Motor temperature estimation method and device based on GP-NARX model

Abstract

The invention provides a motor temperature estimation method and device based on a GP-NARX model. The method comprises the steps that working condition parameters of a motor under a preset working condition are acquired; executing initialization work to obtain an initial temperature value of the motor; and inputting the working condition parameters and the initial temperature value into a GP-NARX model obtained through training in a training stage, and taking a final output value of the GP-NARX model as a real-time temperature estimation value of the motor. The GP-NARX model feeds back the motor temperature value output at the previous moment to the input at the next moment in the working process so as to reflect the influence of the motor temperature change on the real-time temperature of the motor; wherein in the training stage, the multiple working condition parameters are optimized based on the grey relational degree, so that the working condition parameters more related to the motor temperature are screened to serve as training data input into the GP-NARX model, and then the GP-NARX model is trained.

Description

Technical Field

[0001] The present invention relates to the field of motor temperature estimation, and in particular to a motor temperature estimation method and device based on a GP-NARX model. Background Technology

[0002] Due to the advantages of energy conservation, environmental protection, green and low carbon, the market penetration rate of new energy vehicles is constantly increasing, and the market share is also increasing. As the core component of its electric drive system, the drive motor of new energy vehicles has received more attention and research. The motor will lose power during operation, causing its temperature to rise. Excessive temperature will accelerate the aging of the stator winding and the demagnetization of the rotor magnet, and in severe cases, the motor will be scrapped. Therefore, research on how to estimate the motor temperature is of great significance.

[0003] In the prior art, the methods for estimating motor temperature mainly include: finite element method, lumped parameter thermal network method, signal injection method, and flux observer, etc. However, due to the complex internal heat transfer relationship of the automotive drive motor, it is difficult to accurately analyze its heat transfer path, and these motor temperature estimation methods face various difficulties and challenges in practical application. At the same time, in recent years, as artificial intelligence technology has been widely studied and applied in data mining, complex system modeling, parameter identification, etc., solutions for estimating automotive drive motor temperature using artificial intelligence technology have begun to emerge to assist in the research of motor temperature.

[0004] At the same time, Gaussian Process Regression (GPR) is a non-parametric probability model based on Bayesian methods that can simulate the nonlinear behavior of the system. Nonlinear AutoRegressive with eXogenous Input NARX is a recurrent dynamic network, usually used for time series modeling. Its model prediction output can be fed back to the model input port to consider the impact of output value changes on the system. The Gaussian Process Nonlinear AutoRegressive eXogenous (GP-NARX) algorithm can combine GPR and NARX, and has the advantages of both algorithms, and can work well with motor temperature estimation.

[0005] There are many parameters that can affect the temperature during the operation of the motor, and the selection of model input parameters is crucial to modeling. However, manual feature selection is blind and often does not produce ideal results. In addition, there are many hyperparameters to be determined in the GP-NARX model, which directly affect the performance of the model. The parameter optimization methods of existing solutions in the prior art often fail to achieve ideal results.

[0006] To overcome the above-mentioned defects existing in the prior art, there is an urgent need in the art for a method and device for estimating the motor temperature based on the GP-NARX model, which uses the GP-NARX model to perform the motor temperature estimation work. When training the model, the selection of input variables and the solution of hyperparameters of the model are optimized, and the model is specifically improved for the actual working condition problem of motor temperature estimation, so that higher estimation efficiency and estimation accuracy can be obtained when using the GP-NARX model to perform the motor temperature estimation work, and further facilitate the motor development work. Summary of the Invention

[0007] The following gives a brief overview of one or more aspects to provide a basic understanding of these aspects. This overview is not an exhaustive survey of all contemplated aspects, and is neither intended to identify key or decisive elements of all aspects nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description to follow.

[0008] To overcome the above-mentioned defects existing in the prior art, the present invention provides a method for estimating the motor temperature based on the GP-NARX model, which may include: obtaining the working condition parameters of the motor under a predetermined working condition; performing initialization work to obtain the initial temperature value of the motor; and inputting the working condition parameters and the initial temperature value into the GP-NARX model trained through a training stage, and taking the final output value of the GP-NARX model as the real-time temperature estimation value of the motor. In the working process of the GP-NARX model, the motor temperature value output at the previous moment is fed back to the input of the next moment to reflect the influence of the motor temperature change on the real-time temperature of the motor; wherein, in the training stage, multiple working condition parameters are optimized based on the grey relational degree to screen out the working condition parameters more relevant to the motor temperature as the training data input into the GP-NARX model, and then the GP-NARX model is trained.

[0009] In a preferred embodiment, the optimization of multiple working condition parameters based on the grey relational degree includes: calculating the grey relational degree values between multiple working condition parameters and the motor temperature respectively by using the following formula:

[0010]

[0011] wherein, y(k) is the motor temperature, x i (k) are multiple working condition parameters related to the motor temperature, ρ is the grey resolution coefficient; and sorting multiple working condition parameters in descending order according to their corresponding grey relational degree values to obtain the hyperparameter U to be optimized for determining the number of input variables of the GP-NARX model: U = [U 1 , U 2 , …, UM ; where, U 1 represents the operating condition parameter with the largest grey relational grade value with the motor temperature, U 2 represents the operating condition parameters with the grey relational grade values with the motor temperature ranking in the top two, and so on, U M represents the operating condition parameters with the grey relational grade values with the motor temperature ranking in the top M.

[0012] In one embodiment, preferably, the training stage may include: collecting multiple groups of operating condition data of the motor during actual operation, dividing the multiple groups of operating condition data into multiple data sets, and performing optimized sorting based on the grey relational grade. The multiple data sets include a training set; selecting an appropriate kernel function for the GP-NARX model, inputting the operating condition data of the training set sorted by the grey relational grade into the GP-NARX model; and determining the hyperparameters of the GP-NARX model by using the PSO optimization algorithm, so as to complete the model training work.

[0013] In one embodiment, preferably, the multiple groups of operating condition data include the operating condition parameters under multiple operating conditions sorted by the grey relational grade and the real-time temperature of the motor under each operating condition. The training stage may further include: preprocessing the multiple groups of operating condition data sorted by the grey relational grade and organizing them into a time series form of (x i , y i ). x i is the model input, y i is the real-time temperature of the motor collected by the sensor. x i is expressed as:

[0014]

[0015] where, is the operating condition parameter of different operating conditions of the motor for inputting into the GP-NARX model, is the real-time temperature of the motor for feedback to the input of the GP-NARX model. n u and n y are parameters for determining the output regression step length and the input step length of the GP-NARX model respectively.

[0016] In one embodiment, preferably, the real-time temperature y i of the motor is expressed as: y i = f(x i ) + ε i , where f(·) represents the GP-NARX model, and ε i is the added Gaussian white noise with a mean of zero, and ε i satisfies the following expression to prevent model overfitting: where σ y is εi Standard deviation

[0017] In one embodiment, preferably, the GP-NARX model further satisfies the following expression:

[0018] f = f(x i ) ~ N(f|0, K), where K is the covariance matrix, K ij = k(x i , x j ), k(·,·) is the kernel function; selecting an appropriate kernel function for the GP-NARX model may include: selecting the squared exponential function as the kernel function and representing the kernel function with the following formula:

[0019]

[0020] where D is the data dimension, σ f , l are the hyperparameters to be optimized in the kernel function

[0021] In one embodiment, preferably, the hyperparameters of the GP-NARX model are θ = [U, n u , n y , σ f , l, σ y , where U is the model input operating condition parameter, n u , n y are the step sizes of the GP-NARX model input and output feedback respectively, σ f and l are the hyperparameters of the SE kernel function, σ y is the standard deviation of the model Gaussian white noise. Determining the hyperparameters of the GP-NARX model using the PSO optimization algorithm may include: determining the dimension of each particle according to the number of variables of the hyperparameters, randomly generating the initial population of the GP-NARX model hyperparameters, including the initial velocity and initial position of each particle; calculating the fitness value of each particle individual based on the fitness function, and the fitness value of each particle individual is the mean square error f id of the motor temperature prediction; if the mean square error f id of the motor temperature prediction corresponding to a particle is less than the minimum temperature mean square error f(P id ) of the particle in the previous iteration history, then use the mean square error f id of the motor temperature to replace f(P id ) in the previous round, use the particle to replace the particle in the previous round, and update the velocity and position of the particle; if the minimum mean square error f(P id ) of a particle is less than the global minimum mean square error f(P gd) If so, replace the original global minimum mean square error of the motor temperature with the minimum mean square error of the motor temperature of this particle, and save the current position of this particle; and when the preset optimization end condition is satisfied, use the current position of the particle as the optimal input and hyperparameters of the GP-NARX model obtained by optimization.

[0022] In one embodiment, preferably, updating the velocity and position of the particle may include: updating the velocity and position of the particle using the following formula:

[0023] V id (k + 1) = wV id (k) + c 1 r 1 (k)(P id (k) - X id (k)) + c 2 r 2 (k)(P gd (k) - X id (k))

[0024] X id (k + 1) = X id (k) + V id (k + 1)

[0025] Among them, V id (k) is the velocity of the i-th particle in the d-th dimension at the k-th iteration, X id (k) is the position of the i-th particle in the d-th dimension at the k-th iteration, P id is the individual historical optimal position of the i-th particle in the d-th dimension, P gd is the optimal position experienced by all particles, c 1 and c 2 are learning factors determined according to practical experience, r 1 , r 2 are random numbers uniformly distributed between [0, 1], w is the inertia weight that linearly decreases according to the number of iterations, and is expressed by the following formula:

[0026]

[0027] Among them, w max , w min are the maximum and minimum inertia weights respectively, t is the current iteration number, k max is the maximum iteration number.

[0028] In one embodiment, preferably, the multiple data sets further include a test set, and the motor temperature estimation method may further include: after the GP-NARX model is trained in the training stage, the trained GP-NARX model is verified in the test stage; if the trained GP-NARX model meets the preset verification conditions in the test stage, the operating condition parameters and the initial temperature value are input into the GP-NARX model, and the final output value of the model is used as the real-time temperature estimation value of the motor; if the trained GP-NARX model does not meet the preset verification conditions in the test stage, the GP-NARX model is retrained.

[0029] In one embodiment, preferably, the preset verification conditions may include: using the following error function RMSE as an evaluation index to measure the prediction accuracy of the GP-NARX model:

[0030]

[0031] where N represents the number of data groups in the test set, y i represents the real-time temperature value of the motor measured by the sensor, and y i * represents the predicted output value of the GP-NARX model.

[0032] In one embodiment, preferably, inputting the operating condition parameters and the initial temperature value into the GP-NARX model trained in the training stage may include: inputting the operating condition parameter data of the operating condition parameter category optimized and screened based on the grey relational degree in the training stage into the GP-NARX model to predict the real-time temperature estimation value of the motor.

[0033] In one embodiment, preferably, performing the initialization work to obtain the initial temperature value of the motor may include: powering on the vehicle, obtaining the vehicle power-off time, and comparing the vehicle power-off time with a preset time threshold; in response to the vehicle power-off time being not less than the preset time threshold, taking the current temperature of the motor cooling water inlet as the initial temperature value; and in response to the vehicle power-off time being lower than the preset time threshold, selecting a motor cooling curve based on the real-time ambient temperature, and calculating the current temperature of the motor according to the vehicle power-off time and the motor cooling curve as the initial temperature value.

[0034] Another aspect of the present invention also provides a motor temperature estimation device based on a GP-NARX model, including: a memory; and a processor coupled to the memory, the processor being configured to execute the motor temperature estimation method based on the GP-NARX model as described in any one of the above.

[0035] The present invention also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method for estimating the motor temperature based on the GP-NARX model as described in any one of the above. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] After reading the following detailed description of the embodiments of the present disclosure in conjunction with the accompanying drawings, the above features and advantages of the present invention can be better understood. In the drawings, the components are not necessarily drawn to scale, and components having similar relevant characteristics or features may have the same or similar reference numerals.

[0037] Figure 1 FIG. is a schematic flowchart of a method for estimating the motor temperature based on the GP-NARX model according to one aspect of the present invention;

[0038] Figure 2 FIG. is a schematic diagram of the model principle of the GP-NARX model according to an embodiment of the present invention;

[0039] Figure 3 FIG. is a schematic flowchart of the method for estimating the motor temperature based on the GP-NARX model in the model training and testing stages according to an embodiment of the present invention;

[0040] Figure 4 FIG. is a schematic flowchart of the initialization work in the method for estimating the motor temperature according to an embodiment of the present invention;

[0041] Figure 5 FIG. is a schematic diagram of the device structure of a device for estimating the motor temperature based on the GP-NARX model according to another aspect of the present invention; and

[0042] Figure 6 FIG. is a schematic diagram of the hardware device structure for deploying the GP-NARX model according to an embodiment of the present invention.

[0043] For clarity, the following gives a brief description of the reference numerals:

[0044] 601 AMU unit

[0045] 602 DMA unit

[0046] 603 SRAM unit

[0047] 604 Non-volatile storage

[0048] 605 First path

[0049] 606 Second path DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The following specific embodiments illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Although the description of the present invention will be introduced in combination with preferred embodiments, this does not mean that the features of this invention are limited to this implementation manner. On the contrary, the purpose of introducing the invention in combination with the implementation manner is to cover other alternatives or modifications that may be extended based on the claims of the present invention. In order to provide a deep understanding of the present invention, many specific details will be included in the following description. The present invention can also be implemented without using these details. In addition, in order to avoid confusing or obscuring the key points of the present invention, some specific details will be omitted in the description.

[0051] In the description of the present invention, it should be noted that unless otherwise clearly specified and defined, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0052] In addition, the "upper", "lower", "left", "right", "top", "bottom", "horizontal", and "vertical" used in the following description should be understood as the orientations shown in this paragraph and the related drawings. This relative term is only for the convenience of description and does not mean that the device described needs to be manufactured or operated in a specific orientation, so it should not be understood as a limitation to the present invention.

[0053] It can be understood that although terms such as "first", "second", and "third" can be used here to describe various components, regions, layers, and / or parts, these components, regions, layers, and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers, and / or parts. Therefore, the first component, region, layer, and / or part discussed below can be called the second component, region, layer, and / or part without departing from some embodiments of the present invention.

[0054] In order to overcome the above-mentioned defects existing in the prior art, the present invention provides a method and device for estimating the temperature of an electric motor based on a GP-NARX model. The GP-NARX model is used for the work of estimating the temperature of the electric motor. When training the model, the selection of input variables and the solution of hyperparameters of the model are optimized. The model is specifically improved for the actual working condition problem of electric motor temperature estimation, so that higher estimation efficiency and estimation accuracy can be obtained when using the GP-NARX model for the work of estimating the temperature of the electric motor, and further facilitate the development work of the electric motor.

[0055] Figure 1It is a schematic flowchart of a method for estimating the temperature of an electric motor based on a GP-NARX model shown according to an aspect of the present invention.

[0056] Please refer to Figure 1 , the method 100 for estimating the temperature of an electric motor based on a GP-NARX model provided by the present invention may include:

[0057] Step 101: Obtain the operating condition parameters of the electric motor under a predetermined operating condition;

[0058] Step 102: Perform initialization work to obtain the initial temperature value of the electric motor; and

[0059] Step 103: Input the operating condition parameters and the initial temperature value into the GP-NARX model trained through a training stage, and use the final output value of the GP-NARX model as the real-time temperature estimation value of the electric motor. During the operation of the GP-NARX model, the temperature value of the electric motor output at the previous moment is fed back to the input of the next moment to reflect the influence of the change in the electric motor temperature on the real-time temperature of the electric motor; wherein, during the training stage, multiple operating condition parameters are optimized based on the grey relational degree to screen out the operating condition parameters more relevant to the electric motor temperature as the training data input into the GP-NARX model to train the GP-NARX model.

[0060] When the temperature of the automotive drive motor is too high, its parameters will change, affecting its performance. Specific manifestations include, for example, an increase in the resistance of the stator winding, an increase in copper loss, which exacerbates the temperature rise, demagnetization of the rotor permanent magnet, and a decrease in the output torque under the same current. Those skilled in the art can understand that, on the one hand, the change in the electric motor temperature is related to the operating conditions of its own operation, such as: speed, torque, cooling conditions, bus voltage, etc.; on the other hand, the change in the electric motor temperature is also related to the temperature of its body. Therefore, the method for estimating the temperature of the electric motor provided by the present invention uses operating condition parameters such as speed, torque, cooling conditions, and bus voltage as the model input, and the electric motor temperature as the model output, and the temperature value output by the model is fed back to the model input of the next moment, so that the change in the temperature of the automotive drive motor can be affected by both the model input and output at the same time, better simulating the actual influencing factors of the electric motor temperature.

[0061] Figure 2 It is a schematic diagram of the model principle of a GP-NARX model shown according to an embodiment of the present invention.

[0062] Reference may be made in conjunction with Figure 2, the motor temperature estimation method provided by the present invention uses the Gaussian Process Nonlinear AutoRegressive eXogenous (GP-NARX) algorithm to establish a motor thermal model. Gaussian Process (GP) regression is a non-parametric probability model based on Bayesian methods and is a powerful tool for simulating the nonlinear behavior of systems. The Nonlinear AutoRegressive with eXogenous Input (NARX) is a recurrent dynamic network, usually used for time series modeling, and its model prediction output can be fed back to the model input port to consider the influence of output value changes on the system. As mentioned above, the temperature of the automotive drive motor is jointly affected by operating condition parameters and temperature changes. When modeling, it is necessary to consider the feedback of both the model input and output simultaneously. Therefore, the motor temperature estimation method provided by the present invention is based on the GP-NARX model.

[0063] As Figure 2 shown, on the one hand, the model input can be a series of operating condition parameters, such as the coolant temperature t coolt , the DC bus voltage U dc , the three-phase line current amplitude I s , the torque T q , the rotational speed n, etc. On the other hand, the output temperature value of the model is fed back to the input of the next moment to consider the influence of temperature changes on the motor temperature. Here, z in the feedback box represents the delayed time step.

[0064] Figure 3 is a schematic flow chart of the motor temperature estimation method based on the GP-NARX model in the model training and testing stages according to an embodiment of the present invention.

[0065] In a preferred embodiment, the training stage of the GP-NARX model may first include: Step 301, collecting multiple sets of operating condition data of the motor during actual operation and preprocessing the data. The multiple sets of operating condition data may include the operating condition parameters of multiple operating conditions and the real-time temperature of the motor under each operating condition.

[0066] More specifically, for example, an electric drive test bench can be utilized to conduct motor temperature measurement tests under different working conditions. Furthermore, relevant electrical, mechanical, and temperature signals that affect the motor temperature, as well as temperature data of different parts of the motor, can be synchronously collected during the tests. The different working conditions can be different ambient temperatures, different coolant temperatures, different coolant flow rates, different DC bus voltages, different rotational speeds, different torques, Active Short Circuit (ASC) mode, and locked-rotor mode. The parts for motor temperature measurement can be the stator three-phase windings, stator core, rotor permanent magnets, rotor core, bearings, and coolant outlet. The collected electrical, mechanical, and temperature signals can be bus voltage, bus current, power, motor d / q axis voltages, d / q axis currents, three-phase line voltages, three-phase line currents, modulation index, switching frequency, power factor, rotational speed, torque, coolant flow rate, coolant inlet temperature, rotational speed of the oil pump in the electric drive system, oil pump current, etc. Through the above tests, motor temperature field data under different working conditions are obtained, and a motor temperature field data set is established.

[0067] Considering the high cost of bench tests, in addition to using the test bench, in one embodiment, data acquisition can also be combined with a simulation model. For example, first, typical working conditions are selected for bench tests to obtain actual measurement data, and then the simulation model is calibrated using the test data to obtain a high-precision simulation model. Subsequently, different working condition parameters can be used as inputs to the simulation model, and the temperature data of different parts of the motor under different working conditions can be obtained through simulation. This data acquisition method can reduce the cost of bench tests, make it easier to obtain data for working condition points that are difficult to measure in bench tests, and achieve cost reduction and efficiency improvement.

[0068] One of the above two data acquisition methods or a combination of the two can be used to obtain motor temperature field data under different working conditions and establish a motor temperature field data set.

[0069] It is easily understandable that after establishing the motor temperature field data set, preprocessing of the data set is required, such as data filtering, missing value filling, etc. In particular, in the motor temperature estimation method provided by the present invention, during the training stage, multiple such working condition parameters are optimized based on the grey relational degree to screen out the working condition parameters that are more relevant to the motor temperature as the training data for inputting into the GP-NARX model, and then the GP-NARX model is trained.

[0070] For example, in one embodiment, after collecting the motor temperature field data set, the grey relational grade between variables is calculated and sorted. The output of the model is the temperature of a certain part of the motor to be predicted. However, there are many operating parameters related to the motor temperature, and selecting the optimal model input becomes a problem to be optimized. Here, we use the grey relational grade analysis method to select the model input. If there are M operating parameters of the motor, first calculate the grey relational grade values between these operating parameters and the motor temperature, as shown in the following formula:

[0071]

[0072] where y(k) is the motor temperature, x i (k) are multiple such operating parameters related to the motor temperature, and ρ is the grey distinguishing coefficient, which can take a value of 0.5 in this embodiment, for example.

[0073] After calculating the grey relational grade values of each operating parameter with respect to the motor temperature, sort them from largest to smallest according to the grey relational grade values to obtain [ζ 1 , ζ 2 , …, ζ M .

[0074] Therefore, the multiple groups of operating data mentioned in the present invention include the operating parameters under multiple operating conditions sorted by grey relational grade and the real-time temperature of the motor under each operating condition. In one embodiment, the training stage may further include: preprocessing the multiple groups of operating data sorted by grey relational grade and organizing them into a time series form of (x i , y i ), where x i is the model input, y i is the real-time temperature of the motor collected by the sensor, and x i is expressed as:

[0075]

[0076] where are the operating parameters of different operating conditions of the motor for inputting into the GP-NARX model, is the real-time temperature of the motor for feedback to the input of the GP-NARX model, and n u and n y are parameters for determining the output regression step length and input step length of the GP-NARX model, respectively.

[0077] For example, the motor coolant inlet temperature, DC bus voltage, line current, torque, speed, switching frequency, and power can be selected as inputs, and the stator winding temperature can be used as the prediction output to carry out model training.

[0078] Please continue to refer to Figure 3, after data preprocessing, step 302 is executed: The multiple groups of working condition data are divided into multiple data sets, and the multiple data sets can include a training set and a test set. For example, the proportions of the training set and the test set can be 70% and 30% respectively. Among them, the training set is used to train the model to obtain better hyperparameters so as to obtain a better alternative model, and the test set is used to verify the performance of the model after training to determine the optimal model.

[0079] Then step 303 is executed: Initialize the GP-NARX model and select a kernel function. It is easy to understand that since the motor temperature estimation method provided by the present invention needs to feedback the temperature prediction value output by the model to the input, it is necessary to perform initialization work to obtain the initial feedback input value and select a suitable kernel function for the model, and then solve the hyperparameters of the model.

[0080] In a preferred embodiment, the model input x i can actually include the working condition data that determines the change of the motor temperature and the feedback of the temperature prediction value. The relationship between the motor temperature y i and x i can be described by the following formula:

[0081] y i = f(x i ) + ε i

[0082] where f(·) represents the function of the input-output relationship of the GP-NARX model, and ε i is the added Gaussian white noise with a mean of zero, and ε i satisfies the following expression to prevent the model from overfitting:

[0083]

[0084] where σ y is the standard deviation of ε i .

[0085] Furthermore, in a preferred embodiment, when using the GP-NARX model to model the input-output data, it can be assumed that the function f(·) is a zero-mean multivariate Gaussian prior distribution, and can be expressed by the following expression, for example:

[0086] f = f(x i ) ~ N(f|0, K)

[0087] In the formula, K is the covariance matrix, and K ij = k(x i , x j ), and k(·,·) is the kernel function, that is, the covariance function.

[0088] Next, select an appropriate kernel function for the GP-NARX model. The selection of the kernel function is crucial for the model performance and can be flexibly chosen according to the actual problem. For example, in a preferred embodiment, an appropriate kernel function is selected for the GP-NARX model, which may include, for example: selecting the Squared Exponential Kernel (SE Kernel) as the kernel function, and the kernel function can be expressed by the following formula:

[0089]

[0090] where D is the data dimension, and σ f , l are the hyperparameters to be optimized in the kernel function.

[0091] In a preferred embodiment, for the motor temperature estimation method provided by the present invention, after selecting an appropriate kernel function for the model, the working condition data of the training set sorted by the grey relational grade are input into the GP-NARX model; and the PSO optimization algorithm is used to determine the hyperparameters of the GP-NARX model, thereby completing the model training work.

[0092] It can be understood that in the previous steps, the motor temperature estimation method provided by the present invention has sorted multiple working condition parameters based on the grey relational grade. Subsequently, in an embodiment, preferably, according to the grey relational grade values between the sorted motor working condition parameters and the motor temperature, the parameter with the largest grey relational grade value, the parameters with the top two grey relational grade values are sequentially selected until all parameters are selected as the model input. In this way, the hyperparameter U to be optimized for determining the number of model input variables is obtained as follows:

[0093] U = [U 1 , U 2 , …, U M

[0094] In the formula, U 1 represents the working condition parameter with the largest grey relational grade value with the motor temperature, U 2 represents the working condition parameters with the top two grey relational grade values with the motor temperature, and U M represents the working condition parameters with the top M grey relational grade values with the motor temperature.

[0095] At this time, including the model input and the output feedback step length parameters n u , n y , and the hyperparameters σ f , l of the SE kernel function, and the standard deviation σ y of the model Gaussian white noise, a set of hyperparameters to be optimized in the GP-NARX thermal model in the motor temperature estimation method provided by the present invention can be expressed as θ = [U, n u ​,n y ,σ f ,l,σ y 。

[0096] Next, perform step 304: PSO initialization, randomly generate the initial population of the hyperparameters of the GP-NARX model. In one embodiment, first, the dimension of each particle can be determined according to the number of variables of the hyperparameters, and the initial population of the hyperparameters of the GP-NARX model is randomly generated, including the initial velocity and the initial position of each particle.

[0097] For example, in one embodiment, there are 5 variables in total for the hyperparameters θ to be optimized by the model, so the dimension of each particle is 5-dimensional. Each particle vector represents a set of hyperparameters of the GP-NARX model, and different GP-NARX models can be determined. Set the particle swarm size to N. Optionally, take N = 20, and the maximum number of iterations is K generations. Optionally, take K = 1000. The initial position and the initial velocity of the particle can be determined by the following formulas respectively:

[0098] x i0 = rand([U,n u ,n y ,σ f ,l,σ y )

[0099] V i0 = rand(V i )

[0100] In the formula, x i0 is the initial position of the particle randomly generated within the hyperparameter range, and V i0 is the initial velocity of the particle randomly generated.

[0101] Subsequently, perform step 305: Calculate the fitness value of each particle individual based on the fitness function. The fitness value of each particle individual is the mean square error f id of the motor temperature prediction.

[0102] For example, the mean square error of the motor temperature prediction can be calculated using the following formula:

[0103]

[0104] In the formula, f is the mean square error of the model's prediction of the motor temperature, f(x i ) is the predicted motor temperature value of the i-th sample during model training, y i is the true temperature value of the motor of the i-th sample, and M is the number of samples.

[0105] Further, perform step 306: determine whether the optimization termination condition is satisfied; if not, perform step 307: update the velocity and position of the particle, and update the global optimum and local optimum.

[0106] Specifically, for example, if the mean square error f of the motor temperature prediction corresponding to a particle id is less than the minimum temperature mean square error f(P id ) of the particle in the previous iteration history, then use this motor temperature mean square error f id to replace the f(P id ) of the previous round, use this particle to replace the particle of the previous round, and update the velocity and position of the particle. In one embodiment, optionally, the updating of the velocity and position of the particle may include: updating the velocity and position of the particle using the following formula:

[0107] V id (k + 1) = wV id (k) + c 1 r 1 (k)(P id (k) - X id (k)) + c 2 r 2 (k)(P gd (k) - X id (k))

[0108] X id (k + 1) = X id (k) + V id (k + 1)

[0109] where V id (k) is the velocity of the i-th particle in the d-th dimension at the k-th iteration, X id (k) is the position of the i-th particle in the d-th dimension at the k-th iteration, P id is the individual historical optimal position of the i-th particle in the d-th dimension, P gd is the optimal position experienced by all particles, c 1 and c 2 are learning factors determined according to practical experience, r 1 , r 2 are random numbers uniformly distributed between [0, 1], w is an inertia weight that linearly decreases according to the number of iterations, and can be expressed by the following formula:

[0110]

[0111] where w max , w min are the maximum and minimum inertia weights respectively, t is the current iteration number, and k max is the maximum number of iterations.

[0112] If the mean square error of the minimum motor temperature of a particle, f(P id ) is less than the global minimum mean square error of the motor temperature of all particles, f(P gd ), then replace the original global minimum mean square error of the motor temperature with the mean square error of the minimum motor temperature of this particle, and save the current position of this particle:

[0113] x gd = x id = [U id ,n u*id ,n y*id ,σ f*id ,l id ,σ y*id

[0114] where U is the optimal model input variable currently searched, n u and n y are the parameters of the optimal model output regression step length and input step length currently searched respectively, which determine the modeling ability of the model for the dynamic change characteristics of temperature; σ f and l are the optimal hyperparameters of the model kernel function currently searched, which determine the ability of the model to explore and map the law of motor temperature change contained in the training data; σ y The standard deviation of the Gaussian white noise of the currently searched model can enable the model to overcome the influence of measurement noise in the original data acquisition and improve the generalization ability of the model.

[0115] Then return to step 305 to calculate the next particle, and iterate and update in turn until the end condition of the optimization is met. For example, when the maximum number of iterations is reached or global convergence occurs, end the program and return the position of the particle with the smallest mean square error of the current motor temperature, that is, the optimal input and hyperparameters of the model obtained by PSO optimization, namely the optimal model output step length, the optimal input step length, the optimal kernel function parameters, and the optimal Gaussian white noise standard deviation. If the end condition is not met, perform a new round of iteration. At this time, update the position and velocity of the particle, generate a new particle, and return to step 305 until the maximum number of iteration steps or global convergence is reached. The algorithm ends, returns the optimal input variables and hyperparameters of the GP-NARX model, and determines the optimal GP-NARX model, which has the highest prediction accuracy for the motor temperature.

[0116] Using the PSO optimization algorithm to optimize the hyperparameters of the GP-NARX model reduces the difficulty of parameter optimization and speeds up the model training speed.

[0117] So far, the model training stage is completed, and transfer to step 309: enter the testing stage according to the determined GP-NARX model.

[0118] ​In the testing phase, the accuracy of the model is tested using the test set determined in the previous steps. As Figure 3 shown, steps 310 to 312 are executed in sequence: first, calculate the prior distribution and marginal likelihood distribution of the GP-NARX model; subsequently, calculate the posterior distribution of the target value. According to the Bayesian posterior formula, the posterior distribution of the predicted output y* of the model for any input x * satisfies:

[0119]

[0120] where μ * is the mean of the predicted value, is the variance of the predicted value, which can be calculated by the following formula:

[0121]

[0122]

[0123] Next, execute step 313: statistically analyze the test accuracy of the model. After calculating the predicted output y*, select a suitable evaluation index to measure the accuracy of the model. In one embodiment, optionally, the Root Mean Square Error (RMSE) function is used as the evaluation index, and the RMSE is calculated as follows:

[0124]

[0125] where N is the number of data samples in the test set, y*i is the predicted value of the model, and yi is the true value.

[0126] It should be noted that using RMSE as the prediction accuracy evaluation index here is only an exemplary illustration, aiming to more clearly show the specific steps of the motor temperature estimation method provided by the present invention in the model testing phase, rather than limiting the protection scope of the present invention. In fact, other evaluation indexes can also be selected to measure the accuracy of the model prediction.

[0127] If the judged prediction accuracy does not meet the preset requirements, return to step 302 to retrain the model. For example, the proportion or order of the training set and test set in the vehicle dataset can be changed and then retraining work can be carried out again; if after testing, the prediction accuracy of the model can meet the preset requirements, terminate the training and enter step 314: output the model configuration file and save it for use in the formal temperature estimation work.

[0128] The formal use of this GP-NARX model is to carry out the motor temperature estimation work based on this GP-NARX model, that is Figure 1 the content shown.

[0129] Specifically, since the input of the GP-NARX model includes the feedback value of the output, the initialization of the model needs to be performed first to obtain the initial temperature value as the model input. Then, it can return to Figure 1 , where step 102: Perform initialization to obtain the initial temperature value of the motor.

[0130] Figure 4 is a schematic flowchart of the initialization in the motor temperature estimation method according to an embodiment of the present invention. Next, in combination with Figure 4 this, the initialization will be elaborated.

[0131] As shown in Figure 4 , the initialization in the motor temperature estimation method provided by the present invention may include: First, step 401: The vehicle starts and runs, and the vehicle is powered on; after the motor is powered on, step 402 is executed: Obtain the vehicle power-off time tiEcuOff; then step 403 is executed: Compare the vehicle power-off time tiEcuOff with the preset time threshold tiMax_C; in response to the vehicle power-off time being not lower than the preset time threshold, enter step 404: Use the current temperature t coolt at the coolant inlet of the motor as the initial temperature value t init , that is, t init =t coolt ; at the same time, in response to the vehicle power-off time being lower than the preset time threshold, enter step 405: Select a motor cooling curve based on the real-time ambient temperature, and calculate the current temperature of the motor according to the vehicle power-off time and the motor cooling curve as the initial temperature value t init ; finally, step 406: Determine the initial temperature value t init of the model based on the above process as the initial feedback input of the model.

[0132] It should be additionally noted that in the motor temperature estimation method based on the GP-NARX model provided by the present invention, the GP-NARX thermal model is not only applicable to the motors of new energy vehicle drive systems, but also applicable to other occasions where the motor is used as a power source. Its modeling objects can include various types such as permanent magnet synchronous motors, asynchronous motors, reluctance motors, brushless DC motors, etc.

[0133] After the model initialization is completed, the corresponding module in the MCU collects the signals required for the model input, processes them into the format of the model input, calls the motor GP-NARX thermal model for calculation, obtains the real-time estimated value of the motor temperature, and uses it for motor control and thermal protection.

[0134] In addition, the input of the GP-NARX thermal model of the motor can be the operating condition information related to the motor temperature, the temperature predicted by the model, or a combination of such operating condition information. The predicted temperature can be any part of the motor where the temperature needs to be monitored, not limited to the temperatures of the stator three-phase windings, stator core, rotor permanent magnet, rotor core, bearings, coolant outlet, etc.

[0135] In one embodiment, finally, the operating condition parameters and the initial temperature value are input into the GP-NARX model trained through the training stage to estimate the motor temperature. Specifically, it may include: inputting the operating condition parameter data of the operating condition parameter categories optimized and screened based on the grey relational degree in the training stage into the GP-NARX model to predict the real-time temperature estimation value of the motor. That is to say, in the motor temperature estimation method provided by the present invention, the parameters input into the model when using the trained model to estimate the motor temperature are the same parameters optimized and selected through grey relational degree calculation in the training stage, so as to improve the working efficiency of the motor temperature estimation work and at the same time improve the accuracy of the motor temperature estimation.

[0136] The motor temperature estimation method based on the GP-NARX model provided by the present invention uses a data-driven modeling algorithm, avoiding the process of complex motor heat transfer path analysis and thermal network parameter identification, and is especially suitable for occasions where the physical model is particularly complex, improving the development efficiency. At the same time, the above-mentioned modeling process considers the uncertainty between the input and output, and can give the confidence interval of the output prediction result; and multiple-source information can be fused during model training, improving the applicability of the model under various operating conditions. In addition, once the model training is completed, the data-driven model can be used to replace the traditional physical temperature sensor, effectively realizing cost savings in the vehicle development stage.

[0137] Furthermore, the operating condition parameters related to the motor temperature are optimized and selected based on the grey relational degree as the model input variables, and the PSO optimization method is used to optimize the solution of the model parameters, making the motor temperature estimation work more accurate, fast and efficient.

[0138] Although the above methods are illustrated and described as a series of actions to simplify the explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or occur concurrently with other actions shown and described herein or not shown and described herein but understood by those skilled in the art.

[0139] Figure 5 It is a schematic diagram of the device structure of the motor temperature estimation device based on the GP-NARX model shown according to another aspect of the present invention.

[0140] According to another aspect of the present invention, an embodiment of a motor temperature estimation device 500 based on a GP-NARX model is also provided herein.

[0141] As Figure 5 shown, the above-mentioned motor temperature estimation device 500 based on a GP-NARX model provided in this embodiment may include a memory 501 and a processor 502 coupled to the memory 501. The processor 502 may be configured to implement any of the above-mentioned motor temperature estimation methods based on a GP-NARX model.

[0142] According to another aspect of the present invention, an embodiment of a computer storage medium is also provided herein. A computer program is stored on the computer storage medium. When the computer program is executed by a processor, the steps of any of the above-mentioned motor temperature estimation methods based on a GP-NARX model can be implemented.

[0143] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability of hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and the design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention. The processors described in this case can be implemented using electronic hardware, computer software, or any combination thereof. Whether such a processor is implemented as hardware or software will depend upon the particular application and the overall design constraints imposed on the system. As an example, the processors, any part of the processors, or any combination of the processors presented in this disclosure can be implemented using a microprocessor, a microcontroller, a digital signal processor (DSP), a field-programmable gate array (FPGA), a programmable logic device (PLD), a state machine, gated logic, discrete hardware circuits, and other suitable processing components configured to perform the various functions described throughout this disclosure. The functionality of the processors, any part of the processors, or any combination of the processors presented in this disclosure can be implemented using software executed by a microprocessor, a microcontroller, a DSP, or other suitable platform.

[0144] The steps of a method or algorithm described in connection with the embodiments disclosed in this specification can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read from, and write to, the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside as discrete components in a user terminal.

[0145] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions may be stored on or transmitted via a computer-readable medium as one or more instructions or code. The computer-readable medium includes both a computer storage medium and a communication medium including any medium that facilitates transfer of a computer program from one place to another. A storage medium may be any available medium that can be accessed by a computer. By way of example and not limitation, such computer-readable medium can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Any connection is properly termed a computer-readable medium.

[0146] For example, if the software is transmitted from a web site, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. As used herein, disk and disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray disc, where disk generally reproduces data magnetically, while disc uses lasers optically to reproduce data. Combinations of the above should also be included within the scope of computer-readable medium.

[0147] More specifically, for example, the model file completed by PSO optimization training in the previous step can be deployed to the micro - control unit (MCU) of the motor controller, and the motor GP - NARX thermal model can be deployed to run in the central processing unit (CPU) within the MCU.

[0148] Since the computing power required to run the motor GP - NARX model is relatively large, the model can also be accelerated by means of the hardware acceleration unit inside the MCU. The input processing, data conversion and processing of input and output in the NARX structure are run in the CPU. The CPU calls the driver program of the hardware acceleration unit to assist in the calculation of the GP part in the motor GP - NARX thermal model through the hardware acceleration unit, so as to accelerate the operation speed of the thermal model.

[0149] Figure 6 It is a schematic structural diagram of a hardware device for deploying the GP - NARX model according to an embodiment of the present invention.

[0150] As Figure 6 shown, it is the structure of the common hardware acceleration device AMU unit 601 (Advanced Modelling Unit) inside the vehicle - used MCU, which may include: several AMU computing engines AMU0..n (n = 1, 2, 3...), DMA unit 602 (Direct Memory Access), and SRAM unit 603 (System RAM, volatile memory). Among them, the AMU unit 601 can directly obtain the parameters of the motor GP - NARX thermal model from non - volatile memory 604, such as FLASH, for model operation, that is, the first path 605 shown by the arrow; it can also transfer the model parameters from the FLASH area to the SRAM unit 603 of the hardware acceleration unit through the DMA unit 602 of the AMU unit 601, and the AMU unit 601 obtains the parameters of the motor GP - NARX thermal model from the SRAM unit 603 for model operation, that is, the second path 606 shown by the arrow, which can shorten the time for the AMU computing engine to obtain the model parameters and further improve the model operation efficiency.

[0151] In addition, the GP-NARX model can also be deployed to processing units other than the MCU in the motor controller, including: a micro-processing unit (MPU), and other controllers outside the motor controller, such as a regional controller, a domain controller, an in-vehicle computer, and a cloud server, etc. Data interaction can be carried out between them and the motor controller MCU through a communication medium, such as SPI (Serial Peripheral Interface), UART (Universal Asynchronous Receiver / Transmitter), CAN (Controller Area Network), LIN (Local Interconnect Network), ETH (Ethernet), and wireless communication, etc., so as to realize real-time temperature prediction of the motor.

[0152] In one embodiment, after the model is deployed, data can be transmitted to the cloud, and through big data analysis, it can be judged whether there are problems such as excessive scatter and aging of the motor, and the model can be corrected.

[0153] The foregoing description of the disclosure has been provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A motor temperature estimation method based on the GP-NARX model, comprising: Obtaining operating parameters of the motor under predetermined operating conditions; Performing initialization work to obtain an initial temperature value of the motor; as well as The operating condition parameters and the initial temperature value are input into the GP-NARX model obtained through training in the training phase, and the final output value of the GP-NARX model is used as the real-time temperature estimation value of the motor. During operation, the GP-NARX model feeds back the motor temperature value output at the previous moment to the input at the next moment to reflect the influence of the motor temperature change on the real-time temperature of the motor; In the training stage, the plurality of operating parameters are optimized based on the grey correlation degree to select operating parameters that are more relevant to the motor temperature as training data input to the GP-NARX model to train the GP-NARX model.

2. The motor temperature estimation method according to claim 1, characterized in that: The optimizing of the plurality of operating parameters based on the grey correlation degree comprises: The following formulas are used to calculate the grey correlation values ​​of the plurality of operating parameters and the motor temperature: Among them, y(k) is the motor temperature, x i (k) is a plurality of operating parameters related to the motor temperature, and ρ is a gray resolution coefficient; and The plurality of operating condition parameters are sorted in descending order according to their corresponding grey correlation values ​​to obtain the hyperparameter U to be optimized for determining the number of input variables of the GP-NARX model: U=[U1,U2,...,U M ] Among them, U1 represents the operating parameter with the largest gray correlation value with the motor temperature, U2 represents the operating parameters with the top two gray correlation values ​​with the motor temperature, and so on. M Indicates the operating parameters with the top M gray correlation values ​​with the motor temperature.

3. The motor temperature estimation method according to claim 2, characterized in that: The training phase includes: Collecting multiple groups of working condition data of the motor in actual operation, dividing the multiple groups of working condition data into multiple data sets, and performing optimization sorting based on the grey correlation degree, wherein the multiple data sets include a training set; Selecting an adaptive kernel function for the GP-NARX model, and inputting the operating condition data of the training set sorted by grey correlation degree into the GP-NARX model; and The PSO optimization algorithm is used to determine the hyperparameters of the GP-NARX model, thereby completing the model training work.

4. The motor temperature estimation method according to claim 3, characterized in that: The multiple groups of operating condition data include operating condition parameters under multiple operating conditions sorted by grey correlation and the real-time temperature of the motor under each operating condition. The training stage also includes: The multiple groups of working condition data sorted by grey correlation degree are preprocessed and sorted into (x i ,y i ) in time series form, x i is the model input, y i is the real-time temperature of the motor collected by the sensor, x i It is expressed as: in, are the operating parameters of the motor under different operating conditions used to input the GP-NARX model, is the real-time temperature of the motor used as input for feedback to the GP-NARX model, n u and n y are parameters used to determine the output regression step size and input step size of the GP-NARX model respectively.

5. The motor temperature estimation method according to claim 4, characterized in that: The real-time temperature of the motor i It is expressed as: y i =f(x i )+ε i Where, f(·) represents the GP-NARX model, ε i is the added Gaussian white noise with zero mean, and ε i The following expression is satisfied to prevent the model from overfitting: where σ y is i The standard deviation of .

6. The motor temperature estimation method according to claim 5, characterized in that: The GP-NARX model also satisfies the following expression: f=f(x i )~N(f|0,K) Among them, K is the covariance matrix, K ij = k(x i ,x j ), k(·,·) is the kernel function; The step of selecting an adapted kernel function for the GP-NARX model comprises: The square exponential function is selected as the kernel function, and the kernel function is expressed by the following formula: Where D is the data dimension, σ f , l are the hyperparameters to be optimized in the kernel function.

7. The motor temperature estimation method according to claim 6, characterized in that: The hyperparameters of the GP-NARX model are θ = [U, n u ,n y ,σ f ,l,σ y ], where n u 、n y are the step sizes of the input and output feedback of the GP-NARX model respectively. The PSO optimization algorithm is used to determine the hyper parameters of the GP-NARX model, including: Determine the dimension of each particle according to the number of variables of the hyperparameters, and randomly generate an initial population of the hyperparameters of the GP-NARX model, including the initial velocity and initial position of each particle; The fitness value of each individual particle is calculated based on the fitness function, and the fitness value of each individual particle is the mean square error f of the motor temperature prediction. id ; If the mean square error of the motor temperature prediction corresponding to a particle is f id Less than the minimum temperature mean square error f(P id ), then use the motor temperature mean square error f id Replace the previous round of f(P id ), use this particle to replace the particle in the previous round, and update the particle's speed and position; If the minimum motor temperature mean square error f(P id ) is less than the global minimum motor temperature mean square error f(P gd ), the original global minimum motor temperature mean square error is replaced by the minimum motor temperature mean square error of the particle, and the current position of the particle is saved; and When the preset optimization end condition is met, the current position of the particle is used as the optimal GP-NARX model input and hyperparameters obtained by optimization.

8. The motor temperature estimation method according to claim 7, characterized in that: The updating of the speed and position of the particle includes: Use the following formula to update the particle's velocity and position: V id (k+1)=wV id (k)+c1r1(k)(P id (k)-X id (k))+c2r2(k)(P gd (k)-X id (k)) X id (k+1)=X id (k)+V id (k+1) Among them, V id (k) is the velocity of the ith particle in the dth dimension at the kth iteration, X id (k) is the position of the ith particle in the dth dimension at the kth iteration, P id is the individual historical optimal position of the ith particle in the dth dimension, P gd is the optimal position experienced by all particles, c1 and c2 are learning factors determined based on practical experience, r1 and r2 are uniformly distributed random numbers between [0,1], and w is the inertia weight that decreases linearly according to the number of iterations, expressed as follows: Among them, w max 、w min are the maximum and minimum inertia weights respectively, t is the current iteration number, k max is the maximum number of iterations.

9. The motor temperature estimation method according to claim 3, characterized in that: The plurality of data sets also include a test set, and the motor temperature estimation method further includes: After the GP-NARX model is trained in the training phase, the trained GP-NARX model is verified in the testing phase; If the trained GP-NARX model meets the preset verification condition during the test phase, the operating condition parameter and the initial temperature value are input into the GP-NARX model, and the final output value of the model is used as the real-time temperature estimation value of the motor; and If the trained GP-NARX model does not meet the preset verification condition during the testing phase, the GP-NARX model is retrained.

10. The motor temperature estimation method according to claim 9, characterized in that: The preset verification conditions include: The following error function RMSE is used as an evaluation index to measure the prediction accuracy of the GP-NARX model: Where N represents the number of data sets in the test set, y i represents the real-time temperature value of the motor measured by the sensor, y i * indicates the predicted output value of the GP-NARX model.

11. The motor temperature estimation method according to claim 1, characterized in that: The step of inputting the operating condition parameter and the initial temperature value into the GP-NARX model obtained through training in the training phase includes: The operating parameter data of the operating parameter category selected based on the grey correlation degree optimization in the training phase is input into the GP-NARX model to predict the real-time temperature estimation value of the motor.

12. The motor temperature estimation method according to claim 1, characterized in that: The performing of the initialization work to obtain the initial temperature value of the motor includes: The vehicle is powered on, and a power-off time of the vehicle is obtained, and the power-off time of the vehicle is compared with a preset time threshold; In response to the vehicle power-off time being not less than the preset time threshold, taking the current temperature of the motor cooling water inlet as the initial temperature value; and In response to the vehicle power-off time being lower than the preset time threshold, a motor cooling curve is selected based on the real-time ambient temperature, and the current temperature of the motor is calculated according to the vehicle power-off time and the motor cooling curve as the initial temperature value.

13. A motor temperature estimation device based on GP-NARX model, comprising: Memory; as well as A processor coupled to the memory, wherein the processor is configured to execute the motor temperature estimation method based on the GP-NARX model according to any one of claims 1 to 12.

14. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the motor temperature estimation method based on the GP-NARX model as claimed in any one of claims 1 to 12 is implemented.

Images