Electricity-to-oil intelligent digital driving method and system self-adaptive to petroleum drilling working conditions

Abstract

The invention discloses an electricity-to-oil intelligent digital driving method and system adaptive to a petroleum drilling working condition, and belongs to the field of petroleum drilling engineering.The method comprises the steps that petroleum drilling data are collected through a multi-source sensor group, and the collected data are sent to an edge gateway controller to be preprocessed; a multi-source data stream obtained through preprocessing is sent to a cloud end, and a multi-source database is constructed; the intelligent torque prediction model is deployed at the cloud end, real-time estimation of the current drill bit load torque is achieved according to data in the multi-source database, and a drill bit prediction torque value is obtained; the cloud end issues the predicted torque value obtained through calculation to the edge gateway controller, and the edge gateway controller generates an actual control instruction according to the received predicted torque value. Safe and stable operation of drilling operation can be effectively guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of oil drilling engineering, and particularly relates to an electrically-driven intelligent digital method and system for replacing oil in oil drilling. BACKGROUND

[0002] With the increasing development of oil and gas fields, the development range is continuously expanded, the drilling operation environment is increasingly complex, and the formation conditions are variable. In the drilling process, when the drill bit passes through different geological layers underground, the pore fluid pressure is often in a dynamic state. With the development of the while-drilling monitoring technology, by integrating sensors into the drilling tools, the formation pressure signal can be directly collected. However, the existing drilling prediction model still has some problems. First, the traditional drilling method relies on experience and manual judgment, which leads to frequent deviation and steering errors in dynamic environments, and is prone to reduce drilling efficiency and increase safety hazards. Although neural networks have been applied to predict the theoretical trajectory of the wellbore, the existing neural network prediction method often uses a single or multiple indicators to judge the threshold value, and the amount of information is small, and the threshold values of different parameters in different mines are not the same. When multiple indicators are close to the threshold value to different degrees, how to comprehensively judge is still not a good solution. In addition, with the increase of drilling depth, the working conditions of the well are more and more complex, and the working environment of the drilling tools is poor, so it is difficult to accurately know the working state of the drill bit in real time. Therefore, there is an urgent need for a method that can accurately evaluate the working state of the drill bit in real time to improve the efficiency and safety of drilling operations. SUMMARY

[0003] One of the objectives of the present application is to provide an electrically-driven intelligent digital method for replacing oil in oil drilling to solve the problem that the working state of the drill bit cannot be accurately known in real time due to the complex working conditions of the well in the prior art.

[0004] The application is realized by the technical scheme, and the electric-oil intelligent digital driving method adaptive to the working condition of oil drilling comprises the following steps: S100, collecting oil drilling data through a multi-source sensor group, sending the collected data to an edge gateway controller for preprocessing, sending the multi-source data stream obtained through preprocessing to the cloud to construct a multi-source database; S200, deploying an intelligent torque prediction model in the cloud, realizing real-time estimation of the current drill bit load torque according to the data in the multi-source database, and obtaining a drill bit predicted torque value, the intelligent torque prediction model comprises: a feature space for reflecting the essential law of the torque signal in the drilling process, a physical coding sub-model for encoding the logging while drilling data into parameters with clear physical meaning, and a prediction engine for prediction according to the feature space and the physical meaning parameters, the prediction engine adopts a double-flow architecture of fusion of a physical sub-model and a data-driven sub-model, wherein the physical sub-model is a set of differential equations constrained by physical laws constructed according to the Rossler model, used for analyzing the essential characteristics of physics to obtain high-dimensional features of the drill bit torque, and the data-driven sub-model autonomously learns and captures complex patterns that the physical model fails to completely describe from the high-dimensional features through a deep learning network; S300, the cloud sends the calculated predicted torque value to the edge gateway controller, and the edge gateway controller generates actual control instructions according to the received predicted torque value.

[0005] Further, the multi-source data comprises electrical working condition data, mechanical working condition data and geological working condition data; the electrical working condition data comprises three-phase current of a motor stator, bus voltage and rotor position data, the mechanical working condition data comprises vibration acceleration data of drilling tools, real-time drilling pressure, rotation speed and mechanical drilling speed data, and the geological working condition data comprises real-time logging data obtained through a logging while drilling system interface.

[0006] Further, the intelligent torque prediction model further comprises a double-time-scale learning framework set for problems caused by formation mutations, the double-time-scale learning framework comprises a fast cycle core with an extended Kalman filter as the core, used for real-time fine-tuning of the physical sub-model to enable it to quickly track short-term working condition changes; and a slow cycle core with meta-learning as the core, used for quickly reshaping the deep learning network of the data-driven sub-model.

[0007] Further, the intelligent torque prediction model is constructed according to the following steps: S210, converting the multi-source heterogeneous measurement data into a standardized feature space capable of revealing the internal dynamics state of the drilling tool system, and converting the original logging curve into a physical coefficient directly affecting the mechanical process by introducing geological prior knowledge, converting the natural gamma into a dimensionless formation hardness coefficient by constructing a formation hardness mapping function, and establishing a conversion relationship from acoustic travel time to pore pressure gradient; S220, in order to realize accurate prediction of the torque signal, a prediction engine is constructed by dual structure of physics and data, the physical sub-model of the prediction engine is constructed according to the Rossler model, the control parameters in the original Rossler model are coupled into the physical coefficients converted from the logging curve, the mechanical operation parameters are taken as the external driving items of the model, and the physical coefficients are taken as the intrinsic parameters of the state evolution equation, to adjust the endogenous variables of the internal evolution law of the Rossler model; the data-driven sub-model predicts the future torque according to the high-dimensional feature tensor calculated by the physical sub-model through a lightweight adaptive spatio-temporal convolution network.

[0008] Further, the standardized feature space is constructed by the following sub-steps: S211, reconstructing the phase space of the signal of the core dynamic torque time series signal of the drill bit to obtain a reconstructed dynamics state vector; S212, encoding the logging while drilling data curve into a parameter with clear physical meaning according to the physical parameters of the formation, to obtain the encoded formation physical parameters; S213, fusing the reconstructed dynamics state vector and the encoded formation physical parameters to form a standardized feature space containing the current motion state of the system and the physical constraints of the environment.

[0009] Further, the reconstructed dynamics state vector is obtained by phase space reconstruction through time delay coordinate embedding, by combining the torque value at the current time and the torque values at multiple historical times into a high-dimensional vector, the topological structure of the high-dimensional system is recovered from the one-dimensional time series, thereby providing a complete state description.

[0010] Further, the reconstructed dynamics state vector can be as follows:

[0011] ,

[0012] wherein, is the reconstructed dynamics state vector, is the time, indicates the reconstructed dynamics state vector at the time ; is the torque value, indicates the torque value at the current time, is the torque value of the past delay time, is a torque value at a specific time in the past, is an embedding dimension, is a delay time; denotes a set of embedding dimensions; is a vector transpose symbol.

[0013] Further, the formation hardness mapping function is constructed based on a Sigmoid function, realizes encoding of the natural gamma value, and thus converts the natural gamma value into a continuous and normalized hardness parameter; and the conversion relationship from the acoustic travel time to the pore pressure gradient is based on a rock physics model and uses acoustic data for first-order approximation, thus realizing conversion from the acoustic travel time to the pore pressure gradient.

[0014] Further, the formation hardness mapping function can be as shown in the following formula:

[0015] ,

[0016] wherein, is a formation hardness coefficient, which is a dimensionless normalized value; is a natural exponential function, is a coefficient for controlling the steepness of the function curve, is a measured natural gamma value while drilling, is a gamma baseline value of a reference formation.

[0017] Further, the conversion relationship from the acoustic travel time to the pore pressure gradient can be as shown in the following formula:

[0018] ,

[0019] wherein, is a Nabla operator, is a pore pressure, is a pore pressure gradient; is an acoustic travel time conversion coefficient calibrated according to regional geological data, is an acoustic travel time, is a linear conversion coefficient calibrated according to regional geological data.

[0020] Further, the physical sub-model and the data-driven sub-model complement each other through a fusion mechanism, which includes a dynamic fusion weight and a pseudo-Hamilton function. The dynamic fusion weight converts the uncertainty gradient of the physical sub-model, external disturbance and unreliability of the data source into a smooth weight through a Sigmoid function, so as to realize dynamic weighting of the physical mechanism and data-driven. The pseudo-Hamilton function follows the principle of energy conservation and system uncertainty, and a pseudo-Hamilton function representing the total energy of the system is constructed. The gradient size of the pseudo-Hamilton function in the phase space is used to measure the degree of deviation of the system from the steady state. When the system fluctuates violently or encounters high uncertainty of formation information, the fitting ability of the data sub-model is trusted. When the system runs smoothly and the formation characteristics are clear, the physical sub-model with clear physical meaning is trusted.

[0021] Further, the physical sub-model based on the Rossler model can be as follows:

[0022]

[0023]

[0024]

[0025] wherein, is the strain energy, is the inertial kinetic energy, is the energy dissipation, is the coupling strength parameter of the formation hardness coefficient, is the coupling strength parameter of the pore pressure, is the mechanical operating parameter, and are standard Rossler parameters for controlling the basic form of the chaotic attractor; denotes derivation with respect to time.

[0026] Further, the data-driven sub-model based on the adaptive spatiotemporal convolution network can be as follows:

[0027] ,

[0028] wherein, is the predicted torque, is the time parameter, denotes the final predicted future torque; is the dynamic fusion weight; denotes a lightweight adaptive spatiotemporal convolution network, is the fusion tensor processed by the ASTCN; is the pseudo-Hamilton function.

[0029] Further, the dynamic fusion weight and the pseudo Hamiltonian function can be calculated by the following formula:

[0030] ,

[0031] ,

[0032] wherein, is the gradient of the pseudo Hamiltonian function, is the Frobenius norm of the matrix, represents the norm of the Hamiltonian function gradient, reflecting the rate of change of system energy; is the standard deviation of the motor input power in a time window, is the confidence factor of lithology data, which is high when the formation lithology is uniform, and vice versa; is a high-dimensional feature tensor, is an energy transfer coefficient.

[0033] Further, the intelligent torque prediction model further comprises: S230, a dynamic adaptive mechanism based on a double-time-scale learning framework, which first determines whether a new formation is drilled through by a mutation detector based on a chaotic index. When the drill bit encounters a formation mutation, the system dynamics behavior changes dramatically, causing the formation mutation detector to produce a detectable sharp pulse, thereby triggering a slow cycle update. When a mutation is detected, a meta-learning optimizer is activated to achieve rapid model reconstruction through an inner-outer two-layer optimization model.

[0034] Further, the mutation detector based on the chaotic index can be as follows:

[0035]

[0036] wherein, represents the estimated value of the local Lyapunov exponent at time t; is a specific time, is an exponential symbol, is a partial differential symbol, is the length of a short time window; represents the phase space state vector at discrete time step k, represents the Jacobian matrix of the state transition of the system from time k to time k+1.

[0037] Further, the meta-learning optimizer can be as follows:

[0038]

[0039] wherein, is the expected value of the task, is the parameter updated after n steps of gradient descent are performed on the task is the parameter updated after n steps of gradient descent are performed on the task is the parameter of the neural network model, that is, the ASTCN neural network parameter of the data-driven model represents a specific training task sampled from all possible stratigraphic mutation events, for example, a drilling process from shale to sandstone, is the overall distribution of the task is the loss function represents the outer-layer updated task is the loss function of the outer-layer updated task, which is usually the error between the predicted torque and the real torque, and in this embodiment, it can be a Huber loss function is the loss function of the inner-layer updated task is the loss function of the inner-layer updated task is the single-step learning rate of the inner loop

[0040] Further, step S300 can further include that the local Lyapunov index calculated by the intelligent torque prediction model realizes vibration monitoring and early warning, the index is a magnifying glass for measuring dynamic stability of a system, is extremely sensitive to impending instability, and sharp increase of the index is usually much earlier than appearance of a significant peak on an FFT spectrum, so that the index can be used as an early warning signal to realize rapid early warning.

[0041] Another aspect of the present application provides an electric-oil intelligent digital driving system for self-adapting to oil drilling conditions, which comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the electric-oil intelligent digital driving method for self-adapting to oil drilling conditions as described in any of the above aspects when executing the program.

[0042] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0043] 1、The present application realizes comprehensive and real-time monitoring of drilling conditions by collecting dynamic change data of electrical state, mechanical state and downhole geological environment of oil drilling in real time through a multi-source sensor group and constructing a multi-source database, overcomes the limitations of traditional experience and manual judgment, effectively reduces deviation and guidance error rate in the drilling process, and improves safety and efficiency of drilling operations.

[0044] 2、The intelligent torque prediction model of the application adopts a millisecond-level online prediction mechanism of nonlinear torque dynamics, combines data in a multi-source database, converts multi-source heterogeneous measurement data into a standardized feature space, and combines geological prior knowledge to encode logging data into parameters with clear physical meaning, realizes effective fusion and conversion of data, provides comprehensive information support for subsequent prediction, and realizes dynamic adaptation to complex formation conditions in combination with a prediction engine of a double-flow architecture, overcomes the limitations of a single judgment basis, improves the accuracy and reliability of prediction, realizes real-time estimation and high-precision prediction of the current load torque, overcomes the limitations of existing static models and qualitative analysis, and significantly improves the matching degree of the predicted bit torque value and the actual situation.

[0045] 3、The application introduces a double-time-scale learning framework to realize millisecond-level real-time self-adaptation and second-level slow cycle optimization, significantly improve the adaptability and learning efficiency of the model, and effectively solve the real-time problem of downhole engineering parameter evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0046] The drawings described herein are used to provide further understanding of the embodiments of the application, constitute a part of the application, and do not constitute a limitation on the embodiments of the application. In the drawings:

[0047] Figure 1 The method flowchart provided for the embodiment 1 of the application.

[0048] Figure 2 The method timing diagram provided for the embodiment 1 of the application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the embodiments of the application more clear, the technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, not all the embodiments. The components of the embodiments of the application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0050] Embodiment 1

[0051] Figure 1 The method flowchart of the adaptive petroleum drilling working condition electric oil intelligent digital driving method in the embodiment is shown, Figure 2 The timing diagram in the embodiment is shown from Figure 1 It can be seen from the above that the embodiment includes the following steps:

[0052] Step 1: The edge gateway controller of the drilling platform collects real-time data related to oil drilling through a multi-source sensor group. The edge gateway controller preprocesses the collected data to obtain multi-source data streams, and sends the preprocessed multi-source data streams to the cloud to build a multi-source database.

[0053] Specifically, in this embodiment, the relevant data includes electrical state, mechanical state, and even dynamic changes of downhole geological environment that can comprehensively and real-timely reflect the working conditions of oil drilling. For example, electrical condition data, mechanical condition data, and geological condition data.

[0054] Among them, the electrical condition data can be collected by the current transformer and voltage sensor installed at the power electronic controller inlet side or the motor junction box. The rotor position data can be obtained by the rotary encoder installed at the motor shaft end.

[0055] The mechanical condition data can be collected by the MEMS accelerometer installed near the top drive device or the rotary table, which can directly reflect the smoothness of the drilling process. The real-time drilling pressure, rotation speed, and mechanical drilling speed data can be obtained through the drilling platform instrument system interface or the dedicated pressure sensor, which are key parameters for reflecting drilling load and efficiency.

[0056] The geological condition data can be obtained through the MWD (logging while drilling) system interface connected to the edge gateway, which can obtain acoustic travel time, natural gamma, and other logging data in real time. These data are used to sense the formation lithology, hardness, and pore pressure in real time.

[0057] Step 2: An intelligent torque prediction model for realizing millisecond-level online prediction of nonlinear torque dynamics is built in the cloud. This model can combine the data in the multi-source database to realize real-time estimation of the current bit load torque and obtain a high-precision bit prediction torque value.

[0058] Specifically, in this embodiment, the intelligent torque prediction model can be built by the following steps:

[0059] 1) First, the received multi-source heterogeneous measurement data is converted into a standardized feature space that can uniformly reveal the internal dynamics of the drilling tool system.

[0060] The standardized feature space can be built by the following steps:

[0061] Firstly, the signal phase space reconstruction is performed on the core dynamic torque time sequence signal of the drill bit to obtain a reconstructed dynamic state vector; then, the logging while drilling data curve is encoded into a parameter with clear physical meaning according to the physical parameters of the stratum to obtain an encoded stratum physical parameter; finally, the reconstructed dynamic state vector and the encoded stratum physical parameter are fused to form a collaborative feature tensor (i.e., a normalized feature space) containing the current motion state of the system and the physical constraints of the environment, thereby providing comprehensive information input for the subsequent prediction engine.

[0062] Generally, the torque signal usually exhibits strong nonlinear characteristics and chaotic characteristics in the drilling process, and the prior art usually analyzes the time sequence values to capture the torque signal in a future window of time. This way is difficult to capture the essential law of the torque signal in the drilling process. Therefore, in the present embodiment, the one-dimensional time sequence signal of the multi-source heterogeneous measurement data is unfolded into a high-dimensional phase space, so that the hidden dynamic structure is revealed.

[0063] Specifically, for the core dynamic torque time sequence signal, the time delay coordinate embedding technology can be used for phase space reconstruction, that is, the torque value at the current time and the torque values at a plurality of historical times are combined into a high-dimensional vector. Through this method, the topological structure of the high-dimensional system can be recovered from the one-dimensional time sequence, thereby providing a complete state description for subsequent dynamic modeling. In the present embodiment, the embedding dimension and the delay time are not randomly selected, but are adaptively determined through Cao's method and mathematical tools such as minimizing mutual information, so as to ensure that the reconstructed phase space can most truly reflect the attractor morphology of the original dynamic system.

[0064] Exemplarily, in the present embodiment, the phase space reconstruction of the torque signal can be represented by the following formula:

[0065]

[0066] wherein, is a phase space state vector, that is, a reconstructed dynamic state vector, is time, indicates the time reconstructed phase space state vector; is a torque value, indicates the torque value at the current time, which represents the system observation value at the current time, and is a single projection of the drilling process dynamics; is the torque value at the past delay time, is the torque value at the past specific time, and each such value represents a state snapshot of the torque value at the historical time; is the embedding dimension, which can be calculated by the pseudo-nearest neighbor method such as Cao's method; is the delay time; represents a set of embedding dimensions; is the vector transpose symbol. The entire vector is a combination of the current and past series of discrete observations, forming a point in an m-dimensional space, which is no longer an isolated numerical value, but contains the state of the system's recent evolution history.

[0067] It should be noted that the state vector is constructed according to the Takens embedding theorem. For a deterministic dynamic system, even if only one variable of the system can be observed, the phase space equivalent to the original system dynamics behavior in topology can still be reconstructed by delay coordinate embedding of the variable. That is, although the real drilling tool-formation system is determined by numerous variables (temperature, pressure, vibration mode, etc.), the shape of its overall behavior (for example, its chaotic attractor) can be recovered from only the torque time series. Through the shown phase space reconstruction formula, a one-dimensional time series is expanded into a high-dimensional geometric object , so that the chaotic dynamics hidden behind the torque signal can be explicitly expressed and analyzed mathematically, providing an initial prerequisite state for subsequent chaos analysis.

[0068] At the same time, in order to make the model better understand the influence of formation characteristics on drilling tool dynamics, it is necessary to encode the measurement while drilling data (MWD) into parameters with clear physical meaning. In this implementation, by introducing geological prior knowledge, the original logging curve is converted into physical coefficients that directly affect the mechanical process. For example, the hardness of the formation is a key factor affecting the efficiency of rock breaking by the drill bit, and the natural gamma value is usually positively correlated with the shale content, thereby indirectly reflecting the hardness of the rock. Therefore, a mapping function can be constructed to convert the natural gamma into a dimensionless formation hardness coefficient; similarly, the acoustic travel time (Δtc) is closely related to the porosity of the rock, and the porosity directly affects the pore pressure, so a conversion relationship from the acoustic travel time to the pore pressure gradient can be established.

[0069] Exemplarily, in this embodiment, the physical encoding formula can be calculated by the following formula:

[0070]

[0071]

[0072] wherein, is the formation hardness coefficient, which is a dimensionless normalized value; is the natural exponential function, is a coefficient that controls the steepness of the function curve, a natural gamma value for measurement-while-drilling, a gamma baseline value for a reference formation; a Nabla operator, a pore pressure, a pore pressure gradient; a sonic transit time conversion coefficient calibrated according to regional geological data, a sonic transit time, a linear conversion coefficient calibrated according to regional geological data.

[0073] It should be noted that the above two formulas aim to convert the original, abstract logging data into parameters with clear physical meaning and can be directly used by physical models. For the formation hardness coefficient, the natural gamma value is usually related to the shale content, and a high gamma value usually corresponds to mudstone (softer lithology), and a low gamma value corresponds to sandstone or carbonate rock (harder lithology). By using the Sigmoid function to encode the gamma value, the qualitative geological knowledge is converted into a continuous, normalized hardness parameter, so that it can be directly embedded as a multiplier term in the subsequent dynamic equation. For the pore pressure gradient, through an empirical rock physics model, this is a commonly used engineering method for first-order approximation using easily accessible sonic data when pore pressure cannot be directly measured. In this embodiment, the two formulas shown can encode the logging curves into values that can be directly calculated by the physical equation.

[0074] 2) After completing the standardized feature space construction, in order to realize accurate prediction of the torque signal, a core prediction engine that can simulate the complex interaction between the formation and the drilling tool needs to be constructed. The prediction engine can combine the physical mechanism of the feature space and data-driven to realize deep collaboration to complete high-precision prediction.

[0075] In this embodiment, the prediction engine adopts a dual-flow architecture that combines physical models and data-driven models. The physical model is based on mechanical principles and constructs a dynamic model that can generate chaotic torque signals for analyzing the physical nature of the characteristics; and the data-driven model learns and captures complex patterns that the physical model cannot fully describe from the high-dimensional features output by the physical model through a deep learning network.

[0076] Specifically, in the embodiment, the physical model can be constructed based on a common chaotic system, Rossler model, and control parameters in the original Rossler model are replaced or coupled with lithology physical parameters converted from the while-drilling data. Thus, the mechanical drilling speed, drilling pressure and other mechanical operation parameters are taken as external driving items of the model; and the formation hardness coefficient and the pore pressure gradient obtained in the previous step are taken as intrinsic parameters of the state evolution equation to adjust endogenous variables of the internal evolution law of the model, so as to construct a differential equation set constrained by physical laws, and the evolution trajectory of the solution of the equation set in the phase space simulates the physical evolution process of the torque state.

[0077] Exemplarily, in the embodiment, the differential phase space physical model based on the Rossler model can be expressed by the following formula:

[0078]

[0079]

[0080]

[0081] wherein, is the strain energy, is the inertial kinetic energy, is the energy dissipation, is the coupling strength parameter of the formation hardness coefficient, is the coupling strength parameter of the pore pressure, is the mechanical operation parameter, and are standard Rossler parameters for controlling the basic form of the chaotic attractor. denotes derivation with respect to time. The final solution of the formula is: which is the high-dimensional characteristic tensor of the reconstructed phase space state vector.

[0082] It should be noted that in the above formula, is the part responsible for rotation and stretching in the standard Rossler model, which builds the basic framework of chaotic behavior, and the latter is a newly added physical driving item, which is also the core modification of the Rossler model in the embodiment, and the mechanical energy externally applied is introduced as the driving force of the system, and the efficiency of this energy injection is modulated by the formation hardness coefficient; when the hard formation is large, it will cause stronger driving effect, thereby simulating the scene that the drill bit vibrates violently on hard rock. In is the key nonlinear term in the standard Rossler model that produces folding effect, which is the source of chaos, and the latter ​is the second core modification of the Rossler model in this embodiment, the pore pressure gradient is introduced into the nonlinear term to represent that the pore pressure of the rock will affect the breaking mode and the rate of energy release of the rock; high pore pressure can make the rock more easily broken, thereby changing the way of energy dissipation of the system, and then affecting the shape of the chaotic attractor.

[0083] Specifically, in this embodiment, the data-driven model can learn to predict the future torque directly from the fused high-dimensional feature tensor through a lightweight adaptive spatio-temporal convolution network (ASTCN). In order to make the two models complementary to each other, it is necessary to design an organic fusion mechanism. In this embodiment, the fusion mechanism follows the principles of energy conservation and system uncertainty, and a pseudo-Hamilton function representing the total energy of the system is constructed, the gradient size of which in the phase space can measure the degree of deviation of the system from the steady state. When the system fluctuates violently (i.e. the gradient is large) or encounters high uncertainty of formation information, the fitting ability of the data-driven model should be relied on more; on the contrary, when the system runs smoothly and the formation characteristics are clear, the physical model with clear physical meaning should be trusted more.

[0084] Exemplarily, in this embodiment, the data-driven model can be represented by the following formula:

[0085] ,

[0086] wherein, is the predicted torque, is a time parameter, represents the final predicted future torque; is a dynamic fusion weight; represents a lightweight adaptive spatio-temporal convolution network, is a fusion tensor processed by the ASTCN; is a pseudo-Hamilton function.

[0087] Exemplarily, in the data-driven model of this embodiment, the dynamic fusion weight and the pseudo-Hamilton function can be calculated by the following formula:

[0088] ,

[0089] ,

[0090] wherein, is the gradient of the pseudo-Hamilton function, is the Frobenius norm of the matrix, represents the norm of the Hamilton function gradient, reflecting the energy change rate of the system; is the standard deviation of the motor input power in a time window, is a confidence factor of lithology data, which is high when the lithology of the formation is uniform, and low otherwise; is an energy transfer coefficient. The dynamic fusion weight formula converts the uncertainty gradient of the physical model, external disturbance and the unreliability of the data source into a smooth weight through a Sigmoid function, so as to realize the intelligent and dynamic balance between the physical mechanism and the data-driven.

[0091] 3) The core prediction engine constructed is still a static prediction model in nature and cannot well cope with the complex conditions of the formation in the actual drilling process. In order to realize millisecond-level online self-adaptation to cope with the drill tool wear caused by the sudden change of the formation after the prediction engine is established, in the embodiment, an adaptive mechanism capable of online and rapid self-evolution is introduced, so that the model can realize self-adaptive learning.

[0092] In the embodiment, considering that the formation change needs to be responded in seconds (slow cycle) and the torque fluctuation needs to be compensated in real time (fast cycle), a double-time-scale learning framework is introduced to construct a dynamic adaptive mechanism, so as to realize millisecond-level online self-adaptation. In the millisecond-level fast cycle, the learning framework uses extended Kalman filtering as the fast cycle core to real-time fine-tune the mechanical coupling coefficients in the physical model, so that it can quickly track short-term working condition changes. In the second-level slow cycle, the Meta-Learning framework is used as the slow cycle core to realize rapid remodeling of the entire neural network.

[0093] Specifically, in the embodiment, the mechanism first judges whether a new formation is drilled through a mutation detector based on a chaos index. When the drill bit encounters a sudden change in the formation, the dynamic behavior of the system will change dramatically, resulting in a sharp pulse that can be detected by the formation mutation detector, thereby triggering slow cycle updating.

[0094] Once the mutation is detected, the Meta-Learning optimizer is activated. Unlike traditional online learning, the model is trained on various formation change tasks in the pre-training stage, and through internal-external two-layer optimization, the initial parameters learned by the model are forced to be in a special position: from this position, no matter which new formation is encountered next, only a small adjustment (one or two steps of gradient descent) is needed to quickly achieve good performance. Thus, rapid model reconstruction of the formation mutation in seconds or even shorter time is realized, greatly improving the adaptive ability of the model.

[0095] Exemplarily, in the embodiment, the formation mutation detector can be as shown in the following formula:

[0096]

[0097] wherein, represents an estimate of the local Lyapunov exponent at time t; for a specific time instant, for the exponential symbol, for the partial differential symbol, for the length of a short time window; represents the phase space state vector at discrete time step k, represents the Jacobian matrix of the state transition of the system from time k to k+1.

[0098] It should be noted that the formation abruptness detector in the embodiment is constructed based on the Lyapunov exponent. The interaction between the drill bit and the rock is essentially a complex dynamic process. When the drill bit drills in a homogeneous formation (such as a large section of mudstone), the entire dynamic system is in a relatively stable state, which can be imagined as a system running on a particular "attractor". The dynamic behavior (such as sensitivity and degree of chaos) of the system has statistical consistency. When the drill bit suddenly drills from one lithology to another completely different lithology (for example, from soft mudstone to hard and dense limestone), the physical law controlling the dynamic system will change fundamentally. This will inevitably lead to a change in the structure of the dynamic system attractor, which is manifested as a dramatic change in the local divergence / convergence characteristics of the phase space orbit. Therefore, the value of the Lyapunov exponent is a perfect formation abruptness detector. When it increases sharply, it indicates that the underlying dynamic law of the system has changed, and the existing model parameters may no longer be applicable. This signal can be used as a trigger condition to activate a deeper model adaptation mechanism, such as meta-learning in a slow loop.

[0099] Exemplarily, in the embodiment, the meta-learning optimizer updated in the slow loop can be calculated by the following formula:

[0100]

[0101] wherein, is the expected value of the task, is the parameter updated after n steps of gradient descent on the task ; is the parameter of the neural network model, that is, the ASTCN neural network parameter of the data-driven model; represents a specific training task sampled from all possible formation abruptness events, for example, a drilling process from shale to sandstone, is the overall distribution of the task; is the loss function, represents the task a loss function, usually the error between predicted torque and true torque, in this embodiment can be Huber loss function; the task for inner loop update a loss function, a single step learning rate for inner loop.

[0102] It should be noted that in the above formula, the outer loop optimization, which is the core of the whole optimization, its goal is not to minimize the loss of the model on the current task, but to minimize the loss of the model on the new task after the inner loop update, that is, the object of the outer loop optimization is a parameter obtained by the inner loop update , whose optimization goal is not to minimize the direct performance of the model on any single task, but to minimize the expected performance after rapid adaptation. And is the inner loop update part, which describes a simulated and rapid fine-tuning process for any given lithology mutation task , the inner loop update calculates the loss with a small amount of data of the task, and updates the current model parameters one or several times along the gradient direction, to obtain a temporary new parameter that has adapted to the task ; That is, the essence of the inner loop update is to simulate how the model will quickly adjust itself when it actually encounters new lithology on the line.

[0103] Step 3: The cloud end will issue the predicted torque value calculated by the intelligent torque prediction model to the edge gateway controller. The edge gateway controller will generate actual control instructions from the received predicted torque value.

[0104] And according to the local Lyapunov exponent calculated by the intelligent torque prediction model to realize vibration monitoring and early warning, the index is a magnifying glass for measuring the dynamic stability of the system, which is extremely sensitive to impending instability (such as stick-slip vibration), and the sharp increase of the index usually comes much earlier than the appearance of obvious peaks on the FFT spectrum, which can be used as an early warning signal to realize rapid warning.

[0105] A vibration suppression threshold can also be set. When the monitored index exceeds the threshold, the edge gateway controller automatically superimposes a compensation torque instruction on the generated control instruction. The compensation torque instruction is a sine wave or pulse width modulation (PWM) signal with the same frequency and opposite phase as the vibration main frequency, and its amplitude is positively correlated with the monitored vibration amplitude or the over-limit degree of Lyapunov exponent.

[0106] The above detailed description of the specific embodiments of the present application has been given to understand the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An intelligent digital drive method for electric oil drilling that adapts to different oil drilling conditions, characterized in that, The intelligent digital driving method includes, S100: Collects oil drilling data through a multi-source sensor group, sends the collected data to the edge gateway controller for preprocessing, and sends the preprocessed multi-source data stream to the cloud to build a multi-source database; The S200, with its cloud-based intelligent torque prediction model, uses data from multiple databases to estimate the current drill bit load torque in real time and obtain the predicted drill bit torque value. The intelligent torque prediction model includes: The feature space used to reflect the essential laws of torque signals during drilling. Used to encode logging-while-drilling data into physical coding sub-models with parameters of explicit physical meaning. And prediction engines that make predictions based on feature space and physical meaning parameters, The prediction engine employs a dual-stream architecture that integrates a physical sub-model and a data-driven sub-model. The physical sub-model is a set of differential equations constrained by physical laws, constructed based on the Rossler model. It is used to analyze the essential physical characteristics and obtain the high-dimensional features of drill bit torque. The data-driven sub-model, through a deep learning network, autonomously learns from high-dimensional features and captures complex patterns that the physical model fails to fully describe. The S300 and cloud-based systems will send the calculated predicted torque values ​​to the edge gateway controller. The edge gateway controller generates actual control commands based on the received predicted torque value.

2. The intelligent digital drive method for adaptive oil drilling conditions according to claim 1, characterized in that, The multi-source data includes: electrical condition data, mechanical condition data, and geological condition data; The electrical operating data includes: three-phase stator current and bus voltage of the motor, as well as rotor position data. Mechanical operating data includes: drill string vibration acceleration data, real-time drill pressure, rotational speed, and mechanical drilling speed data. Geological condition data includes real-time logging data obtained through the logging-while-drilling system interface.

3. The intelligent digital drive method for adaptive oil drilling conditions according to claim 1, characterized in that, The intelligent torque prediction model also includes: A dual-timescale learning framework was designed to address the problems caused by abrupt changes in geological formations. The dual-timescale learning framework includes: A fast loop kernel based on extended Kalman filtering is used to fine-tune the physical sub-model in real time, enabling it to quickly track short-term operating condition changes. The slow-loop kernel, with meta-learning at its core, enables the rapid reshaping of deep learning networks that drive data-driven sub-models.

4. The adaptive oil drilling condition-based intelligent digital drive method for oil replacement, as described in claim 1, is characterized in that... The intelligent torque prediction model is constructed according to the following steps: S210. Transform multi-source heterogeneous measurement data into a standardized feature space that can reveal the intrinsic dynamic state of the drilling system. Meanwhile, by introducing prior geological knowledge, the original logging curves are transformed into physical coefficients that directly influence the geological process. By constructing a formation hardness mapping function, the natural gamma is transformed into a dimensionless formation hardness coefficient, and the conversion relationship from sonic transit time to pore pressure gradient is established. S220. To achieve accurate prediction of torque signals, a prediction engine is constructed through a dual structure of physical and data components. The physical sub-model of the prediction engine is constructed based on the Rossler model. By coupling the control parameters in the original Rossler model into physical coefficients derived from well logging curves, and by using mechanical operating parameters as external driving terms of the model, The physical coefficients are used as intrinsic parameters of the state evolution equation to adjust the endogenous variables of the internal evolution law of the Rossler model; The data-driven sub-model predicts future torque based on the high-dimensional feature tensor calculated by the physical sub-model using a lightweight adaptive spatiotemporal convolutional network.

5. The adaptive oil drilling condition-based intelligent digital drive method for oil replacement, as described in claim 4, is characterized in that... The standardized feature space is constructed through the following sub-steps: S211. The core dynamic torque timing signal of the drill bit is reconstructed in phase space to obtain the reconstructed dynamic state vector. S212. Based on the physical parameters of the formation, the logging-while-drilling data curves are encoded into parameters with clear physical meanings to obtain the encoded formation physical parameters. S213. The reconstructed dynamic state vector is fused with the encoded formation physical parameters to form a standardized feature space that includes the current motion state of the system and the physical constraints of the environment.

6. The intelligent digital drive method for adaptive oil drilling conditions according to claim 4, characterized in that, The formation hardness mapping function is constructed based on the Sigmoid function, which encodes the natural gamma value, thereby converting the natural gamma value into a continuous, normalized hardness parameter. The conversion relationship between acoustic transit time and pore pressure gradient is based on a rock physics model using acoustic data for a first-order approximation, thereby realizing the conversion between acoustic transit time and pore pressure gradient.

7. The intelligent digital drive method for adaptive oil drilling conditions according to claim 4, characterized in that, The physical sub-model and the data-driven sub-model complement each other through a fusion mechanism. The fusion mechanism includes dynamic fusion weights and a pseudo-Hamilton function. The dynamic fusion weights use a sigmoid function to transform the uncertain gradients of the physical sub-model, external perturbations, and the unreliability of the data source into a smooth weight, thereby achieving a dynamic trade-off between physical mechanisms and data-driven approaches. The pseudo-Hamiltonian function follows the principles of energy conservation and system uncertainty. By constructing a pseudo-Hamiltonian function that characterizes the total energy of the system, the magnitude of the gradient of the pseudo-Hamiltonian function in phase space is used to measure the degree to which the system deviates from steady state. When the system experiences severe fluctuations or encounters high uncertainty in stratigraphic information, the fitting ability of the data sub-model is trusted. When the system operates smoothly and the formation characteristics are clear, then a physical sub-model with clear physical meaning is trusted.

8. The intelligent digital drive method for adaptive oil drilling conditions according to claim 5, characterized in that, The reconstructed dynamic state vector is obtained by phase space reconstruction using time-delayed coordinate embedding. By combining the torque value at the current moment with the torque values ​​at multiple historical moments into a high-dimensional vector, the topology of the high-dimensional system can be recovered from the one-dimensional time series, thus providing a complete state description.

9. The intelligent digital drive method for adaptive oil drilling conditions according to claim 4, characterized in that, The intelligent torque prediction model also includes: S230, A dynamic adaptive mechanism based on a dual-timescale learning framework. The aforementioned dynamic adaptive mechanism first uses a mutation detector based on chaotic indices to determine whether new formations have been encountered during drilling. When the drill bit encounters a sudden change in formation, the dynamic behavior of the system changes drastically, causing the formation change detector to generate a sharp pulse that can be detected, thereby triggering a slow cycle update. When a mutation is detected, the meta-learning optimizer is activated, which optimizes the model through inner and outer layers to achieve rapid model reconstruction.

10. An intelligent digital drive system for electric oil drilling that adapts to different oil drilling conditions, characterized in that: The intelligent digital drive system includes: processor; The memory stores a computer program that, when executed by a processor, implements the intelligent digital drive method for adaptive oil drilling conditions as described in any one of claims 1 to 9.

Images