A Deep Learning-Based Method and System for Predicting Cavitation Intensity in Cavitation Reactors
By using a deep learning-based neural network model, a nonlinear relationship between the shape factor of the cavitation reactor and the cavitation intensity is established, which solves the problem of inaccurate cavitation intensity prediction in existing technologies and achieves high-precision cavitation intensity prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2026-04-03
AI Technical Summary
Existing technologies cannot effectively capture the nonlinear relationship between cavitation reactor shape factors and cavitation intensity, resulting in inaccurate cavitation intensity prediction.
A deep learning-based approach is adopted to construct the shape space of the cavitation reactor, train it with a neural network, establish a nonlinear relationship between the shape factors of the cavitation reactor and the cavitation intensity, and predict the cavitation intensity.
It achieves high-precision cavitation intensity prediction, reduces prediction error and training time, and improves the accuracy and efficiency of the prediction model.
Smart Images
Figure CN119397904B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of cavitation technology, and in particular relates to a method and system for predicting cavitation intensity in cavitation reactors based on deep learning. Background Technology
[0002] The current global energy shortage and environmental pollution necessitate the exploration of innovative, energy-saving, and environmentally friendly process intensification technologies. Fluid dynamic cavitation technology is an emerging process intensification method that has been widely applied in various fields, including wastewater treatment, emulsification, lignocellulosic biomass pretreatment, flotation, biodiesel production, and food processing. Cavitation reactors are crucial for modern industrial applications. Cavitation technology refers to the phase change process of a liquid from the liquid phase to the gas phase when the static pressure of the liquid is lower than its saturated vapor pressure. The collapse of cavitation bubbles generates extreme conditions, including localized high temperatures up to 5000K, pressures reaching 1000 bar, and strong oxidation and reduction effects. Common types of cavitation reactors include orifice plates, Venturi reactors, eddy current diodes, and rotary reactors.
[0003] The prediction of cavitation intensity in cavitation reactors and the influence of reactor shape factors on cavitation intensity are typically analyzed using either the controlled variable method or the surface response method. The controlled variable method aims to eliminate the influence of other factors on cavitation intensity by controlling the shape parameters in the cavitation reactor, thereby obtaining the relationship between the reactor shape factors and cavitation intensity. The surface response method mainly involves designing appropriate experimental points and factor levels, collecting data, and then using a multiple quadratic regression equation to fit the functional relationship between shape factors and cavitation intensity. However, neither of these methods can capture the nonlinear relationship between reactor shape factors and cavitation intensity, nor can they establish the influence law of specific combinations of shape factors on cavitation intensity. Summary of the Invention
[0004] To overcome the shortcomings of the prior art, this invention provides a method and system for predicting cavitation intensity in cavitation reactors based on deep learning, which establishes a nonlinear relationship between the shape factors of the cavitation reactor and the cavitation intensity, thereby enabling cavitation intensity prediction.
[0005] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:
[0006] The first aspect of this invention provides a method for predicting the cavitation intensity of a cavitation reactor based on deep learning.
[0007] A deep learning-based method for predicting cavitation intensity in a cavitation reactor, comprising:
[0008] Obtain the shape parameters of the cavitation reactor to be predicted;
[0009] The shape parameters are input into the trained intensity prediction model to predict the cavitation intensity.
[0010] The prediction results were interpreted to obtain the influence of the interaction between different shape parameters on the cavitation intensity;
[0011] The intensity prediction model is trained using a training set obtained from numerical simulation. The training set is constructed by randomly sampling the shape parameters of the cavitation reactor to construct the shape space of the cavitation reactor. Numerical simulation is then performed on the shape space of the cavitation reactor to obtain the cavitation intensity. The sampled shape parameters and the obtained cavitation intensity are used as a data item to form the training set.
[0012] Furthermore, the cavitation reactor includes a venturi tube and a rotary cavitation reactor;
[0013] The shape parameters of the Venturi tube include the contraction angle, throat diameter, throat length, expansion angle, and inlet and outlet diameters;
[0014] The shape parameters of the rotary cavitation reactor include cylinder diameter, rotor spacing, cylinder depth, and rotor tilt angle.
[0015] Furthermore, the cavitation intensity parameters of the Venturi tube include the pressure drop at the inlet and outlet of the Venturi tube and the time-averaged cavitation volume;
[0016] The cavitation intensity parameter of the rotary cavitation reactor includes the cavitation generation rate.
[0017] Furthermore, the intensity prediction model employs a neural network, including an input layer, a hidden layer, and an output layer;
[0018] The number of nodes in the input layer is the same as the number of shape parameters, and each node represents a specific shape parameter.
[0019] The number of nodes in the output layer is the same as the number of cavitation intensity parameters, with each node representing a specific cavitation intensity parameter.
[0020] Furthermore, the hidden layer is connected according to the influence of specific shape parameter combinations on cavitation intensity, and is activated by the ReLU function before being connected.
[0021] Furthermore, the construction of the cavitation reactor shape space is based on randomly sampled shape parameters to model the cavitation reactor;
[0022] The cavitation intensity was obtained by numerically simulating the modeled cavitation reactor using the commercial software Fluent.
[0023] Furthermore, the method for obtaining the influence of the interaction between different shape parameters on the cavitation intensity involves retaining the network structure with a specific combination of shape parameters, deleting others, and then randomly sampling the nodes of the input layer to obtain the influence of the interaction between different shape parameters on the cavitation intensity.
[0024] A second aspect of the present invention provides a deep learning-based system for predicting the cavitation intensity of a cavitation reactor.
[0025] A deep learning-based cavitation intensity prediction system for cavitation reactors includes an acquisition module, a prediction module, and an interpretation module.
[0026] The acquisition module is configured to acquire the shape parameters of the cavitation reactor to be predicted.
[0027] The prediction module is configured to input shape parameters into a trained intensity prediction model to predict cavitation intensity.
[0028] The interpretation module is configured to interpret the prediction results and obtain the influence of the interaction between different shape parameters on the cavitation intensity.
[0029] The intensity prediction model is trained using a training set obtained from numerical simulation. The training set is constructed by randomly sampling the shape parameters of the cavitation reactor to construct the shape space of the cavitation reactor. Numerical simulation is then performed on the shape space of the cavitation reactor to obtain the cavitation intensity. The sampled shape parameters and the obtained cavitation intensity are used as a data item to form the training set.
[0030] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a deep learning-based method for predicting cavitation intensity in a cavitation reactor as described in the first aspect of the present invention.
[0031] The fourth aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in a deep learning-based method for predicting cavitation intensity in a cavitation reactor as described in the first aspect of the present invention.
[0032] The above one or more technical solutions have the following beneficial effects:
[0033] This invention randomly samples the shape parameters of the cavitation reactor to construct a cavitation reactor shape space, performs numerical simulation on the cavitation reactor shape space to obtain the cavitation intensity, thereby constructing a training set, training the neural network, and establishing a nonlinear relationship between the cavitation reactor shape factors and the cavitation intensity, thus realizing the prediction of cavitation intensity.
[0034] The neural network used in this invention differs from traditional neural network structures. The input layer and hidden layers are not fully connected; instead, they are connected separately based on the influence of specific shape parameter combinations on cavitation intensity. Furthermore, the hidden layer consists of two layers, with the number of nodes in each layer related to the input and output layers. On one hand, this gives the number of nodes in the hidden layer a specific meaning, facilitating the influence of specific shape parameter combinations on cavitation intensity. On the other hand, this structural characteristic closely resembles real-world conditions, enabling better convergence of the neural network training.
[0035] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0036] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0037] Figure 1 This is a diagram showing the shape parameters of the venturi tube in the first embodiment.
[0038] Figure 2 This is a shape space diagram of the first embodiment.
[0039] Figure 3 This is a diagram of the neural network structure of the first embodiment.
[0040] Figure 4 A neural network structure diagram for a specific combination of shape parameters in the first embodiment.
[0041] Figure 5 The diagram shows the effect of a specific combination of shape parameters on cavitation intensity in the first embodiment. Detailed Implementation
[0042] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0043] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0044] Example 1
[0045] One embodiment of this disclosure provides a deep learning-based method for predicting the cavitation intensity of a cavitation reactor, used for predicting the cavitation intensity of a venturi tube.
[0046] Venturi tubes are commonly used cavitation reactors; their shape parameters are as follows: Figure 1 As shown, the main components include the contraction angle β1 and the throat diameter d. th Throat length l th The expansion angle β2 and the inlet and outlet diameters d are considered. For the Venturi tube, this embodiment only considers the first four shape parameters. The cavitation intensity parameters are selected from the inlet and outlet pressure drop Δp and the time-averaged cavitation volume V of the Venturi tube.
[0047] The shape parameters of the venturi tube are randomly sampled, and the cavitation reactor is modeled based on these randomly sampled shape parameters, thus obtaining... Figure 2 The shape space shown, Figure 2 It concerns the contraction angle β1, expansion angle β2, and throat length l. th and throat diameter d th The relationship between them.
[0048] The flow of 100 venturi tubes was simulated using the Fluent platform to obtain their cavitation intensity parameters. After standardizing the shape parameters and cavitation intensity parameters, the shape parameters and cavitation intensity were treated as a single data item to construct a dataset. The dataset was divided into a training set and a validation set. For example, 90 tubes were used as the training set for the neural network and 10 tubes were used as the validation set, with consistent boundary conditions, mesh quality, and other conditions.
[0049] The neural network used in the intensity prediction model, such as Figure 3 As shown, it consists of an input layer, a hidden layer, and an output layer. The input layer has 4 nodes, each representing a shape parameter; the output layer has 2 nodes, each representing a cavitation intensity parameter.
[0050] The hidden layer consists of two layers, with the first hidden layer containing 2 nodes. n -1, the second layer has m nodes, where n represents the number of shape parameters and m represents the number of cavitation intensity parameters. The nodes in the first hidden layer are not fully connected to the input layer (shape parameters), but are connected to specific input nodes based on single-factor, two-factor, ..., n-factor interactions, representing the influence of single-factor, two-factor, ..., n-factor interactions on cavitation intensity, respectively. Each node in the first hidden layer is connected based on the influence of a specific combination of shape parameters on cavitation intensity. Each node in the second hidden layer is fully connected to the first hidden layer and correspondingly connected to the output layer. The two-layer structure has two activation functions in the hidden layers, capable of fitting nonlinear relationships.
[0051] In this embodiment, the number of shape parameters is 4, so the first hidden layer has 15 nodes, representing the influence of single shape parameter factors (4 nodes), dual shape parameter factors (6 nodes), triple shape parameter factors (4 nodes), and quadruple shape parameter factors (1 node) on cavitation intensity. The nodes for single shape parameter factors are connected to one shape parameter node corresponding to the input layer, the nodes for dual shape parameter factors are connected to two shape parameter nodes corresponding to the input layer, and the same applies to triple and quadruple factors. The activation function used in the hidden layer is ReLU, and the output obtained by each connected node is:
[0052] Z [l] =W [l] a [l] +b [l]
[0053] Where W and b are the bias parameter matrices of the weights to be trained, l represents the number of layers in the neural network, and a [l] =δ(z) [l -1] ), where δ is the activation function, z [l-1] The first hidden layer has two nodes, which are fully connected to the nodes in the first hidden layer and connected to the two cavitation intensity parameters of the output layer. The two-layer structure has two activation functions in its hidden layers, thus enabling the fitting of nonlinear relationships.
[0054] The training data is fed into this neural network for training, and then the mean squared error is used as the criterion for evaluating prediction performance, thereby generating a high-precision cavitation intensity prediction model. Because this structure is more in line with the actual impact, the prediction error is 22% lower than that of traditional neural network structures, and the training time is reduced by 14%.
[0055] The trained cavitation intensity prediction model is interpreted to obtain the influence of the interaction between different shape parameters on cavitation intensity. For example, specific nodes in the hidden layer (i.e., nodes with two shape parameters) are retained, while others are deleted, resulting in... Figure 4 The model shown, Figure 4 Figures (a), (b), and (c) show the model structure diagrams with different dual-shape parameter nodes, respectively. Then, by randomly sampling the nodes of the input layer and finally predicting the cavitation intensity, the following can be obtained: Figure 5 The contraction angle β1, expansion angle β2, and throat length l are shown. th and throat diameter d th The influence of pairwise bi-shape parameter combinations among these four shape parameters on cavitation intensity (i.e., the pressure drop Δp at the inlet and outlet of the Venturi tube and the time-averaged cavitation volume V).
[0056] Example 2
[0057] One embodiment of this disclosure provides a deep learning-based method for predicting the cavitation intensity of a cavitation reactor, used for predicting the cavitation intensity of a rotating cavitation reactor.
[0058] The rotary cavitation reactor is a cavitation reactor with high cavitation performance. The cavitation unit is selected as a combination of cylinders and spheres. The shape parameters mainly include cylinder diameter, rotor spacing, cylinder depth and rotor tilt angle. In this embodiment, only the first three shape parameters are considered. The cavitation intensity parameter is selected as the cavitation generation rate.
[0059] The shape parameters of the rotating cavitation reactor are randomly sampled, and the cavitation reactor is modeled based on the randomly sampled shape parameters to obtain the shape space.
[0060] The flow of 200 rotating cavitation reactors was simulated using the Fluent platform to obtain their cavitation intensity parameters. After standardizing the shape parameters and cavitation intensity parameters, the shape parameters and cavitation intensity were treated as a single data item to construct a dataset. The dataset was divided into training and validation sets. For example, 200 samples were used as the training set for a neural network, and 20 samples were used as the validation set, with consistent boundary conditions, mesh quality, and other conditions.
[0061] The intensity prediction model uses a neural network consisting of an input layer, a hidden layer, and an output layer. The input layer has three nodes, each representing a shape parameter, and the output layer has one node, each representing a cavitation intensity parameter.
[0062] The hidden layer adopts a two-layer structure. The first hidden layer has 7 nodes, representing the influence of single shape parameter factors (3 nodes), dual shape parameter factors (3 nodes), and triple shape parameter factors (1 node) on cavitation intensity. The nodes of the single shape parameter factor are connected to one shape parameter node of the input layer, the nodes of the dual shape parameter factor are connected to two shape parameter nodes of the input layer, and the same applies to the triple factor factor. The activation function used in the hidden layer is ReLU. The two-layer structure of the hidden layer has two activation functions, so it can fit nonlinear relationships.
[0063] The training data is fed into the neural network for training, and then the mean squared error is used as the criterion for evaluating prediction performance, thereby creating a high-precision cavitation intensity prediction model; the prediction error is 20% lower than that of traditional neural network structures, and the training time is reduced by 12%.
[0064] The trained cavitation intensity prediction model is interpreted to obtain the influence of the interaction between different shape parameters on cavitation intensity. For example, specific nodes in the hidden layer are retained and others are deleted to obtain a model for a specific combination of shape parameters. Then, nodes in the input layer are randomly sampled, and finally, cavitation intensity prediction is performed to obtain the influence of a specific combination of shape parameters on cavitation intensity.
[0065] Example 3
[0066] One embodiment of this disclosure provides a deep learning-based cavitation intensity prediction system for cavitation reactors, including an acquisition module, a prediction module, and an interpretation module:
[0067] The acquisition module is configured to acquire the shape parameters of the cavitation reactor to be predicted.
[0068] The prediction module is configured to input shape parameters into a trained intensity prediction model to predict cavitation intensity.
[0069] The interpretation module is configured to interpret the prediction results and obtain the influence of the interaction between different shape parameters on the cavitation intensity.
[0070] The intensity prediction model is trained using a training set obtained from numerical simulation. The training set is constructed by randomly sampling the shape parameters of the cavitation reactor to construct the shape space of the cavitation reactor. Numerical simulation is then performed on the shape space of the cavitation reactor to obtain the cavitation intensity. The sampled shape parameters and the obtained cavitation intensity are used as a data item to form the training set.
[0071] Example 4
[0072] The purpose of this embodiment is to provide a computer-readable storage medium.
[0073] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a deep learning-based method for predicting cavitation intensity in a cavitation reactor as described in Embodiment 1 of this disclosure.
[0074] Example 5
[0075] The purpose of this embodiment is to provide an electronic device.
[0076] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in a deep learning-based method for predicting cavitation intensity in a cavitation reactor as described in Embodiment 1 of this disclosure.
[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting cavitation intensity in a cavitation reactor based on deep learning, characterized in that, include: Obtain the shape parameters of the cavitation reactor to be predicted; The shape parameters are input into the trained intensity prediction model to predict the cavitation intensity. The intensity prediction model uses a neural network, including an input layer, a hidden layer, and an output layer; The number of nodes in the input layer is the same as the number of shape parameters, and each node represents a different shape parameter. The number of nodes in the output layer is the same as the number of cavitation intensity parameters, and each node represents a different cavitation intensity parameter; The hidden layer consists of two layers, with the first hidden layer containing 2 nodes. n -1, the second layer has m nodes, where n represents the number of shape parameters and m represents the number of cavitation intensity parameters. The nodes of the first hidden layer are not fully connected to the input layer, but are connected to the input nodes according to single-factor, two-factor, ..., n-factor factors, respectively, representing the influence of single-factor influence, two-factor interaction influence, ..., n-factor interaction influence on cavitation intensity. Each node of the first hidden layer is connected according to the influence of different combinations of shape parameters on cavitation intensity. Each node of the second hidden layer is fully connected to the first hidden layer and correspondingly connected to the output layer. The two-layer structure has two activation functions in the hidden layers, which can fit nonlinear relationships. The hidden layer is connected according to the influence of the combination of shape parameters on the cavitation intensity, and is activated by the ReLU function before being connected. The prediction results were interpreted to obtain the influence of the interaction between different shape parameters on the cavitation intensity; The intensity prediction model is trained using a training set obtained from numerical simulation. The training set is constructed by randomly sampling the shape parameters of the cavitation reactor to construct the shape space of the cavitation reactor. Numerical simulation is then performed on the shape space of the cavitation reactor to obtain the cavitation intensity. The sampled shape parameters and the obtained cavitation intensity are used as a data item to form the training set.
2. The method for predicting cavitation intensity in a cavitation reactor based on deep learning as described in claim 1, characterized in that, The cavitation reactor includes a venturi tube and a rotary cavitation reactor; The shape parameters of the Venturi tube include the contraction angle, throat diameter, throat length, and expansion angle; The shape parameters of the rotary cavitation reactor include cylinder diameter, rotor spacing, cylinder depth, and rotor tilt angle.
3. The method for predicting cavitation intensity in a cavitation reactor based on deep learning as described in claim 2, characterized in that, The cavitation intensity parameters of the Venturi tube include the pressure drop at the inlet and outlet of the Venturi tube and the time-averaged cavitation volume. The cavitation intensity parameter of the rotary cavitation reactor includes the cavitation generation rate.
4. The method for predicting cavitation intensity in a cavitation reactor based on deep learning as described in claim 1, characterized in that, The construction of the cavitation reactor shape space is based on the modeling of the cavitation reactor using randomly sampled shape parameters; The cavitation intensity was obtained by numerically simulating the modeled cavitation reactor using the commercial software Fluent.
5. The method for predicting cavitation intensity in a cavitation reactor based on deep learning as described in claim 1, characterized in that, The method for obtaining the influence of the interaction between different shape parameters on cavitation intensity involves retaining the network structure of the shape parameter combination, deleting others, and then randomly sampling the nodes of the input layer to obtain the influence of the interaction between different shape parameters on cavitation intensity.
6. A deep learning-based cavitation intensity prediction system for a cavitation reactor, employing the method described in claim 1, characterized in that, It includes an acquisition module, a prediction module, and an explanation module: The acquisition module is configured to acquire the shape parameters of the cavitation reactor to be predicted. The prediction module is configured to input shape parameters into a trained intensity prediction model to predict cavitation intensity. The interpretation module is configured to interpret the prediction results and obtain the influence of the interaction between different shape parameters on the cavitation intensity. The intensity prediction model is trained using a training set obtained from numerical simulation. The training set is constructed by randomly sampling the shape parameters of the cavitation reactor to construct the shape space of the cavitation reactor. Numerical simulation is then performed on the shape space of the cavitation reactor to obtain the cavitation intensity. The sampled shape parameters and the obtained cavitation intensity are used as a data item to form the training set.
7. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. Processor, for executing the computer-readable instructions; When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-5.
8. A storage medium, characterized in that, The computer-readable instructions are stored non-temporarily, wherein when the computer-readable instructions are executed by a computer, the method described in any one of claims 1-5 is performed.
Citation Information
Patent Citations
Air cylinder cover structure size determination method based on BP neural network
CN118428006A