A method for determining a post-drilling leakage fluid viscous friction model of a riser disconnection mechanism and data

Abstract

The present application belongs to the technical field of deepwater exploration of ocean oil and gas, and provides a method for determining a model of viscous friction of a riser after separation and drilling leakage, which is fused with mechanism and data, the method comprising: step S1: obtaining original data, including the pressure difference between the top and bottom of the riser, the liquid column height, the cross-sectional area of the inner wall of the pipe, the flow rate of the liquid in the pipe, and the viscous friction of the leakage liquid of the inner wall of the pipe; step S2: analyzing the data characteristics, analyzing and screening the Top-N key features; step S3: processing and classifying the data, using the Min-Max scaling normalization method to perform data feature standardization operation; step S4: determining and optimizing the data-driven model structure, selecting a randomly configured neural network as the core tool; step S5: training the model, alternately executing the network node growth and Dropout regularization algorithm; step S6: testing the model, verifying whether the Top-N key features on the test data are consistent.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of deepwater exploration of offshore oil and gas, and particularly relates to a method for determining a model of viscous friction of a blowout preventer after disconnection of a riser. BACKGROUND

[0002] When the disconnection operation of the riser is performed after the completion of the deepwater riser drilling operation or in an emergency during the drilling process, the viscous friction resistance of the drilling fluid to the inner wall of the pipe needs to be considered, because it directly affects the pump pressure demand, the circulation system design and the well control safety. Factors affecting the friction include the drilling fluid column height, the friction loss being proportional to the volume of the liquid column; the drilling fluid flow rate, the friction pressure drop being proportional to the square of the flow rate. The mechanism research on the viscous friction of the blowout preventer is beneficial to the enrichment and development of the related theory of the deepwater drilling riser and the improvement of the safety of the deepwater drilling work, and provides a new idea for the multi-disciplinary cross-field of petroleum engineering and ocean engineering in China.

[0003] The methods for measuring or calculating the friction mainly include experimental measurement, field monitoring, empirical formula and numerical simulation. The friction calculation method by collecting a large amount of data and analyzing according to the empirical formula includes:

[0004] (1) Darcy-Weisbach equation calculation method, which is suitable for Newtonian fluid (such as water-based drilling fluid), and the key parameters include friction coefficient, hydraulic diameter (riser and drill pipe annulus), drilling fluid density, average flow rate, which are defined as follows:

[0005]

[0006] Among them:

[0007] ΔP is the pressure difference;

[0008] f is the Darcy friction coefficient, which is related to the Reynolds number and the roughness of the pipe;

[0009] L is the length of the pipe;

[0010] D is the inner diameter of the pipe;

[0011] ρ is the density of the liquid;

[0012] v is the flow rate of the liquid.

[0013] (2) Non-Newtonian fluid correction (Herschel-Bulkley model) calculation method, which is suitable for high-viscosity drilling fluid (such as oil-based mud), and the key parameters include consistency coefficient, rheological index and yield stress.

[0014] (3) Numerical simulation (Computational Fluid Dynamics, CFD) method, using CFD software (such as ANSYS Fluent, OpenFOAM) to simulate the flow of drilling fluid in the riser and calculate the friction pressure drop. Its advantages include: it can analyze complex flow states (turbulent flow, secondary flow), and it can analyze actual working conditions such as drill pipe rotation and eccentric annulus; its limitations are high calculation cost and the need for accurate input of rheological parameters.

[0015] Numerical models come in various forms, and the following will be introduced using the full liquid column model as an example. The full liquid column model is a simplified model used to analyze the dynamic behavior of a liquid column in a vertical or inclined pipe, mainly applied in the fields of petroleum engineering, drilling fluid mechanics, and wellbore flow analysis. Its core assumption is to regard the entire liquid column as a continuous, incompressible fluid column, ignoring the complex interaction between gas and liquid phases, suitable for single-phase flow or low-gas-rate liquid column systems.

[0016] The core of the full liquid column model is the hydrostatic equilibrium and momentum conservation, usually combined with the total pressure balance relationship between the hydrostatic column pressure and the flow pressure drop, and its basic form is as follows:

[0017]

[0018] Where:

[0019] f w and f m are the friction coefficients of seawater and mud, respectively;

[0020] ρ w and ρ m are their respective densities;

[0021] v f is the velocity of the fluid relative to the tensioner and its external auxiliary devices;

[0022] a f is the acceleration of the fluid relative to the tensioner and its external auxiliary devices;

[0023] l w is the height of the liquid column of seawater;

[0024] l r is the total length of the riser;

[0025] A e is its hydraulic cross-sectional area, i.e., the difference between the outer ring area and the inner ring area;

[0026] d r is its hydraulic diameter, i.e., the sum of the outer ring circumference and the inner ring circumference.

[0027] Different numerical models have significant differences in applicable conditions, complexity, and considered factors: the full liquid column model is suitable for single-phase liquid column and low gas rate scenarios, has low complexity, and mainly considers hydrostatic pressure and friction loss; the slug flow model is for gas-liquid intermittent flow, has medium to high complexity, and needs to focus on gas-liquid interaction and slug unit dynamic characteristics; the two-fluid model is suitable for high gas rate and complex multiphase flow, has the highest complexity, and needs to finely simulate complex mechanisms such as interphase momentum exchange and interface effects. Each model is selected for use according to actual flow characteristics and precision requirements.

[0028] Note that the traditional modeling method of the viscous friction of the blowout fluid of the riser drilling relies on physical laws or empirical formulas, and has large data sample size, and has problems such as difficulty in parameter calibration and poor adaptability. Combined with existing mechanism analysis, data-driven modeling can significantly improve the generalization ability of the model by mining the internal law of the data. However, there is little existing public data on the blowout fluid of the riser drilling, and the sample size is small, making it difficult to learn large models; at the same time, traditional neural network learning models are prone to overfitting in a small sample environment. In the prior art, the modeling method based on the randomly configured network can be used for dynamic modeling of limited time series data; the Dropout technology combined with the incremental learning characteristics can alleviate the overfitting problem. Note that although the time series data dynamic modeling method of the randomly configured network and the incremental learning characteristics of the Dropout technology have significant advantages in their respective fields, but there is currently no systematic research on the application of the two in the modeling of the viscous friction of the blowout fluid of the riser drilling. SUMMARY

[0029] In order to overcome the dependence of the traditional modeling method of the viscous friction of the blowout fluid of the riser drilling on prior knowledge, and solve the problem of the accuracy of the friction data analysis model under the condition of mechanism fusion data and small sample, the present application proposes a data-driven modeling method for the viscous friction of the blowout fluid of the riser drilling after disengagement, which is a fusion of mechanism and data. The technical solutions adopted to achieve the purpose include the following steps:

[0030] Step S1: Obtain the original data, respectively collect the sample data of the mechanism model and the sensor measurement data, including the pressure difference at the top and bottom of the riser, the liquid column height, the cross-sectional area of the pipe wall, the flow rate of the liquid in the pipe, and the viscous friction of the blowout fluid of the pipe wall, etc.

[0031] Step S2: Analyze the data characteristics, based on the blowout fluid pressure difference, liquid column height, flow rate, and viscous friction data of the riser drilling, analyze and select the Top-N key features, including the peak frequency, time, and equivalent specific time of the entire liquid in the pipe being discharged from the bottom.

[0032] Step S3: processing and classifying data, performing data feature standardization operation using Min-Max scaling normalization method as input and output data of the data-driven model of the RRAM leakage fluid viscous friction in the RRAM, including the pressure difference between the top and bottom of the RRAM, the liquid column height, the flow rate of the liquid in the pipe and the RRAM inner wall leakage fluid viscous friction, the input data selection range is three types of data of liquid column height, flow rate and friction force, and the output data is friction force type data;

[0033] Step S4: determining and optimizing the data-driven model structure, establishing a data-driven model for describing the RRAM inner wall leakage fluid viscous friction, the determination method of the data-driven model for describing the RRAM inner wall leakage fluid viscous friction comprising: fusing multiple mechanism model data and measurement data, selecting a randomly configured neural network as a core tool, and introducing a Dropout algorithm in the deep learning method on the full connection layer structure of the traditional neural network to determine and optimize the model structure;

[0034] The step of establishing a data-driven model for describing the RRAM inner wall leakage fluid viscous friction and determining and optimizing the model structure comprises:

[0035] Step S401: encoding data, encoding the sample data and measurement data set of each model as a training set, and reserving part as a test data set, including the pressure difference between the top and bottom of the RRAM, the liquid column height, the flow rate of the liquid in the pipe, the RRAM inner wall leakage fluid viscous friction;

[0036] Step S402: determining the model architecture, selecting a randomly configured network, gradually adding hidden nodes in an incremental construction manner, and optimizing the number of randomly configured nodes through a supervision mechanism for determining the model structure;

[0037] Step S403: optimizing the model structure, introducing a Dropout algorithm in the process of determining the randomly configured network model structure, randomly discarding hidden layer nodes to realize dynamic sparsification of the network structure for optimizing the model structure;

[0038] Step S5: training the model, training the data-driven model for describing the RRAM inner wall leakage fluid viscous friction by adopting a hybrid training strategy, alternately executing network node growth and Dropout regularization algorithm, and setting: training data x, y, Dropout probability p, randomly configured network constraint parameter λ, and maximum hidden layer node number L max ;

[0039] The output data y of the randomly configured network model applying the Dropout algorithm has a mathematical expression of L node network structure as follows: Wherein x is the input data, y is the output data, L is the number of hidden layer nodes, w l and b lrespectively, are randomly generated weight and bias of the l-th node l is output weight, usually derived by least square method, g() is activation function, usually Sigmoid function can be selected, p is Dropout probability, p l is Dropout mask

[0040] The riser inner wall leakage viscous friction data-driven model training step of the alternating network node growth and Dropout regularization algorithm comprises:

[0041] Step S501: initialize node number L=0, f0(x)=0, e0=y, define residual error: e L =y-f L (x);

[0042] Step S502: incrementally add nodes and randomly generate candidate node parameters, discard part of the nodes until the stop condition is met;

[0043] Each newly added L+1 node generates random weight and bias w L+1 ~U(-a,a), b L+1 ~U(-b,b) to meet uniform sampling or Gaussian distribution, wherein a and b are given constants;

[0044] Calculate the output of the candidate hidden layer node Generate a random Dropout mask for each node, uniformly sample p L+1 ~U(0,1), assign 1 or 0 after comparison operation, which are True or False, i.e. p L+1 =p L+1 >p;

[0045] Update the Dropout hidden layer node output:

[0046] Step S503: calculate the random configuration network output weight Wherein β=[β1…β l ] is the hidden layer output matrix, is the pseudo-inverse of h L (x), h L+1 =[h1…h L+1 ], y is the target output;

[0047] Step S504: update the network and the residual, after adding new nodes, update the network output: f L+1 (x)=f L (x)+β L+ 1h L+1 (x), calculate the new residual: e L+1=y-f L+1 (X);

[0048] Step S505: Check the stop condition: if the training meets the error condition wherein lambda is a preset random configuration network constraint parameter, then stop training; or the network structure reaches the maximum number of nodes L during the training process max , then stop training; otherwise, continue to add nodes.

[0049] Step S6: Test the model, test the data-driven model of the inner wall of the description riser under the leakage liquid viscous friction force, and verify whether the Top-N key features on the test data are consistent.

[0050] The present application improves the accuracy of modeling by combining random configuration network with dynamic Dropout mechanism, and can be applied to the data modeling process of the leakage liquid viscous friction force of the riser drilling. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 : Schematic diagram of random configuration network

[0052] Figure 2 : Data-driven modeling flowchart of drilling leakage liquid viscous friction force integrating mechanism and data

[0053] Figure 3 : Schematic diagram of determination and optimization process of data-driven model of leakage liquid viscous friction force

[0054] Figure 4 : Schematic diagram of training process of data-driven model of leakage liquid viscous friction force

[0055] Figure 5 : Model performance comparison curve

[0056] Figure 6 : Accuracy of data-driven model of leakage liquid viscous friction force of example 1

[0057] Figure 7 : Accuracy of data-driven model of leakage liquid viscous friction force of example 2 DETAILED DESCRIPTION

[0058] In order to clearly and completely describe the technical solutions of the present application and its advantages, the present application will be further described in detail below in combination with the drawings and examples. It should be noted that the specific technical solutions described in the examples herein are only used to explain the present application and do not limit the present application. Based on the examples in the present application, the examples obtained by those skilled in the art using the same inventive concept of the present application should be included in the protection scope of the present application.

[0059] In the existing engineering practice and related research background, combined with the Darcy-Weisbach equation,

[0060]

[0061] and the full liquid column model equation,

[0062]

[0063] According to the published data set of the paper, the model environment parameters and initial conditions are set, including water depth and total liquid column height l r = 1000 m, the water dynamic diameter of the riser d r = 0.4826 m, the friction coefficients of seawater and mud are f w = 0.002, f m = 0.004, the initial conditions of velocity and acceleration are 0, and the simulation results of the simulation data model about the length of the liquid column and the friction force are as follows: the peak time is about 14 s, the peak friction force is about 1.89 × 10 5 N, and the completion time of the equivalent whole pipe drilling fluid from the bottom is about 51 s.

[0064] Note that after the riser is separated, seawater flows into the pipe body from the top valve of the riser, and mixed mud flows into seawater from the bottom of the riser. In order to completely analyze the mechanism characteristics of the viscous friction force of the drilling fluid during the process, the existing mechanism model and data analysis model have the following problems:

[0065] Analysis of the drilling fluid viscous friction force data obtained by the full liquid column model shows that after the mud discharge process is completed, the height of the mud liquid column and the liquid column velocity and acceleration are forced to be 0, i.e. the friction force data after 51 s of completion is not considered. But combined with the actual situation, there is still seawater in the riser, obviously the liquid column velocity and height are not 0, so the full liquid column model of the drilling fluid viscous friction force has part of the mechanism that is not modeled after the completion of the discharge, which needs to be analyzed and the accuracy of the model needs to be improved;

[0066] Analysis of the drilling fluid viscous friction force exponential function model data obtained by model identification on the published data set of the paper shows that the constructed external system can solve the actual problem that the friction force data after completion is not 0. However, this model is completely based on the constructed variable, and does not reflect the length of the liquid column and the velocity state change of the mixed mud and seawater drilling fluid during the discharge, i.e. such friction force time series model has part of the mechanism that is not modeled, and the accuracy of the model needs to be improved.

[0067] To solve the above problems, while dealing with the fact that there is little existing public data on blowout leakage in riser drilling, the sample size is small, and it is difficult to achieve large model learning, the present application proposes a method for determining the blowout leakage viscous friction force model of the riser after disintegration, which describes the relationship between the three types of data of the liquid column height, flow rate and blowout leakage viscous friction force of the blowout leakage in drilling.

[0068] Figure 1 To randomly configure a network diagram, the required data of the network includes the pressure difference between the top and bottom of the riser, the liquid column height, the flow rate of the liquid in the pipe, and the blowout leakage viscous friction force of the inner wall of the riser. The input data selection range is three types of data of the liquid column height, flow rate and friction force, and the output data is the friction force type data.

[0069] Figure 2 The data-driven modeling flowchart of the blowout leakage viscous friction force of the drilling is shown in FIG. 1. Figure 2 To verify the effect of the present application, the following embodiments are provided.

[0070] Embodiment 1

[0071] To verify the effect of the present application, the following embodiments are provided.

[0072] Step S1: Obtain sample data of multiple mechanism models, including the pressure difference between the top and bottom of the riser, the liquid column height, the cross-sectional area of the pipe wall, the flow rate of the liquid in the pipe, and the blowout leakage viscous friction force of the pipe wall, etc.

[0073] Step S2: Analyze the data characteristics, analyze and select the Top-N key features based on the blowout leakage pressure difference, liquid column height, flow rate and viscous friction force data of the riser drilling, and the key features include the peak frequency, time and equivalent whole pipe liquid discharge time of the friction force between the liquid in the pipe and the pipe wall.

[0074] Step S3: Process and classify the data, use the Min-Max scaling normalization method to perform data feature standardization operation, as the input and output data of the blowout leakage viscous friction force data-driven model of the riser drilling, including the liquid column height, the flow rate of the liquid in the pipe, and the blowout leakage viscous friction force of the inner wall of the riser obtained by combining the pressure difference between the top and bottom of the riser. The input data selection is two types of data of the liquid column height and flow rate, and the output data is the friction force type data.

[0075] Step S4: determining and optimizing the data-driven model structure, establishing a data-driven model for describing the viscous friction of the leaked liquid on the inner wall of the riser, the determination method of the data-driven model for describing the viscous friction of the leaked liquid on the inner wall of the riser comprising: fusing multiple mechanism model data, selecting a randomly configured neural network as a core tool, and introducing a Dropout algorithm in deep learning method on the full connection layer structure of the traditional neural network to determine and optimize the model structure;

[0076] Step S5: training the model, training the data-driven model for describing the viscous friction of the leaked liquid on the inner wall of the riser by adopting a hybrid training strategy, alternately executing the network node growth and the Dropout regularization algorithm, and setting:

[0077] The input data x in the training data is the liquid column height and the flow rate, the output data y is the viscous friction data of the leaked liquid, the Dropout probability p is 0.2, the randomly configured network constraint parameter λ is 0.1, and the maximum hidden layer node number L max is 50;

[0078] The data-driven model for describing the viscous friction of the leaked liquid on the inner wall of the riser is trained by alternately executing the network node growth and the Dropout regularization algorithm.

[0079] Step S6: testing the model, testing the trained data-driven model for describing the viscous friction of the leaked liquid on the inner wall of the riser, and verifying whether the Top-N key features on the test data are consistent, the key features including the peak frequency, time and specific time required for the equivalent whole pipe liquid to be discharged from the bottom of the inter-wall friction of the pipe liquid.

[0080] As shown in Figure 3 , the step of establishing the data-driven model for describing the viscous friction of the leaked liquid on the inner wall of the riser and determining and optimizing the model structure comprises:

[0081] Step S401: encoding data, encoding the sample data and the measurement data set of each model as a training set, and reserving part as a test data set, including the pressure difference between the top and bottom of the riser, the liquid column height, the flow rate of the pipe liquid, and the viscous friction of the leaked liquid on the inner wall of the riser;

[0082] Step S402: determining the model architecture, selecting a randomly configured network, gradually adding hidden nodes in an incremental construction manner, and optimizing the number of randomly configured nodes through a supervision mechanism for determining the model structure;

[0083] Step S403: optimizing the model structure, introducing a Dropout algorithm in the process of determining the model structure of the randomly configured network model, randomly discarding hidden layer nodes, realizing dynamic sparsification of the network structure, and optimizing the model structure.

[0084] AsFigure 4 As shown, the casing inner wall leakage viscous friction data-driven model training step of alternately performing network node growth and Dropout regularization algorithm includes:

[0085] Step S501: initialize node number L=0, f0(x)=0, e0=y, define residual error: e L =y-f L (x);

[0086] Step S502: incrementally add nodes and randomly generate candidate node parameters, discard part of the nodes until the stop condition is met;

[0087] Each newly added L+1 node generates random weights and bias w L+1 ~U(-a,a), b L+1 ~U(-b,b) to meet uniform sampling or Gaussian distribution, where a, b are given constants;

[0088] Calculate the candidate hidden layer node output Generate a random Dropout mask for each node, uniformly sample p L+1 ~U(0,1), assign 1 or 0 after comparison operation, True or False, i.e. p L+1 =p L+1 > p;

[0089] Update the Dropout hidden layer node output:

[0090] Step S503: calculate the random configuration network output weight Where β=[β1…β l ] is the hidden layer output matrix, is the pseudo-inverse of h L (x), h L+1 =[h1…h L+1 ], y is the target output;

[0091] Step S504: update the network and the residual, after adding new nodes, update the network output: f L+1 (x)=f L (x)+β L+ 1h L+1 (x), calculate the new residual: e L+1 =y-f L+1 (x);

[0092] Step S505: check the stop condition: if the training meets the error condition , stop training; or the network structure reaches the maximum node number L maxIf yes, stop training; otherwise, continue adding nodes;

[0093] Figure 5 The model performance comparison curve is shown in FIG. 3, in which the data of the three mechanism models are respectively shown, the training data is only from model 1, and the test data is from models 2 and 3. Figure 5 As shown in FIG. 4, compared with the training model, the time to reach the peak frequency and the time to complete the drilling fluid discharge of the output result of the data-driven model are respectively about 0.1s, the modeling result of the Top-N key features meets the requirements, the generalization effect is good, and the data amount is effectively reduced.

[0094] Figure 6 The Mean-Square-Error (MSE) result of the data-driven model of the discharge fluid viscous friction in the embodiment is shown in FIG. 5. Figure 6 As shown in FIG. 5, after the Dropout algorithm is added, the model accuracy is improved, and the modeling accuracy is increased.

[0095] Embodiment 2

[0096] In order to verify the effect of the present application, the discharge fluid viscous friction time series data is used for testing in the embodiment, and the specific steps are as follows:

[0097] Step S1: obtaining sample data, the discharge fluid viscous friction time series data of the inner wall of the riser;

[0098] Step S2: analyzing data features, analyzing and screening Top-N key features, the key features including the peak frequency, time and specific time required for equivalent whole pipe liquid to be discharged from the bottom of the friction between the pipe liquid and the pipe wall;

[0099] Step S3: processing and classifying data, using the Min-Max scaling normalization method to perform data feature standardization operation, as the input and output data of the riser drilling discharge fluid viscous friction data-driven model fused with mechanism and data, the input and output data are both friction time series data;

[0100] Step S4: determining and optimizing the data-driven model structure, establishing the data-driven model for describing the discharge fluid viscous friction of the inner wall of the riser, the data-driven model for describing the discharge fluid viscous friction of the inner wall of the riser determining method includes: selecting a randomly configured neural network as a core tool, and introducing a Dropout algorithm in the deep learning method on the full connection layer structure of the traditional neural network to determine and optimize the model structure;

[0101] Step S5: training the model, adopting a hybrid training strategy to train the data-driven model for describing the discharge fluid viscous friction of the inner wall of the riser, alternately executing network node growth and Dropout regularization algorithm, and setting:

[0102] The input data x in the training data is the liquid column height and the flow rate, the output data y is the leakage liquid viscous friction data, the Dropout probability p is respectively selected as 0.2 and 0.5, the random configuration network constraint parameter λ is 0.1, and the maximum hidden layer node number L max is 50;

[0103] The network node growth and the Dropout regularization algorithm are alternately executed to train the data-driven model of the leakage liquid viscous friction of the inner wall of the riser;

[0104] Step S6: testing the model, the data-driven model of the leakage liquid viscous friction of the inner wall of the riser is tested after training, and whether the Top-N key features on the test data are consistent is verified, the key features including the peak frequency, time and specific time required for the equivalent whole pipe liquid to be discharged from the bottom.

[0105] As Figure 7 shown, the accuracy of the leakage liquid viscous friction data-driven model of example 2 is improved after the Dropout algorithm is added, and the accuracy of modeling on the time series data is effectively improved.

[0106] Although the embodiments of the present application have been described, it can be understood by those skilled in the art that modifications and replacements can be made to the embodiments without departing from the principles of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for determining the viscous friction force model of drilling fluid after riser disintegration, integrating mechanism and data, characterized in that, Includes the following steps: Step S1: Obtain raw data, including sample data of the mechanism model and sensor measurement data, such as pressure difference between the top and bottom of the water-proof pipe, liquid column height, cross-sectional area of ​​the inner wall of the pipe, flow velocity of the liquid in the pipe and viscous friction of the drained liquid in the inner wall of the pipe. Step S2: Analyze data characteristics. Based on the hydraulic pressure difference, fluid column height, flow velocity and viscous friction data of the riser well, analyze and screen its Top-N key features. The key features include the peak frequency and time of the friction between the fluid inside the pipe and the pipe wall, and the specific time required for all the equivalent fluid in the pipe to be discharged from the bottom. Step S3: Process and classify the data, and perform data feature standardization using the Min-Max scaling normalization method. This data serves as the input and output data for the riser drilling downflow fluid viscous friction data-driven model that integrates the mechanism and data. The input data includes the pressure difference between the top and bottom of the riser, the fluid column height, the flow velocity of the fluid inside the riser, and the downflow fluid viscous friction on the inner wall of the riser. The input data selection range includes three types of data: fluid column height, flow velocity, and friction force. The output data is friction force type data. Step S4: Determine and optimize the data-driven model structure, and establish a data-driven model describing the viscous friction of the liquid flowing down the inner wall of the baffle pipe. The method for determining the data-driven model describing the viscous friction of the liquid flowing down the inner wall of the baffle pipe includes: fusing multiple mechanism model data and measurement data, selecting a randomly configured neural network as the core tool, and introducing the Dropout algorithm from deep learning methods on the fully connected layer structure of the traditional neural network to determine and optimize the model structure. Step S5: Train the model. Use a hybrid training strategy to train a data-driven model that describes the viscous friction of the discharge liquid on the inner wall of the riser pipe, and alternately execute the network node growth and Dropout regularization algorithms. Step S6: Test the model. Test the trained data-driven model describing the viscous friction of the drain fluid inside the diaphragm pipe and verify whether the Top-N key features on the test data match.

2. The method for determining the viscous friction force model of drilling fluid after riser dissociation according to claim 1, characterized in that, Step S4, which involves establishing a data-driven model describing the viscous friction of the drain fluid on the inner wall of the riser and determining and optimizing the model structure, includes: Step S401: Encode the data, encode the sample data and measurement dataset of each model as the training set, and reserve a portion as the test dataset, including the pressure difference between the top and bottom of the riser, the height of the liquid column, the flow rate of the liquid in the pipe, and the viscous friction of the drained liquid on the inner wall of the riser. Step S402: Determine the model architecture, select a randomly configured network, and gradually add hidden nodes using an incremental construction method. Optimize the number of randomly configured nodes through a supervision mechanism to determine the model structure. Step S403: Optimize the model structure. In the process of determining the random configuration network model structure, the Dropout algorithm is introduced to randomly discard hidden layer nodes, thereby realizing the dynamic sparsity of the network structure and optimizing the model structure.

3. The method for determining the viscous friction force model of drilling fluid after riser dissociation according to claim 1, characterized in that, Step S5: Train the model. A hybrid training strategy is used to train a data-driven model describing the viscous friction of the discharge fluid inside the riser pipe. Network node growth and Dropout regularization algorithms are executed alternately. The settings are: training data x, y; Dropout probability p; randomly configured network constraint parameters λ; and maximum number of hidden layer nodes L. max ; The mathematical expression for the L-node network structure of the randomly configured network model using the Dropout algorithm to output data y is: Where x represents the input data, y represents the output data, L represents the number of hidden layer nodes, and w l and b l The weights and biases are randomly generated for the l-th node, β. l The output weights are typically derived using the least squares method. `g()` is the activation function, usually a sigmoid function, and `p` is the dropout probability. l This is the Dropout mask.

4. The method for determining the viscous friction force model of drilling fluid after riser dissociation according to claim 3, characterized in that, The training steps for the data-driven model of the viscous friction force of the drain fluid inside the riser, which alternately executes the network node growth and Dropout regularization algorithms, include: Step S501: Initialize the number of nodes L = 0, f0(x) = 0, e0 = y, and define the residual error: e L =yf L (x); Step S502: Add nodes incrementally and randomly generate candidate node parameters, discarding some nodes until the stopping condition is met; Step S503: Calculate the output weights of the randomly configured network. Where β = [β1…β l [ ] represents the hidden layer output matrix. for h L The pseudoinverse of (x), h L+1 =[h1 … h L+1 ], where y is the target output; Step S504: Update the network and residuals. After adding a new node, update the network output: f L+1 (x)=f L (x)+β L+1 h L+1 (x), calculate the new residual: e L+1 =yf L+1 (X); Step S505: Check the stopping condition: If the training meets the error condition Where λ∈(0,1) are preset randomly configured network constraint parameters, training stops; or the network structure reaches the maximum number of nodes L during training. max If the node fails to connect, training will stop; otherwise, continue adding nodes.

5. The method for determining the viscous friction force model of drilling fluid after riser dissociation according to claim 4, characterized in that, Step S502: Add nodes incrementally and randomly generate candidate node parameters, discarding some nodes until the stopping condition is met; For each newly added L+1 node, generate random weights and biases w. L+1 ~U(-a,a),b L+1 ~U(-b,b), satisfying uniform sampling or Gaussian distribution, where a and b are given constants; Calculate the output of candidate hidden layer nodes Dynamically generate random Dropout masks for each node, and uniformly sample to generate p L+1 ~U(0,1) is used for comparison operations, and then assigned the value 1 or 0, which are True or False respectively, i.e., p L+1 =p L+1 >p; Hidden layer node output after Dropout update:

Images