltage signal u is calculated according to the position signal θ(k). a (k) and u b (k) and u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k) and u q (k), the d-axis and q-axis voltage signals u of the k-th sampling period are d (k) and u q (k) is input into the hyperlocal model for weighted processing, and the processing results are respectively compared with the d-axis and q-axis interference observation values ​​output by the neural network interference observer in the kth sampling period. Add to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and The first-order forward Euler discretization method is used to discretize the d-axis and q-axis current differential signals respectively, and the current obtained by current sensor sampling is combined to obtain the d-axis and q-axis discrete current prediction values ​​of the k-th sampling period output by the super-local model. Wherein, k is a positive integer. (2) The predicted values ​​of the d-axis and q-axis discrete currents i of the k+1th sampling period of the hyperlocal model obtained in step (1) d (k+1), i q (k+1) Get the first optimal voltage vector u of the kth sampling period g1d (k) and u g1q (k), according to the optimal voltage vector u g1d (k) and u g1q (k) Calculate the voltage difference ΔE u1d (k), ΔE u1q (k) Obtain the second and third optimal voltage vectors u of the d and q axes according to the voltage difference table g2d (k) and u g2q (k) and u g2q (k) and u g3q (k), use the duty cycle direct calculation method to obtain the d and q axis duty cycles d of the kth sampling period after the hyperlocal model optimization nd (k), d nq (k), and according to the duty cycle d nd (k) and d nq (k) Control the pulse generator to output a drive signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the drive signal, thereby driving the synchronous reluctance motor to operate, where n∈[1,3]. (3) The d-axis and q-axis current signals i of the synchronous reluctance motor in the k+1th sampling cycle are obtained by sampling the current sensor. d (k+1), i q (k+1), according to the d and q axis current signals i d (k+1), i q (k+1) respectively obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period d (k+1) and e q (k+1), obtain the d and q axis error output matrix of the k+1th sampling period according to the obtained d and q axis current estimation errors and the estimation of the d and q axis weights for the k+1th sampling period The error output matrix Estimation with weights Multiply them together to get an estimate of the disturbed observation The synchronous reluctance motor is controlled based on the estimation of the disturbance observation value.

2. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to claim 1, characterized in that: Step (1) includes the following sub-steps: (1-1) The ABC three-phase voltage signal u of the kth sampling period of the synchronous reluctance motor is obtained by sampling with a voltage sensor a (k) and u b (k) and u c (k), obtain the position signal θ(k) of the kth sampling period of the synchronous reluctance motor through the position sensor, and calculate the three-phase voltage signal u of the kth sampling period according to the position signal θ(k) of the kth sampling period a (k) and u b (k) and u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k) and u q (k). (1-2) The d-axis and q-axis voltage signals u of the k-th sampling period obtained in step (1-1) are d (k) and u q (k) Use the corresponding scaling factor α d , α q They are weighted and then compared with the interference observation value of the kth sampling period output by the neural network interference observer. Add together to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and (1-3) Based on the d-axis and q-axis currents of the k-th sampling period and the d-axis and q-axis voltage signals u of the k-th sampling period obtained in step (1-1) d (k) and u q (k), and the first-order forward Euler discretization method is used to respectively calculate the d-axis and q-axis current differential signals of the k-th sampling period obtained in (1-2) Discretization is performed to obtain the predicted values ​​of the d-axis and q-axis discrete currents of the k+1th sampling period output by the hyperlocal model.

3. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to claim 1 or 2, characterized in that: Step (1-2) is to use the following formula: where α d =1 / L d and α q =1 / L q represents the weighting factor, α d =1 / L d and α q =1 / L q These are all values ​​obtained through experience debugging; Steps (1-3) are performed using the following formula: Where T s is the system sampling time.

4. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to any one of claims 1 to 3, characterized in that: d-axis and q-axis interference observation values ​​of the k-th sampling period It is obtained through the following sub-steps: (a) Obtain the current a0, a1, ..., a of the traditional model-predicted current control model at different load starting times x , and according to the current a0, a1, ...., a x Obtain the actual load disturbance current error set F = [d1, d2, ...., d x ], where x represents the total number of loads in the model-predicted current control model. (b) Randomly select multiple load disturbance current errors from the load disturbance current error set F obtained in step (a) to form a data set n times. For each data set p (where p∈[1,3]), obtain all load disturbance current errors in the pth data set The mean of the p-th data set is used as the center of the Gaussian function corresponding to the p-th data set. Where n is a positive integer less than or equal to x, m p represents the total number of load disturbance current errors in the pth data set; (c) For each data set p obtained in step (b), obtain the center of the Gaussian function corresponding to each load disturbance current error in the data set p and the pth data set The Euclidean distance between them is calculated, and all the Euclidean distances are summed up and averaged to obtain the Gaussian function center b corresponding to the pth data set. p , and obtain the load disturbance current error set consisting of all load disturbance current errors in the pth data set obtained in step (1-2) Each load disturbance current error in the pth data set corresponds to the center of the Gaussian function The difference between them is calculated, and all the differences are summed up and averaged to obtain the central width b of the neural network. n . (d) For each data set p obtained in step (b), the m load disturbance current errors in the pth data set are Input Gaussian function to calculate and get multiple error outputs Construct the initial error output matrix H0; (e) Obtain the ideal weight W0 based on the initial error output matrix H0 obtained in step (d) and the actual load disturbance current error set F obtained in step (a) * ; (f) Based on the output matrix H0 obtained in step (d) and the ideal weight W0 obtained in step (e) * Get an estimate of the noise observation for the first sampling period (g) Use the estimate of the interference observation obtained in step (f) As the initial value, the neural network disturbance observer is input for multiple iterations to obtain the disturbance observation value of the kth sampling period.

5. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to claim 4, characterized in that: Step (a) uses the following formula: Where a0 is the current predicted by the model when the current control model is unloaded, d x for a x The corresponding load disturbance current error. Step (e) is to use the following formula: in, is the transpose of the initial error output matrix H0. Step (f) uses the following formula: Where ε is the estimation error of the neural network disturbance observer and satisfies ε≤ε n , ε n is the estimation error of the neural network disturbance observer after n iterations.

6. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to claim 5, characterized in that: Step (2) includes the following sub-steps: (2-1) The predicted values ​​of the d-axis and q-axis discrete currents of the k-th sampling period output by the hyperlocal model obtained in step (1) Get the ideal voltage vector of the d and q axes of the kth sampling period (2-2) The ideal voltage vector of the d and q axes of the kth sampling period obtained according to step (2-1) And the position signal θ(k) of the kth sampling period, obtain the voltage cost function g of the α and β axes α and g β . (2-3) The eight basic voltage vectors obtained according to the on-off state of the inverter are used as the effective voltage vector u of the α and β axes in the kth sampling period α (k) and u β (k) Substitute the α and β axis voltage cost functions obtained in step (2-2) to calculate and obtain the values ​​that make g α and g β The minimum voltage vector is taken as the first optimal voltage vector u of the d and q axes in the kth sampling period. g1d (k) and u g1q (k), the ideal voltage vector of the kth sampling period is respectively compared with the first optimal voltage vector u of the d and q axes g1d (k) and u g1q (k) to obtain the d-axis and q-axis voltage difference ΔE of the k-th sampling period respectively. u1d (k), ΔE u1q (k). (2-4) Divide the voltage vector distribution diagram of the pulse generator into 6 sectors, and calculate the d-axis and q-axis voltage vector errors ΔE obtained in step (2-3) u1d (k), ΔE u1q (k) The corresponding sectors in the voltage vector distribution diagram, and query the corresponding d-axis and q-axis second voltage vectors u of the k-th sampling period in the pre-established second optimal voltage vector table according to the acquired sectors. g2d (k) and u g2q (k), and query the corresponding d-axis and q-axis third voltage vectors u of the k-th sampling period in the pre-established third optimal voltage vector table according to the acquired sectors. g3d (k) and u g3q (k). (2-5) The nth voltage vector u of the d and q axes of the kth sampling period obtained in step (2-4) gnd (k) and u gnd (k), respectively obtain the current prediction value i of the d and q axes corresponding to the k-th sampling period of the n-th voltage vector of the d and q axes dn (k+1) and i qn (k+1). (2-6) The d-axis and q-axis reference currents of the k-th sampling period obtained in step (2-1) are and The current prediction value i of the d and q axes in the kth sampling period corresponding to the nth voltage vector of the d and q axes obtained in step (2-5) respectively dn (k+1) and i qn (k+1) to obtain the nth voltage vector u gnd (k) and u gnd (k) The current error Δε of the d and q axes of the k-th sampling period respectively dn (k) and Δε qn (k). (2-7) The nth voltage vector u obtained according to step (2-6) gnd (k) and u gnd (k) The d-axis and q-axis current errors Δε of the k-th sampling period respectively dn (k), Δε qn (k) Obtain the nth voltage vector u gnd (k) and u gnd (k) The corresponding duty cycle d nd (k) and d nq (k). (2-8) The duty cycle d obtained according to step (2-7) nd (k) and d nq (k) The control pulse generator outputs a driving signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the driving signal, thereby driving the synchronous reluctance motor to operate.

7. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to claim 6, characterized in that: Step (2-1) uses the following formula: in, is the d-axis and q-axis reference current of the k-th sampling period, and e d (k), e q (k) is the estimated error of the d-axis and q-axis currents in the k-th sampling period, and its initial value is 0; m is the error coefficient, and its value is 19000. Step (2-2) uses the following formula: where u α (k) and u β (k) is the effective voltage vector of the kth sampling period of the α and β axes. The calculation of the voltage difference between the α and β axes in the kth sampling period in step (2-3) is performed using the following formula: Steps (2-5) are calculated using the following formula: Steps (2-6) are calculated using the following formula: Steps (2-7) are calculated using the following formula: where d 1d (k) represents d-axis u g1d (k) The corresponding duty cycle, d 2d (k) represents u g2d (k) The corresponding duty cycle, d 3d (k) represents u g3d (k) The corresponding duty cycle, d 1q (k) represents the q-axis u g1q (k) The corresponding duty cycle, d 2q (k) represents u g2q (k) The corresponding duty cycle, d 3q (k) represents u g3q (k) The corresponding duty cycle.

8. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to claim 7, characterized in that: Step (3) includes the following sub-steps: (3-1) Obtain the ABC three-phase current signal i of the k+1th sampling period after the synchronous reluctance motor starts running through the current sensor a (k+1), i b (k+1), i c (k+1), and according to the position signal θ(k+1) of the synchronous reluctance motor in the k+1th sampling period obtained by the position sensor, the d-axis and q-axis current signals i of the k+1th sampling period are obtained. d (k+1) and i q (k+1). (3-2) The ideal voltage vector of the d and q axes of the k-th sampling period obtained according to step (2-1) and Get the d-axis and q-axis current estimates of the k-th sampling period respectively and (3-3) The d-axis and q-axis currents i obtained in the k+1th sampling period in step (3-1) are d (k+1) and i q (k+1) and the d-axis and q-axis current estimates of the k-th sampling period obtained in step (3-2) and To obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period, d (k+1) and e q (k+1). (3-4) Determine the d-axis and q-axis current estimation errors e of the k+1th sampling period obtained in step (3-3) d (k+1), e q (k+1) are all 0. If so, it means that the synchronous reluctance motor runs smoothly and the process ends. Otherwise, it goes to step (3-5); (3-5) The d-axis and q-axis current estimation errors e of the k+1th sampling period obtained in step (3-3) are respectively d (k+1) and e q (k+1), and use the Gaussian function to obtain the error output h corresponding to the d and q axes of the k+1th sampling period respectively d (k+1) and h q (k+1), and use the error output h corresponding to the d and q axes d (k+1) and h q (k+1) Get the error output matrix of the k+1th sampling period (3-6) Obtain an estimate of the weight of the k+1th sampling period based on the error output matrix H(k+1) of the k+1th sampling period obtained in step (3-5) (3-7) Substitute the error output matrix H(k+1) of the k+1th sampling period obtained in step (3-5) with the weight estimate of the k+1th sampling period obtained in step (3-6) Multiply them together to obtain an estimate of the interfering observation value for the k+1th sampling period Then return to step (3-1).

9. The method for realizing model-free predictive current control of a synchronous reluctance motor based on a neural network disturbance observer according to claim 8, characterized in that: Step (3-1) uses the following formula: The d-axis and q-axis current estimates in step (3-2) are obtained using the following formula: Step (3-3) is to calculate the current estimation error using the following formula: Steps (3-5) use the following formula: in, are the centers of Gaussian functions of d and q axes respectively; b d , b q They are the center widths of the neural network on the d and q axes respectively. Steps (3-6) are done using the following formula: in is the derivative of the estimate of the weight at the k+1th sampling period; is the derivative of the error of the weight estimate for the k+1th sampling period and satisfies W * (k) is the ideal value of the weight for the kth sampling period, is the estimate of the weight of the k+1th sampling period; e T (k+1) is the transpose of e(k+1), γ is a positive constant.

10. A system for realizing model-free predictive current control of synchronous reluctance motor based on neural network disturbance observer, characterized in that: include: The first module is used to obtain the ABC three-phase voltage signal u of the kth sampling period of the synchronous reluctance motor through voltage sensor sampling. a (k) and u b (k) and u c (k), the position signal θ(k) of the kth sampling period of the synchronous reluctance motor is obtained by the position sensor, and the three-phase voltage signal u is calculated according to the position signal θ(k). a (k) and u b (k) and u c (k) Coordinate transformation is performed to obtain the d-axis and q-axis voltage signals u of the k-th sampling period respectively. d (k) and u q (k), the d-axis and q-axis voltage signals u of the k-th sampling period are d (k) and u q (k) is input into the hyperlocal model for weighted processing, and the processing results are respectively compared with the d-axis and q-axis interference observation values ​​output by the neural network interference observer in the kth sampling period. Add to obtain the d-axis and q-axis current differential signals of the k-th sampling period respectively and The first-order forward Euler discretization method is used to discretize the d-axis and q-axis current differential signals respectively, and the current obtained by current sensor sampling is combined to obtain the d-axis and q-axis discrete current prediction values ​​of the k-th sampling period output by the super-local model. Wherein, k is a positive integer. The second module is used to obtain the d-axis and q-axis discrete current prediction values ​​i of the k+1th sampling period of the super-local model obtained in the first module. d (k+1), i q (k+1) Get the first optimal voltage vector u of the kth sampling period g1d (k) and u g1q (k), according to the optimal voltage vector u g1d (k) and u g1q (k) Calculate the voltage difference ΔE u1d (k), ΔE u1q (k) Obtain the second and third optimal voltage vectors u of the d and q axes according to the voltage difference table g2d (k) and u g2q (k) and u g2q (k) and u g3q (k), use the duty cycle direct calculation method to obtain the d and q axis duty cycles d of the kth sampling period after the hyperlocal model optimization nd (k), d nq (k), and according to the duty cycle d nd (k) and d nq (k) Control the pulse generator to output a drive signal to the inverter, so that the inverter outputs a voltage to the synchronous reluctance motor according to the drive signal, thereby driving the synchronous reluctance motor to operate, where n∈[1,3]. The third module is used to obtain the d-axis and q-axis current signals i of the synchronous reluctance motor in the k+1th sampling cycle through current sensor sampling. d (k+1), i q (k+1), according to the d and q axis current signals i d (k+1), i q (k+1) respectively obtain the d-axis and q-axis current estimation errors e of the k+1th sampling period d (k+1) and e q (k+1), obtain the d and q axis error output matrix of the k+1th sampling period according to the obtained d and q axis current estimation errors and the estimation of the d and q axis weights for the k+1th sampling period The error output matrix Estimation with weights Multiply them together to get an estimate of the disturbed observation The synchronous reluctance motor is controlled based on the estimation of the disturbance observation value.

Images