An on-line intelligent monitoring method and system for oil content and gas content of a drilling fluid

Abstract

The application discloses an online intelligent monitoring method for oil content and gas content of a drilling fluid, and the method comprises the following steps: collecting image samples of water-based drilling fluid without being separated by a solid-liquid separator by using an online image acquisition device of the drilling fluid, and establishing an image database; further highlighting the gas content and oil content in a unit area, pre-processing the image data, and automatically generating a trainable data set; designing an online intelligent monitoring neural network structure, training and generating an intelligent identification model; packaging the model, and establishing an online intelligent monitoring system. According to the proportion of bubble area, the proportion of oil droplet area and the shape of water-based drilling fluid, the relationship between these characteristics and the gas content and the oil content of the drilling fluid is fitted through a deep residual shrinkage network, so that the volume fraction of the gas content and the oil content in the water-based drilling fluid can be evaluated in real time, and online inquiry and monitoring can be realized through a user interaction module of the monitoring system.

Description

Technical Field

[0001] This invention relates to the field of drilling engineering technology, and in particular to an online intelligent monitoring method and system for oil and gas content in drilling fluid. Background Technology

[0002] Online drilling fluid monitoring technology refers to the real-time monitoring and analysis of drilling fluid, primarily through manual observation supplemented by various sensors. This not only helps operators identify and resolve problems encountered during drilling, optimizing drilling operations to improve production efficiency and safety, but also provides real-time information on certain formations. The gas and oil content in the drilling fluid are crucial pieces of information during the monitoring process.

[0003] Currently, monitoring of gas and oil content in water-based drilling fluids in China primarily relies on manual measurement and sensor measurement. Manual measurement involves observing the presence of abundant foam or a noticeable oil film on the drilling fluid surface, typically indicating problems such as excessive gas or oil contamination. Sensor measurement, on the other hand, uses instruments like pressure sensors, radar level gauges, ultrasonic level gauges, and float level gauges to monitor mud pit level changes in real time, indirectly assessing the oil and gas content in the drilling fluid. While manual measurement offers some real-time capability, it struggles to detect minute bubbles and oil droplets. Although sensor measurement can assess these minute bubbles and droplets, mud pit level monitoring occurs after the drilling fluid has passed through a solid-liquid separator, inherently exhibiting a certain lag. Furthermore, the vibration of the separator releases some gas from the drilling fluid, which dissipates with the airflow above the vibrating screen, thus affecting the sensor's sensitivity in detecting minute bubbles. Summary of the Invention

[0004] This invention proposes an online intelligent monitoring method and system for oil and gas content in drilling fluid. It optimizes the existing methods by addressing the issues of sensitivity and real-time performance in monitoring microbubbles in drilling fluid. Before the drilling fluid enters the solid-liquid separator, image data is collected using an instrument. Based on the characteristics of bubble area, oil droplet area, and morphology in the drilling fluid image data, a depth residual shrinkage network is used to fit the relationship between these characteristics and the gas and oil content of the drilling fluid. This allows for sensitive and timely determination of the oil and gas content in the drilling fluid. While addressing the problems of the existing technology, this invention also provides guidance for the early detection of gas intrusion and overflow.

[0005] In a first aspect, a method for online intelligent monitoring of oil content and gas content in drilling fluid includes the following steps:

[0006] Step S1: Use online drilling fluid image acquisition equipment to acquire image samples of water-based drilling fluid without a solid-liquid separator, and establish an image database;

[0007] Step S2: Preprocess the image data to automatically generate a trainable dataset;

[0008] Step S3: Design the structure of the online intelligent monitoring neural network, train and generate an intelligent recognition model;

[0009] Step S4: Encapsulate the model and establish an online intelligent monitoring system.

[0010] Furthermore, the water-based drilling fluid image sample in step S1 includes a quantitative volume water-based drilling fluid sample collected in the field environment and a laboratory supplementary sample; the laboratory supplementary sample is a quantitative volume water-based drilling fluid prepared in the laboratory environment, and a sample with injected gas and oil droplets.

[0011] Furthermore, the laboratory supplementary sample quantitative volume water-based drilling fluid contains no additional gas or oil droplets. The injected gas and oil droplets are randomly selected from the volume fraction range of 0% to 10% under the quantitative volume standard and injected into the prepared quantitative volume water-based drilling fluid.

[0012] Furthermore, the establishment of the image database in step S1 is completed by using an online drilling fluid image recognition device to photograph the water-based drilling fluid sample after fluorescence irradiation, infrared irradiation, and scaling processing. Each pixel image in the image database is the same size.

[0013] Furthermore, the image preprocessing step S2 includes a bilateral filtering algorithm and a maximum-minimum normalization. The training samples in the trainable dataset include image data and corresponding volume fraction data. The formula for the bilateral filtering algorithm is:

[0014]

[0015] Where ξ is the filtered pixel, x is the surrounding pixel of ξ, f(ξ) and f(x) represent the pixel values ​​of the corresponding pixels, functions c and s are the Closeness spatial proximity function and the Similarity pixel similarity function, respectively, and the weight values ​​in the weighted average are calculated using the Closeness spatial proximity function and the Similarity pixel similarity function. The Closeness spatial proximity function is used to measure the spatial distance between pixels and is described by a Gaussian function: Where d(ξ,x)=‖ξ-x‖; the Similarity pixel similarity function is used to measure the similarity between pixels, and is described by a Gaussian function as: Among them δ(f(ξ),f(x))=‖f(ξ)-f(x)‖.

[0016] Furthermore, the volume fraction is in the form of a three-dimensional vector, [α,β,1-α-β]. T Where α is the gas content of the drilling fluid; β is the oil content of the drilling fluid; the maximum and minimum value normalization includes: finding the minimum value x in the original data. min With the maximum value x max Through formula This allows you to scale the data x to the range [0,1].

[0017] Furthermore, the network structure in step S3 is a neural network structure built upon deep residual shrinkage module units, including an input layer, residual shrinkage calculation block 1, residual shrinkage calculation block 2, residual shrinkage calculation block 3, residual shrinkage calculation block 4, and an output block. The ReLU function is used as the activation function for the non-output blocks, specifically including the following:

[0018] The required image data pixel size for the input layer is 224×224;

[0019] The four residual shrinkage calculation blocks together form a 50-layer deep neural network structure, which extracts the characteristic states of oil and gas content in the image at different volume fractions.

[0020] The network model was trained using Nadam as the optimizer, with learn_rate set to 0.01 and schedule_decay set to 0.06.

[0021] The output block includes a Flatten layer, a fully connected layer 1, and a fully connected layer 2; the softplus function is used as the activation function for the fully connected layer 1, and the number of neurons is set to 1024; the Temperature-Softmax function is used as the activation function for the fully connected layer 2, and the number of neurons is set to 3.

[0022] Furthermore, the Temperature-Softmax function utilizes the feature vector extracted by the deep residual shrinkage network to output feature scores of the oil droplet area, gas area, and oil droplet morphology in the image, representing the drilling fluid's gas and oil content. Its function is to fit the relationship between image features and the gas and oil content in the drilling fluid, expressed as:

[0023]

[0024] Where x represents the input feature vector, n represents the data dimension of the feature vector x, and y represents the feature score vector of drilling fluid gas content and oil content.

[0025] On the other hand, an online intelligent monitoring system for oil and gas content in drilling fluid is provided to realize an online intelligent monitoring method for oil and gas content in drilling fluid, including a data storage module, an intelligent identification module, and a user interaction module.

[0026] The intelligent recognition module is connected to the image database and receives training support from the image database, which provides image data by supplementing laboratory training samples and collecting on-site training samples.

[0027] The data storage module is connected to the intelligent recognition module to receive recognition results, and also includes an intelligent recognition model encapsulated in the drilling fluid online image recognition device and supports remote data storage in the cloud storage terminal;

[0028] The user interaction module and data storage module send query commands and display query results, and also include other query terminals for querying.

[0029] Furthermore, the user interaction module includes a historical data table, a drilling fluid gas content monitoring chart, a drilling fluid oil content monitoring chart, a drilling fluid gas content average value window, and a drilling fluid oil content average value window. The historical data table allows users to view all historical monitoring records of drilling fluid gas content and oil content. The drilling fluid gas content monitoring chart and the drilling fluid oil content monitoring chart display the historical monitoring value change trends of drilling fluid gas content and oil content over a period of time. The drilling fluid gas content average value window and the drilling fluid oil content average value window display the historical monitoring value average of drilling fluid gas content and oil content over a period of time.

[0030] The beneficial effects of this invention are as follows: This invention proposes an online intelligent monitoring method and system for oil and gas content in drilling fluids. The method includes collecting water-based drilling fluid samples and establishing an image database; preprocessing the image data to automatically generate a trainable dataset; training a deep residual shrinkage network to generate an intelligent recognition model; and encapsulating the model to establish an online intelligent monitoring system. Based on the characteristics of the bubble area, oil droplet area, and morphology in water-based drilling fluids, the relationship between these characteristics and the gas and oil content in the drilling fluids is fitted using a deep residual shrinkage network. This allows for real-time assessment of the volume fraction of gas and oil content in the water-based drilling fluids, which can be queried and monitored online through the user interaction module of the monitoring system. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the online intelligent monitoring method for oil and gas content in drilling fluid according to an embodiment of the present invention.

[0032] Figure 2 This is an architecture diagram of an online intelligent monitoring system for oil and gas content in drilling fluid according to an embodiment of the present invention;

[0033] Figure 3 This is a diagram of the deep residual shrinkage network structure in an embodiment of the present invention;

[0034] Figure 4 This is a structural diagram of the user interaction module of the monitoring system in an embodiment of the present invention;

[0035] Figure 5 This is an example diagram of the user interaction module of the monitoring system in an embodiment of the present invention. Detailed Implementation

[0036] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0037] This invention proposes an online intelligent monitoring method and system for oil and gas content in drilling fluid, such as... Figure 1 As shown, the monitoring method mainly includes the following four steps:

[0038] Step S1: Collect water-based drilling fluid samples and establish an image database;

[0039] Step S2: Preprocess the image data to automatically generate a trainable dataset;

[0040] Step S3: Design the structure of the online intelligent monitoring neural network, train and generate an intelligent recognition model;

[0041] Step S4: Encapsulate the model and establish an online intelligent monitoring system.

[0042] In a preferred embodiment, step S1: collecting water-based drilling fluid samples and establishing an image database are as follows;

[0043] In this embodiment, the water-based drilling fluid sample includes a quantitative volume water-based drilling fluid sample collected under field conditions and a supplementary laboratory sample. The supplementary laboratory sample is a quantitative volume water-based drilling fluid prepared under laboratory conditions, with gas and oil droplets injected. The standard volume of the quantitative volume water-based drilling fluid prepared under laboratory conditions is 56mm × 56mm × 0.1mm, and the drilling fluid contains no additional gas or oil droplets. Injecting gas and oil droplets refers to injecting gas and oil droplets into a quantitative volume of 313.6mm. 3 Under standard conditions, the volume fractions of gas and oil droplets are randomly selected from 0% to 10% and injected into the prepared quantitative volume of water-based drilling fluid.

[0044] In this embodiment, the image database refers to the database established by taking pictures of the above-mentioned water-based drilling fluid samples after processing such as fluorescence irradiation, infrared irradiation, and scaling using an online drilling fluid image recognition device. The pixel size of each image data in the database is 224×224.

[0045] In a preferred embodiment, step S2: preprocessing the image data to automatically generate a trainable dataset is as follows;

[0046] In this embodiment, the preprocessing of image data includes bilateral filtering and maximum-minimum normalization. The bilateral filtering algorithm mainly prevents the loss of useful edge information during Gaussian filtering and is an improved version of Gaussian filtering. The maximum-minimum normalization mainly summarizes the statistical distribution of samples. Normalization in the range [0,1] is a statistical probability distribution. It can also limit the preprocessed data to a certain range, eliminate the adverse effects caused by singular sample data, and thus accelerate the convergence speed of the neural network.

[0047] In this embodiment, the bilateral filtering algorithm specifically involves weighted averaging of the surrounding pixel values ​​for each pixel value in the image. Its filtering function can be described as follows:

[0048] Where ξ is the filtered pixel, x is the surrounding pixel of ξ, f(ξ) and f(x) represent the pixel values ​​of the corresponding pixels, and functions c and s are the Closeness and Similarity functions, respectively. The weight values ​​in the weighted average are calculated using the Closeness and Similarity functions.

[0049] The Closeness function measures the spatial distance between pixels and can be described using a Gaussian function: Where d(ξ,x) = ||ξ-x||;

[0050] The Similarity function measures the similarity between pixels and can be described using a Gaussian function: Among them δ(f(ξ),f(x))=‖f(ξ)-f(x)‖.

[0051] In this embodiment, the specific process of normalizing the maximum and minimum values ​​is as follows:

[0052] 1) Find the minimum value x in the original data. min With the maximum value x max ;

[0053] 2) Through the formula This allows data to be scaled to the range [0,1].

[0054] Since the training samples are image data with pixel values ​​distributed in the interval [0, 255], we take x... min =0,x max =255 is sufficient.

[0055] In this embodiment, the automatically generated trainable dataset, each training sample in the dataset can be further interpreted as a combination of preprocessed image data and the volume fraction of injected gas and oil droplets, with a one-to-one correspondence between image data and volume fraction;

[0056] The volume fractions mentioned above are in three-dimensional vector form, i.e., [α,β,1-α-β]. T , where α is the gas content of the drilling fluid; β is the oil content of the drilling fluid.

[0057] In a preferred embodiment, step S3: generating an intelligent recognition model based on training a deep residual shrinking network specifically includes the following.

[0058] In this embodiment, the network structure refers to a neural network structure built based on deep residual shrinkage module units, including an input layer, residual shrinkage calculation block 1, residual shrinkage calculation block 2, residual shrinkage calculation block 3, residual shrinkage calculation block 4, and an output block. The ReLU function is used as the activation function for the non-output blocks and must meet the following criteria:

[0059] The number of training samples is 4000, and the image data in each training sample has a pixel size of 224×224.

[0060] The batch_size value is set to 64, and the epoch value is set to 480.

[0061] The four residual shrinkage calculation blocks together form a 50-layer deep neural network structure;

[0062] The optimizer is set to Nadam, and the learn_rate value is set to 0.01, and the schedule_decay value is set to 0.06;

[0063] The output block contains a Flatten layer, a fully connected layer 1, and a fully connected layer 2. The softplus function is used as the activation function for fully connected layer 1, with 1024 neurons. Fully connected layer 2 uses the Temperature-Softmax function as its activation function, with 3 neurons.

[0064] In this embodiment, the batch_size value is set to 64, indicating that 64 training samples are included in one training iteration, and the epoch value is set to 480, indicating that there are 480 iterations. One iteration includes a complete number of training iterations. If the number of training samples is n, the total number of training iterations can be calculated as: iteration = n / batch_size.

[0065] The deep residual shrinking network is an improved deep residual network that incorporates an attention algorithm with a soft thresholding mechanism. Its overall structure is as follows: Figure 3 As shown, it includes an input block (INPUT) and computation blocks (BLOCK1-4). The input block is a preprocessing step for the input data tensor, where (B, C, W, H) represents the number of samples (Batch_Size), number of channels (Channel), width (Width), and height (Height) of the image in each training iteration, respectively, describing the shape change of the data tensor after each computation node. The input block includes convolution operations (Conv), batch normalization operations (BatchNormalization), and max pooling operations (Max Pool).

[0066] In this embodiment, the convolution operation is based on Figure 3 Taking Conv(7×7,64) as an example, the first element represents the kernel size, and the second element represents the number of kernels. The kernel size affects network performance and masking speed, while the number of kernels affects the feature extraction capability of the convolutional layer. Let the number of kernels be C′, and the kernel tensor be V=[v1,v2,…,v…]. C′ The input data tensor is X = [x1, x2, ..., x]. C ],X∈R C×W×H The output data tensor is U = [x1, u2, ..., u C′ Therefore, the convolution operation can be described as follows:

[0067] The * operator represents masking.

[0068] In this embodiment, the main function of batch normalization is to accelerate model training, improve the model's generalization ability, and reduce gradient vanishing and exploding problems. Its main process is as follows:

[0069] 1) For a batch of input data X = [x1, x2, ..., xn] of size n n For each feature dimension k of the input data x, calculate the mean μ along that dimension. k and variance They are respectively:

[0070]

[0071] 2) For each feature dimension k, perform a normalization operation to obtain the normalized features, which can be described by the following expression:

[0072] Where ∈ takes the value 10 -5 ;

[0073] 3) For the normalized features Perform a linear transformation and translation to obtain the final output feature y. i Its expression is: Here, γ and β are learnable parameters that correspond to scaling and offset operations, respectively, and are used to enhance the expressive power of the data.

[0074] In this embodiment, the main function of max pooling is to extract features, enhance the translation invariance of the model, and reduce the risk of overfitting. Its main process is as follows:

[0075] 1) For the input feature map, divide it into several non-overlapping rectangular regions, each region is called a pooling window;

[0076] 2) For each pooling window, take the maximum value as the value at the corresponding position in the output feature map;

[0077] 3) Continuously move the pooling window and repeat the above process until the entire feature map is traversed to obtain the final output feature map.

[0078] In this embodiment, the computation block contains a large number of standard residual shrinkage network basic modules, with the structure as follows: Figure 3 As shown in RSN1 Block and RSN2 Block, their structures incorporate an attention algorithm combined with a soft thresholding mechanism. The main function of the attention algorithm is to scan global information, discover and enhance useful local information, and suppress redundant information. SENet is one of the classic attention algorithms. The soft thresholding mechanism enhances the feature representation capabilities of different channels in the neural network, thereby improving the performance of the neural network.

[0079] In this embodiment, the attention algorithm incorporating a soft thresholding mechanism replaces the weighting of each feature channel in SENet with soft thresholding of each feature channel. Soft thresholding compresses the feature map of each channel into a single numerical value, representing the importance of that channel. During compression, values ​​below a certain threshold are set to 0, and values ​​greater than or equal to the threshold are reduced by a fixed percentage. This threshold is obtained through continuous training. The specific process of the attention algorithm incorporating a soft thresholding mechanism is as follows: Figure 3 Global average pooling in RSN1 Block and RSN2 Block to F scale This is the calculation process.

[0080] In this embodiment, the main function of global average pooling is to encode the entire spatial features of each channel into global features, so that the C channels are finally compressed into a C×1×1 real number sequence seq, the expression of which is as follows:

[0081]

[0082] Global average pooling to F scale In this computational process, the threshold training part uses a fully connected layer and a sigmoid activation function to learn the nonlinear relationships between each channel. scale Its function is to multiply the data S after each channel is activated by the Sigmoid function with the original features of the input data X, and its expression is: F scale (x c ,s c )=x c ·s c ;

[0083] In this embodiment, the Temperature-Softmax function will use the feature vector extracted by the deep residual shrinkage network to output the feature scores of the oil droplet area, gas area, and oil droplet shape in the image, which represent the gas content and oil content of the drilling fluid. Its function is to fit the relationship between the above features and the gas content and oil content in the drilling fluid.

[0084] The Temperature-Softmax function described above is an improved version of the Softmax function. It introduces a temperature parameter T to control the entropy of the output probability distribution, thereby increasing the model's uncertainty and improving its generalization ability. Its expression is as follows:

[0085] Where x represents the input feature vector, n represents the data dimension of the feature vector x, and y represents the feature score vector of drilling fluid gas content and oil content.

[0086] In a preferred embodiment, step S4: encapsulate the model and establish an online intelligent monitoring system; specifically, it includes the following:

[0087] like Figure 2 The diagram shows a schematic of an online intelligent monitoring system for oil and gas content in drilling fluid, which mainly includes the following three parts:

[0088] 1: Data storage module; 2: Intelligent recognition module; 3: User interaction module.

[0089] The data storage module includes a cloud storage terminal, which allows users to query historical gas and oil content data of drilling fluid for the corresponding well number without geographical limitations.

[0090] The intelligent recognition module includes an intelligent recognition model encapsulated in the online image recognition equipment for drilling fluid.

[0091] like Figure 4As shown, the user interaction module includes a historical data table, a drilling fluid gas content monitoring chart, a drilling fluid oil content monitoring chart, a drilling fluid gas content average window, and a drilling fluid oil content average window.

[0092] In this embodiment, the historical data table allows users to view all historical monitoring records of drilling fluid gas content or oil content, specifically in... Figure 4 Area-04 in the middle;

[0093] The drilling fluid gas content monitoring chart and drilling fluid oil content monitoring chart show the historical monitoring value trends of drilling fluid gas content and oil content over a period of time. Figure 4 Areas 02 and 03 in the middle;

[0094] The drilling fluid gas content average window and drilling fluid oil content average window display the average historical monitoring values ​​of drilling fluid gas content and oil content over a period of time. Figure 4 Area-05 in the middle.

[0095] Area-01 contains the current system identifier and the current system location and time. A specific user interaction module implementation is shown below. Figure 5 As shown.

[0096] This invention proposes an online intelligent monitoring method and system for oil and gas content in drilling fluids. The method includes collecting water-based drilling fluid samples and establishing an image database; preprocessing the image data to automatically generate a trainable dataset; training an intelligent recognition model based on a depth residual contraction network; and encapsulating the model to establish an online intelligent monitoring system. Based on the characteristics of the area and morphology of gas bubbles, oil droplets, etc., in water-based drilling fluids, the relationship between these characteristics and the gas and oil content of the drilling fluid is fitted using a depth residual contraction network. This allows for real-time assessment of the volume fraction of gas and oil content in the water-based drilling fluid, which can be queried and monitored online through the user interaction module of the monitoring system.

[0097] The foregoing description and illustrations have shown and described the basic principles, main features, and advantages of this invention. Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for online intelligent monitoring of oil content and gas content in drilling fluid, characterized in that, Based on the area and morphological characteristics of air bubbles and oil droplets in water-based drilling fluid, the relationship between these characteristics and the gas and oil content of the drilling fluid is fitted using a depth residual shrinkage network. The volume fraction of gas and oil content in the water-based drilling fluid is then evaluated in real time. The online intelligent monitoring method includes the following steps: Step S1: Use an online drilling fluid image acquisition device to acquire water-based drilling fluid image samples without a solid-liquid separator and establish an image database; the water-based drilling fluid image samples include quantitative volume water-based drilling fluid samples acquired in the field environment and laboratory supplementary samples; the laboratory supplementary samples are quantitative volume water-based drilling fluid samples prepared in the laboratory environment and injected with gas and oil droplets. Step S2: Preprocess the image data to automatically generate a trainable dataset; the image preprocessing includes a bilateral filtering algorithm and a maximum-minimum normalization; the training samples in the trainable dataset include image data and corresponding volume fraction data; the formula for the bilateral filtering algorithm is: ； Where ξ is the filtered pixel, x is the surrounding pixel of ξ, f(ξ) and f(x) represent the pixel values ​​of the corresponding pixels, functions c and s are the Closeness spatial proximity function and the Similarity pixel similarity function, respectively, and the weight values ​​in the weighted average are calculated using the Closeness spatial proximity function and the Similarity pixel similarity function. The Closeness spatial proximity function is used to measure the spatial distance between pixels and is described by a Gaussian function: Where d(ξ,x)=‖ξ-x‖; the Similarity pixel similarity function is used to measure the similarity between pixels, and is described by a Gaussian function as: , where δ(f(ξ),f(x))=‖f(ξ)-f(x)‖ Step S3: Design an online intelligent monitoring neural network structure, train and generate an intelligent recognition model; the network structure is a neural network structure based on deep residual shrinkage module units, including an input layer, residual shrinkage calculation block 1, residual shrinkage calculation block 2, residual shrinkage calculation block 3, residual shrinkage calculation block 4 and an output block, using the ReLU function as the activation function for non-output blocks, specifically including the following: The required image data pixel size for the input layer is 224×224; The four residual shrinkage calculation blocks together form a 50-layer deep neural network structure, which extracts the characteristic states of oil and gas content in the image at different volume fractions. The network model was trained using Nadam as the optimizer, with learn_rate set to 0.01 and schedule_decay set to 0.

06. The output block includes a Flatten layer, a fully connected layer 1, and a fully connected layer 2; the softplus function is used as the activation function for the fully connected layer 1, and the number of neurons is set to 1024; the Temperature-Softmax function is used as the activation function for the fully connected layer 2, and the number of neurons is set to 3. The Temperature-Softmax function utilizes the feature vector extracted by the deep residual shrinkage network to output feature scores representing the drilling fluid's gas and oil content, as well as the area of ​​oil droplets, gas area, and droplet morphology in the image. Its function is to fit the relationship between image features and the gas and oil content in the drilling fluid, expressed as: ； Where x represents the input feature vector, n represents the data dimension of the feature vector x, y represents the feature fraction vector about the gas content and oil content of the drilling fluid, and T represents the temperature parameter; Step S4: Encapsulate the model and establish an online intelligent monitoring system.

2. The online intelligent monitoring method for oil and gas content in drilling fluid according to claim 1, characterized in that, The laboratory supplementary sample of the quantitative volume water-based drilling fluid contains no additional gas or oil droplets. The injected gas and oil droplets are randomly selected from the volume fractions of 0% to 10% under the quantitative volume standard and injected into the prepared quantitative volume water-based drilling fluid.

3. The online intelligent monitoring method for oil and gas content in drilling fluid according to claim 2, characterized in that, The establishment of the image database in step S1 is completed by using an online drilling fluid image recognition device to photograph the water-based drilling fluid sample after fluorescence irradiation, infrared irradiation, and scaling. Each pixel image in the image database is the same size.

4. The online intelligent monitoring method for oil and gas content in drilling fluid according to claim 1, characterized in that, The volume fraction is in the form of a three-dimensional vector, [α,β,1-α-β]. T , where α is the gas content of the drilling fluid; β represents the oil content of the drilling fluid; The maximum and minimum value normalization includes: finding the minimum value x in the original data. min With the maximum value x max Through formulas, This allows you to scale the data x to the range [0,1].

5. An online intelligent monitoring system for oil and gas content in drilling fluid, used to implement the online intelligent monitoring method for oil and gas content in drilling fluid as described in any one of claims 1 to 4, characterized in that, It includes a data storage module, an intelligent recognition module, and a user interaction module, specifically including the following: The intelligent recognition module is connected to the image database and receives training support from the image database, which provides image data by supplementing laboratory training samples and collecting on-site training samples. The data storage module is connected to the intelligent recognition module to receive recognition results, and also includes an intelligent recognition model encapsulated in the drilling fluid online image recognition device and supports remote data storage in the cloud storage terminal; The user interaction module and data storage module send query commands and display query results, and also include other query terminals for querying.

6. The online intelligent monitoring system for oil and gas content in drilling fluid according to claim 5, characterized in that, The user interaction module includes a historical data table, a drilling fluid gas content monitoring chart, a drilling fluid oil content monitoring chart, a drilling fluid gas content average value window, and a drilling fluid oil content average value window. The historical data table allows users to view all historical monitoring records of drilling fluid gas content and oil content. The drilling fluid gas content monitoring chart and drilling fluid oil content monitoring chart display the historical monitoring value change trends of drilling fluid gas content and oil content over a period of time. The drilling fluid gas content average value window and drilling fluid oil content average value window display the historical monitoring value average of drilling fluid gas content and oil content over a period of time.

Images