Range extender temperature prediction model training method, range extender control method, and controller

Abstract

The present invention relates to a method for training an extended-range engine temperature prediction model, an extended-range engine control method, and a controller. The training method includes: obtaining sample extended-range engine temperature data corresponding to different time windows under multiple working conditions and constructing a sample extended-range engine temperature feature map; inputting the sample extended-range engine temperature feature map into the extended-range engine temperature prediction model to be trained, extracting graph convolution features of different feature dimensions, and obtaining feedback coefficients of each feature dimension; according to each feedback coefficient, obtaining an active selection factor for each feature dimension and obtaining an optimal feedback coefficient; using the active selection factor of each feature dimension to fuse the graph convolution features of each feature dimension and performing feature enhancement based on the optimal feedback coefficient to obtain time-enhanced features; obtaining predicted extended-range engine temperature data based on the time-enhanced features to obtain a trained extended-range engine temperature prediction model. Using this method can improve the accuracy of extended-range engine temperature prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle control, and particularly to a method for training an extender temperature prediction model, an extender control method, and a controller. Background Art

[0002] With the development of vehicle control technology, a technology has emerged that uses an extender to provide electrical energy for a vehicle when the battery power is insufficient, thereby extending the driving range. When the battery power of an electric vehicle is insufficient, the extender will automatically start to supply power to the battery, enabling the vehicle to continue driving. Moreover, the temperature control of the extender is one of the key factors to ensure its normal operation and extend its lifespan. Therefore, accurately predicting and real-time monitoring the temperature change of the extender is crucial for ensuring its stability and performance.

[0003] In the related art, the temperature prediction of the extender is often limited by the model complexity and real-time performance, and it is difficult to meet the requirements of high-precision and real-time temperature prediction of the extender. With the rapid development of artificial intelligence technology, through the learning of a large amount of data and pattern recognition, the complex relationship between the extender temperature and various factors can be mined, realizing efficient and accurate temperature prediction, and providing strong support for the performance optimization and safe operation of electric vehicles.

[0004] However, currently, for the extender temperature prediction achieved by artificial intelligence technology, the training model usually requires a large number of high-quality samples. In addition, it also faces challenges such as the balance between model complexity and real-time performance, as well as model generalization and stability. Therefore, the currently trained extender temperature prediction model has a low accuracy in predicting the extender temperature. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide a method for training an extender temperature prediction model, an extender control method, a device, a vehicle controller, a storage medium, and a computer program product that can improve the accuracy of extender temperature prediction.

[0006] In a first aspect, the present invention provides a method for training an extender temperature prediction model, including:

[0007] Obtaining sample extender temperature data corresponding to different time windows under multiple working conditions, and constructing a sample extender temperature feature map based on the sample extender temperature data of each time window;

[0008] Inputting the sample extender temperature feature map into the extender temperature prediction model to be trained, extracting graph convolution features corresponding to different feature dimensions of the sample extender temperature feature map through the extender temperature prediction model, and obtaining feedback coefficients of each feature dimension based on each graph convolution feature;

[0009] According to each of the feedback coefficients, obtain the active selection factors for each of the feature dimensions, and obtain the optimal feedback coefficient from each of the feedback coefficients;

[0010] Utilize the active selection factors of each of the feature dimensions to fuse the graph convolution features of each of the feature dimensions to obtain a fused feature, and perform feature enhancement on the fused feature based on the optimal feedback coefficient to obtain a time-enhanced feature;

[0011] Obtain predicted range extender temperature data based on the time-enhanced feature, and train the range extender temperature prediction model using the predicted range extender temperature data to obtain a trained range extender temperature prediction model.

[0012] In one embodiment, the feedback coefficient of each of the feature dimensions is composed of multiple sub-feedback coefficients, and each of the sub-feedback coefficients corresponds to a different time window; the obtaining of the active selection factor for each of the feature dimensions according to each of the feedback coefficients includes: obtaining the current feature dimension and the current sub-feedback coefficient of the current feature dimension corresponding to each time window; the current feature dimension is any one of each of the feature dimensions; according to a preset feedback coefficient threshold and each of the current sub-feedback coefficients, obtain the target time window corresponding to the current feature dimension and the number of the target time windows from each of the time windows; the current sub-feedback coefficient corresponding to the target time window is less than or equal to the feedback coefficient threshold; use the ratio of the number of the target time windows to the total number of windows of each of the time windows as the active selection factor of the current feature dimension.

[0013] In one embodiment, the obtaining of the feedback coefficient of each of the feature dimensions based on each of the graph convolution features includes: inputting the graph convolution feature of the current feature dimension into the fully connected layer of the range extender temperature prediction model, and obtaining the initial predicted range extender temperature data of the current feature dimension corresponding to each time window through the fully connected layer; obtain the difference between the initial predicted range extender temperature data of each of the time windows and the sample range extender temperature data of each of the time windows, and perform normalization processing on the differences of each of the time windows to obtain each of the current sub-feedback coefficients.

[0014] In one embodiment, the utilizing of the active selection factors of each of the feature dimensions to fuse the graph convolution features of each of the feature dimensions to obtain a fused feature includes: obtaining the fusion weight of each of the feature dimensions according to the active selection factor of each of the feature dimensions; wherein, the magnitude of the fusion weight is positively correlated with the magnitude of the active selection factor; utilize the fusion weight of each of the feature dimensions to perform weighted processing on the graph convolution features of each of the feature dimensions to obtain the fused feature.

[0015] In one embodiment, obtaining the optimal feedback coefficient from each of the feedback coefficients includes: obtaining a target feature dimension, and using the feedback coefficient corresponding to the target feature dimension as the optimal feedback coefficient; the active selection factor corresponding to the target feature dimension is the maximum value of each of the active selection factors; based on the optimal feedback coefficient, performing feature enhancement on the fused feature to obtain a time-enhanced feature, including: performing normalization processing on the optimal feedback coefficient, and using the normalized optimal feedback coefficient to perform feature enhancement on the fused feature to obtain the time-enhanced feature.

[0016] In one embodiment, the fused feature is composed of multiple sub-fused features, and each of the sub-fused features corresponds to a different time window; the time-enhanced feature is composed of multiple sub-time-enhanced features, and each of the sub-time-enhanced features corresponds to a different time window; based on the time-enhanced feature, obtaining the predicted range extender temperature data includes: obtaining the sub-time-enhanced feature corresponding to the current time window, and the sub-time-enhanced feature and sub-fused feature of the previous time window of the current time window; the current time window is any one of each of the time windows; using the sub-time-enhanced feature corresponding to the current time window, and the sub-time-enhanced feature and sub-fused feature of the previous time window, to obtain the time-series enhanced feature corresponding to the current time window; inputting the time-series enhanced feature into the fully connected layer of the range extender temperature prediction model, and obtaining the predicted range extender temperature data of the next time window of the current time window through the fully connected layer.

[0017] In one embodiment, based on the sample range extender temperature data of each of the time windows, constructing a sample range extender temperature feature map includes: using each of the time windows as the graph nodes of the sample range extender temperature feature map, and using the sample range extender temperature data of each of the time windows as the node features of each of the graph nodes; using each of the node features, obtaining the similarity degree between each of the graph nodes, and according to the similarity degree, obtaining an adjacency matrix for characterizing the connection relationship between each of the graph nodes; constructing the sample range extender temperature feature map according to the adjacency matrix and each of the node features.

[0018] In a second aspect, the present invention further provides a range extender control method, including:

[0019] Obtaining the historical range extender temperature data of the target range extender corresponding to different time windows, and based on the historical range extender temperature data of each of the time windows, constructing a historical range extender temperature feature map;

[0020] Input the historical range extender temperature feature map into the trained range extender temperature prediction model, and output the predicted range extender temperature data of the target range extender through the range extender temperature prediction model; wherein, the range extender temperature prediction model is trained by the range extender temperature prediction model training method described in any one of the embodiments in the first aspect;

[0021] Use the predicted range extender temperature data to perform temperature control on the target range extender.

[0022] In a third aspect, the present invention also provides a range extender temperature prediction model training device, including:

[0023] A sample feature map construction module, configured to obtain sample range extender temperature data corresponding to different time windows under multiple working conditions, and construct a sample range extender temperature feature map based on the sample range extender temperature data of each time window;

[0024] A feedback coefficient acquisition module, configured to input the sample range extender temperature feature map into the range extender temperature prediction model to be trained, extract the graph convolution features corresponding to different feature dimensions of the sample range extender temperature feature map through the range extender temperature prediction model, and obtain the feedback coefficients of each feature dimension based on each graph convolution feature;

[0025] A selection factor acquisition module, configured to obtain the active selection factors of each feature dimension according to each feedback coefficient, and obtain the optimal feedback coefficient from each feedback coefficient;

[0026] An enhanced feature acquisition module, configured to use the active selection factors of each feature dimension to fuse the graph convolution features of each feature dimension to obtain a fused feature, and perform feature enhancement on the fused feature based on the optimal feedback coefficient to obtain a time-enhanced feature;

[0027] A prediction model training module, configured to obtain predicted range extender temperature data based on the time-enhanced feature, and use the predicted range extender temperature data to train the range extender temperature prediction model to obtain a trained range extender temperature prediction model.

[0028] In a fourth aspect, the present invention also provides a range extender control device, including:

[0029] A historical feature map construction module, configured to obtain the historical range extender temperature data of the target range extender corresponding to different time windows, and construct a historical range extender temperature feature map based on the historical range extender temperature data of each time window;

[0030] A predicted temperature output module, configured to input the historical range extender temperature feature map into a trained range extender temperature prediction model, and output predicted range extender temperature data of the target range extender through the range extender temperature prediction model; wherein, the range extender temperature prediction model is trained by the range extender temperature prediction model training method described in any one of the embodiments in the first aspect;

[0031] A range extender temperature control module, configured to perform temperature control on the target range extender by using the predicted range extender temperature data.

[0032] In a fifth aspect, the present invention further provides a vehicle controller, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method described in any one of the first aspect or the second aspect are implemented.

[0033] In a sixth aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in any one of the first aspect or the second aspect are implemented.

[0034] In a seventh aspect, the present invention further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of the first aspect or the second aspect are implemented.

[0035] The above-mentioned extended-range engine temperature prediction model training method, extended-range engine temperature prediction method, device, vehicle controller, storage medium, and computer program product obtain sample extended-range engine temperature data corresponding to different time windows under multiple working conditions, and construct a sample extended-range engine temperature feature map based on the sample extended-range engine temperature data of each time window; input the sample extended-range engine temperature feature map into the extended-range engine temperature prediction model to be trained, extract the graph convolution features corresponding to different feature dimensions of the sample extended-range engine temperature feature map through the extended-range engine temperature prediction model, and obtain the feedback coefficients of each feature dimension based on each graph convolution feature; according to each feedback coefficient, obtain the active selection factors of each feature dimension, and obtain the optimal feedback coefficient from each feedback coefficient; use the active selection factors of each feature dimension to fuse the graph convolution features of each feature dimension to obtain a fused feature, and perform feature enhancement on the fused feature based on the optimal feedback coefficient to obtain a time-enhanced feature; obtain the predicted extended-range engine temperature data based on the time-enhanced feature, and use the predicted extended-range engine temperature data to train the extended-range engine temperature prediction model to obtain a trained extended-range engine temperature prediction model. By collecting the sample extended-range engine temperature data of different time windows under multiple working conditions during the training of the extended-range engine temperature prediction model, the present invention constructs a feature map and inputs the feature map into the extended-range engine temperature prediction model to be trained, thereby obtaining the graph convolution features of different feature dimensions, and obtaining the feedback coefficients of each feature dimension based on the graph convolution features. Furthermore, the active selection factors of each feature dimension are obtained using the feedback coefficients, so as to achieve feature fusion using the active selection factors, and perform feature enhancement using the optimal feedback coefficient. The extended-range engine temperature prediction is realized by obtaining the time-enhanced feature, thereby completing the model training. Through this method, multi-dimensional feature output and fusion are achieved, the feature representation of the data can be learned more comprehensively, the active selection factors can be obtained based on the feedback coefficients of each feature dimension, feature fusion can be performed based on the active selection factors, and feature enhancement can be performed using the optimal feedback coefficient. The trained extended-range engine temperature prediction model can accurately predict the extended-range engine temperature, so the accuracy of the extended-range engine temperature prediction can be improved, and further the stability of the extended-range engine control can be improved. Description of the Drawings

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a schematic flowchart of the extended-range engine temperature prediction model training method in an embodiment;

[0038] Figure 2Schematic diagram of the process for obtaining the active selection factor in an embodiment;

[0039] Figure 3 Schematic diagram of the process for obtaining the feedback coefficient in an embodiment;

[0040] Figure 4 Schematic diagram of the process for obtaining the predicted range extender temperature data in an embodiment;

[0041] Figure 5 Schematic diagram of the process for constructing the sample range extender temperature feature map in an embodiment;

[0042] Figure 6 Schematic diagram of the process for the range extender control method in an embodiment;

[0043] Figure 7 Schematic diagram of the structure of the active sample selective enhancement temperature prediction model in an embodiment;

[0044] Figure 8 Block diagram of the structure of the range extender temperature prediction model training device in an embodiment;

[0045] Figure 9 Block diagram of the structure of the range extender control device in an embodiment;

[0046] Figure 10 Internal structure diagram of the vehicle controller in an embodiment. Detailed implementation manners

[0047] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0048] In one embodiment, as Figure 1 shown, a method for training a range extender temperature prediction model is provided. In this embodiment, the method is illustrated by taking its application to a vehicle controller as an example. In this embodiment, the method includes the following steps:

[0049] Step S101, obtain sample range extender temperature data corresponding to different time windows under multiple working conditions, and construct a sample range extender temperature feature map based on the sample range extender temperature data of each time window.

[0050] The sample range extender temperature data refers to the range extender temperature data used as training samples for training the range extender temperature prediction model, and the sample range extender temperature data can be composed of the historical temperature data of the range extender. The sample range extender temperature feature map refers to the graph structure used to characterize the sample range extender temperature features.

[0051] Specifically, when the vehicle controller trains the range extender temperature prediction model, it can first collect the historical temperature data of the range extender under multiple working conditions, and then divide the historical temperature data of the range extender according to a specific time window, for example, it can be divided according to the time window T, so as to obtain the sample range extender temperature data corresponding to different time windows, which can be expressed as , where n represents the nth time window. After that, the vehicle controller can further use the sample range extender temperature data corresponding to each time window to construct a graph structure for characterizing the sample range extender temperature characteristics, that is, construct a sample range extender temperature characteristic graph.

[0052] Step S102: Input the sample range extender temperature characteristic graph into the range extender temperature prediction model to be trained, extract the graph convolution features corresponding to different feature dimensions of the sample range extender temperature characteristic graph through the range extender temperature prediction model, and obtain the feedback coefficients of each feature dimension based on each graph convolution feature.

[0053] The range extender temperature prediction model to be trained refers to the range extender temperature prediction model that needs to be trained. This model is mainly used to realize the temperature prediction of the vehicle range extender, and the graph convolution feature refers to the feature output by the graph convolution operation on the sample range extender temperature characteristic graph. This feature can characterize the hidden spatial feature information between time windows, and in this embodiment, the graph convolution feature can correspond to different feature dimensions, for example, it can correspond to 3 feature output dimensions, which are 200, 400, and 600 respectively. The feedback coefficient can be used to characterize the prediction accuracy coefficient of each feature dimension, and this feedback coefficient can be obtained based on the graph convolution features of each feature dimension, that is, the graph convolution feature of feature dimension 1 can obtain the feedback coefficient of feature dimension 1, and similarly, the graph convolution feature of feature dimension 2 can obtain the feedback coefficient of feature dimension 2, and so on.

[0054] Specifically, after the vehicle controller obtains the sample range extender temperature characteristic graph, it can input the sample range extender temperature characteristic graph into the range extender temperature prediction model to be trained. This model can extract the graph convolution features corresponding to different feature dimensions of the sample range extender temperature characteristic graph, and then further use the graph convolution features of each feature dimension to obtain the feedback coefficients of each feature dimension.

[0055] Step S103: Obtain the active selection factors of each feature dimension according to each feedback coefficient, and obtain the optimal feedback coefficient from each feedback coefficient;

[0056] Step S104: Use the active selection factors of each feature dimension to fuse the graph convolution features of each feature dimension to obtain a fused feature, and perform feature enhancement on the fused feature based on the optimal feedback coefficient to obtain a time-enhanced feature.

[0057] The active selection factor can refer to the fusion factor used for fusing the graph convolution features of different feature dimensions, while the optimal feedback coefficient is the enhancement coefficient mainly used for enhancing the features after fusion, that is, the fused features, and the time-enhanced features refer to the features obtained after enhancing the fused features using the optimal feedback coefficient.

[0058] Specifically, after the vehicle controller obtains the feedback coefficients of each feature dimension, it can first obtain the active selection factors of each feature dimension, determine the optimal feedback coefficient from each feedback coefficient, and then use the active selection factors of each feature dimension to fuse the graph convolution features of each feature dimension. After obtaining the fused features, the optimal feedback coefficient is used to enhance the fused features to obtain the time-enhanced features.

[0059] Step S105: Obtain the predicted range extender temperature data based on the time-enhanced features, and use the predicted range extender temperature data to train the range extender temperature prediction model to obtain the trained range extender temperature prediction model.

[0060] The predicted range extender temperature data refers to the predicted data output by the range extender temperature prediction model based on the time-enhanced features. Specifically, after the range extender temperature prediction model obtains the time-enhanced features, it can further output the predicted range extender temperature data through the time-enhanced features, and then use the predicted range extender temperature data to train the range extender temperature prediction model to complete the training of the range extender temperature prediction model, that is, obtain the trained range extender temperature prediction model.

[0061] In the above method for training the range extender temperature prediction model, by obtaining sample range extender temperature data corresponding to different time windows under multiple working conditions, and based on the sample range extender temperature data of each time window, a sample range extender temperature feature map is constructed; the sample range extender temperature feature map is input into the range extender temperature prediction model to be trained, and the graph convolution features corresponding to different feature dimensions of the sample range extender temperature feature map are extracted through the range extender temperature prediction model, and the feedback coefficients of each feature dimension are obtained based on each graph convolution feature; according to each feedback coefficient, the active selection factors of each feature dimension are obtained, and the optimal feedback coefficient is obtained from each feedback coefficient; using the active selection factors of each feature dimension, the graph convolution features of each feature dimension are fused to obtain a fused feature, and the fused feature is feature-enhanced based on the optimal feedback coefficient to obtain a time-enhanced feature; based on the time-enhanced feature, predicted range extender temperature data is obtained, and the range extender temperature prediction model is trained using the predicted range extender temperature data to obtain a trained range extender temperature prediction model. When training the range extender temperature prediction model according to the present invention, sample range extender temperature data of different time windows under multiple working conditions is collected, so as to construct a feature map and input the feature map into the range extender temperature prediction model to be trained, so as to obtain graph convolution features of different feature dimensions, and based on the graph convolution features, the feedback coefficients of each feature dimension are obtained, and then the active selection factors of each feature dimension are obtained using the feedback coefficients, so as to use the active selection factors to achieve feature fusion, and use the optimal feedback coefficient for feature enhancement, and then the range extender temperature prediction is realized after obtaining the time-enhanced feature, so as to complete the model training. Through this method, multi-dimensional feature output and fusion are realized, the feature representation of data can be learned more comprehensively, and the active selection factors can be obtained based on the feedback coefficients of each feature dimension, and feature fusion can be performed based on the active selection factors. The optimal feedback coefficient can also be used for feature enhancement. The trained range extender temperature prediction model can accurately predict the range extender temperature, so the accuracy of range extender temperature prediction can be improved, and further the stability of range extender control can be improved.

[0062] In one embodiment, the feedback coefficient of each feature dimension is composed of a plurality of sub-feedback coefficients, and each sub-feedback coefficient corresponds to a different time window; as Figure 2 shown, step S103 may further include:

[0063] Step S201, obtaining the current feature dimension and the current sub-feedback coefficient of the current feature dimension corresponding to each time window; the current feature dimension is any one of each feature dimension.

[0064] The current feature dimension can refer to any one of multiple feature dimensions, and the current sub-feedback coefficient refers to the sub-feedback coefficient corresponding to the current feature dimension for each time window. In this embodiment, the feedback coefficient can be composed of multiple sub-feedback coefficients, and each sub-feedback coefficient corresponds to a different time window. Specifically, the vehicle controller can first take any one of the multiple feature dimensions as the current feature dimension, and take the sub-feedback coefficients corresponding to each time window in the feedback coefficient corresponding to the current feature dimension as the current sub-feedback coefficients for each time window.

[0065] Step S202: Obtain the target time window corresponding to the current feature dimension and the number of target time windows from each time window according to a preset feedback coefficient threshold and each current sub-feedback coefficient; the current sub-feedback coefficient corresponding to the target time window is less than or equal to the feedback coefficient threshold.

[0066] The feedback coefficient threshold is a preset coefficient threshold. For example, the coefficient threshold can be 0.3, and the target time window corresponding to the current feature dimension can refer to the time window in which the current sub-feedback coefficient is less than or equal to the feedback coefficient threshold. In this embodiment, after obtaining the current sub-feedback coefficients of each time window, the current sub-feedback coefficients can be respectively compared with the feedback coefficient threshold, so as to take the time window in which the current sub-feedback coefficient is less than or equal to the feedback coefficient threshold as the target time window corresponding to the current feature dimension.

[0067] For example, the current sub-feedback coefficients corresponding to the current feature dimension respectively include sub-feedback coefficient 1, sub-feedback coefficient 2, and sub-feedback coefficient 3, corresponding to time window 1, time window 2, and time window 3 respectively. If sub-feedback coefficient 1 and sub-feedback coefficient 2 satisfy being less than or equal to the feedback coefficient threshold, then time window 1 and time window 2 are used as the target time windows.

[0068] Step S203: Take the ratio of the number of target time windows to the total number of windows of each time window as the active selection factor of the current feature dimension.

[0069] After determining the target time window, the number of target time windows can be counted, and the ratio of the number of target time windows to the total number of windows of each time window is taken as the active selection factor of the current feature dimension. For example, if time window 1 and time window 2 are the target time windows, then the number of target time windows is 2, and the total number of time windows is 3, so the active selection factor of the current feature dimension is 2 / 3.

[0070] Taking 0.3 as the feedback coefficient threshold as an example, the active selection factor of the current feature dimension i can be calculated by the following formula:

[0071]

[0072] Among them, represents the current sub-feedback coefficient of the current feature dimension i corresponding to the time window j. C is a counting operation, which can represent the number of target time windows, and n represents the total number of time windows.

[0073] In this embodiment, a pre-set feedback coefficient threshold and the current sub-feedback coefficients of the current feature dimension corresponding to each time window can also be used to screen out the target time windows, and further obtain the active selection factor of the current feature dimension. In this way, the active selection factor can be used to represent whether the number of features within the threshold range of each feature dimension is large enough and whether the quality is high enough. Therefore, the accuracy of obtaining the active selection factor can be improved.

[0074] Furthermore, as Figure 3 shown, step S102 can further include:

[0075] Step S301, input the graph convolutional features of the current feature dimension into the fully connected layer of the range extender temperature prediction model, and obtain the initial predicted range extender temperature data of the current feature dimension corresponding to each time window through the fully connected layer.

[0076] The fully connected layer is the output layer of the range extender temperature prediction model for outputting the predicted range extender temperature data. The initial predicted range extender temperature data is the range extender temperature data predicted by the graph convolutional features of the current feature dimension. And since the number of time windows is multiple, the initial predicted range extender temperature data predicted by the current feature dimension can also be multiple, corresponding to different time windows respectively.

[0077] Specifically, the vehicle controller can input the graph convolutional features of the current feature dimension into the range extender temperature prediction model, and the range extender temperature prediction model outputs the initial predicted range extender temperature data of each time window under the current feature dimension.

[0078] Step S302, obtain the differences between the initial predicted range extender temperature data of each time window and the sample range extender temperature data of each time window, and perform normalization processing on the differences of each time window to obtain each current sub-feedback coefficient.

[0079] After that, the errors between the initial predicted range extender temperature data and the sample range extender temperature data of each time window can be calculated respectively to obtain the errors of each time window under the current feature dimension, and then the above errors are normalized to obtain each current sub-feedback coefficient.

[0080] For example, the feedback coefficient of the current feature dimension i can be expressed by the following formula:

[0081]

[0082] Among them, represents the output data of the current dimensional feature after passing through the fully connected layer. represents the n initial predicted range extender temperature data respectively predicted by n time windows under the current feature dimension, and y is the actual temperature label. is the sample range extender temperature data of n time windows. Then, normalize each difference to obtain n feedback coefficients between (0, 1), that is, obtain the current sub-feedback coefficients under n different time windows.

[0083] In this embodiment, the error between the initial predicted range extender temperature data and the sample range extender temperature data under each time window can be calculated respectively, and the error can be normalized, so as to obtain each current sub-feedback coefficient. By this method, the accuracy of obtaining the current sub-feedback coefficient can be improved.

[0084] In one embodiment, step S104 may further include: obtaining the fusion weights of each feature dimension according to the active selection factors of each feature dimension; wherein, the magnitude of the fusion weight is positively correlated with the magnitude of the active selection factor; using the fusion weights of each feature dimension, perform weighted processing on the graph convolution features of each feature dimension to obtain the fusion features.

[0085] Among them, the fusion weight refers to the feature weight used to fuse the graph convolution features output by each feature dimension, and the magnitude of this feature weight is positively correlated with the magnitude of the active selection factor, that is, the larger the active selection factor of a certain feature dimension, the larger the corresponding fusion weight of this feature dimension will be. In this embodiment, the active selection factor can represent the quantity and quality of features within a certain feedback coefficient threshold range of a feature dimension. If the active selection factor is larger, it means that the quantity of features within a certain feedback coefficient threshold range of this feature dimension is larger and the quality is higher. Therefore, it is necessary to increase the weight of the features of the corresponding dimension, and vice versa, the weight of the corresponding dimension should be reduced.

[0086] There are also various ways to set the fusion weight. For example, the active selection factor itself can be used as the fusion weight, or a certain reference value can be added to the active selection factor as the fusion weight. It can also be used as the fusion weight after increasing or decreasing the active selection factor by a set multiple, as long as it satisfies that the magnitude of the fusion weight is positively correlated with the magnitude of the active selection factor.

[0087] Specifically, after the vehicle controller obtains the active selection factors of each feature dimension, it can use the active selection factors to calculate the fusion weights of the corresponding feature dimensions, and then use each feature weight to perform weighted processing on the graph convolution features of each feature dimension to obtain the fusion features.

[0088] Taking the active selection factor directly as the fusion weight and assuming the number of feature dimensions is 3, the fusion feature can be obtained through the following formula:

[0089]

[0090] where respectively represent the active selection factors of feature dimension 1, feature dimension 2, and feature dimension 3, and respectively represent the graph convolution features of feature dimension 1, feature dimension 2, and feature dimension 3.

[0091] In this embodiment, the fusion weight can also be obtained through the active selection factor, and then the graph convolution features are weighted and fused using the fusion weight. In this way, the weight of feature fusion can be dynamically adjusted, improving the flexibility and adaptability of the model to features of different dimensions.

[0092] Further, step S103 can further include: obtaining the target feature dimension and using the feedback coefficient corresponding to the target feature dimension as the optimal feedback coefficient; the active selection factor corresponding to the target feature dimension is the maximum value of each active selection factor; step S104 can further include: performing normalization processing on the optimal feedback coefficient and using the normalized optimal feedback coefficient to perform feature enhancement on the fusion feature to obtain the time-enhanced feature.

[0093] The target feature dimension refers to the feature dimension corresponding to the maximum value of the active selection factor, and the optimal feedback coefficient is the feedback coefficient of this target feature dimension. In this embodiment, the optimal feedback coefficient can also be determined through the active selection factors of each feature dimension. For example, the maximum active selection factor can be first found from the active selection factors of each feature dimension, and the feedback coefficient of the feature dimension corresponding to this active selection factor is used as the optimal feedback coefficient. For example, respectively represent the active selection factors of feature dimension 1, feature dimension 2, and feature dimension 3. Suppose , then feature dimension 2 is used as the target feature dimension, and thus the feedback coefficient of feature dimension 2 is used as the optimal feedback coefficient.

[0094] After obtaining the optimal feedback coefficient, normalization processing can be first performed on the optimal feedback coefficient, and the normalized optimal feedback coefficient is used to achieve feature enhancement of the fusion feature, and the time-enhanced feature is obtained in this way.

[0095] For example, the time enhancement feature is calculated as follows:

[0096]

[0097] Among them, represents the optimal feedback coefficient. Then represents the normalized optimal feedback coefficient, while represents the fused feature.

[0098] In this embodiment, the feedback coefficient of the feature dimension corresponding to the largest active selection factor can be used as the optimal feedback coefficient, and the enhancement of the fused feature can be achieved by normalizing the optimal feedback coefficient. This method can further improve the acquisition accuracy of the time enhancement feature.

[0099] In addition, the fused feature is composed of multiple sub-fused features, and each sub-fused feature corresponds to a different time window; the time enhancement feature is composed of multiple sub-time enhancement features, and each sub-time enhancement feature corresponds to a different time window; as Figure 4 shown, step S105 may further include:

[0100] Step S401, obtaining the sub-time enhancement feature corresponding to the current time window, and the sub-time enhancement feature and sub-fused feature of the previous time window of the current time window; the current time window is any one of the time windows.

[0101] In this embodiment, the fused feature can also be composed of multiple sub-fused features, each sub-fused feature corresponding to a different time window. Similarly, the time enhancement feature can also be composed of multiple sub-time enhancement features, and each sub-time enhancement feature also corresponds to a different time window, and the current time window can refer to any one of the time windows.

[0102] Specifically, when predicting the temperature of the range extender, the current time window can also be determined from multiple time windows first, so as to obtain the sub-time enhancement feature of the current time window, the sub-time enhancement feature of the previous time window of the current time window, and the sub-fused feature without feature enhancement.

[0103] Step S402, using the sub-time enhancement feature corresponding to the current time window, and the sub-time enhancement feature and sub-fused feature of the previous time window, to obtain the time series enhancement feature corresponding to the current time window.

[0104] After obtaining the sub-time enhanced features corresponding to the current time window, as well as the sub-time enhanced features and sub-fusion features of the previous time window, the above features can be used to obtain the temporal enhanced features corresponding to the current time window, and the temporal enhanced features can be used to predict the extender temperature data of the next time window of the current time window.

[0105] For example, the current time window The temporal enhanced features of Can be characterized by the following formula:

[0106]

[0107] Among them, Represents the time enhanced features under the current time window , Is the weight matrix, Represents the previous time window of the current time window, that is The time enhanced features under Represents the previous time window of the current time window, that is The un-time-enhanced fusion features under Is the bias matrix.

[0108] Step S403, input the temporal enhanced features into the fully connected layer of the extender temperature prediction model, and obtain the predicted extender temperature data of the next time window of the current time window through the fully connected layer.

[0109] After obtaining the temporal enhanced features corresponding to the current time window, the temporal enhanced features can be input into the fully connected layer of the extender temperature prediction model used to output the predicted extender temperature data, and the predicted extender temperature data of the next time window of the current time window can be obtained through the fully connected layer.

[0110] In this embodiment, the temporal enhanced features corresponding to the current time window can also be obtained by combining the sub-time enhanced features and sub-fusion features of the previous time window, so as to output the predicted extender temperature data of the next time window of the current time window. Through this method, a recursive feature selection mechanism can be implemented, enabling the model to dynamically adjust the importance of features according to feedback information, thereby better adapting to the changes and dynamics of the data and improving the adaptive ability of the model.

[0111] In one embodiment, as Figure 5 Shown, step S101 can further include:

[0112] Step S501, take each time window as a graph node of the sample extender temperature feature map, and take the sample extender temperature data of each time window as the node feature of each graph node.

[0113] In this embodiment, the graph nodes of the sample range extender temperature feature map can represent different time windows of the sample range extender temperature data, and the sample range extender temperature data of each time window can be used as the node features of each graph node. For example, if the sample range extender temperature data of each time window contains 1000 data samples, then the above data samples can be used as the node features of each graph node.

[0114] Step S502: Use each node feature to obtain the similarity degree between each graph node, and based on the similarity degree, obtain an adjacency matrix for characterizing the connection relationship between each graph node.

[0115] Step S503: Construct a sample range extender temperature feature map according to the adjacency matrix and each node feature.

[0116] After that, the similarity degree between each graph node can be obtained by using the node features of each graph node. For example, it can be calculated through cosine similarity, so as to obtain an adjacency matrix, which can represent the connection relationship between each graph node. Taking n time windows as an example, the number of graph nodes is n, and the obtained adjacency matrix at this time can be an n×n adjacency matrix. Then, a sample range extender temperature feature map can be constructed according to the constructed adjacency matrix and each node feature.

[0117] For example, the graph structure can be represented by G=(A,X), where X is the node feature, that is, the sample range extender temperature data of each time window, and A represents the adjacency matrix of graph G, representing the connection relationship between nodes, which can be obtained by calculating the similarity degree between each graph node.

[0118] In this embodiment, the sample range extender temperature data of each time window can also be used as the node feature of the graph node, and the similarity degree between the graph nodes can be obtained through the node feature, so as to construct an adjacency matrix to form a sample range extender temperature feature map. By this method, the efficiency of constructing the sample range extender temperature feature map can be improved.

[0119] In one embodiment, as Figure 6 shown, a range extender control method is also provided. In this embodiment, this method is exemplified by being applied to a vehicle controller. In this embodiment, the method includes the following steps:

[0120] Step S601: Obtain the historical range extender temperature data of the target range extender corresponding to different time windows, and construct a historical range extender temperature feature map based on the historical range extender temperature data of each time window.

[0121] Among them, the target range extender refers to the range extender for which the range extender temperature prediction needs to be performed, the historical range extender temperature data refers to the temperature data of the target range extender under different time windows, and the historical range extender temperature feature map is the feature map constructed based on the above temperature data.

[0122] Specifically, when the temperature prediction of the target range extender is required, first, the historical range extender temperature data corresponding to different time windows of the range extender can be collected, and then the corresponding historical range extender temperature feature map can be constructed based on the above historical range extender temperature data.

[0123] Step S602: Input the historical range extender temperature feature map into the trained range extender temperature prediction model, and output the predicted range extender temperature data of the target range extender through the range extender temperature prediction model; among them, the range extender temperature prediction model is trained by the range extender temperature prediction model training method of any one of the above embodiments.

[0124] After that, the vehicle controller can input the historical range extender temperature feature map into the trained range extender temperature prediction model, and the predicted range extender temperature data of the target range extender is output by the range extender temperature prediction model.

[0125] Among them, the predicted range extender temperature data can be obtained through the following process. First, the graph convolution features corresponding to different feature dimensions of the historical range extender temperature feature map are extracted by the range extender temperature prediction model, and the feedback coefficients of each graph convolution feature are obtained. It can be that each graph convolution feature is input into the fully connected layer of the range extender temperature prediction model. After obtaining the initial prediction data of each time window, the feedback coefficients of each graph convolution feature are obtained based on the error between the initial prediction data and the historical range extender temperature data. Then, according to the above feedback coefficients, the active selection factors and optimal feedback coefficients of each feature dimension are obtained. It can be that each feedback coefficient is compared with a preset feedback coefficient threshold first, and then the number of time windows in which each feedback coefficient is less than or equal to the feedback coefficient threshold is counted to calculate the active selection factors of each feature dimension, and further the feedback coefficient of the feature dimension with the largest active selection factor is used as the optimal feedback coefficient. Next, the active selection factors can be used to fuse each graph convolution feature, and the fusion feature is feature-enhanced based on the optimal feedback coefficient. It can be that the fusion weights are obtained by using the active selection factors for weighted fusion, and the fusion weights are feature-enhanced by normalizing the optimal feedback coefficient and then using the normalized optimal feedback coefficient. Finally, the predicted range extender temperature data is obtained by using the time-enhanced feature. For example, it can be that the time-enhanced feature of the previous time interval of the current time interval, the fusion feature, and the time-enhanced feature of the current time interval are used to output the predicted range extender temperature data of the next time window.

[0126] Step S603, perform temperature control on the target range extender using the predicted range extender temperature data.

[0127] In the above range extender control method, by obtaining the historical range extender temperature data of the target range extender corresponding to different time windows, and based on the historical range extender temperature data of each time window, constructing a historical range extender temperature feature map; inputting the historical range extender temperature feature map into the trained range extender temperature prediction model, and outputting the predicted range extender temperature data of the target range extender through the range extender temperature prediction model; wherein, the range extender temperature prediction model is trained by the range extender temperature prediction model training method described in any of the above embodiments; perform temperature control on the target range extender using the predicted range extender temperature data. In the present invention, by constructing the historical range extender temperature feature map of the target range extender and combining the trained range extender temperature prediction model to output the predicted range extender temperature data, since the multi-dimensional feature output and fusion are realized during the training of the range extender temperature prediction model, it can learn the feature representation of the data more comprehensively, and can obtain the active selection factor based on the feedback coefficient of each feature dimension, and perform feature fusion based on the active selection factor, and can also use the optimal feedback coefficient for feature enhancement. The trained range extender temperature prediction model can accurately predict the range extender temperature, so the accuracy of the range extender temperature prediction can be improved, and further the stability of the range extender control can be improved.

[0128] In one embodiment, a range extender temperature prediction method based on active sample selective enhancement is also provided, as Figure 7 shown. This method designs a multi-dimensional feature enhancement mechanism, adaptively selects the optimal dimension of feature output for different working condition data, and secondly designs a recursive feature selection mechanism to effectively enhance the time series feature extraction based on the feedback coefficient, improve the generalization ability and adaptability of the prediction model under different working conditions, and improve the prediction accuracy. Specifically, it includes the following steps:

[0129] First, collect the historical temperature data of the range extender under multiple working conditions, and then design a multi-dimensional feature enhancement mechanism. Divide the historical temperature data according to the time window T, T = [t1, t2,..., t n], where n represents the nth time window, and each time window contains 1000 data samples. First, we need to construct a graph for the historical temperature of the range extender. The graph structure is the input of the graph convolutional network. The graph structure consists of nodes, features corresponding to the nodes, and connecting lines between nodes. The graph structure is represented by G=(A,X), where X is the node feature and A represents the adjacency matrix of the graph G, which represents the connection relationship between the nodes. Therefore, each time window is regarded as a node, and the 1000 data samples in the time window are used as the node features corresponding to the node, so that n nodes corresponding to n time windows and 1000 data samples in each time window can be obtained as the node features of the corresponding nodes. Then, the similarity between the nodes is calculated by cosine similarity, so that an n×n adjacency matrix can be obtained, that is, A and X in the above graph structure are node features. Then, the graph structure is subjected to graph convolution operation to mine the hidden spatial feature information between time windows, and construct feature outputs of three dimensions to capture the features of the data from different angles, so that the model can learn the feature representation of the data more comprehensively. The calculation process is as follows:

[0130]

[0131]

[0132]

[0133] in The output feature dimension is 200. The output feature dimension is 400. The output feature dimension is 600. , , are the corresponding weight matrices, To perform symmetric normalization on the adjacency matrix A, is the activation function. In this way, we can obtain n nodes with node features of 200, 400, and 600, which can be expressed as n×200, n×400, and n×600. Then, the three dimensional features are input into the fully connected layer for preliminary prediction, and the feedback coefficient of the output feature of each dimension is calculated. The calculation process is as follows:

[0134]

[0135] Among them, i=1, 2, 3, respectively represent , , Three dimensional features, Represents the initial temperature prediction value obtained after the three dimensional features pass through the fully connected layer, Denote the n temperature prediction values obtained by predicting n nodes in the i-th dimension, and y is the actual temperature label. These are n actual temperature values. Then, normalize the differences to obtain 3×n feedback coefficients between (0, 1).

[0136] Construct an active selection factor based on the feedback coefficients. Calculate the ratios of the feedback coefficients of the three dimensions respectively. By setting a threshold, calculate the number of nodes that meet the threshold, so as to obtain the fusion weights corresponding to different dimensional features. The calculation process is as follows:

[0137]

[0138] where i = 1, 2, 3 correspond to the three dimensions. Denote the feedback coefficient of the j-th initial temperature prediction value predicted in the i-th feature dimension. 0.3 is the threshold, and C is the counting operation. In this way, the active selection factor for each feature dimension can be obtained. Through this ratio calculation, a high Z value represents a large number and high quality of features within the threshold range. Therefore, the weight of the corresponding dimensional feature needs to be increased, and vice versa, the weight of the corresponding dimension should be reduced. Therefore, the active selection factor is given to dynamically adjust the weight of feature fusion, improving the flexibility and adaptability of the model to different dimensional features. The multi-dimensional feature fusion process is as follows:

[0139]

[0140] Secondly, a recursive feature selection mechanism is designed. Since the above operations are mainly used to mine the spatial correlation information between time series signals and do not consider the temporal correlation, first enhance each moment feature based on the feedback coefficients. First, extract the feedback coefficients in the optimal dimension, that is The corresponding optimal feedback coefficient is denoted by Then, normalize , that is . Since The smaller it is, the more accurate the prediction. Therefore, multiply by each moment feature to achieve time feature enhancement. The corresponding calculation process of the time-enhanced feature is as follows:

[0141]

[0142] Furthermore, an effective sample selection factor is designed to select sample features of the features at the previous moment. In this process, the features at the unenhanced moment are retained, and features are adaptively selected through the effective sample selection factor to ensure that the model can comprehensively utilize all feature information during the entire training process and improve the global performance of the model. The calculation process is as follows:

[0143]

[0144] Among them, represents the time enhancement feature under the time window, denotes the feature without time enhancement under the time window, represents the time series enhancement feature obtained after mining various feature information at the previous moment, is the bias matrix. Finally, the feature is input into the fully connected layer to output the predicted temperature of the range extender in the next time period, thereby realizing the temperature prediction task.

[0145] Through this embodiment, aiming at the problem that the existing range extender temperature prediction model has high requirements for the quality of data samples and the model is complex, resulting in limitations in the temperature prediction accuracy and stability of the model, a range extender temperature prediction method based on active sample selective enhancement is proposed. The main methods include: first, collecting the historical temperature data of the range extender under multiple working conditions, and then designing a multi-dimensional feature enhancement mechanism. The historical temperature data is divided according to the time window, and each time window contains 1000 data samples. The hidden spatial feature information between time windows is mined through graph convolution feature aggregation, and three-dimensional feature outputs are constructed to capture the features of the data from different angles, enabling the model to learn the feature representation of the data more comprehensively. Then, calculate the feedback coefficient of each dimension output feature, and then construct an active selection factor to obtain the fusion weights corresponding to different dimension features. In this way, the weights of feature fusion are dynamically adjusted, improving the flexibility and adaptability of the model to different dimension features. Then, fuse different dimension features according to the fusion weights to obtain multi-dimensional fusion features; secondly, design a recursive feature selection mechanism to perform sample enhancement on the features at the previous moment based on the feedback coefficient and the effective sample selection factor. The model can dynamically adjust the importance of features according to the feedback information, thereby better adapting to the changes and dynamics of the data and improving the adaptive ability of the model.

[0146] It should be understood that although the steps in the flowcharts involved in the above-mentioned embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or steps in other steps.

[0147] Based on the same inventive concept, embodiments of the present invention further provide a training device for an extender temperature prediction model for implementing the extender temperature prediction model training method involved above, and a prediction device for an extender temperature for implementing the extender temperature prediction method involved above. The implementation solutions provided by the device for solving problems are similar to those described in the above method. Therefore, the specific limitations in one or more embodiments of the training device for an extender temperature prediction model and embodiments of the prediction device for an extender temperature provided below can refer to the limitations on the extender temperature prediction model training method and the extender temperature prediction method in the above text, and will not be repeated here.

[0148] In one embodiment, as Figure 8 shown, a training device for an extender temperature prediction model is provided, including: a sample feature map construction module 801, a feedback coefficient acquisition module 802, a selection factor acquisition module 803, an enhanced feature acquisition module 804, and a prediction model training module 805, where:

[0149] The sample feature map construction module 801 is configured to obtain sample extender temperature data corresponding to different time windows under multiple operating conditions, and construct a sample extender temperature feature map based on the sample extender temperature data of each time window;

[0150] The feedback coefficient acquisition module 802 is configured to input the sample extender temperature feature map into the extender temperature prediction model to be trained, extract graph convolution features corresponding to different feature dimensions of the sample extender temperature feature map through the extender temperature prediction model, and obtain feedback coefficients for each feature dimension based on each graph convolution feature;

[0151] The selection factor acquisition module 803 is configured to obtain an active selection factor for each feature dimension according to each feedback coefficient, and obtain an optimal feedback coefficient from each feedback coefficient;

[0152] The enhanced feature acquisition module 804 is configured to use the active selection factor for each feature dimension to fuse the graph convolution features for each feature dimension to obtain a fused feature, and perform feature enhancement on the fused feature based on the optimal feedback coefficient to obtain a time-enhanced feature;

[0153] The prediction model training module 805 is configured to obtain predicted extender temperature data based on the time-enhanced feature, and train the extender temperature prediction model by using the predicted extender temperature data to obtain a trained extender temperature prediction model.

[0154] In one embodiment, the feedback coefficient of each feature dimension is composed of multiple sub-feedback coefficients, and each sub-feedback coefficient corresponds to a different time window; the selection factor acquisition module 803 is further configured to acquire the current feature dimension and the current sub-feedback coefficient of the current feature dimension corresponding to each time window; the current feature dimension is any one of the feature dimensions; according to the preset feedback coefficient threshold and each current sub-feedback coefficient, the target time window corresponding to the current feature dimension and the number of target time windows are acquired from each time window; the current sub-feedback coefficient corresponding to the target time window is less than or equal to the feedback coefficient threshold; the ratio of the number of target time windows to the total number of windows of each time window is used as the active selection factor of the current feature dimension.

[0155] In one embodiment, the feedback coefficient acquisition module 802 is further configured to input the graph convolution feature of the current feature dimension into the fully connected layer of the extended-range engine temperature prediction model, and obtain the initial predicted extended-range engine temperature data of the current feature dimension corresponding to each time window through the fully connected layer; acquire the difference between the initial predicted extended-range engine temperature data of each time window and the sample extended-range engine temperature data of each time window, and perform normalization processing on the differences of each time window to obtain each current sub-feedback coefficient.

[0156] In one embodiment, the enhanced feature acquisition module 804 is further configured to acquire the fusion weight of each feature dimension according to the active selection factor of each feature dimension; wherein, the magnitude of the fusion weight is positively correlated with the magnitude of the active selection factor; the graph convolution features of each feature dimension are weighted by using the fusion weights of each feature dimension to obtain the fused features.

[0157] In one embodiment, the selection factor acquisition module 803 is further configured to acquire the target feature dimension and use the feedback coefficient corresponding to the target feature dimension as the optimal feedback coefficient; the active selection factor corresponding to the target feature dimension is the maximum value of each active selection factor; the enhanced feature acquisition module 804 is further configured to perform normalization processing on the optimal feedback coefficient and perform feature enhancement on the fused features by using the normalized optimal feedback coefficient to obtain the time-enhanced features.

[0158] In one embodiment, the fusion feature is composed of a plurality of sub-fusion features, and each sub-fusion feature corresponds to a different time window; the time enhancement feature is composed of a plurality of sub-time enhancement features, and each sub-time enhancement feature corresponds to a different time window; the prediction model training module 805 is further configured to obtain the sub-time enhancement feature corresponding to the current time window, and the sub-time enhancement feature and the sub-fusion feature of the previous time window of the current time window; the current time window is any one of the time windows; using the sub-time enhancement feature corresponding to the current time window, and the sub-time enhancement feature and the sub-fusion feature of the previous time window, obtain the time series enhancement feature corresponding to the current time window; input the time series enhancement feature into the fully connected layer of the range extender temperature prediction model, and obtain the predicted range extender temperature data of the next time window of the current time window through the fully connected layer.

[0159] In one embodiment, the sample feature map construction module 801 is further configured to use each time window as a graph node of the sample range extender temperature feature map, and use the sample range extender temperature data of each time window as the node feature of each graph node; use each node feature to obtain the similarity degree between each graph node, and according to the similarity degree, obtain an adjacency matrix for characterizing the connection relationship between each graph node; construct a sample range extender temperature feature map according to the adjacency matrix and each node feature.

[0160] In one embodiment, as Figure 9 shown, a range extender control device is provided, including: a historical feature map construction module 901, a predicted temperature output module 902, and a range extender temperature control module 903, where:

[0161] The historical feature map construction module is configured to obtain the historical range extender temperature data corresponding to the target range extender for different time windows, and construct a historical range extender temperature feature map based on the historical range extender temperature data of each time window;

[0162] The predicted temperature output module 902 is configured to input the historical range extender temperature feature map into the trained range extender temperature prediction model, and output the predicted range extender temperature data of the target range extender through the range extender temperature prediction model; wherein, the range extender temperature prediction model is trained by the range extender temperature prediction model training method of any one of the above embodiments;

[0163] The range extender temperature control module 903 is configured to perform temperature control on the target range extender by using the predicted range extender temperature data.

[0164] Each module in the above-mentioned range extender temperature prediction model training device and range extender temperature prediction device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above modules can be embedded in the processor in the vehicle controller in hardware form or be independent of it, or can be stored in the memory in the vehicle controller in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0165] In one embodiment, a vehicle controller is provided. The vehicle controller can be a terminal, and its internal structure diagram can be as Figure 10 shown. The vehicle controller includes a processor, a memory, an input / output interface, and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the vehicle controller is used to provide computing and control capabilities. The memory of the vehicle controller includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the vehicle controller is used to exchange information between the processor and external devices. The communication interface of the vehicle controller is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for training a range extender temperature prediction model and a method for predicting the temperature of a range extender.

[0166] Those skilled in the art can understand that Figure 10 the structure shown in

[0167] is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the vehicle controller to which the solution of the present invention is applied. The specific vehicle controller may include more or fewer components than those shown in the figure, or combine certain components, or have a different component layout.

[0168] In one embodiment, a vehicle controller is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in each of the above method embodiments are implemented.

[0168] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in each of the above method embodiments are implemented.

[0169] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in each of the above method embodiments are implemented.

[0170] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of the relevant data need to comply with the relevant regulations.

[0171] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided by the present invention can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided by the present invention can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0172] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0173] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several variations and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the appended claims.

Claims

1. A method for training an extended-range engine temperature prediction model, characterized in that The method includes: Obtaining sample range extender temperature data corresponding to different time windows under multiple working conditions, and constructing a sample range extender temperature feature map based on the sample range extender temperature data of each of the time windows; Inputting the sample range extender temperature feature map into a range extender temperature prediction model to be trained, extracting graph convolution features corresponding to different feature dimensions of the sample range extender temperature feature map through the range extender temperature prediction model, and obtaining feedback coefficients for each of the feature dimensions based on each of the graph convolution features; The feedback coefficients for each of the feature dimensions include a plurality of sub-feedback coefficients, and each of the sub-feedback coefficients corresponds to a different time window respectively; The current sub-feedback coefficient corresponding to the current feature dimension for each time window represents the difference between the initial predicted range extender temperature data corresponding to the current feature dimension for each time window and the sample range extender temperature data; The current feature dimension is any one of each of the feature dimensions; Obtaining an active selection factor for each of the feature dimensions according to each of the feedback coefficients, and obtaining an optimal feedback coefficient from each of the feedback coefficients; wherein, the active selection factor for the current feature dimension includes the ratio of the number of target time windows to the total number of windows of each of the time windows; wherein, the target time window is a time window in each of the time windows where the current sub-feedback coefficient is less than or equal to a preset feedback coefficient threshold; Using the active selection factors for each of the feature dimensions to fuse the graph convolution features for each of the feature dimensions to obtain a fused feature, and performing feature enhancement on the fused feature based on the optimal feedback coefficient to obtain a time-enhanced feature; Obtaining predicted range extender temperature data based on the time-enhanced feature, and training the range extender temperature prediction model using the predicted range extender temperature data to obtain a trained range extender temperature prediction model.

2. The method according to claim 1, wherein The obtaining an active selection factor for each of the feature dimensions according to each of the feedback coefficients includes: Obtaining the current feature dimension and the current sub-feedback coefficient corresponding to the current feature dimension for each time window; Obtaining the target time window corresponding to the current feature dimension and the number of the target time windows from each of the time windows according to a preset feedback coefficient threshold and each of the current sub-feedback coefficients; Taking the ratio of the number of the target time windows to the total number of windows of each of the time windows as the active selection factor for the current feature dimension.

3. The method according to claim 2, wherein The obtaining feedback coefficients for each of the feature dimensions based on each of the graph convolution features includes: Inputting the graph convolution feature of the current feature dimension into the fully connected layer of the range extender temperature prediction model, and obtaining the initial predicted range extender temperature data corresponding to the current feature dimension for each time window through the fully connected layer; Obtaining the difference between the initial predicted range extender temperature data for each of the time windows and the sample range extender temperature data for each of the time windows, and performing normalization processing on the differences for each of the time windows to obtain each of the current sub-feedback coefficients.

4. The method according to claim 1, wherein The using the active selection factors for each of the feature dimensions to fuse the graph convolution features for each of the feature dimensions to obtain a fused feature includes: Obtain the fusion weights for each of the feature dimensions according to the active selection factors of each of the feature dimensions; wherein, the magnitude of the fusion weight is positively correlated with the magnitude of the active selection factor; Use the fusion weights of each of the feature dimensions to perform weighted processing on the graph convolution features of each of the feature dimensions to obtain the fusion feature.

5. The method according to claim 4, wherein The obtaining of the optimal feedback coefficient from each of the feedback coefficients includes: Obtain the target feature dimension and use the feedback coefficient corresponding to the target feature dimension as the optimal feedback coefficient; the active selection factor corresponding to the target feature dimension is the maximum value of each of the active selection factors; The feature enhancement of the fusion feature based on the optimal feedback coefficient to obtain the time-enhanced feature includes: Perform normalization processing on the optimal feedback coefficient and use the normalized optimal feedback coefficient to perform feature enhancement on the fusion feature to obtain the time-enhanced feature.

6. The method according to claim 1, wherein The fusion feature is composed of multiple sub-fusion features, and each of the sub-fusion features corresponds to a different time window; the time-enhanced feature is composed of multiple sub-time-enhanced features, and each of the sub-time-enhanced features corresponds to a different time window; the obtaining of the predicted range extender temperature data based on the time-enhanced feature includes: Obtain the sub-time-enhanced feature corresponding to the current time window, and the sub-time-enhanced feature and sub-fusion feature of the previous time window of the current time window; the current time window is any one of each of the time windows; Use the sub-time-enhanced feature corresponding to the current time window, and the sub-time-enhanced feature and sub-fusion feature of the previous time window to obtain the time-series enhanced feature corresponding to the current time window; Input the time-series enhanced feature into the fully connected layer of the range extender temperature prediction model, and obtain the predicted range extender temperature data of the next time window of the current time window through the fully connected layer.

7. The method according to any one of claims 1 to 6, characterized in that The construction of the sample range extender temperature feature map based on the sample range extender temperature data of each of the time windows includes: Use each of the time windows as the graph nodes of the sample range extender temperature feature map, and use the sample range extender temperature data of each of the time windows as the node features of each of the graph nodes; Use each of the node features to obtain the similarity degree between each of the graph nodes, and according to the similarity degree, obtain the adjacency matrix for characterizing the connection relationship between each of the graph nodes; Construct the sample range extender temperature feature map according to the adjacency matrix and each of the node features.

8. A range extender control method, characterized in that, The method includes: Obtain the historical range extender temperature data of the target range extender corresponding to different time windows, and construct a historical range extender temperature feature map based on the historical range extender temperature data of each of the time windows; Input the historical range extender temperature feature map into the trained range extender temperature prediction model, and output the predicted range extender temperature data of the target range extender through the range extender temperature prediction model; wherein, the range extender temperature prediction model is trained by the range extender temperature prediction model training method according to any one of claims 1 to 7; Temperature control of the target range extender is performed using the predicted range extender temperature data.

9. A vehicle controller, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images