Automobile passenger compartment modeling method based on order reduction of neural network model

Abstract

The invention is suitable for the technical field of automobiles, and provides an automobile passenger compartment modeling method based on neural network model order reduction, and the method comprises the steps: building a three-dimensional passenger compartment and dummy model, combining with a human body thermal comfort evaluation index to obtain data, carrying out the steady-state order reduction through a neural network, and verifying the precision through a decision coefficient and an error index. And carrying out transient order reduction by adopting a long-short-term memory network optimized by a genetic algorithm to finally obtain a one-dimensional model. According to the method, the problems that traditional three-dimensional modeling is time-consuming in simulation and the one-dimensional modeling precision is insufficient are solved, the temperature and wind speed determination coefficient of each position of the established model reaches about 0.94, the model can be coupled with an air conditioning system, the simulation operation time is shortened, and support is provided for design of an automobile thermal management control strategy. The method comprises the steps of model establishment, steady-state and transient order reduction, working condition analysis, model verification and the like, and has the advantages of precision and efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of automobile technology, and in particular relates to a method for modeling an automobile passenger compartment based on neural network model order reduction. Background Art

[0002] During vehicle development, the air conditioning system must be continuously optimized and verified to ensure thermal comfort within the passenger compartment. Thermal management system modeling can provide valuable assistance for optimization and operation.

[0003] Air conditioning system simulations typically utilize one-dimensional modeling, as highly accurate results can be quickly obtained when modeling various components using one-dimensional methods. The passenger compartment model primarily includes three-dimensional modeling, one-dimensional modeling, and numerical modeling. The disadvantage of using three-dimensional modeling for the passenger compartment is long simulation times and high computational resource utilization. Modeling based on the one-dimensional lumped parameter method, on the other hand, assumes uniform temperature distribution within the cabin, resulting in less accurate simulation data. However, through simplification, the model can be quickly coupled with the air conditioning system, enabling rapid prediction and evaluation of thermal performance within the passenger compartment. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for modeling a vehicle passenger compartment based on neural network model order reduction, aiming to solve the technical problems existing in the prior art identified in the background technology.

[0005] The present invention is implemented as follows: a method for modeling a vehicle passenger compartment based on neural network model order reduction, comprising: 1. Establish a three-dimensional model and thermal comfort evaluation model: Construct a 1:1 3D model of the SUV passenger compartment and a dummy model, and perform geometric simplification and partitioning. Combined with the human thermal comfort evaluation index (PMV), the human body is divided into three zones: upper, middle and lower, and the weighted average PMV value is calculated. 2. Establishment and verification of steady-state environmental reduced-order model: The data is acquired through STAR CCM, using air conditioning parameters (wind speed, temperature, angle), solar radiation intensity, vehicle speed, and external temperature as inputs and human body temperature and humidity as outputs. The neural network training model was used to determine the coefficient of determination (R 2 ≥0.9), mean absolute error (MAE), and root mean square error (RMSE) verification accuracy.

[0006] 3. Steady-state operating condition analysis: Analyze the impact of internal (air supply parameters) and external (solar radiation, etc.) variables on thermal comfort and identify key factors.

[0007] 4. Establishment of transient environment reduced-order model: A long short-term memory network (LSTM) optimized by genetic algorithms is used to process time series data through a "gate" structure. 5. Transient model verification and one-dimensional model generation: Verification conditions: Temperature and wind speed determination coefficient R at each location 2 ≥0.94, the error meets the requirements, and a one-dimensional model that can be coupled with the air-conditioning system is generated.

[0008] The beneficial effects of the present invention are: This method combines a three-dimensional passenger compartment and dummy model with human thermal comfort evaluation indicators and utilizes a neural network for steady-state and transient order reduction. The resulting one-dimensional model achieves a coefficient of determination of approximately 0.94 for temperature and wind speed at each location, with minimal error between simulated and predicted values. This model combines the accuracy of a three-dimensional model with the efficiency of a one-dimensional model. This method uses a genetic algorithm to optimize a long-short-term memory network to handle transient conditions, accurately capturing the temporal dependencies of data. The resulting one-dimensional model can be directly coupled with the air conditioning system and thermal management system, significantly reducing the simulation computation time for the passenger compartment coupled to the air conditioning system. This provides high-precision and efficient modeling support for the design of automotive thermal management control strategies, enhancing the feasibility of optimizing the thermal comfort environment in the passenger compartment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flowchart of a method for modeling a vehicle passenger compartment based on neural network model order reduction provided by an embodiment of the present invention; Figure 2 To establish a three-dimensional passenger compartment model and dummy model; Figure 3 is the absolute coefficient of the reduced-order model of the crew cabin steady-state environment; Figure 4 This is the working condition analysis result diagram; Figure 5 The dummy is divided into zones and the thermal sensation index value of each zone; Figure 6 It is a static neural network model; Figure 7 This is the LSTM neuron diagram. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, 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 intended to limit the present invention.

[0011] like Figure 1 As shown, an embodiment of the present invention provides a method for modeling a vehicle passenger compartment based on neural network model order reduction, comprising the following steps: S101. Build a three-dimensional passenger compartment and dummy model Model construction: Taking the 1:1 SUV vehicle model as the object, the passenger compartment part is extracted, small structures such as seams and screws are simplified, and the inner surface areas are divided (A-pillars, B-pillars, instrument panel, auxiliary instrument panel, steering wheel, front and rear doors, tailgate, floor, roof, luggage compartment, front and rear left and right seats, rear seats, front and rear windshields, left and right front windows, left and right rear windows, left and right rear triangular windows, window frames, etc.). For the vehicle's exterior surface model, the radar, grille, seams, etc. are also simplified or filled, and the surface is cleaned and repaired. At the same time, after processing the passenger compartment model, the dummy model is adjusted and placed in the passenger compartment, and the partitions are divided. After adjusting the dummy's posture, it is placed in the cabin and the partitions are divided (such as Figure 2 shown).

[0012] Thermal Comfort Fusion: The human body is divided into three areas: upper, middle and lower. The PMV value of each area is calculated. The weighted formula is: ; The specific PMV calculation involves the following parameters: ; ; ; ; ; ; in: M is the metabolic rate; W The heat consumed in doing external work; is the thermal resistance of clothing; f cl It is the ratio of a person's body surface area when clothed to that when naked; is the air temperature; is the mean radiant temperature; is the air flow rate; p a is the water vapor partial pressure; is the convective heat transfer coefficient; is the surface temperature of the clothing; is the relative humidity of air; p s is the saturation pressure of moist air.

[0013] Data acquisition: Boundary conditions (such as air supply parameters and solar radiation) are set through STAR CCM, and the temperature and wind speed data of each monitoring point are obtained using field functions.

[0014] S102: Reduced-order model training in a steady-state environment Based on the passenger cabin model combined with a thermal comfort dummy, a data set is obtained through its three-dimensional model, and a reduced-order model is established using a neural network model.

[0015] For the steady-state condition of the passenger cabin thermal environment, the thermal comfort analysis is performed when the thermal environment reaches a stable state. It does not involve the transient state of dynamic time, so a static neural network model is established for the steady-state environment. The neural network model used in the steady-state environment is as follows: Figure 6 shown.

[0016] Establish evaluation indicators: The evaluation indicators include the following: model fidelity (Fidelity), mean absolute error (MAE) reflecting the average error value of the prediction results, mean absolute percentage error (MAPE) reflecting the percentage of error in the prediction value, root mean square error (RMSE) reflecting the degree of deviation between the prediction value and the measured value, and determination coefficient (R 2 ). The calculation formula is as follows: ; ; in is the standardized root mean square error, is the correlation coefficient of its model (e.g. linear correlation factor), is the true value, is the number of training models, is the predicted output value of the model. When one of the feature inputs of the model is a constant, the correlation coefficient is omitted.

[0017] ; ; ; ; in, is the average of the true values.

[0018] Based on the evaluation index evaluation model, The mean absolute error reflects the average error value of the prediction results. The smaller the value, the better the model fit. It reflects the degree of deviation between the predicted value and the measured value. The smaller the value, the better the fit of the model. It reflects the degree of fit between the predicted value and the measured value. The closer it is to 1, the better the fit. Generally, it is more than 0.9, which means the model is reliable. 2 ,Finally, the neural network reduced-order model was selected as the degradation model of the passenger compartment under steady-state environment.

[0019] S103: Steady-state operating analysis The resulting reduced-order model of the passenger cabin under steady-state conditions was compared with a three-dimensional passenger cabin model for operating condition analysis. The main focus was on the impact of internal and external cabin environmental variables on the thermal comfort of the cabin. The analysis included internal variables such as supply air temperature, air velocity, and air angle, as well as external variables such as solar radiation intensity, vehicle speed, and cabin external temperature.

[0020] By changing the input values ​​of the variables, the output results of the steady-state reduced-order model of the passenger compartment and the three-dimensional model of the passenger compartment, as well as the thermal comfort evaluation value of the passenger compartment, are obtained. The accuracy of the reduced-order model in the steady-state environment of the passenger compartment and the factors that have a significant impact on the thermal comfort of the passenger compartment are analyzed and compared. This can verify the feasibility of using neural networks to reduce the order of the three-dimensional model of the passenger compartment. Furthermore, factors that have a significant impact on the thermal comfort of the passenger compartment can be selected as verification conditions for the training model in the transient environment of the passenger compartment. (For example, Figure 4 (as shown in the working condition analysis results).

[0021] S104: Establishment of reduced-order model for transient environment Model selection: A neural network model was selected as the training scheme for the reduced-order model of the transient environment in the passenger compartment. Because transient conditions are absolutely time-dependent, a genetically optimized LSTM neural network was used to reduce the order of transient conditions. This gate structure includes the following: Forget Gate: ; Input Gate: ; Output Gate: ; The input and output of the LSTM model have an additional cell state. Figure 7 From the top, there is a straight line crossing the left and right sides above the neuron, namely and , this is for the previous one After performing some simple linear operations, this method can easily remember the historical information of the last time point. Figure 7 Middle, first from left to right Is the forget gate, which is a sigmoid function, its role is to decide which information to discard. and Represent the current input features and the features of the previous moment respectively. The forget gate converts the two into a vector between 0 and 1. The dimensions are the same, indicating How much information should be passed through, 0 means completely discarded, 1 means completely preserved.

[0022] (Sigmoid function) and The function is the input gate, which determines what information to input into the cell. The sigmoid layer determines what information to update. The output is . Function calculation generates a new vector , after combining the two, update the cell and get .

[0023] The latter part is the output gate, which is used to determine the output of the neuron and is affected by the cell state. First, after the sigmoid layer, the output gate is determined. Which features will be output. Then, through The layer maps the result to (-1, 1) and finally multiplies the two outputs to get the final output .

[0024] S105: Transient Model Verification and One-Dimensional Model Generation After obtaining the reduced-order model for the passenger compartment under transient conditions, its accuracy needed to be verified. Supply air temperature, wind speed, and wind angle, which exhibit significant impacts on passenger compartment thermal comfort under steady-state conditions, were selected as input variables. By comparing the results of the three-dimensional model with the reduced-order model and combining evaluation indicators, the credibility of the reduced-order model for the passenger compartment under transient conditions was confirmed. The resulting reduced-order model can be used as a one-dimensional model for vehicle thermal management performance simulation analysis.

[0025] A thermal balance analysis of the passenger compartment steady-state reduced-order model was conducted. The effects of air conditioning air velocity, air temperature, air outlet angle, solar radiation intensity, vehicle speed, and ambient temperature outside the passenger compartment on comfort were analyzed. Results were compared with those obtained from the three-dimensional model to verify the accuracy of the steady-state reduced-order model and prepare for reduced-order analysis under transient conditions. Further reduced-order training was conducted under transient conditions, using a neural network model as the training model. A genetic algorithm was also used for optimization, and an LSTM neural network was selected as the training model.

[0026] After completing the model training of the transient environment of the passenger compartment, the accuracy of the model was also verified. After confirming that the accuracy of the reduced-order model of the transient environment of the passenger compartment was reliable, it can be used as a one-dimensional model of the passenger compartment for coupled simulation with the air-conditioning system and the vehicle thermal management system. This solves the problems of long simulation time and high computing resource consumption when using a three-dimensional model of the passenger compartment, as well as low data accuracy and low simulation reliability when using the lumped parameter method to establish a one-dimensional model of the passenger compartment.

[0027] Accuracy verification: Taking the step change of supply air temperature as an example, the temperature and wind speed curves of each part of the three-dimensional model and the reduced-order model are compared. The mean value is 0.94, and the error is within 5% (as shown in Figure 4).

[0028] One-dimensional model application: This model can be coupled with the air conditioning system for thermal management strategy simulation, reducing calculation time compared to three-dimensional modeling.

[0029] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0030] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0031] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for modeling a car passenger compartment based on neural network model order reduction, characterized in that: The method comprises: Establish a three-dimensional passenger cabin model combined with a human thermal comfort evaluation method, and obtain reduced-order data based on simulation of the combined model; Using the reduced-order data to establish a steady-state environmental reduced-order model through a neural network, and evaluating the credibility of the steady-state environmental reduced-order model through the coefficient of determination, mean absolute error, and root mean square error; Based on the steady-state environmental reduced-order model, the influence of the internal and external environmental variables of the passenger cabin on thermal comfort under steady-state conditions is analyzed to verify the accuracy of the steady-state environmental reduced-order model; Select a neural network model with a coefficient of determination not less than 0.9 and a mean absolute error and root mean square error that meet the preset accuracy threshold after verification under steady-state conditions, use a genetic algorithm to optimize the long-short-term memory network, and establish a transient environment reduced-order model; The transient environment reduced-order model is trained and verified, and the one-dimensional model of the passenger compartment is obtained and its accuracy is verified.

2. The method according to claim 1, characterized in that The establishment of a three-dimensional passenger cabin model and its integration with a human thermal comfort evaluation method specifically includes: Build a 1:1 SUV 3D passenger compartment model and dummy model, perform geometric simplification and partitioning; The human body is divided into three zones: upper, middle, and lower. The PMV values ​​of each zone are calculated based on the zones set in Figure 5 and the local thermal sensation index of each zone. The weighted PMV values ​​are then weighted to obtain the overall average PMV. The weighted calculation formula is: ； in, represents the overall average PMV, Indicates the PMV value of each partition; is the regional local thermal sensation index; The calculation formula for the PMV value of each partition is: ； ； ； ； ； ； in: M is the metabolic rate; W The heat consumed in doing external work; is the thermal resistance of clothing; f cl It is the ratio of a person's body surface area when clothed to that when naked; is the air temperature; is the mean radiant temperature; is the air flow rate; p a is the water vapor partial pressure; is the convective heat transfer coefficient; is the surface temperature of the clothing; is the relative humidity of air; p s is the saturation pressure of moist air.

3. The method according to claim 1, characterized in that The establishment of the steady-state environment reduced-order model specifically includes: The internal and external environmental variables of the passenger compartment are used as inputs to the steady-state environmental reduced-order model; The neural network is trained using the stochastic gradient descent method, and the activation functions and hidden nodes of each hidden layer are adjusted according to different feature output results; The accuracy of the obtained steady-state environmental reduced-order model was verified by analyzing the coefficient of determination of the validation set, the error comparison between the true value and the predicted value in the validation set, and other error evaluation methods.

4. The method according to claim 3, characterized in that The internal and external environmental variables of the passenger cabin include: air conditioning air supply wind speed, temperature, angle, solar radiation intensity, vehicle speed, external ambient temperature, and the characteristic output is the temperature and wind speed of various parts of the human body.

5. The method according to claim 4, characterized in that The establishment of the transient environment reduced-order model specifically includes: The transient data set obtained by the 3D passenger compartment model in STAR CCM is used as the training data set for training the reduced-order model of the passenger compartment transient environment; Conduct steady-state verification to select neural network models with a coefficient of determination not less than 0.9 and a mean absolute error and root mean square error that meet the preset accuracy threshold, and establish a transient environment reduced-order model; The prepared dataset is used to train the transient reduced-order model of the passenger compartment to obtain the final transient environment reduced-order model.

6. The method according to claim 1, characterized in that The transient environment order reduction model adopts an LSTM neural network optimized by a genetic algorithm to process time series data through a forget gate, an input gate, and an output gate.

Images