A method and system for stable operation of compressors in the field weakening region based on angle compensation

Abstract

This application discloses a method and system for stable operation of a compressor in the field of magnetic field weakening region based on angle compensation, relating to the field of compressor magnetic field weakening region control technology. The method includes: estimating the rotor angle using a phase-locked loop and generating an estimated angle; processing the operating parameters of the target compressor, the estimated rotor angle, and electromagnetic data generated by finite element analysis using a machine learning model to generate a predicted angle; and adjusting the compressor's operating parameters based on the predicted angle. The machine learning model compensates for the influence of multiple frequency components on the angle prediction by combining spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals. Furthermore, the electromagnetic data generated by finite element analysis further enhances the machine learning model's adaptability to complex electromagnetic environments. This invention can dynamically adjust the compressor's operating parameters under complex operating conditions, improving operational stability and energy efficiency.

Description

Technical Field

[0001] This application relates to the field of compressor magnetic field weakening zone control technology, and in particular to a method and system for stable operation of compressor magnetic field weakening zone based on angle compensation. Background Technology

[0002] In modern compressor motor control systems, accurate rotor angle estimation is crucial for stable compressor operation. Especially during operation in the weak magnetic field region, the weakened rotor magnetic field can cause unstable motor torque output, leading to reduced compressor efficiency. This typically occurs when motor speed is high and voltage is close to the rated voltage limit. Traditional control systems rely on phase-locked loops (PLLs) to estimate the rotor angle and adjust based on phase errors to maintain stable motor operation. However, under complex operating conditions, especially with multiple frequency interference signals such as space harmonics, high-frequency components, and noise, traditional phase error detection methods may struggle to cope, resulting in inaccurate rotor angle estimation and consequently affecting the overall system control performance.

[0003] For example, Chinese patent CN114738977B discloses a method, device, and storage medium for controlling the magnetic field weakening angle of an air conditioner. The method includes: in response to the air conditioner entering a magnetic field weakening control mode, acquiring the operating frequency of the air conditioner compressor and monitoring the outdoor ambient temperature during the compressor's operation; and adjusting the magnetic field weakening angle of the air conditioner compressor based on the operating frequency and the outdoor ambient temperature. This disclosure allows for precise control of the magnetic field weakening by differentiating between different outdoor ambient temperature ranges and the difference in operating frequency between the air conditioner compressor and the outdoor ambient temperature range.

[0004] This technical solution is limited to coarse adjustments based on the compressor's operating conditions and does not fully consider the influence of various frequency components in rotor angle estimation, resulting in limited control accuracy. Furthermore, existing control systems typically lack intelligent learning capabilities, making it difficult to adapt to different operating states and complex environmental conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application discloses a method and system for stable operation of a compressor in the weak magnetic region based on angle compensation.

[0006] Firstly, this application provides a method for stable operation of a compressor in the field weakening region based on angle compensation, including:

[0007] The rotor angle of the target compressor is estimated based on the phase-locked loop, and the estimated angle of the rotor is generated.

[0008] Using a pre-defined machine learning model, a predicted angle is generated based on the operating parameters of the target compressor and the estimated angle of the rotor; wherein, the machine learning model is used to compensate for the influence of multiple frequency components on the predicted angle; the multiple frequency components include: spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals;

[0009] Based on the predicted angle, the operating parameters of the target compressor are adjusted.

[0010] As an optional implementation, generating the estimated angle of the rotor further includes:

[0011] A high-frequency voltage signal is injected into the stator winding of the target compressor motor;

[0012] Acquire the current response signal of the stator winding and extract the spatial harmonic components from the current response signal;

[0013] The spatial harmonic components and the back electromotive force signal are weighted and combined, and then fed to the phase-locked loop for phase error detection to generate the estimated angle.

[0014] As an optional implementation, it also includes: modeling the electromagnetic characteristics of the target compressor using the finite element analysis method to generate electromagnetic data of the target compressor under the target operating state;

[0015] Using a pre-defined machine learning model, the predicted angle is generated based on the operating parameters of the target compressor, the estimated angle of the rotor, and the electromagnetic data generated through finite element analysis.

[0016] As an optional implementation method, the finite element analysis method employs a pre-calculation approach to pre-calculate the electromagnetic characteristics of the target compressor motor, generating electromagnetic data under different operating conditions;

[0017] The electromagnetic data includes: magnetic flux density distribution, current density distribution, magnetic flux coupling, and electromagnetic loss characteristics.

[0018] The pre-calculated electromagnetic data is stored through a lookup table or database and dynamically retrieved during real-time operation based on the compressor's current operating status.

[0019] As an optional implementation, it also includes:

[0020] Based on the electromagnetic data pre-calculated using finite element analysis, the frequency and amplitude of the high-frequency voltage signal are determined.

[0021] When the magnetic flux density is lower than the first threshold, a high-frequency voltage signal with a frequency within the first preset range is selected for injection;

[0022] When the current density is higher than the second threshold, the amplitude of the high-frequency voltage signal is adjusted to the second preset range.

[0023] As an optional implementation, the input data of the machine learning model includes: the target compressor's rotational speed, load, stator current, back electromotive force signal, estimated rotor angle, and electromagnetic data pre-calculated through finite element analysis;

[0024] The machine learning model includes: a long short-term memory network model;

[0025] The long short-term memory network model is used to process the input data in real time and generate predicted angles;

[0026] The Long Short-Term Memory (LSTM) network model generates a prediction angle by analyzing historical data and current operating parameters.

[0027] As an optional implementation, the machine learning model is trained using a truncated backpropagation algorithm, including:

[0028] The historical operating parameters of the target compressor are segmented based on a preset length;

[0029] The preset length is determined based on the system's calculated load and real-time performance targets; the historical operating parameters include: time-series data of rotational speed, load, stator current, and back electromotive force signal;

[0030] Local gradient calculations are performed on the time-series data of the historical operating parameters of each compressor segment, and the weights of the machine learning model are updated based on the results of the local gradient calculations.

[0031] The local gradient calculation limits the number of time steps for each backpropagation by truncating backpropagation.

[0032] After the calculation of each segment of time series data is completed, the hidden state and memory state of the current segment of time series data are used as the initial state for the calculation of the next segment of time series data.

[0033] As an optional implementation, the preset length is determined based on the system's computational load and real-time performance targets, including:

[0034] The system computing load of the target compressor is monitored; wherein the system computing load is determined based on a weighted average of the processor utilization and memory occupancy.

[0035] When the weighted average of the system's calculated load is greater than or equal to a preset first load threshold, the preset length is shortened.

[0036] When the weighted average of the system's calculated load is less than the first load threshold, the preset length is extended.

[0037] Secondly, this application also provides a compressor magnetic field weakening region stable operation system based on angle compensation, including:

[0038] The estimation unit estimates the rotor angle of the target compressor based on the phase-locked loop and generates the estimated rotor angle.

[0039] The prediction unit uses a preset machine learning model to generate a predicted angle based on the operating parameters of the target compressor and the estimated angle of the rotor; wherein, the machine learning model is used to compensate for the influence of multiple frequency components on the predicted angle; the multiple frequency components include: spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals;

[0040] The control unit adjusts the operating parameters of the target compressor based on the predicted angle.

[0041] Compared with existing technologies, the advantages of this invention are: It utilizes a machine learning model to process multiple input parameters in real time, including rotational speed, load, stator current, back EMF signal, and electromagnetic data, effectively compensating for the effects of spatial harmonics, high-frequency signals, and noise, generating a more accurate rotor prediction angle. Through comprehensive analysis of multi-source data, it can dynamically adapt to complex electromagnetic environments and operating condition changes, especially improving the stable operation performance of the compressor in weak magnetic fields. Based on the predicted angle, it adjusts the compressor's rotational speed, load, and current in real time, ensuring the system operates optimally under different load conditions, further improving energy efficiency and operational reliability. The pre-calculation method of finite element analysis and the truncated backpropagation algorithm during machine learning training reduce the computational burden in real-time operation, enabling the system to maintain high-precision prediction while ensuring real-time performance. Attached Figure Description

[0042] Figure 1 A flowchart illustrating a method for stable operation of a compressor in the field weakening region based on angle compensation, provided in an embodiment of this application.

[0043] Figure 2 A flowchart for generating the estimated angle of the rotor provided in an embodiment of this application;

[0044] Figure 3 A schematic diagram of a compressor magnetic field weakening region stable operation system based on angle compensation provided in an embodiment of this application. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0046] This invention relates to a method and system for stable operation control of compressors in weak magnetic regions, particularly achieving precise control through rotor angle compensation. When a compressor operates in a weak magnetic region, errors in the rotor angle can lead to operational instability, thereby affecting system efficiency and lifespan. To address this issue, this invention proposes a rotor angle prediction method based on a machine learning model and compensation for multiple frequency components.

[0047] The applicable target compressor of this invention can use a permanent magnet synchronous motor (PMSM) as a power source, whose rotor rotates through electromagnetic induction. Compressors typically operate at high speeds and high loads, under which conditions, it is necessary to accurately estimate the actual rotor angle to achieve precise torque control.

[0048] The stator windings of the compressor motor are connected to the control system, and the dynamic operating status of the motor can be obtained by collecting stator current and voltage signals. Specifically, during motor operation, the back electromotive force (Back-EMF) signal on the motor stator windings is detected to provide a real-time estimate of the rotor position.

[0049] Angle compensation refers to the process of adjusting the relative angle (i.e., phase difference) between the rotor and stator magnetic fields during motor control to compensate for insufficient magnetic flux or torque caused by various factors (such as high-speed motor operation or load changes), thereby ensuring that the motor can still maintain stable operation in specific areas (such as weak magnetic fields).

[0050] The field weakening region refers to the area where a motor enters high-speed operation. As the speed increases, the stator voltage approaches the upper limit of the motor's rated voltage, making it impossible to continue increasing the voltage to maintain the magnetic flux, resulting in a weakened magnetic field. At this point, the motor's magnetic flux is insufficient, thus affecting the output torque, and the motor enters the so-called field weakening region.

[0051] See Figure 1 The diagram shows a flowchart of a method for stable operation of a compressor in the field weakening region based on angle compensation, provided in an embodiment of this application. The method includes steps S101 to S103, wherein:

[0052] S101: Estimate the rotor angle of the target compressor based on the phase-locked loop and generate the estimated rotor angle.

[0053] S102: Using a preset machine learning model, a predicted angle is generated based on the operating parameters of the target compressor and the estimated angle of the rotor; wherein, the machine learning model is used to compensate for the influence of multiple frequency components on the predicted angle; the multiple frequency components include: spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals.

[0054] S103: Based on the predicted angle, adjust the operating parameters of the target compressor.

[0055] A phase-locked loop (PLL) is a feedback control system capable of detecting and adjusting signal phase errors, which is used in this invention to estimate the rotor angle of the target compressor.

[0056] In practice, real-time operating signals can be acquired using stator sensors on the compressor motor. The phase-locked loop (PLL) system compares the input signal with a reference signal to calculate the phase error. Subsequently, the system adjusts in real time based on the phase error to generate an estimated angle corresponding to the actual operating angle of the rotor.

[0057] To further improve the operating stability of the compressor in the weak magnetic field region, this invention utilizes a pre-set machine learning model to compensate for and predict the rotor angle. The input parameters of the machine learning model include the compressor's operating parameters (such as speed, stator current, and back electromotive force signal) and the estimated rotor angle. By processing these input data in real time, the machine learning model generates a predicted angle value to compensate for angle errors caused by multiple frequency components (spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals).

[0058] Based on the predicted angle, the system adjusts the compressor's operating parameters in real time to ensure stable operation of the compressor in the weak magnetic field region. For example, when the predicted angle deviation is large, the system can ensure the normal operation of the motor by adjusting the magnitude of the stator current or the frequency of the voltage.

[0059] It should be noted that the phase-locked loop in this disclosure can also correspond to a phase-locked loop system or a phase-locked loop module.

[0060] As an optional implementation, the phase-locked loop in this disclosure includes: a phase detector, a low-pass filter, a voltage-controlled oscillator, and a feedback loop.

[0061] The system includes: a phase detector (PD) for comparing the phase difference between the reference signal and the feedback signal and generating a phase error signal; a low-pass filter (LPF) for filtering the phase error signal, removing high-frequency noise, and obtaining a smooth error signal; a voltage-controlled oscillator (VCO) for controlling its output frequency through the error signal, gradually locking the phase of the output signal to the phase of the back electromotive force signal; and a feedback loop for feeding the VCO output signal back to the phase detector, forming a closed-loop control.

[0062] In practical implementation, when the compressor starts, the stator windings are energized, and the motor rotor gradually enters a stable rotational state. At this time, the back electromotive force (EMF) signal generated by the stator windings is collected and transmitted as an input signal to the phase detector of the phase-locked loop (PLL). The phase detector simultaneously receives a feedback signal from the VCO, calculates the phase difference between the feedback signal and the back EMF signal, and outputs a corresponding phase error signal. This error signal is processed by a low-pass filter to remove high-frequency interference, and the output smooth error signal controls the voltage-controlled oscillator (VCO) to gradually approximate the frequency and phase of the back EMF signal. When the PLL enters the locked state, the VCO output signal is synchronized with the back EMF signal, and the output phase of the VCO at this time is the real-time angle of the motor rotor.

[0063] After the phase-locked loop (PLL) is locked, the phase signal output by the VCO can be sent to the control system or central processing unit deployed within the compressor as an estimate of the rotor angle. The control system can then perform further torque or speed control based on this estimated angle, thereby ensuring that the compressor maintains stable torque output even when operating at high speeds.

[0064] To further improve the accuracy of rotor angle estimation, this invention employs a method of injecting a high-frequency voltage signal into the stator winding of the target compressor motor. By acquiring the current response signal of the stator winding and extracting the spatial harmonic components, the accuracy of rotor angle estimation in the weak magnetic field region can be effectively improved.

[0065] See Figure 2 The diagram shown is a flowchart for generating the estimated angle of the rotor according to an embodiment of this application. As an optional implementation, generating the estimated angle of the rotor further includes steps S201 to S203, wherein:

[0066] S201: Inject a high-frequency voltage signal into the stator winding of the target compressor motor;

[0067] S202: Acquire the current response signal of the stator winding and extract the spatial harmonic components in the current response signal;

[0068] S203: The spatial harmonic components and the back electromotive force signal are weighted and combined, and provided to the phase-locked loop for phase error detection to generate the estimated angle.

[0069] The purpose of injecting a high-frequency voltage signal is to obtain a richer current response signal, especially including harmonic component information related to the rotor angle.

[0070] In this invention, the high-frequency voltage signal refers to a voltage signal with a frequency higher than the operating frequency of the motor. In a preferred embodiment, the frequency of the high-frequency voltage is in the range of 1kHz to 10kHz. This signal is injected through the stator winding of the compressor motor to excite the motor to generate a current response signal in the high-frequency range, thereby obtaining the harmonic components related to the rotor angle. The frequency and amplitude of this high-frequency voltage signal are dynamically adjusted according to the motor's operating state and electromagnetic characteristics.

[0071] The current response signal is the current feedback signal acquired by a current sensor after a high-frequency voltage signal is injected into the stator windings of a motor. This signal reflects the transient electromagnetic state of the motor, including harmonic components related to the rotor position. The acquisition frequency of the current response signal should be at least twice the frequency of the high-frequency voltage signal to ensure effective capture of the high-frequency components. Further processing of this signal can extract spatial harmonic components, which can be used to generate estimates of the rotor angle.

[0072] For example, the control system acquires the current response signal of the stator winding using a current sensor at a sampling rate of at least 20 kHz. This signal is processed by a digital bandpass filter, the passband of which is set to correspond to the frequency of the injected high-frequency voltage signal. The processed signal is then subjected to spectral analysis using a Fast Fourier Transform (FFT) to extract the main spatial harmonic components for subsequent angle estimation.

[0073] Spatial harmonic components (SHPCs) refer to the high-frequency components in a motor's current response signal that are related to the rotor angle. These are typically caused by the motor's structural characteristics and the electromagnetic interaction between the stator and rotor. In this invention, SHPCs are obtained through spectral analysis and filtering of the current response signal. SHPCs can reflect changes in the rotor angle, and by weighting them with the back electromotive force (EMF) signal, the accuracy of rotor angle estimation is further improved.

[0074] In practice, the compressor motor's control system injects a high-frequency voltage signal with a preset frequency and amplitude into the stator windings. After the signal is injected, the stator windings generate a current response signal, which contains high-frequency harmonic components related to the rotor angle. Following the injection of the high-frequency voltage signal, the system acquires the stator winding's current response signal using a current sensor. To ensure the accuracy of signal acquisition, the current sensor's sampling rate is at least 20kHz to capture the high-frequency components. The acquired current response signal is then filtered to extract the spatial harmonic components.

[0075] In practical implementation, the control system of the target compressor can be equipped with an embedded or external signal generator to generate a high-frequency voltage signal with a preset frequency and amplitude. The output signal of this signal generator is applied to the stator winding of the motor through the power inverter of the control system. The control system can modulate the output voltage signal by controlling the power inverter, thereby generating a high-frequency voltage signal. Typically, the switching frequency of the inverter can be adjusted to meet the required frequency range, such as 1kHz to 10kHz. Furthermore, the frequency and amplitude of the high-frequency voltage signal can be dynamically adjusted through a closed-loop control algorithm. The control system can control the frequency and amplitude based on the real-time monitored motor operating status (such as voltage, current, speed, and load). For example, a frequency suitable for the current operating status can be dynamically selected through a lookup table, and the signal amplitude can be automatically adjusted according to the magnetic flux density or current density.

[0076] For example, the compressor's control system incorporates a digital signal processor (DSP) that generates a high-frequency voltage signal at a preset frequency. This high-frequency voltage signal, ranging from 1 kHz to 10 kHz, is injected into the stator windings by pulse-width modulation (PWM) control of the inverter's switching state. The signal amplitude is then dynamically adjusted based on the motor's real-time load and flux density.

[0077] Furthermore, the extracted spatial harmonic components are weighted and combined with the back electromotive force (EMF) signal to further improve the accuracy of rotor angle estimation. The back EMF signal originates from the self-induced voltage of the motor stator windings during operation and contains information related to rotor motion. The system weights the spatial harmonic components and the back EMF signal, with the weighting coefficients dynamically adjusted according to the operating conditions. The combined signal is then provided to the phase-locked loop (PLL) system for phase error detection.

[0078] The back EMF signal can be obtained through the same acquisition circuit as the spatial harmonic component, and the high-frequency component can be extracted after filtering. The weighting coefficients can be dynamically adjusted according to the current operating state. For example, if the motor load is large, the weight of the back EMF signal can be increased to reflect the impact of rotor position changes. The weighting coefficients can be obtained through a preset lookup table or real-time calculation.

[0079] In this way, the extracted spatial harmonic components and back electromotive force signals are linearly weighted and combined using weighting coefficients. The combined signal will simultaneously contain angular information from both the harmonic components and the back electromotive force, reducing errors caused by a single signal.

[0080] The combined signal is input to the phase-locked loop (PLL) system for real-time phase error detection. The PLL adjusts the signal phase to minimize the phase error between the input and reference signals, thereby generating an estimated rotor angle. This estimated angle reflects the actual rotor position of the motor in the field-weakening region and can be used for further angle compensation and control.

[0081] For example, after the combined signal is input into the phase-locked loop (PLL), the phase detector calculates the phase difference between this signal and the expected rotor angle reference signal. The phase error signal is smoothed by a loop filter and then transmitted to the voltage-controlled oscillator (VCO). The VCO adjusts the output angle in real time based on the phase error signal to ensure that the estimated angle matches the actual rotor position. Finally, an accurate rotor estimated angle is generated for subsequent control steps.

[0082] In this way, by injecting a high-frequency voltage signal into the stator winding, the spatial harmonic components related to the rotor angle are obtained, and then weighted and combined with the back electromotive force signal, higher precision rotor angle estimation is achieved. Dynamically adjusting the frequency and amplitude of the high-frequency voltage signal allows it to adapt to different motor operating states, improving the system's adaptability and robustness. Phase error detection using a phase-locked loop ensures real-time accurate angle estimation, particularly improving the compressor's operational stability and control precision in the weak magnetic field region.

[0083] As an optional implementation, the method further includes: modeling the electromagnetic characteristics of the target compressor using the finite element analysis method to generate electromagnetic data of the target compressor under the target operating state;

[0084] Using a pre-defined machine learning model, the predicted angle is generated based on the operating parameters of the target compressor, the estimated angle of the rotor, and the electromagnetic data generated through finite element analysis.

[0085] Finite element analysis (FEA) is a numerical computational tool used to solve complex physical problems. In the field of motors, it is often used to analyze electromagnetic field distribution and electromagnetic performance. By inputting the structure, materials, geometry, and operating conditions of a compressor motor into a finite element analysis model, electromagnetic data of the motor under different operating conditions can be generated.

[0086] The electromagnetic data mentioned in this disclosure refers to electromagnetic performance parameters obtained through finite element analysis that describe the motor under specific operating conditions, including: magnetic flux density distribution, magnetic linkage coupling, induced electromotive force, current density distribution, and electromagnetic losses. These data can accurately reflect the motor's operation under different load, speed, and voltage conditions.

[0087] In practice, the first step is to create a 3D model of the compressor motor's geometry. Then, specific boundary conditions, such as voltage or current, are applied to the model using finite element analysis software. Next, electromagnetic field simulation is performed using the motor's material properties, such as the conductivity and permeability of the stator and rotor. Finally, electromagnetic data under the target operating conditions are obtained through calculation.

[0088] In addition, a machine learning model is used to generate the predicted rotor angle. The model's inputs include compressor operating parameters (such as speed, load, and stator current), the estimated rotor angle generated via a phase-locked loop, and electromagnetic data generated through finite element analysis. By combining these inputs, the machine learning model can output a more accurate predicted angle.

[0089] The input variables include the compressor's real-time operating parameters, the rotor estimated angle generated by the phase-locked loop (PLL), and electromagnetic data obtained from finite element analysis. This combination of data provides a more comprehensive angle prediction input, compensating for the limitations of a single signal source.

[0090] For example, the geometry of a compressor motor can be obtained in several ways. For instance, it can be obtained directly from the motor's CAD model, or extracted from technical drawings and design documents provided by the motor manufacturer.

[0091] For example, finite element analysis software commonly used for electromagnetic simulation includes ANSYS Maxwell, COMSOL Multiphysics, JMAG, and Altair Flux. In specific implementations, the appropriate software can be selected based on the specific application scenario and its unique characteristics.

[0092] It is important to note that in finite element simulation, accurate setting of boundary conditions is crucial. Boundary conditions typically include: voltage or current excitation, periodic boundary conditions, air gap settings, and material properties.

[0093] In compressor motors, AC voltage or current is typically applied to the stator windings. This is usually achieved by defining an AC source in simulation software. Common boundary condition settings include setting the voltage waveform of the stator windings, such as a sinusoidal voltage (amplitude set according to the motor's rated voltage).

[0094] Furthermore, for motors with symmetrical structures, periodic boundary conditions can be used to reduce the amount of simulation computation. For example, for a 12-slot stator, only 1 / 12 of the part needs to be simulated, and the rest is automatically mapped by periodic boundary conditions.

[0095] Material properties are crucial input data for the simulation, including the electrical conductivity and permeability of materials used in the stator, rotor, and coils. Specific parameters can be obtained through the following methods:

[0096] Materials Database: Most simulation software comes with a built-in materials database. Data such as conductivity and permeability of common motor materials (such as silicon steel sheets, copper wires, permanent magnet materials, etc.) can be directly extracted from the software's database.

[0097] Before simulation, the motor's geometry needs to be discretized, i.e., divided into finite element meshes. The mesh fineness determines the simulation accuracy and computation time. For example, critical areas such as the air gap require smaller mesh elements to improve accuracy, while flatter areas can use larger mesh elements to reduce computation.

[0098] During the simulation, the software calculates the electromagnetic field of the motor, solving for parameters such as the magnetic flux between the stator and rotor, the current density distribution in the stator coils, and the induced electromotive force of the motor. The dynamic characteristics of the electromagnetic field over time are also simulated during the calculation, especially under conditions of rotor rotation or load changes.

[0099] By combining finite element analysis with machine learning models, the prediction accuracy of rotor angles can be significantly improved. The electromagnetic data generated by finite element analysis provides richer physical information for the machine learning model, which, by combining electromagnetic data, operating parameters, and the estimated rotor angle, can generate accurate predicted angles in real time. Compared to traditional methods that rely solely on empirical formulas or single signal sources, this invention provides more robust prediction results under different operating conditions, especially in the weak magnetic region, where compressor control is more stable, improving the overall energy efficiency and dynamic response performance of the system.

[0100] Understandably, the finite element analysis method used for real-time data simulation analysis will impose a certain computational burden, which may have some impact on the compressor's built-in control system with weak computing power.

[0101] As an optional implementation, the finite element analysis method employs a pre-calculation approach to pre-calculate the electromagnetic characteristics of the target compressor motor, generating electromagnetic data under different operating conditions. The electromagnetic data includes: magnetic flux density distribution, current density distribution, magnetic flux coupling, and electromagnetic loss characteristics. The pre-calculated electromagnetic data is stored through a lookup table or database and dynamically retrieved during real-time operation based on the compressor's current operating state.

[0102] Pre-calculation refers to using finite element simulation to calculate the electromagnetic characteristics of the motor under various operating conditions during the design phase, and then storing the calculation results. The benefit of pre-calculation is that it reduces the computational burden during real-time operation, especially in real-time control systems, by minimizing the time delay of complex electromagnetic calculations.

[0103] In practice, the pre-calculated electromagnetic data is stored in a lookup table or database. When the compressor is running in real time, the system can dynamically call up the pre-stored electromagnetic data according to the current operating status (such as speed, load, etc.) to quickly respond to system needs.

[0104] Among these methods, lookup tables are a common and efficient storage technique. They allow for the pre-calculation and storage of complex results, which can then be retrieved directly at runtime, significantly improving system efficiency. Databases can also be used for larger-scale storage, especially when dealing with large datasets, where they can be managed and retrieved more effectively.

[0105] For example, during real-time control, the system automatically retrieves pre-calculated electromagnetic data based on the compressor's current operating status (such as motor speed, load, voltage, etc.) measured by sensors. This data has already been generated and stored during the finite element simulation stage, so the system no longer needs to perform complex finite element calculations in real time, thereby improving the system's response speed and real-time performance.

[0106] As an optional implementation, the frequency and amplitude of the high-frequency voltage signal can also be determined based on the electromagnetic data pre-calculated by finite element analysis; when the magnetic flux density is lower than a first threshold, a high-frequency voltage signal with a frequency within a first preset range is selected for injection; when the current density is higher than a second threshold, the amplitude of the high-frequency voltage signal is adjusted to a second preset range.

[0107] In practical implementation, the electromagnetic characteristics of the motor under different loads and speeds are pre-calculated during the finite element analysis simulation. Based on this data, the system can quickly find the optimal high-frequency voltage signal frequency and amplitude using a lookup table. For example, when the magnetic flux density is low, a higher frequency signal is needed to excite the space harmonics; when the current density is high, the signal amplitude needs to be appropriately reduced to prevent motor overload.

[0108] For example, by storing frequency and amplitude lookup table data, the system automatically matches the corresponding frequency and amplitude based on real-time operating status information acquired by sensors during system operation. This method requires minimal computation, has a fast response time, and is suitable for real-time control. Furthermore, dynamic algorithms (such as fuzzy control and PID control) can be used to calculate suitable high-frequency voltage signal frequencies and amplitudes in real time, providing a more flexible control scheme in conjunction with electromagnetic data.

[0109] For example, if the current magnetic flux density is detected to be low, the lookup table indicates that the frequency should be adjusted to 5kHz, with an amplitude of 80% of the rated voltage. If the current density is high, the system automatically reduces the amplitude of the high-frequency voltage signal to a second preset range.

[0110] In practice, by monitoring the magnetic flux density of the motor in real time, when the magnetic flux density is lower than the set first threshold, a high-frequency voltage signal within a first preset range is automatically injected.

[0111] The first preset range can be determined based on experimental or simulation data of the target compressor, and is typically a high-frequency range that ensures stable motor operation. By selecting this frequency, it is possible to ensure that the magnetic field can effectively excite the spatial harmonics of the stator and rotor under low magnetic flux density conditions, thereby improving the accuracy of rotor angle estimation.

[0112] For example, in actual operation, the system monitors the magnetic flux density of the motor in real time using a magnetic flux sensor and determines whether it is below a first threshold. If it is below, the system immediately selects the appropriate frequency using a lookup table.

[0113] The frequency of the high-frequency voltage signal in the stator winding is adjusted by a modulation controller (such as a frequency converter or inverter) to ensure that the injected signal meets the requirements of the current magnetic flux state of the motor.

[0114] In practice, the current density of the stator winding is also monitored in real time. When the current density exceeds the set second threshold, the system automatically adjusts the amplitude of the high-frequency voltage signal to the second preset range to prevent the motor from overloading or overheating.

[0115] The second preset range is a safe amplitude range, which can be set based on experimental or empirical data for the target compressor, such as 50% to 80% of the rated voltage, to ensure stable operation of the motor under high load. The purpose of adjusting the amplitude is to control the signal amplitude without reducing the signal frequency to avoid the motor windings carrying excessive current.

[0116] For example, the current density of the motor windings is monitored in real time using a current sensor. If the detected current density exceeds a safety threshold (e.g., exceeding 10A / mm²), the sensor will detect the current. 2The system immediately reduces the amplitude of the high-frequency voltage signal.

[0117] The system automatically adjusts the amplitude of the high-frequency voltage signal through the inverter or power regulation module to ensure that the adjustment amplitude is within the preset safety range.

[0118] By dynamically adjusting the frequency and amplitude of the high-frequency voltage signal using electromagnetic data generated from finite element analysis, this invention can automatically select and adjust the optimal parameters of the high-frequency voltage signal based on the current magnetic flux density and current density of the compressor. This dynamic adjustment mechanism effectively improves the system's adaptability to different operating conditions. Especially under conditions of low magnetic flux density or high current density, by adjusting the frequency and amplitude, it can ensure the effectiveness of the signal while avoiding motor overload or excessive losses, thereby improving the stability and energy efficiency of the compressor operation.

[0119] As an optional implementation, the input data of the machine learning model includes: the target compressor's rotational speed, load, stator current, back electromotive force signal, estimated rotor angle, and electromagnetic data pre-calculated through finite element analysis; the machine learning model includes: a long short-term memory network model; the long short-term memory network model is used to process the input data in real time and generate a predicted angle; wherein, the long short-term memory network model generates the predicted angle by analyzing historical data and current operating parameters.

[0120] The machine learning model is input to multiple operating parameters of the target compressor and electromagnetic data generated through finite element analysis. The input data is multi-source and describes the compressor's current operating state and electromagnetic environment.

[0121] Rotational speed, or the real-time rotational speed of the target compressor, can be monitored by sensors and used as input to a machine learning model to capture dynamic changes in the rotor.

[0122] Load refers to the mechanical load borne by the compressor, which can be obtained by monitoring the compressor's power or torque. Changes in load affect electromagnetic characteristics and are therefore used as input parameters.

[0123] Stator current reflects the current status of the stator windings and is monitored in real time by a current sensor. Changes in stator current are directly related to the motor's torque and electromagnetic losses.

[0124] The back electromotive force (EMF) signal represents the voltage induced in the stator coils of a motor and is an important feedback signal of the motor's operating status. The back EMF signal is directly related to rotor motion and can provide information about the rotor angle.

[0125] The estimated rotor angle is the preliminary rotor angle calculated by the phase-locked loop (PLL) system, which is used to assist the machine learning model in generating a more accurate predicted angle.

[0126] The electromagnetic data generated by finite element analysis, i.e. pre-calculated electromagnetic data, includes information such as magnetic flux density and current density under different operating conditions of the motor, which is used to enhance the machine learning model's understanding of changes in the electromagnetic environment.

[0127] These input data provide a detailed description of the compressor's real-time status and historical operation, helping machine learning models capture the dynamic changes in the system.

[0128] In practice, the machine learning model uses a Long Short-Term Memory (LSTM) network, which has the advantage of being able to process input data with time correlation and to memorize and analyze historical data over a long period of time, thereby generating predictive perspectives.

[0129] LSTM is a type of recurrent neural network (RNN) capable of processing time-series data. By memorizing historical information, it can handle the complex relationships between the current operating state and historical states. In compressor operation, not only are current parameters crucial for predicting the rotor angle, but past operating states also influence future angle changes; LSTM can effectively handle this information.

[0130] The LSTM model receives time-series data including speed, load, and stator current, and combines it with pre-calculated electromagnetic data from finite element analysis and estimated rotor angles to generate more accurate predicted angles. This model processes this data in real time and updates the prediction results at each time step. By processing input data online, the LSTM model can generate predicted angles during the real-time operation of the compressor, ensuring the system's real-time control performance.

[0131] Furthermore, LSTM, through its internal memory cells, can retain important historical data for extended periods and fully utilize this data in current predictions. For example, trends in rotational speed and load significantly influence future rotor angle changes, and LSTM can capture these trends and use them for prediction. Simultaneously, the LSTM model also processes current real-time data. By analyzing real-time states such as rotational speed, load, and stator current, the model can dynamically adjust itself in conjunction with historical data, resulting in more accurate predictions.

[0132] For example, the input layer in this disclosure accepts multiple input parameters, including rotational speed, load, stator current, back electromotive force signal, estimated rotor angle, and electromagnetic data generated through finite element analysis. These data constitute the time-series input, with each timestep providing a set of data for the current state. The LSTM hidden layer consists of multiple units, each capable of capturing the temporal dependencies of the input data. The hidden layer is the core of the entire LSTM architecture, responsible for processing and memorizing the time-series data.

[0133] Following the LSTM layer, a fully connected layer is used to map the LSTM output to the desired output space. In this invention, the output of this layer is the predicted rotor angle. The output layer provides the final prediction result, i.e., the predicted rotor angle. This layer outputs the actual angle value based on the calculations of the previous layer.

[0134] For example, the input dimension is equal to the number of input parameters, such as speed, load, stator current, back EMF signal, rotor estimated angle and electromagnetic data. That is, the data dimension of each time step is 6, corresponding to 6 features.

[0135] The time step refers to how many time steps of data the LSTM model processes in a sequence. Choosing an appropriate time step is crucial for understanding the dynamic changes in a motor system.

[0136] Furthermore, the loss function is used to measure the error between the predicted angle and the true angle. For prediction problems, the loss function can be the mean squared error (MSE), which measures the squared error between the predicted and actual values.

[0137] For example, in training an LSTM model, dataset preparation can include: training data: historical operating parameters, including rotational speed, load, stator current, etc., and actual rotor angles. Through time-series input of this data, the model can learn the dynamic relationships between the data; input windows: dividing the time-series data into multiple input windows, for example, inputting data of a preset time step each time as input to the model, and corresponding to an actual rotor angle as output.

[0138] As an optional implementation, the input signal can also be preprocessed to extract frequency domain features including spatial harmonic components and high-frequency components of the back EMF signal; the long short-term memory network predicts the rotor angle by combining the frequency domain features with the time domain features; the input dimension of the machine learning model is extended based on the time domain features and frequency domain features, adding frequency features of spatial harmonic components and high-frequency components of the back EMF to compensate for the influence of multiple frequency components on the predicted angle.

[0139] In practical implementation, LSTM has the advantage of processing time series data and its long-term and short-term dependencies, but special feature processing and model architecture are required to compensate for the influence of different frequency components.

[0140] Spatial harmonic components are closely related to rotor angle variations. To enable LSTM to process this information, harmonic components can be included as part of the timing input. By performing frequency domain analysis (such as Fourier transform) on the current response signal, spatial harmonic components are extracted from the signal, and these harmonic components are used as input to the LSTM. In this way, the LSTM model can learn these timing harmonic characteristics to compensate for the impact of harmonic components on angle prediction.

[0141] The high-frequency components in the back EMF signal reflect the transient dynamic changes of the rotor. Therefore, by performing high-frequency filtering on the back EMF signal, the high-frequency components are extracted and used as input to an LSTM. The LSTM can learn the correlation between the changes in the high-frequency components and the rotor angle, thus making the prediction of the rotor angle more accurate.

[0142] Noise is typically a random interference that affects signal clarity. To improve the noise immunity of LSTM, the input signal can be preprocessed, including denoising filtering (such as low-pass filtering or adaptive filtering) to reduce noise components. Alternatively, noise can be directly used as a temporal input feature, allowing LSTM to learn how to identify and compensate for the impact of noise on angle prediction.

[0143] In practical implementation, the architecture of the LSTM network can be appropriately adjusted to better handle the influence of different frequency components on the prediction angle.

[0144] For example, an LSTM model can be designed with multiple input channels, each responsible for processing different frequency characteristics. For instance, spatial harmonic components, high-frequency components of the back electromotive force, and time-domain signals (such as rotational speed and load) can be passed to the LSTM through different input paths. With this structure, the LSTM can focus on variations in different signal frequencies and learn how to combine this information.

[0145] Alternatively, a hierarchical LSTM architecture can be used, allowing different LSTM layers to handle different frequency features. For example, the first LSTM layer can specifically handle time-domain signals (such as rotational speed and load), while the second LSTM layer can handle frequency-domain signals (such as spatial harmonics and high-frequency components). This hierarchical structure helps the network better learn the contribution of different frequency components to the prediction angle.

[0146] To capture local features within different frequency components, a convolutional neural network (CNN) can be added before the LSTM, specifically for extracting spatial harmonic and high-frequency components from the input signal. Convolutional layers effectively capture local frequency features in the signal, while LSTM excels at handling long-term dependencies. The combination of CNN and LSTM can effectively improve the model's predictive power.

[0147] While LSTM excels at time-series modeling, its compensation capabilities can be significantly enhanced by performing frequency domain transformation on the input signal to extract spatial harmonics and high-frequency components. Combining these frequency domain features with time-domain data can further improve the model's performance. For example, Fourier transform or wavelet transform can be used to convert the time-domain signal into frequency-domain features, which can then be used as input to the LSTM.

[0148] By optimizing the loss function, the influence of different frequency components can be weighted. For example, giving higher weights to high-frequency components (such as spatial harmonics and high-frequency components of back electromotive force) ensures that the LSTM pays more attention to these key frequency features when predicting angles.

[0149] It is important to note that when high-frequency components of spatial harmonics or back electromotive force are extracted as independent features, the input dimension increases accordingly to ensure that the model can better compensate for the impact of these frequency components on angle prediction.

[0150] Thus, by employing a Long Short-Term Memory (LSTM) network as a machine learning model, time-series data can be effectively processed. By combining historical operating parameters and current electromagnetic characteristics, accurate rotor prediction angles can be generated. Compared to traditional methods, LSTM can utilize past operating information and process the current state in real time, thereby providing more stable and accurate prediction results. Furthermore, by combining pre-calculated electromagnetic data generated from finite element analysis with rotor angle estimation from a phase-locked loop, the prediction accuracy and real-time system response are further improved. This invention significantly enhances the operating stability and energy efficiency control of the compressor in the weak magnetic region.

[0151] In Long Short-Term Memory (LSTM) networks or other recurrent neural network (RNN) models, traditional backpropagation algorithms propagate errors back along the entire time series and update model weights. However, excessively long time series can lead to high computational costs and are prone to gradient vanishing or exploding problems. Therefore, this disclosure employs a Truncated Backpropagation Through Time (TBPTT) algorithm, which reduces computational complexity by processing time series data in segments and limits the number of backpropagation steps through local gradient calculations.

[0152] As an optional implementation, the machine learning model is trained using a truncated backpropagation algorithm, including: segmenting the historical operating parameters of the target compressor into segments based on a preset length; wherein the preset length is determined based on the system computational load and real-time target; the historical operating parameters include time-series data of speed, load, stator current, and back electromotive force signal; performing local gradient calculation on the time-series data of each segment of the compressor's historical operating parameters, and updating the weights of the machine learning model based on the local gradient calculation results; wherein the local gradient calculation limits the number of time steps in each backpropagation by truncating backpropagation; after the calculation of each time series data is completed, the hidden state and memory state of the current time series data are used as the initial state for the calculation of the next time series data.

[0153] This method divides long-term data into multiple shorter segments and performs backpropagation and gradient updates within each segment. The time steps for each backpropagation are limited, thus eliminating the need to backpropagate along the entire sequence and significantly reducing computational overhead. Furthermore, after each time-series computation is completed, the hidden and memory states are preserved and passed to the next time-series segment, ensuring no information is lost.

[0154] Each time series is processed individually with its local gradient calculated, and the weights are updated based on the local error of that segment. Ultimately, the weights of the entire model gradually approach their optimum after multiple local updates.

[0155] In practice, the historical operating parameters of the target compressor (such as speed, load, stator current, back electromotive force, etc.) are divided into segments with a time step length of 50. Each segment represents the compressor's operation within 50 time steps. The segment length can be dynamically adjusted according to the system's calculated load and real-time requirements.

[0156] For each data segment spanning 50 time steps, the model performs forward propagation, then calculates the error using local gradients, and performs backpropagation and weight updates within these 50 time steps. The backpropagation time steps are limited to 50, and longer historical data are not traced back.

[0157] After each 50 time steps of data computation, the hidden and memory states of the LSTM are preserved and used as the initial input states for the next 50 time steps. This maintains the contextual continuity of the time series and ensures that the model captures long-term dependencies.

[0158] By segmenting the long-term time series, the computational complexity during backpropagation is significantly reduced. This segmented processing avoids the need to compute the entire time series, reducing memory usage and computation time, and improving the system's real-time performance.

[0159] As an optional implementation, the preset length is determined based on system computing load and real-time target, including: monitoring the system computing load of the target compressor; wherein the system computing load is determined based on the weighted average of the processor utilization and memory occupancy; when the weighted average of the system computing load is greater than or equal to a preset first load threshold, the preset length is shortened; when the weighted average of the system computing load is less than the first load threshold, the preset length is extended.

[0160] In this disclosure, the system's computational load is assessed by real-time monitoring of the weighted average of processor utilization and memory usage. The processor (such as a CPU or GPU) of a compressor system consumes significant computational resources when running an LSTM model. By monitoring the system's resource usage, the model's operating parameters can be dynamically adjusted to ensure the system meets real-time performance targets.

[0161] In practice, processor utilization (CPU / GPU usage) can be collected in real time through the operating system or dedicated monitoring software to indicate the current computational load on the processor. Memory usage is assessed by monitoring system memory utilization. The training and inference processes of LSTM models consume a significant amount of memory, especially when processing large-scale time series data, where memory usage can become a bottleneck. Computational load is measured by a weighted average of processor utilization and memory utilization. Specifically, the weights can be set according to the system's design requirements, for example, processor utilization and memory utilization could be weighted at 70% and 30%, respectively.

[0162] Based on the monitoring results of the system's computational load, the time series segment length (i.e., the preset length) of the LSTM model is dynamically adjusted. When the load is high, the length of each time series is shortened to avoid exceeding the computational resource limit, thereby reducing the computational pressure on the model; when the load is low, the time step can be appropriately extended.

[0163] Based on the historical operating data of the target compressor, a first load threshold is set as the critical point for calculating the load. If the calculated load is greater than or equal to this threshold, it indicates that system resources are relatively strained, and the time series length should be shortened; if the load is lower than the threshold, it indicates that system resources are not being fully utilized, and the series length can be extended to improve prediction accuracy.

[0164] For example, key performance indicators such as compressor system response time, task processing time, and model inference time can be recorded under different load scenarios. Particular attention should be paid to whether the system's computing power is fully utilized under high load, and whether there are issues such as latency, processor overheating, or memory overflow. Once this data is collected, the next step is to set a critical point for the computing load (the first load threshold) based on the system's real-time requirements. This threshold should reflect whether the system can maintain the required response speed under high load without exceeding the capacity of its computing resources.

[0165] For example, when the computational load exceeds a first load threshold (e.g., exceeding 70%), the system automatically shortens the time step. For instance, a sequence originally with 50 time steps can be shortened to 30 time steps to reduce the model's computational load and improve real-time response. When the computational load is below the first load threshold, the system can appropriately extend the time step to increase the model's ability to process historical data. For example, the original 30 time steps can be extended to 50 time steps to improve the model's prediction accuracy.

[0166] In this way, by dynamically adjusting the preset length of the time series, the present invention can automatically adapt to system resources under different computing loads, ensuring that the system can still maintain real-time performance under high load and avoiding model computation overload.

[0167] Regarding the above S103:

[0168] In practical implementation, the following key operating parameters can be adjusted based on the prediction perspective:

[0169] Rotor speed: The predicted angle of the rotor reflects the deviation of the motor speed. By providing feedback on this angle, the compressor system can adjust the speed to ensure that the compressor maintains stable operating efficiency under different load conditions.

[0170] Torque / Load: If the predicted angle indicates that the rotor angle deviates from the expected value, the system can adjust the torque by controlling the magnetic field in the stator windings by adjusting the current or voltage, thereby ensuring that the compressor load is adapted to the working condition.

[0171] Stator current: Predicted angle helps determine if the motor is under high load. Under high load conditions, the system can adjust the stator current based on angle feedback to prevent overload or motor losses.

[0172] Operating frequency: If the system detects irregular fluctuations in the rotor prediction angle, it may indicate that the system load is changing significantly. The system can automatically adjust the operating frequency to match the current operating requirements and ensure that the compressor operates at the optimal frequency.

[0173] In this way, by adjusting parameters such as compressor speed, load, and current based on the rotor prediction angle, the compressor can maintain efficient operation under different loads and operating conditions, avoiding system instability or overload problems caused by mismatched operating parameters.

[0174] It is important to emphasize that in the phase-locked loop (PLL) system, the space harmonic components and back EMF signals have already been weighted and combined, and a preliminary rotor angle estimate has been generated through phase error detection. This process effectively utilizes these signals and corrects the phase error through a feedback control mechanism, generating a relatively accurate estimated angle. At this stage, the PLL can process noise and high-frequency components in the combined space harmonic and back EMF signals, thus obtaining a reliable angle estimate.

[0175] The introduction of machine learning models is primarily aimed at further optimizing or compensating for estimated angles, especially improving prediction accuracy under complex operating conditions. Deep learning models like LSTM excel at handling time-series data, capturing more complex time dependencies and historical information, thus compensating for the shortcomings of traditional feedback control systems (such as PLLs) in certain nonlinear or highly dynamic scenarios. Specifically, LSTM can: utilize broader historical data and real-time inputs to compensate for errors in PLL angle estimation caused by complex nonlinear operating conditions; handle the complex influences of multi-source data from the system (such as electromagnetic characteristics, rotational speed, load, etc.) on angle estimation, characteristics that PLL systems may not be able to effectively handle; and provide a more flexible processing method than a single feedback mechanism based on phase error, especially under complex operating conditions or with higher uncertainties.

[0176] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0177] Based on the same inventive concept, this application also provides a compressor magnetic field weakening region stable operation system based on angle compensation. Since the principle of the system in this application to solve the problem is similar to the compressor magnetic field weakening region stable operation method based on angle compensation described above in this application, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0178] Reference Figure 3 The diagram shown is a schematic of a compressor magnetic field weakening region stable operation system based on angle compensation provided in an embodiment of this application. The system includes:

[0179] The estimation unit 10 estimates the rotor angle of the target compressor based on the phase-locked loop and generates the estimated angle of the rotor.

[0180] The prediction unit 20 uses a preset machine learning model to generate a predicted angle based on the operating parameters of the target compressor and the estimated angle of the rotor; wherein, the machine learning model is used to compensate for the influence of multiple frequency components on the predicted angle; the multiple frequency components include: spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals;

[0181] The control unit 30 adjusts the operating parameters of the target compressor based on the predicted angle.

[0182] The description of the processing flow of each module in the system and the interaction flow between each module can be found in the relevant descriptions in the above method embodiments, and will not be detailed here.

[0183] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this disclosure, it should be understood that the disclosed system and method can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0184] 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.

[0185] In addition, the functional units in the various embodiments of this disclosure 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.

[0186] Finally, it should be noted that the above-described embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A method for stable operation of a compressor in the field weakening region based on angle compensation, characterized in that, include: The rotor angle of the target compressor motor is estimated based on the phase-locked loop, and the estimated angle of the rotor is generated. Using a pre-defined machine learning model, a predicted angle is generated based on the operating parameters of the target compressor motor, the estimated angle of the rotor, and electromagnetic data generated through finite element analysis. The machine learning model is used to compensate for the influence of multiple frequency components on the predicted angle. The multiple frequency components include: spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals. Based on the predicted angle, adjust the operating parameters of the target compressor motor; The process of generating the estimated angle of the rotor also includes: A high-frequency voltage signal is injected into the stator winding of the target compressor motor; Acquire the current response signal of the stator winding and extract the spatial harmonic components from the current response signal; The spatial harmonic components and the back electromotive force signal are weighted and combined, and then fed to the phase-locked loop for phase error detection to generate the estimated angle.

2. The method according to claim 1, characterized in that, It also includes: modeling the electromagnetic characteristics of the target compressor motor using the finite element analysis method, and generating electromagnetic data of the target compressor motor under the target operating state.

3. The method according to claim 2, characterized in that, The finite element analysis method employs a pre-calculation approach to pre-calculate the electromagnetic characteristics of the target compressor motor, generating electromagnetic data under different operating conditions. The electromagnetic data includes: magnetic flux density distribution, current density distribution, magnetic flux coupling, and electromagnetic loss characteristics. The pre-calculated electromagnetic data is stored through a lookup table or database and dynamically retrieved during real-time operation based on the compressor's current operating status.

4. The method according to claim 3, characterized in that, Also includes: Based on the electromagnetic data pre-calculated using finite element analysis, the frequency and amplitude of the high-frequency voltage signal are determined. When the magnetic flux density is lower than the first threshold, a high-frequency voltage signal with a frequency within the first preset range is selected for injection; When the current density is higher than the second threshold, the amplitude of the high-frequency voltage signal is adjusted to the second preset range.

5. The method according to claim 4, characterized in that, The input data for the machine learning model includes: the speed, load, stator current, back electromotive force signal of the target compressor motor, the estimated angle of the rotor, and electromagnetic data pre-calculated through finite element analysis; The machine learning model includes: a long short-term memory network model; The long short-term memory network model is used to process the input data in real time and generate predicted angles; The Long Short-Term Memory (LSTM) network model generates a prediction angle by analyzing historical data and current operating parameters.

6. The method according to claim 5, characterized in that, The machine learning model is trained using a truncated backpropagation algorithm, including: The historical operating parameters of the target compressor motor are segmented based on a preset length. The preset length is determined based on the system's calculated load and real-time performance targets; the historical operating parameters include: time-series data of rotational speed, load, stator current, and back electromotive force signal; Local gradient calculations are performed on the time-series data of the historical operating parameters of each compressor segment, and the weights of the machine learning model are updated based on the results of the local gradient calculations. The local gradient calculation limits the number of time steps for each backpropagation by truncating backpropagation. After the calculation of each segment of time series data is completed, the hidden state and memory state of the current segment of time series data are used as the initial state for the calculation of the next segment of time series data.

7. The method according to claim 6, characterized in that, The preset length is determined based on system computational load and real-time targets, including: The system computational load of the target compressor motor is monitored; wherein the system computational load is determined based on a weighted average of processor utilization and memory occupancy. When the weighted average of the system's calculated load is greater than or equal to a preset first load threshold, the preset length is shortened. When the weighted average of the system's calculated load is less than the first load threshold, the preset length is extended.

8. A compressor magnetic field weakening region stable operation system based on angle compensation, characterized in that, include: The estimation unit estimates the rotor angle of the target compressor motor based on the phase-locked loop, and generates the estimated rotor angle. The prediction unit uses a preset machine learning model to generate a predicted angle based on the operating parameters of the target compressor motor, the estimated angle of the rotor, and electromagnetic data generated through finite element analysis. The machine learning model is used to compensate for the influence of multiple frequency components on the predicted angle. The multiple frequency components include: spatial harmonic components, high-frequency components of the back electromotive force signal, and noise signals. The control unit adjusts the operating parameters of the target compressor motor based on the predicted angle; The process of generating the estimated angle of the rotor also includes: A high-frequency voltage signal is injected into the stator winding of the target compressor motor; Acquire the current response signal of the stator winding and extract the spatial harmonic components from the current response signal; The spatial harmonic components and the back electromotive force signal are weighted and combined, and then fed to the phase-locked loop for phase error detection to generate the estimated angle.

Images