A method and system for controlling motor rotor temperature prediction

Abstract

The application relates to a motor rotor temperature prediction method and system for control, and the method comprises the following steps: rotor temperature measurement tests are carried out on the motor under different working conditions; during the motor test, the electrical, mechanical and temperature time sequence data of the tested motor are synchronously collected to establish a sample data set; according to the sample data set, a pre-established deep learning model is trained to obtain a trained motor rotor temperature prediction model; motor parameters are collected in real time, then are loaded into the motor rotor temperature prediction model to obtain an electronic rotor temperature prediction result, and motor control is carried out according to the electronic rotor temperature prediction result. Compared with the prior art, the actual motor operation data is used as the sample data set, the deep learning method is used to mine data features and rules, the complex coupling relationship between the motor parameters and the rotor temperature is accurately modeled, the precision of the motor rotor temperature prediction is improved, and the performance of the motor control is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of motor control, in particular to a motor rotor temperature prediction method and system for control. BACKGROUND

[0002] As an important device for electromechanical energy conversion, motor has the advantages of simple structure, high power density and high torque density, and is widely used in electric vehicles, electric ships, wind power generation and other fields. However, in the process of motor operation, losses are inevitable, and the losses are converted into heat in the motor. The accumulated heat makes the temperature of each part of the motor rise. High temperature will cause damage to the motor, mainly in the following two aspects:

[0003] (1) High temperature causes the insulation of stator winding to age and fall off, and in severe cases, the stator winding is burned out.

[0004] (2) High temperature causes irreversible demagnetization of rotor permanent magnet or rotor winding failure, reduces the torque output capability of the motor, and in severe cases, the motor loses working ability and is scrapped.

[0005] For the stator winding of the motor, as a non-rotating part, its temperature monitoring is relatively easy, and a certain number of temperature sensors are generally embedded in the stator winding to measure the winding temperature in real time. However, as a high-speed rotating part, it is very difficult to directly measure the temperature of the rotor. Generally, a corresponding measurement scheme needs to be customized according to the structure of the motor, including a specific thermocouple, a rotor shaft with a hole, a high-speed slip ring or a wireless signal transmission device, etc. These greatly increase the cost of motor manufacturing, and are limited by space in actual application. The above-mentioned direct measurement method for rotor temperature is only limited to laboratory or special application occasions. Therefore, it is necessary to study the indirect measurement method of motor rotor temperature.

[0006] At present, the research on motor rotor temperature prediction methods mainly includes flux linkage observer, signal injection method and lumped parameter thermal network model.

[0007] (1) The flux linkage observer uses the known relationship between the permanent magnet flux linkage of the permanent magnet motor and the rotor temperature to estimate the permanent magnet flux linkage through the design of the flux linkage observer, and then inversely calculates the rotor permanent magnet temperature. However, this method is very sensitive to measurement errors, and is not suitable for motor static and low-speed working conditions.

[0008] (2) The signal injection method injects a direct current or high-frequency voltage signal into the motor, and determines the rotor temperature according to the response of the system to the injected signal. However, injecting the corresponding signal into the motor will interfere with the operation of the motor and cause additional losses.

[0009] (3) lumped parameter thermal network model analog circuit model, the points with similar temperature in the motor are equivalent to a node, the nodes are connected through thermal resistance, representing different heat transfer paths in the motor, the heat storage efficiency of different materials is replaced by heat capacity, and finally a thermal network model describing the temperature distribution of each node is obtained. This method requires professional knowledge of motors, understanding of the geometric structure and parameters of the motor, and an accurate motor loss model, which limits the application of this method. SUMMARY

[0010] The purpose of the present application is to overcome the defects of the prior art, i.e., the need for professional knowledge of motors and an accurate motor loss model, which limits the application of the prior art method, and to provide a motor rotor temperature prediction method and system for control.

[0011] The purpose of the present application can be achieved by the following technical solutions:

[0012] A motor rotor temperature prediction method for control, comprising the following steps:

[0013] Motor test step: measuring the rotor temperature of the motor under different operating conditions, including the motor operating in different states, different speeds and output torques, different DC bus voltages, different cooling liquid flow rates, different cooling liquid temperatures, and different ambient temperatures;

[0014] Data acquisition step: synchronously collecting electrical, mechanical, and temperature time series data of the test motor during the motor test step, and establishing a sample data set;

[0015] Data modeling step: training a pre-established deep learning model according to the sample data set to obtain a trained motor rotor temperature prediction model;

[0016] Model application step: real-time acquisition of motor parameters, then loading into the motor rotor temperature prediction model, obtaining electronic rotor temperature prediction results, and performing motor control according to the electronic rotor temperature prediction results.

[0017] Further, the motor includes a permanent magnet motor, an induction motor, a reluctance motor, and a DC motor.

[0018] Further, in the data acquisition step, the synchronous collection of electrical, mechanical, and temperature time series data of the test motor is specifically: setting the data acquisition frequency, collecting electrical, mechanical, and temperature parameters of the motor under different operating conditions at certain time intervals, and saving the time series data in the form of time series data, and the time reference of the time series data is synchronized.

[0019] Further, the data modeling step specifically includes the following sub-steps:

[0020] a data preprocessing step of filling missing values, removing outliers, smoothing data, normalizing data, and arranging data into a form suitable for input into a machine learning model;

[0021] a data division step of dividing the sample data set processed by the data preprocessing step into a training set, a validation set, and a test set;

[0022] a model training step of loading the training set into a pre-established deep learning model for training, the model training process including a forward error calculation and an error backpropagation, the model parameter values being updated during the error backpropagation, and the model being validated by the validation set to determine whether a preset model training termination condition is met, thereby determining the model parameters;

[0023] a model validation step of loading the test set into the trained deep learning model, evaluating the model performance using a preset model performance evaluation index, and verifying the generalization ability of the deep learning model.

[0024] Further, in the model training step, an optimization algorithm for model training is used to solve an optimization problem of the deep learning model with a loss function as the target, find a set of model parameter values that minimize the target function value, update the model parameter values, and the optimization algorithm includes an SGD algorithm, an Adagrad algorithm, an RMSProp algorithm, or an Adam algorithm.

[0025] Further, the model performance evaluation index includes a mean absolute error (MAE), a root mean square error (RMSE), a standardized root mean square error (SRMSE), a standard deviation (STDDEV), a mean square error (MSE), a mean absolute percentage error (MAPE), or a maximum absolute error (MAX).

[0026] Further, the deep learning model is an LSTM network, which includes a fully connected layer and a plurality of LSTM memory unit combinations, each of which is used to process a type of time series data in the sample data set, and the output of each LSTM memory unit combination is connected to the fully connected layer and then outputs a motor rotor temperature prediction value.

[0027] The LSTM memory unit combination includes a first LSTM memory unit and a second LSTM memory unit, the input of the first LSTM memory unit is the time series data at the current time and the output of the first LSTM memory unit hidden layer at the previous time, and the output of the first LSTM memory unit is the output of the first LSTM memory unit hidden layer at the current time.

[0028] The input of the second LSTM memory unit is the output of the first LSTM memory unit hidden layer at the current time and the output of the second LSTM memory unit hidden layer at the previous time, and the output is the output of the second LSTM memory unit hidden layer at the current time.

[0029] Further, the first LSTM memory unit and the second LSTM memory unit are both LSTM unit structures, the LSTM unit structure comprises a forgetting gate, an input gate, an output gate and a current candidate memory unit, and the calculation expression of the hidden layer output of the LSTM unit structure is:

[0030] h t =o t ⊙tanh(c t )

[0031]

[0032] In the formula, h t is the output of the hidden layer at the current time t, o t is the output result of the output gate at the current time t, c t is the state information of the current memory unit at the current time t, f t is the output result of the forgetting gate at the current time t, c t-1 is the state information of the current memory unit at t-1, i t is the output result of the input gate at the current time t, is the tanh layer state of the current candidate memory unit at the current time t.

[0033] Further, the termination condition of the model training is that the loss function no longer decreases with the increase of the iteration round number, at which time the model training is stopped.

[0034] The application also provides a motor rotor temperature prediction system using the motor rotor temperature prediction method for control.

[0035] A motor test unit is used to measure the rotor temperature of the motor under different working conditions, wherein the different working conditions include different states, different rotating speeds and output torques, different DC bus voltages, different cooling liquid flow rates, different cooling liquid temperatures and different ambient temperatures.

[0036] A data acquisition and processing unit is used to synchronously acquire the electrical, mechanical and temperature time series data of the tested motor during the motor test step, and to establish a sample data set.

[0037] The workstation is used for training a pre-established deep learning model according to the sample data set, so as to obtain a trained motor rotor temperature prediction model; motor parameters are collected in real time, and then loaded into the motor rotor temperature prediction model to obtain an electronic rotor temperature prediction result;

[0038] The man-machine interaction unit is used for displaying the electronic rotor temperature prediction result, generating motor control parameters according to the electronic rotor temperature prediction result, and performing motor control;

[0039] The data transmission unit is used to realize data transmission among the data acquisition and processing unit, the workstation and the man-machine interaction unit.

[0040] Compared with the prior art, the motor is tested under various working conditions, and the motor rotor temperature and related electrical, mechanical and environmental temperature parameters under different working conditions are synchronously collected to form a data set for establishing a machine learning model. The related electrical, mechanical and environmental temperature parameters are used as model inputs, and the motor rotor temperature is used as model output, so that the machine learning model is trained and verified, the trained model is called to realize real-time online prediction of the motor rotor temperature, and the following advantages are obtained:

[0041] (1) The data generated in the actual operation of the motor is used for modeling, more uncertain factors are considered in the time series data, and the obtained model has stronger robustness;

[0042] (2) Multiple factors affecting the motor rotor temperature are fused into a model and used as model inputs, and the obtained model has stronger adaptability to working conditions;

[0043] (3) No professional background knowledge in the motor field is needed, and the internal structure of the motor does not need to be clearly understood, and the data-driven method is directly used for modeling;

[0044] (4) The related parameters in the operation of the motor are obtained as model inputs, the data characteristics and rules are mined by using the deep learning method, the complex coupling relationship between the motor parameters and the rotor temperature is accurately modeled, the rotor temperature change of the motor is predicted in real time, the predicted temperature is used for motor control, the precision of the motor rotor temperature prediction is improved, and the performance of the motor control is improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 It is a structure schematic view of a motor rotor temperature prediction system for control provided in the embodiment of the application;

[0046] Figure 2 It is a RNN neural network structure schematic view provided in the embodiment 1 of the application;

[0047] Figure 3A memory cell structure diagram of an LSTM provided in embodiment 1 of the present application;

[0048] Figure 4 A schematic diagram of an LSTM network model provided in embodiment 1 of the present application for motor rotor temperature prediction;

[0049] Figure 5 A flow chart of an LSTM network training process provided in embodiment 1 of the present application;

[0050] Figure 6 A schematic diagram of loss function changes in an LSTM training process provided in embodiment 1 of the present application;

[0051] Figure 7 A schematic diagram of model predicted values and real values provided in embodiment 1 of the present application;

[0052] Figure 8 A schematic diagram of an LSTM motor thermal model for motor control provided in embodiment 2 of the present application;

[0053] Figure 9 A block diagram of a motor control with an LSTM motor thermal model provided in embodiment 2 of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0055] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0056] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0057] The present embodiment provides a motor rotor temperature prediction method for control, comprising the following steps:

[0058] Motor test step: rotor temperature measurement test is carried out on the motor under different working conditions, including different states of the motor, different rotating speeds and output torques, different DC bus voltages, different cooling liquid flow rates, different cooling liquid temperatures and different ambient temperatures, comprehensively considering the influencing factors of the rotor temperature of the motor; the motor includes a permanent magnet motor, an induction motor, a reluctance motor and a DC motor;

[0059] Data acquisition step: during the motor test step, the electrical, mechanical and temperature time series data of the tested motor are synchronously collected by using voltage sensors, current sensors, temperature sensors and rotating speed-torque sensors, and a sample data set for machine learning model training and verification is established;

[0060] Data modeling step: according to the sample data set, the pre-established deep learning model is trained to obtain a trained motor rotor temperature prediction model;

[0061] Model application step: real-time motor parameters are collected, then loaded into the motor rotor temperature prediction model to obtain electronic rotor temperature prediction results, and the motor is controlled according to the electronic rotor temperature prediction results.

[0062] The data modeling step specifically includes the following sub-steps:

[0063] Data preprocessing step: the sample data set is filled with missing values, outliers are removed, data is smoothed, data is standardized, and the data is arranged into a form suitable for machine learning model input;

[0064] Data division step: the sample data set processed by the data preprocessing step is divided into a training set, a validation set and a test set; wherein the training set and the validation set are used for machine learning model training to determine model parameters, and the test set is used to test the generalization ability of the model to determine whether the model can be used finally;

[0065] Model training step: the training set is loaded into the pre-established deep learning model for training, and the model training process includes forward error calculation and error backpropagation, the model parameter values are updated in the error backpropagation process, and the model is verified by the validation set to determine whether the preset model training termination condition is met, thereby determining the model parameters;

[0066] Model verification step: the test set is loaded into the trained deep learning model, and the model performance is evaluated by using a preset model performance evaluation index to verify the generalization ability of the deep learning model. The model performance evaluation index can be selected from the group consisting of mean absolute error MAE, root mean square error RMSE, standardized root mean square error SRMSE, standard deviation STDDEV, mean square error MSE, mean absolute percentage error MAPE and maximum absolute error MAX.

[0067] In the data acquisition step, the electrical, mechanical and temperature time series data of the motor of the subject are synchronously collected, specifically: setting the data acquisition frequency, collecting the electrical, mechanical, temperature and other parameters of the motor under different operating conditions at certain time intervals, and saving the time series data in the form of time series data, and the time reference of the time series data is synchronized.

[0068] In the model training step, an optimization algorithm for model training is used to solve the optimization problem of the deep learning model with the loss function as the target, find a set of model parameter values that minimize the target function value, update the model parameter values, and the optimization algorithm includes SGD algorithm, Adagrad algorithm, RMSProp algorithm or Adam algorithm.

[0069] The termination condition of model training is that the loss function no longer decreases with the increase of the number of iterations, at which point the model training is stopped.

[0070] In recent years, artificial intelligence algorithms such as machine learning have developed rapidly and been widely applied, especially in complex system modeling, which has shown unparalleled advantages. Therefore, when the time series data of the relevant electrical, mechanical, temperature and other parameters of the motor under different operating conditions is obtained, the thermal model for motor rotor temperature prediction is trained based on the above artificial intelligence algorithm, and the motor rotor temperature is predicted. Compared with other models, the machine learning algorithm can extract the implicit nonlinear relationship between multiple variables through "learning" of the training data, accurately model the complex relationship between different variables, and improve the accuracy of motor rotor temperature prediction.

[0071] Example 1

[0072] In this embodiment, a long short-term memory network (LSTM) model is selected as a deep learning model for rotor temperature prediction, and the LSTM is improved based on a recurrent neural network (RNN).

[0073] RNN network is a deep learning neural network suitable for processing time series data, which is connected by weight between hidden layer neurons, realizes weight sharing, can process time series correlation of data, and the typical RNN structure is shown in the accompanying Figure 2 However, due to the simple structure of RNN, it cannot solve the long-term time dependence between data, and the problem of gradient disappearance or explosion will occur during model training. Therefore, people improve RNN and propose LSTM network. LSTM introduces a gating unit in its own structure, controls the flow of information by controlling the state of the forget gate, input gate and output gate, forms a processing method for long-term memory or forgetting of time series data, processes long-term dependence in time series data, and solves the problem of gradient disappearance or gradient explosion in RNN training process.

[0074] In this embodiment, the LSTM network adopted includes a full connection layer and a plurality of LSTM memory unit combinations, each of which is used to process one kind of time series data in the sample data set, and the output of each LSTM memory unit combination is connected to the full connection layer to output the motor rotor temperature prediction value;

[0075] The LSTM memory unit combination includes a first LSTM memory unit and a second LSTM memory unit, the input of the first LSTM memory unit is the time series data at the current time and the output of the first LSTM memory unit hidden layer at the previous time, and the output is the output of the first LSTM memory unit hidden layer at the current time.

[0076] The input of the second LSTM memory unit is the output of the first LSTM memory unit hidden layer at the current time and the output of the second LSTM memory unit hidden layer at the previous time, and the output is the output of the second LSTM memory unit hidden layer at the current time.

[0077] The first LSTM memory unit and the second LSTM memory unit are both LSTM unit structures, which include a forgetting gate, an input gate, an output gate and a current candidate memory unit, as shown in the accompanying drawings. Figure 3 The internal structure of the LSTM unit structure is shown in the accompanying drawings, where x t is the model input data at the current t time, h t-1 is the hidden layer output at the previous time, is the current candidate memory unit state, c t is the current memory unit state information, i t is the input gate, f t is the forgetting gate, o t is the output gate, and h t is the hidden layer output at the current time.

[0078] The forgetting gate f t decides how much the state of the memory unit at the last time is forgotten, and controls the degree of information transfer from the previous time to the current time, as shown in the following formula:

[0079] f t = σ (W f x t + U f h t-1 + b f )

[0080] The input gate i t controls how much the state of the current candidate memory unit is updated to the state of the current memory unit, and supplements the memory from the current input information, as shown in the following formula:

[0081] it = σ(W i x t + U i h t-1 + b i )

[0082] Output gate o t controls how much of the current output depends on the state of the current memory cell, as follows:

[0083] o t = σ(W o x t + U o h t-1 + b o )

[0084] State information of the current candidate memory cell The calculation formula is as follows:

[0085]

[0086] State information c t of the current memory cell is determined by two parts: (1) the product of the output result i t of the input gate and the output result of the tanh layer; and (2) the product of the output result f t of the forget gate and the state c t-1 of the memory cell at the previous moment. The calculation formula of the state c t of the current memory cell is as follows:

[0087]

[0088] The current hidden layer output h t is determined by the product of the output result o t of the output gate and the state information tanh(c t ) of the current memory cell after the tanh layer, and the calculation formula is as follows:

[0089] h t = o t ⊙ tanh(c t )

[0090] In the above formula, W and U represent the corresponding weight matrices, b is the corresponding bias vector, σ represents the activation function of the three gate units, which is the Sigmoid function, h t is the output of the hidden layer at the current moment t, o t is the output result of the output gate at the current moment t, c t is the state information of the current memory cell at the current moment t, f tis the output result of the forget gate at the current time t, c t-1 is the state information of the current memory cell at time t-1, i t is the output result of the input gate at the current time t, is the tanh layer state of the current candidate memory cell at the current time t.

[0091] According to the foregoing, as an example but not limitation, as shown in the accompanying drawings, a motor rotor temperature prediction method for control can be implemented by an LSTM network, and the specific process is as follows: Figure 4

[0092] (1) Motor test: measure the rotor temperature of the motor under various different working conditions to generate time series data of the motor operation. Specifically, in embodiment 1, different working conditions include different states of the motor (motor, generator), different speeds and output torques, different cooling liquid temperatures, and different ambient temperatures.

[0093] (2) Data acquisition: use voltage sensors, current sensors, temperature sensors, and speed-torque sensors to synchronously collect time series data of electrical, mechanical, and temperature of the tested motor in the motor test. Specifically, in embodiment 1, the above parameters include motor d-q axis voltage, d-q axis current, motor speed ω, motor cooling water temperature T water , and motor ambient temperature T amb , to establish a sample data set for machine learning model training and verification.

[0094] (3) Data modeling, in embodiment 1, specifically includes the following steps:

[0095] 31) Data preprocessing: data preprocessing refers to filling missing values, removing outliers, smoothing data, and standardizing data. Specifically, in embodiment 1, the mean and standard deviation of time series data are used for standardization, which can be calculated by the following formula:

[0096]

[0097] In the formula, x t ' is the standardized time series data, is the mean of the time series data, and σ is the standard deviation of the time series data.

[0098] For the data input form of the LSTM network, the time series sample data is arranged as follows: [sample data, time series length, feature number].

[0099] ​32) Data division: the data division includes dividing the above-mentioned pre-processed data into a training set, a validation set and a test set. In this embodiment 1, the training set accounts for 60% of the sample data, the validation set accounts for 20% of the sample data, and the test set accounts for 20% of the sample data.

[0100] 33) Model training: the training process of the LSTM network model of this embodiment 1 is shown in FIG. 6. The optimization algorithm of model training is Adam algorithm. The process of model training includes forward error calculation and error back propagation. The model parameter value is updated once in the process of error back propagation. In this embodiment 1, the LSTM network is a three-layer structure, and the random initialization method is used for parameter initialization. The loss function of the LSTM network model is defined as the mean square error function MSE, which can be calculated by the following formula: Figure 5

[0101]

[0102] In the formula, y i is the predicted value of the model for the input, is the true value corresponding to the input data, and N is the sample number. Referring to FIG. 7, the change of the loss function MSE with the iteration number in the training process of the LSTM model can be seen. It can be seen that the loss function continuously decreases with the increase of the iteration number. When the loss function no longer decreases with the increase of the iteration number, the model training is terminated, and the model validation link is entered. Figure 6

[0103] 34) Model validation: in order to verify the generalization ability of the model, this embodiment 1 selects the root mean square error RMSE as the evaluation index of the model. The evaluation index RMSE can be calculated by the following formula:

[0104]

[0105] In the formula, y i is the predicted value of the model for the input, is the true value corresponding to the input data, and N is the sample number.

[0106] The model trained in step 33) is tested on the validation set. Once the model output predicted value is compared with the true value, as shown in FIG. 8, when the evaluation index RMSE reaches the set threshold, the model training link is ended, the model is saved, and the result is input to the data transmission unit and displayed on the human-computer interaction unit. Figure 7

[0107] (4) Model application: the trained and validated model is saved. By calling the model, the collected motor related parameters are used as the model input, the motor rotor temperature is predicted in real time through model operation, the temperature signal is input to the motor controller, and the control of the motor is realized. ​​​

[0108] As shown in the accompanying drawings, the motor ambient temperature, motor cooling water temperature, motor speed, motor d-q axis voltage, motor d-q axis current are used as model inputs, and the motor rotor temperature is used as model output. A LSTM-based thermal model is trained and verified in the workstation. Figure 8

[0109] A control block diagram for using the above LSTM thermal model for motor control is shown in the accompanying drawings. The LSTM thermal model predicts the motor rotor temperature in real time and feeds the temperature signal back to the motor controller, which can be used for motor output torque compensation control and motor thermal management. Figure 9

[0110] Embodiment 2

[0111] The embodiment provides a motor rotor temperature prediction system for control, which comprises:

[0112] (1) Motor test unit: used for temperature measurement test of the motor under various different working conditions, to test the temperature change of the motor under different working conditions, especially the motor rotor temperature change of the motor under different working conditions; equivalent to performing the above motor test step.

[0113] (2) Data acquisition and processing unit: including voltage, current, speed-torque, temperature sensors, and data acquisition equipment, which can synchronously acquire the motor electrical, mechanical, temperature and other related parameters during the motor test process, and can set the data acquisition frequency. The unit can also convert the voltage / current signals output by the sensors into corresponding quantities with actual physical meaning; equivalent to performing the above data acquisition step.

[0114] (3) Data transmission unit: used to realize data transmission between the data acquisition and processing unit, the workstation, and the man-machine interaction unit. As an example but not limitation, the data transmission mode can be CAN bus, and the data transmission protocol can be TCP / IP protocol.

[0115] (4) Workstation: used to perform various steps of data modeling, including time series data preprocessing, data division, machine learning model training, machine learning model testing, model saving, and outputting model prediction results to the data transmission unit; equivalent to performing the above data modeling steps and model application steps.

[0116] (5) Man-machine interaction unit: used for data display and input of system related control parameters. As an example but not limitation, the man-machine interaction unit can be composed of a display, a keyboard, and a mouse.

[0117] ​​The preferred embodiments of the present application have been described above in detail. It should be understood that modifications and variations to the preferred embodiments could be made by those skilled in the art in light of the teachings above without departing from the spirit of the present application. It is therefore to be understood that what is desired to be protected by letters patent is defined by the scope of the claims that follow and that on behalf of which the summary of the application is intended to serve.

Claims

1. A method for controlling a temperature prediction of a rotor of an electric machine, characterized in that The method comprises the following steps: Motor test step: rotor temperature measurement test is performed on the motor under different working conditions, including different states, different speeds and output torques, different DC bus voltages, different cooling liquid flow rates, different cooling liquid temperatures, and different ambient temperatures; Data acquisition step: during the motor test step, the electrical, mechanical, and temperature time series data of the test motor are synchronously collected to establish a sample data set; Data modeling step: according to the sample data set, a pre-established deep learning model is trained to obtain a trained motor rotor temperature prediction model; Model application step: real-time motor parameters are collected, then loaded into the motor rotor temperature prediction model to obtain an electronic rotor temperature prediction result, and motor control is performed according to the electronic rotor temperature prediction result; The data modeling step specifically comprises the following sub-steps: Data preprocessing step: the sample data set is filled with missing values, outliers are removed, data is smoothed, data is standardized, and the data is arranged into a form suitable for machine learning model input; Data division step: the sample data set processed by the data preprocessing step is divided into a training set, a validation set, and a test set; Model training step: the training set is loaded into a pre-established deep learning model for training, the model training process includes forward error calculation and error backpropagation, the model parameter values are updated during error backpropagation, and the model is verified through the validation set to determine whether the preset model training termination condition is met, thereby determining the model parameters; Model verification step: the test set is loaded into the trained deep learning model, a preset model performance evaluation index is used to evaluate the model performance, and the generalization ability of the deep learning model is verified; In the model training step, an optimization algorithm for model training is used to solve the optimization problem of the deep learning model with a loss function as the target, find a set of model parameter values that minimize the target function value, update the model parameter values, and the optimization algorithm includes SGD algorithm, Adagrad algorithm, RMSProp algorithm, or Adam algorithm; The deep learning model is an LSTM network, which comprises a full connection layer and a plurality of LSTM memory unit combinations, each of which is used to process a kind of time series data in the sample data set, and the output of each LSTM memory unit combination is connected to the full connection layer and then outputs the motor rotor temperature prediction value; The LSTM memory unit combination comprises a first LSTM memory unit and a second LSTM memory unit, the input of the first LSTM memory unit is the time series data at the current time and the output of the first LSTM memory unit hidden layer at the previous time, and the output of the first LSTM memory unit hidden layer at the current time. The input of the second LSTM memory unit is the output of the first LSTM memory unit hidden layer at the current time and the output of the second LSTM memory unit hidden layer at the previous time, and the output is the output of the second LSTM memory unit hidden layer at the current time. The motor includes a permanent magnet motor, an induction motor, a reluctance motor and a direct current motor; In the data acquisition step, the synchronous acquisition of the electrical, mechanical and temperature time series data of the test motor specifically includes: setting the data acquisition frequency, collecting the electrical, mechanical and temperature parameters of the motor under different operating conditions at certain time intervals, and saving the time series data in the form of time series data, and the time reference of the time series data is synchronized; The model performance evaluation indicators include mean absolute error (MAE), root mean square error (RMSE), standardized root mean square error (SRMSE), standard deviation (STDDEV), mean square error (MSE), mean absolute percentage error (MAPE) or maximum absolute error (MAX); The first LSTM memory unit and the second LSTM memory unit are both LSTM unit structures, which include a forgetting gate, an input gate, an output gate and a current candidate memory unit, and the calculation expression of the hidden layer output of the LSTM unit structure is: h t = o t ⊙ tanh(c t ) wherein h t is the output of the hidden layer at the current time t, o t is the output result of the output gate at the current time t, c t is the state information of the current memory cell at the current time t, f t is the output result of the forget gate at the current time t, c t-1 is the state information of the current memory cell at the time t-1, i t is the output result of the input gate at the current time t, is the tanh layer state of the current candidate memory cell at the current time t; The termination condition of the model training is that the loss function no longer decreases with the increase of the iteration number, at which time the model training is stopped.

2. A motor rotor temperature prediction system employing a motor rotor temperature prediction method for control according to claim 1, characterized by, It comprises: A motor test unit for rotor temperature measurement test of the motor under different operating conditions, the different operating conditions including different states of the motor, different rotational speeds and output torques, different DC bus voltages, different cooling liquid flow rates, different cooling liquid temperatures and different ambient temperatures; A data acquisition and processing unit for synchronously acquiring electrical, mechanical and temperature time series data of the test motor during the motor test step, and establishing a sample data set; A workstation for training a pre-established deep learning model according to the sample data set to obtain a trained motor rotor temperature prediction model, real-time acquisition of motor parameters, and then loading the motor rotor temperature prediction model to obtain electronic rotor temperature prediction results; A human-computer interaction unit for displaying the electronic rotor temperature prediction results, generating motor control parameters according to the electronic rotor temperature prediction results, and controlling the motor; A data transmission unit for realizing data transmission among the data acquisition and processing unit, the workstation and the human-computer interaction unit. A data transmission unit for realizing data transmission among the data acquisition and processing unit, the workstation and the human-computer interaction unit.

Images