An improved sensorless control method for permanent magnet synchronous motor

Abstract

The application discloses an improved position sensorless control method of a permanent magnet synchronous motor, adopts a sliding mode control in a speed loop part, and links electromagnetic torque with speed. By using a Lagrange extreme value method, a group of optimal combinations are found in an electromagnetic torque formula of the permanent magnet synchronous motor, and maximum torque current ratio control is realized. An observation value of back electromotive force is fed back to a current observation link, a full-order sliding mode observer is established, and a recursive probability wavelet fuzzy neural network is used to copy an s-type function in the full-order sliding mode observer. Finally, rotor position information is obtained from the back electromotive force based on a phase-locked loop. The application uses the sliding mode variable structure in the speed loop to improve disturbance resistance and control precision of speed regulation, MTPA control reduces copper loss, improves efficiency of the permanent magnet synchronous motor, the new full-order sliding mode observer avoids phase lag caused by low-pass filtering, and an adaptive law designed by the application minimizes observation error and chattering under condition change.

Description

Technical Field

[0001] This invention relates to the field of motor control, and specifically to an improved sensorless control method for a permanent magnet synchronous motor. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) possess advantages such as high efficiency, high power density, high torque density, and good dynamic response. They are widely used in industrial automation and renewable energy fields. Especially in electric vehicles, due to their low energy consumption, high acceleration performance, and ability to provide longer range, they have become a common type of drive motor. PMSM control systems require fast and accurate speed control capabilities and good dynamic response characteristics. By comparing the desired speed with the actual speed, the control strategy is adjusted to ensure stable operation of the motor within a predetermined speed range. Speed ​​is a crucial parameter in PMSM control systems. In traditional control systems, position or speed sensors are used to accurately acquire the position and speed information of the motor rotor. However, sensor installation and calibration require additional cost and workload and are also affected by environmental conditions and sensor lifespan. In contrast, sensorless control of PMSMs reduces system cost and complexity. Furthermore, it avoids sensor failure, thereby improving system reliability and robustness.

[0003] Sensorless control technology for permanent magnet synchronous motors can be divided into two basic types: high-frequency injection-based methods and observation-based methods. High-frequency injection refers to injecting high-frequency voltage or current, superimposing the high-frequency signal onto the base excitation. By detecting the high-frequency response, the motor position can be estimated even at low speeds or when stationary. However, this method relies on the accuracy of high-frequency signal detection. Furthermore, the injection of high-frequency excitation is prone to generating high-frequency noise, thus degrading system performance. Observation-based methods are mainly used for medium- and high-speed applications. They use the motor current as the input to an observer, which is designed to estimate the rotor position. Observation methods include Extended Kalman Filter (EKF), Model Reference Adaptive Control (MRAC), and Sliding Mode Observer (SMO). Among these, EKF and MRAC methods rely on the mathematical model of the motor, and the accuracy of the model directly affects the accuracy of rotor position estimation. In contrast, the SMO method has a lower dependence on the motor model and is robust to changes in motor parameters and external disturbances, showing broad application prospects. However, the traditional SMO method uses the current error as the input signal to the observer and generates a pulse signal as the estimated back electromotive force (EMF). This necessitates the design of filtering techniques to process pulse signals, placing high demands on the system's sampling frequency and filtering characteristics. Furthermore, the filtering process introduces delay, requiring the design of a compensator to mitigate this issue, thus increasing the system's complexity. Summary of the Invention

[0004] Purpose of the invention: In order to solve the problems existing in the prior art, the present invention provides an improved sensorless control method for permanent magnet synchronous motors, which improves the anti-disturbance and control accuracy of speed regulation, enhances the description of the dynamic behavior and control requirements of the system, and can obtain high-precision rotor speed and position information with less chattering even under speed or load changes.

[0005] Technical solution: This invention discloses an improved sensorless control method for permanent magnet synchronous motors, comprising the following steps:

[0006] Step 1: Establish the electrical angular velocity error e, design the sliding surface s, and use sliding mode control to control the electrical angular velocity ω in the speed loop. e With electromagnetic torque T e Connect with;

[0007] Step 2: The electromagnetic torque T in Step 1 e The input is fed to the MTPA control module, and the Lagrange extremum method is used to find a set of optimal values ​​for i. d and i q The combination results in electromagnetic torque T e Maximum, stator current i s Minimize copper loss;

[0008] Step 3: Establish the full-order state equations of the permanent magnet synchronous motor in a two-phase stationary coordinate system;

[0009] Step 4: Based on the full-order state equations established in Step 3, construct a full-order sliding mode observer to measure the voltage u of the permanent magnet synchronous motor in the two-phase stationary coordinate system. s The output stator current is observed by a full-order sliding mode observer. Based on the stator current observation value With stator current i s The difference is used as the current observation error.

[0010] Step 5: Adjust the current observation error from Step 4. After simulating the s-function module using a recursive probabilistic wavelet fuzzy neural network (RPWFNN), the sliding mode control rate V(s) of the permanent magnet synchronous motor in the two-phase stationary coordinate system is output. i );

[0011] Step 6: Convert the sliding mode control law V(s) from Step 5. i The feedback gain matrix G is fed back to the full-order sliding mode observer as its input, and the back electromotive force observation value is output after passing through the full-order sliding mode observer.

[0012] Step 7: Calculate the back electromotive force observation value from Step 6. The back EMF observation is fed back as a state variable to the full-order sliding mode observer for update calculation. After processing by the phase-locked loop module, the observed value of the permanent magnet synchronous motor rotor position is output. and electric angular velocity observations The rotor position observation value The observed electrical angular velocity is fed back to the phase-locked loop module as its input. It is also fed back to the full-order sliding mode observer as its input.

[0013] Furthermore, in step 1, the speed loop based on sliding mode control is designed as follows:

[0014] The electrical angular velocity error e is as follows:

[0015] e = ω e * -ω e

[0016] Where: ω e * To set the electric angular velocity, ω e Electric angular velocity;

[0017] The sliding surface s is as follows:

[0018]

[0019] in: Let be the first derivative of the electric angular velocity, and c be a positive constant;

[0020] The sliding mode control law is as follows:

[0021]

[0022] Where α and β are positive constants. J is the moment of inertia, n p For extreme logarithms, It is the first derivative of the electromagnetic torque.

[0023] Furthermore, in step 2, the MTPA control module is designed as follows:

[0024]

[0025] Where: L d L q i represents the equivalent inductance of the permanent magnet synchronous motor along the d-axis and q-axis. d For the d-axis stator current, i q Let ψ be the q-axis stator current. f It is a permanent magnet flux linkage.

[0026] Furthermore, the full-order state equation in step 3 is:

[0027] The model of the permanent magnet synchronous motor in a two-phase stationary coordinate system is as follows:

[0028]

[0029] Where: u α For the α-axis stator voltage, u β For the β-axis stator voltage, i α Let i be the α-axis stator current. β Let θ be the stator current along the β axis. e For the rotor position of the permanent magnet synchronous motor, e α e is the back electromotive force along the α axis. β R is the back electromotive force along the β axis. s This refers to the stator resistance value of a permanent magnet synchronous motor.

[0030] The extended back electromotive force and its rate of change satisfy the following relationship:

[0031]

[0032] The full-order state equations of the permanent magnet synchronous motor are:

[0033]

[0034] in: A 12 =diag(-1 / L) d -1 / L d ), B1 = diag(1 / L) d 1 / L d ), u s =[u α u β ] T Let e ​​be the stator voltage in a two-phase stationary coordinate system, e = [e α e β ] T For the back electromotive force, i s =[i α i β ] T Stator current;

[0035] Stator current i s The formula is as follows:

[0036]

[0037] Where: i a i b and i c This refers to the three-phase stator current of the permanent magnet synchronous motor.

[0038] Furthermore, the full-order sliding mode observer in step 4 is:

[0039]

[0040] in: A 12 =diag(-1 / L) d -1 / L d B1 = diag(1 / L) d 1 / L d ), It is the sliding mode control rate. It is the feedback gain matrix. It is the sliding mode surface for the α-axis and β-axis stator currents, k i k e It is the switching gain of the sliding mode observer. These are stator current observations. These are the observed values ​​of the α-axis stator current. These are the observed values ​​of the β-axis stator current. The back electromotive force observation value, This is the observed value of the back electromotive force along the α axis. This represents the observed value of the β-axis electromotive force.

[0041] Furthermore, the sliding mode control rate V(s) output by the recursive probabilistic wavelet fuzzy neural network (RPWFNN) in step 5... i )for:

[0042]

[0043] Where: N is the total number of rules. Indicates the output. This represents the l-th output of the rule layer. This represents the adjustable weights between the rule layer and the output layer;

[0044] The adaptive law design for RPWFNN parameters is as follows:

[0045]

[0046]

[0047]

[0048]

[0049] in: The wavelet weights of the wavelet layer, m ij and c ijThese represent the center point and width of the i-th input, the j-th Gaussian function, and the distance to the node in this layer.

[0050] Furthermore, the phase-locked loop operation process in step 7 is as follows:

[0051] According to the formula for extended back electromotive force, the rotor position error is:

[0052]

[0053] When steady state is reached, the error between the actual angle and the estimated angle of the motor is very small. Therefore, the above equation can be equivalent to:

[0054]

[0055] In the formula, E ex To expand the back electromotive force amplitude,

[0056] After the rotor position error signal ε is amplified and integrated, the observed electrical angular velocity of the permanent magnet synchronous motor is obtained. Observations of the electric angular velocity of the motor The rotor position observations are obtained by performing integration. Rotor position observations The feedback signals for the phase-locked loop (PLL) module are repeated sequentially to form a complete PLL position tracking structure.

[0057] Beneficial effects:

[0058] This invention utilizes the sliding mode variable structure in the speed loop, which eliminates the need for precise mathematical models and exhibits strong robustness, thereby improving the disturbance rejection and control accuracy of speed regulation. MTPA control enables the stator current i s Minimizing the amplitude reduces copper losses and improves the efficiency of the permanent magnet synchronous motor. Full-order SMO (FSMO) is an improvement on the traditional SMO method, enhancing the description of system dynamics and control requirements by introducing extended back electromotive force (EMF) state variables. FSMO also possesses inherent second-order filtering characteristics, eliminating the need for additional filtering stages. Considering the approximation, learning, and inference capabilities of the recursive probabilistic wavelet fuzzy neural network (RPWFNN) for arbitrary smooth functions, RPWFNN can be used to simulate the sigmoid function used in FSMO. Combining the adaptive law of network parameters, in the fuzzy neural network, the sliding surface is the input variable, and the output of the fuzzy neural network is the equivalent control law. The designed RPWFNN can adaptively update the network parameters according to the actual operating state of the permanent magnet synchronous motor system. Therefore, even under conditions of speed or load variations, high-precision rotor speed and position information with less chattering can be obtained. Therefore, this invention has the following advantages:

[0059] 1. Strong robustness. When internal parameters change or are subject to external disturbances, it generates a well-defined sinusoidal waveform for the estimated back EMF, exhibiting lower harmonic interference over a wider speed range. This reduces the impact of current errors on the back EMF and improves anti-interference capability.

[0060] 2. High accuracy. The designed recursive probabilistic wavelet fuzzy neural network (RPWFNN) can adaptively update network parameters according to the actual operating state of the permanent magnet synchronous motor system. Therefore, even under conditions of speed or load variation, high-precision rotor speed and position information with minimal chattering can be obtained.

[0061] 3. Wide applicability. The designed full-order sliding mode observer is applicable to permanent magnet synchronous motors of different power in different fields. It can significantly reduce the overshoot and static error of various motor performance indicators and effectively reduce system chattering.

[0062] 4. Simple and practical. Sliding mode control is a commonly used nonlinear control method. It does not require an accurate system model, but only requires designing a sliding mode controller to achieve stable control of the system according to actual control requirements. Attached Figure Description

[0063] Figure 1 A simplified control structure diagram for a permanent magnet synchronous motor;

[0064] Figure 2 This is a schematic diagram of a full-order sliding mode observer structure based on the improved RPWFNN.

[0065] Figure 3 This is a schematic diagram of an orthogonal phase-locked loop position tracker.

[0066] Figure 4 A schematic diagram of the RPWFNN neural network;

[0067] Figure 5 A comparison of the speed observation error under constant load speed change conditions between the traditional algorithm and the simulation algorithm of this invention;

[0068] Figure 6 A comparison of rotor position observation errors under constant load and speed change conditions using traditional algorithms and the simulation algorithm of this invention;

[0069] Figure 7 A comparison of the speed observation error under constant speed load sudden change in the traditional algorithm and the simulation algorithm of this invention;

[0070] Figure 8 This paper compares the rotor position observation error under the traditional algorithm and the simulation algorithm of this invention when the load changes abruptly at a constant speed. Detailed Implementation

[0071] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0072] This invention discloses an improved sensorless control method for a permanent magnet synchronous motor, comprising the following steps:

[0073] Step 1. Design of the sliding mode regulator for the vector system of the permanent magnet synchronous motor

[0074] Let e ​​be the rotational speed error, then e = ω e * -ω e ,but:

[0075]

[0076] In the formula: The rate of change of velocity error, T is the second derivative of the velocity error. L This represents the load torque.

[0077] make J is the moment of inertia, n p For extreme logarithms, Then equation (1) above can be rearranged as:

[0078]

[0079] The sliding surface is designed as follows: Where c must satisfy the Hurwitz condition, i.e., c > 0. The Lyapunov function is defined as follows: but:

[0080]

[0081] According to the Lyapunov stability criterion, a system is stable when V is positive definite and has continuous first-order partial derivatives, and the first-order partial derivatives of V are semi-negative definite. To ensure this... The sliding mode control law is designed as follows:

[0082]

[0083] but:

[0084]

[0085] The above analysis shows that the designed control law can ensure that the system asymptotically stabilizes and approaches the sliding mode.

[0086] Step 2. Design of the MTPA control module

[0087] Inductance (L) on the direct shaft of the permanent magnet synchronous motor d) and the inductance on the quadrature axis (L) q The values ​​are not equal. Using maximum torque-to-current ratio control (MTPA) can effectively utilize the additional torque generated by the structure of this type of motor. In the electromagnetic torque formula of the PMSM, i... d and i q Treating i as variables, find an optimal set of i. d and i q Combined, the electromagnetic torque T is obtained. e The maximum value can be obtained by using the Lagrange multiplication table to find the extrema, specifically the value of i corresponding to the maximum electromagnetic torque. d and i q Combining these elements allows for MTPA control.

[0088] The process of finding the extrema using the Lagrange multiplier is as follows: The formula for the stator current of the motor is shown in the equation:

[0089]

[0090] Setting the electromagnetic torque formula to zero as the constraint λ of the Lagrange extremum method, the form is as shown in the equation:

[0091]

[0092] Among them, L d L q i represents the equivalent inductance of the permanent magnet synchronous motor along the d-axis and q-axis. d For the d-axis stator current, i q Let ψ be the q-axis stator current. f For permanent magnet flux linkage, under condition i s Find T at a given time e The maximum value of can be expressed as follows according to the Lagrange extreme value method:

[0093]

[0094] Perform i on the newly created function H respectively d i q And by taking the partial derivative with respect to λ, we can obtain:

[0095]

[0096] According to Lagrange's method for finding extrema, setting the above three equations to 0, i.e. The final result can be obtained:

[0097]

[0098]

[0099] Substituting equation (10) into the electromagnetic torque equation (11), we get:

[0100] 9n p 2 (L d -L q ) 2 i q 4 +6T e ψ f n p i q -4T e 2 =0

[0101] Take n p =4 gives:

[0102]

[0103] via i d and i q The relational expression is used to calculate i. d and i q This allows for MTPA control.

[0104] Step 3. Model establishment of the permanent magnet synchronous motor in a two-phase stationary coordinate system:

[0105]

[0106] Where: u α For the α-axis stator voltage, u β For the β-axis stator voltage, i α Let i be the α-axis stator current. β For the β-axis stator current, e α e is the back electromotive force along the α axis. β R is the back electromotive force along the β axis. s L is the stator resistance of the permanent magnet synchronous motor. d L q ω represents the equivalent inductance of the permanent magnet synchronous motor along the d-axis and q-axis. e Let θ be the electric angular velocity. e This indicates the rotor position of the permanent magnet synchronous motor.

[0107] Typically, the mechanical time constant of a system is much larger than its electromagnetic time constant. Therefore, within the PWM control cycle, the angular velocity can be considered a constant value. At this point, the extended back electromotive force and its rate of change satisfy the following relationship:

[0108]

[0109] Combining equations (13) and (14), the full-order state equations of the permanent magnet synchronous motor are:

[0110]

[0111] in: A 12 =diag(-1 / L) d -1 / L d ), B1 = diag(1 / L) d 1 / L d ), u s =[u α u β ] T Let e ​​be the stator voltage in a two-phase stationary coordinate system, e = [e α e β ] T For the back electromotive force, i s =[i α i β ] T This refers to the stator current.

[0112] Stator current i s The formula is as follows:

[0113]

[0114] Where: i a i b and i c This refers to the three-phase stator current of the permanent magnet synchronous motor.

[0115] Step 4. Design of a full-order sliding mode observer

[0116] From the full-order state equation in equation (15), it can be seen that the rotor's position information exists only in the electromotive force (EMF). Therefore, we can use the extended EMF to observe the rotor's position. In the state equation, the stator current is the only measurable physical quantity. Therefore, the sliding surface can be defined as:

[0117]

[0118] in: These are the observed values ​​of the α-axis stator current. These are the observed values ​​of the β-axis stator current.

[0119] Based on the above IPMSM full-order state equations, a full-order sliding mode observer with stator current and extended back electromotive force as state variables is established, as shown in equation (18):

[0120]

[0121] in: A 12=diag(-1 / L) d -1 / L d B1 = diag(1 / L) d 1 / L d ), It is the sliding mode control rate. It is the feedback gain matrix, k i k e It is the switching gain of the sliding mode observer. These are stator current observations. The back electromotive force observation value, This is the observed value of the back electromotive force along the α axis. This represents the observed value of the β-axis electromotive force.

[0122] Then the observation vector of EMF(e) in equation (18) can be expressed as:

[0123]

[0124] Where: s is the Laplace operator.

[0125] When equation (18) is subtracted from equation (15), and under normal circumstances, when the sliding mode occurs, the observed rotational speed is... It can converge to the actual rotational speed ω e ,Right now Then the dynamic equation of the observation error can be obtained, as shown in equation (20):

[0126]

[0127] in: The observation error is due to the back electromotive force.

[0128] When the system reaches the sliding mode state, the observer value of the stator current converges to the actual value, i.e. The error of the electromotive force observer can be obtained from the dynamic equation of the current observation error in equation (20):

[0129]

[0130] Substituting equation (21) into equation (20), the dynamic equation for the electromotive force observation error can be rewritten as:

[0131]

[0132] Step 5. Proof of stability of the full-order sliding mode observer

[0133] Consider positive definite Lyapunov candidate functions Substituting (20) and (22) into the derivative of V1 with respect to time, we get:

[0134]

[0135] If the feedback gain k is selected i Satisfying k i >max(|e α |,|e β The condition |) is met. This ensures that It is a negative semidefinite function, which means that It is a bounded function, and according to Lyapunov stability theory and Barbalat's lemma, the designed sliding mode surface converges to zero as t→∞. Therefore, the sliding motion throughout the entire observation process can be guaranteed, and the designed FSMO is stable even in the presence of uncertainties in the system.

[0136] As shown in (19), the FSMO constructed based on the full-order model of the permanent magnet synchronous motor includes both current observation and electromotive force (EMF), and simultaneously feeds back the observed EMF value to the current observation loop. Furthermore, the observed EMF value inherently possesses the second-order low-pass filtering characteristic known from (22), which can effectively filter out high-frequency noise. Compared with traditional sliding mode observers, no additional low-pass filter is required, thus avoiding the phase lag problem in rotor position observation and improving observation accuracy. Rearranging equation (22), we can obtain the dynamic equation for the EMF observation error as a second-order system, where the variables between the two axes are independent and have the same transient response process. The dynamic equation can be written as:

[0137]

[0138] in: If the designed FSMO(k i and k e The gain selection ensures that the eigenvalues ​​of the corresponding characteristic equation (23) lie strictly in the left half of the complex plane, and the optimal damping ratio is obtained. This ensures the observer's expected convergence speed.

[0139] Step 6. Design of the s-function based on the recursive probabilistic wavelet fuzzy neural network (RPWFNN):

[0140] 6-layer RPWFNN Figure 4 As shown, it includes an input layer, a membership layer, a probability layer, a wavelet layer, a rule layer, and an output layer, which are used to implement the design of the RPWFNN of this invention.

[0141] (1) The network and output of the neuron input layer are as follows:

[0142]

[0143]

[0144] in: and These are the input and output of the i-th neuron, respectively, and N is the total number of rules.

[0145] The input to the proposed RPWFNN is the stator current observation error. and its derivative

[0146] (2) The membership layer uses a Gaussian function to implement the fuzzification operation of RPWFNN. The input and output of the nodes in this layer are represented as follows:

[0147]

[0148]

[0149] in: and These represent the center point and width of the i-th input, the j-th Gaussian function, and the distance to the node in this layer. This represents the output of the j-th neuron in the membership layer.

[0150] (3) The probability layer also chooses a Gaussian function as the acceptor field function for this layer, expressed as:

[0151]

[0152]

[0153] in: and These represent the center point and width of the Gaussian function from the j-th input to the p-th term of the node in this layer. This represents the output of the j-th neuron in the probability layer.

[0154] (4) The wavelet layer consists of k wavelet functions as shown below:

[0155]

[0156]

[0157] in: The wavelet weights of the wavelet layer; For the i-th k-th term, output to the wavelet and layer node, ψ k (N) represents the output of the k-th wavelet layer node.

[0158] (5) In the rule layer, each node is represented by addition, and a product operation is performed to obtain the Mamdani inference set. Considering the fuzzy level group as the independent variable, Bayes' theorem is used to process the probability information, as shown in (33). In addition, in order to achieve the cyclic characteristic, the output of each rule node is fed back to itself as input. Therefore, the previous values ​​can be memorized through the feedback structure shown in (34). In addition, the wavelet output is also processed in (34). The node output of this layer is represented as:

[0159]

[0160]

[0161]

[0162] in: The connection weight between the rule layer and the probability layer is set to 1; The connection weights of nodes are used to achieve recursion. It is the output of the rules layer.

[0163] (6) Each node in the output layer is the sum of the product of the output of the rule layer and the adjustable weights between the rule layer and the output layer. The output of this layer is:

[0164]

[0165]

[0166] in: The output of RPWFNN is the sliding mode control rate V(s) of the permanent magnet synchronous motor in the two-phase stationary coordinate system. i ), This represents the connection weights between the rule layer and the output layer.

[0167] The weights and parameters of the recursive probabilistic wavelet fuzzy neural network based on gradient descent, and the objective function E(N) can be defined as follows:

[0168]

[0169] The main objective is to minimize the error function E(N) to obtain the online learning parameters. The online learning algorithm is described in detail below:

[0170] 1) Layer 6: The propagation error term is represented as:

[0171]

[0172] The update amount of the connection weights in this layer is described as follows:

[0173]

[0174] In the formula, η1 is the learning rate. Therefore, the connection weights are updated according to formula (40).

[0175]

[0176] 2) Layer 5: The two error terms propagated in this layer are:

[0177]

[0178]

[0179] 3) Layer 4: The error terms that need to be propagated in Layer 4 are represented as follows:

[0180]

[0181] According to the chain rule, the update pattern for the connection weights of the 4th layer is calculated as follows:

[0182]

[0183] In the formula, η2 is the learning rate. This leads to the updated connection weights. for:

[0184]

[0185] 4) Layer 2: The propagation error term for Layer 2 is calculated as follows:

[0186]

[0187] The update values ​​of the mean and standard deviation of the second-level membership functions are obtained by the chain rule in the following formulas:

[0188]

[0189]

[0190] Where η3 and η4 are the learning rates. Therefore, the center point and width of the membership function in the second layer are updated according to the following formula:

[0191]

[0192]

[0193] Due to the uncertainty of the storage system, it is impossible to accurately calculate the Jacobian matrix of the storage system. Therefore, to solve this problem and improve the online learning speed of network parameters, the following incremental adaptive law is adopted:

[0194]

[0195] Step 7. Phase-locked loop design:

[0196] A quadrature phase-locked loop (PLL) is used to process and expand the back electromotive force (EMF) observation information to obtain the rotor position angle observation value. The rotor position angle error signal is then normalized to simplify the PLL parameter design. This scheme can quickly track changes in rotor position and speed. Figure 3 As shown.

[0197] According to the formula for extended back electromotive force, the rotor position error is:

[0198]

[0199] When steady state is reached, the error between the actual angle and the estimated angle of the motor is very small. Therefore, the above equation can be equivalent to:

[0200]

[0201] In the formula, E ex To expand the back electromotive force amplitude,

[0202] The formula shows that the rotor position error obtained by the phase detector contains the amplitude of the extended back electromotive force (EMF). When the rotational speed changes, the extended back EMF of the motor also changes accordingly, thus causing the amplitude of the extended back EMF to change as well. However, what is needed when extracting rotor position information is the angle error passed to the loop filter (…). Figure 3 (PI link) and voltage-controlled oscillator ( Figure 3 The integral term (in the middle stage) is used instead of expanding the amplitude of the back EMF. Therefore, it is necessary to eliminate the amplitude of the expanded back EMF, which also eliminates the influence of the rotational speed, thereby improving the accuracy of rotor position extraction.

[0203] The position error after eliminating the influence of rotational speed is:

[0204]

[0205] After eliminating the rotation factor, the phase-locked loop transfer function, which is independent of the motor speed, is obtained as follows:

[0206]

[0207] After the rotor position error signal ε is amplified and integrated, the observed electrical angular velocity of the permanent magnet synchronous motor is obtained. Observations of the electric angular velocity of the motor The rotor position observations are obtained by performing integration. Rotor position observations The feedback signals for the phase-locked loop (PLL) module are repeated sequentially to form a complete PLL position tracking structure.

[0208] The following is an experimental simulation of the control method described above in this invention. See [link / reference]. Figure 5 This is a comparison of the speed observation error under constant load and speed change under traditional algorithms and the simulation algorithm of this invention. Figure 6 This paper compares the rotor position observation errors under constant load and sudden speed changes using the traditional algorithm and the simulation algorithm of this invention. The traditional algorithm uses a full-order sliding mode observer (FSMO) control, while the algorithm of this invention uses a recursive probabilistic wavelet fuzzy neural network (RPWFNN) with a full-order sliding mode observer (FSMO) control. As shown in the figure, under constant load and sudden speed changes, the RPWFNN-FSMO algorithm of this invention exhibits smaller speed observation errors and rotor position observation errors compared to the full-order sliding mode observer (FSMO).

[0209] Figure 7 To compare the speed observation error under constant speed load sudden change with the traditional algorithm and the simulation algorithm of this invention, Figure 8 This paper compares the rotor position observation errors under constant speed and load abrupt changes using the traditional algorithm and the simulation algorithm of this invention. The traditional algorithm uses a full-order sliding mode observer (FSMO) control, while the algorithm of this invention uses a recursive probabilistic wavelet fuzzy neural network (RPWFNN) with a full-order sliding mode observer (FSMO) control. As shown in the figure, under constant speed and load abrupt changes, the RPWFNN-FSMO algorithm of this invention exhibits more stable and smaller speed and rotor position observation errors compared to the full-order sliding mode observer (FSMO).

[0210] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An improved sensorless control method for a permanent magnet synchronous motor, characterized in that, Includes the following steps: Step 1: Establish electrical angular velocity error Design of sliding surface Sliding mode control is used to control the electrical angular velocity in the speed loop. With electromagnetic torque Connect with; Step 2: The electromagnetic torque from Step 1... The input is fed into the MTPA control module, and the Lagrange extremum method is used to find a set of optimal values. and Combination, resulting in electromagnetic torque Maximum stator current Minimize copper loss; Step 3: Establish the full-order state equations of the permanent magnet synchronous motor in a two-phase stationary coordinate system; Step 4: Based on the full-order state equations established in Step 3, construct a full-order sliding mode observer to measure the voltage of the permanent magnet synchronous motor in the two-phase stationary coordinate system. The output stator current is observed by a full-order sliding mode observer. The stator current observation value With stator current The difference is used as the current observation error. ; Step 5: Current observation error in step 4 After the s function module simulated by the recursive probabilistic wavelet fuzzy neural network RPWFNN, the output is the sliding mode control rate in the two-phase stationary coordinate system of the permanent magnet synchronous motor ; Step 6: Adjust the sliding mode control rate from Step 5. The feedback gain matrix G is fed back to the full-order sliding mode observer as its input, and the back electromotive force observation value is output after passing through the full-order sliding mode observer. ; Step 7: Calculate the back electromotive force observation value from Step 6. The back EMF observation is fed back as a state variable to the full-order sliding mode observer for update calculation. After processing by the phase-locked loop module, the observed value of the permanent magnet synchronous motor rotor position is output. Electric angular velocity observations The rotor position observation value The observed electrical angular velocity is fed back to the phase-locked loop module as its input. It is also fed back to the full-order sliding mode observer as its input.

2. The improved sensorless control method for a permanent magnet synchronous motor according to claim 1, characterized in that, In step 1, the speed loop based on sliding mode control is designed as follows: electrical angular velocity error as follows: ; wherein: is the set electrical angular velocity, is the electrical angular velocity; Slip surface As follows: ； wherein: is the first derivative of the electrical angular velocity error, ; The sliding mode control law is as follows: ； wherein: is a normal number, , is the moment of inertia, is the number of pole pairs, , is the first derivative of the electromagnetic torque.

3. The improved sensorless control method for a permanent magnet synchronous motor according to claim 1, characterized in that, In step 2, the MTPA control module is designed as follows: ； in: For permanent magnet synchronous motors in shaft and The equivalent inductance of the shaft, for Shaft stator current, It is a permanent magnet flux linkage. For extreme logarithms, .

4. An improved sensorless control method for a permanent magnet synchronous motor according to claim 1, characterized in that, The process of establishing the full-order state equations of the permanent magnet synchronous motor in the two-phase stationary coordinate system in step 3 is as follows: The model of the permanent magnet synchronous motor in a two-phase stationary coordinate system is as follows: ； in: for Shaft stator voltage, for Shaft stator voltage, for Shaft stator current, for Shaft stator current, This refers to the rotor position of the permanent magnet synchronous motor. for Shaft back electromotive force, for Shaft back electromotive force, This refers to the stator resistance value of the permanent magnet synchronous motor. For permanent magnet synchronous motors in shaft and The equivalent inductance of the shaft, for Shaft stator current, for Shaft stator current, It is a permanent magnet flux linkage. Electric angular velocity; The extended back electromotive force and its rate of change satisfy the following relationship: ； The full-order state equations of the permanent magnet synchronous motor are: ； in: , , , , The stator voltage is in a two-phase stationary coordinate system. It is the back electromotive force. Stator current; Stator current The formula is as follows: ； in: , and This refers to the three-phase stator current of the permanent magnet synchronous motor.

5. An improved sensorless control method for a permanent magnet synchronous motor according to claim 4, characterized in that, The full-order sliding mode observer in step 4 is: ； in: , , , , , It is the sliding mode control rate. It is the feedback gain matrix. yes shaft and Sliding surface of shaft stator current, , It is the switching gain of the sliding mode observer. These are stator current observations. for Shaft stator current observations for Shaft stator current observations The back electromotive force observation value, for Observed back electromotive force of the shaft, for Observed values ​​of shaft electromotive force.

6. An improved sensorless control method for a permanent magnet synchronous motor according to claim 1, characterized in that, The recursive probabilistic wavelet fuzzy neural network (RPWFNN) mentioned in step 5 includes an input layer, a membership layer, a probability layer, a wavelet layer, a rule layer, and an output layer, outputting the sliding mode control law. for: ； in: For the total number of rules, Indicates the output. This represents the l-th output of the rule layer. This represents the adjustable weights between the rule layer and the output layer. This represents the output of the output layer network; The adaptive law design for RPWFNN parameters is as follows: ； ； ； ； in: This indicates that the connection weights are updated at layer 6. This represents the updated connection weights at layer 4. This represents the wavelet weights of the i-th input and k-th wavelet function in the 4th layer. This indicates the update amount of the connection weights in layer 6. This indicates the update pattern of the connection weights in layer 4. , These represent the center point and width after the membership function of the second layer are updated, respectively. and Let $\mathbf{i}$ be the center point and width of the $j$ Gaussian function from the $i$-th input to the membership layer node. and These represent the update amounts of the mean and standard deviation of the membership function at the second level, respectively.

7. An improved sensorless control method for a permanent magnet synchronous motor according to claim 1, characterized in that, The phase-locked loop operation process in step 7 is as follows: According to the formula for extended back electromotive force, the rotor position error is: ； When steady state is reached, the error between the actual angle and the estimated angle of the motor is very small. Therefore, the above equation can be equivalent to: ； In the formula, To expand the back electromotive force amplitude, ; The rotor position error signal After scaling up and integrating, the observed electric angular velocity of the permanent magnet synchronous motor is obtained. Observations of the electric angular velocity of the motor The rotor position observations are obtained by performing integration. , based on rotor position observations The feedback signals for the phase-locked loop (PLL) module are repeated sequentially to form a complete PLL position tracking structure.

Images