Gas-liquid interface accurate positioning method based on artificial intelligence

Abstract

The invention discloses a gas-liquid interface accurate positioning method based on artificial intelligence. The method comprises the steps of acquiring gas-liquid interface positioning data, optimizing original data, constructing a multi-layer gas-liquid interface positioning model, enhancing the performance of the gas-liquid interface positioning model and performing real-time accurate positioning on a gas-liquid interface. The invention relates to the technical field of data processing, in particular to a gas-liquid interface accurate positioning method based on artificial intelligence, according to the scheme, a multi-layer structure model is innovatively designed, and the model improves the positioning accuracy, real-time performance and adaptability of a gas-liquid interface by comprehensively analyzing image data, time sequence data and environment data; an image data processing method of a multi-window self-attention mechanism and a multi-window offset self-attention mechanism is provided, and the accuracy of gas-liquid interface positioning is improved; the algorithm for obtaining the optimal parameters of the model is improved by adopting a cosine transform initialization method, a dynamic spiral search strategy and a dynamic adjustment item factor, an optimal parameter combination is obtained, and the positioning accuracy of the gas-liquid interface is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and specifically refers to a method for accurately positioning the gas-liquid interface based on artificial intelligence. Background Art

[0002] The gas-liquid interface refers to the interface where gas and liquid contact. It widely exists in nature and various industrial applications. The accurate positioning of the gas-liquid interface is crucial for many scientific and engineering problems. Therefore, a method for accurately positioning the gas-liquid interface based on artificial intelligence has emerged. This method automatically identifies and accurately locates the position of the gas-liquid interface by applying artificial intelligence technology. Through intelligent analysis of the images or sensor data of the gas-liquid interface, it can achieve high-precision positioning of the gas-liquid interface position, providing more advanced technical support for related industries and scientific research.

[0003] However, there are technical problems in traditional gas-liquid interface accurate positioning methods, such as untimely dynamic response, unconsidered environmental factors, and single data source analysis, which lead to inaccurate gas-liquid interface positioning accuracy; there are technical problems in traditional gas-liquid interface accurate positioning methods in image data analysis, such as insufficient extraction of global information and capture of deep relationship of features, which lead to inaccurate gas-liquid interface positioning accuracy; in existing multi-level gas-liquid interface positioning models, there are technical problems of improper setting of built-in parameters, which lead to inaccurate output results of the models. Summary of the Invention

[0004] In view of the above situation, to overcome the defects of the prior art, the present invention provides an accurate positioning method for gas-liquid interfaces based on artificial intelligence. Aiming at the technical problems of untimely dynamic response, unconsidered environmental factors, and single data source analysis in the traditional accurate positioning method for gas-liquid interfaces, which lead to inaccurate positioning accuracy of gas-liquid interfaces, this solution innovatively designs a multi-level structure model. This model solves the problems of accuracy and adaptability in the gas-liquid interface positioning of traditional methods by comprehensively analyzing image data, time-series data, and environmental data. The image data is processed by the image data processing layer to provide the initial position of the gas-liquid interface. The time-series data is processed by the dynamic time-series data processing layer to capture the dynamic trend of the gas-liquid interface changing over time in real time and accurately track the position of the interface. The environmental data is processed by the environmental data processing layer to be able to adapt to different complex environments. Through the effective combination of the above three-layer structure, the positioning accuracy, real-time performance, and adaptability of the gas-liquid interface are significantly improved. Aiming at the technical problem of insufficient extraction of global information and capture of deep relationships between features in the image data analysis of the traditional accurate positioning method for gas-liquid interfaces, which leads to inaccurate positioning accuracy of gas-liquid interfaces, this solution innovatively proposes an image data processing method based on a multi-window self-attention mechanism and a multi-window offset self-attention mechanism, enhancing the adaptability of gas-liquid interface positioning in complex and dynamically changing environments and improving the positioning accuracy of gas-liquid interfaces. Through the multi-window self-attention mechanism, accurate weighting of local features is ensured, and global information can be effectively extracted, solving the problem that traditional methods fail to consider local dependencies in global feature fusion, thereby improving the positioning accuracy of gas-liquid interfaces. Through the multi-window offset self-attention mechanism, the window selection range is optimized, and deep relationships between features are effectively captured, further improving the accuracy of gas-liquid interface positioning. Aiming at the technical problem that in the existing multi-level gas-liquid interface positioning model, improper setting of built-in parameters leads to inaccurate model output results, this solution uses a cosine transform initialization method, a dynamic spiral search strategy, and a dynamic adjustment term factor to improve the algorithm for obtaining the optimal parameters of the model, obtains the optimal parameter combination, improves the accuracy of the model output results, and thus improves the precision of gas-liquid interface positioning.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an accurate positioning method for gas-liquid interfaces based on artificial intelligence, and this method includes the following steps:

[0006] Step S1: Obtain gas-liquid interface positioning data;

[0007] Step S2: Optimize and process the original data;

[0008] Step S3: Build a multi-level gas-liquid interface positioning model;

[0009] Step S4: Enhance the performance of the gas-liquid interface positioning model;

[0010] Step S5: Real-time and precise positioning of the gas-liquid interface.

[0011] Further, in step S1, the acquisition of the gas-liquid interface positioning data specifically refers to collecting the original gas-liquid interface positioning data through an industrial camera and sensors; the original gas-liquid interface positioning data includes historical gas-liquid interface data and real-time gas-liquid interface data; both the historical gas-liquid interface data and the real-time gas-liquid interface data include gas-liquid interface image data and environmental data; the historical gas-liquid interface data also includes historical gas-liquid interface position data.

[0012] Further, in step S2, the optimization processing of the original data is used to optimize the original gas-liquid interface positioning data, specifically by performing image data optimization processing, environmental data optimization processing, and data time synchronization on the original gas-liquid interface positioning data to obtain optimized gas-liquid interface positioning data;

[0013] Step S21: Image data optimization processing, which is used to obtain an optimized gas-liquid interface image, specifically including image denoising, image enhancement, and image registration;

[0014] Step S22: Environmental data optimization processing, specifically including data cleaning and standardization processing;

[0015] Step S23: Data time synchronization, which is used to analyze the relationship between the gas-liquid interface and environmental changes; specifically, according to the timestamps of the image data and environmental data, each image data frame is synchronized with the environmental data in the corresponding time period.

[0016] Further, in step S3, the construction of the multi-level gas-liquid interface positioning model is used to identify the position of the gas-liquid interface through the model, specifically by designing an image data processing layer, a dynamic time series data processing layer, an environmental data processing layer, and a model output layer to obtain a multi-level gas-liquid interface positioning model; including the following steps:

[0017] Step S31: Design the image data processing layer, which is used to obtain the preliminary position of the gas-liquid interface from the gas-liquid interface image data, specifically including the following steps:

[0018] Step S311: Gas-liquid interface image feature extraction, specifically including the following steps:

[0019] Step S3111: Gas-liquid interface local feature extraction, which is used to extract local spatial features from the gas-liquid interface image, specifically by using a convolutional kernel with a size of to perform convolution processing on the optimized gas-liquid interface image, adjust the number of channels of the input feature map, and then use The convolutional kernel further extracts local spatial features. After each convolution operation is completed, batch normalization is used to standardize the convolution output, and then the ReLU activation function is used to perform a non-linear transformation on the convolution output to obtain the local feature map of the gas-liquid interface. The formula used is as follows:

[0020] ;

[0021] In the formula, represents the value of the output feature map after the (l + 1)-th layer of convolution operation, represents the ReLU activation function, represents the normalization operation, represents the weight value of the (l + 1)-th layer of convolutional kernel, represents the value at the position of in the output feature map after the l-th layer of convolution operation, m represents the number of channels of the input feature map, 、 and represent the size of the convolutional kernel, 、 and represent the offset of the convolutional kernel in the spatial dimension, 、 and represent the stride when the convolutional kernel slides in each dimension, represents the bias term of the (l + 1)-th layer of convolution operation;

[0022] Step S3112: Extraction of macroscopic features of the gas-liquid interface, which is used to identify the macroscopic features of the gas-liquid interface. Specifically, dilated convolution is introduced to perform convolution processing on the optimized image of the gas-liquid interface. Batch normalization is performed on the feature map after dilated convolution processing, and the ReLU activation function is used to introduce non-linear transformation to the processed feature map, and finally the macroscopic feature map of the gas-liquid interface is obtained;

[0023] Step S3113: Feature splicing, specifically, splicing the local feature map of the gas-liquid interface and the macroscopic feature map of the gas-liquid interface to obtain the comprehensive feature map of the gas-liquid interface ;

[0024] Step S312: Obtaining global dependencies of the gas-liquid interface, specifically including the following steps:

[0025] Step S3121: Preliminary fusion of gas-liquid interface features, specifically, first performing multi-window self-attention calculation on the comprehensive feature map of the gas-liquid interface to obtain the window merged feature , performing layer normalization on , inputting it into the MLP module for feature enhancement, and finally obtaining the preliminarily fused feature map through residual connection , the multi-window self-attention calculation is specifically based on the gas-liquid interface comprehensive feature map size, divide it into multiple non-overlapping small windows with a size of small windows, , and respectively represent the depth, height and width of the window. Perform self-attention calculation operations on each window to obtain the weighted features of each window, merge the weighted features of each window, and output the features through residual connection; the formula used is as follows:

[0026] ;

[0027] ;

[0028] In the formula, represents the preliminary fusion feature, represents the layer normalization function, represents the multi-window self-attention calculation function, represents the multi-layer perceptron function, represents the window merged feature;

[0029] Step S3122: Gas-liquid interface feature depth fusion, specifically, first perform multi-window offset self-attention calculation on the preliminary fusion feature to obtain the offset window merged feature , perform layer normalization processing on , and input it into the MLP module for feature enhancement. Finally, through residual connection, obtain the preliminary fusion feature map , the window offset self-attention calculation is specifically based on the gas-liquid interface comprehensive feature map size, divide it into multiple non-overlapping small windows with a size of small windows, and at the same time set the offset , perform window offset, , and respectively represent the depth, height and width offsets of the window. Perform self-attention calculation on each offset window to obtain the weighted features of each offset window, merge the weighted features of each offset window, and output the features through residual connection; the formula used is as follows:

[0030] ;

[0031] ;

[0032] In the formula, represents the depth fusion feature, represents the window offset self-attention calculation function, Indicates the offset window merging feature;

[0033] Step S313: Obtain the preliminary position of the gas-liquid interface. The formula used is as follows:

[0034] ;

[0035] In the formula, Indicates the preliminary position of the gas-liquid interface, Indicates the output weight matrix of the preliminary position of the gas-liquid interface, Indicates the output bias term parameter of the preliminary position of the gas-liquid interface, Indicates the Sigmoid activation function;

[0036] Step S32: Design a dynamic time series data processing layer to track the change of the gas-liquid interface position over time and optimize the positioning result of the gas-liquid interface. Specifically, process the time series information of the historical gas-liquid interface position data through a long short-term memory network to obtain the dynamic time series features of the gas-liquid interface ; The formula used is as follows:

[0037] ;

[0038] In the formula, Indicates the unit operation function, Indicates the hidden state at the previous moment, Indicates the input historical gas-liquid interface position data at the current moment;

[0039] Step S33: Design an environmental data processing layer to optimize the positioning result of the gas-liquid interface through environmental data. Specifically, process the environmental data through a fully connected layer to obtain the environmental features of the gas-liquid interface ; The formula used is as follows:

[0040] ;

[0041] In the formula, Indicates the weight matrix of the environmental data, Indicates the bias term parameter of the environmental data, Indicates the environmental data;

[0042] Step S34: Design a model output layer to obtain the final position of the gas-liquid interface; The formula used is as follows:

[0043] ;

[0044] In the formula, Indicates the final position of the gas-liquid interface, Indicates the output weight matrix of the final position of the gas-liquid interface, Represents the offset term parameter of the output of the final position of the gas-liquid interface.

[0045] Further, in step S4, enhancing the performance of the gas-liquid interface positioning model specifically includes training a multi-level gas-liquid interface positioning model, obtaining the optimal hyperparameters of the model, and adjusting the hyperparameters of the model to obtain an optimized multi-level gas-liquid interface positioning model; it includes the following steps:

[0046] Step S41: Training the multi-level gas-liquid interface positioning model, specifically using the historical gas-liquid interface data to train the multi-level gas-liquid interface positioning model to obtain a trained multi-level gas-liquid interface positioning model;

[0047] Step S42: Obtaining the optimal hyperparameters of the model, used to obtain the optimal hyperparameter combination of the trained multi-level gas-liquid interface positioning model, specifically using an improved optimization algorithm to obtain the optimal hyperparameter combination, including the following steps:

[0048] Step S421: Initializing the position of the population individuals, specifically using the cosine transform initialization method to initialize the search individuals; the formula used is as follows:

[0049] ;

[0050] In the formula, represents the value of the i-th chaotic sequence, represents a random number uniformly distributed within the range of ; represents a random number uniformly distributed within the range of ; represents the initialization position of the i-th individual, and represent the search lower limit and upper limit respectively;

[0051] Step S422: Calculating the fitness value, specifically calculating the fitness value f i of the individuals in the population; taking the performance of the trained multi-level gas-liquid interface positioning model established based on the individual position as the fitness value of the individual, sorting the individuals from the best to the worst according to the fitness value, and obtaining the position of the individual with the highest current global fitness value ;

[0052] Step S423: The global search behavior of the individual, the formula specifically used is as follows:

[0053] ;

[0054] In the formula, t represents the current iteration number, represents the position of the i-th individual in the t-th generation population, represents the position of the i-th individual in the (t + 1)-th generation population, represents the position of the $i$-th individual in the population of the $(t - 1)$-th generation, represents the position of the individual with the worst fitness value in the population during the $t$-th iteration, represents a random number uniformly distributed within the range, represents a random number uniformly distributed within the range, represents the dynamic coefficient, represents the offset coefficient, and the value range is , represents the obstacle threshold, represents that no obstacle is encountered, represents that an obstacle is encountered;

[0055] Step S424: Optimal solution derivation behavior, specifically, the optimal solution derivation behavior is carried out through the dynamic spiral search strategy, and the formula used is as follows:

[0056] ;

[0057] ;

[0058] ;

[0059] In the formula, represents the position of the $i$-th derived individual in the population of the $(t + 1)$-th generation, represents the position of the $i$-th derived individual in the population of the $t$-th generation, represents the position of the individual with the best fitness value in the population during the $t$-th iteration, represents the lower boundary of the derived individual's position, represents the upper boundary of the derived individual's position, and respectively represent the maximum and minimum values of the solution space, and represent a random number uniformly distributed within the range, represents the constant for controlling path update, represents the maximum number of iterations;

[0060] Step S425: Individual local exploration behavior, specifically, the search behavior is adjusted through the dynamic adjustment term factor, and the formula used is as follows:

[0061] ;

[0062] ;

[0063] ;

[0064] In the formula, Represents the lower boundary of the individual's local exploration, Represents the upper boundary of the individual's local exploration, Represents a random number subject to a normal distribution, Represents between A random number uniformly distributed within the range, Represents a dynamic adjustment term factor for adjusting the individual's local exploration behavior;

[0065] Step S426: Competitive global perturbation behavior, and the specific formula used is as follows:

[0066] ;

[0067] In the formula, Represents between A random number uniformly distributed within the range;

[0068] Step S427: Obtain the optimal hyperparameter combination of the model. Specifically, when the individual fitness value f i Is higher than the fitness threshold and reaches the maximum number of iterations, terminate the search and obtain the global optimal position of the individual to obtain the optimal hyperparameter combination of the model;

[0069] Step S43: Adjust the hyperparameters of the model. Specifically, adjust the hyperparameters of the trained multi-level gas-liquid interface positioning model according to the optimal hyperparameter combination of the model to obtain an optimized multi-level gas-liquid interface positioning model.

[0070] Furthermore, in step S5, the real-time precise positioning of the gas-liquid interface specifically means inputting the real-time gas-liquid interface data into the optimized multi-level gas-liquid interface positioning model to obtain the real-time precise gas-liquid interface position result, realizing the continuous monitoring of the change trend of the interface.

[0071] The beneficial effects obtained by the present invention using the above solution are as follows:

[0072] (1) Aiming at the technical problems of untimely dynamic response, unconsidered environmental factors, and single data source analysis in the traditional gas-liquid interface precise positioning method, which leads to inaccurate gas-liquid interface positioning accuracy, this solution innovatively designs a multi-level structure model. This model solves the accuracy and adaptability problems of the traditional method in gas-liquid interface positioning by comprehensively analyzing image data, time series data, and environmental data. The image data processing layer processes the image data to provide the initial gas-liquid interface position, the dynamic time series data processing layer processes the time series data to capture the dynamic trend of the gas-liquid interface changing with time in real time and accurately track the position of the interface, and the environmental data processing layer processes the environmental data to be able to adapt to different complex environments. Through the effective combination of the above three-layer structure, the positioning accuracy, real-time performance, and adaptability of the gas-liquid interface are significantly improved.

[0073] (2) Aiming at the technical problems existing in the traditional precise positioning method of the gas-liquid interface in image data analysis, namely the insufficient extraction of global information and the inability to capture the deep relationship between features, which leads to inaccurate positioning accuracy of the gas-liquid interface. This solution innovatively proposes an image data processing method based on the multi-window self-attention mechanism and the multi-window offset self-attention mechanism, enhancing the adaptability of gas-liquid interface positioning in complex and dynamically changing environments and improving the positioning accuracy of the gas-liquid interface. Through the multi-window self-attention mechanism, precise weighting of local features is ensured, and global information can be effectively extracted, solving the problem that traditional methods fail to consider local dependencies in global feature fusion, thereby improving the positioning accuracy of the gas-liquid interface. Through the multi-window offset self-attention mechanism, the window selection range is optimized, and the deep relationship between features is effectively captured, further improving the accuracy of gas-liquid interface positioning.

[0074] (3) Aiming at the technical problem in the existing multi-level gas-liquid interface positioning model that improper setting of built-in parameters leads to inaccurate model output results, this solution uses the cosine transform initialization method, the dynamic spiral search strategy, and the dynamic adjustment term factor to improve the algorithm for obtaining the optimal parameters of the model, obtaining the optimal parameter combination, improving the accuracy of the model output results, and thus improving the precision of gas-liquid interface positioning. Brief Description of the Drawings

[0075] Figure 1 It is a schematic flowchart of the method for precise positioning of the gas-liquid interface based on artificial intelligence provided by the present invention;

[0076] Figure 2 It is a schematic flowchart of step S2;

[0077] Figure 3 It is a schematic flowchart of step S3;

[0078] Figure 4 It is a schematic flowchart of step S4;

[0079] Figure 5 It is a schematic flowchart of step S31;

[0080] Figure 6 It is a schematic flowchart of step S42;

[0081] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. Detailed Embodiments

[0082] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0083] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0084] Embodiment 1, refer to Figure 1 , the technical solution adopted by the present invention is as follows: The method for precise positioning of the gas-liquid interface based on artificial intelligence provided by the present invention includes the following steps:

[0085] Step S1: Obtain gas-liquid interface positioning data, specifically collect the original gas-liquid interface positioning data through an industrial camera and a sensor;

[0086] Step S2: Optimize the original data, which is used to optimize the original gas-liquid interface positioning data. Specifically, perform image data optimization processing, environmental data optimization processing, and data time synchronization on the original gas-liquid interface positioning data to obtain optimized gas-liquid interface positioning data;

[0087] Step S3: Build a multi-level gas-liquid interface positioning model, which is used to identify the position of the gas-liquid interface through the model. Specifically, design an image data processing layer, a dynamic time series data processing layer, an environmental data processing layer, and a model output layer to obtain a multi-level gas-liquid interface positioning model;

[0088] Step S4: Enhance the performance of the gas-liquid interface positioning model, specifically including training the multi-level gas-liquid interface positioning model, obtaining the optimal hyperparameters of the model, and adjusting the hyperparameters of the model to obtain an optimized multi-level gas-liquid interface positioning model;

[0089] Step S5: Real-time precise positioning of the gas-liquid interface, specifically input the real-time data into the optimized model to obtain the real-time precise gas-liquid interface position result.

[0090] Embodiment 2, refer to Figure 1, this embodiment is based on the above embodiment. In step S1, the acquisition of gas-liquid interface positioning data specifically refers to collecting the original gas-liquid interface positioning data through an industrial camera and a sensor; the original gas-liquid interface positioning data includes historical gas-liquid interface data and real-time gas-liquid interface data; both the historical gas-liquid interface data and the real-time gas-liquid interface data include gas-liquid interface image data and environmental data; the gas-liquid interface image data includes an upper gas-liquid interface image, a side gas-liquid interface image, and an oblique side gas-liquid interface image; the environmental data includes temperature data and pressure data; the historical gas-liquid interface data also includes historical gas-liquid interface position data.

[0091] Embodiment III, refer to Figure 1 and Figure 2 , this embodiment is based on the above embodiment. In step S2, the optimization processing of the original data is used to optimize the original gas-liquid interface positioning data, and specifically includes the following steps;

[0092] Step S21: Image data optimization processing, which is used to obtain an optimized gas-liquid interface image, and specifically includes image denoising, image enhancement, and image registration; the image denoising specifically refers to removing random noise and background interference in the image through a Gaussian filtering algorithm to ensure the clarity of the image; the image enhancement specifically refers to enhancing the visual features of the gas-liquid interface image through contrast enhancement, brightness adjustment, and edge enhancement techniques; the image registration is used to ensure the data consistency and accuracy of multi-view images, and specifically refers to performing an image alignment operation on the gas-liquid interface images taken from different angles through a feature point matching algorithm.

[0093] Step S22: Environmental data optimization processing, which specifically includes data cleaning and standardization processing; the data cleaning is to process missing values, outliers, and duplicate values in the environmental data; the standardization processing is to standardize the data based on the maximum-minimum normalization method.

[0094] Step S23: Data time synchronization, which is used to analyze the relationship between the gas-liquid interface and environmental changes; specifically, according to the timestamps of the image data and the environmental data, each image data frame is synchronized with the environmental data in the corresponding time period.

[0095] Embodiment IV, refer to Figure 1 and Figure 3 , this embodiment is based on the above embodiment. In step S3, the construction of the multi-level gas-liquid interface positioning model specifically includes the following steps;

[0096] Step S31: Design an image data processing layer, which is used to extract the position of the gas-liquid interface from the gas-liquid interface image data to obtain the preliminary position of the gas-liquid interface.

[0097] Step S32: Design a dynamic time-series data processing layer to track the temporal changes in the gas-liquid interface position and optimize the positioning result of the gas-liquid interface. Specifically, process the temporal information of the historical gas-liquid interface position data through a long short-term memory network to obtain the dynamic time-series features of the gas-liquid interface ; The formula used is as follows:

[0098] ;

[0099] In the formula, represents the unit operation function, represents the hidden state at the previous moment, represents the input historical gas-liquid interface position data at the current moment;

[0100] Step S33: Design an environmental data processing layer to optimize the positioning result of the gas-liquid interface through environmental data. Specifically, process the environmental data through a fully connected layer to obtain the environmental features of the gas-liquid interface ; The formula used is as follows:

[0101] ;

[0102] In the formula, represents the weight matrix of the environmental data, represents the bias term parameter of the environmental data, represents the environmental data;

[0103] Step S34: Design a model output layer to obtain the final position of the gas-liquid interface; The formula used is as follows:

[0104] ;

[0105] In the formula, represents the final position of the gas-liquid interface, represents the output weight matrix of the final position of the gas-liquid interface, represents the output bias term parameter of the final position of the gas-liquid interface.

[0106] By performing the above operations, aiming at the technical problems in the traditional precise positioning method of the gas-liquid interface, such as untimely dynamic response, unconsidered environmental factors, and single data source analysis, which lead to inaccurate positioning accuracy of the gas-liquid interface, this solution innovatively designs a multi-level structure model. This model solves the accuracy and adaptability problems of the traditional method in gas-liquid interface positioning by comprehensively analyzing image data, time-series data, and environmental data. The image data processing layer processes the image data to provide the initial position of the gas-liquid interface. The dynamic time-series data processing layer processes the time-series data to capture the dynamic trend of the gas-liquid interface changing with time in real time and accurately track the position of the interface. The environmental data processing layer processes the environmental data to be able to adapt to different complex environments. Through the effective combination of the above three-layer structure, the positioning accuracy, real-time performance, and adaptability of the gas-liquid interface are significantly improved.

[0107] Example Five, refer to Figure 1 and Figure 5 , based on the above example, in step S31, the design of the image data processing layer specifically includes the following steps:

[0108] Step S311: Extract the image features of the gas-liquid interface, which specifically includes the following steps:

[0109] Step S3111: Extract the local features of the gas-liquid interface, which is used to extract the local spatial features from the gas-liquid interface image. Specifically, a convolution kernel with a size of is used to perform convolution processing on the optimized image of the gas-liquid interface to adjust the number of channels of the input feature map. Then, a convolution kernel is used to further extract the local spatial features. After each convolution operation is completed, batch normalization is used to standardize the convolution output, and then the ReLU activation function is used to perform a non-linear transformation on the convolution output to obtain the local feature map of the gas-liquid interface. The formula used is as follows:

[0110] ;

[0111] In the formula, represents the value of the output feature map after the (l + 1)-th layer of convolution operation, represents the ReLU activation function, represents the normalization operation, represents the weight value of the (l + 1)-th layer of convolution kernel, represents the value at the position of in the output feature map after the l-th layer of convolution operation, m represents the number of channels of the input feature map, , and represent the size of the convolution kernel, , and Denotes the offset of the convolutional kernel in the spatial dimension. 、 and Denote the stride when the convolutional kernel slides in each dimension. Denotes the bias term of the (l + 1)-th layer convolutional operation;

[0112] Step S3112: Macroscopic feature extraction of the gas-liquid interface, which is used to identify the macroscopic features of the gas-liquid interface. Specifically, dilated convolution is introduced to perform convolution processing on the optimized image of the gas-liquid interface. Batch normalization operation is performed on the feature map after dilated convolution processing, and the ReLU activation function is used to introduce non-linear transformation to the processed feature map, and finally the macroscopic feature map of the gas-liquid interface is obtained; The specific parameter settings of the dilated convolution are to select a convolutional kernel of size and set the dilation rate to 3;

[0113] Step S3113: Feature splicing, specifically splicing the local feature map of the gas-liquid interface and the macroscopic feature map of the gas-liquid interface to obtain the comprehensive feature map of the gas-liquid interface ;

[0114] Step S312: Obtaining global dependencies of the gas-liquid interface, which specifically includes the following steps:

[0115] Step S3121: Preliminary fusion of gas-liquid interface features. Specifically, first perform multi-window self-attention calculation on the comprehensive feature map of the gas-liquid interface to obtain the window merged feature , perform layer normalization on , input it into the MLP module for feature enhancement, and finally obtain the preliminarily fused feature map through residual connection. The multi-window self-attention calculation is specifically based on the size of the comprehensive feature map of the gas-liquid interface , divide it into multiple non-overlapping small windows of size . 、 and respectively represent the depth, height, and width of the window. Perform self-attention calculation operations on each window to obtain the weighted features of each window, merge the weighted features of each window, and output the features through residual connection; The formula used is as follows:

[0116] ;

[0117] ;

[0118] In the formula, denotes the preliminarily fused feature, denotes the layer normalization function, denotes the multi-window self-attention calculation function. represents a multi-layer perceptron function, represents window merged features;

[0119] Step S3122: Deep fusion of gas-liquid interface features, specifically, first perform multi-window offset self-attention calculation on the preliminarily fused features to obtain offset window merged features , perform layer normalization on and input it into the MLP module for feature enhancement, and finally obtain the preliminarily fused feature map through residual connection , the specific window offset self-attention calculation is based on the gas-liquid interface comprehensive feature map size, divide it into multiple non-overlapping small windows of size , and at the same time set the offset , perform window offset, , and respectively represent the depth, height, and width offset amounts of the window, perform self-attention calculation on each offset window to obtain the weighted features of each offset window, merge the weighted features of each offset window, and output the features through residual connection; the formula used is as follows:

[0120] ;

[0121] ;

[0122] In the formula, represents the deep fusion feature, represents the window offset self-attention calculation function, represents the offset window merged feature;

[0123] Step S313: Obtain the preliminary position of the gas-liquid interface, and the formula used is as follows:

[0124] ;

[0125] In the formula, represents the preliminary position of the gas-liquid interface, represents the output weight matrix of the preliminary position of the gas-liquid interface, represents the output bias term parameter of the preliminary position of the gas-liquid interface, represents the Sigmoid activation function.

[0126] By performing the above operations, aiming at the technical problems of insufficient extraction of global information and capture of deep-level relationships between features in the traditional gas-liquid interface precise positioning method in image data analysis, which leads to inaccurate gas-liquid interface positioning accuracy, this solution innovatively proposes an image data processing method based on a multi-window self-attention mechanism and a multi-window offset self-attention mechanism, enhancing the adaptability of gas-liquid interface positioning in complex and dynamically changing environments and improving the gas-liquid interface positioning accuracy. Through the multi-window self-attention mechanism, precise weighting of local features is ensured, and global information can be effectively extracted, solving the problem that the traditional method fails to consider local dependencies in global feature fusion, thus improving the accuracy of gas-liquid interface positioning. Through the multi-window offset self-attention mechanism, the window selection range is optimized, and deep-level relationships between features are effectively captured, further improving the accuracy of gas-liquid interface positioning.

[0127] Example Six, refer to Figure 1 、 Figure 4 and Figure 6 This example is based on the above example. In step S4, enhancing the performance of the gas-liquid interface positioning model specifically includes the following steps:

[0128] Step S41: Training the multi-level gas-liquid interface positioning model, specifically using the historical gas-liquid interface data to train the multi-level gas-liquid interface positioning model to obtain the trained multi-level gas-liquid interface positioning model;

[0129] Step S42: Obtaining the optimal hyperparameters of the model to obtain the optimal hyperparameter combination of the trained multi-level gas-liquid interface positioning model. Specifically, using an improved optimization algorithm to obtain the optimal hyperparameter combination, including the following steps:

[0130] Step S421: Initializing the positions of the population individuals, specifically using the cosine transform initialization method to initialize the search individuals; the formula used is as follows:

[0131] ;

[0132] In the formula, represents the i-th chaotic sequence value, represents a random number uniformly distributed within the range of , represents a random number uniformly distributed within the range of , represents the initialization position of the i-th individual, and represent the search lower limit and upper limit respectively;

[0133] Step S422: Calculating the fitness value, specifically calculating the fitness value f of the individuals in the population i; The performance of the trained multi-level gas-liquid interface positioning model established based on the individual position is used as the fitness value of the individual. The individuals are sorted from the best to the worst according to the fitness value, and the position of the individual with the highest current global fitness value is obtained. ;

[0134] Step S423: Individual global search behavior, and the specific formula is as follows:

[0135] ;

[0136] In the formula, t represents the current iteration number, represents the position of the i-th individual in the t-th generation population, represents the position of the i-th individual in the (t + 1)-th generation population, represents the position of the i-th individual in the (t - 1)-th generation population, represents the position of the individual with the worst fitness value in the t-th iteration of the population, represents between a uniformly distributed random number within the range, represents between a uniformly distributed random number within the range, represents the dynamic coefficient, represents the offset coefficient, and the value range is , represents the obstacle threshold, represents not encountering an obstacle, represents encountering an obstacle;

[0137] Step S424: Optimal solution derivation behavior, specifically, the optimal solution derivation behavior is carried out through a dynamic spiral search strategy, and the formula used is as follows:

[0138] ;

[0139] ;

[0140] ;

[0141] In the formula, represents the position of the i-th derived individual in the (t + 1)-th generation population, represents the position of the i-th derived individual in the t-th generation population, represents the position of the individual with the best fitness value in the t-th iteration of the population, represents the lower boundary of the derived individual position, represents the upper boundary of the derived individual position, and respectively represent the maximum and minimum values of the solution space, and represent between A random number uniformly distributed in the range, represents the constant that controls the path update, Indicates the maximum number of iterations;

[0142] Step S425: Individual local exploration behavior, specifically adjusting the search behavior by dynamically adjusting the item factor, the formula used is as follows:

[0143] ;

[0144] ;

[0145] ;

[0146] In the formula, represents the lower boundary of individual local exploration, represents the upper boundary of individual local exploration, represents random numbers that follow a normal distribution, Indicates between A random number uniformly distributed in the range, Represents the dynamic adjustment factor, which is used to adjust the individual local exploration behavior;

[0147] Step S426: Competitive global perturbation behavior, the specific formula used is as follows:

[0148] ;

[0149] In the formula, Indicates between Random numbers uniformly distributed within a range;

[0150] Step S427: Obtain the optimal hyperparameter combination of the model, specifically when the individual fitness value f i When the fitness threshold is higher than the maximum number of iterations, the search is terminated and the global optimal position of the individual is obtained, and the optimal hyperparameter combination of the model is obtained;

[0151] Step S43: adjusting the hyperparameters of the model, specifically adjusting the hyperparameters of the trained multi-level gas-liquid interface positioning model according to the optimal hyperparameter combination of the model to obtain an optimized multi-level gas-liquid interface positioning model.

[0152] By performing the above operations, in order to address the technical problem of improper built-in parameter settings in existing multi-level gas-liquid interface positioning models, which leads to inaccurate model output results, this solution adopts a cosine transform initialization method, a dynamic spiral search strategy and a dynamic adjustment item factor to improve the algorithm for obtaining the optimal model parameters, obtain the optimal parameter combination, improve the accuracy of the model output results, and thus improve the accuracy of gas-liquid interface positioning.

[0153] Embodiment VI. Refer to Figure 1 , based on the above embodiment, in step S5, the real-time accurate positioning of the gas-liquid interface specifically means inputting the real-time gas-liquid interface data into the optimized multi-level gas-liquid interface positioning model to obtain the real-time accurate gas-liquid interface position result, so as to continuously monitor the change trend of the interface.

[0154] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0155] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention.

[0156] The above describes the present invention and its implementation manners. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. All in all, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments to this technical solution without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An artificial intelligence-based precise positioning method for gas-liquid interfaces, characterized in that: The method includes the following steps: Step S1: Obtain gas-liquid interface positioning data. By collecting data, obtain the original gas-liquid interface positioning data. Step S2: Optimize and process the original data, which is used to optimize the original gas-liquid interface positioning data. Specifically, perform image data optimization processing, environmental data optimization processing, and data time synchronization on the original gas-liquid interface positioning data to obtain the optimized gas-liquid interface positioning data. Step S3: Construct a multi-level gas-liquid interface positioning model, which is used to accurately identify the position of the gas-liquid interface. Specifically, by comprehensively analyzing image data, time-series data, and environmental data, design and construct a multi-level structure model to obtain a multi-level gas-liquid interface positioning model. This model first analyzes the gas-liquid interface image data through the designed image data processing layer to obtain the preliminary position of the gas-liquid interface, and then processes the time-series data and environmental data of the historical gas-liquid interface positions, and integrates the processing results of the time-series data and environmental data in the model output layer to further adjust the preliminary position of the gas-liquid interface to achieve accurate positioning of the gas-liquid interface. Step S4: Enhance the performance of the gas-liquid interface positioning model, which is used to optimize the training of the model and improve the hyperparameters of the model, and improve the accuracy of the output results of the multi-level gas-liquid interface positioning model. Train the model using historical data, and adopt the cosine transform initialization method, dynamic spiral search strategy, and dynamic adjustment term factor to improve the algorithm for obtaining the optimal parameters of the model, obtain the optimal hyperparameter combination of the model, adjust the hyperparameters of the model according to the optimal hyperparameter combination of the model, and finally obtain the optimized multi-level gas-liquid interface positioning model. Step S5: Real-time and accurate positioning of the gas-liquid interface. Specifically, input the real-time data into the optimized model to obtain the real-time accurate gas-liquid interface position result.

2. The method for accurately positioning the gas-liquid interface based on artificial intelligence according to claim 1, wherein: In step S3, the construction of the multi-level gas-liquid interface positioning model is specifically to design an image data processing layer, a dynamic time-series data processing layer, an environmental data processing layer, and a model output layer to obtain a multi-level gas-liquid interface positioning model. It includes the following steps: Step S31: Design an image data processing layer for obtaining the preliminary position of the gas-liquid interface from the gas-liquid interface image data ; Step S32: Design a dynamic time-series data processing layer for tracking the change of the gas-liquid interface position over time and optimizing the positioning result of the gas-liquid interface. Specifically, process the time-series information of the historical gas-liquid interface position data through a long short-term memory network to obtain the dynamic time-series features of the gas-liquid interface ; The formula used is as follows: ； In the formula, represents the unit operation function, represents the hidden state at the previous moment, represents the input historical gas-liquid interface position data at the current moment; Step S33: Design an environmental data processing layer for optimizing the positioning result of the gas-liquid interface through environmental data. Specifically, process the environmental data through a fully connected layer to obtain the gas-liquid interface environmental features ; The formula used is as follows: ； In the formula, represents the weight matrix of environmental data, represents the bias term parameter of environmental data, represents environmental data; Step S34: Design the model output layer, which is used to obtain the final position of the gas-liquid interface. The formula used is as follows: ； In the formula, represents the final position of the gas-liquid interface, represents the output weight matrix of the final position of the gas-liquid interface, represents the output bias term parameter of the final position of the gas-liquid interface.

3. The method for precise positioning of gas-liquid interface based on artificial intelligence according to claim 1, characterized in that: In step S31, the design of the image data processing layer specifically includes the following steps: Step S311: Extract the gas-liquid interface image features, which specifically includes the following steps: Step S3111: Local feature extraction of the gas-liquid interface, which is used to extract local spatial features from the gas-liquid interface image. Specifically, a convolution kernel with a size of is used to perform convolution processing on the optimized gas-liquid interface image to adjust the number of channels of the input feature map. Then, the convolution kernel is used to further extract local spatial features. After each convolution operation, batch normalization is used to standardize the convolution output, and then the ReLU activation function is used to perform a non-linear transformation on the convolution output to obtain the local feature map of the gas-liquid interface. The formula used is as follows: ； In the formula, represents the value of the output feature map after the (l + 1)-th layer convolution operation, represents the ReLU activation function, represents the normalization operation, represents the weight value of the (l + 1)-th layer convolution kernel, represents the value at the position in the output feature map after the l-th layer convolution operation, m represents the number of channels of the input feature map, , and represent the size of the convolution kernel, , and represent the offset of the convolution kernel in the spatial dimension, , and represent the stride when the convolution kernel slides in each dimension, represents the bias term of the (l + 1)-th layer convolution operation; Step S3112: Extract the macroscopic features of the gas-liquid interface, which is used to identify the macroscopic features of the gas-liquid interface. Specifically, introduce dilated convolution to perform convolution processing on the optimized gas-liquid interface image, perform batch normalization operation on the feature map after dilated convolution processing, and use the ReLU activation function to introduce nonlinear transformation to the processed feature map, and finally obtain the macroscopic feature map of the gas-liquid interface. Step S3113: Feature splicing, specifically splicing the local feature map of the gas-liquid interface and the macroscopic feature map of the gas-liquid interface to obtain a comprehensive feature map of the gas-liquid interface ; Step S312: Obtain the global dependence of the gas-liquid interface, which specifically includes the following steps: Step S3121: Preliminary fusion of gas-liquid interface features. Specifically, first, for the comprehensive gas-liquid interface feature map perform multi-window self-attention calculation to obtain window merged features , perform layer normalization on , input it into the MLP module for feature enhancement, and finally obtain the preliminarily fused feature map through residual connection; the multi-window self-attention calculation is specifically as follows: according to the size of the comprehensive gas-liquid interface feature map , divide it into multiple non-overlapping small windows of size , , and respectively represent the depth, height, and width of the window. Perform self-attention calculation operations on each window to obtain the weighted features of each window, merge the weighted features of each window, and output the features through residual connection; the formula used is as follows: ； ； In the formula, represents the preliminary fusion feature, represents the layer normalization function, represents the multi-window self-attention calculation function, represents the multi-layer perceptron function, represents the window merging feature; Step S3122: Deep fusion of gas-liquid interface features, specifically, first, perform multi-window offset self-attention calculation on the preliminary fusion features to obtain the offset window merged features , perform layer normalization on and input it into the MLP module for feature enhancement. Finally, through residual connection, obtain the preliminarily fused feature map ; The window offset self-attention calculation is specifically as follows: According to the size of the gas-liquid interface comprehensive feature map , divide it into multiple non-overlapping small windows of size , and at the same time set the offset , perform window offset, , and respectively represent the depth, height, and width offset amounts of the window. Perform self-attention calculation on each offset window to obtain the weighted features of each offset window. Merge the weighted features of each offset window and output the features through residual connection; The formula used is as follows: ； ； In the formula, represents the deep fusion feature, represents the window offset self-attention calculation function, represents the offset window merged feature; Step S313: Obtain the preliminary position of the gas-liquid interface. The formula used is as follows: ； In the formula, represents the initial position of the gas-liquid interface, represents the output weight matrix of the initial position of the gas-liquid interface, represents the output bias term parameter of the initial position of the gas-liquid interface, represents the Sigmoid activation function.

4. The method for precise positioning of gas-liquid interface based on artificial intelligence according to claim 1, characterized in that: In step S4, the enhancement of the performance of the gas-liquid interface positioning model specifically includes training the multi-level gas-liquid interface positioning model, obtaining the optimal hyperparameters of the model, and adjusting the hyperparameters of the model to obtain the optimized multi-level gas-liquid interface positioning model. It includes the following steps: Step S41: Training the multi-level gas-liquid interface positioning model, specifically training the multi-level gas-liquid interface positioning model using historical gas-liquid interface data to obtain the trained multi-level gas-liquid interface positioning model; Step S42: Obtaining the optimal hyperparameters of the model, which is used to obtain the optimal hyperparameter combination of the trained multi-level gas-liquid interface positioning model. Specifically, using an improved optimization algorithm to obtain the optimal hyperparameter combination, including the following steps: Step S421: Initializing the position of the population individuals, specifically using the cosine transform initialization method to initialize the search individuals; the formula used is as follows: ； Wherein, represents the value of the i-th chaotic sequence, represents a random number uniformly distributed within the range of ; represents a random number uniformly distributed within the range of ; represents the initial position of the i-th individual, and represent the lower search limit and the upper search limit, respectively; Step S422: Calculate the fitness value, specifically calculate the fitness value f of the individuals in the population i ; Take the performance of the trained multi-level gas-liquid interface positioning model established based on the individual's position as the fitness value of the individual, sort the individuals from the best to the worst according to the fitness value, and obtain the position of the individual with the highest current global fitness value ; Step S423: The global search behavior of individuals, specifically the formula used is as follows: ； where t represents the current iteration number, represents the position of the i-th individual in the population at the t-th generation, represents the position of the i-th individual in the population at the (t + 1)-th generation, represents the position of the i-th individual in the population at the (t - 1)-th generation, represents the position of the individual with the worst fitness value in the population at the t-th iteration, represents a random number uniformly distributed within the range of ; represents a random number uniformly distributed within the range of ; represents the dynamic coefficient, represents the offset coefficient, and its value range is , represents the obstacle threshold, represents that no obstacle is encountered, represents that an obstacle is encountered; Step S424: The optimal solution derivation behavior, specifically performing the optimal solution derivation behavior through a dynamic spiral search strategy, and the formula used is as follows: ； ； ； Wherein, represents the position of the i-th derived individual in the (t + 1)-th generation population, represents the position of the i-th derived individual in the t-th generation population, represents the position of the individual with the best fitness value in the population at the t-th iteration, represents the lower boundary of the derived individual position, represents the upper boundary of the derived individual position, and respectively represent the maximum and minimum values of the solution space, and represent random numbers uniformly distributed within the range, represents a constant for controlling path update, represents the maximum number of iterations; Step S425: The local exploration behavior of individuals, specifically adjusting the search behavior by dynamically adjusting the item factor, and the formula used is as follows: ； ； ； wherein, represents the lower boundary of individual local exploration, represents the upper boundary of individual local exploration, represents a random number subject to a normal distribution, represents a random number uniformly distributed within the range of ; represents a dynamic adjustment term factor for adjusting the individual local exploration behavior; Step S426: The competitive global perturbation behavior, specifically the formula used is as follows: ； In the formula, represents a random number uniformly distributed within the range of ; Step S427: Obtain the optimal hyperparameter combination of the model. Specifically, when the individual fitness value f i is higher than the fitness threshold and the maximum number of iterations is reached, terminate the search and obtain the global optimal position of the individual to get the optimal hyperparameter combination of the model; Step S43: Adjusting the hyperparameters of the model, specifically adjusting the hyperparameters of the trained multi-level gas-liquid interface positioning model according to the optimal hyperparameter combination of the model to obtain the optimized multi-level gas-liquid interface positioning model.

5. The method for accurately positioning the gas-liquid interface based on artificial intelligence according to claim 1, wherein: In step S5, the real-time precise positioning of the gas-liquid interface specifically means inputting the real-time gas-liquid interface data into the optimized multi-level gas-liquid interface positioning model to obtain the real-time precise gas-liquid interface position result, so as to continuously monitor the change trend of the interface.

6. The method for accurately positioning the gas-liquid interface based on artificial intelligence according to claim 1, wherein: In step S1, the acquisition of the gas-liquid interface positioning data specifically means collecting the original gas-liquid interface positioning data through an industrial camera and sensors; the original gas-liquid interface positioning data includes historical gas-liquid interface data and real-time gas-liquid interface data; both the historical gas-liquid interface data and the real-time gas-liquid interface data include gas-liquid interface image data and environmental data; the historical gas-liquid interface data also includes historical gas-liquid interface position data.

7. The method for accurately positioning the gas-liquid interface based on artificial intelligence according to claim 1, wherein: In step S2, the optimization processing of the original data is used to optimize the original gas-liquid interface positioning data. Specifically, it performs image data optimization processing, environmental data optimization processing, and data time synchronization on the original gas-liquid interface positioning data to obtain the optimized gas-liquid interface positioning data; Step S21: Image data optimization processing, which is used to obtain the optimized gas-liquid interface image, specifically including image denoising, image enhancement, and image registration; Step S22: Environmental data optimization processing, specifically including data cleaning and standardization processing; Step S23: Data time synchronization, which is used to analyze the relationship between the gas-liquid interface and environmental changes; specifically, according to the timestamps of the image data and environmental data, synchronize each image data frame with the environmental data in the corresponding time period.

Images