A method for predicting energy consumption distribution in a thermal management system for plug-in hybrid electric vehicles

Abstract

This invention discloses a method for predicting the energy consumption distribution of a plug-in hybrid electric vehicle's thermal management system. The method includes the following steps: extracting relevant features of the core energy-consuming modules of the vehicle's thermal management system from driving data under various operating conditions based on historical data, and preprocessing the feature data; proposing a modeling strategy for real-to-ideal scenario conversion based on the complex coupling of the energy-consuming module's operating scenarios, and reconstructing the feature set; creating pseudo-labels for unlabeled data based on a semi-supervised learning framework under scenario mapping relationships, using the label values ​​to represent energy consumption values, and predicting the energy consumption distribution of different core energy-consuming modules in the thermal management system; inputting real vehicle test data into the proposed framework to verify the reliability of the proposed method. This invention, based on a semi-supervised learning framework, creates pseudo-labels for unlabeled data, using the pseudo-label values ​​as predicted energy consumption values, and can predict the energy consumption distribution of different core energy-consuming modules in the thermal management system.

Description

Technical Field

[0001] This invention belongs to the field of vehicle engineering, and specifically relates to a method for predicting energy consumption distribution in the thermal management system of a plug-in hybrid electric vehicle. Background Technology

[0002] Currently, the use of new energy vehicles generates a large amount of operational data. Storing, mining, and analyzing this data through cloud computing technology enables in-depth analysis of the characteristics of new energy vehicle products. The application of big data analytics in the automotive cloud can help automakers deeply understand product pain points, optimize product design, and provide a better driving experience, thus offering safer, more convenient, and intelligent solutions for people's travel.

[0003] The vehicle thermal management system is a complex system composed of multiple interconnected subsystems. It coordinates, optimizes, and controls the matching of vehicle components and subsystems from a holistic vehicle perspective. For example, efficient engine cooling significantly reduces fuel consumption, while the heating and cooling efficiency of the air conditioning system directly affects passenger experience. An excellent thermal management system effectively solves vehicle-wide thermal issues, ensuring that each functional module operates within its optimal temperature range, improving overall vehicle economy and performance, and guaranteeing safe driving.

[0004] In recent years, with the development of big data in electric vehicles and the rise of intelligent algorithms, data-driven prediction algorithms have gradually become a hot topic in automotive thermal management energy consumption prediction research. Data-driven methods extract information from driving data under different operating conditions and use methods such as neural networks and Gaussian regression to learn the potential relationships between this information and the thermal management energy consumption of each module, thereby achieving the effect of energy consumption distribution prediction. However, the difficulty of data-driven methods lies in how to select the features most correlated with thermal management energy consumption from massive amounts of driving data and predict results with the highest possible reliability. On the other hand, the machine learning training process for prediction purposes generally has a high dependence on labels; how to achieve effective prediction in scenarios with a small amount of labeled data and a large amount of unlabeled data is also one of the challenges.

[0005] The current problems with this method are that when the number of selected feature data labels is insufficient, the learning effect and prediction accuracy of the data-driven method will decrease significantly; when several energy-consuming modules share the same energy source, it is difficult to extract the energy consumption data of different modules independently; and when there are other energy-consuming devices in the energy-consuming modules that cannot be recorded, it is difficult to extract the relevant feature data for training and learning.

[0006] A battery thermal management method based on model predictive control (CN202010062895.7) establishes an electrical system including the transmission system and battery pack. hot The method involves a system model including aging multi-state estimation and the cooling system; designing a state estimator and cost function for the model predictive controller; establishing the coupling between the vehicle speed prediction model and the control system; determining the optimal battery temperature reference values ​​under different ambient temperatures and coupling them with the control system; and achieving vehicle battery thermal management. While this method is low in complexity and highly feasible, and considers battery temperature management, aging management, and cooling system energy consumption management within the control system, it does not extend to the overall vehicle thermal management system level. Other energy-consuming modules within the thermal management system, such as the motor and cabin, are not analyzed or studied. Summary of the Invention

[0007] This invention proposes a method for predicting energy consumption distribution in the thermal management system of plug-in hybrid electric vehicles. First, relevant features of the three energy-consuming modules—motor, battery, and cabin—are extracted from historical data under various operating conditions. These features reflect the module's voltage, current, temperature, and switching status. The feature data undergoes preprocessing, including feature selection, missing value filling, label calculation, and representation of unknown energy consumption data. An ideal operating scenario is constructed by checking unlabeled feature data. Next, a semi-supervised learning model is trained, and model parameters are optimized. High-reliability pseudo-labels generated after training are recorded as energy consumption values.

[0008] The present invention is achieved by at least one of the following technical solutions.

[0009] A method for predicting energy consumption distribution in a thermal management system for a plug-in hybrid electric vehicle includes the following steps: Driving data under various operating conditions will be input into the trained semi-supervised learning model to predict the energy consumption distribution of different energy-consuming modules; the training of the semi-supervised learning model includes the following steps: Step 1: Extract historical data features for all vehicles and preprocess them; the extracted features include status features and other features. Step 2: Add additional feature data, check the working status of each energy-consuming module, and perform feature engineering according to the working scenario; Step 3: Set the label values ​​for each energy-consuming module; construct feature label datasets for real and ideal working scenarios respectively; Step 4: Divide the dataset into training and testing sets, perform preliminary training on the semi-supervised learning model, and retrain the model using the feature values ​​and label values ​​extracted from the historical data generated by the target vehicle. Step 5: After the model training is completed, record the generated pseudo-label values.

[0010] Furthermore, in step 1, the state characteristics include the operating voltage, current, temperature, and switch state variables of the electric vehicle's historical data, with the variable values ​​being the measured values ​​of the state variables at different times.

[0011] Furthermore, in step 1, other features include battery thermal demand level, driver's seat temperature setting, and rear seat temperature setting.

[0012] Furthermore, in step 2, the additional feature refers to the operating status of other energy-consuming devices that may not be measured or recorded.

[0013] Furthermore, in step 2, feature engineering assigns only 0 or 1 to the additional features, where 0 indicates that other energy-consuming devices are not working, and vice versa. In the tag data, the additional features indicate the presence of unknown devices.

[0014] Furthermore, in step 4, the preliminary training of the semi-supervised learning model refers to ignoring the calculation of unlabeled loss and only considering the training process with labeled loss, including freezing shallow feature extraction, model parameter tuning, checking data fit, and setting physical consistency loss. Retraining refers to resuming the calculation of unlabeled loss and retraining the initially trained model, while keeping the selected features and model parameters consistent.

[0015] Furthermore, the loss function setting for the semi-supervised learning model includes an uncertainty-based filtering component and a similarity-based pseudo-label calibration component.

[0016] Furthermore, the pseudo-label values ​​refer to the calibrated pseudo-labels generated based on unlabeled data after the model training is completed, including pseudo-labels for the motor module, battery module, and cockpit module.

[0017] A computer device according to the present invention includes a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, which, when executed by the processor, causes the processor to implement the method described herein.

[0018] The present invention provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor implements the method described herein.

[0019] Compared with the prior art, the beneficial effects of the present invention are at least as follows: 1. This invention establishes a feature set that includes different operating temperatures, different operating voltages, and different operating currents, reflecting the differences in working conditions under different operating conditions.

[0020] 2. This invention proposes a modeling strategy for real-to-ideal scenario transformation. By constructing an ideal working scenario from an unlabeled dataset, different energy-consuming modules with the same energy supply are processed separately, achieving accurate prediction of the energy consumption of each module under limited data. Visualizing potential energy-consuming devices in feature data improves the reliability of the prediction results.

[0021] 3. This invention provides a basis for estimating the energy consumption distribution of an automotive thermal management system. It uses a semi-supervised learning framework to create pseudo-labels for unlabeled data and uses the pseudo-label values ​​as predicted energy consumption values, which can predict the energy consumption distribution of different core energy-consuming modules in the thermal management system. Attached Figure Description

[0022] Figure 1 This is a flowchart of the overall process of predicting the energy consumption distribution of the thermal management system according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the energy consumption structure of the vehicle thermal management system according to an embodiment of the present invention; Figure 3 This is a diagram illustrating the "real-ideal" work scenario modeling strategy of this invention. Figure 4a This is a graph showing the estimated energy consumption distribution of the battery module under low-temperature conditions according to an embodiment of the present invention.

[0023] Figure 4b This is a graph showing the estimated energy consumption distribution of the cockpit module under low-temperature conditions according to an embodiment of the present invention.

[0024] Figure 5a This is a graph showing the estimated energy consumption distribution of the battery module under high-temperature conditions according to an embodiment of the present invention.

[0025] Figure 5b This is a graph showing the estimated energy consumption distribution of the cockpit module under high-temperature conditions according to an embodiment of the present invention.

[0026] Figure 6 This is a graph showing the energy consumption distribution estimation results under normal operating conditions in an embodiment of the present invention. Detailed Implementation

[0027] To make the technical solutions and objectives of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific implementation steps described herein are only used to better illustrate the application of the present invention, but the technical features involved in the embodiments of the present invention are not limited thereto.

[0028] Please see Figures 1-2 The energy consumption distribution prediction method for the thermal management system of a plug-in hybrid electric vehicle provided by the present invention includes the following steps: Step 1: Extract historical data features from all vehicles and preprocess them. The extracted features include state features and other features. State features refer to the operating voltage, current, temperature, and on / off status variables of the electric vehicle's historical data; these variables are measured values ​​at different times. Other features include battery thermal demand level, driver's seat temperature setting, and rear seat temperature setting.

[0029] In some embodiments of the present invention, please refer to the appendix for the vehicle thermal management system architecture. Figure 2 ; In some embodiments of the present invention, the data preprocessing method in step 1 includes: 1. Extracting features at the time when all feature data of each module exists. 2. Filling zero values ​​with the mean value for the feature data based on the working state of the battery and cockpit modules. Taking the battery voltage feature data of the battery module as an example, the average value a1 of the battery voltage during the working period of the battery module is calculated, and the average value a1 is used to replace the feature data where the battery voltage is 0 during the working period of the battery module; similarly, the average value a2 of the battery voltage during the stopping period of the battery module is calculated, and the average value a2 is used to replace the feature data where the battery voltage is 0 during the stopping period of the battery module.

[0030] Step 2: Add additional feature data, check the working status of each energy-consuming module, and perform feature engineering according to the working scenario.

[0031] In some embodiments of the present invention, step 2 adds an additional feature data type, "Others," to the feature data of the battery module and cockpit module to characterize the energy consumption of other potential devices. Similar to the feature data indicating the module's operating status, the additional feature data "Others" is assigned only a value of 0 or 1, where 0 indicates that other energy-consuming devices are not working, and vice versa. By adding the additional feature column "Others," the model learns the impact of unknown devices on system energy consumption during training. In labeled data, the additional feature "Others" indicates the presence of unknown devices; in unlabeled data, by setting the additional feature to zero, the model predicts based solely on known features, indirectly estimating the energy consumption of known modules.

[0032] Step 3: Set the label values ​​for each energy-consuming module; construct feature label datasets for real and ideal working scenarios respectively.

[0033] Constructing a realistic working scenario ensures that, with labeled data centralized, the working status of each module and device at any given time matches the real operating environment for the feature data. Constructing an ideal working scenario means ensuring that, with unlabeled data, for any given time, at most one of several modules based on the same energy source is in a working state.

[0034] In some embodiments of the present invention, the label setting method in step 3 is as follows: For the motor module, the product of motor current and motor voltage is used as the tag value. For the battery module and cockpit module, the operating conditions of both being powered by PTC or ECP under different operating conditions are considered; specifically, PTC heating is used under low-temperature conditions, and ECP cooling is used under normal and high-temperature conditions. After data inspection, the operating ranges of the battery module and cockpit module do not overlap under low-temperature conditions, meaning there is no time when the two modules work simultaneously; therefore, PTC power is used as the tag for both the battery module and cockpit module. Under normal operating conditions, the battery module does not require cooling; therefore, ECP power is used as the tag for the cockpit module. Under high-temperature conditions, the operating ranges of the battery module and cockpit module overlap; after merging the feature data of the two modules into a single feature set, ECP power is set as the tag.

[0035] In some embodiments of the present invention, the method for constructing the "real-ideal" work scenario in step 3 is as follows: In a real working scenario, there should be other energy-consuming devices, so the additional feature Others is 1; in an ideal working scenario, other energy-consuming devices are in a shutdown state, so the additional feature Others is 0.

[0036] In low-temperature operating conditions, a complete real-world operating scenario includes tagged data on the motor module, tagged data on the battery module, and tagged data on the cockpit module; an ideal operating scenario includes tagged data on the battery module and tagged data on the cockpit module.

[0037] Under normal operating conditions, a complete real-world working scenario includes both the motor module and the cockpit module having tagged data; an ideal working scenario includes the cockpit module having no tagged data.

[0038] Under high-temperature conditions, in addition to numerical processing of the additional "Others" column, we also process feature data representing whether the module is operational (for the battery module, the feature is a data request; for the cockpit module, the feature is the panel air conditioning switch; a non-zero feature value indicates operation, and a zero feature value indicates inoperability). We check and modify data to ensure that at most one module (either the battery module or the cockpit module) is operational at any given time. A complete real-world high-temperature operating scenario includes tagged data on both the motor module and the battery / cockpit module; an ideal high-temperature operating scenario includes no tagged data on either the battery module or the cockpit module.

[0039] In some embodiments of this invention, historical data from real-world multi-source operating scenarios are used to train the model, learning the potential relationship between the feature data of each module and energy consumption. As the model is trained until convergence, the unlabeled dataset is forced to construct an "ideal operating scenario" through feature engineering—that is, artificially set so that only the target module is working while other related modules and unknown devices are in a non-working state. Its predictive behavior is constrained and guided towards reasoning about the idealized, single-module operating state. Therefore, the pseudo-labels ultimately generated by the model can be directly interpreted as the independent energy consumption values ​​of each core module under ideal operating conditions.

[0040] Step 4: Divide the dataset into training and testing sets to perform initial training on the semi-supervised learning model. Then, retrain the model using feature values ​​and label values ​​extracted from the historical data generated by the target vehicle.

[0041] In some embodiments of the present invention, the semi-supervised learning model in step 4 improves the reliability of pseudo-labels in the regression environment by applying strong and weak enhancements to the unlabeled data and then inputting it into the model along with the labeled data. The loss function setting includes two parts: an uncertainty-based filtering part and a similarity-based pseudo-label calibration part.

[0042] Based on the uncertainty filtering component, Monte Carlo dropout technique is used to calculate the variance of the predicted values. Specifically, it can be expressed as:

[0043] In the formula, Indicates the number of model predictions. Indicates the first m In the first round of predictions j Predicted values ​​for weakly enhanced samples, Indicates the first j A weakly enhanced sample The average of the predictions from each round is used as the pseudo-label for the sample in subsequent processing.

[0044] In the similarity-based pseudo-label calibration section, the similarity distribution of unlabeled samples is defined. This leads to the introduction of similarity-based pseudo-labels. The specific calculation formula is expressed as follows:

[0045]

[0046] In the formula, Indicates the first j A weakly enhanced sample, Indicates the first i A representation vector of a labeled sample, Indicates the first k A representation vector of a labeled sample, T Indicates temperature hyperparameter and T >0 is used to control the sharpness of the distribution. This indicates the number of labeled samples. Indicates the first i The label values ​​of a labeled sample.

[0047] The formula for calculating the unsupervised regression loss function is:

[0048]

[0049] In the formula, This indicates the loss for unlabeled samples. Indicates the number of unlabeled samples. This represents the relative uncertainty threshold and τ>0. Indicates the first j Predicted values ​​of a strongly enhanced sample, Indicates the first j The pseudo-labels after calibration of a weakly enhanced sample This represents the balancing hyperparameter and is located between 0 and 1.

[0050] The supervised loss function is expressed as:

[0051] In the formula, This indicates that there is a loss of labeled samples. , They represent the first i The true label value and the predicted value of each sample.

[0052] In some embodiments of the present invention, the physical consistency regularization setting of the semi-supervised learning model in step 4 includes: First, since the predicted energy consumption value is a physical power output, it must lie within a non-negative reasonable space. Therefore, the energy consumption prediction value of the model is forced to satisfy non-negativity, that is:

[0053] In the formula, This represents a nonnegative physical consistency regularization term.

[0054] Secondly, considering that both the battery module and the cockpit module are heated by PTC or cooled by ECP under different operating conditions, a power upper limit constraint is set to ensure that the predicted energy consumption of a single core module does not exceed the total output power of its corresponding energy supply under the current operating conditions, thus avoiding the model from generating predictions that exceed the physical capabilities of the thermal management system.

[0055] In the formula, This indicates a power output physical consistency regularization term. This indicates the total power of the PTC or ECP.

[0056] Furthermore, the model's prediction results must be strictly synchronized with the actual operating state of the core energy-consuming modules. This is determined by reading the feature values ​​representing the operating state of the corresponding energy-consuming modules: when a module is off, its predicted energy consumption is forced to zero; only when the module is on is the model allowed to output non-zero energy consumption.

[0057] In the formula, Regular expressions indicating physical consistency of working status. This indicates the working state of the j-th sample, and is either 1 or 0. =1 indicates that it is in working status. =0 indicates that it is in a non-working state.

[0058] In some embodiments of the present invention, the overall loss function of the semi-supervised learning model in step 4 is:

[0059]

[0060] In the formula, This represents the unlabeled loss weight coefficient. This represents the physical consistency regularization loss.

[0061] Step 5: After model training is complete, record the generated pseudo-label values. The significance of the pseudo-labels lies in the power value of the corresponding module at each moment under the current operating conditions, including the pseudo-labels of the motor module, battery module, and cockpit module.

[0062] In some embodiments of the present invention, the reliability verification of the pseudo-tags in step 5 uses root mean square error, mean absolute error, and coefficient of determination as evaluation indicators, specifically expressed as follows:

[0063]

[0064]

[0065] In the formula, M Indicates the number of samples. , They represent the first i The true value and predicted value of a sample This represents the average value.

[0066] The root mean square error (RMSE) and mean absolute error (MAE) quantify the degree of deviation between the true and predicted values. Lower RMSE and MAE values ​​indicate that the model's predictions are generally more accurate and do not produce severe prediction distortion. The coefficient of determination (R²) 2 It reflects the model's ability to explain the volatility of the target variable. Its numerical range is between negative infinity and 1. The closer it is to 1, the higher the model's goodness of fit to the data changes, and the more consistent the trend of the predicted value with the actual value.

[0067] The invention will be further described below with reference to the accompanying drawings and experimental examples.

[0068] Based on actual collected data and the vehicle thermal management system architecture, representative feature data of each core energy-consuming module are selected as shown in Table 1. During data extraction, considering factors such as avoiding missing values ​​and ensuring consistent timestamps, only the moments when all feature data are simultaneously present are retained. Zero values ​​in the data are handled as follows: based on the operating switch parameters of each module, the average value of each feature data during both the module's operating and non-operating periods is calculated. Then, a mean-filling method is used to replace zero values, achieving data restoration of the core energy-consuming modules under different operating conditions. Three basic feature sets are constructed for the motor module, battery module, and cabin module. Table 1 Feature Data

[0069] A feature dataset based on a real-world, ideal working scenario is constructed. The dataset distribution is shown in Table 2. For the relationship between the datasets of each module and the model, please refer to [link / reference]. Figure 3 .

[0070] Table 2 Feature Dataset Distribution

[0071] In the formula, , , These represent labeled datasets for the motor module, battery module, and cockpit module under low-temperature operating conditions, respectively. , These represent labeled datasets for the motor module and cockpit module under normal operating conditions, respectively. , These are labeled datasets representing the motor module, battery, and cockpit common module under high-temperature operating conditions, respectively. , These represent the unlabeled datasets for the battery module and cockpit module under low-temperature operating conditions, respectively. This represents the unlabeled dataset of the cockpit module under normal operating conditions. , These represent the unlabeled datasets for the battery module and cockpit module under high-temperature operating conditions, respectively.

[0072] The parameter settings of the semi-supervised learning framework under scene mapping are shown in Table 3. A ResNet50 model was trained for 100 epochs using the Adamw optimizer with a learning rate of 0.0001. The unlabeled loss weight coefficient was set to 0.01, the temperature parameter T was set to 1, and the equilibrium parameter β was set to 0.5, as shown in Table 3. Each sample was predicted 5 times. An ablation experiment was also set up. The original data was not decoupled and was directly input into the semi-supervised learning framework to predict the energy consumption values ​​of each core module as a comparison method. This comparative analysis aims to highlight the advantages of the proposed method in improving the accuracy of energy consumption distribution estimation in automotive thermal management systems.

[0073] Table 3 Hyperparameter Settings

[0074] In the formula, Epoch represents the training epoch, Optimizer represents the optimizer type, Learning rate represents the learning rate, and Batch size represents the sample size of each batch. λ u This represents the unlabeled loss weight coefficient. T This indicates temperature hyperparameters. β Indicates the balance hyperparameters, n This indicates the number of predicted values ​​from the model.

[0075] To verify the superiority of the proposed method, the original data were not decoupled and were directly input into a semi-supervised learning framework to predict the energy consumption values ​​of each core module as a comparison method. This comparative analysis aims to highlight the advantages of the proposed method in improving the accuracy of energy consumption distribution estimation in automotive thermal management systems.

[0076] Because the power-consuming devices outside the core modules of the PTC and ECP are not decoupled, the PTC / ECP power values ​​cannot be directly used as labels for the feature data of the battery module and cockpit module. Therefore, the PTC / ECP power values ​​will be input into the model as one of the feature data of the battery module and cockpit module. Thus, only labeled datasets for the motor module and unlabeled datasets for the battery module and cockpit module are obtained.

[0077] In Table 4, the comparison results of evaluation metrics after training are shown. The proposed method achieved an R-value close to 1 on the validation set under all three operating conditions. 2The proposed method accurately reproduces the actual energy consumption variation patterns. Simultaneously, both RMSE and MAE values ​​remain at low levels, with MAE slightly lower than RMSE, indicating a relatively concentrated prediction error distribution and the absence of numerous extreme error points, demonstrating high stability and reliability. The proposed method consistently outperforms the comparative methods under all operating conditions. (Including MAE, RMSE, and coefficient of determination R...) 2 All evaluation indicators, including their average values, outperformed the direct prediction method.

[0078] Table 4 Comparison of Energy Consumption Distribution Estimation Results

[0079] Finally, for the pseudo-label values ​​after training, i.e., the energy consumption values ​​of each core energy-consuming module under different operating conditions, please refer to [link / reference needed]. Figures 4a-6 .from Figures 4a-4b It can be seen that under low-temperature operating conditions, the management energy consumption of the battery module is generally between 1500W and 3000W, with a maximum operating time of about 20 minutes; the management energy consumption of the cockpit module is generally between 1400W and 2000W, with a maximum operating time of about 5 minutes. Compared to the former, the battery clearly accounts for a higher proportion of energy consumption in the PTC module and operates for a longer period. Figure 6 As can be seen, under normal operating conditions, due to the near-normal temperature, the raw data shows that the battery module's flow request is consistently 0, meaning the battery module is not working (not dissipating heat). Under normal operating conditions, the cockpit module's power is typically between 150 and 700W, with a maximum single operating time of approximately 20 minutes. From... Figures 5a-5b It can be seen that under high-temperature conditions, the power range of the battery module is approximately 300~2000W, while that of the cockpit module is approximately 200~3500W. The power range of the two modules is quite large, and the peak values ​​both appear in the later stages of the measurement. The cockpit module exhibits a clear operating characteristic of power decreasing from high to low.

[0080] This invention addresses the problems of existing methods for estimating energy consumption distribution in automotive thermal management systems applied in vehicle engineering, which rely on large amounts of labeled data and struggle to decouple multiple factors in complex operating environments. It designs a mapping mechanism from real-world scenarios to ideal scenarios. The model is first trained on real-world operating scenario data containing various unknown disturbances and module coupling. By introducing additional feature data ("Others"), the model learns the overall energy consumption characteristics and potential patterns of the system. Based on this, an ideal operating scenario is actively constructed from an unlabeled dataset, guiding the trained model to isolate coupling factors and random disturbances from real-world operation. Finally, the pseudo-labels output by the model represent pure estimates of the specific energy consumption of each core module under ideal, isolated conditions, achieving accurate estimation of the independent energy consumption of core energy-consuming modules in automotive thermal management systems under ideal operating scenarios.

[0081] It should be noted that although the implementation of the present invention has been described in detail with reference to examples, those skilled in the art will readily understand that any modifications, substitutions, and improvements made without departing from the spirit and principles of the invention as set forth in the appended claims should be included within the scope of protection of the present invention.

Claims

1. A method for predicting energy consumption distribution in a thermal management system for a plug-in hybrid electric vehicle, characterized in that, Includes the following steps: Driving data under various operating conditions will be input into the trained semi-supervised learning model to predict the energy consumption distribution of different energy-consuming modules; the training of the semi-supervised learning model includes the following steps: Step 1: Extract historical data features for all vehicles and preprocess them; the extracted features include status features and other features. Step 2: Add additional feature data, check the working status of each energy-consuming module, and perform feature engineering according to the working scenario; Step 3: Set the label values ​​for each energy-consuming module; construct feature label datasets for real and ideal working scenarios respectively; Step 4: Divide the dataset into training and testing sets, perform preliminary training on the semi-supervised learning model, and retrain the model using the feature values ​​and label values ​​extracted from the historical data generated by the target vehicle. Step 5: After the model training is completed, record the generated pseudo-label values.

2. The energy consumption distribution prediction method for a plug-in hybrid electric vehicle thermal management system according to claim 1, characterized in that, In step 1, the state characteristics include the operating voltage, current, temperature, and switch state variables of the electric vehicle's historical data, with the variable values ​​being the measured values ​​of the state variables at different times.

3. The energy consumption distribution prediction method for a plug-in hybrid electric vehicle thermal management system according to claim 1, characterized in that, In step 1, other features include battery thermal demand level, driver's seat temperature setting, and rear seat temperature setting.

4. A method for predicting energy consumption distribution in a thermal management system for a plug-in hybrid electric vehicle as described in claim 1, characterized in that, In step 2, the additional feature refers to the operating status of other energy-consuming devices that may not be measured or recorded.

5. A method for predicting energy consumption distribution in a thermal management system for a plug-in hybrid electric vehicle based on claim 1, characterized in that, In step 2, feature engineering assigns only 0 or 1 to the additional features, where 0 indicates that other energy-consuming devices are not working, and vice versa. In the tag data, the additional features indicate the presence of unknown devices.

6. A method for predicting energy consumption distribution in a thermal management system for a plug-in hybrid electric vehicle based on claim 1, characterized in that, In step 4, the preliminary training of the semi-supervised learning model refers to ignoring the calculation of unlabeled loss and only considering the training process with labeled loss, including freezing shallow feature extraction, model parameter tuning, checking data fit, and setting physical consistency loss. Retraining refers to resuming the calculation of unlabeled loss and retraining the initially trained model, while keeping the selected features and model parameters consistent.

7. A method for predicting energy consumption distribution in a thermal management system for a plug-in hybrid electric vehicle as described in claim 1, characterized in that, The loss function setting for the semi-supervised learning model includes an uncertainty-based filtering component and a similarity-based pseudo-label calibration component.

8. A method for predicting energy consumption distribution in a thermal management system for a plug-in hybrid electric vehicle based on claim 1, characterized in that, The pseudo-label values ​​refer to the calibrated pseudo-labels generated based on unlabeled data after the model training is completed, including pseudo-labels for the motor module, battery module, and cockpit module.

9. A computer device comprising a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, characterized in that: When the computer program is executed by the processor, it causes the processor to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor implements the method as described in any one of claims 1 to 8.

Images