Method for estimating temperature of drive motor based on machine learning and double kalman filter

Abstract

The application discloses a driving motor temperature estimation method based on machine learning and double Kalman filtering and relates to the technical field of motor temperature monitoring and control. The application comprises the following steps: constructing a finite element simulation model, collecting motor loss data and temperature response data under multiple working conditions through finite element simulation, and constructing an offline parameter database through a physical information machine learning method; based on the offline parameter database, a second-order lumped parameter thermal network model is constructed, and the second-order lumped parameter thermal network model is converted into a linear parameter variation state space model. The application switches parameters in real time through an LPV model, dynamically corrects model errors in combination with a double Kalman filter, overcomes parameter drift caused by temperature and rotating speed, and solves the problem that the precision of an offline modeling open-loop estimator is insufficient and accurate real-time estimation under all working conditions cannot be realized due to modeling errors, time-varying thermal parameters and other reasons in the traditional rotor temperature estimation method under wide-range complex operating conditions of an electric vehicle.

Description

Technical Field

[0001] This invention relates to the field of motor temperature monitoring and control technology, specifically a method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering. Background Technology

[0002] With the rapid development of transportation electrification, the drive motor, as a core component of electric vehicles, is of paramount importance in terms of operational reliability and safety. Temperature is a key indicator reflecting the thermal state of the motor, directly affecting its efficiency, lifespan, and operational safety. Real-time acquisition of its transient temperature characteristics is crucial to ensuring stable motor operation; therefore, efficient temperature monitoring technology is particularly important for drive motors.

[0003] Obtaining rotor temperature by embedding sensors increases production costs and raises process requirements. Furthermore, the rotor temperature field is difficult to describe by sensors with a point-based measurement range. Therefore, sensorless estimation methods have become mainstream. However, existing sensorless methods have several problems: First, modeling relies heavily on prior knowledge; parameter definitions and heat source determination depend on global information about the motor structure, material properties, and cooling system. Second, motor parameters vary significantly across different models and operating conditions, making model transfer difficult. Moreover, thermal parameters are time-varying; thermal conductivity and heat capacity change dynamically with speed, current, and temperature, easily leading to model errors. More importantly, these methods generally rely on large amounts of measured motor data, while obtaining full-condition data in bench testing is costly and complex, especially in the complex operating environment of electric vehicles, making it difficult to cover all conditions. Simultaneously, model versatility is poor; offline-trained models struggle to respond in real-time to parameter mutations or drifts, resulting in decreased estimation accuracy.

[0004] To address this, the present invention proposes a method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering. Summary of the Invention

[0005] The purpose of this invention is to provide a drive motor temperature estimation method based on machine learning and dual Kalman filtering. This method can solve the problems of insufficient accuracy of the open-loop estimator in offline modeling under the wide range of complex operating conditions of electric vehicles due to modeling errors and time-varying thermal parameters, which makes it impossible to achieve accurate real-time estimation under all operating conditions. It also addresses the problems of traditional methods relying on a large amount of bench test data, which is difficult and costly to acquire.

[0006] According to a first aspect of the present invention, in order to achieve the above-mentioned objective, the present invention provides the following technical solution: a method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering, comprising the following steps:

[0007] S1. Construct a finite element simulation model, collect motor loss data and temperature response data under multiple working conditions through finite element simulation, and construct an offline parameter database;

[0008] S2. Based on the offline parameter database, construct a second-order lumped parameter thermal network model, and transform the second-order lumped parameter thermal network model into a linear parameter variation state space model to obtain a linear variable parameter thermal network model;

[0009] S3. Receive the actual motor current, speed and oil temperature data parameters, input the current, speed and oil temperature data parameters into the linear variable parameter-thermal network model to perform online temperature estimation of the motor, and construct a dual Kalman filter based on the linear variable parameter-thermal network model. The dual Kalman filter performs closed-loop adjustment of parameters and state during the temperature estimation process to optimize the online temperature estimation value.

[0010] Furthermore, a finite element simulation model was constructed, and motor loss data and temperature response data under multiple operating conditions were collected through finite element simulation. An offline parameter database was then built using physical information machine learning methods, as detailed below:

[0011] S11. Based on the nominal parameters of the experimental motor, a finite element model of the oil-cooled induction motor was constructed in the simulation software Motor-CAD. At the same time, the geometric dimensions were normalized with the stator outer diameter as the reference.

[0012] S12. Select multiple operating conditions with speeds ranging from 0-15000 rpm and speed intervals of 1000 rpm, and stator currents ranging from 0-120 A and current intervals of 10 A, and collect two types of data: loss data and temperature data.

[0013] Loss data acquisition: Stator losses under different operating conditions are obtained through magnetic-thermal coupling simulation. and rotor losses It covers copper loss, iron loss, and mechanical loss;

[0014] Temperature data acquisition: Stator temperature data for 3000 seconds were collected under typical operating conditions at low speed (1000 rpm), medium speed (6000 rpm), and high speed (10000 rpm). Rotor temperature and coolant temperature Data was collected at 0.1s intervals, and the collected dataset was used for thermal network parameter identification.

[0015] S13. Construct a loss model based on the data collected in step S12, and learn the coefficients of the loss model through a fully connected neural network. The specific inputs are rotational speed and current, and the output is the stator or rotor loss calculation coefficients.

[0016] Furthermore, the loss model constructed in step S13 learns the coefficients of the loss model through a fully connected neural network, transforming the loss into a function of observable variables, as follows:

[0017] S131. Establish the drive motor loss formula and learn the coefficients through a fully connected neural network. The loss model is transformed into the following function using parametric modeling:

[0018]

[0019] In the formula, Where is the stator current, and n is a coefficient related to the current and rotational speed. It is an allocation coefficient, and its constraint is: and determine bearing loss. and wind resistance loss ; It is the stator copper loss coefficient, used to characterize the proportional relationship between stator copper loss and the square of current; It is a speed-related index of stator copper loss; The rotor copper loss coefficient represents the proportional relationship between rotor copper loss and stator copper loss. This is the basic coefficient for iron loss, used to characterize the overall magnitude of total iron loss; The current-related index of iron loss reflects the degree of influence of stator current on total iron loss; It is the speed-related index of iron loss, reflecting the degree of influence of speed on total iron loss; This is the iron loss allocation factor, used to distribute the total iron loss between the stator and rotor, satisfying 0. <k7<1; is the bearing loss coefficient in mechanical loss, which characterizes the component of mechanical loss that is related to the first power of the rotational speed. The drag loss coefficient in mechanical losses represents the component of mechanical losses that is related to the cube of the rotational speed.

[0020] Final stator total loss for:

[0021]

[0022] Total rotor loss for

[0023]

[0024] In the formula, For mechanical wear, For stator copper loss, For rotor copper losses, For stator iron loss, For stator iron loss;

[0025] S132. The training process uses the L-BFGS optimization algorithm, and the loss function is... L 1 is:

[0026]

[0027] In the formula, The stator loss prediction values ​​for the training samples. The stator loss simulation values ​​for the training samples. The rotor loss prediction values ​​for the training samples. The simulated rotor loss values ​​are for the training samples;

[0028] Next, thermal parameters were identified. Based on temperature simulation data, the L-BFGS algorithm was used to identify the thermal parameters of the lumped parameter thermal network model. The input is the current temperature status. The output is the predicted temperature value for the next time step. loss function L 2 is:

[0029]

[0030] In the formula, , For the heat capacity of the stator and rotor, , , Thermal conductivity; T i+1 represents the temperature state at the next moment.

[0031] Furthermore, the state equations of the second-order lumped-parameter thermal network model are as follows:

[0032]

[0033]

[0034] In the formula , For the heat capacity of the stator and rotor, , , Thermal conductivity; , For stator / rotor losses, , , These are the stator, rotor, and cooling oil temperatures, respectively.

[0035] Furthermore, the second-order lumped-parameter thermal network model is transformed into a linear parameter-varying state-space model, as follows:

[0036] First, define the time-varying parameter vector. for:

[0037]

[0038] in:

[0039]

[0040] in, The stator temperature change rate, The rotor temperature change rate, Let be the state derivative vector. For state vectors, For the input vector, The state matrix, The input matrix; , For the heat capacity of the stator and rotor, , , Thermal conductivity; , For stator / rotor losses, , , These are the stator, rotor, and cooling oil temperatures, respectively.

[0041] Furthermore, the dual Kalman filter includes a state Kalman filter and a parameter Kalman filter. Updating the state of the dual Kalman filter enables closed-loop adjustment of parameters and state, which is used to optimize the online temperature estimate, as detailed below:

[0042] The state update equation is established as follows:

[0043]

[0044]

[0045]

[0046] Then, the parameter update equations are established as follows:

[0047]

[0048]

[0049]

[0050] In the formula, : Represents the discrete time step; The current measurement output, i.e., the actual measured stator winding temperature; : Measurement output matrix, used to extract stator temperature from state vector; The covariance matrix of the measurement noise characterizes the statistical properties of the noise during the measurement process; : A 2×2 identity matrix, used for identity transformation in matrix operations; T represents matrix transpose;

[0051] : The predicted value of the current state vector, including the predicted values ​​of stator temperature and rotor temperature; The updated value of the state vector at the current moment; : The covariance matrix of the current state prediction error; : The covariance matrix of the current state update error; Kalman gain for state updates, used to weigh the weights of predicted and measured values; : The partial derivative matrix of the state with respect to the measurement output;

[0052] The predicted value of the model parameter vector at the current moment; The updated value of the model parameter vector at the current moment; : The covariance matrix of the parameter prediction error at the current time; : The covariance matrix of the parameter update error at the current moment; Kalman gain for parameter updates is used to balance the weights of predicted and measured parameter values, minimizing parameter update errors. : The partial derivative matrix of the parameters with respect to the measurement output.

[0053] According to a second aspect of the present invention, the present invention provides a machine learning and dual Kalman filter-based induction motor rotor temperature estimation system for implementing the drive motor temperature estimation method based on machine learning and dual Kalman filtering described in the first aspect, comprising:

[0054] The database construction module is used to build finite element simulation models. It collects motor loss data and temperature response data under multiple operating conditions through finite element simulation and builds an offline parameter database.

[0055] The model building module is used to construct a second-order lumped parameter thermal network model based on an offline parameter database, and to transform the second-order lumped parameter thermal network model into a linear parameter variation state space model to obtain a linear variable parameter thermal network model. Based on the rotational speed, the module divides the range and assigns corresponding linear parameter variation state space model parameters to each range to achieve adaptive switching under all operating conditions.

[0056] The real-time estimation module receives actual motor current, speed, and oil temperature data parameters, inputs these parameters into a linear variable parameter-thermal network model for online motor temperature estimation, and constructs a dual Kalman filter based on the linear variable parameter-thermal network model. The dual Kalman filter performs closed-loop adjustment of parameters and states during the temperature estimation process to optimize the online temperature estimate.

[0057] According to a third aspect of the present invention, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor loads and executes the computer program, it employs the drive motor temperature estimation method based on machine learning and dual Kalman filtering described in the first aspect.

[0058] The present invention has at least the following beneficial effects:

[0059] 1. This invention overcomes parameter drift caused by temperature and speed by switching parameters in real time using the LPV model and dynamically correcting model errors using a dual Kalman filter. It solves the problem that traditional rotor temperature estimation methods, under the wide range of complex operating conditions of electric vehicles, suffer from insufficient accuracy of the open-loop estimator in offline modeling due to modeling errors and time-varying thermal parameters, thus failing to achieve accurate real-time estimation under all operating conditions.

[0060] 2. The present invention is based on a modeling strategy that utilizes finite element analysis (FEA) simulation data and machine learning based on physical information. This strategy avoids the complex physical modeling process involving expert knowledge and eliminates the dependence on experimental data. It solves the problems of traditional methods relying on a large amount of bench test data, which makes data acquisition difficult and costly.

[0061] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating the estimation method described in this invention;

[0063] Figure 2 This is a schematic diagram illustrating the framework of the estimation method described in this invention.

[0064] Figure 3 This refers to the rotor temperature estimation results and errors in this invention. Detailed Implementation

[0065] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0066] Example 1:

[0067] Please see Figures 1-3 This invention provides a technical solution: a method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering, comprising the following steps:

[0068] Step 1: Establish a simulation model of the oil-cooled induction motor based on the finite element analysis (FEA) software Motor-CA. Input the motor's geometric parameters (stator / rotor outer diameter, number of slots, winding configuration, etc.) and cooling system parameters (oil cooling method, inlet oil temperature, etc.), simulate multiple operating conditions, and collect motor loss data and temperature response data. The motor loss data includes stator loss and rotor loss, covering copper loss, iron loss and mechanical loss, including simulation data of stator temperature, rotor temperature and cooling oil temperature.

[0069] Based on the nominal parameters of the experimental motor, a finite element model of an oil-cooled induction motor was constructed in the simulation software Motor-CAD. At the same time, the geometric dimensions were normalized with the stator outer diameter as the reference. A total of 180 operating conditions were selected, with a speed of 0-15000rpm and a speed interval of 1000rpm, and a stator current of 0-120A and a current interval of 10A. Two types of data were collected: loss data and temperature data.

[0070] First, loss data was collected: stator losses under different operating conditions were obtained through magnetic-thermal coupling simulation. and rotor losses It covers copper loss, iron loss, and mechanical loss;

[0071] Next, temperature data was collected: stator temperature data for 3000 seconds were selected under typical operating conditions at low speed (1000 rpm), medium speed (6000 rpm), and high speed (10000 rpm). Rotor temperature and coolant temperature Data was collected at 0.1s intervals, and the collected dataset was used for thermal network parameter identification.

[0072] A loss model is constructed based on the collected motor loss data. A physical information machine learning approach is used to build an offline parameter database. Fully connected neural networks (FCNN) are then used to learn the loss model parameters. The inputs to the loss model are rotational speed and current, and the outputs are the stator / rotor loss calculation coefficients, as detailed below:

[0073] (1) Establish the loss formula of the drive motor and learn the coefficients through a fully connected neural network. The loss model is transformed into the following function using parametric modeling:

[0074]

[0075] In the formula, Where is the stator current, and n is a coefficient related to the current and rotational speed. It is an allocation coefficient, and its constraint is: and determine bearing loss. and wind resistance loss ; It is the stator copper loss coefficient, used to characterize the proportional relationship between stator copper loss and the square of current; It is a speed-related index of stator copper loss; The rotor copper loss coefficient represents the proportional relationship between rotor copper loss and stator copper loss. This is the basic coefficient for iron loss, used to characterize the overall magnitude of total iron loss; The current-related index of iron loss reflects the degree of influence of stator current on total iron loss; It is the speed-related index of iron loss, reflecting the degree of influence of speed on total iron loss; This is the iron loss allocation factor, used to distribute the total iron loss between the stator and rotor, satisfying 0. <k7<1; is the bearing loss coefficient in mechanical loss, which characterizes the component of mechanical loss that is related to the first power of the rotational speed. The drag loss coefficient in mechanical losses represents the component of mechanical losses that is related to the cube of the rotational speed.

[0076] Final stator total loss for:

[0077]

[0078] Total rotor loss for

[0079]

[0080] In the formula, For mechanical wear, For stator copper loss, For rotor copper losses, For stator iron loss, For stator iron loss;

[0081] (2) The training process uses the L-BFGS optimization algorithm, and the loss function is:

[0082]

[0083] In the formula, The stator loss prediction values ​​for the training samples. The stator loss simulation values ​​for the training samples. The rotor loss prediction values ​​for the training samples. The simulated rotor loss values ​​are for the training samples;

[0084] Next, thermal parameters were identified. Based on temperature simulation data, the L-BFGS algorithm was used to identify the thermal parameters of the lumped parameter thermal network model. The input is the current temperature status. The output is the predicted temperature value for the next time step. The loss function is:

[0085]

[0086] In the formula, , For the heat capacity of the stator and rotor, , , Thermal conductivity; T i+1 The temperature state at the next moment;

[0087] Step 2: Based on the offline parameter database built in Step 1, construct a second-order lumped parameter thermal network model (LPTN) to describe the thermal balance relationship between the stator and rotor. Then, transform the second-order LPTN model into a linear parameter variation (LPV) state-space model to achieve the fusion of the second-order lumped parameter thermal network model (LPTN) and the linear parameter variation (LPV) state-space model, resulting in a linear variable parameter thermal network model (LPV-LPTN model).

[0088] Based on the offline parameter database from step one, a second-order LPTN model is constructed to describe the thermal balance relationship between the stator and rotor:

[0089]

[0090]

[0091] in , These are the stator and rotor heat capacities, respectively. , , For thermal conductivity, , For stator / rotor losses, , , These are the stator, rotor, and cooling oil temperatures, respectively.

[0092] Identifying thermal parameters of lumped-parameter thermal network models using the L-BFGS algorithm The input is the current temperature status. The output is the predicted temperature value for the next time step. This is used to train a second-order lumped-parameter hot network model until the loss function is minimized. The specific loss function is:

[0093]

[0094] In the formula, , For the heat capacity of the stator and rotor, , , Thermal conductivity;

[0095] Furthermore, the second-order LPTN model is transformed into a linear parameter variation (LPV) state-space model, as follows:

[0096] First, define the time-varying parameter vector. for

[0097]

[0098] in:

[0099]

[0100] in, The stator temperature change rate, The rotor temperature change rate, Let be the state derivative vector. For state vectors, For the input vector, The state matrix, The input matrix;

[0101] Step 3: Receive the actual motor current, speed, and oil temperature data parameters, and input these data parameters into the linear variable parameter-thermal network model. Then, design a dual Kalman filter (DKF). Through the DKF iteration process, update the state to achieve closed-loop regulation of parameters and state, and finally realize online temperature estimation of the drive motor.

[0102] The dual Kalman filter comprises a state Kalman filter and a parametric Kalman filter, which are designed collaboratively based on the state-space equations of the linear variable parametric-thermal network model.

[0103] Among them, the state Kalman filter is used to refine the linear variable parameter-thermal network model with the latest updated thermal parameters and estimate the motor's state variables, namely the rotor and stator temperatures, in real time.

[0104] The parametric Kalman filter is used to identify and update thermal parameters in real time based on the error between the state estimate and the actual oil temperature measurement, in order to compensate for the time-varying model parameters caused by changes in operating conditions.

[0105] First, the state update equation is established as follows:

[0106]

[0107] Then, the parameter update equation is established as follows:

[0108]

[0109] In the formula, : Represents the discrete time step; : The measured output at the current moment (i.e., the actual measured stator winding temperature); The measurement output matrix is ​​used to extract the stator temperature from the state vector (since the state vector x contains the stator temperature). and rotor temperature ); The covariance matrix of the measurement noise characterizes the statistical properties of the noise during the measurement process; : A 2×2 identity matrix, used for identity transformation in matrix operations; T represents matrix transpose;

[0110] Specifically, for the parameters of the state update equation, : The predicted value of the current state vector (including the predicted values ​​of stator temperature and rotor temperature). : The updated value of the state vector at the current moment (based on the stator temperature and rotor temperature after measurement correction); : Covariance matrix of the current state prediction error (error statistics during the prediction phase); : Covariance matrix of the current state update error (corrected error statistics); The Kalman gain for state updates is used to weigh the predictions against the measurements in order to minimize the state update error. : The partial derivative matrix of the state with respect to the measurement output;

[0111] Unique to the parameter update equation parameters, The predicted value of the model parameter vector at the current moment; : The updated value of the model parameter vector at the current moment (based on the thermal parameters after measurement correction); : Covariance matrix of parameter prediction error at the current moment (statistics of parameter error during the prediction phase); : Covariance matrix of parameter update error at the current moment (corrected parameter error statistics); Kalman gain for parameter updates is used to balance the weights of predicted and measured parameter values, minimizing parameter update errors. : The partial derivative matrix of the parameters with respect to the measurement output.

[0112] like Figure 2 As shown, in the upper-level offline modeling stage, a finite element model of an oil-cooled induction motor is constructed using Motor-CAD software. Simulation data such as loss and temperature are obtained by simulating operating conditions. Then, a fully connected neural network (input speed and current, output coefficients k1-k9, after normalization) combined with the L-BFGS algorithm is used to construct a model containing loss coefficients and thermal parameters. The offline parameter database (etc.); in the middle-level model transformation stage, the second-order lumped parameter thermal network model (second-order LPTN model) describing the stator-rotor-cooling oil thermal balance is transformed into a linear parameter-changing state-space model to adapt to time-varying parameters; in the lower-level online estimation stage, dual Kalman filtering adjusts the temperature state (T) through dual closed loops. s The system updates the state by inputting the actual effective values ​​of motor current, speed, winding temperature, and oil temperature, and then updating the state through the DKF iteration process. This achieves closed-loop regulation of parameters and state, ultimately enabling online temperature estimation of the drive motor and outputting the rotor temperature estimate, thus achieving full-condition adaptive operation.

[0113] like Figure 3 As shown, there are two sub-plots (left plot: rotor temperature curve, right plot: estimation error curve). The data shows that the estimation error is ≤5℃ under all operating conditions, which verifies the effectiveness of the method in this embodiment under dynamic operating conditions WLTC (speed 0-9800rpm, torque 0-200Nm).

[0114] Example 2:

[0115] This embodiment provides a machine learning and dual Kalman filter-based induction motor rotor temperature estimation system to implement the drive motor temperature estimation method based on machine learning and dual Kalman filtering described in the first aspect, including:

[0116] The database construction module is used to build finite element simulation models. It collects motor loss data and temperature response data under multiple operating conditions through finite element simulation and builds an offline parameter database through physical information machine learning methods.

[0117] The model building module is used to construct a second-order lumped parameter thermal network model based on an offline parameter database, and to transform the second-order lumped parameter thermal network model into a linear parameter variation state space model to obtain a linear variable parameter thermal network model. Based on the rotational speed, the module divides the range and assigns corresponding linear parameter variation state space model parameters to each range to achieve adaptive switching under all operating conditions.

[0118] The real-time estimation module receives actual motor current, speed, and oil temperature data parameters, inputs these parameters into a linear variable parameter-thermal network model for online motor temperature estimation, and constructs a dual Kalman filter based on the linear variable parameter-thermal network model. The dual Kalman filter performs closed-loop adjustment of parameters and states during the temperature estimation process to optimize the online temperature estimate.

[0119] Example 3:

[0120] The present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program capable of running on the processor. When the processor loads and executes the computer program, it adopts the drive motor temperature estimation method based on machine learning and dual Kalman filtering described in Embodiment 1.

[0121] It should be noted that the terminal device can be a computer device such as a desktop computer, a laptop computer, or a cloud server, and the terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices, and buses.

[0122] Furthermore, the processor can be a central processing unit (CPU). Of course, depending on the actual use, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be used. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it in this regard.

[0123] Example 4:

[0124] The present invention provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the drive motor temperature estimation method based on machine learning and dual Kalman filtering described in Embodiment 1.

[0125] The computer program can be stored in a computer-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain middleware. The computer-readable medium includes any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the computer-readable medium includes, but is not limited to, the above-mentioned components.

[0126] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0127] For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on," "mounted on," "fixed to," or "set on" another element, it may be directly on the other element or there may be an intermediate element present. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element present. The terms "vertical," "horizontal," "upper," "lower," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible embodiments.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0129] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering, characterized in that, Includes the following steps: S1. Construct a finite element simulation model, collect motor loss data and temperature response data under multiple working conditions through finite element simulation, and construct an offline parameter database; S2. Based on the offline parameter database, construct a second-order lumped parameter thermal network model, and transform the second-order lumped parameter thermal network model into a linear parameter variation state space model to obtain a linear variable parameter thermal network model; S3. Receive the actual motor current, speed and oil temperature data parameters, input the current, speed and oil temperature data parameters into the linear variable parameter-thermal network model to perform online temperature estimation of the motor, and construct a dual Kalman filter based on the linear variable parameter-thermal network model. The dual Kalman filter performs closed-loop adjustment of parameters and state during the temperature estimation process to optimize the online temperature estimation value. A finite element simulation model was constructed, and motor loss and temperature response data under multiple operating conditions were collected through finite element simulation. An offline parameter database was then built using physical information machine learning methods, as detailed below: S11. Based on the nominal parameters of the experimental motor, a finite element model of the oil-cooled induction motor was constructed in the simulation software Motor-CAD. At the same time, the geometric dimensions were normalized with the stator outer diameter as the reference. S12. Select multiple operating conditions with speeds ranging from 0-15000 rpm and speed intervals of 1000 rpm, and stator currents ranging from 0-120 A and current intervals of 10 A, and collect two types of data: loss data and temperature data. Loss data acquisition: Stator losses under different operating conditions are obtained through magnetic-thermal coupling simulation. and rotor losses It covers copper loss, iron loss, and mechanical loss; Temperature data acquisition: Stator temperature data for 3000 seconds were collected under typical operating conditions at low speed (1000 rpm), medium speed (6000 rpm), and high speed (10000 rpm). Rotor temperature and coolant temperature Data was collected at 0.1s intervals, and the collected dataset was used for thermal network parameter identification. S13. Construct a loss model based on the data collected in step S12, and learn the coefficients of the loss model through a fully connected neural network. The specific inputs are rotational speed and current, and the output is the stator or rotor loss calculation coefficients. The loss model constructed in step S13 uses a fully connected neural network to learn the coefficients of the loss model, transforming the loss into a function of observable variables, as follows: Establish the loss formula for the drive motor and learn the coefficients through a fully connected neural network. The loss model is transformed into the following function using parametric modeling: In the formula, Where is the stator current, and n is a coefficient related to the current and rotational speed. It is an allocation coefficient, and its constraint is: and determine bearing loss. and wind resistance loss ; It is the stator copper loss coefficient, used to characterize the proportional relationship between stator copper loss and the square of current; It is a speed-related index of stator copper loss; The rotor copper loss coefficient represents the proportional relationship between rotor copper loss and stator copper loss. This is the basic coefficient for iron loss, used to characterize the overall magnitude of total iron loss; The current-related index of iron loss reflects the degree of influence of stator current on total iron loss; It is the speed-related index of iron loss, reflecting the degree of influence of speed on total iron loss; This is the iron loss allocation factor, used to distribute the total iron loss between the stator and rotor, satisfying 0. <k7<1; is the bearing loss coefficient in mechanical losses, representing the component of mechanical losses that is related to the first power of the rotational speed; It is the wind resistance loss coefficient in mechanical losses, representing the component of mechanical losses that is related to the cube of the rotational speed; Final stator total loss for: Total rotor loss for: In the formula, For mechanical wear, For stator copper loss, For rotor copper losses, For stator iron loss, For stator iron loss; In step S2, the training process of the second-order lumped parameter heat network model adopts the L-BFGS optimization algorithm, and the loss function is... L 1 is: In the formula, The stator loss prediction values ​​for the training samples. The stator loss simulation values ​​for the training samples. The rotor loss prediction values ​​for the training samples. The simulated rotor loss values ​​are for the training samples; Identifying thermal parameters of lumped-parameter thermal network models using the L-BFGS algorithm ; The input to the second-order lumped-parameter thermal network model is the current temperature state. The output is the predicted temperature value for the next time step. loss function L 2 is: In the formula, , These are the stator and rotor heat capacities, respectively. , , Thermal conductivity; T i+1 represents the temperature state at the next moment.

2. The method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering according to claim 1, characterized in that: The state equations of the second-order lumped-parameter thermal network model are as follows: In the formula , For the heat capacity of the stator and rotor, , , Thermal conductivity; , For stator / rotor losses, , , These are the stator, rotor, and cooling oil temperatures, respectively.

3. The method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering according to claim 2, characterized in that: The second-order lumped-parameter thermal network model is transformed into a linear parameter-varying state-space model, as follows: First, define the time-varying parameter vector. for: in: in, The stator temperature change rate, The rotor temperature change rate, Let be the state derivative vector. For state vectors, For the input vector, The state matrix, The input matrix; , For the heat capacity of the stator and rotor, , , Thermal conductivity; , For stator / rotor losses, , , These are the stator, rotor, and cooling oil temperatures, respectively.

4. The method for estimating the temperature of a drive motor based on machine learning and dual Kalman filtering according to claim 3, characterized in that: The dual Kalman filter includes a state Kalman filter and a parameter Kalman filter. The state of the dual Kalman filter is updated to achieve closed-loop adjustment of parameters and state, which is used to optimize the online temperature estimate, as detailed below: The state update equation is established as follows: Then, the parameter update equations are established as follows: In the formula, : Represents the discrete time step; The current measurement output, i.e., the actual measured stator winding temperature; : Measurement output matrix, used to extract stator temperature from state vector; The covariance matrix of the measurement noise characterizes the statistical properties of the noise during the measurement process; A 2×2 identity matrix used for identity transformation in matrix operations; T denotes matrix transpose; : The predicted value of the current state vector, including the predicted values ​​of stator temperature and rotor temperature; The updated value of the state vector at the current moment; : The covariance matrix of the current state prediction error; : The covariance matrix of the current state update error; Kalman gain for state updates, used to weigh the weights of predicted and measured values; : The partial derivative matrix of the state with respect to the measurement output; The predicted value of the model parameter vector at the current moment; The updated value of the model parameter vector at the current moment; : The covariance matrix of the parameter prediction error at the current time; : The covariance matrix of the parameter update error at the current moment; Kalman gain for parameter updates is used to balance the weights of predicted and measured parameter values, minimizing parameter update errors. : The partial derivative matrix of the parameters with respect to the measurement output.

5. A machine learning and dual Kalman filter-based induction motor rotor temperature estimation system, used to implement the drive motor temperature estimation method based on machine learning and dual Kalman filtering as described in any one of claims 1 to 4, characterized in that, include: The database construction module is used to build finite element simulation models. It collects motor loss data and temperature response data under multiple operating conditions through finite element simulation and builds an offline parameter database. The model building module is used to construct a second-order lumped parameter thermal network model based on an offline parameter database, and to transform the second-order lumped parameter thermal network model into a linear parameter variation state space model to obtain a linear variable parameter thermal network model. Based on the rotational speed, the module divides the range and assigns corresponding linear parameter variation state space model parameters to each range to achieve adaptive switching under all operating conditions. The real-time estimation module receives actual motor current, speed, and oil temperature data parameters, inputs these parameters into a linear variable parameter-thermal network model for online motor temperature estimation, and constructs a dual Kalman filter based on the linear variable parameter-thermal network model. The dual Kalman filter performs closed-loop adjustment of parameters and states during the temperature estimation process to optimize the online temperature estimate.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor loads and executes the computer program, it employs the drive motor temperature estimation method based on machine learning and dual Kalman filtering as described in any one of claims 1 to 4.

7. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the drive motor temperature estimation method based on machine learning and dual Kalman filtering as described in any one of claims 1 to 4.

Images