Physical information and data driven based powder fuel mass flow rate prediction model

Abstract

The application relates to a powder fuel mass flow rate prediction model based on physical information and data driving, and relates to the technical field of laser spectrum application. A laser system generates a laser beam, an optical modulation and shaping system modulates the laser beam, a signal detection system suppresses background light noise, a camera system collects a scattering image, an image preprocessing module performs clipping preprocessing, a feature extraction module learns image features through a deep belief network, an XGBoost regression prediction module considers physical constraint parameters, and the learned image features and real-time mass flow rate data are input into an XGBoost model for training calculation to construct a nonlinear mass flow rate prediction model. The scattering image and real-time mass flow rate are used as original data, physical constraint parameters are considered, image features are extracted through a deep belief network, and an XGBoost model is used for training to construct a nonlinear mass flow rate prediction model based on physical information and data driving, so that the accuracy of mass flow rate prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of laser spectroscopy application technology, specifically to a powder fuel mass flow rate prediction model based on physical information and data. Background Technology

[0002] Powder fuel engines offer advantages such as adjustable flow rate, high combustion efficiency, and high specific impulse, making them suitable for combat missions involving wide airspace, broad speed ranges, and multiple ballistic trajectories. They have broad application scenarios in next-generation long-range air-to-air missiles, air-to-ground missiles, and air defense missiles. These engines use high-energy metal or boron powders as fuel. Under the action of fluidizing gas, the powdered fuel enters the combustion chamber in a gas-solid two-phase form for combustion, thereby generating thrust. As a core component of powder fuel engines, the performance of the fuel supply system directly determines the engine's thrust adjustment capability and multi-pulse start-up characteristics. To adapt to the complex requirements of multi-mission scenarios, the fuel supply system not only needs flexible adjustment of mass flow rate but also must maintain high flow rate stability under constant operating conditions. Therefore, achieving real-time, high-precision prediction of powdered fuel mass flow rate is crucial for developing highly reliable fuel supply technology and achieving precise control of engine thrust. Accurate prediction of powdered fuel mass flow rate can provide key data support for the control strategy of the fuel supply system, thereby optimizing the overall engine performance and ensuring its stable and efficient operation under diverse working conditions.

[0003] Current research on the measurement of mass flow rate of powdered fuels mainly focuses on pneumatic transport pipeline environments, with limited research conducted on measurements under specific fuel supply methods characteristic of powdered fuel engines. Such studies are limited to steady-state measurement methods such as Coriolis flowmeters and gravimetric methods. However, the mass flow rate of dense gas-solid two-phase flow fields is affected by various factors, such as fluidizing gas pressure, piston velocity, powdered fuel characteristics, and fluidization parameters. This leads to significant differences in the characteristics of dense gas-solid two-phase flow fields, and the importance of these characteristics is unclear. Therefore, existing models developed based on steady-state measurement results cannot achieve high-precision prediction of the mass flow rate of powdered fuels.

[0004] In recent years, machine learning methods have demonstrated great potential as a new approach to predictive modeling, especially in situations involving multi-factor coupling and strong interactions between variables. Machine models show particular advantages in these scenarios. They focus more on the correlations and relationships between data samples, mining potential patterns and connections between input and output variables in high-dimensional space to establish complex mapping relationships. Significant progress has been made in characterizing the mass flow rate of powdered fuels. For example, support vector machines and convolutional neural networks have been used to mine feature parameters and obtain the mass flow rate of powdered fuels under the influence of multiple factors. This has been applied in pneumatic transport pipelines with good results. However, to date, no related research reports have been published in the field of powdered fuel engines. Although machine learning models offer new ideas for constructing mass flow rate prediction models due to their powerful data mining capabilities, the reliability of the prediction results remains questionable as they are black-box systems lacking physical constraints.

[0005] In view of the above analysis, in order to achieve high-precision prediction of powdered fuel mass flow rate, it is urgent to establish a powdered fuel mass flow rate prediction model based on physical information and data-driven approach. By combining the physical information to describe the mechanism of powdered fuel transportation process and the data-driven ability to fit complex nonlinear relationships, reliable technical support can be provided for the optimized control of fuel supply system. Summary of the Invention

[0006] To address the shortcomings of the prior art, this invention provides a powdered fuel mass flow rate prediction model based on physical information and data-driven methods. This model uses scattering images and real-time mass flow rates as raw data, while also considering physical constraint parameters. It extracts image features through a deep belief network, trains the model based on an XGBoost model, and constructs a nonlinear mass flow rate prediction model based on physical information and data-driven methods, thereby improving the accuracy of mass flow rate prediction.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a powdered fuel mass flow rate prediction model based on physical information and data-driven, including a data acquisition system and a computer system. The data acquisition system includes a laser system, an optical modulation and shaping system, a signal detection system, and a camera system. The computer system includes an image preprocessing module, a feature extraction module, and an XGBoost regression prediction module.

[0008] The laser system generates a laser beam by determining the wavelength based on the type of powdered fuel in the combustion chamber.

[0009] The optical modulation and shaping system is arranged between the laser system and the engine to modulate the laser beam;

[0010] The signal detection system is positioned behind the area to be tested to suppress background light noise;

[0011] The camera system acquires scattered images of powdered fuel particles in the combustion chamber, and combines the particle density and particle flow velocity to obtain the real-time mass flow rate of the scattered images.

[0012] The image preprocessing module performs cropping preprocessing on the acquired scattering image;

[0013] The feature extraction module learns the image features of the scattering image through a deep belief network;

[0014] The XGBoost regression prediction module considers physical constraint parameters and inputs the learned image features and real-time mass flow rate data into the XGBoost model for training and calculation to construct a nonlinear mass flow rate prediction model.

[0015] The specific implementation steps are as follows:

[0016] Step 1: Collection and Processing of Raw Data

[0017] Multiple sets of scattered images of powdered fuel were acquired using a camera system. The particle density of the powdered fuel was measured based on the extinction principle. The particle velocity was measured using PIV and optical flow algorithms. The real-time mass flow rate of the scattered images was obtained by combining the particle density and particle velocity. The scattered images were labeled with physical constraint parameters as raw data.

[0018] Step 2: Extracting deep image features from the data

[0019] The scattering image is cropped into an image format suitable for use as input to deep learning. Image features are extracted through a deep belief network without considering labels. The network consists of multiple stacked RBMs. Through layer-by-layer training, the features learned by each RBM are used as input to the next layer.

[0020] Step 3: Construct a nonlinear mass flow rate prediction model

[0021] The extracted image features, real-time quality flow rate considering labels, and physical constraint parameters form a dataset as input variables, and the predicted quality flow rate is used as the output variable. This dataset is trained using an XGBoost model with the following objective function: Where n is the total number of training data points, and K is the total number of decision trees trained. For loss function, y i For the i-th real-time quality flow rate data, For the i-th predicted mass flow rate data, Ω(f k ) represents the regularization term. f k For the decision tree trained in the kth round, T k w represents the number of leaf nodes in the decision tree trained in the kth round. jLet γ be the weight of the j-th leaf node, and λ be the hyperparameters.

[0022] The XGBoost model optimizes in a forward stepwise manner, training a new regression tree f in each round. k (x) is used to optimize the residuals from the previous round, then the predicted value for the k-th round is... f is the mass flow rate predicted in the first k-1 rounds. k (x i ) represents the newly trained decision tree in the kth round, paired with input x. i The predicted value, x i Let be the image features and physical constraint parameters in the i-th set of data;

[0023] The newly trained regression tree f k (x) Optimize the prediction error by fitting the negative gradient of the current loss, where the negative gradient of the i-th data in the k-th round is... Through continuous iteration, the final predicted quality flow rate data is represented by a weighted sum of all decision trees. A well-trained nonlinear mass flow rate prediction model is obtained;

[0024] Step 4: Model Accuracy Assessment

[0025] The preprocessed scattering image of the powdered fuel is input into the trained nonlinear mass flow rate prediction model, which outputs the predicted mass flow rate of the powdered fuel. The prediction accuracy is evaluated by comparing the predicted mass flow rate with the real-time mass flow rate considering the label. If the requirements are not met, the hyperparameters γ and λ in the objective function are adjusted.

[0026] Furthermore, the physical constraint parameters include the piston movement speed of the engine, the bore size, and the particle size of the powdered fuel particles.

[0027] Furthermore, the RBM consists of a visible layer and a hidden layer, and the weight value between any two connected neurons in the visible layer and the hidden layer is w. ij The neuron states in the visible and hidden layers are v, respectively. i The energy function of RBM, defined as hj, is: E(v,h)=-∑a i v i -∑b j h j -∑v i w ij h j Among them, ai and b j This is a bias term.

[0028] Furthermore, the accuracy assessment uses the average error as the criterion. If the average error is ≤5%, the prediction result of the nonlinear mass flow rate prediction model is considered to meet the requirements.

[0029] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention utilizes a camera system to acquire scattering images of powdered fuel and uses a deep belief network to perform feature representation learning on the original image data. The extracted features are input into a supervised learning regression model to establish a mapping relationship between the scattering image and the mass flow rate. The mass flow rate is calibrated using real-time flow rate data obtained by the extinction method and is used as the target value (label) of the regression model. At the same time, physical parameters closely related to the mass flow rate, such as the piston movement speed of the engine, the diameter of the nozzle, and the particle size of the powdered fuel particles, are considered as auxiliary physical constraints. A powdered fuel mass flow rate prediction model based on physical information and data-driven methods is constructed, which helps to improve the accuracy of mass flow rate prediction and solve the problem of unstable operation of powdered fuel engines caused by mass flow rate deviation. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the predictive model structure of the present invention;

[0031] Figure 2 This is a schematic diagram illustrating the relationship between RBM and DBN in this invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0033] like Figures 1-2 As shown, a powdered fuel mass flow rate prediction model based on physical information and data-driven methods includes a data acquisition system and a computer system. The data acquisition system includes a laser system, an optical modulation and shaping system, a signal detection system, and a camera system, used to acquire scattering images of powdered fuel in real time. The computer system includes an image preprocessing module, a feature extraction module, and an XGBoost regression prediction module, used for image preprocessing and model training.

[0034] The laser system generates a laser beam by determining the wavelength based on the type of powdered fuel in the combustion chamber.

[0035] The optical modulation and shaping system is arranged between the laser system and the engine and is used to modulate the laser beam;

[0036] The signal detection system is positioned behind the area to be measured at the same horizontal level as the laser system to suppress unmodulated background light noise and improve the signal-to-noise ratio of signal detection.

[0037] The camera system acquires scattered images of powdered fuel particles in the combustion chamber, and combines the particle density and particle flow velocity to obtain the real-time mass flow rate of the scattered images.

[0038] The image preprocessing module performs cropping preprocessing on the acquired scattering image;

[0039] The feature extraction module learns the image features of the scattering image through a deep belief network;

[0040] The XGBoost regression prediction module considers physical constraints such as the piston movement speed, nozzle size, and particle size of the powdered fuel, and inputs the learned image features and real-time mass flow rate data into the XGBoost model for training and calculation, thereby constructing a nonlinear mass flow rate prediction model to predict the mass flow rate.

[0041] When applying the prediction model of this invention, firstly, multiple sets of scattering images of powdered fuel and their associated real-time mass flow rates are acquired as raw data. Simultaneously, physical constraint parameters such as engine piston speed, nozzle size, and powdered fuel particle size are considered. These scattering images are labeled. Then, the acquired scattering images are input into a deep belief network for unsupervised learning without considering the labels, extracting image features from the raw data to obtain feature patterns closer to the original essence of the data. Finally, a nonlinear mass flow rate prediction model is constructed using a supervised learning-based XGBoost model, utilizing a dataset containing image features, real-time mass flow rates considering labels, and physical constraint parameters. The specific implementation steps are as follows:

[0042] Step 1: Collection and Processing of Raw Data

[0043] The laser beam generated by the laser system, after passing through an optical modulation and shaping system, is used as a light source and incident on the test area of ​​the combustion chamber to excite powdered fuel particles in the same plane. A camera system acquires multiple sets of scattered images of the powdered fuel. The particle density of the powdered fuel is measured by comparing the intensity of transmitted light and incident light based on the extinction principle using a signal detection system. For small particles and low-concentration areas, particle velocity is measured using PIV (Piston-Volume Interval). For dense gas-solid two-phase flow fields, an optical flow algorithm is used to calculate the motion of pixels or feature points in the image sequence between consecutive image frames to obtain the velocity vector field. Combining particle density and particle velocity, the real-time mass flow rate of the scattered images is obtained. Considering physical constraints such as the piston speed, nozzle size, and particle size of the powdered fuel, these scattered images are labeled and used as raw data for subsequent XGBoost model training.

[0044] Step 2: Extracting deep image features from the data

[0045] The scattering image is cropped and shaped into an image format suitable for deep learning input by the image preprocessing module, and then input into the feature extraction module without considering labels. The feature extraction module uses a deep belief network (DBN) to extract image features, combined with... Figure 2 As shown, it is composed of multiple Restricted Boltzmann Machines (RBMs) stacked together. Each RBM consists of a visible layer and a hidden layer. The number of neurons in the first visible layer is determined by the dimension of the input data. The weight between any two connected neurons in the visible and hidden layers is w. ij The neuron states in the visible and hidden layers are v, respectively. i and h j The energy function of the RBM is defined as: E(v,h)=-∑a i v i -∑b j h j -∑v i w ij h j Where ai and bj are bias terms. Through layer-by-layer training, the features learned by each RBM are used as the input to the next layer, and a DBN is formed by concatenating multiple RBMs. The DBN uses a non-linear structure to extract image features and map data from a high-dimensional space to a low-dimensional space, finding the best feature representation that can reconstruct the original data to the greatest extent possible;

[0046] Step 3: Construct a nonlinear mass flow rate prediction model

[0047] The extracted image features, real-time mass flow rate considering the label, and physical constraint parameters are combined into a dataset and used as input variables. The predicted mass flow rate is used as the output variable, and the XGBoost regression prediction module is used to predict the mass flow rate of powdered fuel. The core of the XGBoost model is to iteratively train multiple decision tree models and combine them into a powerful ensemble model. Its objective function consists of a loss function and a regularization term, which can be expressed as follows: Where n is the total number of training data points, and K is the total number of decision trees trained. The loss function is the mean squared error (MSE). y i For the i-th real-time quality flow rate data, For the i-th predicted mass flow rate data, Ω(f k ) is a regularization term used to prevent overfitting, defined as: f k For the decision tree trained in the kth round, T k w represents the number of leaf nodes in the decision tree trained in the kth round. jLet f be the weight of the j-th leaf node, and γ and λ be hyperparameters used to control the complexity of the model. k This represents the entire decision tree and is used to calculate the structural complexity of the tree, but the specific input x is not involved at this point.

[0048] The XGBoost model optimizes in a forward stepwise manner, training a new regression tree f in each round. k (x) is used to optimize the residuals from the previous round, then the predicted value for the k-th round is... f is the mass flow rate predicted in the first k-1 rounds. k (x i ) represents the newly trained decision tree in the kth round, paired with input x. i The predicted value, x i Let f be the image features and physical constraint parameters in the i-th data set. k (x) represents the decision tree f k The predicted value applied to the input x, and the value of f in the regularization term. k The meanings are different.

[0049] The newly trained regression tree f k (x) The prediction error is optimized by fitting the negative gradient of the current loss. The negative gradient of the i-th data in the k-th round is defined as follows: Through continuous iteration, the final predicted quality flow rate data is represented by a weighted sum of all decision trees. A well-trained nonlinear mass flow rate prediction model is obtained;

[0050] Step 4: Model Accuracy Assessment

[0051] The preprocessed powdered fuel scattering image is input into the trained nonlinear mass flow rate prediction model, which outputs the predicted mass flow rate of the powdered fuel. The predicted mass flow rate is then compared with the real-time mass flow rate considering the label to evaluate the accuracy of the prediction. The average error is used as the basis for accuracy judgment. If the average error is ≤5%, the prediction result of the nonlinear mass flow rate prediction model is considered accurate. Otherwise, the two hyperparameters γ and λ in the objective function are adjusted until the requirements are met.

[0052] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0053] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A powdered fuel mass flow rate prediction model based on physical information and data-driven methods, characterized in that: It includes a data acquisition system and a computer system. The data acquisition system includes a laser system, an optical modulation and shaping system, a signal detection system, and a camera system. The computer system includes an image preprocessing module, a feature extraction module, and an XGBoost regression prediction module. The laser system generates a laser beam by determining the wavelength based on the type of powdered fuel in the combustion chamber. The optical modulation and shaping system is arranged between the laser system and the engine to modulate the laser beam; The signal detection system is positioned behind the area to be measured to suppress background light noise; The camera system acquires scattered images of powdered fuel particles in the combustion chamber, and combines the particle density and particle flow velocity to obtain the real-time mass flow rate of the scattered images. The image preprocessing module performs cropping preprocessing on the acquired scattering image; The feature extraction module learns the image features of the scattering image through a deep belief network; The XGBoost regression prediction module considers physical constraint parameters and inputs the learned image features and real-time mass flow rate data into the XGBoost model for training and calculation to construct a nonlinear mass flow rate prediction model. The specific implementation steps are as follows: Step 1: Collection and Processing of Raw Data Multiple sets of scattered images of powdered fuel were acquired using a camera system. The particle density of the powdered fuel was measured based on the extinction principle. The particle velocity was measured using PIV and optical flow algorithms. The real-time mass flow rate of the scattered images was obtained by combining the particle density and particle velocity. The scattered images were labeled with physical constraint parameters as raw data. Step 2: Extracting deep image features from the data The scattering image is cropped into an image format suitable for use as input to deep learning. Image features are extracted through a deep belief network without considering labels. The network consists of multiple stacked RBMs. Through layer-by-layer training, the features learned by each RBM are used as input to the next layer. Step 3: Construct a nonlinear mass flow rate prediction model The extracted image features, real-time quality flow rate considering labels, and physical constraint parameters form a dataset as input variables, and the predicted quality flow rate is used as the output variable. This dataset is trained using an XGBoost model with the following objective function: Where n is the total number of training data points, and K is the total number of decision trees trained. For loss function, y i For the i-th real-time quality flow rate data, For the i-th predicted mass flow rate data, Ω(f k ) represents the regularization term. f k For the decision tree trained in the kth round, T k w represents the number of leaf nodes in the decision tree trained in the kth round. j Let γ be the weight of the j-th leaf node, and λ be the hyperparameters. The XGBoost model optimizes in a forward stepwise manner, training a new regression tree f in each round. k (x) is used to optimize the residuals from the previous round, then the predicted value for the k-th round is... f is the mass flow rate predicted in the first k-1 rounds. k (x i ) represents the newly trained decision tree in the kth round, paired with input x. i The predicted value, x i Let be the image features and physical constraint parameters in the i-th set of data; The newly trained regression tree f k (x) Optimize the prediction error by fitting the negative gradient of the current loss, where the negative gradient of the i-th data in the k-th round is... Through continuous iteration, the final predicted quality flow rate data is represented by a weighted sum of all decision trees. A well-trained nonlinear mass flow rate prediction model is obtained; Step 4: Model Accuracy Assessment The preprocessed scattering image of the powdered fuel is input into the trained nonlinear mass flow rate prediction model, which outputs the predicted mass flow rate of the powdered fuel. The prediction accuracy is evaluated by comparing the predicted mass flow rate with the real-time mass flow rate considering the label. If the requirements are not met, the hyperparameters γ and λ in the objective function are adjusted.

2. The powdered fuel mass flow rate prediction model based on physical information and data-driven according to claim 1, characterized in that: The physical constraint parameters include the engine's piston movement speed, bore size, and particle size of the powdered fuel.

3. The powder fuel mass flow rate prediction model based on physical information and data-driven according to claim 1, characterized in that: The RBM consists of a visible layer and a hidden layer, and the weight between any two connected neurons in the visible and hidden layers is w. ij The neuron states in the visible and hidden layers are vi and hj, respectively. The energy function of the RBM is defined as: E(v,h)=-∑a i v i -∑b j h j -∑v i w ij h j , where a i and b j This is a bias term.

4. The powdered fuel mass flow rate prediction model based on physical information and data-driven method according to claim 1, characterized in that: The accuracy is assessed using the average error as the criterion. If the average error is ≤5%, the prediction results of the nonlinear mass flow rate prediction model are considered to meet the requirements.

Images