A speed prediction method and an inductance-free control system of a permanent magnet synchronous motor

Abstract

The application relates to the technical field of motor control, and discloses a rotating speed prediction method of a permanent magnet synchronous motor and an inductive control system. α β a voltage instruction and a current signal under a stationary coordinate system, and a load torque observation value; a gated recurrent neural network is constructed; the multi-dimensional time sequence characteristics are input into the gated recurrent neural network to obtain a rotating speed prediction result of the permanent magnet synchronous motor; the gated recurrent neural network comprises multiple gated recurrent layers, each gated recurrent layer is composed of multiple gated recurrent units, and each gated recurrent unit comprises an update gate and a reset gate. The application realizes smooth and stable control in a full speed range.​

Description

Technical Field

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

[0002] Permanent magnet synchronous motors (PMSMs) have become a key drive technology in many fields due to their advantages of high power density, high torque density, high efficiency and high reliability. Sensorless control of PMSMs is an indispensable core component of servo and frequency conversion drive systems.

[0003] However, since a single PMSM sensorless control strategy cannot cover the entire speed range, it is impossible to achieve smooth and stable control across the entire speed range, making it difficult to ensure stable operation of the PMSM in scenarios such as startup, low-speed high torque, high-speed field weakening, and continuous operation across the entire speed range. Summary of the Invention

[0004] This invention provides a speed prediction method and a sensorless control system for permanent magnet synchronous motors (PMSMs) to solve the problem that a single sensorless control strategy for PMSMs cannot achieve smooth and stable control across the entire speed range.

[0005] In a first aspect, the present invention provides a method for predicting the speed of a permanent magnet synchronous motor, the method comprising: Multidimensional time-series characteristics of a permanent magnet synchronous motor during stable operation are collected; these multidimensional time-series characteristics include... α - β Voltage commands and current signals in a stationary coordinate system, as well as load torque observations; Construct a gated recurrent neural network; Multidimensional time-series features are input into a gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor. The gated recurrent neural network includes multiple gated recurrent layers, each of which consists of multiple gated recurrent units, and each gated recurrent unit includes an update gate and a reset gate.

[0006] This invention provides a speed prediction method for permanent magnet synchronous motors (PMSMs). It utilizes a gated recurrent neural network (GRNN) to predict the speed of the PMSM. By employing the GRNN to learn the characteristics of different speed ranges, it eliminates the need for logical switching between low-speed and high-speed ranges. This solves the problems of estimation jumps and transient oscillations at speed boundaries in hybrid control strategies, as well as the inability of a single PMSM sensorless control strategy to cover the entire speed range, thus achieving smooth and stable control across the entire speed range.

[0007] In one optional implementation, multidimensional time-series features are input into a gated recurrent neural network to obtain the predicted speed of the permanent magnet synchronous motor, including: Hierarchical feature extraction of multidimensional temporal features is performed using multiple gated recurrent layers to obtain hidden state features; By mapping the hidden state features to a fully connected layer, the speed prediction results of the permanent magnet synchronous motor are obtained.

[0008] This invention provides a method for predicting the rotational speed of a permanent magnet synchronous motor (PMSM). It utilizes a hierarchical feature extraction mechanism to progressively mine the electromagnetic dynamic characteristics of the motor across its entire speed range from coarse to fine granular. It also constructs a data-driven model by combining multi-source input features such as voltage, current, and load torque observations. This method does not rely on a mathematical model of the motor and effectively solves the problems of low-speed observation failure, large high-speed error, and switching instability in traditional sensorless control. Finally, it avoids periodic jumps by outputting sine and cosine electrical angles, thus achieving accurate prediction of the rotor speed and electrical angle across the entire speed range of the PMSM.

[0009] In one optional implementation, multi-layer gated recurrent layers are used to perform hierarchical feature extraction on multi-dimensional temporal features to obtain hidden state features, including: Obtain the hidden state from the previous time step, linearly transform the concatenation matrix corresponding to the hidden state from the previous time step and the multidimensional temporal features, and then use the sigmoid activation function to calculate the values ​​of the update gate and the reset gate respectively. The value of the reset gate is used to selectively forget the hidden state of the previous time step, and then concatenated with multidimensional temporal features. The concatenated matrix is ​​then linearly transformed and the tanh activation function is used to calculate the candidate hidden state. By updating the gate value, the hidden state and candidate hidden states of the previous time step are weighted and fused to obtain the hidden state features of the current gated recurrent layer.

[0010] The present invention provides a speed prediction method for a permanent magnet synchronous motor, which adaptively adjusts the update gate and the reset gate through a gated cycle mechanism, thereby achieving dynamic timing adaptation for different operating conditions such as low speed, high speed, acceleration and deceleration.

[0011] In one optional implementation, before inputting the multidimensional time-series features into the gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor, the method further includes: Historical time-series sample data is obtained, and the gated recurrent neural network is trained using the historical time-series sample data to obtain the trained gated recurrent neural network.

[0012] This invention provides a method for predicting the speed of a permanent magnet synchronous motor (PMSM). It uses historical time-series sample data to train a gated recurrent neural network (RNN). The RNN learns the inherent mapping law between the speed and electrical angle of the PMSM across the entire speed range. The trained RNN is then used to predict the speed of the PMSM, thus improving the accuracy and rationality of the prediction.

[0013] In one optional implementation, the gated recurrent neural network is trained using historical time-series sample data to obtain the trained gated recurrent neural network, including: Data preprocessing is performed on historical time series sample data to obtain preprocessed historical time series sample data; Initialize the network parameters of the gated recurrent neural network; The preprocessed historical time series sample data is input into the gated recurrent neural network after the network parameters are initialized to obtain the speed prediction value. The loss error is calculated based on the predicted speed value and the actual speed value corresponding to the historical time series sample data. The network parameters are updated based on backpropagation of the loss error, and the gated recurrent neural network is iteratively trained based on the updated network parameters until the preset number of iterations is reached, and the trained gated recurrent neural network is output.

[0014] This invention provides a speed prediction method for permanent magnet synchronous motors (PMSMs). Addressing the issues of single sensorless control strategies failing to cover the entire speed range and the complex switching logic of hybrid control strategies, which are prone to estimation jumps and transient oscillations at speed boundaries, this method learns the intrinsic mapping law between speed and electrical angle of the PMSM across the entire speed range from the multidimensional time-series characteristics composed of voltage commands and current signals in the α-β stationary coordinate system, as well as load torque observations. This achieves full-speed-range prediction and sensorless control of the PMSM.

[0015] In one optional implementation, the gated recurrent neural network is trained using historical time-series sample data to obtain the trained gated recurrent neural network, and further includes: Obtain the learning rate for the current iteration of training, adaptively adjust the learning rate for the current iteration of training to obtain the network parameter update step size, and use the network parameter update step size to iteratively train the gated recurrent neural network.

[0016] The present invention provides a method for predicting the rotational speed of a permanent magnet synchronous motor. By limiting the step size of parameter updates using the learning rate, the method avoids the learning results from deteriorating due to excessively large parameter update steps in the later stages of training, thereby improving the generalization ability of the model and ensuring stable convergence of network parameters.

[0017] Secondly, the present invention provides a sensorless control system for a permanent magnet synchronous motor, the system comprising: a prediction module, a sensorless control module, and a permanent magnet synchronous motor; The prediction module is used to execute the speed prediction method of the permanent magnet synchronous motor according to the first aspect or any corresponding embodiment above, and to obtain the speed prediction result of the permanent magnet synchronous motor. The sensorless control module is used to perform sensorless control of the permanent magnet synchronous motor based on the predicted motor speed and electrical angle values ​​from the speed prediction results of the permanent magnet synchronous motor.

[0018] This invention provides a sensorless control system for a permanent magnet synchronous motor. The prediction module achieves accurate prediction of the rotor speed and electrical angle of the permanent magnet synchronous motor across the entire speed domain through a gated recurrent neural network. Then, the sensorless control module performs sensorless control of the permanent magnet synchronous motor based on the predicted motor speed and electrical angle values, replacing the mechanical encoder, reducing system cost, improving reliability, eliminating the need for a mathematical model of the motor, and exhibiting strong resistance to parameter perturbations and load disturbances. It achieves smooth and stable control across the entire speed domain and solves the inherent defects of sensorless control.

[0019] In one alternative implementation, the sensorless control module includes: a speed controller, a current controller, an inverse Park converter, a pulse width modulator, and an inverter. The speed controller, connected to the output of the prediction module, is used to acquire the initial speed command and perform error adjustment based on the initial speed command and the predicted motor speed value. d - q Current command in a rotating coordinate system; The current controller, connected to the output of the speed controller, is used to obtain... d - q Predicting current in a rotating coordinate system, based on d - q Current command in rotating coordinate system and d - q Predicted current calculation in rotating coordinate system d - q Voltage command in a rotating coordinate system; The inverse Park converter is connected to the output of the current controller and the prediction module, respectively, and is used to predict the electrical angle value. d - q The voltage command in the rotating coordinate system is subjected to an inverse Park transformation to obtain... α - β Voltage commands in stationary coordinates will α - β Voltage command input prediction module in stationary coordinate system; A pulse width modulator, connected to the output of the inverse Park converter, is used to... α - β The voltage command in the stationary coordinate system is subjected to space vector pulse width modulation to obtain the switching signal; The inverter is connected to the output of the pulse width modulator and the permanent magnet synchronous motor respectively, and is used to drive and control the operation of the permanent magnet synchronous motor based on the switching signal.

[0020] This invention provides a sensorless control system for a permanent magnet synchronous motor. The sensorless control module constructs a dual-closed-loop vector control architecture through the coordinated operation of a speed controller, a current controller, an inverse Park converter, a pulse width modulator, and an inverter. Based on the speed and electrical angle predicted by a gated recurrent neural network, it realizes sensorless coordinate transformation and closed-loop control, improving voltage utilization and control stability, ensuring stable operation of the permanent magnet synchronous motor across the entire speed range, reducing system hardware costs, and improving environmental adaptability.

[0021] In one optional implementation, the sensorless control module further includes: The Clark converter is used to acquire the three-phase stator current of a permanent magnet synchronous motor, and then perform a Clark converter on the three-phase stator current to obtain... α - β Current signal in a stationary coordinate system.

[0022] This invention provides a sensorless control system for a permanent magnet synchronous motor, in which a Clark converter transforms the stator three-phase current into... α - β The current signal in the stationary coordinate system automatically filters out the zero-sequence component of the stator three-phase current during the transformation process, effectively suppressing the interference caused by grid imbalance and sensor error, improving the purity of the current signal, realizing coordinate dimensionality reduction and decoupling, simplifying control complexity, providing standardized input for gated recurrent neural networks, and improving prediction accuracy.

[0023] In one optional implementation, the sensorless control module further includes: The Park converter is connected to the output of the prediction module, the output of the Clark converter, and the input of the current controller, respectively, and is used to convert the predicted electrical angle value into... α - β The current signal in the stationary coordinate system is subjected to the Park transform to obtain... d - q Predicting current in a rotating coordinate system.

[0024] This invention provides a sensorless control system for a permanent magnet synchronous motor, in which the Park converter, based on the predicted electrical angle, converts... α - β The transformation of the current signal in the stationary coordinate system is as follows: d - q Predicting the current in a rotating coordinate system enables decoupled control of AC to DC quantities, providing feedback for the current closed loop and constructing a complete dual-closed-loop vector control architecture. Attached Figure Description

[0025] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the structure of a sensorless control system for a permanent magnet synchronous motor according to an embodiment of the present invention; Figure 2 This is a schematic flowchart of a first method for predicting the speed of a permanent magnet synchronous motor according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the hierarchical structure of a gated recurrent neural network according to an embodiment of the present invention; Figure 4 This is a second flowchart illustrating a method for predicting the speed of a permanent magnet synchronous motor according to an embodiment of the present invention. Figure 5 This is a schematic diagram of a gated loop unit calculating hidden state features according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the third process of a speed prediction method for a permanent magnet synchronous motor according to an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0029] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0030] Permanent magnet synchronous motors have attracted widespread attention in unmanned underwater vehicles (UUVs), where energy efficiency, space constraints, and operational reliability are extremely demanding. However, UUVs face unique challenges in underwater operations due to high pressure, strong corrosion, limited heat dissipation, and sensitivity to electromagnetic noise.

[0031] Sensorless technology solves the problems of mechanical failure, complexity and reliability caused by position sensors. The main PMSM sensorless control technologies include the back EMF observer method, the high-frequency signal injection method, the artificial intelligence fusion method and the hybrid strategy method.

[0032] Among them, the back EMF observer method is mainly used in the medium and high speed region. It obtains the electric angle by reconstructing the stator back EMF and extracting phase information from it. It mainly includes: sliding mode observer, super-spiral sliding mode observer, model reference adaptive system and extended Kalman filter.

[0033] The high-frequency signal injection method effectively solves the estimation failure problem caused by the small back EMF amplitude in the zero-speed and extremely low-speed regions. By injecting a rotating or pulsating high-frequency voltage signal into the motor, the rotor position information is extracted from the response current by utilizing the salient pole effect or saturation effect of the motor magnetic circuit.

[0034] Artificial intelligence fusion combines neural networks with observers, utilizing the powerful nonlinear mapping capabilities of neural networks to adjust observer parameters or compensate for model uncertainties in real time, thereby achieving better performance under complex operating conditions.

[0035] The emerging approach of hybrid control strategies and integrated intelligent algorithms is a future development trend. Hybrid control strategies achieve full-speed-domain operation by smoothly switching between different control strategies during the startup / low-speed phase and the medium-to-high-speed phase.

[0036] The core objective of sensorless control for PMSMs is to achieve full-speed-range, highly reliable, and robust control under typical operating conditions such as cruising, hovering, or maneuvering of unmanned underwater vehicles. However, the aforementioned sensorless control technology for PMSMs has the following drawbacks: In low-speed and zero-speed conditions, the back EMF observer method suffers from a sharp deterioration in signal-to-noise ratio due to the back EMF amplitude approaching zero, resulting in a severe decrease in observation accuracy or even complete failure. This makes it impossible to fundamentally solve the problems of zero-speed start-up and extremely low-speed operation.

[0037] High-frequency signal injection can effectively achieve zero-speed start-up and low-speed stable operation, but the injected high-frequency signal will introduce additional current harmonics, resulting in a significant increase in electromagnetic vibration and acoustic noise. In the medium and high speed range, the performance will drop sharply due to the attenuation of high-frequency response.

[0038] Artificial intelligence fusion methods typically introduce neural networks combined with observers and other non-sensory control strategies to achieve observer parameter adjustment or model uncertainty compensation, but this leads to a significant increase in algorithm complexity and computational overhead.

[0039] Hybrid strategies combine the advantages of different methods and can theoretically cover the entire speed domain. However, the switching logic is complex, and the switching process may introduce unstable factors. It is prone to prediction jumps and transient oscillations at speed boundaries, making it difficult to balance robustness and smoothness.

[0040] This invention provides a method for predicting the speed of a permanent magnet synchronous motor (PMSM). By utilizing the nonlinear mapping and timing signal learning capabilities of neural networks, sensorless prediction of the speed and electrical angle of the PMSM across the entire speed domain can be achieved.

[0041] As an optional application scenario of this invention, such as Figure 1 As shown, the sensorless control system of the permanent magnet synchronous motor includes: a prediction module 101, a sensorless control module 102, and a permanent magnet synchronous motor 103; the sensorless control module 102 includes: a speed controller 1021, a current controller 1022, an inverse Park converter 1023, a pulse width modulator 1024, an inverter 1025, a Clark converter 1026, and a Park converter 1027; the prediction module 101 can execute a speed prediction method for the permanent magnet synchronous motor.

[0042] According to an embodiment of the present invention, a method for predicting the speed of a permanent magnet synchronous motor is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0043] This embodiment provides a speed prediction method for a permanent magnet synchronous motor, which can be used in the prediction module described above. Figure 2 This is a flowchart of a speed prediction method for a permanent magnet synchronous motor according to an embodiment of the present invention, as shown below. Figure 2 As shown, the process includes the following steps: Step S201: Collect multi-dimensional time-series characteristics of the permanent magnet synchronous motor during stable operation; wherein, the multi-dimensional time-series characteristics include α - β Voltage commands and current signals in a stationary coordinate system, as well as load torque observations.

[0044] Specifically, under stable motor operation conditions, the output of the inverse Park converter is collected. α - β Voltage command in stationary coordinate system and The current signal output by the Park converter and and load torque observations obtained through the load disturbance observer. .

[0045] Furthermore, the multidimensional time-series features are cleaned, data timestamps are aligned, missing data is filled in using interpolation, and outliers and sensor noise are eliminated using methods such as median filtering and moving average filtering.

[0046] Specifically, based on the highest sampling rate, low sampling rate signals are timestamped to ensure that voltage, current, position, and load data at the same moment correspond one-to-one, thus cleaning and aligning multidimensional time-series features. Linear interpolation or spline interpolation is used to complete missing data caused by sensor packet loss and anomalies, ensuring the continuity of the time series. Median filtering is used to eliminate pulse-type outliers, and moving average filtering is used to suppress Gaussian noise from the sensor. Continuous time-series data is divided into fixed-length time window samples to obtain the multidimensional time-series features of the gated recurrent neural network input.

[0047] Step S202: Construct a gated recurrent neural network.

[0048] Specifically, the gated recurrent neural network is a GRU (Gated Recurrent Unit) neural network. The GRU neural network introduces two gating mechanisms, an update gate and a reset gate, to capture the long-short-term dependencies in the input sequence. This can solve the gradient vanishing or gradient explosion problems that recurrent neural networks are prone to when processing long sequences. At the same time, compared with long short-term memory neural networks, the GRU neural network can still achieve similar performance with fewer parameters and higher training efficiency. Therefore, the GRU neural network is used to learn the intrinsic mapping law between the rotational speed and electrical angle of the PMSM in the full speed domain, so as to realize the full speed domain prediction and sensorless control of the PMSM based on the GRU neural network.

[0049] Step S203: Input the multidimensional time-series features into the gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor; wherein, the gated recurrent neural network includes multiple gated recurrent layers, each gated recurrent layer is composed of multiple gated recurrent units, and each gated recurrent unit includes an update gate and a reset gate.

[0050] Specifically, set the length of the sliding window. Then gated recurrent neural network The input features at time step 1 can be represented as: (1) in, It represents multidimensional time series features.

[0051] Furthermore, such as Figure 3 As shown, the gated recurrent neural network uses four gated recurrent layers (GUR layers), with 32, 16, 8, and 4 hidden units in each layer, respectively.

[0052] Furthermore, the above-mentioned multidimensional time series features Feature extraction is performed on the input gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor. The speed prediction result of the permanent magnet synchronous motor includes the predicted motor speed value. Electric angle prediction value , The output of the time-matter neural network is the speed prediction result of the permanent magnet synchronous motor. It can be represented as: (2) in, and It represents the sine and cosine values ​​of the electrical angle.

[0053] This embodiment provides a speed prediction method for a permanent magnet synchronous motor (PMSM). It utilizes a gated recurrent neural network (GRNN) to predict the speed of the PMSM. By employing the GRNN to learn characteristics of different speed ranges, it eliminates the need for logical switching between low-speed and high-speed regions. This solves the problems of estimation jumps and transient oscillations at speed boundaries in hybrid control strategies, as well as the inability of a single PMSM sensorless control strategy to cover the entire speed range. It achieves smooth and stable control across the entire speed range. Furthermore, the use of electrical angle sine and cosine encoding output effectively eliminates boundary jump phenomena related to angle period changes.

[0054] This embodiment provides a speed prediction method for a permanent magnet synchronous motor, which can be used in the prediction module described above. Figure 4 This is a flowchart of a speed prediction method for a permanent magnet synchronous motor according to an embodiment of the present invention, as shown below. Figure 4 As shown, the process includes the following steps: Step S401: Collect multi-dimensional time-series characteristics of the permanent magnet synchronous motor during stable operation; wherein, the multi-dimensional time-series characteristics include α - β Voltage commands and current signals in a stationary coordinate system, as well as observed load torque values. For details, please refer to [link to relevant documentation]. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0055] Step S402: Construct a gated recurrent neural network. See details below. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0056] Step S403: Input the multidimensional time-series features into the gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor; wherein, the gated recurrent neural network includes multiple gated recurrent layers, each gated recurrent layer is composed of multiple gated recurrent units, and each gated recurrent unit includes an update gate and a reset gate.

[0057] Specifically, step S403 includes: Step S4031: Hierarchical feature extraction of multidimensional temporal features is performed using multi-layer gated recurrent layers to obtain hidden state features.

[0058] Specifically, such as Figure 3 As shown, assuming the input to the first gated recurrent layer (32 hidden units) is a 20-dimensional temporal feature, for... The input at each time step is sequentially processed by gated loop units to capture coarse-grained, long-term global electromagnetic dynamics, outputting a coarse-grained hidden state sequence of length 32. The input of the second gated loop layer (16 hidden units) is the hidden state sequence of the first layer, which further extracts medium-grained voltage-current-load coupling features, compresses the feature dimension, and outputs a medium-grained hidden state sequence of length 16. The input of the third gated loop layer (8 hidden units) is the hidden state sequence of the second layer, which extracts fine-grained local features related to rotor position and speed, outputting a fine-grained hidden state sequence of length 8. The input of the fourth gated loop layer (3 hidden units) is the hidden state sequence of the third layer, and finally outputs a final hidden state of length 3.

[0059] In some alternative implementations, such as Figure 5 As shown, step S4031 above includes: Step a1: Obtain the hidden state of the previous time step. After linearly transforming the concatenation matrix corresponding to the hidden state of the previous time step and the multidimensional temporal features, calculate the values ​​of the update gate and the reset gate using the sigmoid activation function.

[0060] Specifically, such as Figure 5 As shown, the previous state is hidden. With current input features ( Vector concatenation, using a weight matrix After linear transformation, it is activated by the sigmoid activation function (a non-linear activation function widely used in neural networks). Output 0 Update gate between 1 The value of is calculated using the following formula: (3) in, The closer to 1, the less historical information is retained and the more new information is introduced; the closer to 0, the more historical information is retained.

[0061] Furthermore, hide the state from the previous moment. With current input features Vector concatenation, using a weight matrix After linear transformation, it is activated by the sigmoid function. Output 0 Reset door between 1 The value of is calculated using the following formula: (4) in, The closer to 0, the more historical information is forgotten; the closer to 1, the more historical information is retained.

[0062] Step a2: Selectively forget the hidden state of the previous time step using the value of the reset gate, and concatenate it with the multidimensional temporal features. After linear transformation, the concatenated matrix is ​​used to calculate the candidate hidden state using the tanh activation function.

[0063] Specifically, first, element-wise multiplication is performed using the reset gate. Hidden state from the previous moment Perform selective forgetting, then compare with the current input. splicing, via weight matrix After linear transformation, candidate hidden states are generated using the tanh activation function (hyperbolic tangent function). , Candidate hidden state The calculation formula is as follows: (5) Among them, candidate hidden state Used to extract valid electrical / load characteristics of the current time step and fuse historical information after reset.

[0064] Step a3: By updating the gate value, the hidden state and candidate hidden states of the previous time step are weighted and fused to obtain the hidden state features of the current gated loop layer.

[0065] Specifically, by updating the door Hidden state in the previous moment and candidate hidden state Perform weighted fusion. Control the weight of historical information. Controlling the weights of new information, and the hidden state features of the current gated recurrent layer. The calculation formula is as follows: (6) Furthermore, the hidden state features of the current gated loop layer mentioned above... It serves as the input for the next time step and is simultaneously passed to the next gated loop layer.

[0066] Step S4032: Map the hidden state features to a fully connected layer to obtain the speed prediction result of the permanent magnet synchronous motor.

[0067] Specifically, the hidden state of the last time step of the fourth gated loop layer is selected and mapped to the speed prediction result of the permanent magnet synchronous motor through a fully connected layer. .

[0068] This embodiment provides a method for predicting the speed of a permanent magnet synchronous motor (PMSM). It utilizes a hierarchical feature extraction mechanism to progressively mine the electromagnetic dynamic characteristics of the motor across its entire speed range, from coarse to fine granular. A GRU gating mechanism is used to adaptively adjust the update and reset gates, enabling dynamic timing adaptation to different operating conditions such as low speed, high speed, and acceleration / deceleration. A data-driven model is constructed by combining multi-source input features such as voltage, current, and load torque observations, without relying on the motor's mathematical model. This effectively solves the problems of low-speed observation failure, large high-speed errors, and switching instability inherent in traditional sensorless control. Finally, by using sine and cosine output electrical angles to avoid periodic jumps, accurate prediction of the PMSM's rotor speed and electrical angle across the entire speed range is achieved.

[0069] This embodiment provides a speed prediction method for a permanent magnet synchronous motor, which can be used in the prediction module described above. Figure 6 This is a flowchart of a speed prediction method for a permanent magnet synchronous motor according to an embodiment of the present invention, as shown below. Figure 6 As shown, the process includes the following steps: Step S601: Collect multi-dimensional time-series characteristics of the permanent magnet synchronous motor during stable operation; wherein, the multi-dimensional time-series characteristics include α - β Voltage commands and current signals in a stationary coordinate system, as well as observed load torque values. For details, please refer to [link to relevant documentation]. Figure 4 Step S401 of the illustrated embodiment will not be described again here.

[0070] Step S602: Construct a gated recurrent neural network. See details below. Figure 4 Step S402 of the illustrated embodiment will not be described again here.

[0071] Step S603: Obtain historical time series sample data, and use the historical time series sample data to train the gated recurrent neural network to obtain the trained gated recurrent neural network.

[0072] Specifically, step S603 includes: Step S6031: Perform data preprocessing on the historical time series sample data to obtain preprocessed historical time series sample data.

[0073] Specifically, historical time-series sample data includes α - β Voltage command in stationary coordinate system and The current signal output by the Park converter and Load torque observations obtained through the load disturbance observer and the actual rotational speed measured by the encoder. With electrical angle signal .

[0074] Furthermore, the historical time-series sample data is cleaned, the data timestamps are aligned, missing data is filled in using interpolation, outliers are eliminated and sensor noise is reduced using methods such as median filtering and moving average filtering, and time-series samples for training (i.e., historical time-series sample data after data preprocessing) are constructed, and the input features are normalized.

[0075] Step S6032: Initialize the network parameters of the gated recurrent neural network.

[0076] Specifically, network parameters include weights, biases, and hidden states.

[0077] Step S6033: Input the preprocessed historical time series sample data into the gated recurrent neural network after network parameter initialization to obtain the speed prediction value.

[0078] Step S6034: Calculate the loss error based on the predicted speed value and the actual speed value corresponding to the historical time series sample data.

[0079] Specifically, the loss error is calculated using a loss function, the expression of which is: (7) (8) in, Indicates the loss error. This represents the actual rotational speed, i.e., the rotational speed measured by the encoder. With electrical angle signal .

[0080] Step S6035: Update the network parameters based on backpropagation of the loss error, and iteratively train the gated recurrent neural network based on the updated network parameters until the preset number of iterations is reached, and output the trained gated recurrent neural network.

[0081] Specifically, the loss error is passed back to the GRU layer from the output layer layer by layer, the gradient of each weight matrix is ​​calculated, and the parameter values ​​are updated in the direction of the negative gradient of the network parameters to reduce the prediction error. The Adam optimizer (Adaptive Moment Estimation, an optimization algorithm) is used to update the network parameters according to the gradient and minimize the loss function.

[0082] Furthermore, the learning rate of the current iteration of training is obtained, and the learning rate of the current iteration of training is adaptively adjusted to obtain the network parameter update step size. The gated recurrent neural network is then iteratively trained using the network parameter update step size.

[0083] Furthermore, the formula for calculating the learning rate is as follows: (9) in, This represents the actual learning rate. This represents the initial learning rate. Indicates the number of iterations.

[0084] Furthermore, the learning rate is used to limit the step size of parameter updates, avoiding excessively large step sizes in the later stages of training that could lead to deterioration of the learning results. A larger initial learning rate is used in the early stages of training to achieve rapid convergence, while the learning rate is gradually reduced as the number of iterations increases in the later stages of training to avoid overfitting, improve the model's generalization ability, and ensure stable convergence of network parameters.

[0085] Furthermore, the prediction accuracy of the gated recurrent neural network was verified using an independent test set.

[0086] Step S604: Input the multi-dimensional time-series features into the gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor; wherein, the gated recurrent neural network includes multiple gated recurrent layers, each gated recurrent layer consists of multiple gated recurrent units, and each gated recurrent unit includes an update gate and a reset gate. For details, please refer to... Figure 4 Step S403 of the illustrated embodiment will not be described again here.

[0087] This embodiment provides a speed prediction method for a permanent magnet synchronous motor (PMSM). Addressing the issues of single sensorless control strategies failing to cover the entire speed domain and the complex switching logic of hybrid control strategies, which are prone to estimation jumps and transient oscillations at speed boundaries, this method learns the intrinsic mapping law between speed and electrical angle of the PMSM across the entire speed domain from multidimensional time-series features composed of voltage commands and current signals in the α-β stationary coordinate system and load torque observations. Then, it utilizes a trained gated recurrent neural network for PMSM speed prediction, improving the accuracy and rationality of the prediction and achieving full-speed domain prediction and sensorless control of the PMSM.

[0088] This embodiment provides a sensorless control system for a permanent magnet synchronous motor, such as... Figure 1 As shown, it includes: a prediction module 101, a sensorless control module 102, and a permanent magnet synchronous motor 103; The prediction module 101 is used to execute the speed prediction method of the permanent magnet synchronous motor 103 and obtain the speed prediction result of the permanent magnet synchronous motor 103.

[0089] Specifically, the prediction module 101 predicts based on the output of the inverse Park transformer. α - β Voltage command in stationary coordinate system and The current signal output by the Park converter and and load torque observations obtained through the load disturbance observer. Predict the rotational speed of the PMSM rotor With electrical angle sine and cosine values and .

[0090] The sensorless control module 102 is used to perform sensorless control on the permanent magnet synchronous motor 103 based on the predicted motor speed and electrical angle values ​​in the speed prediction results of the permanent magnet synchronous motor 103.

[0091] Specifically, sensorless control of the PMSM based on the GRU neural network is achieved by using the predicted values ​​of motor speed and electrical angle.

[0092] This embodiment provides a sensorless control system for a permanent magnet synchronous motor. The prediction module achieves accurate prediction of the rotor speed and electrical angle of the permanent magnet synchronous motor across the entire speed domain through a gated recurrent neural network. Then, the sensorless control module performs sensorless control of the permanent magnet synchronous motor based on the predicted motor speed and electrical angle values, replacing the mechanical encoder, reducing system cost, improving reliability, eliminating the need for a mathematical model of the motor, and exhibiting strong resistance to parameter perturbations and load disturbances. It achieves smooth and stable control across the entire speed domain, solving the inherent defects of sensorless control.

[0093] In some alternative implementations, the sensorless control module 102 includes: a speed controller 1021, a current controller 1022, an inverse Park converter 1023, a pulse width modulator 1024, and an inverter 1025. The speed controller 1021, connected to the output of the prediction module 101, is used to acquire the initial speed command, and perform error adjustment based on the initial speed command and the predicted motor speed value to obtain... d - q Current command in rotating coordinate system.

[0094] Specifically, the speed controller 1021 responds to the speed command. With predicted speed Calculated d - q Current command in rotating coordinate system and .

[0095] The current controller 1022 is connected to the output terminal of the speed controller 1021 and is used to obtain... d - q Predicting current in a rotating coordinate system, based on d - q Current command in rotating coordinate system and d - q Predicted current calculation in rotating coordinate system d - q Voltage command in a rotating coordinate system.

[0096] Specifically, the current controller 1022 responds to the current command. and With predicted current and Calculated d - q Voltage command in rotating coordinate system and .

[0097] The inverse Park converter 1023 is connected to the output of the current controller 1022 and the prediction module 101, respectively, and is used to adjust the current based on the predicted electrical angle value. d - q The voltage command in the rotating coordinate system undergoes an inverse Park transformation (a transformation that converts the current or voltage vector in the rotating coordinate system back to the stationary coordinate system) to obtain... α - β Voltage commands in stationary coordinates will α - β Voltage command input prediction module 101 in stationary coordinate system.

[0098] Specifically, the inverse Park converter 1023 is based on the sine and cosine values ​​of the electrical angle. and ,Will and Inverse Park transform, we get α - β Voltage command in stationary coordinate system and .

[0099] Pulse width modulator 1024, connected to the output of inverse Park converter 1023, is used for...α - β The voltage command in the stationary coordinate system is subjected to space vector pulse width modulation to obtain the switching signal.

[0100] Specifically, the pulse width modulator 1024 uses SVPWM (Space Vector Pulse Width Modulation) to... α - β Voltage command in stationary coordinate system and Modulation is performed to obtain a switching signal. .

[0101] Inverter 1025 is connected to the output of pulse width modulator 1024 and permanent magnet synchronous motor 103 respectively, and is used to drive and control permanent magnet synchronous motor 103 to run based on switching signals.

[0102] Specifically, the inverter 1025 modulates the switching signal obtained by the pulse width modulator 1024. and DC bus voltage Generate the actual voltage, i.e., the three-phase AC voltage, and use the three-phase AC voltage to drive the PMSM to operate.

[0103] This embodiment provides a sensorless control system for a permanent magnet synchronous motor. The sensorless control module constructs a dual-closed-loop vector control architecture through the coordinated operation of a speed controller, a current controller, an inverse Park converter, a pulse width modulator, and an inverter. Based on the speed and electrical angle predicted by a gated recurrent neural network, it realizes sensorless coordinate transformation and closed-loop control, improving voltage utilization and control stability, ensuring stable operation of the permanent magnet synchronous motor across the entire speed range, reducing system hardware costs, and improving environmental adaptability.

[0104] In some alternative implementations, the sensorless control module 102 further includes: Clark converter 1026 is used to acquire the stator three-phase current of permanent magnet synchronous motor 103, and perform Clark transformation (a coordinate transformation method to convert three-phase electrical signals into two-phase electrical signals) on the stator three-phase current to obtain... α - β Current signal in a stationary coordinate system.

[0105] Specifically, the Clark converter 1026 converts the PMSM three-phase current acquired by the current sensor into... α - β Current signal in stationary coordinate system and .

[0106] This embodiment provides a sensorless control system for a permanent magnet synchronous motor, in which a Clark converter transforms the stator three-phase current into... α - β The current signal in the stationary coordinate system automatically filters out the zero-sequence component of the stator three-phase current during the transformation process, effectively suppressing the interference caused by grid imbalance and sensor error, improving the purity of the current signal, realizing coordinate dimensionality reduction and decoupling, simplifying control complexity, providing standardized input for gated recurrent neural networks, and improving prediction accuracy.

[0107] In some alternative implementations, the sensorless control module 102 further includes: Park converter 1027 is connected to the output of prediction module 101, the output of Clark converter 1026, and the input of current controller 1022, respectively, and is used to convert the predicted electrical angle value into the current. α - β The current signal in the stationary coordinate system is subjected to the Park transform (a transformation that converts the current or voltage vector in the stationary coordinate system to a coordinate system that rotates synchronously with the rotor magnetic field) to obtain... d - q Predicting current in a rotating coordinate system.

[0108] Specifically, the Park converter 1027 is based on the sine and cosine values ​​of the electrical angle. and ,Will α - β Current signal in stationary coordinate system and Perform the Park transformation to obtain d - q Predicting current in a rotating coordinate system and .

[0109] This embodiment provides a sensorless control system for a permanent magnet synchronous motor. The Park converter, based on the predicted electrical angle value, will... α - β The transformation of the current signal in the stationary coordinate system is as follows: d - q Predicting the current in a rotating coordinate system enables decoupled control of AC to DC quantities, providing feedback for the current closed loop and constructing a complete dual-closed-loop vector control architecture.

[0110] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the embodiments of this application.

[0111] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0112] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0115] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of this application, essentially, or the parts that contribute to the prior art, or parts of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for predicting the speed of a permanent magnet synchronous motor, characterized in that, The method includes: Collect multi-dimensional timing characteristics of a permanent magnet synchronous motor during stable operation; wherein, the multi-dimensional timing characteristics include α - β Voltage commands and current signals in a stationary coordinate system, as well as load torque observations; Construct a gated recurrent neural network; The multidimensional time-series features are input into the gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor; wherein, the gated recurrent neural network includes multiple gated recurrent layers, each gated recurrent layer is composed of multiple gated recurrent units, and each gated recurrent unit includes an update gate and a reset gate.

2. The method according to claim 1, characterized in that, The step of inputting the multidimensional time-series features into the gated recurrent neural network to obtain the predicted speed of the permanent magnet synchronous motor includes: Hierarchical feature extraction is performed on the multidimensional temporal features using multiple gated recurrent layers to obtain hidden state features; The hidden state features are mapped using a fully connected layer to obtain the speed prediction result of the permanent magnet synchronous motor.

3. The method according to claim 2, characterized in that, The process of extracting hidden state features by performing hierarchical feature extraction on the multidimensional temporal features using multiple gated recurrent layers includes: Obtain the hidden state of the previous time step, and after linearly transforming the concatenation matrix corresponding to the hidden state of the previous time step and the multidimensional temporal features, use the sigmoid activation function to calculate the values ​​of the update gate and the reset gate respectively; The value of the reset gate is used to selectively forget the hidden state of the previous time step, and then concatenated with the multidimensional temporal features. The concatenated matrix is ​​then linearly transformed and the tanh activation function is used to calculate the candidate hidden state. The hidden state features of the current gated loop layer are obtained by weighted fusion of the previous hidden state and the candidate hidden state using the value of the updated gate.

4. The method according to claim 1, characterized in that, Before inputting the multidimensional time-series features into the gated recurrent neural network to obtain the speed prediction result of the permanent magnet synchronous motor, the method further includes: Historical time-series sample data is acquired, and the gated recurrent neural network is trained using the historical time-series sample data to obtain the trained gated recurrent neural network.

5. The method according to claim 4, characterized in that, The step of training the gated recurrent neural network using the historical time-series sample data to obtain the trained gated recurrent neural network includes: The historical time series sample data is preprocessed to obtain preprocessed historical time series sample data. Initialize the network parameters of the gated recurrent neural network; The preprocessed historical time series sample data is input into the gated recurrent neural network after the network parameters are initialized to obtain the speed prediction value. Based on the predicted speed value and the actual speed value corresponding to the historical time series sample data, the loss error is calculated. The network parameters are updated based on the backpropagation of the loss error, and the gated recurrent neural network is iteratively trained based on the updated network parameters until a preset number of iterations is reached, and the trained gated recurrent neural network is output.

6. The method according to claim 5, characterized in that, The step of training the gated recurrent neural network using the historical time-series sample data to obtain the trained gated recurrent neural network further includes: Obtain the learning rate of the current iteration of training, adaptively adjust the learning rate of the current iteration of training to obtain the network parameter update step size, and use the network parameter update step size to iteratively train the gated recurrent neural network.

7. A sensorless control system for a permanent magnet synchronous motor, characterized in that, The system includes: a prediction module, a sensorless control module, and a permanent magnet synchronous motor; The prediction module is used to execute the speed prediction method of the permanent magnet synchronous motor according to any one of claims 1 to 6, and obtain the speed prediction result of the permanent magnet synchronous motor. The sensorless control module is used to perform sensorless control on the permanent magnet synchronous motor based on the predicted motor speed and electrical angle values ​​in the speed prediction results of the permanent magnet synchronous motor.

8. The system according to claim 7, characterized in that, The sensorless control module includes: a speed controller, a current controller, an inverse Park converter, a pulse width modulator, and an inverter; The speed controller is connected to the output of the prediction module and is used to acquire an initial speed command, perform error adjustment based on the initial speed command and the predicted motor speed value, and obtain... d - q Current command in a rotating coordinate system; The current controller is connected to the output terminal of the speed controller and is used to obtain... d - q Predicting current in a rotating coordinate system, based on the above d - q Current command in rotating coordinate system and the d - q Predicted current calculation in rotating coordinate system d - q Voltage command in a rotating coordinate system; The inverse Park converter is connected to the output of the current controller and the prediction module, respectively, and is used to adjust the current based on the predicted electrical angle value. d - q The voltage command in the rotating coordinate system is subjected to an inverse Park transformation to obtain... α - β The voltage command in the stationary coordinate system will... α - β The voltage command in the stationary coordinate system is input into the prediction module; The pulse width modulator is connected to the output of the inverse Park converter and is used to control the pulse width modulator. α - β The voltage command in the stationary coordinate system is subjected to space vector pulse width modulation to obtain the switching signal; The inverter is connected to the output of the pulse width modulator and the permanent magnet synchronous motor respectively, and is used to drive and control the operation of the permanent magnet synchronous motor based on the switching signal.

9. The system according to claim 8, characterized in that, The contactless control module also includes: The Clark converter is used to acquire the three-phase stator current of the permanent magnet synchronous motor, and to perform Clark conversion on the three-phase stator current to obtain... α - β Current signal in a stationary coordinate system.

10. The system according to claim 9, characterized in that, The contactless control module also includes: The Park converter is connected to the output of the prediction module, the output of the Clark converter, and the input of the current controller, respectively, and is used to convert the electrical angle prediction value into the current. α - β The current signal in the stationary coordinate system is subjected to Park transform to obtain the... d - q Predicting current in a rotating coordinate system.

Images