A fuel atomization particle size distribution prediction method, device, equipment and medium

Abstract

This invention provides a method, apparatus, device, and medium for predicting fuel atomization particle size distribution. It relates to the field of image processing technology. The method includes: acquiring a sequence of atomization images during the lateral jet process of fuel under different atomization conditions; extracting droplet size data from each atomization image and fitting it to a log-normal distribution to obtain the logarithmic mean and logarithmic standard deviation characterizing each atomization image; training a neural network using atomization conditions and time series as inputs and distribution parameters as outputs, through a loss function including operating condition constraint loss and temporal constraint loss; the operating condition constraint loss is used to constrain the distribution parameters of similar operating conditions at the same time to have no significant deviation; the temporal constraint loss is used to constrain the distribution parameters of continuous time under the same operating condition to have no abrupt changes; inputting the target atomization conditions and time into the trained neural network, outputting predicted distribution parameters, and reconstructing the fuel atomization particle size distribution at the target time based on the predicted distribution parameters.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, apparatus, device, and medium for predicting the particle size distribution of fuel atomization. Background Technology

[0002] Optimizing combustion efficiency and precisely controlling pollutant emissions from fuel transverse jet atomization has become a core research direction and key technological challenge in power equipment fields such as aerospace and automotive engines. Particle size distribution, as a crucial quantitative evaluation indicator for measuring the effectiveness of fuel transverse jet atomization, directly determines the decision-making direction and implementation effect of atomization process optimization through accurate characterization and prediction. This is of great significance for improving the combustion efficiency of power equipment and reducing pollutant emissions.

[0003] Existing fuel atomization particle size analysis and prediction technologies are still limited to the traditional scope of static or steady-state prediction. First, atomization images or droplet data are collected through various detection methods, and then the data is processed manually or by simple algorithms to obtain full discrete droplet particle size data. Then, this type of high-dimensional and massive full discrete droplet particle size data is used as the prediction target for model construction and training.

[0004] This type of technical solution not only requires processing massive amounts of discrete data, resulting in high model training complexity, but also fails to consider the time dynamic evolution characteristics of the atomization process, making it difficult to adapt to the dynamic changes in different atomization conditions and different atomization times. It cannot achieve dynamic and accurate prediction of the atomization particle size distribution of the fuel transverse jet, and is no longer able to meet the high precision and high efficiency requirements of atomization process optimization in aerospace, automotive engine and other fields. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, equipment, and medium for predicting the particle size distribution of fuel atomization in response to the above-mentioned technical problems.

[0006] The following technical solution is adopted in this specification: This specification provides a method for predicting fuel atomization particle size distribution, including: Under different atomization conditions, a sequence of atomization images during the lateral jet of fuel is acquired according to a preset time series. Each atomization image in the sequence corresponds to a time point in the time series. Droplet size data are extracted from each atomized image, and the droplet size data are fitted with a log-normal distribution to obtain distribution parameters characterizing the particle size morphology distribution in each atomized image. The distribution parameters include the logarithmic mean and the logarithmic standard deviation. A neural network prediction model is constructed, taking atomization condition parameters and time series as inputs and distribution parameters as outputs. The neural network prediction model is trained using a loss function that includes condition constraint loss and time series constraint loss. The condition constraint loss is used to constrain the distribution parameters of similar conditions under the same time to have no significant deviation; the time series constraint loss is used to constrain the distribution parameters of continuous time under the same condition to have no abrupt changes. The target atomization condition parameters and target time are input into the trained neural network prediction model, which outputs the predicted distribution parameters. Based on the predicted distribution parameters, the fuel atomization particle size distribution at the target time is reconstructed.

[0007] Furthermore, the step of training the neural network prediction model using a loss function that includes both operational condition constraint loss and temporal constraint loss specifically includes: Construct the loss function, with the following expression: ; ; ; ; in, This represents the total loss value. The basic prediction loss is given, where m is the number of training samples. , The first The predicted log standard deviation and predicted log mean of each training sample , The first The true log standard deviation and true log mean of each training sample; For time-constrained loss, This refers to the number of consecutive time points under the same operating condition. , For adjacent time points under the same working condition, , They are respectively The standard deviation and mean of the predicted logarithm at time 1. , They are respectively The standard deviation and mean of the predicted logarithm at time 1; Losses due to operating conditions constraints The number of similar operating condition groups is determined by the following method: when the Euclidean distance of the atomizing operating condition parameters is less than a preset threshold. ε When the condition is similar, it is determined to be a similar working condition; For the first The set of predicted log-standard deviations for each working condition in a group of similar working conditions. For the first The variance of the predicted logarithmic standard deviation under similar working conditions; No. The set of predicted log-mean values ​​for each working condition in a group of similar working conditions. For the first Variance of the predicted logarithmic mean under similar working conditions; , Separate time-series constraint strength coefficients and operating condition constraint strength coefficients; With the goal of minimizing the total loss value of the loss function, the network parameters of the neural network prediction model are iteratively updated until the total loss value of the loss function converges or the model training reaches the preset number of iterations, thus completing the training of the neural network prediction model.

[0008] Furthermore, the neural network prediction model is a generalized regression neural network.

[0009] Furthermore, the step of extracting droplet size data from each atomized image includes: For each atomized image, grayscale processing, noise suppression processing, and adaptive threshold binarization processing are performed sequentially to generate a binary image with salient target droplet region; Based on the binary image, an eight-neighbor connected component labeling algorithm is used to identify each droplet region, and the area equivalent diameter of each droplet region is calculated as the initial particle size value. Based on preset nozzle structure parameters, a particle size upper limit threshold is set, and continuous liquid phase regions with an equivalent diameter exceeding the particle size upper limit threshold are eliminated; at the same time, non-spherical interference regions with a shape factor lower than the preset roundness threshold are eliminated, and effective droplet regions that conform to morphological characteristics are retained. The set of equivalent diameter values ​​corresponding to all effective droplet regions constitutes a droplet size dataset that uniquely corresponds to the atomization image.

[0010] Furthermore, the reconstructing of the particle size distribution at the target time based on the predicted log mean and predicted log standard deviation specifically includes: predict the log mean and the predicted log standard deviation Substituting into the log-normal probability density function, the expression for the log-normal probability density function is: ; in, The droplet size to be predicted; Set the preset particle size statistical interval [ , It is divided into multiple continuous and non-overlapping particle size sub-intervals; Substitute the predicted log-mean value into each particle size sub-interval. and the predicted log standard deviation The log-normal probability density function is integrated, and the integral result of each particle size sub-interval is used as the distribution probability of droplet appearance in the corresponding particle size sub-interval. The discrete droplet size distribution histogram and the continuous droplet size distribution curve at the target time are generated based on the distribution probability of each droplet size sub-interval.

[0011] Furthermore, the step of acquiring the atomization image sequence during the fuel transverse jet process according to a preset time series under different atomization conditions includes: Multiple discrete level values ​​are set for each atomization operation parameter, and multiple operation parameter combinations are formed by combining the discrete level values ​​of each parameter. For each combination of operating parameters, during the fuel transverse jet atomization process, the atomization start time is continuously collected at fixed time intervals from the atomization start time to the preset end time, and a single frame atomization image corresponding to each time sampling point is obtained, forming a time series atomization image set. For each combination of operating parameters and each time sampling point, the image acquisition operation is independently and repeatedly performed a preset number of times to obtain multiple sets of repeated fogged image data corresponding to that acquisition node.

[0012] Furthermore, the atomization condition parameters include at least one or more of the following: fuel injection pressure, fuel physical property parameters, airflow velocity, and injection structure parameters.

[0013] This specification provides a fuel atomization particle size distribution prediction device, comprising: The data acquisition module is used to acquire a sequence of atomized images during the lateral jet of fuel under different atomization conditions and parameters, according to a preset time series. Each atomized image in the atomized image sequence corresponds to a time point in the time series. The particle size distribution fitting module is used to extract droplet particle size data from each atomization image and fit the droplet particle size data to a log-normal distribution to obtain distribution parameters characterizing the particle size morphology distribution in each atomization image. The distribution parameters include the logarithmic mean and the logarithmic standard deviation. The prediction model building module is used to construct a neural network prediction model. It takes atomization condition parameters and time series as inputs and distribution parameters as outputs. The neural network prediction model is trained through a loss function that includes condition constraint loss and time series constraint loss. The condition constraint loss is used to ensure that the distribution parameters of similar conditions under the same time do not have significant deviations. The time series constraint loss is used to ensure that the distribution parameters of continuous time under the same condition do not have abrupt changes. The distribution prediction module is used to input the target atomization condition parameters and the target time into the trained neural network prediction model, output the predicted distribution parameters, and reconstruct the fuel atomization particle size distribution at the target time based on the predicted distribution parameters.

[0014] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described fuel atomization particle size distribution prediction method.

[0015] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described fuel atomization particle size distribution prediction method.

[0016] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: This invention reduces the dimensionality of high-dimensional, massive discrete particle size data into two low-dimensional distribution parameters—logarithmic mean and logarithmic standard deviation—by fitting a log-normal distribution to the droplet size data of fuel transverse jet atomization. These parameters replace the entire discrete data as the model's prediction target, reducing data processing volume and model training complexity. Simultaneously, the model uses atomization condition parameters and time series as joint inputs, and introduces operating condition constraint loss and temporal constraint loss during training. This constrains the consistency of distribution parameters under similar operating conditions at the same time and limits the abrupt changes in distribution parameters over continuous time under the same operating condition. It fully captures the temporal dynamic evolution characteristics and operating condition correlation of the atomization process, allowing the model to fully learn the mapping law of particle size distribution dynamic evolution over time under different atomization condition parameters. This enables the model to capture the dynamic evolution characteristics of the time dimension, adapting to the dynamic changes of different atomization conditions and atomization times, and achieving dynamic prediction of fuel transverse jet atomization particle size distribution. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 This is a flowchart illustrating a method for predicting the particle size distribution of fuel atomization provided in this specification. Figure 2 This is a schematic diagram of the image processing results in a transverse jet atomization particle size distribution reconstruction method provided in this specification; Figure 3 This is a schematic diagram illustrating the fitting of particle size parameter distribution to a log-normal distribution, as provided in this specification. Figure 4 This is a schematic diagram of a neural network prediction result provided in this specification; Figure 5 This is a schematic diagram of a fuel atomization particle size distribution prediction device provided in this specification; Figure 6This is a schematic diagram of a computer device provided for this specification. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0020] In the field of analysis, primarily targeting fuel transverse jet atomization scenarios, this technology integrates automatic image recognition, intelligent parameter calculation, and neural network distribution trend prediction to achieve dynamic and accurate prediction and reconstruction of atomized particle size distribution. This effectively meets the core needs of aerospace, automotive engines, and other fields driven by dual carbon objectives for optimizing fuel transverse jet atomization combustion efficiency and controlling pollutant emissions. The method for predicting fuel atomization particle size distribution of the present invention is described below with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart illustrating a method for predicting fuel atomization particle size distribution provided in this specification, as shown below. Figure 1 As shown, the method includes the following: S1. Under different atomization conditions, acquire a sequence of atomization images during the lateral jet of fuel according to a preset time series. Each atomization image in the sequence corresponds to a time point in the time series.

[0022] For example, in the embodiments of this application, when acquiring the original atomization images of the fuel transverse jet atomization process under different atomization conditions, the original atomization images in conventional image format can be obtained through various methods such as experimental acquisition and numerical simulation output. First, multiple discrete level values ​​are set for each atomization condition parameter and combined to form multiple working condition parameter combinations. Then, for each working condition parameter combination, the fuel transverse jet atomization process is continuously acquired from the atomization start time to the preset end time at fixed time intervals to obtain single-frame atomization images corresponding to each time sampling point. This constitutes a time-series atomization image set that can capture the complete dynamic process of liquid column breakup, droplet evolution, and steady-state atomization. At the same time, for each working condition parameter combination and each time sampling point, the image acquisition operation is independently repeated a preset number of times at the acquisition node to obtain multiple sets of repeated atomization image data corresponding to that acquisition node.

[0023] S2. Extract droplet size data from each atomized image and fit the droplet size data to a log-normal distribution to obtain distribution parameters characterizing the particle size distribution in each atomized image. The distribution parameters include the logarithmic mean and the logarithmic standard deviation.

[0024] For example, extracting droplet size data from the original atomized image may include: performing grayscale and binarization preprocessing on the original atomized image sequentially to remove noise interference; analyzing the preprocessed image using a connected component analysis algorithm to automatically identify independent droplets in the image, and extracting the pixel area and roundness features of each droplet; converting the pixel features of the droplets into actual particle size and actual area based on the pixel calibration scale input by the user (i.e., N pixels corresponding to the actual length L); filtering effective droplets according to preset conditions, removing the jet liquid column portion with a particle size larger than the nozzle diameter, and non-spherical interference objects with a roundness smaller than a preset roundness threshold (e.g., set to 0.1), thus eliminating invalid interference data; counting the total number of remaining effective droplets after filtering, and calculating the Sottle mean diameter (SMD), finally outputting the effective droplet size dataset, which is used as the basis for subsequent log-normal distribution fitting of droplet size data.

[0025] For example, in this embodiment of the application, a log-normal distribution is fitted to the droplet size data to obtain the logarithmic mean and logarithmic standard deviation used to characterize the size distribution morphology. Specifically, this may include: using the effective droplet size dataset as input data, calling the log-normal distribution fitting function, specifying the distribution type as "Lognormal", and performing fitting calculations according to the log-normal distribution probability density formula; through the fitting calculation process, directly outputting the logarithmic mean and logarithmic standard deviation that can completely characterize the size distribution morphology of the fuel transverse jet atomization; simultaneously generating an original size distribution histogram based on the original droplet size data, and generating a log-normal distribution fitting curve by combining the fitted logarithmic mean and logarithmic standard deviation, and comparing the histogram with the fitting curve to intuitively verify the fitting effect of the log-normal distribution, thus achieving a simplified characterization that replaces massive discrete droplet size data with two parameters: the logarithmic mean and the logarithmic standard deviation.

[0026] S3. Construct a neural network prediction model, taking atomization condition parameters and time series as inputs and distribution parameters as outputs. Train the neural network prediction model using a loss function that includes condition constraint loss and time series constraint loss. The condition constraint loss is used to constrain the distribution parameters of similar conditions under the same time to have no significant deviation; the time series constraint loss is used to constrain the distribution parameters of continuous time under the same condition to have no abrupt changes.

[0027] Based on any of the above embodiments, for example, a neural network prediction model is constructed to learn the dynamic mapping relationship between atomization condition parameters, atomization process time parameters, and particle size distribution characteristic parameters. The specific implementation steps are as follows: First, a generalized regression neural network is selected as the basic architecture of the model, and a composite loss function containing temporal constraint loss term and working condition constraint loss term is constructed so that the model can simultaneously satisfy the physical constraints of temporal evolution continuity and working condition response consistency during the training process.

[0028] Secondly, training samples covering multiple combinations of atomization operating parameters and multiple atomization time points are collected. The input features of each sample group consist of atomization operating parameters and normalized time parameters. The atomization operating parameters include at least one of jet pressure, fuel viscosity, fuel density, and airflow velocity. The normalized time parameters characterize the time process after atomization starts. The output labels include the logarithmic mean, logarithmic standard deviation, and the number of effective droplets at the corresponding operating conditions and time points obtained through the aforementioned log-normal distribution fitting step.

[0029] Next, the input features are normalized preprocessed to scale all parameters to a standardized numerical range, thereby eliminating dimensional differences and improving the stability and convergence efficiency of model training.

[0030] Subsequently, the preprocessed training samples are input into the generalized regression neural network for iterative training. The objective is to minimize the total loss value of the loss function, with the logarithmic mean and logarithmic standard deviation as the core outputs and the effective droplet number as an auxiliary supervision signal. By optimizing the composite loss function, the model accurately captures the nonlinear dynamic evolution law between atomization conditions, time process, and particle size distribution characteristics. When the total loss value of the loss function converges or the model training reaches the preset number of iterations, the training of the neural network prediction model is completed. The formula for the dual-constraint loss function used in model training is as follows:

[0031] ; ; ; ; in, This represents the total loss value. The basic prediction loss is represented by the mean squared error (MSE), where m is the number of training samples. , The first The predicted log standard deviation and predicted log mean of each training sample , The first The true log standard deviation and true log mean of each training sample; For time-constrained loss, This refers to the number of consecutive time points under the same operating condition. , For adjacent time points under the same working condition, , They are respectively The standard deviation and mean of the predicted logarithm at time 1. , They are respectively The standard deviation and mean of the predicted logarithm at time 1; Losses due to operating conditions constraints The number of similar operating condition groups is determined by the following method: when the Euclidean distance of the atomizing operating condition parameters is less than a preset threshold. ε When the condition is similar, it is determined to be a similar working condition; For the first The set of predicted log-standard deviations for each working condition in a group of similar working conditions. For the first The variance of the predicted logarithmic standard deviation under similar working conditions; No. The set of predicted log-mean values ​​for each working condition in a group of similar working conditions. For the first Variance of the predicted logarithmic mean under similar working conditions; , The time-series constraint strength coefficient and the working condition constraint strength coefficient are respectively selected, with a value range of 0.1~0.5. The optimal value is determined by cross-validation to balance the basic prediction accuracy and constraint strength.

[0032] S4. Input the target atomization condition parameters and target time into the trained neural network prediction model, output the predicted distribution parameters, and reconstruct the fuel atomization particle size distribution at the target time based on the predicted distribution parameters.

[0033] For example, input the predicted log mean and predicted log standard deviation into the log-normal probability density function: ; in, The droplet size to be predicted is denoted as .

[0034] Based on the preset particle size statistical interval [ , ] and the number of interval divisions, will count the intervals [ , The particle size distribution is divided into multiple continuous and non-overlapping sub-intervals; the probability density integral value of each sub-interval is calculated as the frequency distribution, generating discrete particle size distribution histograms and continuous particle size distribution curves, realizing the mapping from low-dimensional parameter prediction to high-dimensional distribution reconstruction.

[0035] The prediction and reconstruction of the fuel transverse jet atomization particle size distribution is achieved through a GRNN neural network prediction model. The specific execution process is as follows: After normalizing the target atomization condition parameters and target time parameters required for actual application, the data is input into the trained GRNN neural network prediction model. Based on the mapping rules obtained from training, the model directly outputs the predicted logarithmic mean and predicted logarithmic standard deviation for the target operating condition and target time, and also outputs the predicted effective droplet number N. Substituting the predicted logarithmic mean and predicted logarithmic standard deviation into the log-normal distribution probability density formula, the model automatically calculates the frequency distribution of droplet size within the user-defined particle size statistical interval. Based on the calculated frequency distribution, the model generates a frequency histogram and frequency line graph of droplet size in real time, and automatically back-generates the complete fuel transverse jet atomization particle size distribution curve for the target operating condition and target time, thereby realizing the technical closed loop of parameter prediction and distribution reconstruction. In addition, the above model retains the interactive adjustment function of binarization threshold and particle size / roundness screening range. It can dynamically update the fitting results of log mean and log standard deviation and particle size prediction distribution according to the user's operation instructions, and can adapt to the needs of different fogging image quality and various practical application scenarios.

[0036] This embodiment also verifies the effectiveness of the above method through experimental simulation, and the corresponding experimental simulation results are as follows: Figures 2-4 As shown, where Figure 2 This is a schematic diagram of the image processing results in the reconstruction method of transverse jet atomization particle size distribution. It is an atomization image after binarization. The black area clearly shows the dynamic process of the fuel jet column gradually breaking down from a continuous shape on the left and spreading to the upper right into a large number of discrete droplets. The droplet density gradually decreases with the diffusion distance, which intuitively reflects the atomization evolution morphology. Figure 3 This is a schematic diagram showing the fitting of the particle size parameter distribution to the log-normal distribution. Figure 4 This is a schematic diagram of the prediction results of the neural network.

[0037] The fuel atomization particle size distribution prediction device provided by the present invention is described below. The fuel atomization particle size distribution prediction device described below can be referred to in correspondence with the fuel atomization particle size distribution prediction method described above.

[0038] Figure 5 This is a schematic diagram of the structure of a fuel atomization particle size distribution prediction device provided by the present invention. For example, please refer to [link to schematic diagram]. Figure 5 As shown, the fuel atomization particle size distribution prediction device may include: The data acquisition module is used to acquire a sequence of atomized images during the transverse jet process of fuel under different atomization conditions and parameters, according to a preset time series. Each atomized image in the atomized image sequence corresponds to a time point in the time series.

[0039] The particle size distribution fitting module is used to extract droplet particle size data from each atomization image and fit the droplet particle size data to a log-normal distribution to obtain distribution parameters characterizing the particle size morphology distribution in each atomization image. The distribution parameters include the logarithmic mean and the logarithmic standard deviation.

[0040] The prediction model building module is used to construct a neural network prediction model. It takes atomization condition parameters and time series as inputs and distribution parameters as outputs. The neural network prediction model is trained through a loss function that includes condition constraint loss and time series constraint loss. The condition constraint loss is used to ensure that the distribution parameters of similar conditions under the same time do not have significant deviations. The time series constraint loss is used to ensure that the distribution parameters of continuous time under the same condition do not have abrupt changes.

[0041] The distribution prediction module is used to input the target atomization condition parameters and the target time into the trained neural network prediction model, output the predicted distribution parameters, and reconstruct the fuel atomization particle size distribution at the target time based on the predicted distribution parameters.

[0042] Specific limitations regarding the fuel atomization particle size distribution prediction device can be found in the limitations on fuel atomization particle size distribution prediction described above, and will not be repeated here. Each module in the aforementioned fuel atomization particle size distribution prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0043] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A method for predicting fuel atomization particle size distribution is provided.

[0044] This instruction manual also provides Figure 6 The schematic diagram of the computer device shown is as follows: Figure 6 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 A method for predicting fuel atomization particle size distribution is provided.

[0045] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0046] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method of predicting a fuel oil atomized particle size distribution, characterized by, include: Under different atomization conditions, a sequence of atomization images during the lateral jet of fuel is acquired according to a preset time series. Each atomization image in the sequence corresponds to a time point in the time series. Droplet size data are extracted from each atomized image, and the droplet size data are fitted with a log-normal distribution to obtain distribution parameters characterizing the particle size morphology distribution in each atomized image. The distribution parameters include the logarithmic mean and the logarithmic standard deviation. A neural network prediction model is constructed, taking atomization condition parameters and time series as inputs and distribution parameters as outputs. The neural network prediction model is trained using a loss function that includes condition constraint loss and time series constraint loss. The condition constraint loss is used to constrain the distribution parameters of similar conditions under the same time to have no significant deviation; the time series constraint loss is used to constrain the distribution parameters of continuous time under the same condition to have no abrupt changes. The target atomization condition parameters and target time are input into the trained neural network prediction model, which outputs the predicted distribution parameters. Based on the predicted distribution parameters, the fuel atomization particle size distribution at the target time is reconstructed.

2. The fuel atomized particle size distribution prediction method according to claim 1, characterized by, The training of the neural network prediction model using a loss function that includes both operational condition constraint loss and temporal constraint loss specifically includes: Construct the loss function, with the following expression: ； ； ； ； in, This represents the total loss value. The basic prediction loss is given, where m is the number of training samples. , The first The predicted log standard deviation and predicted log mean of each training sample , The first The true log standard deviation and true log mean of each training sample; For time-constrained loss, This refers to the number of consecutive time points under the same operating condition. , For adjacent time points under the same working condition, , They are respectively The standard deviation and mean of the predicted logarithm at time 1. , They are respectively The standard deviation and mean of the predicted logarithm at time 1; Losses due to operating conditions constraints The number of similar operating condition groups is determined by the following method: when the Euclidean distance of the atomizing operating condition parameters is less than a preset threshold. ε When the condition is similar, it is determined to be a similar working condition; For the first The set of predicted log-standard deviations for each working condition in a group of similar working conditions. For the first The variance of the predicted logarithmic standard deviation under similar working conditions; No. The set of predicted log-mean values ​​for each working condition in a group of similar working conditions. For the first Variance of the predicted logarithmic mean under similar working conditions; , Separate time-series constraint strength coefficients and operating condition constraint strength coefficients; With the goal of minimizing the total loss value of the loss function, the network parameters of the neural network prediction model are iteratively updated until the total loss value of the loss function converges or the model training reaches the preset number of iterations, thus completing the training of the neural network prediction model.

3. The fuel atomization particle size distribution prediction method as described in claim 1, characterized in that, The neural network prediction model is a generalized regression neural network.

4. The fuel atomization particle size distribution prediction method as described in claim 1, characterized in that, The step of extracting droplet size data from each atomized image includes: For each atomized image, grayscale processing, noise suppression processing, and adaptive threshold binarization processing are performed sequentially to generate a binary image with salient target droplet region; Based on the binary image, an eight-neighbor connected component labeling algorithm is used to identify each droplet region, and the area equivalent diameter of each droplet region is calculated as the initial particle size value. Based on preset nozzle structure parameters, a particle size upper limit threshold is set, and continuous liquid phase regions with an equivalent diameter exceeding the particle size upper limit threshold are eliminated; at the same time, non-spherical interference regions with a shape factor lower than the preset roundness threshold are eliminated, and effective droplet regions that conform to morphological characteristics are retained. The set of equivalent diameter values ​​corresponding to all effective droplet regions constitutes a droplet size dataset that uniquely corresponds to the atomization image.

5. The fuel atomization particle size distribution prediction method as described in claim 1, characterized in that, The reconstructing of the particle size distribution at the target time based on the predicted log mean and predicted log standard deviation specifically includes: predict the log mean and the predicted log standard deviation Substituting into the log-normal probability density function, the expression for the log-normal probability density function is: ； in, The droplet size to be predicted; Set the preset particle size statistical interval [ , It is divided into multiple continuous and non-overlapping particle size sub-intervals; Substitute the predicted log-mean value into each particle size sub-interval. and the predicted log standard deviation The log-normal probability density function is integrated, and the integral result of each particle size sub-interval is used as the distribution probability of droplet appearance in the corresponding particle size sub-interval. The discrete droplet size distribution histogram and the continuous droplet size distribution curve at the target time are generated based on the distribution probability of each droplet size sub-interval.

6. The fuel atomization particle size distribution prediction method as described in claim 1, characterized in that, The step of acquiring the atomization image sequence of the fuel transverse jet process according to a preset time series under different atomization conditions includes: Multiple discrete level values ​​are set for each atomization operation parameter, and multiple operation parameter combinations are formed by combining the discrete level values ​​of each parameter. For each combination of operating parameters, during the fuel transverse jet atomization process, the atomization start time is continuously collected at fixed time intervals from the atomization start time to the preset end time, and a single frame atomization image corresponding to each time sampling point is obtained, forming a time series atomization image set. For each combination of operating parameters and each time sampling point, the image acquisition operation is independently and repeatedly performed a preset number of times to obtain multiple sets of repeated fogged image data corresponding to that acquisition node.

7. The fuel atomization particle size distribution prediction method as described in claim 1, characterized in that, The atomization condition parameters include at least one or more of the following: fuel injection pressure, fuel physical property parameters, airflow velocity, and injection structure parameters.

8. A fuel atomization particle size distribution prediction device, characterized in that, include: The data acquisition module is used to acquire a sequence of atomized images during the lateral jet of fuel under different atomization conditions and parameters, according to a preset time series. Each atomized image in the atomized image sequence corresponds to a time point in the time series. The particle size distribution fitting module is used to extract droplet particle size data from each atomization image and fit the droplet particle size data to a log-normal distribution to obtain distribution parameters characterizing the particle size morphology distribution in each atomization image. The distribution parameters include the logarithmic mean and the logarithmic standard deviation. The prediction model building module is used to construct a neural network prediction model. It takes atomization condition parameters and time series as inputs and distribution parameters as outputs. The neural network prediction model is trained through a loss function that includes condition constraint loss and time series constraint loss. The condition constraint loss is used to ensure that the distribution parameters of similar conditions under the same time do not have significant deviations. The time series constraint loss is used to ensure that the distribution parameters of continuous time under the same condition do not have abrupt changes. The distribution prediction module is used to input the target atomization condition parameters and the target time into the trained neural network prediction model, output the predicted distribution parameters, and reconstruct the fuel atomization particle size distribution at the target time based on the predicted distribution parameters.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the fuel atomization particle size distribution prediction method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fuel atomization particle size distribution prediction method as described in any one of claims 1 to 7.

Images