Method for predicting spraying characteristics of internal combustion engine

Abstract

The invention discloses a method for predicting spraying characteristics of an internal combustion engine. The method comprises the following steps: establishing an original data set comprising data of working condition parameters and data of spraying characteristic parameters; calculating the deviation between the measured value and the calculated value of the existing spray characteristic formula, and correcting the existing spray characteristic formula; constructing a physical constraint system of spraying characteristics; extrapolating and expanding by adopting a Gaussian process regression method; performing interpolation enhancement by adopting a generative adversarial network; constructing a spray prediction model through deep learning; according to the method, an existing spraying characteristic formula is corrected through the deviation value, and the data expansion accuracy is improved; the limitation of small sample experimental data is effectively solved through data enhancement, Gaussian process regression extrapolation with priori knowledge is utilized to expand and cover limit working conditions, a correction formula and physical constraints are added into a generative adversarial network, details are further complemented, data breakpoints are eliminated, and a high-quality data set is constructed. And the generalization ability, the prediction precision and the physical rationality of the deep learning model are improved.

Description

Technical Field

[0001] This invention relates to the field of internal combustion engine simulation and calculation technology, specifically to a method for predicting the spray characteristics of an internal combustion engine. Background Technology

[0002] As a key factor affecting combustion, the development process and atomization quality of spray directly affect the mixing effect of fuel and air, thereby affecting combustion efficiency and pollutant emissions. Therefore, in order to meet the requirements of speed and accuracy in combustion prediction, establishing an accurate and efficient spray characteristic prediction model is a necessary foundation for further in-depth research on the combustion process and emission reduction.

[0003] In the prior art, a method for predicting the penetration distance of diesel spray with a variable injection rate, as described in application number 202010556843.5, is based on jet mechanics theory. It establishes an analytical expression for the change of effective injection velocity with injection velocity, replacing the injection velocity with the effective injection velocity, and introduces it into a spray model suitable for constant injection rate calculations, thus obtaining a spray model suitable for variable injection rates for predicting penetration distance. Another method, as described in application number 202311080763.7, proposes an in-cylinder fuel liquid phase spray penetration distance prediction method. This method sets a brightness coefficient and a liquid phase spray penetration distance prediction model, then calculates the acceleration of the fuel droplets based on their Sauter average diameter and mass, and finally calculates the penetration distance based on the liquid phase spray penetration distance prediction model. The method and apparatus for predicting dynamic spray characteristics of a liquid-liquid coaxial centrifugal injector, with application number 202410827380.X, first establishes a three-dimensional model of the liquid-liquid coaxial centrifugal injector and performs mesh generation. Then, a conversion mechanism between the continuous phase and the discrete phase is established. Based on the mesh generation and the conversion mechanism, the flow simulation of the working medium in the injector within a preset time period is performed. During the simulation, the first pulsation is applied upstream and downstream of the injector. The prediction model is then obtained by optimizing the BP neural network and training it in combination with the simulation results, and finally, the dynamic spray characteristics prediction is realized, which significantly improves the analysis efficiency and accuracy.

[0004] Existing spray prediction methods still have the following shortcomings and limitations, including: 1. Limited data collection under limited conditions. Expanding the dataset through CFD simulation or experimental acquisition requires a lot of time and effort. The accuracy of the data generated by the expanded dataset is affected by prior and posterior formulas, but the calculation results of existing spray characteristic formulas deviate from reality, affecting the accuracy of the expanded dataset; 2. Limited variable parameters, restricting the prediction scenarios for spray characteristics. Spray characteristics are affected by the coupling of multiple variable parameters such as injector structure, environmental pressure, and environmental temperature; 3. Insufficient model generalization ability. The model is built based on specific operating condition parameters. When the change exceeds the validation range, the model prediction accuracy will decrease; 4. High computational cost. The iterative computation under rapid changes in variable parameters is large, the data regression fitting computation is time-consuming, and the pre-simulation calculation consumes computational resources, all of which become bottlenecks to improving computational efficiency and are difficult to adapt to the real-time, fast, and accurate prediction requirements of dynamic spray.

[0005] Therefore, there is an urgent need for a spray characteristic prediction method that effectively expands the dataset based on the modified spray characteristic formula and physical constraint system, coupled with the influence of multi-dimensional parameters, and establishes an accurate and fast prediction model through machine learning. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for predicting the spray characteristics of internal combustion engines, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] A method for predicting the spray characteristics of an internal combustion engine includes the following contents and steps:

[0009] S1. Obtain spray test data and establish a raw dataset including operating condition parameters and spray characteristic parameters;

[0010] S2. Introduce the existing spray characteristic formulas based on the spray characteristic parameters, calculate the deviation between the measured and calculated values ​​of the spray characteristic parameters, and revise the existing spray characteristic formulas accordingly.

[0011] ;

[0012] in, Let be the deviation of the i-th spray characteristic parameter. The existing spray characteristic formula is for the i-th spray characteristic parameter. Here is the spray characteristic correction formula for the i-th spray characteristic parameter;

[0013] S3. Based on the spray mechanism and the original dataset, construct a physical constraint system for spray characteristics. This system includes trend constraints, correlation constraints, and limit constraints. Trend constraints constrain the changing trends of spray characteristic parameters with the core parameters of the operating conditions. Correlation constraints are based on the spray characteristic correction formula. The spray characteristics of the core carrier are related to the law, and the limit constraints are the physical boundary values ​​of the spray characteristic parameters;

[0014] S4. Set the boundary parameters for the extreme operating conditions, using the spray characteristic correction formula F. i As a priori formula, the Gaussian process regression method is used to extrapolate and extend the original dataset based on the physical constraint system to obtain an extrapolated extended dataset covering all working conditions.

[0015] S5. Generative adversarial networks are used to perform interpolation enhancement on the extrapolated dataset to complete the details and eliminate breakpoints, thereby obtaining the training dataset. The prior formula and physical constraint system are then embedded into the generator's total loss function.

[0016] S6. Input the data from the training dataset into the deep learning network model to learn the changing patterns of spray characteristics and the relationship between operating parameters and spray characteristic parameters. Embed the physical constraint system to train and validate the spray prediction model.

[0017] In one exemplary embodiment of this disclosure, the operating parameters include environmental pressure. Ambient temperature Injection pressure and injector orifice diameter Spray characteristic parameters include spray penetration distance. Spray cone angle and spray particle size .

[0018] In one exemplary embodiment of this disclosure, the deviation of the spray characteristics is fitted as a linear relationship related to the operating parameters:

[0019] ;

[0020] in, , , , ... represent the influence coefficients of various operating condition parameters. denoted as the deviation of the i-th spray characteristic parameter.

[0021] In an exemplary embodiment of this disclosure, the spray is divided into a near-orifice region, a breakup region, and a far-orifice region. In step S3, different main core control and physical constraint equations are configured according to the dominant physical processes of the three regions. The main core control and physical constraint equations of the near-orifice region include the multiphase flow volume fraction transport equation and the momentum conservation equation. The main core control and physical constraint equations of the breakup region include the droplet breakup model and the gas-liquid two-phase momentum exchange equation. The main core control and physical constraint equations of the far-orifice region include the gas phase component transport equation, the drag force equation, and the droplet evaporation model.

[0022] In one exemplary embodiment of this disclosure, step S4 transforms the partitioned physical constraint losses into residual terms, incorporates them into the negative log-likelihood function of Gaussian process regression, and establishes a final loss function with a unified optimization objective. :

[0023] ;

[0024] in, For likelihood loss, The physical constraint global equilibrium coefficient, , and These are the residuals for the near-orifice region, the fragmentation region, and the far-orifice region, respectively. , , These are the contribution weighting coefficients for the near-orifice region, the fragmentation region, and the far-orifice region, respectively.

[0025] In one exemplary embodiment of this disclosure, the contribution weight coefficient is derived from the input operating parameters to determine the contribution of the three regions, and the partition constraint strength is adaptively adjusted.

[0026] In one exemplary embodiment of this disclosure, the loss function of the generator in step S5 for:

[0027] ;

[0028] in, To combat the losses, For data loss, To generate the loss of the deviation between the data and the mapping values ​​of the prior formula, The total loss due to the physical constraints of the partition. , and These are the weighting coefficients for each type of loss.

[0029] In one exemplary embodiment of this disclosure, the total loss of the partitioned physical constraints is obtained by weighted summation of the residuals of each region:

[0030] ;

[0031] in, , and These are the regional weighting coefficients for the near-orifice region, the breakup region, and the far-orifice region, respectively, adjusted according to the degree of influence of different regions on spray performance.

[0032] Compared with the prior art, the internal combustion engine spray characteristic prediction method of the present invention has the following beneficial effects:

[0033] This method corrects existing physical constraints by adjusting the deviation amount, thereby improving the accuracy of data expansion. It effectively addresses the limitations of small sample experimental data through data augmentation, expands coverage of extreme working conditions by using Gaussian process regression extrapolation with prior knowledge, and incorporates the correction formula and physical constraints into the generative adversarial network to further complete the details, eliminate data breakpoints, construct a high-quality dataset, and improve the generalization ability, prediction accuracy, and physical rationality of the deep learning model.

[0034] A refined partitioned physical constraint system is constructed, taking into account trend, correlation and limit constraints, avoiding the distortion problem of global constraints, improving the physical rationality of data and models, and reducing computational complexity. At the same time, the deviation is fitted into a linear relationship related to the operating parameters, which is suitable for multi-structure design of fuel injectors and application scenarios with variable operating temperature or pressure. Attached Figure Description

[0035] Figure 1 This is a flowchart of the spray characteristic prediction method disclosed in this invention;

[0036] Figure 2 Generate a flowchart for the training dataset. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely the best embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] The term "embodiment" as used herein means that a particular method, step, or content described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0039] This embodiment provides a method for predicting the spray characteristics of internal combustion engines, mainly addressing the problems of limited experimental sample data, low data quality, inability to cover extreme operating conditions, and poor generalization ability of deep learning. It includes the following content and steps:

[0040] S1. Based on the raw spray test data obtained from the experiment, the raw spray data is preprocessed, and data related to spray characteristics are extracted to establish a raw dataset containing operating condition parameter data and spray characteristic parameter data. The operating condition parameters include, but are not limited to, ambient pressure P, ambient temperature T, and injection pressure T. in Injector orifice diameter D, injector length-to-diameter ratio Q, number of injector orifices N; spray characteristic parameters, including but not limited to spray penetration distance. Spray cone angle and spray particle size ;

[0041] S2. Considering the common calculation deviations in existing spray characteristic formulas, such as the deviation between calculated and actual values ​​caused by the inability to cover extreme working conditions, the deviation between calculated and actual values ​​caused by measurement errors, and the deviation between calculated and actual values ​​caused by temperature differences between experimental environments and real-world scenarios, the calculation deviations are statistically analyzed and the existing spray characteristic formulas are corrected.

[0042] Based on classical spray theory or simulation empirical formulas, existing spray characteristic formulas corresponding to spray characteristic parameters are defined. Operating condition parameter data is input to determine the calculated values ​​of the spray characteristic parameters. Actual measured values ​​of the spray characteristic parameters under the corresponding operating conditions are collected through experimental methods, and the deviation under single-target spray characteristic parameters is calculated. ;

[0043] Deviation The data is fitted to a linear relationship related to operating parameters to adapt to the multi-structure design of fuel injectors and application scenarios with variable operating temperature or pressure, thereby improving the generalization ability of deep learning.

[0044] ;

[0045] in, , , , ... in order are ambient pressure P, ambient temperature T, and injection pressure T. in The influence coefficients of operating parameters such as injector orifice diameter D, etc., are fitted and regressed to constant values. In this embodiment, the deviation of the i-th spray characteristic parameter is... This is the deviation in spray penetration distance. This is the spray cone angle deviation. This refers to the deviation in spray particle size.

[0046] According to the deviation The existing spray characteristic formula is modified to obtain the spray characteristic correction formula:

[0047] ;

[0048] in, The existing spray characteristic formula is for the i-th spray characteristic parameter; In this embodiment, the spray characteristic correction formula for the i-th spray characteristic parameter is as follows: This is the corrected formula for spray penetration distance. This is the corrected formula for the spray cone angle. This is the correction formula for spray particle size;

[0049] S3. Based on the relevant mechanisms of spraying and data analysis of the original dataset, according to the flow field structure, energy transfer law and phase transition characteristics of spraying in the spatial propagation process, a physical constraint system for constraining the characteristics and development of spraying is constructed. The physical constraint system includes trend constraints, correlation constraints and limit constraints.

[0050] Among them, trend constraints: constrain the changing trend of spray characteristic parameters with the core parameters of the operating conditions, such as constraining the changing trend of the spray cone angle decreasing with the increase of spray pressure; limit constraints: constrain the physical boundary values ​​of relevant spray characteristic parameters to avoid generating invalid data, such as constraining the droplet diameter range.

[0051] Correlation constraints: Correction formula based on spray characteristics Physical constraints reflecting the inherent correlation laws of spray characteristics are established for the core carrier. These physical constraints include, but are not limited to, transport equations, momentum conservation equations, droplet breakup models, gas-liquid two-phase momentum exchange equations, drag force equations, and droplet evaporation models. Considering the significant differences in the dominant physical processes and dominant chemical laws of spray in different regions, the spray is divided into three continuous regions with significant differences in physical characteristics: the near-orifice region, the breakup region, and the far-orifice region. Core control and physical constraint equations are then configured accordingly.

[0052] In the near-nozzle region, the spray jet has just left the injector outlet and presents a continuous liquid column shape. The multiphase flow volume fraction transport equation and momentum conservation equation dominate in this region.

[0053] In the breakup zone, droplets break up and form a high-concentration droplet swarm, and the momentum exchange between the gas and liquid phases is significant. This region is dominated by droplet breakup models and gas-liquid two-phase momentum exchange equations.

[0054] In the far-hole region, the droplet breakup process is basically completed at this stage. The main focus is on the convection, diffusion, and evaporation of droplets in the airflow. Therefore, this region is mainly based on the gas phase component transport equation, drag force equation, and droplet evaporation model.

[0055] By employing the refined partitioning constraint strategy that aligns with the evolution of spray patterns, each region only needs to be constrained or have its weight allocation set to primarily constrain its dominant mechanism. This avoids the constraint distortion caused by uniform constraints across the entire region and cross-regional physical contradictions, making the constraints more precise. By configuring on demand, it adapts to the differentiated optimization needs of different regions and specifically avoids wasting computing resources in unimportant regions, thereby improving computing efficiency.

[0056] S4, such as Figure 2 As shown, the boundary parameters of the extreme working conditions are set, and the Gaussian process regression method is used to perform the first stage of extrapolation expansion on the original dataset to complete the working condition samples; in order to avoid generating invalid samples, the physical constraint system constructed in step S3 is added to the expansion process:

[0057] First, using the spray characteristic correction formula F i As a prior formula, since the existing small sample experimental data is difficult to cover all working conditions, especially extreme working conditions, in order to reduce the dependence of the prediction model on the existing data, the prior formula is used as the mean function to cover extreme working conditions and improve the rationality of extrapolation. Optionally, a hyperparameter correction term can be added to the prior formula for further accurate fine-tuning.

[0058] Based on existing experimental data, relying on prior formulas or finely tuned prior formulas, spray characteristic parameters under extreme working conditions not covered by existing data are derived through Gaussian process regression, and extrapolated datasets are obtained.

[0059] During the expansion process, physical constraint extrapolation results are used to avoid generating physically invalid spray data. A partitioned physical constraint system is used as the posterior, and the core control and physical constraint losses are transformed into residual terms, which are then incorporated into the negative log-likelihood function of Gaussian process regression. This constructs a loss function that integrates data fitting error and physical constraint error, achieving a dual optimization of data fit and physical rationality.

[0060] ;

[0061] in, Likelihood loss measures the degree of fit between the Gaussian process regression prediction results and the experimental data. The physical constraint global equilibrium coefficient;

[0062] To calculate the residuals in the near-orifice region using the multiphase flow volume fraction transport equation and momentum conservation equation as the main governing equations, To calculate the residuals within the fragmentation region, primarily based on the droplet fragmentation model and the gas-liquid two-phase momentum exchange equations, To calculate the residuals in the region of the far-spray nozzle, which are mainly based on the droplet breakup model and the gas-liquid two-phase momentum exchange equation;

[0063] , , The contribution weight coefficients for the near-orifice region, the breaking region, and the far-orifice region are respectively. Each weight coefficient is derived from the input working condition parameters to make adaptive adjustments to the contribution of the three regions, so as to achieve adaptive adjustment of the zoning constraint strength of the working condition. For example, the weight coefficient of the near-orifice region increases with the increase of the injection pressure, and the weight coefficient of the far-orifice region increases with the increase of the ambient temperature.

[0064] S5. Further data augmentation is performed on the extrapolated dataset. By interpolating and completing the detailed data, potential discontinuities in the dataset are eliminated, ensuring the continuity of the data distribution and providing a high-quality training dataset for deep learning. Generative adversarial networks are used to implement the second stage of data augmentation, embedding the prior formula and physical constraint system into the generator's total loss function, as detailed below:

[0065] A generator loss function is constructed that incorporates both physical constraint loss and empirical formula loss to ensure that the generated data closely matches the baseline laws of prior knowledge and adheres to the physical mechanism of spraying.

[0066] ;

[0067] in, For the total loss, To combat the loss, which stems from the game between the generator and the discriminator, the deviation of the generator output data from the experimental data in terms of statistical characteristics is calculated. The data loss is the numerical deviation between the quantization generator output data and the experimental data; To account for the loss due to the deviation between the generated data and the mapping values ​​of the prior formula; The total loss due to the physical constraints of the partitions is obtained by weighted summation of the residuals of each spray region: ; , and These are the regional weighting coefficients for the near-orifice region, the fragmentation region, and the far-orifice region, respectively. These can be adjusted based on the degree of influence of different regions on spray performance. For example, the fragmentation region has the greatest impact on spray atomization quality, so the spray particle size parameter should be appropriately increased when generating the spray particle size parameter. The weight value;

[0068] , and The weighting coefficients for each loss should be set appropriately based on the data sample and prediction requirements; they should be increased when the data sample is sufficient. Conversely, appropriately increase and This ensures that the generated data conforms to physical laws; alternatively, iterative optimization methods can be used to adjust the weight combination to the optimal level.

[0069] S6. Input the training dataset obtained in step S5 into the deep learning network model. The deep learning network model adopts a deep neural network (DNN) architecture. Its network structure is designed as follows: the input layer is the spray condition parameters, the hidden layer extracts the deep features of the condition parameters and the coupling relationship between features, and the output layer is the spray characteristic parameter data at different times. The deep learning network model extracts and mines the potential correlation features between the condition parameters and the spray characteristics, learns the changing law of the spray characteristics and the potential correlation law between the condition parameters and the spray characteristic parameters, and constructs a spray characteristic prediction model.

[0070] During model training, a physical constraint term is added to the loss function to ensure that the model adheres to the core physical laws of spraying. At the same time, a two-dimensional verification mechanism is established to validate the spray prediction model and further ensure the physical rationality of its prediction results. The two-dimensional verification mechanism includes: 1. Verifying the data accuracy of the model's prediction results through accuracy indicators; 2. Verifying whether the prediction results follow physical rules based on the spray physical constraint system. Accordingly, the spray characteristic prediction model is trained in a closed loop until the model meets the accuracy and physical constraint requirements.

[0071] Through the above description of the embodiments, those skilled in the art can clearly understand that the various embodiments of this application can be implemented by means of software or software combined with necessary general-purpose hardware platforms, and of course, they can also be implemented by hardware functions. Based on this understanding, the technical solution of this application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions to cause a computer device, such as including but not limited to a personal computer, server, or network device, to execute all or part of the steps of the method described in any embodiment of this application.

[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the spray characteristics of an internal combustion engine, characterized in that, Includes the following steps and content: S1. Obtain spray test data and establish a raw dataset including operating condition parameters and spray characteristic parameters; S2. Introduce the existing spray characteristic formula for the spray characteristic parameters, calculate the deviation between the measured and calculated values ​​of the spray characteristic parameters, and correct the existing spray characteristic formula: ； in, The deviation amount of the spray characteristic parameter described in the i-th item. The existing spray characteristic formula for the spray characteristic parameter mentioned in the i-th item, The spray characteristic correction formula is for the spray characteristic parameters mentioned in the i-th item; S3. Based on the spray mechanism and the original dataset, construct a physical constraint system for the spray characteristics. This system includes trend constraints, correlation constraints, and limit constraints. The trend constraints constrain the changing trend of the spray characteristic parameters with respect to the core parameters of the operating conditions. The correlation constraints are based on the spray characteristic correction formula... Establish core control and physical constraint equations for the core carrier, where the limit constraints are the physical boundary values ​​of the spray characteristic parameters; S4. Set the boundary parameters for the extreme operating conditions, using the spray characteristic correction formula F. i As a priori formula, the Gaussian process regression method is used to extrapolate and extend the original dataset based on the physical constraint system to obtain an extrapolated and extended dataset covering all working conditions. S5. Using a generative adversarial network, the extrapolated augmented dataset is interpolated and enhanced to complete the details and eliminate breakpoints, thereby obtaining a training dataset. The prior formula and the physical constraint system are then embedded into the generator's total loss function. S6. Input the data from the training dataset into the deep learning network model to learn the variation law of spray characteristics and the relationship between the operating condition parameters and the spray characteristic parameters, embed the physical constraint system, and train and verify the spray prediction model.

2. The method for predicting the spray characteristics of an internal combustion engine according to claim 1, characterized in that: The operating parameters include ambient pressure. Ambient temperature Injection pressure and injector orifice diameter The spray characteristic parameters include spray penetration distance. Spray cone angle and spray particle size .

3. The method for predicting the spray characteristics of an internal combustion engine according to claim 2, characterized in that: The deviation is fitted as a linear relationship related to the operating condition parameters: ； in, , , , ... are the influence coefficients of each of the aforementioned operating condition parameters. The deviation of the spray characteristic parameter described in the i-th item.

4. The method for predicting the spray characteristics of an internal combustion engine according to any one of claims 1 to 3, characterized in that: The spray is divided into a near-orifice region, a breakup region, and a far-orifice region. In step S3, different main core control and physical constraint equations are configured according to the dominant physical processes of the three regions. The main core control and physical constraint equations of the near-orifice region include the multiphase flow volume fraction transport equation and the momentum conservation equation. The main core control and physical constraint equations of the breakup region include the droplet breakup model and the gas-liquid two-phase momentum exchange equation. The main core control and physical constraint equations of the far-orifice region include the gas phase component transport equation, the drag force equation, and the droplet evaporation model.

5. The method for predicting the spray characteristics of an internal combustion engine according to claim 4, characterized in that: In step S4, the partitioned physical constraint losses are converted into residual terms and incorporated into the negative log-likelihood function of Gaussian process regression to establish the final loss function with a unified optimization objective. : ； in, For likelihood loss, The physical constraint global equilibrium coefficient, , and These are the residuals of the near-orifice region, the broken region, and the far-orifice region, respectively. , , These are the contribution weighting coefficients for the near-orifice region, the fragmentation region, and the far-orifice region, respectively.

6. The method for predicting the spray characteristics of an internal combustion engine according to claim 5, characterized in that: The contribution weight coefficient is derived from the input operating parameters to determine the contribution of the three regions, and the partition constraint strength is adaptively adjusted.

7. The method for predicting the spray characteristics of an internal combustion engine according to claim 5, characterized in that: The total loss function of the generator in step S5 for: ； in, To combat the losses, For data loss, To generate the deviation loss between the data and the mapping value of the prior formula, The total loss due to the physical constraints of the partition. , and These are the weighting coefficients for each type of loss.

8. The method for predicting the spray characteristics of an internal combustion engine according to claim 7, characterized in that: The total loss of the partitioned physical constraints is obtained by weighted summation of the residuals of each region: ； in, , and The regional weighting coefficients for the near-spray hole region, the breaking region, and the far-spray hole region are adjusted according to the degree of influence of different regions on the spray performance parameters.

Images