Method for establishing temperature monitoring model for permanent magnet synchronous motor and temperature monitoring method

Abstract

The application discloses a temperature monitoring model establishment method and a temperature monitoring method for a permanent magnet synchronous motor, and belongs to the field of synchronous motor state monitoring, and comprises the following steps: constructing a full-condition operation data set of the permanent magnet synchronous motor; constructing a prediction network, and training the prediction network by using the full-condition operation data set to update the parameters of a prediction output model; and obtaining the temperature monitoring model for the permanent magnet synchronous motor after the training is completed; the prediction network comprises: a calibrated thermal network model of the permanent magnet synchronous motor, which takes motor state data as input and is used for estimating the temperature of a core node; and a prediction output model, which is a deep learning model, takes the core node temperature estimation result output by the thermal network model and corresponding motor state data as input, and is used for predicting the temperature of the core node. The application reduces the dependence on a large amount of motor state data in the machine learning process, improves the interpretability of the model, and thus improves the accuracy and reliability of the temperature monitoring of the permanent magnet synchronous motor.

Description

Technical Field

[0001] This invention belongs to the field of synchronous motor condition monitoring, and more specifically, relates to a method for establishing a temperature monitoring model and a temperature monitoring method for permanent magnet synchronous motors. Background Technology

[0002] With the rapid popularization of electric and hybrid vehicles, the electric motor, as a key component of the electric vehicle's power transmission system, naturally faces more development opportunities and challenges. Compared to traditional industrial motors, electric vehicle motors must have higher torque and power density to meet the constraints of vehicle interior space, and also must have a wider operating speed range to meet the vehicle's traction requirements. Among all types of motors, permanent magnet synchronous motors have become the preferred choice for electric vehicle main drive motors due to their strong overload capacity and wide-range field weakening capability.

[0003] With the dual objectives of maximizing power / torque density and satisfying thermal constraints on internal components, the thermal mechanism of permanent magnet synchronous motors (PMSMs) has become a key area of ​​research in both academic studies and industrial applications. Accurate modeling of the internal thermal state of the motor and real-time temperature monitoring are crucial for improving the efficiency of PMSMs, extending their lifespan, and reducing the failure rate of internal components, especially the permanent magnets and windings.

[0004] Traditional methods for direct temperature measurement based on physical sensors achieve this by embedding temperature sensors in key components of the motor. While simple to implement and widely used in industrial applications, this method lacks analysis of the complex internal thermal characteristics of the motor, limiting temperature monitoring to a limited number of critical stator-side components. Furthermore, it is only suitable for industrial scenarios with constant speed, constant operating environment, and small load variations. Additionally, the extra sensors are susceptible to electromagnetic interference, which can lower the winding insulation level and reduce the overall system reliability.

[0005] Compared to methods that directly measure temperature using temperature sensors, mathematical modeling methods based on electrical parameters or thermal networks can identify the temperature of components, including the rotor, without requiring internal sensors. However, the application of such methods is limited by the fact that temperature estimation results are overly sensitive to errors in electrical parameters and the problem of determining the internal parameters of the thermal network.

[0006] To address the limitations of traditional temperature estimation based on electrical parameters and thermal network models, some researchers have proposed using machine learning to bypass traditional physical models. This involves training a machine learning model using a large amount of training data from a statistical perspective, and then using the trained model to achieve a low-order fit for temperature estimation under actual operating conditions. However, to ensure good predictive performance, a large amount of training data is required. In practical applications, the available motor state data is limited, making it difficult to meet the training requirements of the machine learning model and thus failing to achieve the desired prediction accuracy. Furthermore, the proposed machine learning model lacks interpretability and may contain singularities in temperature estimation, reducing the algorithm's security. Summary of the Invention

[0007] To address the shortcomings and improvement needs of existing technologies, this invention provides a method for establishing a temperature monitoring model and a temperature monitoring method for permanent magnet synchronous motors. The purpose is to reduce the reliance on a large amount of motor state data during machine learning and improve the interpretability of the model, thereby improving the accuracy and reliability of temperature monitoring for permanent magnet synchronous motors.

[0008] To achieve the above objectives, according to one aspect of the present invention, a method for establishing a temperature monitoring model for a permanent magnet synchronous motor is provided, comprising:

[0009] Construct a full-condition operation dataset for permanent magnet synchronous motors; the full-condition dataset includes motor status data and corresponding core node temperatures of the permanent magnet synchronous motor under all operating conditions; the core nodes include the stator, windings and rotor;

[0010] Construct a prediction network; the prediction network includes: a calibrated permanent magnet synchronous motor thermal network model and a prediction output model; the thermal network model takes motor state data as input to estimate the temperature of the core node; the prediction output model is a deep learning model, which takes the core node temperature estimation result output by the thermal network model and the corresponding motor state data as input to correct the error in the core node temperature estimation result and output the corrected core node temperature.

[0011] The prediction network is trained using a full-condition operating status dataset to update the parameters of the prediction output model. After training, the trained prediction network is output as a temperature monitoring model for permanent magnet synchronous motors.

[0012] Furthermore, the full-condition operation data also includes: the extended node temperature corresponding to the motor status data;

[0013] Furthermore, the predicted output model includes:

[0014] The correction branch takes the core node temperature prediction results output by the thermal network model and the corresponding motor state data as input, and is used to correct the errors in the core node temperature prediction results, and outputs the corrected core node temperature.

[0015] And an extended node temperature prediction branch, which takes the core node temperature prediction results output by the thermal network model and the corresponding motor state data as input, to predict the temperature of the extended nodes of the permanent magnet synchronous motor.

[0016] The extended nodes include at least one of the following: bearings, end windings, different slot positions, and motor end covers.

[0017] Furthermore, both the corrected branch and the extended node temperature prediction branch are long short-term memory networks.

[0018] Furthermore, the heat network model is established as follows:

[0019] The casing, stator, windings, rotor, and external environment of the permanent magnet synchronous motor are represented as nodes in a thermal network model, denoted as nodes c, s, w, r, and am, respectively.

[0020] An equivalent thermal resistance R between the stator and the outer casing is added between nodes c and s. c,s An equivalent thermal resistance R between the stator and winding is added between nodes s and w. s,w An equivalent convective thermal resistance R between the stator and rotor is added between nodes s and r. s,r An equivalent convective thermal resistance R between the rotor and the external environment is added between nodes r and am. r,am ;

[0021] The equivalent heat capacity C of the stator is added between node s and the zero-degree equipotential surface. s and loss P s The equivalent heat capacity C of the winding is added between node w and the zero equipotential surface. w and loss P w The equivalent heat capacity C of the rotor is added between node r and the zero-degree equipotential surface. r and loss P r ;

[0022] A first heat source is added between the shell node c and the zero-degree equipotential surface to provide θ. c The temperature difference is considered, and a second heat source is added between the ambient temperature node am and the zero-degree equipotential surface to provide θ. am The temperature difference is used to establish the thermal network model;

[0023] Where, θ c and θ am These represent the temperatures of the outer casing and the external environment, respectively.

[0024] According to another aspect of the present invention, a method for establishing a temperature monitoring model for a permanent magnet synchronous motor is provided, comprising:

[0025] Construct a full-condition operation dataset for permanent magnet synchronous motors; the full-condition dataset includes motor status data and corresponding core node temperatures of the permanent magnet synchronous motor under all operating conditions; the core nodes include the stator, windings and rotor;

[0026] Construct a prediction network; the prediction network includes: a calibrated permanent magnet synchronous motor thermal network model, and a prediction output model;

[0027] A prediction network is constructed, comprising: a calibrated thermal network model of a permanent magnet synchronous motor and a prediction output model; the thermal network model takes motor state data as input to estimate the temperature of the core node; the prediction output model is a deep learning model, including an error prediction module and a residual network link; the error prediction module takes the core node temperature estimation result output by the thermal network model and the corresponding motor state data as input to predict the error in the core node temperature estimation result; the residual network link is used to link the output of the thermal network model to the output of the error prediction module to correct the error in the core node temperature estimation result and output the corrected core node temperature.

[0028] The prediction network is trained using a full-condition operating status dataset to update the parameters of the prediction output model. After training, the trained prediction network is output as a temperature monitoring model for permanent magnet synchronous motors.

[0029] Furthermore, the full-condition operation data also includes: the extended node temperature corresponding to the motor status data;

[0030] Furthermore, the prediction output model also includes:

[0031] The extended node temperature prediction module takes the core node temperature prediction results output by the thermal network model and the corresponding motor status data as input to predict the temperature of the extended nodes of the permanent magnet synchronous motor.

[0032] The extended nodes include at least one of the following: bearings, end windings, different slot positions, and motor end covers.

[0033] Furthermore, both the error prediction module and the extended node temperature prediction module are long short-term memory networks.

[0034] Furthermore, the heat network model is established as follows:

[0035] The casing, stator, windings, rotor, and external environment of the permanent magnet synchronous motor are represented as nodes in a thermal network model, denoted as nodes c, s, w, r, and am, respectively.

[0036] An equivalent thermal resistance R between the stator and the outer casing is added between nodes c and s. c,s An equivalent thermal resistance R between the stator and winding is added between nodes s and w. s,w An equivalent convective thermal resistance R between the stator and rotor is added between nodes s and r. s,r An equivalent convective thermal resistance R between the rotor and the external environment is added between nodes r and am. r,am ;

[0037] The equivalent heat capacity C of the stator is added between node s and the zero-degree equipotential surface. s and loss P s The equivalent heat capacity C of the winding is added between node w and the zero equipotential surface. w and loss P w The equivalent heat capacity C of the rotor is added between node r and the zero-degree equipotential surface. r and loss P r ;

[0038] A first heat source is added between the shell node c and the zero-degree equipotential surface to provide θ. c The temperature difference is considered, and a second heat source is added between the ambient temperature node am and the zero-degree equipotential surface to provide θ. am The temperature difference is used to establish the thermal network model;

[0039] Where, θ c and θ am These represent the temperatures of the outer casing and the external environment, respectively.

[0040] According to another aspect of the present invention, a method for monitoring the temperature of a permanent magnet synchronous motor is provided, comprising:

[0041] The operating status data of the permanent magnet synchronous motor is acquired in real time, input into the temperature monitoring model of the permanent magnet synchronous motor established by the temperature monitoring model establishment method of the permanent magnet synchronous motor provided by the present invention, and the output of the temperature monitoring model of the permanent magnet synchronous motor is collected to complete the temperature monitoring.

[0042] According to another aspect of the present invention, a computer-readable storage medium is provided, including a stored computer program; when the computer program is executed by a processor, it controls the device where the computer-readable storage medium is located to execute the above-described method for establishing a temperature monitoring model for a permanent magnet synchronous motor provided by the present invention, and / or the above-described method for monitoring the temperature of a permanent magnet synchronous motor provided by the present invention.

[0043] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0044] (1) The machine learning model for temperature monitoring of permanent magnet synchronous motors established in this invention consists of two main parts: a thermal network model and a deep learning model, which are connected in series. The thermal network model describes the internal thermal characteristics of the permanent magnet synchronous motor, and the temperature of the core node of the permanent magnet synchronous motor output by the model is used as the input of the deep learning model. This provides a physically meaningful model benchmark for the entire machine learning model architecture, enhances the interpretability of the entire model architecture, avoids safety issues, and improves the reliability of temperature monitoring. At the same time, it can significantly reduce the dependence of the subsequent deep learning process on complex networks and a large amount of training data, and achieve low-order fitting of the internal physical model. Thus, the accuracy of the temperature detection results of the entire model can be guaranteed even when the actual training data is limited.

[0045] In one approach, the deep learning component reconstructs the entire thermophysical process, directly correcting errors in the core node temperature estimates output by the thermal network model. In another approach, the deep learning component is used only to predict errors in the core node temperature estimates output by the thermal network model. A residual network link is introduced between the output of the thermal network model and the output of the deep learning component. This allows the deep learning component to focus on learning the residual parts that the network model fails to cover, rather than reconstructing the entire thermophysical process. This improves the ensemble effect of the model and achieves better prediction results with limited training data.

[0046] (2) In the preferred embodiment of the present invention, the established temperature monitoring model can also predict the temperature of key parts that the thermal network model may ignore, such as bearings and end windings, so as to more comprehensively realize the temperature monitoring of permanent magnet synchronous motor.

[0047] (3) In a preferred embodiment of the present invention, a long short-term memory network is used to implement the deep learning part. The long short-term memory neural network can integrate the information of the current and previous time steps, and can better learn the mapping relationship between (motor status data, core node temperature estimation results) and the final temperature monitoring results, thereby further improving the accuracy of temperature monitoring.

[0048] (4) In a preferred embodiment of the present invention, the thermal network model selects the shell, stator, winding, rotor, and external environment as equivalent nodes in the permanent magnet motor thermal network model. Based on the constant temperature of the shell and the external environment, heat sources of corresponding temperatures are introduced between the shell nodes, the external environment nodes, and the zero-degree equipotential surface. Based on the thermal constraints of the internal components of the permanent magnet motor, equivalent crosstalk thermal resistances are introduced between the shell nodes and the stator nodes, and between the stator nodes and the winding nodes. Equivalent convection thermal resistances are introduced between the stator nodes and the rotor nodes, and between the rotor nodes and the external environment nodes. Furthermore, equivalent crosstalk thermal resistances are introduced between the stator nodes, the winding nodes, the rotor nodes, and the external environment nodes. The zero-degree equation surface introduces corresponding equivalent heat capacity and losses, enabling the thermal network model to accurately model the internal thermal characteristics of the permanent magnet synchronous motor. Furthermore, this thermal network model only contains three nodes with unknown temperatures: stator nodes, winding nodes, and rotor nodes. Therefore, the thermal network model established in this invention is specifically a third-order thermal network model. Analysis and verification show that all parameters in this third-order thermal network model are observable and converge to their unique solution. Based on the third-order thermal network model established in this invention, the temperature of the permanent magnet synchronous motor can be accurately and reliably estimated without relying on the motor's geometric parameters and material properties. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of a temperature monitoring model for a permanent magnet synchronous motor provided in Embodiment 1 of the present invention;

[0050] Figure 2 This is a schematic diagram of the heat network model provided in Embodiment 1 of the present invention;

[0051] Figure 3 This is a schematic diagram of a temperature monitoring model for a permanent magnet synchronous motor provided in Embodiment 2 of the present invention;

[0052] Figure 4 This is a schematic diagram of a temperature monitoring model for a permanent magnet synchronous motor provided in Embodiment 3 of the present invention;

[0053] Figure 5 This is a schematic diagram of a temperature monitoring module for a permanent magnet synchronous motor provided in Embodiment 4 of the present invention;

[0054] Figure 6 This invention provides a schematic diagram of the operating state of a permanent magnet synchronous motor.

[0055] Figure 7 for Figure 6 The diagram shows the monitoring results of the temperature monitoring performance of the motor windings, stator and rotor nodes under random operation spectrum.

[0056] Figure 8 for Figure 6 The diagram shows the temperature monitoring results of the extended node of the tested motor under a random operating spectrum. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0058] In this invention, the terms "first," "second," etc. (if present) in the invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0059] To address the technical issues of existing machine learning-based temperature monitoring methods for permanent magnet synchronous motors (PMSMs) relying on large amounts of training data and exhibiting poor model interpretability, this invention provides a method for establishing a temperature monitoring model and a temperature monitoring method for PMSMs. The overall approach involves proposing a novel machine learning model with two key components: a thermal network model and a deep learning model, arranged in a series. The thermal network model describes the internal thermal characteristics of the PMSM, and its output—the temperature of the PMSM's core nodes—serves as input to the deep learning model. This provides a physically meaningful benchmark for the entire machine learning model architecture, enhancing its interpretability and significantly reducing the reliance on complex networks and large amounts of training data in subsequent deep learning processes.

[0060] The following is an example.

[0061] Example 1:

[0062] A method for establishing a temperature monitoring model for a permanent magnet synchronous motor includes:

[0063] Construct a full-condition operation dataset for permanent magnet synchronous motors; the full-condition dataset includes motor status data and corresponding core node temperatures of the permanent magnet synchronous motor under all operating conditions; the core nodes include the stator, windings and rotor;

[0064] Constructing prediction networks; such as Figure 1 As shown, the prediction network includes: a calibrated permanent magnet synchronous motor thermal network model and a prediction output model; the thermal network model takes motor state data as input to estimate the temperature of the core node; the prediction output model is a deep learning model, which takes the core node temperature estimation result output by the thermal network model and the corresponding motor state data as input to correct the error in the core node temperature estimation result and output the corrected core node temperature.

[0065] The prediction network is trained using a full-condition operating status dataset to update the parameters of the prediction output model. After training, the trained prediction network is output as a temperature monitoring model for permanent magnet synchronous motors.

[0066] It is easy to understand that motor status data refers to data that affects the temperature to be monitored during motor operation; optionally, in this embodiment, motor status data specifically includes: speed, current, and ambient temperature.

[0067] The machine learning model for temperature monitoring of a permanent magnet synchronous motor established in this embodiment consists of two main parts: a thermal network model and a deep learning model, which are connected in series. The thermal network model describes the internal thermal characteristics of the permanent magnet synchronous motor, and its output temperature of the core node of the permanent magnet synchronous motor is used as the input of the deep learning model. This provides a physically meaningful model benchmark for the entire machine learning model architecture, enhancing the interpretability of the entire model architecture. At the same time, it can significantly reduce the dependence of the subsequent deep learning process on complex networks and large amounts of training data, achieving low-order fitting of the internal physical model. Thus, the accuracy of the temperature detection results of the entire model can be guaranteed even with limited actual training data.

[0068] The basic idea of ​​the thermal network model for permanent magnet synchronous motors is to group motor components with similar characteristics into a simplified common equivalent node, transforming the partial differential equation of heat, as shown in Equation (1), into an ordinary differential equation, as shown in Equation (2). Therefore, it effectively avoids complex gradient calculations in real-time temperature estimation and reduces the number of parameters that need to be identified during calibration.

[0069]

[0070]

[0071] In formula (1), ρ is the mass density, and c p Let C be the specific heat capacity, θ be the scalar temperature field, p be the heat generated at a specific point, λ be the direction-dependent thermal conductivity, and ▽ be the gradient operator. In formula (2), C i P i and θ i These represent the thermal capacity, total power consumption, and average temperature of node i, respectively. θ ex,k R represents the boundary temperature, which is the ambient air and coolant temperature used as the model boundary. ij and R ik The equivalent thermal resistance between different nodes. ζ(t) is a vector with respect to operating conditions and time, used to indicate that the parameters in the lumped parameter thermal network model are time-varying variables.

[0072] According to formula (2), the challenge of temperature monitoring becomes finding the actual thermal resistance, heat capacity, and loss parameters of the motor. The thermal resistance, heat capacity, and loss parameters in formula (1) can be identified through actual motor operating data or extracted from a pre-designed simulation model.

[0073] Due to errors caused by internal parameters identified from actual operating data or inferred from simulations, and additional uncertainties generated during the establishment of the thermal network model, the temperature estimation results output by the thermal network model often contain a certain degree of error. This embodiment found that this error is closely related to parameters such as speed and current in the motor state data. Therefore, this embodiment uses the core node temperature estimation results output by the thermal network model and the motor state data as inputs to the deep learning model. This is beneficial for the deep learning model to learn errors, further reduce dependence on training data, and accelerate the training efficiency of the model.

[0074] Considering the current methods for calculating the parameters of permanent magnet synchronous motors (PMSMs) using finite element methods, precise motor geometry and material thermal properties are required for accurate thermal network models. However, these precise parameters are often difficult to obtain. While global parameter identification methods can identify parameters online, eliminating reliance on readily available motor geometry and material thermal properties, existing thermal network models are complex, and global parameter identification methods cannot guarantee simultaneous convergence of all parameters, thus posing a risk of divergence in parameter identification and temperature monitoring. To address this issue, this embodiment proposes a novel, simplified, and efficient thermal network model as a preferred implementation. This model accurately models the internal thermal characteristics of the motor while ensuring the observability of all parameters and guaranteeing that all parameters converge to their unique solutions, providing a reliable basis for real-time temperature estimation of PMSMs. Specifically, in this embodiment, the thermal network model is established as follows:

[0075] The casing, stator, windings, rotor, and external environment of the permanent magnet synchronous motor are represented as nodes in a thermal network model, denoted as nodes c, s, w, r, and am, respectively.

[0076] An equivalent thermal resistance R between the stator and the outer casing is added between nodes c and s. c,s An equivalent thermal resistance R between the stator and winding is added between nodes s and w. s,w An equivalent convective thermal resistance R between the stator and rotor is added between nodes s and r. s,r An equivalent convective thermal resistance R between the rotor and the external environment is added between nodes r and am. r,am ;

[0077] The equivalent heat capacity C of the stator is added between node s and the zero-degree equipotential surface. s and loss P sThe equivalent heat capacity C of the winding is added between node w and the zero equipotential surface. w and loss P w The equivalent heat capacity C of the rotor is added between node r and the zero-degree equipotential surface. r and loss P r ;

[0078] A first heat source is added between the shell node c and the zero-degree equipotential surface to provide θ. c The temperature difference is considered, and a second heat source is added between the ambient temperature node am and the zero-degree equipotential surface to provide θ. am The temperature difference is used to establish the thermal network model;

[0079] Where, θ c and θ am These represent the temperatures of the outer casing and the external environment, respectively.

[0080] The heat network model established in this embodiment is as follows: Figure 2 As shown in the figure, this thermal network model can accurately model the internal thermal characteristics of a permanent magnet synchronous motor. Furthermore, this thermal network model only contains three nodes with unknown temperatures: the stator node, the winding node, and the rotor node. Therefore, the thermal network model established in this embodiment is specifically a third-order thermal network model. Analysis and verification show that all parameters in this third-order thermal network model are observable and converge to their unique solution. Based on the third-order thermal network model established in this invention, the temperature estimation of a permanent magnet synchronous motor can be accurately and reliably achieved without relying on the motor's geometric parameters and material properties. The observability and convergence of the parameters in this third-order thermal network model are verified and analyzed below.

[0081] The state equation of the third-order thermal network model established in this invention can be expressed as shown in formula (3).

[0082]

[0083] The matrices and vectors in formula (3) are shown in formula (4).

[0084]

[0085] The following describes the convergence characteristics of the internal parameters of the thermal network model established in this invention. The motor's operating states are divided into two types: extremely low-speed operating state and full-condition operating state.

[0086] At extremely low speeds, the motor losses are mainly concentrated in the copper losses at the winding nodes, which can be calculated using formula (5). At this time, the stator losses and rotor losses can be considered as zero. Meanwhile, since the rotor speed is extremely low, the air thermal resistance inside the air gap is mainly conducted, which can be calculated using formula (6).

[0087]

[0088]

[0089] In the above formula, R s,0 Represents DC resistance, θ w and θ w,0 i represents the average temperature and reference temperature of winding node w, respectively. d and i q These represent the actual d-axis and q-axis currents of the motor, respectively, and α Cu d represents the temperature coefficient of winding resistance. i The outer diameter of the rotor, d air λ represents the thickness of the air gap. air The value represents the thermal conductivity of air, and L represents the length of the motor shaft.

[0090] The remaining parameters, namely R c,s ,R s,w ,R r,am C w C s C r These are the parameters that need to be identified.

[0091] The threshold used to distinguish between extremely low-speed operation and full-condition operation can be determined based on the actual operating characteristics of the motor, ensuring that in the extremely low-speed operation state, motor losses are mainly concentrated in the copper losses at the winding nodes. Optionally, in the following embodiments, the preset threshold is 10 revolutions per minute. If the motor speed is lower than this preset threshold, the motor is determined to be in an extremely low-speed operation state; otherwise, the motor is determined to be in a full-condition operation state.

[0092] Under full operating conditions, the heat loss and thermal resistance of a motor dynamically change with varying operating conditions, unlike under extremely low-speed conditions. To accurately capture this dynamic change and verify the convergence of the model parameters, this invention first proposes a state-space equation for a lumped-parameter thermal network model of the motor under extremely low-speed operation. This equation not only encompasses the thermal characteristics of the motor under extremely low-speed conditions but also lays the foundation for further analysis of the motor's thermal characteristics under full operating conditions. Through this method, this invention can accurately simulate and analyze the changes in heat loss and thermal resistance across the entire operating range of the motor (from extremely low speed to full operating conditions), thus providing a more comprehensive and in-depth understanding of motor temperature control and thermal management. This verification of the convergence characteristics of the thermal network model parameters under different operating conditions has significant practical application value for improving motor efficiency and reliability.

[0093] To prove the convergence characteristics of the motor thermal network model parameters under different states, the state-space equations of the lumped-parameter thermal network model of the motor under extremely low-speed operation are first written:

[0094]

[0095] In the above formula, the superscript "^" indicates the identification value of the parameter. In formula (7), the winding node loss P w Thermal conductivity G between stator and rotor s,r Calculate directly using formulas (5) and (6).

[0096] The present invention derives the state space equation (8) for extremely low speed error by calculating the difference between the reference model (7) and the actual thermophysical model (3) of the motor.

[0097]

[0098] The error correlation matrix Φ(Φ) in formula (8) LowSpd ) and error vector The relevant definition is given in (9).

[0099]

[0100] By examining the matrix Φ related to the parameter error vector in (9), Low Spd. It can be observed that when the motor operates in a steady state for a long time, the motor temperature reaches a steady state, at which point Φ LS2 =0 3x3 (Third-order zero matrix), Φ LS1 It is full rank. This indicates that the lumped-parameter thermal network model effectively simplifies to a pure thermal resistance network. In this state, given the pre-calculated G... s、r and P w Values ​​of all thermal resistances (R) s,w ,R c,s ,R r,am Thermal conductivity (G) s,w G c,s G r,am The solution is unique in any steady-state time interval because equation Φ has a unique solution in this time interval. LS1 It is full rank. A key advantage of this thermal model structure is that the capacitance does not affect the steady-state temperature. Therefore, once the thermal resistance is accurately identified, the heat capacity (C) is... w C s C r It can also be identified through any transient temperature state, because Φ LS2 It is a diagonal matrix.

[0101] Once all heat capacities (C) are accurately identified using data from extremely low-speed operation... w Cs C r ) and thermal resistance (R) c,s ,R s,w Then the state equation of the full operating condition can be reformulated as shown in (10).

[0102]

[0103] The state equation for the full-operational-condition running state shown in formula (10) differs from the state equation for the extremely low-speed running state shown in formula (7) in the following ways: the winding nodes are omitted because the thermal resistance R connected to the winding at this time is... s,w and heat capacity C w The identification is completed, and the quantities are known and independent of the equations; the thermal resistance R of the stator and windings. s,r It is necessary to identify, rather than directly calculate, the losses of stator nodes and rotor nodes need to be identified.

[0104] In equation (10), by identifying the parameters under extremely low speed conditions, the conduction thermal resistance and heat capacity are fixed as known constants, and the focus of parameter identification shifts to determining the convection thermal resistance and power loss. In equation (10), P v This represents the total loss of the motor, which is equal to the sum of the losses of the windings, stator, and rotor at the equivalent nodes.

[0105] Based on the parameters associated with the winding nodes identified and fixed in advance according to the ultra-low speed operation data, and the winding copper loss in the electrical model Copper loss of windings under lumped parameter thermal network model equal The winding loss can be expressed by formula (11) according to Kirchhoff's current law.

[0106]

[0107] In formula (11), R s,ac Let f1 represent the AC resistance, and f2 represent the frequency characteristic of the resistance, which can be expressed as formula (12). Where the fitting coefficient β0... Cu,1 and β Cu,2 Fitting is required during the identification phase. n represents the actual motor speed. rated This corresponds to the rated speed of the motor being tested.

[0108]

[0109] The error vector at this time and error correlation matrix Φ(Φ FullCond. ) can be expressed as formula (13).

[0110]

[0111] In formula (13), the convective thermal resistance is characterized by its sensitivity to rotational speed, reflecting the process of heat exchange between the rotor and the environment surrounding the test bench through the air gap and stator. Therefore, these thermal resistances are characterized in the model as variables related to rotational speed, independent of current. Conversely, the power losses at the stator and rotor nodes are sensitive to both rotational speed and current. When the motor operates at different speeds under light load, the loss term Pv-Pw is relatively low, making temperature more sensitive to changes in rotational-related convective thermal resistance. By determining the global optimal solution for the convective thermal resistance at different speeds during light load operation, the ratio between stator and rotor losses and the total motor losses can be accurately distinguished. This is a key factor in differentiating stator and rotor losses within the total motor losses.

[0112] The convective thermal resistance and losses to be identified are modeled. Based on analytical calculations, the convective thermal resistance R... s,r and R r,am The simplified model can be represented as follows.

[0113]

[0114] The remaining loss (P) v -P w The ratio α of stator and rotor losses is divided into these two components. P Assuming it is a first-order polynomial function, the function depends on the motor speed and current shown in (15).

[0115]

[0116] In the formula, i s i is the motor operating current. rated This corresponds to the rated current of the motor being tested. Where β... P,00 ,β P,10 ,β P,01 ,β P,11 The fitting coefficients for the corresponding terms can be obtained through parameter identification.

[0117] Total motor loss P v Polynomial functions are also used for modeling, as shown in (16).

[0118]

[0119] In the formula, k v,i ,k v,n ,k v,in represents the fitting coefficient for the corresponding term.

[0120] Based on the above analysis, the thermal network model for permanent magnet motors established in this invention has parameters to be identified under different operating conditions, as shown in Table 1.

[0121] This represents the parameters to be identified under different operating conditions.

[0122]

[0123] Under different conditions, a cost function is designed to characterize the error between the actual temperature of the nodes in the thermal network model and the temperature calculated based on the parameters. The parameter identification can be completed by optimizing and solving the cost function.

[0124] It should be noted that, Figure 2 The thermal network model shown is only a preferred embodiment of this invention and should not be construed as the only limitation of the invention. In other embodiments of the invention, existing thermal network models based on the internal thermal characteristics of permanent magnet synchronous motors and capable of estimating the core node temperature, or thermal network models that complete parameter identification in other ways, can all be used in the invention.

[0125] In this embodiment, the deep learning model is used to learn the intrinsic mapping relationship between input and output. In practical applications, the specific structure of the deep learning model can be flexibly selected. As a preferred implementation, a Long Short-Term Memory (LSTM) network is selected to implement the deep learning model. The structural features of the LSTM neural network include block input, forget gate, and output gate. These components can integrate information from the current and previous time steps, enabling better learning of the mapping relationship between (motor state data, core node temperature estimation results) and the final temperature monitoring result, further improving the accuracy of temperature monitoring. The complete algorithm of the LSTM network is described in detail below:

[0126] c′ t =φ(W cx x t +W ch h t-1 +b c )

[0127] f t =σ(W fx x t +W fh h t-1 +b f )

[0128] i t =σ(W ix x t +W ih h t-1 +b i )

[0129] c t =c′ t ⊙i t +ft ⊙c t-1

[0130] o t =σ(W ox x t +W oh h t-1 +b o )

[0131] h t =φ(c t )⊙o t

[0132] In the formula, W and b represent the weight matrix and bias vector of each gate, respectively. t and h t It is the input and output vector of the node in the current layer, where x t Includes the temperature and operating status output from the thermal network model. t f t i t c t o t and h t These represent the block input, forgotten input, input gate, cell, output gate, and final output of the long short-term memory neural network, respectively, representing the actual time step t.

[0133] Since this work deals with a regression problem, mean squared error (MSE) is chosen as the loss function for the training process. The function expression is as follows:

[0134]

[0135] Among them, Y i and These represent the actual temperature and the temperature prediction result output by the temperature detection model for permanent magnet synchronous motors, respectively.

[0136] During training, the weight matrix W and bias vector b are adjusted to minimize the mean square error, thereby optimizing the network parameters until the error between the model's output and the actual measured temperature data is minimized, thus completing the model training.

[0137] In summary, this embodiment provides a novel machine learning model architecture that, while utilizing deep learning for temperature monitoring of permanent magnet synchronous motors, improves model interpretability and reduces reliance on large amounts of motor state data, effectively enhancing the accuracy and reliability of temperature detection. Through improvements to the thermal network model, accurate and reliable temperature estimation of permanent magnet synchronous motors can be achieved without relying on motor geometry and material properties, which is beneficial for improving the training effect of the deep learning component.

[0138] Example 2:

[0139] A method for establishing a temperature monitoring model for a permanent magnet synchronous motor. This embodiment is similar to Embodiment 1 above, except that, in order to perform more comprehensive temperature monitoring of the permanent magnet synchronous motor, this embodiment, while monitoring the temperature of the core nodes, also monitors temperature nodes that may be ignored by the thermal network model, such as bearings, end windings, different slot positions, and key areas like the motor end cover, as extended temperature nodes. Specifically, in this embodiment, the full-condition operating data also includes: the extended node temperatures corresponding to the motor status data;

[0140] Furthermore, the structure of the predicted output model is as follows: Figure 3 As shown, it specifically includes two branches, which are as follows:

[0141] The correction branch takes the core node temperature prediction results output by the thermal network model and the corresponding motor state data as input, and is used to correct the errors in the core node temperature prediction results, and outputs the corrected core node temperature.

[0142] And an extended node temperature prediction branch, which takes the core node temperature prediction results output by the thermal network model and the corresponding motor state data as input, to predict the temperature of the extended nodes of the permanent magnet synchronous motor.

[0143] In practical applications, the extended nodes can be set according to observation needs. Optionally, in this embodiment, the extended nodes specifically include: bearings, end windings, different slot positions, and motor end covers.

[0144] In this embodiment, both the correction branch and the extended node temperature prediction branch are long short-term memory networks.

[0145] In this embodiment, the specific implementation of the remaining steps can be referred to the description in Embodiment 1 above, and will not be repeated here.

[0146] Example 3:

[0147] A method for establishing a temperature monitoring model for a permanent magnet synchronous motor includes:

[0148] Construct a full-condition operation dataset for permanent magnet synchronous motors; the full-condition dataset includes motor status data and corresponding core node temperatures of the permanent magnet synchronous motor under all operating conditions; the core nodes include the stator, windings and rotor;

[0149] Constructing prediction networks; such as Figure 4 As shown, the prediction network includes: a calibrated permanent magnet synchronous motor thermal network model and a prediction output model;

[0150] A prediction network is constructed, comprising: a calibrated thermal network model of a permanent magnet synchronous motor and a prediction output model; the thermal network model takes motor state data as input to estimate the temperature of the core node; the prediction output model is a deep learning model, including an error prediction module and a residual network link; the error prediction module takes the core node temperature estimation result output by the thermal network model and the corresponding motor state data as input to predict the error in the core node temperature estimation result; the residual network link is used to link the output of the thermal network model to the output of the error prediction module to correct the error in the core node temperature estimation result and output the corrected core node temperature.

[0151] The prediction network is trained using a full-condition operating status dataset to update the parameters of the prediction output model. After training, the trained prediction network is output as a temperature monitoring model for permanent magnet synchronous motors.

[0152] This embodiment is similar to Embodiment 1 above, except that a residual network link is introduced between the outputs of the thermal network model and the deep learning model. This allows the long short-term memory neural network to focus on learning the residual parts that the thermal model fails to cover, rather than reconstructing the entire thermophysical process. This method is particularly suitable for situations where the training dataset is limited, thereby improving the ensemble effect of the model and achieving better prediction results with limited training data.

[0153] Similarly, in this embodiment, the thermal network model is Figure 2 The third-order heat network model shown can be established using the method described in Example 1 above. It should also be noted that... Figure 2 The thermal network model shown is only a preferred embodiment of this invention and should not be construed as the only limitation of the invention. In other embodiments of the invention, existing thermal network models based on the internal thermal characteristics of permanent magnet synchronous motors and capable of estimating the core node temperature, or thermal network models that complete parameter identification in other ways, can all be used in the invention.

[0154] In this embodiment, the error prediction module in the prediction data model is specifically a long short-term memory network.

[0155] Example 4:

[0156] A method for establishing a temperature monitoring model for a permanent magnet synchronous motor. This embodiment is similar to Embodiment 3 above, except that, in order to perform more comprehensive temperature monitoring of the permanent magnet synchronous motor, this embodiment, while monitoring the temperature of the core nodes, also monitors temperature nodes that may be ignored by the thermal network model, such as bearings, end windings, different slot positions, and key areas like the motor end cover, as extended temperature nodes. Specifically, in this embodiment, the full-condition operating data also includes: the extended node temperatures corresponding to the motor status data;

[0157] Furthermore, the predicted output model is as follows: Figure 5 As shown, in addition to the error prediction module and residual network connection in Example 1, it also includes:

[0158] The extended node temperature prediction module takes the core node temperature prediction results output by the thermal network model and the corresponding motor state data as input to predict the temperature of the extended nodes of the permanent magnet synchronous motor. Optionally, in this embodiment, the extended node temperature prediction module is also a long short-term memory network.

[0159] In practical applications, the extended nodes can be set according to observation needs. Optionally, in this embodiment, the extended nodes specifically include: bearings, end windings, different slot positions, and motor end covers.

[0160] In this embodiment, the specific implementation of the remaining steps can be referred to the description in Embodiment 1 above, and will not be repeated here.

[0161] Example 5:

[0162] A method for monitoring the temperature of a permanent magnet synchronous motor, comprising:

[0163] Real-time operating status data of the permanent magnet synchronous motor is acquired and input into the temperature monitoring model for the permanent magnet synchronous motor established by the temperature monitoring model establishment method for the permanent magnet synchronous motor provided in Embodiments 1 to 4 above, and the output of the temperature monitoring model for the permanent magnet synchronous motor is collected to complete temperature monitoring.

[0164] Example 6:

[0165] A computer-readable storage medium includes a stored computer program; when the computer program is executed by a processor, it controls the device where the computer-readable storage medium is located to execute the temperature monitoring model establishment method for permanent magnet synchronous motors provided in any one of embodiments 1 to 4 above, and / or the temperature monitoring method for permanent magnet synchronous motors provided in embodiment 5 above.

[0166] The following analysis further verifies the beneficial effects of the technical solution provided by this invention based on specific temperature monitoring results of a permanent magnet synchronous motor. Using the temperature monitoring model for the permanent magnet synchronous motor established in Example 4 above, motor status data from a 60kW experimental prototype was input into the model, and its output was collected to obtain temperature monitoring results including the core node temperature and the extended node temperature. The motor status data is as follows: Figure 6 As shown, the temperature monitoring results of each core node are as follows: Figure 7 The temperature monitoring results of each extended node are shown below. Figure 8 As shown. According to Figure 7 and Figure 8 As shown in the temperature observation results, the temperature monitoring method provided by this invention has a temperature observation error of no more than 5°C for both the core node and the extended node, demonstrating high monitoring accuracy.

[0167] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for establishing a temperature monitoring model for a permanent magnet synchronous motor, characterized in that, include: Construct a full-condition operation dataset for permanent magnet synchronous motors; the full-condition operation dataset includes motor status data and corresponding core node temperatures of the permanent magnet synchronous motor under all operating conditions; The core nodes include the stator, windings, and rotor; A prediction network is constructed; the prediction network includes: a calibrated permanent magnet synchronous motor thermal network model and a prediction output model; the thermal network model takes motor state data as input and is used to estimate the temperature of the core node; the prediction output model is a deep learning model, which takes the core node temperature estimation result output by the thermal network model and the corresponding motor state data as input, and is used to correct the error in the core node temperature estimation result and output the corrected core node temperature. The prediction network is trained using the full-condition operating status dataset to update the parameters of the prediction output model. After training, the trained prediction network is output as a temperature monitoring model for permanent magnet synchronous motors. The thermal network model is established in the following way: The casing, stator, windings, rotor, and external environment of the permanent magnet synchronous motor are represented as nodes in the thermal network model, and are respectively denoted as nodes. c , s , w , r and am ; At the node c and s The equivalent thermal resistance between the stator and the housing is added between them. R c,s At the node s and w The equivalent thermal resistance between the stator and the winding is added between them. R s,w At the node s and r Adding the equivalent convective thermal resistance between the stator and the rotor R s,r At the node r and am The equivalent convective thermal resistance between the rotor and the external environment is added between them. R r,am ; The equivalent heat capacity of the stator is added between node s and the zero-degree equipotential surface. C s and loss P s The equivalent heat capacity of the winding is added between node w and the zero-degree equipotential surface. C w and loss P w At the node r The equivalent heat capacity of the rotor is added between the rotor and the zero-degree equipotential surface. C r and loss P r ; At the outer shell node c A first heat source is added between the zero-degree equipotential surface to provide c Temperature difference, and at external environmental nodes am A second heat source is added between the zero-degree equipotential surface and the surface to provide heat. am The temperature difference is used to establish the thermal network model. in, c and am These represent the temperatures of the outer casing and the external environment, respectively.

2. The method for establishing a temperature monitoring model for a permanent magnet synchronous motor as described in claim 1, characterized in that, The full-condition operating data also includes: the temperature of the extended node corresponding to the motor status data; Furthermore, the prediction output model includes: The correction branch takes the core node temperature prediction result output by the thermal network model and the corresponding motor status data as input, and is used to correct the error in the core node temperature prediction result, and outputs the corrected core node temperature. And an extended node temperature prediction branch, which takes the core node temperature prediction results output by the thermal network model and the corresponding motor status data as input, to predict the temperature of the extended nodes of the permanent magnet synchronous motor. The extended node includes at least one of the following: bearing, end winding, different slot positions, and motor end cover.

3. The method for establishing a temperature monitoring model for a permanent magnet synchronous motor as described in claim 2, characterized in that, Both the correction branch and the extended node temperature prediction branch are long short-term memory networks.

4. A method for establishing a temperature monitoring model for a permanent magnet synchronous motor, characterized in that, include: Construct a full-condition operation dataset for permanent magnet synchronous motors; the full-condition operation dataset includes motor status data and corresponding core node temperatures of the permanent magnet synchronous motor under all operating conditions; The core nodes include the stator, windings, and rotor; Construct a prediction network; the prediction network includes: a calibrated permanent magnet synchronous motor thermal network model, and a prediction output model; A prediction network is constructed, comprising: a calibrated thermal network model of a permanent magnet synchronous motor and a prediction output model; the thermal network model takes motor state data as input to estimate the temperature of the core node; the prediction output model is a deep learning model, including an error prediction module and a residual network link; the error prediction module takes the core node temperature estimation result output by the thermal network model and the corresponding motor state data as input to predict the error in the core node temperature estimation result; the residual network link is used to link the output of the thermal network model to the output of the error prediction module to correct the error in the core node temperature estimation result and output the corrected core node temperature; The prediction network is trained using the full-condition operating status dataset to update the parameters of the prediction output model. After training, the trained prediction network is output as a temperature monitoring model for permanent magnet synchronous motors. The thermal network model is established in the following way: The casing, stator, windings, rotor, and external environment of the permanent magnet synchronous motor are represented as nodes in the thermal network model, and are respectively denoted as nodes. c , s , w , r and am ; At the node c and s The equivalent conductive thermal resistance between the stator and the outer casing is added between them. R c,s At the node s and w The equivalent thermal resistance between the stator and the winding is added between them. R s,w At the node s and r Adding the equivalent convective thermal resistance between the stator and the rotor R s,r At the node r and am The equivalent convective thermal resistance between the rotor and the external environment is added between them. R r,am ; The equivalent heat capacity of the stator is added between node s and the zero-degree equipotential surface. C s and loss P s The equivalent heat capacity of the winding is added between node w and the zero-degree equipotential surface. C w and loss P w At the node r The equivalent heat capacity of the rotor is added between the rotor and the zero-degree equipotential surface. C r and loss P r ; At the outer shell node c A first heat source is added between the zero-degree equipotential surface to provide c Temperature difference, and at external environmental nodes am A second heat source is added between the zero-degree equipotential surface and the surface to provide heat. am The temperature difference is used to establish the thermal network model. in, c and am These represent the temperatures of the outer casing and the external environment, respectively.

5. The method for establishing a temperature monitoring model for a permanent magnet synchronous motor as described in claim 4, characterized in that, The full-condition operating data also includes: the temperature of the extended node corresponding to the motor status data; Furthermore, the prediction output model also includes: The extended node temperature prediction module takes the core node temperature prediction results output by the thermal network model and the corresponding motor status data as input to predict the temperature of the extended nodes of the permanent magnet synchronous motor. The extended node includes at least one of the following: bearing, end winding, different slot positions, and motor end cover.

6. The method for establishing a temperature prediction model for a permanent magnet synchronous motor as described in claim 5, characterized in that, Both the error prediction module and the extended node temperature prediction module are long short-term memory networks.

7. A method for monitoring the temperature of a permanent magnet synchronous motor, characterized in that, include: Real-time operating status data of the permanent magnet synchronous motor is acquired and input into the temperature monitoring model for the permanent magnet synchronous motor established by the method for establishing a temperature monitoring model for the permanent magnet synchronous motor according to any one of claims 1 to 6, and the output of the temperature monitoring model for the permanent magnet synchronous motor is collected to complete temperature monitoring.

8. A computer-readable storage medium, characterized in that, Includes a stored computer program; when the computer program is executed by a processor, it controls the device containing the computer-readable storage medium to execute the temperature monitoring model establishment method for permanent magnet synchronous motors according to any one of claims 1 to 6, and / or the temperature monitoring method for permanent magnet synchronous motors according to claim 7.

Images