Knowledge-embedded drilling speed intelligent prediction system and method thereof
The knowledge-embedded drilling speed prediction system integrates drilling mechanisms with AI, constrained by physical laws, to address instability and unreliability in existing models, enhancing prediction accuracy and reliability.
Patent Information
- Application Number
- CN202411258209.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2024-12-10
AI Technical Summary
Existing drilling speed intelligent prediction models suffer from instability, low reliability, and poor mobility due to their reliance on unexplainable 'black box' artificial intelligence, and lack of domain knowledge integration, especially in the drilling field where downhole data is scarce and rock crushing mechanisms are unclear.
A knowledge-embedded drilling speed prediction system that integrates drilling mechanisms with AI, using neural networks constrained by physical laws and equations, incorporating drilling knowledge through structural optimization and fusion, to ensure stable and reliable predictions.
The system enhances the stability, reliability, and accuracy of drilling speed predictions by adhering to physical laws and incorporating domain knowledge, improving the model's performance and reducing errors.
Smart Images
Figure CN119106244A_ABST
Abstract
Description
Technical Field
[0001] The invention provides a knowledge-embedded drilling speed intelligent prediction system and method thereof, belonging to the technical field of drilling engineering. Background Art
[0002] As a high-risk, high-investment project, drilling models must be stable, reliable and explainable, which are necessary conditions for the on-site promotion and application of all drilling models. Otherwise, it is very likely to mislead drilling planning and increase drilling costs and risks. However, with the continuous deepening of the process of intelligent drilling, a large number of mechanical drilling speed intelligent prediction models have been applied and tested at the drilling site. The shortcomings of the existing models, such as weak migration and poor stability, have also been revealed, which has seriously restricted the process of intelligent drilling.
[0003] The conventional drilling speed intelligent prediction model inputs data into an unexplainable artificial intelligence "black box model" to obtain the corresponding output results. The calculation process is uncontrollable and may not conform to the already clear physical laws. The input results have low stability, weak mobility, and poor reliability. Once the input data differs greatly from the training set, the accuracy of the model will be greatly reduced.
[0004] CN118396043A provides a mechanical drilling speed prediction method based on dual-branch incremental learning. Although the patent proposes a prediction model based on a new artificial intelligence algorithm, which improves the prediction accuracy of the model, the field requires not only high accuracy, but also high reliability and mobility. The stability, reliability and mobility of this method are difficult to guarantee.
[0005] In terms of improving the stability and transferability of the drilling speed intelligent prediction model, a reliable method is to embed the knowledge of the drilling field into the artificial intelligence model. Knowledge embedding of the artificial intelligence model refers to representing physical knowledge in a form that can be processed by computers, so that the established intelligent model follows objective physical laws or facts during the working process, avoiding extremely unreasonable prediction results. In other industrial fields, relevant research on domain knowledge embedding has been carried out.
[0006] However, there is still a lack of research in the field of drilling rate prediction. Compared with other industrial fields, the embedding of domain knowledge in the drilling rate intelligent prediction model faces many difficulties. On the one hand, due to the high cost of downhole drilling measurement tools and the lack of downhole data, the downhole state cannot be accurately described, and the mechanical drilling rate can only be directly predicted based on surface logging and mud logging data; on the other hand, most of the current drilling rate equations are regression equations obtained through data statistics, and the rock crushing mechanism is still not completely clear, lacking guidance from regular equations such as the Bernoulli equation and the Darcy equation. The knowledge-embedded drilling rate intelligent prediction model is also one of the most important challenges that the drilling intelligent transformation process needs to face. Summary of the invention
[0007] The present invention extracts complex knowledge such as the correlation between drilling parameters and mechanical penetration rate, boundary conditions, confidence intervals, etc., forms knowledge embedding methods through structural optimization, physical constraints and model fusion, builds a coupling framework of multiple drilling knowledge embedding methods, and proposes a knowledge-embedded penetration rate intelligent prediction system and method.
[0008] The present invention integrates artificial intelligence algorithms with drilling mechanisms, constrains the calculation process of the "black box" drilling speed intelligent prediction model, ensures that the calculations are performed and the results are output within the defined drilling physics laws and confidence intervals, and improves the stability, migration performance and reliability of the model.
[0009] The specific technical solution is
[0010] First, the present invention provides a knowledge-embedded drilling speed intelligent prediction system, including intelligent drilling equipment, a data management module, a drilling knowledge module, a neural network algorithm module, a network structure optimization module, a neural network physical constraint module, a mechanism and data model fusion module, and a model training and prediction module;
[0011] Intelligent drilling equipment for real-time monitoring, measurement and collection of surface and downhole drilling data;
[0012] The data management module is used to receive and store the measurement data of the intelligent drilling equipment, and perform data cleaning, data extraction, and feature analysis on the received drilling data;
[0013] Drilling knowledge module, with built-in boundary conditions of drilling multiple parameters, the influence of multiple parameters on mechanical drilling speed, commonly used drilling speed equations, and expert knowledge input interface;
[0014] Neural network algorithm module, embedded with multiple network structures such as fully connected neural network, linear cross network, and linear network;
[0015] The network structure optimization module calls and combines different neural network algorithms in different neural network algorithm modules based on the correlation between multiple parameters and mechanical drilling speed, thereby fully fitting the complex correlation between multiple parameters and mechanical drilling speed;
[0016] The neural network physical constraint module extracts the influence relationship between multiple drilling parameters and mechanical drilling speed, confidence intervals, and boundary condition knowledge based on drilling knowledge, and converts it into a form that the neural network can understand. By optimizing the neural network loss function and output neuron activation function, soft constraints and hard constraints are added respectively, so that the neural network model conforms to the already defined physical mechanism;
[0017] The mechanism and data model fusion module uses an integrated learning algorithm to fuse the drilling rate prediction physical mechanism model and the neural network model to build the final drilling rate prediction model driven by the mechanism and data.
[0018] The model training and prediction module analyzes, extracts and processes the target well data according to the needs of drilling rate prediction of the target well, divides the data into training set and test set, and builds an intelligent drilling rate prediction model embedded with drilling knowledge based on hyperparameter search and model early stopping; inputs the drilling parameters of the target well into the established model, calculates and outputs the final predicted drilling rate, and provides a reference for drilling operators.
[0019] Secondly, the present invention provides a knowledge-embedded drilling speed intelligent prediction method, which adopts the knowledge-embedded drilling speed intelligent prediction system; the method comprises the following steps:
[0020] (1) The intelligent drilling equipment collects the data measured on the ground and underground and outputs it to the data management module;
[0021] (2) The data management module receives the measurement data of the intelligent drilling equipment, and stores, processes and analyzes it; transmits the feature analysis results to the network structure optimization module; receives the command of the model training and prediction module, extracts the drilling data after relevant data processing and transmits it to the model training and prediction module;
[0022] (3) The drilling knowledge module receives commands from the model training and prediction module, extracts the drilling speed equation according to requirements and transmits it to the model training and prediction module; receives instructions from the neural network physical constraint module, and inputs the boundary conditions of drilling multi-parameters, the influence of multi-parameters on mechanical drilling speed, commonly used drilling speed equations, and expert knowledge into the neural network physical constraint module according to requirements;
[0023] (4) The neural network algorithm module receives the command from the network structure optimization module, calls the relevant neural network according to the requirements and passes it to the network structure optimization module;
[0024] (5) The network structure optimization module receives the feature analysis results of the data management module and forms a new neural network architecture according to the multi-parameter correlation relationship; according to the new neural network architecture, the network model in the neural network algorithm module is called to combine and form a new network model; and the combined neural network model is input into the neural network physical constraint module;
[0025] (6) The neural network physical constraint module receives the drilling knowledge from the drilling rate prediction physical mechanism module and converts it into a form that can be understood by the computer. It converts the response knowledge into soft constraints and hard constraints of the neural network through special methods; it receives the neural network model from the network structure optimization module, imposes physical constraints on it, and passes the constrained neural network model to the mechanism and data model fusion module;
[0026] (7) The mechanism and data model fusion module receives the neural network model under physical constraints from the neural network physical constraint module, forms a model architecture driven by the mechanism and data based on integrated learning, and outputs the architecture to the model training and prediction module;
[0027] (8) The model training and prediction module analyzes the data management module data according to the drilling speed prediction requirements, extracts the adjacent well data and performs standardization processing, and divides the data into a training set and a test set according to the proportion; calls the drilling speed equation in the drilling knowledge module according to the expert requirements, builds the optimal drilling speed prediction model based on the drilling speed equation based on the training set and the test set data, and inputs the drilling speed equation into the model architecture driven by the mechanism and data; receives the model architecture driven by the mechanism and data of the mechanism and data model fusion module, and the formed optimal drilling speed prediction model based on the drilling speed equation, embeds the optimal drilling speed equation based on the drilling speed equation into the model architecture driven by the mechanism and data, uses the training set and the test set to intelligently optimize the hyperparameters and train to build the optimal knowledge-embedded drilling speed prediction model; receives the drilling speed parameters that the operator wants to predict, and inputs them into the optimal knowledge-embedded drilling speed prediction model.
[0028] The present invention provides a knowledge-embedded drilling speed intelligent prediction system and method thereof, which greatly improves the stability, reliability and accuracy of conventional drilling speed intelligent prediction and optimization models. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a schematic diagram of the system structure of the present invention;
[0030] Figure 2 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0031] The specific technical solution of the present invention is described in conjunction with the accompanying drawings.
[0032] like Figure 1 As shown, a knowledge-embedded drilling speed intelligent prediction system includes intelligent drilling equipment, a data management module, a drilling knowledge module, a neural network algorithm module, a network structure optimization module, a neural network physical constraint module, a mechanism and data model fusion module, and a model training and prediction module;
[0033] Intelligent drilling equipment is used to monitor, measure and collect surface and underground drilling data in real time.
[0034] The data management module is used to receive and store the measurement data of the intelligent drilling equipment, and perform data cleaning, data extraction, and feature analysis on the received drilling data.
[0035] The drilling knowledge module has built-in boundary conditions of multiple drilling parameters, the influence of multiple parameters on mechanical drilling speed, commonly used drilling speed equations, expert knowledge input interface, etc.
[0036] The neural network algorithm module embeds various network structures such as fully connected neural network, linear cross network, and linear network.
[0037] The network structure optimization module calls and combines different neural network algorithms in different neural network algorithm modules based on the correlation between multiple parameters and mechanical drilling speed, thereby fully fitting the complex correlation between multiple parameters and mechanical drilling speed.
[0038] The neural network physical constraint module extracts the influence relationship between multiple drilling parameters and mechanical drilling speed, confidence intervals, boundary conditions and other knowledge based on drilling knowledge, and converts it into a form that can be understood by the neural network. By optimizing the neural network loss function, output neuron activation function, etc., soft constraints and hard constraints are added respectively, so that the neural network model conforms to the already defined physical mechanism.
[0039] The mechanism and data model fusion module uses an integrated learning algorithm to fuse the drilling rate prediction physical mechanism model and the neural network model to build the final drilling rate prediction model driven jointly by mechanism and data.
[0040] The model training and prediction module analyzes, extracts and processes the target well data according to the needs of drilling rate prediction of the target well, divides the data into training set and test set, and builds an intelligent drilling rate prediction model embedded with drilling knowledge based on hyperparameter search and model early stopping; inputs the drilling parameters of the target well into the established model, calculates and outputs the final predicted drilling rate, and provides a reference for drilling operators.
[0041] like Figure 2 As shown, a knowledge-embedded drilling speed intelligent prediction method includes the following steps:
[0042] (1) The intelligent drilling equipment collects the data measured on the ground and underground and outputs it to the data management module.
[0043] (2) The data management module receives the measurement data of the intelligent drilling equipment, and stores, processes, and analyzes it; transmits the feature analysis results to the network structure optimization module; receives the command of the model training and prediction module, extracts the drilling data after relevant data processing, and transmits it to the model training and prediction module;
[0044] (3) The drilling knowledge module receives commands from the model training and prediction module, extracts the drilling speed equation according to requirements and transmits it to the model training and prediction module; receives instructions from the neural network physical constraint module, and inputs the boundary conditions of drilling multi-parameters, the influence of multi-parameters on mechanical drilling speed, commonly used drilling speed equations, expert knowledge, etc. into the neural network physical constraint module according to requirements;
[0045] (4) The neural network algorithm module receives the command from the network structure optimization module, calls the relevant neural network according to the requirements and passes it to the network structure optimization module;
[0046] (5) The network structure optimization module receives the feature analysis results of the data management module and forms a new neural network architecture according to the multi-parameter correlation relationship; according to the new neural network architecture, the network model in the neural network algorithm module is called to combine and form a new network model; and the combined neural network model is input into the neural network physical constraint module.
[0047] (6) The neural network physical constraint module receives the drilling knowledge from the drilling rate prediction physical mechanism module and converts it into a form that can be understood by the computer. It converts the response knowledge into soft constraints and hard constraints of the neural network through special methods. It receives the neural network model from the network structure optimization module and imposes physical constraints on it, and passes the constrained neural network model to the mechanism and data model fusion module.
[0048] (7) The mechanism and data model fusion module receives the neural network model under physical constraints from the neural network physical constraint module, forms a model architecture driven by the joint mechanism and data based on integrated learning, and outputs the architecture to the model training and prediction module.
[0049] (8) The model training and prediction module analyzes the data management module data according to the drilling speed prediction requirements, extracts the neighboring well data and performs standardization processing, and divides the data into training set and test set according to a certain ratio; calls the drilling speed equation in the drilling knowledge module according to the expert requirements, builds the optimal drilling speed prediction model based on the drilling speed equation based on the training set and test set data, and inputs the drilling speed equation into the model architecture driven by the mechanism and data; receives the model architecture driven by the mechanism and data fusion module, and the formed optimal drilling speed prediction model based on the drilling speed equation, embeds the optimal drilling speed equation based on the drilling speed equation into the model architecture driven by the mechanism and data, uses the training set and test set to intelligently optimize the hyperparameters and train to build the optimal knowledge-embedded drilling speed prediction model; receives the drilling speed parameters that the operator wants to predict, and inputs them into the optimal knowledge-embedded drilling speed prediction model.
Claims
1. A knowledge-embedded drilling speed intelligent prediction system, characterized in that: It includes intelligent drilling equipment, data management module, drilling knowledge module, neural network algorithm module, network structure optimization module, neural network physical constraint module, mechanism and data model fusion module, model training and prediction module; Intelligent drilling equipment for real-time monitoring, measurement and collection of surface and downhole drilling data; The data management module is used to receive and store the measurement data of the intelligent drilling equipment, and perform data cleaning, data extraction, and feature analysis on the received drilling data; Drilling knowledge module, with built-in boundary conditions of drilling multiple parameters, the influence of multiple parameters on mechanical drilling speed, commonly used drilling speed equations, and expert knowledge input interface; Neural network algorithm module, embedded with multiple network structures such as fully connected neural network, linear cross network, and linear network; The network structure optimization module calls and combines different neural network algorithms in different neural network algorithm modules based on the correlation between multiple parameters and mechanical drilling speed, thereby fully fitting the complex correlation between multiple parameters and mechanical drilling speed; The neural network physical constraint module extracts the influence relationship between multiple drilling parameters and mechanical drilling speed, confidence intervals, and boundary condition knowledge based on drilling knowledge, and converts it into a form that the neural network can understand. By optimizing the neural network loss function and output neuron activation function, soft constraints and hard constraints are added respectively, so that the neural network model conforms to the already defined physical mechanism; The mechanism and data model fusion module uses an integrated learning algorithm to fuse the drilling rate prediction physical mechanism model and the neural network model to build the final drilling rate prediction model driven by the mechanism and data. The model training and prediction module analyzes, extracts and processes the target well data according to the needs of drilling rate prediction of the target well, divides the data into training set and test set, and builds an intelligent drilling rate prediction model embedded with drilling knowledge based on hyperparameter search and model early stopping; inputs the drilling parameters of the target well into the established model, calculates and outputs the final predicted drilling rate, and provides a reference for drilling operators.
2. A knowledge-embedded drilling speed intelligent prediction method, characterized in that: The drilling speed intelligent prediction system with knowledge embedding according to claim 1 is adopted; the method comprises the following steps: (1) The intelligent drilling equipment collects the data measured on the ground and underground and outputs it to the data management module; (2) The data management module receives the measurement data of the intelligent drilling equipment, and stores, processes and analyzes it; transmits the feature analysis results to the network structure optimization module; receives the command of the model training and prediction module, extracts the drilling data after relevant data processing and transmits it to the model training and prediction module; (3) The drilling knowledge module receives commands from the model training and prediction module, extracts the drilling speed equation according to requirements and transmits it to the model training and prediction module; receives instructions from the neural network physical constraint module, and inputs the boundary conditions of drilling multi-parameters, the influence of multi-parameters on mechanical drilling speed, commonly used drilling speed equations, and expert knowledge into the neural network physical constraint module according to requirements; (4) The neural network algorithm module receives the command from the network structure optimization module, calls the relevant neural network according to the requirements and passes it to the network structure optimization module; (5) The network structure optimization module receives the feature analysis results of the data management module and forms a new neural network architecture according to the multi-parameter correlation relationship; according to the new neural network architecture, the network model in the neural network algorithm module is called to combine and form a new network model; and the combined neural network model is input into the neural network physical constraint module; (6) The neural network physical constraint module receives the drilling knowledge from the drilling rate prediction physical mechanism module and converts it into a form that can be understood by the computer. It converts the response knowledge into soft constraints and hard constraints of the neural network through special methods; it receives the neural network model from the network structure optimization module, imposes physical constraints on it, and passes the constrained neural network model to the mechanism and data model fusion module; (7) The mechanism and data model fusion module receives the neural network model under physical constraints from the neural network physical constraint module, forms a model architecture driven by the mechanism and data based on integrated learning, and outputs the architecture to the model training and prediction module; (8) The model training and prediction module analyzes the data management module data according to the drilling speed prediction requirements, extracts the adjacent well data and performs standardization processing, and divides the data into a training set and a test set according to the proportion; calls the drilling speed equation in the drilling knowledge module according to the expert requirements, builds the optimal drilling speed prediction model based on the drilling speed equation based on the training set and the test set data, and inputs the drilling speed equation into the model architecture driven by the mechanism and data; receives the model architecture driven by the mechanism and data fusion module, and the formed optimal drilling speed prediction model based on the drilling speed equation, embeds the optimal drilling speed equation based on the drilling speed equation into the model architecture driven by the mechanism and data, uses the training set and the test set to intelligently optimize the hyperparameters and train to build the optimal knowledge-embedded drilling speed prediction model; receives the drilling speed parameters that the operator wants to predict, and inputs them into the optimal knowledge-embedded drilling speed prediction model.