Speed-increase domain motor temperature estimation method and system based on data mechanism fusion

Abstract

This invention belongs to the field of motor thermal state estimation technology, and discloses a method and system for estimating the temperature of a speed-domain motor based on data mechanism fusion. The method includes the following steps: acquiring the motor's operating data under a first operating condition, where the speed is less than or equal to a first threshold and the torque is less than or equal to a second threshold; establishing a thermal network model and a loss model for the motor based on the operating data, and using the thermal network model and loss model as prior thermal estimation models. The thermal network model describes the thermal dynamic characteristics of the motor stator and rotor, and the loss model calculates the stator and rotor losses as input to the thermal network model. A model correction factor is introduced to correct the model parameters in the prior thermal estimation model. This effectively solves the problems of excessive reliance on full-speed-domain calibration data in existing technologies, and the decrease in temperature estimation accuracy due to model mismatch when applying low-speed-range prior models to unknown operating conditions.

Description

Technical Field

[0001] This invention belongs to the field of motor thermal state estimation technology, specifically relating to a method and system for estimating the temperature of extended speed domain motors based on data mechanism fusion. Background Technology

[0002] As a core component of new energy vehicles and high-performance electric drive systems, the thermal state of the drive motor directly affects system efficiency, reliability, and operational safety. With the continuous development of motors towards higher power density, internal losses and thermal loads increase significantly, making rotor overheating a particularly prominent risk. However, due to the rotating structure, directly measuring rotor temperature through embedded sensors faces severe limitations in terms of cost and stability, hindering mass production applications. Therefore, model-based sensorless temperature estimation has become the mainstream technical approach. Existing solutions largely rely on extensive bench calibration data across the entire speed domain and all operating conditions to establish accurate loss and thermal network models. However, full-speed domain calibration is not only costly and time-consuming, making it difficult to match the rapid iteration of motor products, but also presents calibration blind spots due to limitations in test safety and equipment capabilities at higher speed or torque boundary conditions. This results in reliable data readily available in engineering being concentrated in lower speed and torque operating conditions. When the model trained based on this low-load region data is extrapolated to uncalibrated high-load or high-speed unknown operating conditions, severe model mismatch easily occurs due to changes in loss mechanisms and time-varying thermal parameters, leading to a sharp decline in temperature estimation accuracy. Therefore, existing technologies suffer from problems such as excessive reliance on full-speed domain calibration data and decreased temperature estimation accuracy due to model mismatch when applying low-speed region prior models to unknown operating conditions. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for estimating the temperature of motors in the extended speed domain based on data mechanism fusion. This solves the problems of excessive reliance on full-speed domain calibration data and the decrease in temperature estimation accuracy due to model mismatch when applying low-speed prior models to unknown operating conditions.

[0004] The objective of this invention can be achieved through the following technical solutions: A method for estimating the temperature of a speed-domain motor based on data mechanism fusion includes the following steps: Acquire the motor's operating data under the first operating condition, which is the condition where the speed is less than or equal to the first threshold and the torque is less than or equal to the second threshold; A thermal network model and a loss model of the motor are established based on the operating data. The thermal network model and the loss model are used as a priori thermal estimation models. The thermal network model is used to describe the thermal dynamic characteristics of the motor stator and rotor, and the loss model is used to calculate the stator loss and rotor loss of the motor as input to the thermal network model. The prior thermal estimation model is applied to the current operating conditions, and model correction is introduced to correct the model parameters in the prior thermal estimation model in order to compensate for the model mismatch of the prior thermal estimation model under the current operating conditions. The system collects the current, speed, stator temperature, and cooling medium temperature of the motor under the current operating conditions in real time. It uses a dual extended Kalman filter to jointly update the motor's thermal state and model correction online, outputting the estimated stator temperature and rotor temperature at the current moment. The updated thermal state and model correction are then used for temperature prediction at the next moment. The thermal state includes both stator temperature and rotor temperature.

[0005] Furthermore, the first threshold is 4000 rpm, and the second threshold is 100 Nm.

[0006] Furthermore, the operating data includes motor speed. n electromagnetic torque Trq D-axis current i d Q-axis current i q stator temperature T s Rotor temperature T r Cooling medium temperature T c .

[0007] Furthermore, based on the operating data, a thermal network model and a loss model for the motor are established, specifically including the following steps: Based on the stator temperature in the operating data T s Rotor temperature T r Changes, establishing a system to describe stator temperature T s Rotor temperature T r The state equations of a second-order lumped-parameter thermal network model for thermal dynamics are as follows: In the formula, K cs , K cr , K sr These are the equivalent thermal conductance from the stator node to the cooling boundary, the equivalent thermal conductance from the rotor node to the cooling boundary, and the equivalent thermal conductance between the stator and the rotor, respectively. , These are the stator and rotor heat capacities, respectively. , These are stator and rotor losses, respectively. Based on the variation characteristics of electromagnetic torque and stator temperature in the operating data, the first operating condition is divided into multiple sub-operating condition segments, including load transient operating condition, load steady-state operating condition, unloading transient operating condition, and unloading steady-state operating condition. Based on sub-operating condition sections, an alternating identification method using thermal parameters and equivalent losses is employed, based on the stator temperature of each sub-operating condition section. T s Rotor temperature T r Cooling medium temperature T c The equivalent stator loss value and equivalent rotor loss value of each sub-operating condition segment are obtained by reverse calculation, and the equivalent stator loss value and equivalent rotor loss value are used as the loss label of the sub-operating condition segment; at the same time, the thermal parameters in the thermal network model are identified. Using the equivalent stator loss labels and equivalent rotor loss labels of each sub-operating condition segment as supervisory information, and taking the current, speed and stator temperature in the operating data as input, a loss model is trained.

[0008] Furthermore, the model corrections include , , , , ,in, and These are the correction factors for stator loss and rotor loss, respectively. , and These are correction factors for the thermal conductivity from the stator to the cooling boundary, the thermal conductivity from the rotor to the cooling boundary, and the thermal conductivity of the stator-rotor coupling, respectively. A model correction parameter is introduced to correct the model parameters in the prior heat estimation model. The corrected effective parameters satisfy the following: In the formula, P s,eff and P r,eff These are the effective stator loss and effective rotor loss under the current operating conditions. K sc,eff , K rc,eff and K sr,effThese are the corrected thermal conductance from the stator to the cooling boundary, the thermal conductance from the rotor to the cooling boundary, and the thermal conductance coupled between the stator and rotor, respectively. Substituting the model corrections into the second-order lumped-parameter thermal network model, we obtain the online corrected thermal state equation for the current operating condition: In the formula, T s,k and T r,k They are respectively k The stator temperature and rotor temperature at that moment. T c,k for k The temperature of the cooling medium at any given time, Δ t The sampling period.

[0009] Furthermore, the thermal state of the motor and the model corrections are jointly updated online using a dual extended Kalman filter, specifically including the following steps: The thermal state vector and the correction state vector are defined as follows: in, Let be the thermal state vector. Let k be the state vector of the correction value. Based on the thermal state vector and the correction state vector, the state equation, parameter equation, and observation equation are constructed for the dual extended Kalman filter, where: The state equation is constructed by incorporating model corrections into the prior heat estimation model, and is expressed as: In the formula, For state process noise, x k and x k-1 They are respectively k Time and k-1 The thermal state vector at time t, θ k for k The model correction state vector at time 1. u k-1 for k-1 The external input vector at time t; The parametric equations, expressed in random walk form, are as follows: In the formula, This is parameter process noise, reflecting the variation characteristics of the correction amount between adjacent sampling times; θ k-1 for k- 1 The model correction state vector at time t; The observation equation, with stator temperature as the sole observable, is expressed as: In the formula, y k for k Stator temperature observation at time t, H For the observation matrix, For measuring noise; The dual extended Kalman filter performs online joint updates of thermal state and model corrections based on state equations, parameter equations, and observation equations.

[0010] Furthermore, the dual extended Kalman filter includes a state filter and a parameter filter; The state filter is used to predict and update the thermal state at the current moment. Its update equation is as follows: In the formula, K k,x Let K be the Kalman gain of the state filter at time k. and These are the state prediction covariance matrix and the updated covariance matrix, respectively. R To measure the noise covariance matrix, and These are the predicted and updated values ​​for the thermal state, respectively. I It is the identity matrix; The parameter filter is used to predict and update the model correction at the current time step. Its update equation is: In the formula, For parameter filters in k Kalman gain at time step and These are the predicted covariance matrix and the updated covariance matrix of the corrected state, respectively. The linearized observation matrix of the parameter filter, and These are the predicted and updated values ​​of the model correction, respectively.

[0011] A speed-domain motor temperature estimation system based on data mechanism fusion, employing a motor temperature estimation method to achieve motor temperature estimation, including: The data acquisition module is used to acquire the motor's operating data under the first operating condition; The offline training module, connected to the data acquisition module, is used to build a priori heat estimation model based on the running data; The online correction module, connected to the offline training module, is used to apply the prior heat estimation model to the current operating conditions and introduce model correction parameters for correction. The dual Kalman filter module, connected to the online correction module, is used to jointly update the motor's thermal state and model correction values ​​online, and output stator temperature estimates and rotor temperature estimates.

[0012] The beneficial effects of this invention are: 1. This application obtains the operating data of the motor in the low-speed, low-load region where the speed and torque are not higher than the preset threshold, establishes a prior thermal estimation model including a thermal network model and a loss model, and applies the prior thermal estimation model to the current operating condition. Then, a model correction amount is introduced to correct the model parameters. At the same time, a dual extended Kalman filter is used to jointly update the thermal state composed of stator temperature and rotor temperature and the model correction amount online. The updated thermal state and model correction amount are then used for temperature prediction at the next moment. Thus, it can effectively compensate for the model mismatch under unknown operating conditions by relying only on the prior data of the low-speed, low-load region, and achieve continuous and accurate estimation of the motor thermal state under unknown operating conditions. This significantly reduces the dependence on full-speed domain full-condition bench calibration data, shortens the development cycle and reduces testing costs. 2. The method and system proposed in this application utilize an architecture that combines offline training of low-speed, low-load region data with online joint updating of dual extended Kalman filters. This effectively extrapolates readily available prior knowledge of the low-speed, low-load region to uncalibrated, unknown operating conditions such as high load or high speed, solving the model mismatch problem caused by calibration blind spots in traditional solutions. Furthermore, during online operation, only easily measurable signals such as current, speed, stator temperature, and cooling medium temperature need to be collected, eliminating the need for rotor temperature sensors and reducing hardware costs and system complexity. The closed-loop recursive mechanism of the dual extended Kalman filters enables the model correction to smoothly track changes in operating conditions, avoiding abrupt changes in estimated values ​​and improving the stability, accuracy, and engineering practicality of motor temperature estimation under complex operating conditions across a wide speed range. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the overall process of the motor temperature estimation method of the present invention; Figure 2 This is a schematic diagram of the overall structure of the motor temperature estimation system of the present invention; Figure 3 This is a schematic diagram comparing the stator temperature estimation results with the measured results in an embodiment of the present invention; Figure 4 This is a schematic diagram comparing the rotor temperature estimation result with the measured result in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] like Figure 1 As shown, the extended speed domain motor temperature estimation method based on data mechanism fusion includes the following steps: Acquire the motor's operating data under the first operating condition, which is the condition where the speed is less than or equal to the first threshold and the torque is less than or equal to the second threshold; A thermal network model and a loss model of the motor are established based on the operating data. The thermal network model and the loss model are used as a priori thermal estimation models. The thermal network model is used to describe the thermal dynamic characteristics of the motor stator and rotor, and the loss model is used to calculate the stator loss and rotor loss of the motor as input to the thermal network model. The prior thermal estimation model is applied to the current operating conditions, and model correction is introduced to correct the model parameters in the prior thermal estimation model in order to compensate for the model mismatch of the prior thermal estimation model under the current operating conditions. The current, speed, stator temperature, and cooling medium temperature of the motor are collected in real time under the current operating conditions. The thermal state and model correction of the motor are jointly updated online using a dual extended Kalman filter. The estimated stator temperature and rotor temperature at the current moment are output, and the updated thermal state and model correction are used for temperature prediction at the next moment. The thermal state includes stator temperature and rotor temperature. This method only requires obtaining the operating data of the motor in the low-speed, low-load region (first operating condition) to establish a priori thermal estimation model, avoiding the strong dependence of traditional methods on full-speed, full-condition bench calibration data, and significantly reducing testing costs and development cycle. By introducing model correction to compensate for model mismatch and using a dual extended Kalman filter to jointly update the thermal state and model correction online, it can achieve real-time and accurate estimation of stator and rotor temperatures under unknown high-speed operating conditions, and recursively use the updated results for prediction at the next moment, forming a closed-loop adaptive mechanism, which significantly improves the stability and applicability of thermal state estimation in the high-speed region.

[0017] The first threshold is 4000 rpm, and the second threshold is 100 Nm. By limiting the speed threshold of the first working condition to 4000 rpm and the torque threshold to 100 Nm, the specific range of the low-speed and low-load zone is clarified, making the collection of the prior sample set operable and repeatable, which is convenient for engineering implementation and batch application.

[0018] Operating data includes motor speed n electromagnetic torque Trq D-axis current i d Q-axis current i q stator temperature T s Rotor temperature T r Cooling medium temperature T c By collecting complete operating data, including rotational speed, electromagnetic torque, D / Q shaft current, stator temperature, rotor temperature, and cooling medium temperature, sufficient supervisory information is provided for subsequent thermal network model identification and loss model training, ensuring the identifiability and accuracy of model parameters.

[0019] The thermal network model and loss model of the motor are established based on operational data, specifically including the following steps: Based on the stator temperature in the operating data T s Rotor temperature T r Changes, establishing a system to describe stator temperature T s Rotor temperature T r The state equations of a second-order lumped-parameter thermal network model for thermal dynamics are as follows: In the formula, K cs , Kcr , K sr These are the equivalent thermal conductance from the stator node to the cooling boundary, the equivalent thermal conductance from the rotor node to the cooling boundary, and the equivalent thermal conductance between the stator and the rotor, respectively. , These are the stator and rotor heat capacities, respectively. , These are stator and rotor losses, respectively. In this embodiment, the thermal parameter vector is defined as follows: In the formula, This is a vector of thermal parameters; The above heat network equations can be rewritten in state-space form as follows: In the formula, The state matrix, b c The coefficient vector corresponding to the cooling medium temperature input item represents the coefficient of effect of cooling medium temperature on the rate of change of thermal state of stator nodes and rotor nodes; Wherein, the state matrix and coefficient vector b c The expression is as follows: When the system reaches thermal equilibrium, the steady-state consistency constraint is expressed as: In the formula, Let Θ be the steady-state uniform residual corresponding to the thermal parameter vector Θ. A The state matrix, It is a column vector of all 1s. This is the coefficient vector corresponding to the input term of the cooling medium temperature. The stability constraint is expressed as: In the formula, State matrix A The i 1 eigenvalue, This indicates the operation of taking the real part; Based on the variation characteristics of electromagnetic torque and stator temperature in the operating data, the first operating condition is divided into multiple sub-operating condition segments, including load transient operating condition, load steady-state operating condition, unloading transient operating condition, and unloading steady-state operating condition. Based on sub-operating condition sections, an alternating identification method using thermal parameters and equivalent losses is employed, based on the stator temperature of each sub-operating condition section. Ts Rotor temperature T r Cooling medium temperature T c The equivalent stator loss value and equivalent rotor loss value of each sub-operating condition segment are obtained by reverse calculation, and the equivalent stator loss value and equivalent rotor loss value are used as the loss label of the sub-operating condition segment; at the same time, the thermal parameters in the thermal network model are identified. In this embodiment, the specific implementation of the alternating identification method of thermal parameters and equivalent losses is as follows: First, under the current thermal parameters Under the following assumptions, the equivalent stator and rotor losses for each sub-operating condition are deduced by inversely using the temperature derivative and thermal balance relationship: In the formula, and The first m Instantaneous stator loss and instantaneous rotor loss within each operating condition. and These are the stator heat capacity and rotor heat capacity corresponding to this sub-operating condition, respectively. , and These are the stator-to-cooling-boundary thermal conductance, rotor-to-cooling-boundary thermal conductance, and stator-rotor coupling thermal conductance corresponding to this sub-operating condition. and , respectively, are the derivatives of stator temperature and rotor temperature with respect to time, and m is the sub-condition number; Within each sub-condition, the instantaneous losses are weighted and averaged to construct the equivalent loss label for that sub-condition. Separate labels are constructed for the load phase and the unload phase: in, and The first m Equivalent stator loss labels and equivalent rotor loss labels for each sub-operating condition load stage. and The first m Equivalent stator loss labels and equivalent rotor loss labels for each sub-operating condition unloading stage; For the first j The temperature fitting error for each sub-condition is defined as: In the formula, and Let $\frac{j}{k}$ be the stator temperature fitting error and the rotor temperature fitting error, respectively, for the $j$-th sub-condition at the $k$-th sampling point. and These are the stator temperature and rotor temperature fitted values ​​obtained from the model calculation, respectively. and The measured rotor temperature and rotor temperature are respectively. The objective function for thermal parameter identification is: In the formula, w j (k) The sample weights within the sub-condition are... r ss For steady-state consistency constraints, r stab For stability constraints; Through multiple rounds of alternating iterations, the first... n During round iteration, at the current thermal parameters Downward inversion of equivalent loss label : Fix the wheel wear label and update the thermal parameters: In the formula, This is the hot parameter vector updated in the (n+1)th iteration. For the equivalent loss label in the nth round The objective function for identifying thermal parameters is as follows. For the first n The equivalent loss label set obtained by round-by-round iterative back-reasoning; Repeat the above two steps, and re-infer the equivalent loss label based on the updated thermal parameters and continue iterating until the convergence condition is met, finally obtaining a set of thermal parameters and equivalent loss labels consistent with the temperature response of the low-speed sub-condition.

[0020] Using the equivalent stator loss label and equivalent rotor loss label of each sub-operating condition segment as supervision information, and the current, speed and stator temperature in the operating data as input, the loss model is trained. In this embodiment, the stator loss model adopts a basis function without explicit speed terms to conform to the mechanism characteristics that stator loss is mainly dominated by current and temperature; the rotor loss model retains the cross terms and higher-order terms related to speed to characterize the variation law of rotor loss in the high-speed region. Preferably, the stator loss basis function is defined as: In the formula, Let be the stator loss basis function vector. and Normalized d shaft current andq shaft current, The current amplitude, This is the normalized temperature of the cooling medium; The stator loss model parameter vector is defined as follows: In the formula, This is the parameter vector for the stator loss model. to These are the coefficients to be identified corresponding to each component of the stator loss basis function; The predicted stator loss then satisfies: In the formula, This is the predicted value for stator loss. This is the linear output of the stator loss model. It is a non-negative activation function, used to ensure that the stator loss prediction value is non-negative; Preferably, the rotor loss basis function is defined according to the source of rotor loss as follows: In the formula, Let the rotor loss basis function vector be . This represents the normalized motor speed. The rotor loss model parameter vector is defined as follows: In the formula, This is the parameter vector for the rotor loss model. to These are the coefficients to be identified corresponding to each component of the rotor loss basis function; The predicted rotor loss value then satisfies: In the formula, This is the linear output of the rotor loss model. This is the predicted value for rotor loss; Both loss models employ a parameterized structure with non-negative activation outputs to ensure that the loss outputs are non-negative.

[0021] Both loss models employ a parameterized structure with non-negative activation outputs to ensure that the loss outputs are non-negative.

[0022] The overall objective function is constructed using the label fitting term, parameter regularization term, higher-order constraint term, and rotational direction monotonicity constraint term. The model parameter vector is then trained to obtain a loss mapping relationship that combines physical rationality and high-speed extrapolation capability.

[0023] Let the loss label obtained by back-calculation under low-speed sub-condition be denoted as Then the overall objective function of the loss model can be written as: In the formula, L represents the overall objective function of the loss model, L fit For the label fitting term, L L2 For L2 regularization terms, L high For higher-order constraints, L mono This is a monotonicity constraint term in the direction of rotational speed; The label fitting term is: In the formula, For sample weights, This is the rotor loss error weighting coefficient. , To output the normalized scale.

[0024] To suppress overfitting and high-speed extrapolation instability caused by excessively large loss model parameters, a parameter is introduced into the training objective of the loss model. L 2. Regular expression terms, the regular expression terms with added parameters are: In the formula, These are the L2 regularization weights; To prevent higher-order terms from being over-amplified during high-speed extrapolation, stronger regularization is applied to some higher-order basis functions: In the formula, These are the constraint weight coefficients for higher-order terms. , These are the indices of higher-order terms in the stator and rotor models, respectively.

[0025] To ensure that the rotor loss extrapolation conforms to the physical model, the monotonicity constraint regarding the rotational speed direction can be expressed as: In the formula, and These are the monotonicity constraint weighting coefficients for stator loss and rotor loss, respectively. By establishing a second-order lumped-parameter thermal network model, the thermal dynamic characteristics of the stator and rotor are described using a low-order model, reducing computational complexity. The thermal response samples are enriched by subdividing the low-speed operating conditions into multiple sub-conditions such as load transient, load steady-state, unloading transient, and unloading steady-state. The alternating identification method of thermal parameters and equivalent losses enables simultaneous identification of thermal parameters and loss labels under conditions where losses are completely unknown, solving the technical challenge of independent identification of coupled thermal parameters and losses. The loss model is trained using the back-derived loss labels as supervisory information, enabling the low-speed identification results to be effectively extrapolated to high-speed operating conditions, laying the model foundation for high-speed temperature estimation.

[0026] Model corrections include , , , , ,in, and These are the correction factors for stator loss and rotor loss, respectively. , and These are correction factors for the thermal conductivity from the stator to the cooling boundary, the thermal conductivity from the rotor to the cooling boundary, and the thermal conductivity of the stator-rotor coupling, respectively. A model correction parameter is introduced to correct the model parameters in the prior heat estimation model. The corrected effective parameters satisfy the following: In the formula, P s,eff and P r,eff These are the effective stator loss and effective rotor loss under the current operating conditions. K sc,eff , K rc,eff and K sr,eff These are the corrected thermal conductance from the stator to the cooling boundary, the thermal conductance from the rotor to the cooling boundary, and the thermal conductance coupled between the stator and rotor, respectively. Substituting the model corrections into the second-order lumped-parameter thermal network model, we obtain the online corrected thermal state equation for the current operating condition: In the formula, T s,k andT r,k They are respectively k The stator temperature and rotor temperature at that moment. T c,k for k The temperature of the cooling medium at any given time, Δ t The sampling period.

[0027] By introducing model correction factors that include stator loss correction factors, rotor loss correction factors, and multiple heat conduction correction factors, the key parameters in the prior thermal estimation model are equivalently scaled and corrected. This effectively characterizes the deviation from the low-speed nominal model under high-speed unknown operating conditions. The corrected online thermal equation of state directly describes the recursive relationship between stator temperature and rotor temperature, providing an accurate state transition model for the dual extended Kalman filter, thereby compensating for the estimation error caused by model mismatch.

[0028] The thermal state of the motor and model corrections are jointly updated online using a dual extended Kalman filter, specifically including the following steps: The thermal state vector and the correction state vector are defined as follows: in, Let be the thermal state vector. Let k be the state vector of the correction value. It should be noted that the correction quantity state vector The parameters in the parameters can be selected for use by one or more depending on the actual application scenario; it is not necessary to use all of them.

[0029] Based on the thermal state vector and the correction state vector, the state equation, parameter equation, and observation equation are constructed for the dual extended Kalman filter, where: The state equation is constructed by incorporating model corrections into the prior heat estimation model, and is expressed as: In the formula, For state process noise, x k and x k-1 They are respectively k Time and k-1 The thermal state vector at time t, θ k for k The model correction state vector at time 1. u k-1 for k-1 The external input vector at time t; The parametric equations, expressed in random walk form, are as follows: In the formula, This is parameter process noise, reflecting the variation characteristics of the correction amount between adjacent sampling times; θ k-1 for k- 1 The model correction state vector at time t; The observation equation, with stator temperature as the sole observable, is expressed as: In the formula, y k for k Stator temperature observation at time t, H For the observation matrix, For measuring noise; The dual extended Kalman filter performs online joint updates of thermal state and model corrections based on state equations, parametric equations, and observation equations. By defining the thermal state vector and the correction state vector, and constructing state equations, parametric equations, and observation equations for the dual extended Kalman filter, the thermal state estimation and model correction estimation are unified into the same filtering framework. The state equation describes the evolution of the thermal state, the parametric equation adopts a random walk form to adapt to the slow time-varying characteristics of the model correction, and the observation equation only requires the stator temperature as the only observation, eliminating the need to install a rotor temperature sensor, thus reducing hardware costs and system complexity. The dual extended Kalman filter performs joint estimation of the thermal state and correction based on the three equations, realizing adaptive online correction of the model.

[0030] The dual extended Kalman filter includes a state filter and a parameter filter; The state filter is used to predict and update the thermal state at the current moment. Its update equation is as follows: In the formula, K k,x Let K be the Kalman gain of the state filter at time k. and These are the state prediction covariance matrix and the updated covariance matrix, respectively. R To measure the noise covariance matrix, and These are the predicted and updated values ​​for the thermal state, respectively. I It is the identity matrix; The parameter filter is used to predict and update the model correction at the current time step. Its update equation is: In the formula, Let K be the Kalman gain of the parameter filter at time k. and These are the predicted covariance matrix and the updated covariance matrix of the corrected state, respectively. The linearized observation matrix of the parameter filter, and These are the predicted and updated values ​​of the model correction, respectively. By decomposing the dual extended Kalman filter into a state filter and a parameter filter, the thermal state and model correction are predicted and updated respectively. The two filters are executed alternately to form a closed-loop control mechanism. The state filter outputs the estimated values ​​of the stator temperature and rotor temperature at the current moment, while the parameter filter updates the model correction in real time to adapt to changes in operating conditions. This alternating update structure can effectively suppress noise interference, improve estimation accuracy, and ensure that the model correction can smoothly track the model offset under high-speed operating conditions, avoiding instability caused by sudden changes in the estimated value.

[0031] like Figure 2 As shown, a speed-domain motor temperature estimation system based on data mechanism fusion is used to estimate the motor temperature, including: The data acquisition module is used to acquire the motor's operating data under the first operating condition; The offline training module, connected to the data acquisition module, is used to build a priori heat estimation model based on the running data; The online correction module, connected to the offline training module, is used to apply the prior heat estimation model to the current operating conditions and introduce model correction parameters for correction. The dual Kalman filter module, connected to the online correction module, is used to jointly update the motor's thermal state and model correction values ​​online, and output stator temperature estimates and rotor temperature estimates.

[0032] By constructing a system architecture that includes a data acquisition module, an offline training module, an online correction module, and a dual Kalman filter module, a complete functional closed loop from data acquisition and model training to online estimation is achieved. The connections between the modules are clear, and offline training and online operation are separated, which not only ensures the basic accuracy of the model, but also has the ability to adapt to unknown operating conditions, making it easy to embed into the motor controller or vehicle controller to achieve real-time temperature monitoring.

[0033] To verify the motor temperature estimation method of this application, the following experiment was conducted: To verify the effectiveness of the method proposed in this application, experiments were conducted on a permanent magnet synchronous motor (PMSM) test bench. The test bench included the PMSM under test, a motor controller, a dynamometer, a cooling circulation device, and a data acquisition system. During the experiment, motor speed, electromagnetic torque, d-axis current, q-axis current, stator temperature, rotor temperature, and cooling medium temperature were simultaneously collected. First, operating data under low-speed, low-load conditions was collected, with speed not exceeding 4000 rpm and torque not exceeding 100 Nm. Based on this data, a priori thermal estimation model was established. Subsequently, the model was applied to extended-speed operating conditions, and a dual extended Kalman filter was used to jointly estimate the motor thermal state and model corrections online. Experimental results are as follows Figure 3 , Figure 4 As shown, Figure 3 The rotor temperature estimation curves of the permanent magnet synchronous motor under full-speed operating conditions are presented, where the solid line represents the measured rotor temperature value and the dashed line represents the rotor temperature estimation value obtained by the method of this application. Figure 4 The corresponding estimation error curve is given, with the vertical axis representing the deviation between the estimated and measured values. From Figure 3 It can be seen that, under the premise of offline training using only data from the low-speed, low-load region with a rotational speed not exceeding 4000 rpm and a torque not exceeding 100 Nm, the rotor temperature curve estimated by the method of this application is highly consistent with the measured temperature curve across the entire speed range, and can accurately track the changing trends of temperature rise, fall, and steady-state stages. Figure 4 It can be seen that the estimation error under all operating conditions is consistently controlled within ±5℃, indicating that the method of this application can still maintain high estimation accuracy under unknown high-speed and high-load operating conditions without calibration. The above experimental results show that the technical solution of this application, which combines low-speed region prior modeling with online joint updating of dual extended Kalman filters, can effectively solve the problems of excessive dependence on full-speed domain calibration data and model mismatch leading to decreased estimation accuracy under unknown operating conditions in traditional methods, verifying the applicability and reliability of the method of this application on different types of drive motors.

[0034] 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 the invention. In this specification, 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.

[0035] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for temperature estimation of an extended speed range electric machine based on data mechanism fusion, characterized in that, Includes the following steps: Acquire the motor's operating data under the first operating condition, which is the condition where the speed is less than or equal to the first threshold and the torque is less than or equal to the second threshold; A thermal network model and a loss model of the motor are established based on the operating data. The thermal network model and the loss model are used as a priori thermal estimation models. The thermal network model is used to describe the thermal dynamic characteristics of the motor stator and rotor, and the loss model is used to calculate the stator loss and rotor loss of the motor as input to the thermal network model. The prior thermal estimation model is applied to the current operating conditions, and model correction is introduced to correct the model parameters in the prior thermal estimation model in order to compensate for the model mismatch of the prior thermal estimation model under the current operating conditions. The system collects the current, speed, stator temperature, and cooling medium temperature of the motor under the current operating conditions in real time. It uses a dual extended Kalman filter to jointly update the motor's thermal state and model correction online, outputting the estimated stator temperature and rotor temperature at the current moment. The updated thermal state and model correction are then used for temperature prediction at the next moment. The thermal state includes both stator temperature and rotor temperature.

2. The method of claim 1, wherein, The first threshold is 4000 rpm, and the second threshold is 100 Nm.

3. The method of claim 1, wherein, Operating data includes motor speed n electromagnetic torque Trq D-axis current i d Q-axis current i q stator temperature T s Rotor temperature T r Cooling medium temperature T c .

4. The method of claim 3, wherein, The thermal network model and loss model of the motor are established based on operational data, specifically including the following steps: Based on the stator temperature in the operating data T s Rotor temperature T r Changes, establishing a system to describe stator temperature T s Rotor temperature T r The state equations of a second-order lumped-parameter thermal network model for thermal dynamics are as follows: In the formula, K cs , K cr , K sr These are the equivalent thermal conductance from the stator node to the cooling boundary, the equivalent thermal conductance from the rotor node to the cooling boundary, and the equivalent thermal conductance between the stator and the rotor, respectively. , These are the stator and rotor heat capacities, respectively. , These are stator and rotor losses, respectively. Based on the variation characteristics of electromagnetic torque and stator temperature in the operating data, the first operating condition is divided into multiple sub-operating condition segments, including load transient operating condition, load steady-state operating condition, unloading transient operating condition, and unloading steady-state operating condition. Based on sub-operating condition sections, an alternating identification method using thermal parameters and equivalent losses is employed, based on the stator temperature of each sub-operating condition section. T s Rotor temperature T r Cooling medium temperature T c The equivalent stator loss value and equivalent rotor loss value of each sub-operating condition segment are obtained by reverse calculation, and the equivalent stator loss value and equivalent rotor loss value are used as the loss label of the sub-operating condition segment; at the same time, the thermal parameters in the thermal network model are identified. Using the equivalent stator loss labels and equivalent rotor loss labels of each sub-operating condition segment as supervisory information, and taking the current, speed and stator temperature in the operating data as input, a loss model is trained.

5. The method of claim 4, wherein, Model corrections include , , , , ,in, and These are the correction factors for stator loss and rotor loss, respectively. , and These are correction factors for the thermal conductivity from the stator to the cooling boundary, the thermal conductivity from the rotor to the cooling boundary, and the thermal conductivity of the stator-rotor coupling, respectively. A model correction parameter is introduced to correct the model parameters in the prior heat estimation model. The corrected effective parameters satisfy the following: In the formula, P s,eff and P r,eff These are the effective stator loss and effective rotor loss under the current operating conditions. K sc,eff , K rc,eff and K sr,eff These are the corrected thermal conductance from the stator to the cooling boundary, the thermal conductance from the rotor to the cooling boundary, and the thermal conductance coupled between the stator and rotor, respectively. Substituting the model corrections into the second-order lumped-parameter thermal network model, we obtain the online corrected thermal state equation for the current operating condition: In the formula, T s,k and T r,k They are respectively k The stator temperature and rotor temperature at that moment. T c,k for k The temperature of the cooling medium at any given time, Δ t The sampling period.

6. The method of claim 5, wherein the data mechanism fusion based extended speed domain motor temperature estimation method is characterized by, The thermal state of the motor and model corrections are jointly updated online using a dual extended Kalman filter, specifically including the following steps: The thermal state vector and the correction state vector are defined as follows: wherein, is the thermal state vector, is k is the time instant correction state vector; Based on the thermal state vector and the correction state vector, the state equation, parameter equation, and observation equation are constructed for the dual extended Kalman filter, where: The state equation is constructed by incorporating model corrections into the prior heat estimation model, and is expressed as: In the formula, For state process noise, x k and x k-1 They are respectively k Time and k-1 The thermal state vector at time t, θ k for k The model correction state vector at time 1. u k-1 for k-1 The external input vector at time t; The parametric equations, expressed in random walk form, are as follows: In the formula, This is parameter process noise, reflecting the variation characteristics of the correction amount between adjacent sampling times; θ k-1 for k-1 The model correction state vector at time t; The observation equation, with stator temperature as the sole observable, is expressed as: wherein y k is the stator temperature observation at time k, H is the observation matrix, is the measurement noise; The dual extended Kalman filter performs online joint updates of thermal state and model corrections based on state equations, parameter equations, and observation equations.

7. The method of claim 6, wherein the data mechanism fusion based extended speed domain motor temperature estimation method is characterized by, The dual extended Kalman filter includes a state filter and a parameter filter; The state filter is used to predict and update the thermal state at the current moment. Its update equation is as follows: In the formula, K k,x For state filters in k Kalman gain at time step and These are the state prediction covariance matrix and the updated covariance matrix, respectively. R To measure the noise covariance matrix, and These are the predicted and updated values ​​for the thermal state, respectively. I It is the identity matrix; The parameter filter is used to predict and update the model correction at the current time step. Its update equation is: In the formula, Let K be the Kalman gain of the parameter filter at time k. and These are the predicted covariance matrix and the updated covariance matrix of the corrected state, respectively. The linearized observation matrix of the parameter filter, and These are the predicted and updated values ​​of the model correction, respectively.

8. A temperature estimation system for an extended speed range electric machine based on data mechanism fusion, which implements the temperature estimation method of any one of claims 1 to 7 to estimate the temperature of the electric machine, characterized in that, include: The data acquisition module is used to acquire the motor's operating data under the first operating condition; The offline training module, connected to the data acquisition module, is used to build a priori heat estimation model based on the running data; The online correction module, connected to the offline training module, is used to apply the prior heat estimation model to the current operating conditions and introduce model correction parameters for correction. The dual Kalman filter module, connected to the online correction module, is used to jointly update the motor's thermal state and model correction values ​​online, and output stator temperature estimates and rotor temperature estimates.

Images