Vehicle air conditioning system simulation model order reduction method and storage medium

Abstract

The invention relates to an order reduction method for a simulation model of a vehicle air-conditioning system. The order reduction method comprises the following steps: determining an input variable, an output variable, a value range of the input variable and an inspection range of the output variable of the simulation model of the air-conditioning system; performing binning operation on the value range of the input variable, and constructing an initial training condition set according to a box set obtained by the binning operation; determining a plurality of effective training conditions, and constructing an effective training data set; determining a plurality of verification working conditions, and constructing an effective verification data set; constructing an artificial neural network model taking the input variable as input and taking the output variable as output; and training the artificial neural network model by using the effective training data set, and verifying the artificial neural network model by using the effective verification data set to obtain a reduced-order model. The invention further provides a storage medium. According to the method, the simulation model of the air conditioning system can be simplified, and the reduced-order model with higher calculation efficiency is constructed on the premise that necessary simulation precision is reserved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle simulation analysis, and particularly relates to a vehicle air conditioning system simulation model reduction method and a storage medium. BACKGROUND

[0002] The vehicle air conditioning system is responsible for providing a good driving and riding environment for the driver and passengers. On the one hand, the refrigeration and heating performance of the vehicle air conditioning system under different boundary conditions directly determines the thermal comfort performance of the passenger compartment. On the other hand, when the defrosting and defogging functions of the vehicle air conditioning system perform poorly, it may cause driving safety problems. Therefore, the development of the vehicle air conditioning system is an essential part of the development of the whole vehicle product.

[0003] Computer-aided simulation is an important means to assist the development of the vehicle air conditioning system, which usually uses a vehicle air conditioning system simulation model to simulate and analyze the performance of the vehicle air conditioning system design scheme under various operating conditions. Compared with the traditional development method based on physical tests, computer-aided simulation has many advantages such as fast speed, low cost, and easy to modify the design scheme, and has been widely applied in recent years.

[0004] Although the application of computer-aided simulation in the development of the vehicle air conditioning system has the above-mentioned advantages, it also faces the following technical problems in the actual application process:

[0005] Using a high-precision vehicle air conditioning system simulation model for simulation analysis can obtain accurate vehicle air conditioning system simulation analysis results. However, although the high-precision vehicle air conditioning system simulation model can help improve the accuracy of the simulation analysis results, it has the problem of low computational efficiency.

[0006] In the calibration and development of the vehicle air conditioning system control strategy, in order to seek the optimal control strategy of the vehicle air conditioning system under certain operating conditions, iterative simulation is often needed to compare the differences in thermal management performance of various strategies. In recent years, when new intelligent control algorithms such as reinforcement learning algorithm and model predictive control algorithm are applied, a large number of simulations based on the vehicle air conditioning system simulation model are needed to explore the response of the vehicle air conditioning system under various operating conditions. Using a high-precision vehicle air conditioning system simulation model for iterative simulation and a large number of simulations requires a large amount of computing resources, and therefore is not suitable for control strategy optimization and rapid analysis and evaluation of design schemes. SUMMARY

[0007] The purpose of the present application is to provide a vehicle air conditioning system simulation model reduction method and a storage medium to alleviate or eliminate at least one of the above technical problems.

[0008] The vehicle air conditioning system simulation model reduction method provided by the present application comprises the following steps:

[0009] determining input variables, output variables, value ranges of the input variables and observation ranges of the output variables of a simulation model of an air conditioning system;

[0010] performing a binning operation on the value ranges of the input variables, determining a plurality of initial training operating conditions according to a bin set obtained by the binning operation, and constructing an initial training operating condition set; the binning operation comprises: performing an equal-interval binning operation on the value ranges of the input variables to obtain a plurality of basic bins, determining all basic bin combinations composed of the basic bins arranged continuously, and merging all the basic bin combinations respectively to obtain a plurality of combined bins;

[0011] determining a first boundary value operating condition and a second boundary value operating condition in the initial training operating condition set, the input variable of the first boundary value operating condition taking the lower limit of the basic bin, the input variable of the second boundary value operating condition taking the upper limit of the basic bin, performing simulation based on the first boundary value operating condition and the second boundary value operating condition respectively by using the simulation model of the air conditioning system, and determining an average value of a simulation duration of the first boundary value operating condition and a simulation duration of the second boundary value operating condition as an effective simulation duration T0;

[0012] determining a transition simulation duration T1 of the first boundary value operating condition according to a simulation result of the first boundary value operating condition, determining a transition simulation duration T2 of the second boundary value operating condition according to a simulation result of the second boundary value operating condition, and determining a larger value T max from T1 and T2 as a transition simulation duration T max when T max is less than T0, determining a complete simulation duration as T0+T

[0013] performing simulation based on the initial training operating condition set respectively by using the simulation model of the air conditioning system, and respectively intercepting simulation data after 0 to T max from simulation results of a plurality of the initial training operating conditions as interception data, determining average values of the output variables in a plurality of the initial training operating conditions based on a plurality of the interception data respectively, taking an initial training operating condition whose average value of the output variable meets a preset condition as an effective training operating condition, determining an effective training operating condition set, and constructing an effective training data set by using a plurality of the interception data corresponding to a plurality of the effective training operating conditions respectively;

[0014] selecting part of the effective training operating conditions from the effective training operating condition set as verification operating conditions, and determining a verification operating condition set;

[0015] performing simulation based on the verification operating condition set by using the simulation model of the air conditioning system, and respectively intercepting simulation data after 0 to T maxThe simulation data after the time period is used to build an effective validation data set;

[0016] An artificial neural network model is built, which takes the input variables as input and outputs the output variables;

[0017] The artificial neural network model is trained using the effective training data set and validated using the effective validation data set, and a reduced-order model is obtained.

[0018] Optionally, there are multiple input variables, and the box set obtained by the binning operation determines multiple initial training operating conditions, including the following steps: performing Cartesian product combination on the box set of multiple input variables to obtain multiple box combinations, and determining multiple initial training operating conditions each corresponding to a box combination.

[0019] Optionally, after determining multiple initial training operating conditions each corresponding to a box combination, the following step is further included: classifying the multiple initial training operating conditions according to the value range size characteristics of each box in the multiple box combinations to determine multiple operating condition categories.

[0020] The step of selecting part of the effective training operating conditions as validation operating conditions from the effective training operating condition set includes the following step: randomly selecting one effective training operating condition as the validation operating condition from each operating condition category.

[0021] Optionally, the following steps are further included: normalizing the value range of the input variables to a first preset interval, and normalizing the observation range of the output variables to a second preset interval.

[0022] The preset condition is that the average value of the output variables is within the second preset interval after normalization.

[0023] Optionally, the step of determining the transition simulation duration T1 of the first boundary value operating condition according to the simulation result of the first boundary value operating condition includes the following steps: calculating the first average value signal of the output variables in the simulation result of the first boundary value operating condition, the value at the termination time of the first average value signal is M1, the last time when the value of the first average value signal is equal to 0.95*M1 is T 11 , the last time when the value of the first average value signal is equal to 1.05*M1 is T 12 , and T1=Max(T 11 , T 12 );

[0024] The step of determining the transition simulation time length T2 of the second edge value working condition according to the simulation result of the second edge value working condition comprises the following steps: calculating a second average value signal of the output variable in the simulation result of the second edge value working condition, the value of the second average value signal at the end time is M2, and the last time when the value of the second average value signal is equal to 0.95*M2 is T 21 , the last time when the value of the second average value signal is equal to 1.05*M2 is T 22 , and T2 = Max(T 21 , T 22 ).

[0025] Optionally, when the air conditioning system simulation model is used to perform simulation based on the first edge value working condition and the second edge value working condition respectively, the input variable of the air conditioning system simulation model is a uniformly distributed sampling signal, and the upper and lower limits of the uniformly distributed sampling signal are determined by the basic box of the input variable in the first edge value working condition and the basic box of the input variable in the second edge value working condition.

[0026] Optionally, the artificial neural network model comprises a first full connection layer, an RNN layer and a second full connection layer connected in sequence, the number of units of the first full connection layer is: the number of input variables * the number of output variables * K, K is a training parameter, the number of units of the RNN layer is: (the number of input variables + 1) * (the number of output variables * K) / 2, the activation function of the RNN layer is a ReLU function, and the number of units of the second full connection layer is: the number of output variables * K, and the activation function of the second full connection layer is a Tanh function.

[0027] Optionally, the step of training the artificial neural network model by using the effective training data set comprises the following steps: the initial value of the training parameter K is 2, when the artificial neural network model does not meet the convergence standard, the value of K is gradually increased, the value of K is increased by 2 each time, until the artificial neural network model meets the convergence standard or reaches the set training end condition.

[0028] Optionally, the input variable comprises compressor speed, evaporator inlet air mass flow and condenser inlet air mass flow, and the output variable comprises passenger cabin temperature.

[0029] The application further provides a storage medium storing one or more computer readable programs, when the computer readable programs are called and executed by one or more controllers, the steps of the vehicle air conditioning system simulation model reduction method described above can be realized.

[0030] The application provides a vehicle air conditioning system simulation model reduction method, first, the input and output variables of the model are determined, and the value range and observation range of the input and output variables are normalized into the interval of [-1, 1]. Secondly, the binning interval of the input variable value range is determined, the initial training working condition set is constructed according to the binning interval and the binning range, and the complete simulation duration of a single working condition is determined. Then, the air conditioning system simulation model is used to simulate the initial training working condition set, the effective training working condition is selected from the simulation result, the effective training data is obtained, and the effective training data set is formed. Then, the effective validation working condition set is constructed based on the effective training working condition set, the air conditioning system simulation model is used to simulate the effective validation working condition set, and the effective validation data set is obtained. Finally, a three-layer (not counting the input layer) artificial neural network model is constructed, and the effective training data set and the effective validation data set are used for iterative training, and the final reduction model is obtained. The application has the following characteristics:

[0031] The vehicle air conditioning system simulation model reduction method provided by the application can simplify the originally complex air conditioning system simulation model, construct a reduction model with higher calculation efficiency under the premise of retaining necessary simulation accuracy, and facilitate design scheme rectification rapid analysis and control strategy optimization;

[0032] The vehicle air conditioning system simulation model reduction method provided by the application can avoid the interference of transient disturbance in the initial simulation stage on the reduction of the air conditioning system simulation model, and can improve the fitting ability of the reduction model to the performance of the original air conditioning system simulation model;

[0033] The vehicle air conditioning system simulation model reduction method provided by the application can be applied to various forms of input variables, including but not limited to random noise, uniform distribution sampling and sine sweep, and has the characteristics of wide application type. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The flowchart of the vehicle air conditioning system simulation model reduction method described in some embodiments;

[0035] Figure 2 The flowchart of the vehicle air conditioning system simulation model reduction method described in a specific example;

[0036] Figure 3 The schematic diagram of an air conditioning system simulation model;

[0037] Figure 4 The average value signal of the output variable passenger cabin temperature in the first boundary value working condition simulation result;

[0038] Figure 5 The average value signal of the output variable passenger cabin temperature in the second boundary value working condition simulation result;

[0039] Figure 6 The loss function curves for the effective training dataset and the effective validation dataset when the training parameter k equals 2;

[0040] Figure 7 The loss function curves for the effective training dataset and the effective validation dataset when the training parameter k equals 4;

[0041] Figure 8 The loss function curves for the effective training dataset and the effective validation dataset when the training parameter k equals 6;

[0042] Figure 9 The loss function curves for the effective training dataset and the effective validation dataset when the training parameter k equals 8. Detailed Implementation

[0043] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0044] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0045] like Figure 1 The method for reducing the order of a vehicle air conditioning system simulation model, as shown, includes the following steps:

[0046] S100: Determine the input variables, output variables, value range of the input variables, and scope of consideration of the output variables for the air conditioning system simulation model;

[0047] S200: Perform binning on the range of values ​​of the input variable, determine multiple initial training scenarios based on the bin set obtained from the binning operation, and construct an initial training scenario set; the binning operation includes: dividing the range of values ​​of the input variable into bins at equal intervals to obtain multiple base bins, determining all combinations of base bins consisting of consecutively arranged base bins, merging all base bin combinations to obtain multiple combination bins, and the multiple combination bins and multiple base bins of each input variable constitute the bin set of that input variable, where both combination bins and base bins are bins in the bin set;

[0048] S300: determining a first boundary condition and a second boundary condition in the initial training condition set, the input variable of the first boundary condition taking the lower limit of the base box, the input variable of the second boundary condition taking the upper limit of the base box, simulating based on the first boundary condition and the second boundary condition respectively by using the air conditioning system simulation model, and determining the average value of the simulation time length of the first boundary condition and the simulation time length of the second boundary condition as the effective simulation time length T0;

[0049] S400: determining the transition simulation time length T1 of the first boundary condition according to the simulation result of the first boundary condition, determining the transition simulation time length T2 of the second boundary condition according to the simulation result of the second boundary condition, and determining the larger value T max compared with T0, when T max is less than T0, determining the complete simulation time length as T0+T max and performing the subsequent steps, otherwise, returning to the previous step to increase the effective simulation time length by T0;

[0050] S500: simulating based on the initial training condition set respectively by using the air conditioning system simulation model, and intercepting the simulation data after 0 to T max time period in the simulation result of each of the initial training conditions as the intercepted data, determining the average value of the output variable in the initial training conditions based on the plurality of intercepted data respectively, and taking the initial training condition whose average value of the output variable meets the preset condition as the effective training condition, so as to determine a plurality of effective training condition sets, and constructing an effective training data set by using the plurality of intercepted data respectively corresponding to the plurality of effective training conditions;

[0051] S600: selecting part of the effective training conditions as the verification conditions from the effective training condition set, and determining a verification condition set;

[0052] S700: simulating based on the verification condition set by using the air conditioning system simulation model, and constructing an effective verification data set by intercepting the simulation data after 0 to T max time period in the simulation result of each of the verification conditions respectively;

[0053] S800: constructing an artificial neural network model, the input variable being taken as the input of the artificial neural network model, and the output variable being taken as the output of the artificial neural network model;

[0054] S900: training the artificial neural network model by using the effective training data set, verifying the artificial neural network model by using the effective verification data set, and obtaining a reduced order model.

[0055] By adopting the technical solution, the value range of the input variable is reasonably binned, the meaningful effective training set is focused, the effective effective training data set and the effective verification data set are screened out, the computationally efficient artificial neural network model is used, the key behavior of the original air conditioning system simulation model is kept in high-precision fitting, the originally complex air conditioning system simulation model is simplified, the necessary simulation accuracy is reserved, and a higher-order model with higher computational efficiency is constructed, and the simulation speed is significantly improved. This makes it possible to use the reduced-order model for rapid analysis in the design scheme rectification rapid analysis and control strategy optimization scenes that require a large number of simulation iterations.

[0056] Moreover, the vehicle air conditioning system simulation model reduction method determines the transient transition time T max , and discards the data in the [0, T max ] time period in the generation of subsequent training and verification data, effectively filtering out the strong transient disturbance in the initial simulation stage. By adopting the above-mentioned vehicle air conditioning system simulation model reduction method, the disturbance of the initial simulation stage transient disturbance to the air conditioning system simulation model reduction can be avoided, the influence of the non-stationary random process on the parameter identification of the reduced-order model can be avoided, the fitting ability of the reduced-order model to the performance of the original air conditioning system simulation model can be more accurately controlled, and the efficiency and accuracy of the reduced-order model training can be improved.

[0057] In some embodiments, the input variable has multiple, and the box set obtained by the binning operation determines multiple initial training working conditions including the following steps: Cartesian product combination of the box set of multiple input variables is performed to obtain multiple box combinations, and multiple initial training working conditions corresponding to multiple box combinations are determined. By adopting the technical solution, the systematization of the initial training working condition set can be ensured, and the performance boundary and key coupling effect of the vehicle air conditioning system can be more fully investigated.

[0058] In some embodiments, after the determination of the plurality of initial training working conditions each corresponding to one of the plurality of bin combinations, the method further comprises the step of: classifying the plurality of initial training working conditions according to the range size characteristics of each bin in the plurality of bin combinations, to determine a plurality of working condition categories; and the step of selecting part of the effective training working conditions from the effective training working condition set as the validation working conditions comprises the step of: randomly selecting one of the effective training working conditions from each of the working condition categories as the validation working condition. With the above technical solution, the plurality of initial training working conditions in the initial training working condition set are classified according to the range size characteristics of the bin combinations, and a stratified sampling strategy is adopted, i.e., one of the effective training working conditions in each working condition category is randomly selected to form the effective validation working condition set, which can improve the representativeness of the effective validation working condition set and the reliability of the model evaluation, and ensure that the effective validation working condition set can more evenly cover each type of bin combination, thereby reducing the possibility of evaluation bias.

[0059] In a specific implementation, the range size characteristics of each bin in the bin combination include the range of each input variable in the bin combination. If the range of the base bin is considered as 1, then the range of the combination bin is a multiple of the range of the base bin. For example, the range of the combination bin formed by merging two base bins is 2, the range of the combination bin formed by merging three base bins is 3, and the range of the combination bin formed by merging Q base bins is Q.

[0060] As a specific example, the air conditioning system simulation model has two input variables, i.e., a first input variable and a second input variable, and the ranges of the first input variable and the second input variable are [In min1 , In max1 ] and [In min2 , In max2 ], respectively. The bin interval of the first input variable and the second input variable is determined as bin1 and bin2, respectively. Then, [In min1 , In max1 ] is equally divided into [In min1 , In min1 +bin1], [In min1 , In min1 +bin1*2], …, [In min1 +floor((In max1 -In min1 ) / bin1)*bin1, In max1 ], a total of floor((In max1 -In min1 ), bin1)+1 base bins, denoted as N1, where floor() is a floor function. Similarly, the second input variable has floor((In max2 -Inmin2 There are 1 basic bins, denoted as N2. The range of values ​​for each bin in the bin combination of two input variables includes the range of values ​​n1 of the first input variable's bin and the range of values ​​n2 of the second input variable's bin. Therefore, multiple initial training conditions divided into A categories can be constructed, as shown in Table 1.

[0061] Table 1 Initial Training Scenario Set (Number of Input Variables = 2)

[0062] Working condition category value range n1 Value range n2 Working condition number First working condition category 1 1 [N1*N2] Second working condition category 1 2 [N1*(N2-1)] Third working condition category 2 2 (N1-1)*(N2-1) Fourth working condition category 1 3 [N1*(N2-2)] … … … … A working condition category [N1] [N2] 1*1

[0063] When the number of input variables is greater than 2, the same principle applies.

[0064] Furthermore, since the value range n1 of the bin for the first input variable in the first working condition category is 1, and the value range n2 of the bin for the second input variable is 1, both the first and second boundary value working conditions fall within the first working condition category. In the first boundary value working condition, each bin in its bin combination is a base bin containing the lower limit of the input variable; in the second boundary value working condition, each bin in its bin combination is a base bin containing the upper limit of the input variable.

[0065] In some embodiments, the method for reducing the order of a vehicle air conditioning system simulation model further includes the following steps: normalizing the value range of the input variable to a first preset interval as the input variable of the artificial neural network model; normalizing the examination range of the output variable to a second preset interval as the output variable of the artificial neural network model; the preset condition is that the average value of the output variable, after normalization, is within the second preset interval. In specific implementation, the first preset interval is [-1, 1], and the second preset interval is [-1, 1]. By normalizing the value range of the input variable and the examination range of the output variable, the training stability, accuracy reliability, and deployment convenience of the reduced-order model can be ensured.

[0066] As a specific example, the input variable of the air conditioning system simulation model is In. 仿真 Its value range is [In min In max ], then the input variable In of the reduced-order model 降阶 Through the formula: In 降阶 =2 / (In) max -In min )*In 仿真 +1-In max *2 / (In max -In min ) was calculated.

[0067] As a specific example, the determining the transition simulation time length T1 of the first boundary operating condition according to the simulation result of the first boundary operating condition comprises the following steps: calculating a first average value signal of the output variable in the simulation result of the first boundary operating condition, the value of the first average value signal at the end time is M1, the last time when the value of the first average value signal is equal to 0.95*M1 is T 11 , the last time when the value of the first average value signal is equal to 1.05*M1 is T 12 , T1 = Max(T 11 , T 12 ); the determining the transition simulation time length T2 of the second boundary operating condition according to the simulation result of the second boundary operating condition comprises the following steps: calculating a second average value signal of the output variable in the simulation result of the second boundary operating condition, the value of the second average value signal at the end time is M2, the last time when the value of the second average value signal is equal to 0.95*M2 is T 21 , the last time when the value of the second average value signal is equal to 1.05*M2 is T 22 , T2 = Max(T 21 , T 22 ).

[0068] In some embodiments, when the simulation is performed based on the first boundary operating condition and the second boundary operating condition respectively, the input variable of the air conditioning system simulation model is a uniformly distributed sampling signal, and the upper and lower limits of the uniformly distributed sampling signal are determined by the basic box of the input variable in the first boundary operating condition and the basic box of the input variable in the second boundary operating condition.

[0069] In some embodiments, the artificial neural network model comprises an input layer, a first full connection layer, an RNN layer and a second full connection layer connected in sequence; the number of units of the first full connection layer is: the number of input variables * the number of output variables * K, K is a training parameter; the number of units of the RNN layer is: (the number of input variables + 1) * (the number of output variables * K) / 2, and the activation function of the RNN layer is a ReLU function; the number of units of the second full connection layer is: the number of output variables * K, and the activation function of the second full connection layer is a Tanh function.

[0070] Specifically, the expression of the ReLU function is: ReLU(z) = max(0, z); and the expression of the Tanh function is:

[0071] In some embodiments, the training of the artificial neural network model using the effective training data set comprises the following steps: the initial value of the parameter K is 2, when the artificial neural network model does not meet the convergence criteria, the value of K is gradually increased, the value of K is increased by 2 each time, until the artificial neural network model meets the convergence criteria or reaches the set training end condition. This training method minimizes the trial and error cost and maximizes the fault diagnosability, ensuring that the reduced-order model has lightweight and robustness under the premise of meeting the accuracy requirement.

[0072] In specific implementation, the convergence criteria is that the fidelity index of the effective training data set and the effective validation data set is greater than or equal to 80%. The set training end condition can be that K reaches a set maximum value.

[0073] As a specific example, the input variables include compressor speed, evaporator inlet air mass flow, and condenser inlet air mass flow, and the output variables include passenger cabin temperature.

[0074] The application also provides a storage medium storing one or more computer readable programs, which, when invoked and executed by one or more controllers, can implement the steps of the vehicle air conditioning system simulation model reduction method described above.

[0075] The vehicle air conditioning system simulation model reduction method proposed in the application is described in more detail below with a specific example.

[0076] Figure 3 The simulation model of the air conditioning system of a certain vehicle built in the multi-physical field simulation software Amesim is reduced, and the convergence criteria of the reduced-order model is: the fidelity index FI (Fidelity Index) is used as a statistic, as shown in equations (1) and (2), wherein cc is the correlation coefficient, N is the sample size, y is the output value of the air conditioning system simulation model, y p The fidelity index FI of the effective training data set and the effective validation data set is required to be greater than or equal to 80%.

[0077] FI[%]=|100×(1-NRMSE)×cc| (1)

[0078]

[0079] As Figure 2 shown, in this specific example, the vehicle air conditioning system simulation model reduction method comprises the following steps:

[0080] Step one: determine the three input variables of the air conditioning system simulation model as the compressor speed, the condenser inlet air mass flow rate and the evaporator inlet air mass flow rate, take the three input variables as the input variables of the reduced order model; determine the output variable of the air conditioning system simulation model as the passenger cabin temperature, take the output variable as the output variable of the reduced order model.

[0081] Step two: determine the value range In 仿真1 of the compressor speed of the air conditioning system simulation model as [1000, 4000], the value range In 仿真2 of the condenser inlet air mass flow rate as [0.4, 1.6], and the value range In 仿真3 of the evaporator inlet air mass flow rate as [0.4, 1], convert In 仿真1 , In 仿真2 and In 仿真3 based on formulas (3) to (5) respectively, and normalize them into the interval [-1, 1], take the normalized In 降价1 , In 降价2 and In 降价3 as the value range of the input variables of the reduced order model;

[0082] In 降阶1 = 2 / (4000-1000)*In 仿真1 +1-4000*2 / (4000-1000) (3)

[0083] In 降阶2 = 2 / (1.6-0.4)*In 仿真2 +1-1.6*2 / (1.6-0.4) (4)

[0084] In 降阶3 = 2 / (1-0.4)*In 仿真3 +1-1*2 / (1-0.4) (5)

[0085] Determine the investigation range Out 仿真 of the passenger cabin temperature as [10, 40], convert Out 仿真 based on formula (6) to Out 降阶 , normalize it into the interval [-1, 1], and take the normalized Out 降阶 as the investigation range of the output variable of the reduced order model.

[0086] Out 降阶 = 2 / (40-10)*Out 仿真 +1-40*2 / (40-10) (6)

[0087] Step three: the bin interval of the compressor speed value range is 1000, the bin interval of the condenser inlet air mass flow value range is 0.4, and the bin interval of the evaporator inlet air mass flow value range is 0.2. Then the compressor speed value range is divided into [1000, 2000], [2000, 3000], [3000, 4000] for a total of 3 basic bins, that is, the first input variable basic bin number N1 = 3. The condenser inlet air mass flow value range is divided into [0.4, 0.8], [0.8, 1.2], [1.2, 1.6] for a total of 3 basic bins, that is, the second input variable basic bin number N2 = 3. The evaporator inlet air mass flow value range is divided into [0.4, 0.6], [0.6, 0.8], [0.8, 1] for a total of 3 basic bins, that is, the third input variable basic bin number N3 = 3. Thus, the initial training working condition set shown in Table 2 can be constructed. In the first working condition category, the compressor speed [1000, 2000], the condenser inlet air mass flow [0.4, 0.8], and the evaporator inlet air mass flow [0.4, 0.6] are defined as the first edge value working condition; the compressor speed [3000, 4000], the condenser inlet air mass flow [1.2, 1.6], and the evaporator inlet air mass flow [0.8, 1] are defined as the second edge value working condition. n1 is the value range of the first input variable bin, n2 is the value range of the second input variable bin, and n3 is the value range of the third input variable bin.

[0088] Table 2: Initial training working condition set

[0089]

[0090]

[0091] Step four: the first edge value working condition and the second edge value working condition are simulated by using the air conditioning system simulation model, and the sum of the simulation time lengths of the first edge value working condition and the second edge value working condition is 3600s. The input variables of the air conditioning system simulation model are uniformly distributed sampling signals, and the upper and lower limits of the uniform distribution are determined by the bins of the first edge value working condition and the second edge value working condition, as shown in Table 3. The effective simulation time T0 of a single working condition is determined to be 1800s.

[0092] Table 3: Upper and lower limits of uniform distribution

[0093]

[0094] Step five: the average value signals of the output variable passenger cabin temperature in the first edge value working condition and the second edge value working condition are shown in Figure 4 and Figure 5 respectively. The value M1 of the signal at the end time in the first edge value working condition is 27.44, and the last time T 11T = 606 s, the last time T when the average value signal equals 1.05*M2 = 28.81 is T = 1 s, and the last time T when the average value signal equals 1.05*M2 = 28.81 is T = 606 s. 12 T1 = max(T 11 , T 12 ) = 606 s. The value M2 of the signal at the end time of the second boundary condition is 19.62, and the last time T when the signal equals 0.95*M2 = 18.64 is T = 1 s. 21 T = 606 s, the last time T when the average value signal equals 1.05*M2 = 28.81 is T = 1 s, and the last time T when the average value signal equals 1.05*M2 = 28.81 is T = 606 s. 22 T2 = max(T 21 , T 22 ) = 606 s, and finally T max = max(T1, T2) = 816 s.

[0095] Step six: T max = 816 s, T0 = 1800 s, T max <T0, continue to step seven.

[0096] Step seven: determine the complete simulation time T of a single condition T = T0 + T3 = 1800 s + 816 s = 2616 s, and simulate the air conditioning system simulation model based on the initial training set conditions determined in Table 2.

[0097] Step eight: for each initial training set condition, truncate the simulation data from time 0 to 816 s, retain the simulation data after 0 to 816 s and count the average value of the converted output variable, and eliminate the conditions whose average value exceeds the interval [-1, 1], as shown in Table 4. The remaining conditions with average values within the interval [-1, 1] are used as effective training set conditions, and the effective training data set is constructed based on the simulation data after time 0 to 816 s in the effective training set conditions.

[0098] Table 4 Effective training conditions and effective validation conditions

[0099]

[0100]

[0101] Step nine: randomly select one condition from the effective training conditions of each condition category shown in Table 4 as the effective validation condition of the condition category, and the B validation conditions of the B condition categories are summarized to form the effective validation condition set.

[0102] Step ten: simulate the effective validation condition set in Table 4 using the air conditioning system simulation model to obtain simulation data, truncate the simulation data from time 0 to T max , and retain the simulation data from 0 to T maxThe simulation data after the time period is taken as the effective validation data, and an effective validation data set is constructed.

[0103] Step eleven: an artificial neural network model is constructed based on a 3-layer (not including an input layer) artificial neural network, the first layer of the artificial neural network is a full connection layer, the number of units is 3*1*K, wherein K is a training parameter, and the activation function is a ReLU function, and the expression is as follows:

[0104] ReLU(z) = max(0, z)

[0105] The second layer of the artificial neural network is an RNN (Recurrent Neural Network) layer, the number of units is (3+1)*(1*K) / 2, and the activation function is a ReLU function; the third layer of the artificial neural network is a full connection layer, the number of units is 1*K, and the activation function is a Tanh function, and the expression is as follows.

[0106]

[0107] Step twelve: the artificial neural network model constructed in step eleven is trained based on the effective training data set and the effective validation data set, the small batch random gradient descent method is used for training, the learning rate is 0.001, the batch size is 64, the upper limit of the number of single training iterations is 10000, the value of the training parameter K starts from 2, and when K is 6, the convergence standard is met, and the training is ended. The effective training data set FI and the effective validation data set FI in the training process are shown in Table 5, and the loss function value curve of the effective training data set and the effective validation data set is shown in FIG. 2. Figure 6 to Figure 9

[0108] Table 5: effective training data set FI and effective validation data set FI

[0109] K Valid training data set FI Valid verification data set FI Is the training finished? 2 64.9％ 78.2％ No 4 82.3％ 79.5％ No 6 81.5％ 79.3％ No 8 86.1％ 80.5％ Yes

[0110] The above embodiments are only preferred embodiments for fully illustrating the present application, and the protection scope of the present application is not limited thereto. Any equivalent replacement or transformation made by those skilled in the art on the basis of the present application is within the protection scope of the present application. In the description of the specification, the description of the terms “one embodiment”, “some embodiments”, “example”, “specific example” or “some examples” means that the specific features, structures, materials or characteristics combined with the embodiment or example are contained in at least one embodiment or example of the present application. In the description of the specification, the illustrative description of the above terms is not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in the specification.​

Claims

1. A method for reducing the order of a vehicle air conditioning system simulation model, characterized in that, Includes the following steps: Determine the input variables, output variables, value range of the input variables, and scope of consideration of the output variables for the air conditioning system simulation model; The range of values ​​of the input variable is binned, and multiple initial training scenarios are determined based on the bin set obtained from the binning operation to construct an initial training scenario set. The binning operation includes: dividing the range of values ​​of the input variable into bins at equal intervals to obtain multiple basic bins, determining all combinations of basic bins consisting of consecutively arranged basic bins, and merging all basic bin combinations to obtain multiple combined bins. In the initial training case set, a first boundary condition and a second boundary condition are determined. The input variable of the first boundary condition is taken from the base box where the lower limit is located, and the input variable of the second boundary condition is taken from the base box where the upper limit is located. The air conditioning system simulation model is used to perform simulations based on the first boundary condition and the second boundary condition respectively. The average of the simulation duration of the first boundary condition and the simulation duration of the second boundary condition is determined as the effective simulation duration T0. Based on the simulation results of the first boundary condition, determine the transition simulation duration T1 for the first boundary condition; based on the simulation results of the second boundary condition, determine the transition simulation duration T2 for the second boundary condition; and select the larger value T1 from T2. max Compared with T0, when T max When the time is less than T0, the complete simulation duration is determined to be T0+T. max Then proceed with the subsequent steps; otherwise, return to the previous steps and increase the effective simulation time by T0. Using the aforementioned air conditioning system simulation model, simulations were performed based on the initial training condition set, and the simulation results from 0 to T were extracted from multiple initial training conditions. max Simulation data after a time period is used as intercepted data. The average value of the output variable in multiple initial training scenarios is determined based on multiple intercepted data. The initial training scenarios in which the average value of the output variable meets the preset conditions are taken as effective training scenarios. The effective training scenario set is determined. The effective training dataset is constructed using multiple intercepted data corresponding to multiple effective training scenarios. A subset of valid training scenarios are selected from the set of valid training scenarios to serve as validation scenarios, thus determining the set of validation scenarios. Using the aforementioned air conditioning system simulation model, simulations were performed based on the aforementioned verification condition set. Simulation results from 0 to T were extracted from the simulation results of multiple verification conditions. max Construct an effective validation dataset using simulation data from periods beyond the specified time. Construct an artificial neural network model, wherein the artificial neural network model takes the input variable as input and the output variable as output; The artificial neural network model is trained using the effective training dataset and validated using the effective validation dataset to obtain a reduced-order model.

2. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 1, characterized in that, The input variables are multiple, and the determination of multiple initial training conditions based on the bin sets obtained by the binning operation includes the following steps: combining the bin sets of the multiple input variables by Cartesian product to obtain multiple bin combinations, and determining multiple initial training conditions that correspond one-to-one with the multiple bin combinations.

3. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 2, characterized in that, After determining the multiple initial training conditions that correspond one-to-one with the multiple bin combinations, the method further includes the following steps: classifying the multiple initial training conditions according to the value range characteristics of each bin in the multiple bin combinations, and determining multiple condition categories. The step of selecting a subset of valid training conditions from the set of valid training conditions as verification conditions includes the following steps: randomly selecting one of the valid training conditions from each of the condition categories as the verification condition.

4. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 1, characterized in that, It also includes the following steps: The range of values ​​of the input variables is normalized to a first preset interval, and the range of values ​​of the output variables is normalized to a second preset interval. The preset condition is that the average value of the output variable, after normalization, is within the second preset interval.

5. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 1, characterized in that, The step of determining the transition simulation duration T1 of the first boundary condition based on the simulation results of the first boundary condition includes the following steps: calculating the first average value signal of the output variable in the simulation results of the first boundary condition, the value of the first average value signal at the termination time being M1, and the last time when the value of the first average value signal is equal to 0.95*M1 being T. 11 The last time the value of the first average signal is equal to 1.05*M1 is T. 12 T1=Max(T 11 T 12 ); The step of determining the transition simulation duration T2 of the second boundary condition based on the simulation results of the second boundary condition includes the following steps: calculating the second average value signal of the output variable in the simulation results of the second boundary condition, the value of the second average value signal at the termination time being M2, and the last time when the value of the second average value signal is equal to 0.95*M2 being T. 21 The last time the value of the second average signal is equal to 1.05*M2 is T. 22 T2=Max(T) 21 T 22 ).

6. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 1, characterized in that, When the air conditioning system simulation model is used to perform simulations based on the first boundary condition and the second boundary condition, the input variable of the air conditioning system simulation model is a uniformly distributed sampled signal. The upper and lower limits of the uniformly distributed sampled signal are determined by the base box of the input variable in the first boundary condition and the base box of the input variable in the second boundary condition.

7. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 1, characterized in that, The artificial neural network model includes a first fully connected layer, an RNN layer, and a second fully connected layer connected sequentially. The number of units in the first fully connected layer is: the number of input variables * the number of output variables * K, where K is the training parameter. The number of units in the RNN layer is: (the number of input variables + 1) * (the number of output variables * K) / 2, and the activation function of the RNN layer is the ReLU function. The number of units in the second fully connected layer is: the number of output variables * K, and the activation function of the second fully connected layer is the Tanh function.

8. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 7, characterized in that, The training of the artificial neural network model using the effective training dataset includes the following steps: the initial value of the training parameter K is 2. When the artificial neural network model does not meet the convergence criterion, the value of K is gradually increased by 2 each time until the artificial neural network model meets the convergence criterion or reaches the set training termination condition.

9. The method for reducing the order of a vehicle air conditioning system simulation model according to claim 1, characterized in that, The input variables include compressor speed, evaporator inlet air mass flow rate, and condenser inlet air mass flow rate, and the output variable includes passenger compartment temperature.

10. A storage medium, characterized in that, It stores one or more computer-readable programs, which, when executed by one or more controllers, can implement the steps of the vehicle air conditioning system simulation model reduction method as described in any one of claims 1 to 9.

Images