Accurate prediction method of transient temperature field of electric drive transmission considering time-space correlation characteristics

Through the spatiotemporal graph convolution thermal neural network model and C-mean clustering algorithm based on least squares method, combined with temperature sensor data and thermal network topological edge weight matrix, the accuracy and cost problems of the prediction of transient temperature field of the electric drive transmission are solved. , realizes high-precision temperature field prediction.

CN115081308BActive Publication Date: 2024-11-12TONGJI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210471076.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-29
Publication Date
2024-11-12
Estimated Expiration
2042-04-29

AI Technical Summary

Technical Problem

The prior art cannot accurately and quickly predict the transient temperature field of the electric drive gearbox, especially when running at high loads and high speeds, which leads to difficulty in thermal management. The existing methods fail to effectively consider the spatial and temporal correlation characteristics, making it difficult to apply to the electric drive gearbox. In the real car box.

Method used

The spatiotemporal graph convolution thermal neural network (OLS-TGCN) model is adopted based on least squares method, combining temperature sensor data and thermal network topological edge weight matrix, and the spatiotemporal correlation characteristics of temperature nodes are captured through graph convolution and gated cycle units to establish Transient temperature field prediction model, and error compensation is performed using C-mean clustering algorithm to improve prediction accuracy.

Benefits of technology

It realizes high-precision and low-cost transient temperature field prediction of electric drive gearboxes, which can accurately predict temperature changes in a certain time in the future, reduce thermal prediction costs, and maximize prediction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115081308B_ABST
    Figure CN115081308B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for accurately predicting the transient temperature field of an electric drive gearbox considering the time-space correlation characteristics, comprising: step 1: obtaining real transient temperature field data and preprocessing the obtained data; step 2: establishing a time-space graph convolutional thermal neural network transient temperature field prediction model based on the least squares method; step 3: using partially obtained sensor information and the thermal network topology edge weight matrix as input, and obtaining the transient temperature field data of the electric drive gearbox within a certain time scale in the future through the temperature prediction model established in step 2; using the possible C-means clustering algorithm to explore the error distribution law of the prediction model, and performing error compensation on the predicted temperature value, so as to further improve the prediction accuracy of the temperature prediction model. Compared with the prior art, the present invention has the advantages of high prediction accuracy and low cost.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of transient temperature field prediction of an electric drive gearbox for a vehicle, and in particular to a method for accurately predicting the transient temperature field of an electric drive gearbox taking into account time-space correlation characteristics. Background Art

[0002] As a key component of new energy vehicles such as pure electric vehicles, hybrid electric vehicles, and plug-in hybrid electric vehicles, electric drive transmissions have been widely used. Electric drive transmissions have evolved from preliminary structural integration to deep system integration, from the initial two-in-one design to a three-in-one design and a multi-in-one design, gradually achieving low redundancy and high cost performance of electric drive systems. However, while electric drive transmissions are moving towards a higher degree of integration, they also face corresponding problems: components such as motors and inverters will generate high heat losses when working, especially when running at high loads and high speeds. If the heat is not dissipated in time, the high temperature will cause the permanent magnets of the motors to demagnetize, and in severe cases, the motors and inverters will have over-temperature faults. Therefore, accurate monitoring and prediction of the transient temperature field of the electric drive transmission can achieve effective thermal management of the electric drive transmission and ensure the thermal safety of the electric drive transmission.

[0003] Prior art CN113379041A discloses a method for predicting the temperature of a train traction motor based on a neural network algorithm built on PyTorch, which mainly improves the temperature prediction accuracy from the perspective of data preprocessing such as temperature data resampling and sampling using the nearest method, but because it does not simultaneously consider the spatiotemporal correlation characteristics between data points, and the prediction time interval it targets is relatively long (in days), it is not suitable for short-term rapid temperature prediction of an electric drive gearbox (in seconds). CN105160109A discloses a method for predicting the temperature rise of a motor based on an RBF neural network, which predicts the temperature rise parameters of a modern automobile window motor by constructing an RBF neural network, and at the same time substitutes the parameters predicted by the neural network into the constructed motor temperature rise mathematical model to achieve real-time prediction of the motor temperature. However, the invention does not consider the spatiotemporal correlation characteristics between the input parameters, and is unable to predict the temperature of important motor components and the overall transient temperature field changes. CN110412462A discloses a method for predicting transient temperature of permanent magnet synchronous motor for aviation, which is based on curve fitting and loss calculation of current source method / voltage source method, motor three-dimensional model design and finite element model meshing, cooling and heat dissipation conditions and fluid-solid coupling analysis, and temperature distribution of each motor component for transient temperature prediction to obtain high-precision temperature distribution. However, this method requires detailed motor material and structural parameters to be known in advance, and the use of finite element software takes a lot of time, and cannot be directly applied to the rapid prediction of transient temperature field of electric drive gearbox in real vehicles. CN202011402714.7 discloses a method for predicting temperature of permanent magnet synchronous motor, which uses a data set to train the constructed PSNLSTMs model, and the left and right sides of the trained PSNLSTMs model are input according to the same weight for regression prediction to obtain the predicted temperature of the permanent magnet synchronous motor at the next moment. This method only considers the time series correlation characteristics between temperature nodes, does not consider its spatial correlation characteristics, and ignores the factors that affect each other in the real temperature field. At the same time, it cannot realize the estimation of overall temperature node information using partial temperature node information. CN113720495A discloses a real-time temperature estimation method for a motor rotor. The method constructs a motor temperature estimation model based on a motor loss model and a thermal circuit model. The method uses the motor current, speed, ambient temperature, and other hot spot temperatures that can be measured by low-cost sensors during operation as inputs to estimate key hot spot temperatures such as the motor winding temperature, stator core temperature, and rotor magnetic steel temperature. The invention has a relatively complex process for solving the loss model and the thermal circuit model, does not consider the temporal and spatial correlation characteristics between temperature nodes, and cannot predict future temperatures. The above-mentioned prior arts only predict the temperature of a single motor. It is difficult to migrate such methods to a more complex electric drive gearbox, and their applicability has certain limitations.

[0004] In fact, the current mainstream trend is to use some temperature sensors and temperature estimation methods to monitor the temperature of motors and inverters, and to cool them by active cooling. However, due to layout and cost constraints, most of the temperature points are measured by a small number of temperature sensors, which makes it difficult to accurately estimate the overall temperature field. In addition, since the future transient temperature field of the electric drive transmission cannot be quickly and accurately predicted, it is impossible to effectively and actively suppress abnormal temperature rise in the future. Therefore, it is very important to propose a prediction method that can effectively predict the overall transient temperature field changes of the electric drive transmission within a certain time scale in the future by relying only on a small amount of temperature sensor data. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method for accurately predicting the transient temperature field of an electric drive transmission with high prediction accuracy and low cost and taking into account the time-space correlation characteristics.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for accurately predicting the transient temperature field of an electric drive transmission considering the temporal and spatial correlation characteristics, the prediction method comprising:

[0008] Step 1: Obtain real transient temperature field data and preprocess the acquired data;

[0009] Step 2: Establish a transient temperature field prediction model of spatiotemporal graph convolutional thermal neural network based on least squares method;

[0010] Step 3: Take part of the acquired sensor information and the thermal network topology edge weight matrix as input, and obtain the transient temperature field data of the electric drive transmission within a certain time scale in the future through the temperature prediction model established in step 2; use the possible C-means clustering algorithm to explore the error distribution law of the prediction model, and perform error compensation on the predicted temperature value to further improve the prediction accuracy of the temperature prediction model.

[0011] Preferably, the step 1 is specifically:

[0012] Build an electric drive gearbox transient temperature field data acquisition platform, arrange temperature sensors at key points inside the electric drive gearbox, obtain real transient temperature field data, and classify, filter and normalize the acquired data.

[0013] More preferably, the method for obtaining the real transient temperature field data in step 1 is specifically:

[0014] Step 1-1: Keep the ambient temperature constant by using a temperature control device in a semi-anechoic chamber environment;

[0015] Step 1-2: perform data calibration;

[0016] Step 1-3: The battery simulator supplies high voltage to the motors MG1 and MG2 and the oil pump in the electric drive gearbox, and the PUMA system controls the output voltage of the battery simulator through CAN; the rapid prototype controller communicates with the oil pump controller through CAN C to control the speed and direction of the oil pump motor; the three groups of working conditions of NEDC, UDDS and WLTC are loaded in the PUMA system respectively; the rapid prototype controller calculates the torque distribution value of the motors MG1 and MG2 at the next moment according to the speed information, and controls the motor controller in the electric drive gearbox in real time through CAN A;

[0017] Step 1-4: Electric drive gearbox transient temperature field data acquisition;

[0018] Step 1-5: Preprocess all types of collected data, save them as original data sets according to sampling time and data category, filter and normalize the acquired data sets, and divide the data sets into known temperature node data and temperature node data to be predicted. At the same time, divide the entire data set into test set, verification set and training set according to working conditions.

[0019] More preferably, the filtering and normalization processing method is:

[0020] The original multidimensional time series data set collected in step 1 is filtered to delete low-quality data points, and all data are normalized for model training and testing.

[0021] More preferably, the raw data include stator winding temperature, stator yoke temperature, stator tooth temperature, rotor permanent magnet temperature, coolant temperature, housing temperature of motors MG1 and MG2, air temperature inside motors MG1 and MG2, inverter junction temperature, inverter internal temperature, front end, rear end, upper end and lower end temperatures outside the gearbox housing, front end, rear end, upper end and lower end temperatures outside the gearbox, three-phase currents of motors MG1 and MG2, and actual speed and torque information of motors MG1 and MG2.

[0022] Preferably, the step 2 is specifically:

[0023] Step 2-1: Establish a temporal-spatial graph convolutional thermal neural network OLS-TGCN transient temperature field prediction model based on the least squares method;

[0024] Step 2-2: In order to minimize the error between the temperature field data actually measured by each sensor and the predicted value, the loss function of the OLS-TGCN model is constructed, the prediction model is trained and tested through the error back propagation algorithm, and hyperparameter optimization is used to improve the prediction accuracy of the OLS-TGCN model.

[0025] More preferably, the transient temperature field prediction model is specifically:

[0026]

[0027] in, represents the temperature estimation process based on the least squares method; Indicates that based on some temperature sensor data X at time t t Estimated other temperature node data; X t1 Represents all temperature node information at time t β is the model coefficient; f(A,X t1 ) represents the graph convolution process, A represents the edge weight matrix of the hot network topology, represents the preprocessing step; is a matrix with a self-connected structure, is the degree matrix; u t 、r t With c t They are the update gate, reset gate and memory gate at time t, respectively. t represents the hidden layer output at time t, h t-1 represents the hidden layer output at time t-1, W u , W r , W c , W0 both represent the weights of the prediction model, b u 、b r 、b c , b represents the bias of the prediction model, σ(·), tanh(), and Relu() represent activation functions.

[0028] Preferably, the step 3 is specifically:

[0029] Step 3-1: Construct a spatial correlation measurement function that considers the similarity of temperature rise and the similarity of heating rate, quantify the spatial correlation between the temperature of the temperature node to be predicted and the temperatures of other nodes, and solve the topological edge weight matrix of the thermal network of the electric drive gearbox;

[0030] Step 3-2: Import part of the acquired sensor information and the thermal network topology edge weight matrix as input into the OLS-TGCN transient temperature field prediction model to predict the transient temperature field data of the electric drive transmission within a certain time scale in the future;

[0031] Step 3-3: Use the possible C-means clustering algorithm to cluster the historical temperature data set predicted by the spatiotemporal graph convolutional neural network temperature prediction model to obtain the membership matrix and cluster center value;

[0032] Step 3-4: Calculate the average of the third dimension error values ​​of each clustered class and use it as the error range of the predicted temperature data;

[0033] Step 3-5: Select new input data, calculate its membership matrix and cluster center according to step 3-3, calculate the similarity between the new data and the historical data, and classify and mark the new input;

[0034] Step 3-6: Calculate the error to be compensated by taking into account that the error of the temperature predicted by the new input data and the historical input has a similar error distribution characteristic;

[0035] Step 3-7: Perform temperature compensation on the temperature value predicted by the OLS-TGCN prediction model according to the calculated error to be compensated, and output it as the final prediction result.

[0036] More preferably, the step 3-1 is specifically as follows:

[0037] Step 3-1-1: Construct a spatial correlation measurement function that considers the similarity of temperature rise and the similarity of heating rate, and quantify the spatial correlation between the temperature of the temperature node to be predicted and the temperatures of other nodes;

[0038] The temperature rise similarity T Rn The expression is:

[0039]

[0040] Among them, X0(t) and X p (t) are the temperature change history data sets of the target heat path and the surrounding associated heat path in a certain period of time;

[0041] The temperature rise rate similarity TV n for:

[0042]

[0043] Among them, T V0 (i) and T Vn (i) are the temperature rise rates of the target heat path and the adjacent heat path in the period [tT, t], and They represent the average temperature rise rates of the target heat path and the surrounding associated heat path in the period [tT, t] respectively;

[0044] The spatial correlation metric function is:

[0045]

[0046] Among them, corr n represents the spatial correlation metric value of the nth associated hot path and the target hot path;

[0047] Step 3-1-2: Construct a complete thermal network graph containing all temperature nodes based on the training data set, and learn the spatial correlation of each node through the spatial correlation measurement function described in step 3-1-1, prune the thermal paths with low correlation, reduce the complexity of the thermal network, and use the spatial correlation measurement values ​​between each temperature node as weights to determine the thermal network topology edge weight matrix.

[0048] More preferably, the step 3-1 is specifically as follows:

[0049] Step 3-2-1: Use the least squares method to estimate the information of l2 temperature nodes to be predicted at time i through the information of l1 known temperature nodes at time i in the historical data;

[0050] Let p = l 1+ l2, taking the information of p temperature nodes at time i as input, the prediction error of p temperature points at time j is obtained by the spatiotemporal graph convolutional thermal neural network temperature prediction model:

[0051]

[0052] in, x p,j is the actual temperature value of the pth temperature node at the jth moment; is the predicted value of the pth temperature node at the jth moment;

[0053] Step 3-2-2: Input the p temperature nodes at time i The prediction errors of the predicted values ​​of p temperature nodes at j moments form a new data set:

[0054]

[0055] Step 3-2-3: Combine the thermal network topology edge weight matrix obtained in step 3-1-2 and the complete temperature field data obtained in step 3-2-2 as input and import them into the OLS-TGCN model to output the transient temperature field data of the electric drive transmission within a certain time scale in the future.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] High prediction accuracy and low cost: The spatiotemporal graph convolutional network transient temperature field prediction model based on least squares method proposed in the present invention can estimate the transient temperature field of the electric drive gearbox by relying only on data collected in real time by a small number of sensors, further reducing the thermal prediction cost of the transient temperature field of the electric drive gearbox. At the same time, by considering the spatiotemporal correlation characteristics of different temperature nodes and using temperature compensation to correct the prediction results of the prediction model, the prediction accuracy of the transient temperature field change of the electric drive gearbox can be maximized. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic flow chart of the method for accurately predicting the transient temperature field of an electric drive gearbox in the present invention;

[0059] Figure 2 Schematic diagram of a transient temperature field data acquisition platform for an electric drive gearbox in an embodiment of the present invention;

[0060] Figure 3 Schematic diagram of the structure of a gated cycle unit in an embodiment of the present invention;

[0061] Figure 4 Schematic diagram of the structure of a spatiotemporal graph convolutional thermal neural network based on the least squares method in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0063] like Figure 1 As shown, the accurate prediction method of the transient temperature field of the electric drive gearbox considering the time-space correlation characteristics includes three parts, namely, the transient temperature field data collection and processing of the electric drive gearbox, the establishment of the temperature field prediction model, and the transient temperature field prediction of the electric drive gearbox. At the same time, the prediction method specifically includes the following steps:

[0064] Step 1: Build an electric drive gearbox transient temperature field data acquisition platform, arrange temperature sensors at key points inside the electric drive gearbox, obtain real transient temperature field data, and classify, filter and normalize the acquired data.

[0065] Step 2: Considering that the least squares method has the advantage of quickly estimating other temperature information through a small amount of temperature information, and combining the characteristics of graph convolutional neural network and gated recurrent unit, which have the characteristics of learning the spatial correlation characteristics and temporal correlation characteristics of each temperature node data respectively, a spatiotemporal graph convolutional thermal neural network transient temperature field prediction model based on the least squares method is established, and the prediction accuracy of the temperature prediction model is improved through model training and hyperparameter optimization.

[0066] Step 3: Use part of the acquired sensor information and the thermal network topology edge weight matrix as input, and obtain the transient temperature field data of the electric drive transmission within a certain time scale in the future through the temperature prediction model established in step 2; at the same time, use the possible C-means clustering algorithm to explore the error distribution law of the prediction model, perform error compensation on the predicted temperature value, and further improve the prediction accuracy of the temperature prediction model.

[0067] Step 1 is as follows:

[0068] Step 1-1: In a semi-anechoic chamber, the ambient temperature is kept constant by using a temperature control device.

[0069] Step 1-2: Host computer 1 completes data calibration through CANape software and hardware.

[0070] Step 1-3: The battery simulator supplies high voltage to the motors MG1 and MG2 and the oil pump in the electric drive transmission, and the PUMA system controls the output voltage of the battery simulator through CAN. The rapid prototype controller communicates with the oil pump controller through CAN C to control the speed and direction of the oil pump motor. The three groups of working conditions, NEDC, UDDS and WLTC, are loaded in the PUMA system respectively. The rapid prototype controller calculates the torque distribution value of the motors MG1 and MG2 at the next moment based on the speed information, and controls the motor controller in the electric drive transmission in real time through CAN A.

[0071] Step 1-4: Data collection of transient temperature field of electric drive gearbox. The data collection principle is as follows:

[0072] The rapid prototype controller receives the feedback signal of the motor controller in the electric drive transmission (the actual speed and torque information of the MG1 and MG2 motors) in real time through CAN A and collects data in the host computer 1 through CANape. The PUMA system receives the output voltage and current of the battery simulator and the signal of the speed and torque sensor on the rear axle in real time through CAN. The host computer 2 communicates with the USB to RS485 through Modbus to monitor and record the values ​​of various temperature sensors inside and outside the electric drive transmission in real time.

[0073] Step 1-5: Sort all kinds of data and save them into original data sets according to sampling time and data category. Filter and normalize the acquired data sets, and divide the data sets into known temperature node data and temperature node data to be predicted. At the same time, divide the entire data set into test set, validation set and training set according to working conditions.

[0074] The data filtering and normalization process described in steps 1-5 is specifically as follows:

[0075] The original multi-dimensional time series data set collected in steps 1-4 is filtered, low-quality data points are deleted, and all data are normalized. The original data includes the stator winding temperature of motor MG1 and MG2, the stator yoke temperature, the stator tooth temperature, the rotor permanent magnet temperature, the coolant temperature, the housing temperature of motor MG1 and MG2, the air temperature inside motor MG1 and MG2, the junction temperature of inverter, the internal temperature of inverter, the front end, rear end, upper end and lower end temperature of the outer housing of gearbox, the front end, rear end, upper end and lower end temperature of the outer air of gearbox, the three-phase current of motor MG1 and MG2, the actual speed and torque of motor MG1 and MG2, etc. According to actual needs, the stator yoke temperature, stator tooth temperature, rotor permanent magnet temperature and inverter internal temperature of motor MG1 and MG2 are used as the temperature node information to be predicted, and the remaining data are used as the known temperature node information. The entire data set is divided into a test set, a validation set, and a training set according to the working conditions. 70% of the transient temperature field test data based on NEDC and UDDS working conditions is used as a training set, and the remaining 30% of the data is used as a validation set to verify its generalization ability during the iterative training of the model to determine whether to stop training the model. At the same time, the transient temperature field test data based on the WLTC working condition is used as a test set to evaluate the generalization ability of the model.

[0076] Step 2 is as follows:

[0077] Step 2-1: Establish a transient temperature field prediction model based on the spatiotemporal graph convolutional thermal neural network (OLS-TGCN) based on the least squares method.

[0078] Step 2-2: In order to minimize the error between the temperature field data actually measured by each sensor and the predicted value, the loss function of the OLS-TGCN model is constructed, the prediction model is trained and tested through the error back propagation algorithm, and hyperparameter optimization is used to improve the prediction accuracy of the OLS-TGCN model.

[0079] Step 2-1 is as follows:

[0080] Step 2-1-1: Analyze the spatial correlation characteristics between each temperature node through the graph convolution neural network model. The graph convolution process can be defined as follows:

[0081]

[0082] Where X represents the feature matrix with information of each temperature node, A represents the edge weight matrix of the thermal network topology, represents the preprocessing step, is a matrix with a self-connected structure, is the degree matrix. W0 indicates that the matrix represents the weight from the input to the output layer, and Relu() indicates the activation function.

[0083] Step 2-1-2: Use Figure 3 The gated recurrent unit shown in FIG. 1 obtains the time series characteristics of the temperature field change, thereby predicting the temperature field change in the short term in the future based on some historical data, where h t-1 represents the hidden state at time t-1; x t Represents the temperature field information at time t; r t is the reset gate, which is used to control the degree of ignoring the state information of the previous moment; u t is the update gate, which is used to control the extent to which the state information of the previous moment enters the current state; c t is the memory content stored at all times (memory gate); h t is the output state (output gate) at the moment. This model not only captures the transient temperature field information at the current moment, but also retains the changing trend of the historical temperature information, and has the ability to capture time correlation.

[0084] Step 2-1-3: Combining the corresponding characteristics of the least squares method, graph convolutional neural network and gated recurrent unit, according to the spatiotemporal correlation characteristics of temperature field changes, a spatiotemporal graph convolutional thermal neural network (OLS-TGCN) transient temperature field prediction model based on the least squares method is established. The specific model expression is:

[0085]

[0086] in represents the temperature estimation process based on the least squares method, Indicates that based on some temperature sensor data X at time t t Estimated other temperature node data, X t1 represents all temperature node information at time t, β is the model coefficient; f(A,X t1 ) represents the graph convolution process, see step 2-1-1 for a detailed explanation. t 、r t With c t They are the update gate, reset gate and memory gate at time t, respectively. t represents the hidden layer output at time t, h t-1 represents the hidden layer output at time t-1, W u, W r , W c , W0 both represent the weights of the prediction model, b u , b r , b c , b represents the deviation of the prediction model, σ(·), tanh(), Relu() represent activation functions. At the same time, the model structure diagram is as follows Figure 4 As shown in the figure, the upper side of the spatiotemporal graph convolutional thermal neural network prediction model is the transient temperature field prediction process, and the lower side is the OLS-TGCN neuron. OLS represents the process of temperature estimation using the least squares method, TGCN is the graph convolution process, and X t2 represents the temperature node information predicted based on all temperature node information at time t, Y t Represents the temperature prediction output at time t.

[0087] Step 3 is as follows:

[0088] Step 3-1: Construct a spatial correlation measurement function that considers the similarity of temperature rise and the similarity of heating rate, quantify the spatial correlation between the temperature of the temperature node to be predicted and the temperatures of other nodes, and solve the topological edge weight matrix of the thermal network of the electric drive transmission.

[0089] Step 3-2: Import part of the acquired sensor information and the thermal network topology edge weight matrix as input into the OLS-TGCN transient temperature field prediction model to predict the transient temperature field data of the electric drive transmission within a certain time scale in the future.

[0090] Step 3:3: Use the Possibility C-means clustering algorithm to cluster the historical temperature dataset predicted by the spatiotemporal graph convolutional neural network temperature prediction model to obtain the membership matrix and cluster center values.

[0091] Step 3-4: Calculate the average of the third dimension error values ​​of each clustered class and use it as the error range of the predicted temperature data.

[0092] Step 3-5: Select new input data, calculate its membership matrix and cluster center according to step 3-1, calculate the similarity between the new data and the historical data, and classify and label the new input.

[0093] Step 3-6: Calculate the error to be compensated by taking into account that the error between the new input data and the temperature predicted by the historical input has a similar error distribution characteristic.

[0094] Step 3-7: Perform temperature compensation on the temperature value predicted by the OLS-TGCN prediction model according to the calculated error to be compensated, and output it as the final prediction result.

[0095] Step 3-1 is as follows:

[0096] Step 3-1-1: Construct a spatial correlation measurement function that considers the similarity of temperature rise and the similarity of heating rate to quantify the spatial correlation between the temperature of the temperature node to be predicted and the temperatures of other nodes.

[0097] The temperature rise similarity T Rn The expression is:

[0098]

[0099] Where X0(t) and X p (t) are the historical data sets of temperature changes (temperature rise) of the target heat path and the surrounding associated heat paths within a certain period of time.

[0100] The temperature rise rate similarity TV n for:

[0101]

[0102] Among them, T V0 (i) and T Vn (i) are the temperature rise rates of the target heat path and the adjacent heat path in the period [tT, t], and They represent the average temperature rise rates of the target heat path and the surrounding associated heat path within the period [tT, t] respectively.

[0103] The spatial correlation metric function is:

[0104]

[0105] Among them, corr n Indicates the spatial correlation measure between the nth associated hot path and the target hot path.

[0106] Step 3-1-2: Construct a complete thermal network graph containing all temperature nodes based on the training data set, and learn the spatial correlation of each node through the spatial correlation measurement function described in step 3-1-1, and prune the thermal paths with low correlation to reduce the complexity of the thermal network. At the same time, use the spatial correlation measurement values ​​between each temperature node as weights to determine the thermal network topology edge weight matrix.

[0107] Step 3-2 is as follows:

[0108] Step 3-2-1: Combine the least squares method to estimate the information of l2 temperature nodes to be predicted at time i through the information of l1 known temperature nodes at time i in the historical data. Let p = l 1+l2, taking the information of p temperature nodes at time i as input, the prediction error of p temperature points at time j is obtained by the spatiotemporal graph convolutional thermal neural network temperature prediction model

[0109] in x p,j is the actual temperature value of the pth temperature node at the jth moment, is the predicted value of the pth temperature node at the jth moment.

[0110] Step 3-2-2: Input the p temperature nodes at time i The prediction errors of the predicted values ​​of p temperature nodes at j moments form a new data set

[0111] Step 3-2-3: Combine the thermal network topology edge weight matrix obtained in step 2-1-5 and the complete temperature field data obtained in step 2-3-2 as input and import them into the OLS-TGCN model to output the transient temperature field data of the electric drive transmission within a certain time scale in the future.

[0112] Step 3-3 is as follows:

[0113] Step 3-3-1: Create a membership matrix U = [u z,s ], where u z,s ∈(0,1), indicating the degree of membership of the zth temperature data to the sth class.

[0114] Step 3-3-2: Set the objective function of the Possibility C-means clustering algorithm to:

[0115]

[0116] Where i+j is the size of the data set, K is the number of cluster centers, and m is the weighted index.

[0117] Step 3-3-3: Set the constraints and optimization objectives of the possible C-means clustering algorithm. The constraints are The optimization goal is

[0118] Step 3-3-4: Determine the membership matrix U and cluster center C of the fuzzy C-means algorithm according to the Lagrangian method, where u z,s for c k for

[0119] Step 3-3-5: Calculate the cluster center C by initializing the parameters and update the membership matrix U. Take the optimal objective function as the iteration condition, continuously update C and U until their values ​​converge, and then output the values ​​of U and C.

[0120] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics, characterized in that: The prediction method includes: Step 1: Obtain real transient temperature field data and preprocess the acquired data; Step 2: Establish a transient temperature field prediction model of spatiotemporal graph convolutional thermal neural network based on least squares method; Step 3: Take part of the acquired sensor information and the thermal network topology edge weight matrix as input, and obtain the transient temperature field data of the electric drive transmission within a certain time scale in the future through the temperature prediction model established in step 2; use the possible C-means clustering algorithm to explore the error distribution law of the prediction model, and perform error compensation on the predicted temperature value to further improve the prediction accuracy of the temperature prediction model; The step 2 is specifically as follows: Step 2-1: Establish a temporal-spatial graph convolutional thermal neural network OLS-TGCN transient temperature field prediction model based on the least squares method; Step 2-2: In order to minimize the error between the temperature field data actually measured by each sensor and the predicted value, the loss function of the OLS-TGCN model is constructed, the prediction model is trained and tested through the error back propagation algorithm, and the hyperparameter optimization is used to improve the prediction accuracy of the OLS-TGCN model; The transient temperature field prediction model is specifically: in, represents the temperature estimation process based on the least squares method; Indicates that based on the partial temperature sensor data X at time t t Other estimated temperature node data; X t1 represents all temperature node information at time t; β is the model coefficient; f(A,X t1 ) represents the graph convolution process, A represents the edge weight matrix of the electric drive transmission thermal network topology solved after quantifying the spatial correlation between the temperature of the temperature node to be predicted and the temperatures of other nodes, represents the preprocessing step; is a matrix with a self-connected structure, I N is the identity matrix, D is the degree matrix; u t 、r t With c t They are the update gate, reset gate and memory gate at time t, respectively. t represents the hidden layer output at time t, h t-1 represents the hidden layer output at time t-1, W u , W r , W c , W0 both represent the weights of the prediction model, b u , b r , b c , b represents the bias of the prediction model, σ(·), tanh(), and Relu() represent activation functions.

2. The method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics according to claim 1 is characterized in that: The step 1 is specifically as follows: Build an electric drive gearbox transient temperature field data acquisition platform, arrange temperature sensors at key points inside the electric drive gearbox, obtain real transient temperature field data, and classify, filter and normalize the acquired data.

3. The method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics according to claim 2 is characterized in that: The method for obtaining the real transient temperature field data in step 1 is specifically as follows: Step 1-1: Keep the ambient temperature constant by using a temperature control device in a semi-anechoic chamber environment; Step 1-2: perform data calibration; Step 1-3: The battery simulator supplies high voltage to the motors MG1 and MG2 and the oil pump in the electric drive gearbox. The PUMA system controls the output voltage of the battery simulator through CAN. The rapid prototype controller communicates with the oil pump controller through CAN C to control the speed and direction of the oil pump motor. The three operating conditions of NEDC, UDDS and WLTC are loaded in the PUMA system respectively. The rapid prototype controller calculates the torque distribution value of motors MG1 and MG2 at the next moment based on the speed information, and controls the motor controller in the electric drive transmission in real time through CAN A; Step 1-4: Electric drive gearbox transient temperature field data acquisition; Step 1-5: Preprocess all types of collected data, save them as original data sets according to sampling time and data category, filter and normalize the acquired data sets, and divide the data sets into known temperature node data and temperature node data to be predicted. At the same time, divide the entire data set into training set, verification set and test set according to working conditions.

4. The method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics according to claim 2 is characterized in that: The filtering and normalization processing method is: The original multidimensional time series data set collected in step 1 is filtered to delete low-quality data points, and all data are normalized for model training and testing.

5. The method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics according to claim 3 is characterized in that: The raw data include stator winding temperature, stator yoke temperature, stator tooth temperature, rotor permanent magnet temperature, coolant temperature, housing temperature of motors MG1 and MG2, air temperature inside motors MG1 and MG2, inverter junction temperature, inverter internal temperature, front end, rear end, top end and bottom end temperature of the exterior of the gearbox housing, front end, rear end, top end and bottom end temperature of the exterior air of the gearbox, three-phase current of motors MG1 and MG2, and actual speed and torque information of motors MG1 and MG2.

6. The method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics according to claim 1 is characterized in that: The step 3 is specifically as follows: Step 3-1: Construct a spatial correlation measurement function that considers the similarity of temperature rise and the similarity of heating rate, quantify the spatial correlation between the temperature of the temperature node to be predicted and the temperatures of other nodes, and solve the topological edge weight matrix of the thermal network of the electric drive gearbox; Step 3-2: Import part of the acquired sensor information and the thermal network topology edge weight matrix as input into the OLS-TGCN transient temperature field prediction model to predict the transient temperature field data of the electric drive transmission within a certain time scale in the future; Step 3-3: Use the possible C-means clustering algorithm to cluster the historical temperature data set predicted by the spatiotemporal graph convolutional neural network temperature prediction model to obtain the membership matrix and cluster center value; Step 3-4: Calculate the average of the third dimension error values ​​of each clustered class and use it as the error range of the predicted temperature data; Step 3-5: Select new input data, calculate its membership matrix and cluster center according to step 3-3, calculate the similarity between the new data and the historical data, and classify and mark the new input; Step 3-6: Calculate the error to be compensated by taking into account that the error of the temperature predicted by the new input data and the historical input has a similar error distribution characteristic; Step 3-7: Perform temperature compensation on the temperature value predicted by the OLS-TGCN prediction model according to the calculated error to be compensated, and output it as the final prediction result.

7. The method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics according to claim 6 is characterized in that: The step 3-1 is specifically as follows: Step 3-1-1: Construct a spatial correlation measurement function that considers the similarity of temperature rise and the similarity of temperature rise rate, and quantify the spatial correlation between the temperature of the temperature node to be predicted and the temperatures of other nodes; The temperature rise similarity T Rn The expression is: Among them, X0(t) and X p (t) are the temperature change history data sets of the target heat path and the surrounding associated heat path in a certain period of time; The temperature rise rate similarity T Vn for: Among them, T V0 (i) and T Vn (i) are the temperature rise rates of the target heat path and the adjacent heat path in the period [tT, t], and They represent the average temperature rise rates of the target heat path and the surrounding associated heat path in the period [tT, t] respectively; The spatial correlation metric function is: Among them, corr n represents the spatial correlation metric value of the nth associated hot path and the target hot path; Step 3-1-2: Construct a complete thermal network graph containing all temperature nodes based on the training data set, and learn the spatial correlation of each node through the spatial correlation measurement function described in step 3-1-1, prune the thermal paths with low correlation, reduce the complexity of the thermal network, and use the spatial correlation measurement values ​​between each temperature node as weights to determine the thermal network topology edge weight matrix.

8. The method for accurately predicting the transient temperature field of an electric drive gearbox considering the temporal and spatial correlation characteristics according to claim 6 is characterized in that: The step 3-2 is specifically as follows: Step 3-2-1: Use the least squares method to estimate the information of l2 temperature nodes to be predicted at time i through the information of l1 known temperature nodes at time i in the historical data; Let p = l 1+ l2, taking the information of p temperature nodes at time i as input, the prediction error of p temperature points at time j is obtained by the spatiotemporal graph convolutional thermal neural network temperature prediction model: in, x p,j is the actual temperature value of the pth temperature node at the jth moment; is the predicted value of the pth temperature node at the jth moment; Step 3-2-2: Input the p temperature nodes at time i The prediction errors of the predicted values ​​of p temperature nodes at j moments form a new data set: Step 3-2-3: Combine the thermal network topology edge weight matrix obtained in step 3-1-2 and the complete temperature field data obtained in step 3-2-2 as input and import them into the OLS-TGCN model to output the transient temperature field data of the electric drive transmission within a certain time scale in the future.