A control method and system for a rotary steerable system

Abstract

The present application relates to the technical field of control adjustment, in particular to a control method and system of a rotary steering system, comprising the following steps: obtaining original attitude data of the rotary steering system; obtaining an attitude estimation value; obtaining an attitude angle; obtaining a control instruction; obtaining a filtered signal and a derivative; and outputting a control signal of the rotary steering system; through data preprocessing, attitude calculation, fault compensation, signal smoothing and time lag cancellation, a closed-loop optimized path is formed, the control precision and stability of the rotary steering system are effectively improved, the influence of faults on the control performance is reduced, the smooth processing of the control instruction and the time lag compensation of the extreme learning machine effectively suppress the oscillation of the control signal, and the dynamic response characteristics of the system are improved; the multi-level control architecture can adapt to the changes of the complex downhole environment, ensure the rotary steering system to accurately drill according to the predetermined trajectory, and improve the drilling efficiency and wellbore quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of control adjustment technology, in particular to a control method and system of a rotary steering system. BACKGROUND

[0002] Rotary steerable drilling technology (RSS) is a key technology in the field of oil and gas exploration and development, which allows the drilling tools to adjust the direction of the drill bit accurately while rotating continuously, to achieve complex well trajectory control.

[0003] Compared with traditional sliding directional drilling, RSS can significantly improve drilling efficiency, well trajectory smoothness and wellbore cleanliness, and is particularly suitable for the development of unconventional oil and gas resources such as deep sea and shale gas.

[0004] However, the existing control method of rotary steering system still faces many challenges. First, the downhole geological environment is extremely complex, with factors such as formation hardness, ground stress, friction between drilling tools and well wall, downhole vibration and high temperature, which make the system show strong nonlinearity, time-varying, hysteresis and uncertainty.

[0005] In particular, the strong vibration noise near the drill bit position can seriously drown the original signal of the sensor, resulting in a significant decrease in attitude measurement accuracy. Although traditional PID, robust control and other methods can solve some problems to a certain extent, they are strongly dependent on accurate models and prior knowledge, and are difficult to cope with these complex unknown disturbances and model uncertainties. In addition, in the design of the control system, it is also difficult to efficiently process the multi-modal data processing problem caused by multi-variable and multi-link coupling. SUMMARY

[0006] The main purpose of the present application is to provide a control method of a rotary steering system, which aims to solve the problem of control time lag when the existing control method in the prior art faces vibration noise.

[0007] To achieve the above-mentioned purpose, the present application provides a control method of a rotary steering system, which comprises the following steps:

[0008] Obtain the original attitude data of the rotary steering system to adjust the filter parameters through the Kalman filtering algorithm, and output the attitude estimation value;

[0009] According to the attitude estimation value, output the attitude angle of the rotary steering system through the rotation speed compensation;

[0010] Output the control instruction of the attitude angle after fault compensation through the radial basis function neural network;

[0011] Smooth the control instruction, and obtain the filter signal of the control instruction and the derivative of the filter signal;

[0012] According to the filtered signal and the derivative of the filtered signal, the time delay is compensated by an extreme learning machine, and a control signal of the rotary steering system is output.

[0013] Optionally, the acquiring the original attitude data of the rotary steering system to adjust the filtering parameter by the Kalman filtering algorithm and outputting the attitude estimation value comprises the following steps:

[0014] The analog output signal of the acquisition module is obtained, and after being digitized, it is transmitted to the control module through the communication interface;

[0015] The dynamic model of the rotary steering system is obtained, and the optimal covariance at the last time is transmitted to the current using the dynamic model, and after adding the process noise covariance, an updated mean square difference matrix is output;

[0016] According to the mean square difference matrix, the predicted state is corrected by the Kalman gain, and the attitude estimation value is output.

[0017] Optionally, according to the attitude estimation value, the attitude angle of the rotary steering system is output by speed compensation, which comprises the following steps:

[0018] The attitude estimation value is obtained, and the optimal covariance matrix is updated for attitude solution to output the attitude angle of the rotary steering system, wherein the attitude angle includes the inclination angle and the tool face angle.

[0019] Optionally, the control instruction compensated for the attitude angle by the radial basis function neural network comprises the following steps:

[0020] The fault dynamic function is obtained, the attitude estimation value is mapped to a high-dimensional space by a nonlinear activation function, and then multiplied by a weight vector to obtain an estimated value of the fault signal;

[0021] The observation error is obtained, and the observation error and the integral of the observation error are added to obtain an integral sliding surface;

[0022] The fault-tolerant control law is obtained, and the control instruction is generated by superimposing the compensating force to converge the attitude estimation value of the rotary steering system to the integral sliding surface.

[0023] Optionally, the fault-tolerant control law is obtained, and the control instruction is generated by superimposing the compensating force to converge the attitude estimation value of the rotary steering system to the integral sliding surface, which comprises the following steps:

[0024] The nominal control law is obtained;

[0025] The fault-tolerant control law is generated by detecting and estimating the fault signal by the radial basis function neural network, and the fault-tolerant control law is used to generate a compensating force to offset the influence of the fault signal on the rotary steering system;

[0026] The nominal control law is superimposed with the fault-tolerant control law to output a control instruction.

[0027] Optionally, the compensating for the execution time delay through the extreme learning machine according to the filtered signal and the derivative of the filtered signal, and outputting the control signal of the rotary steering system comprises the following steps:

[0028] An approximation function of the filtered signal and the derivative of the filtered signal under the extreme learning is obtained, and the execution time delay is modeled;

[0029] An adaptive law of the extreme learning machine is obtained to update the network weight, and a compensation signal is obtained to offset the execution time delay;

[0030] The filtered signal, the derivative and the compensation signal are combined to obtain a control signal, and the control signal is used to drive the rotary steering system.

[0031] To achieve the above object, the application further provides a control system, which comprises:

[0032] A collection module is configured to obtain original attitude data of the rotary steering system, adjust a filtering parameter through a Kalman filtering algorithm, and output an attitude estimation value;

[0033] A compensation module is configured to output an attitude angle of the rotary steering system through speed compensation according to the attitude estimation value;

[0034] A control module is configured to output a control instruction compensated for faults through a radial basis function neural network according to the attitude angle;

[0035] The control module is further configured to smooth the control instruction, and obtain a filtered signal of the control instruction and a derivative of the filtered signal;

[0036] The control module is further configured to compensate for an execution time delay through an extreme learning machine according to the filtered signal and the derivative of the filtered signal, and output a control signal of the rotary steering system.

[0037] Optionally, the control module comprises an observer embedded with a radial basis function neural network, and the radial basis function neural network is configured to detect an estimated fault signal.

[0038] Optionally, the control module further comprises a command filter configured to obtain the filtered signal and the derivative of the filtered signal.

[0039] Optionally, the control module further comprises an extreme learning machine configured to obtain an approximation function of the filtered signal and the derivative of the filtered signal under the extreme learning, and model the execution time delay;

[0040] The extreme learning machine is also used to obtain an adaptive law to update network weights and obtain a compensation signal to offset execution time lag.

[0041] The extreme learning machine is also used to obtain a control signal by combining a filtered signal, a derivative and a compensation signal, and the control signal is used to drive the rotary steering system.

[0042] The present application can achieve the following beneficial effects:

[0043] The present application forms a closed-loop optimization path through data preprocessing, attitude calculation, fault compensation, signal smoothing and time lag offsetting, effectively improves the control accuracy and stability of the rotary steering system, and further effectively improves the control accuracy and stability of the rotary steering system in a complex downhole environment. The Kalman filter algorithm of dynamic parameter adjustment significantly improves the anti-noise capability of attitude estimation, reduces the attitude angle measurement error, the radial basis function neural network realizes adaptive compensation for system faults, reduces the influence of faults on control performance, the smoothing processing of control instructions and the time lag compensation of extreme learning machine effectively suppress the oscillation of control signal, and improve the dynamic response characteristics of the system. The multi-level control architecture can adapt to the changes of the complex downhole environment, ensure the rotary steering system to accurately drill according to the predetermined trajectory, and improve the drilling efficiency and wellbore quality. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale.

[0045] Figure 1 The flowchart of the control method in embodiment 1 of the present application.

[0046] The implementation of the present application, functional features and advantages will be further described with reference to the drawings. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0048] It should be noted that all directionality indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, motion condition, etc. between components in a certain posture, and if the certain posture changes, the directionality indications will also change accordingly.

[0049] In the present application, unless otherwise explicitly specified and limited, the terms "connection", "fixation" and the like should be understood broadly, for example, "connection" can be fixed connection, or detachable connection, or integral; can be mechanical connection, or electrical connection; can be direct connection, or indirect connection through an intermediate medium; can be internal communication of two elements or interaction relationship between two elements, unless otherwise explicitly limited. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0050] In addition, if the present application has a description of "first", "second", etc., the description of "first", "second", etc. is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing throughout the text includes three parallel solutions. Taking "A and / or B" as an example, it includes A solution, or B solution, or A and B solution. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor within the scope of protection claimed by the present application.

[0051] Embodiment 1

[0052] Referring to Figure 1 The present embodiment provides a control method of a rotary steering system, the control method comprising the following steps:

[0053] Obtaining original attitude data of the rotary steering system to adjust filtering parameters by a Kalman filtering algorithm and outputting an attitude estimation value;

[0054] According to the attitude estimation value, outputting an attitude angle of the rotary steering system by a rotation speed compensation;

[0055] Outputting a control instruction for the attitude angle after fault compensation by a radial basis function neural network;

[0056] Smoothing the control instruction, and obtaining a filtering signal of the control instruction and a derivative of the filtering signal;

[0057] According to the filtered signal and the derivative of the filtered signal, the time delay in execution is compensated by the extreme learning machine, and a control signal of the rotary steering system is output.

[0058] It should be noted that in the conventional existing rotary steering drilling technology, the original attitude data of the sensor is seriously disturbed by vibration noise under the complex downhole environment, the time-varying characteristics of the formation parameters cause the mismatch of the dynamic model, and the multi-modal data generated by the multi-link coupling is difficult to be efficiently processed by the traditional control architecture. The angle estimation error is accumulated due to the insufficient noise suppression capability in the attitude solving process, the control command oscillation is caused by the dynamic compensation of the fault signal and the execution time delay, and finally the trajectory tracking accuracy of the rotary steering system is reduced and the dynamic response is delayed.

[0059] For example, in deep sea drilling operation, when the drilling tool rotates at a speed of 150 revolutions per minute, the signal-to-noise ratio of the original signal collected by the triaxial accelerometer near the drill bit is less than 20 dB, and the Kalman filter with fixed parameters cannot adaptively adjust the process noise covariance, resulting in that the attitude estimation value deviates from the true value by more than 0.5 degrees. The nonlinear change of the drill-string-wall contact force caused by the sudden change of the formation hardness cannot be matched by the fixed weight vector of the radial basis function neural network in real time, and the observation error integral sliding surface is over-adjusted; the second derivative signal output by the command filter is not associated with the execution mechanism time delay model, the phase of the control signal lags by 50 milliseconds, and a steady-state deviation is generated between the set value and the angle of the steering wing rib.

[0060] Based on the above problems, the embodiment provides a control method of a rotary steering system, which combines Kalman filtering, radial basis function neural network and extreme learning machine to construct a multi-stage compensation mechanism, and solves the attitude control problem of the rotary steering system under strong nonlinear and time-varying interference through the synergistic effect of noise suppression, fault tolerance and time delay compensation.

[0061] Specifically, the original attitude data of the rotary steering system is obtained, and then the Kalman filtering algorithm is used to dynamically adjust the filtering parameters and output the attitude estimation value. The Kalman filtering algorithm optimizes the state estimation in real time through two stages of prediction and update. In the prediction stage, the current state is predicted by using the system model, and in the update stage, the prediction result is corrected by combining the measurement data. The dynamic adjustment of the filtering parameters can adapt to the change of the system noise characteristics and improve the attitude estimation accuracy.

[0062] After obtaining the attitude estimation value, the attitude angle of the rotary steering system is output through speed compensation. The speed compensation considers the influence of the drilling tool rotation on the attitude measurement, and then the measurement error caused by the rotation is eliminated by the compensation algorithm to obtain a more accurate attitude angle.

[0063] Next, the attitude angle is obtained, and a control instruction compensated for failure is output through a radial basis function neural network. The radial basis function neural network has good nonlinear mapping capability, can adaptively learn the failure characteristics of the system, and through online training, the neural network continuously optimizes its weight parameters, realizes real-time compensation for failure, and outputs more reliable control instructions.

[0064] Then, the control instruction is smoothed to obtain a filtered signal of the control instruction and a derivative of the filtered signal. The smoothing can reduce high-frequency oscillation of the control instruction and improve system stability. The filtered signal and the derivative information of the filtered signal provide a basis for subsequent time delay compensation.

[0065] Finally, an actuating time delay is compensated for through a limit learning machine to output a control signal of the rotary steering system. The limit learning machine has fast learning capability and can model and compensate for the time delay characteristics of the actuator in real time. By combining the filtered signal, the derivative information, and the time delay compensation, a final control signal is generated to drive the rotary steering system to perform accurate control.

[0066] The above multi-level control architecture forms a closed-loop optimization path through data preprocessing, attitude calculation, failure compensation, signal smoothing, and time delay cancellation, effectively improving the control accuracy and stability of the rotary steering system. Furthermore, the control accuracy and stability of the rotary steering system in complex downhole environments are effectively improved. The Kalman filter algorithm with dynamic parameter adjustment significantly improves the noise resistance of attitude estimation, reduces attitude angle measurement error, and the radial basis function neural network realizes adaptive compensation for system failure, reducing the impact of failure on control performance. The smoothing of the control instruction and the limit learning machine time delay compensation effectively suppress the oscillation of the control signal and improve the dynamic response characteristics of the system. This multi-level control architecture can adapt to changes in complex downhole environments and ensure that the rotary steering system accurately drills along the predetermined trajectory, improving drilling efficiency and wellbore quality.

[0067] In some embodiments, the Kalman filter algorithm refers to an algorithm that uses linear system state equations and observation data to optimally estimate the state of the system. Specifically, the filter parameters can be recursively adjusted in two steps of prediction and update, noise suppression is achieved by minimizing the mean square error matrix, and the attitude estimation accuracy is improved in complex noise environments.

[0068] In some embodiments, the rotation speed compensation refers to dynamic correction of the attitude angle of the rotary steering system. Specifically, the optimal covariance matrix can be updated by using a dynamic model combined with process noise covariance to eliminate measurement lag and error caused by rotation, ensuring the accuracy of attitude calculation.

[0069] In some embodiments, the radial basis function neural network refers to a neural network with a single hidden layer feedforward structure, which can specifically map the input to a high-dimensional space using a nonlinear activation function, linearly combine the input through a weight vector to approximate the fault signal, realize online estimation and compensation of the fault dynamic function, and enhance the fault tolerance of the system.

[0070] In some embodiments, the smoothing process refers to filtering and differentiating the control instruction, which can specifically generate a continuous and derivable filtered signal and its derivative using a command filter to avoid oscillation of the actuator caused by instruction mutation and improve control stability.

[0071] In some embodiments, the extreme learning machine refers to a training method of a single hidden layer feedforward neural network, which can specifically initialize the input weight randomly and adaptively adjust the output weight to approximate the nonlinear function of the execution time delay, generate a compensation signal to offset the time delay effect, and ensure the real-time performance of the control signal.

[0072] In the embodiment, the original attitude data of the rotary steering system is acquired to adjust the filtering parameters by the Kalman filtering algorithm, and the attitude estimation value is outputted, including the following steps:

[0073] The analog output signal of the acquisition module is acquired, digitized, and transmitted to the control module through the communication interface;

[0074] The dynamic model of the rotary steering system is acquired to use the dynamic model to transmit the optimal covariance at the last time to the current, add the process noise covariance to output the updated mean square error matrix;

[0075] The mean square error matrix is acquired to output the attitude estimation value by correcting the predicted state through the Kalman gain.

[0076] It should be noted that the analog signal of the acquisition module can reduce the distortion in the signal transmission process after digital processing, the communication interface ensures efficient data transmission to the control module, the dynamic model uses the optimal covariance at the last time for state prediction, and the mean square error matrix is updated after superimposing the process noise covariance. This process can quantify the influence of model error on prediction. The Kalman gain is dynamically adjusted according to the updated mean square error matrix, and the predicted state is linearly weighted and corrected, wherein the larger the gain value is, the smaller the influence of measurement noise on the estimation result is. For example, the state transition matrix can be constructed based on the quaternion attitude update model, the process noise covariance is set to a diagonal matrix in the range of 0.01-0.1 according to the downhole vibration intensity, and the mean square error matrix is updated and the Kalman gain is dynamically adjusted, which can effectively suppress the accumulation of model error in a strong vibration noise environment and improve the convergence speed and accuracy of the attitude estimation value.

[0077] It also needs to be explained that based on the above process, the strong noise interference in the complex downhole environment is effectively handled, the accuracy and reliability of the attitude measurement are improved, thereby the rotary steering system can more accurately perceive its own attitude, provide reliable basic data for subsequent trajectory control, and further improve the overall drilling efficiency and wellbore quality.

[0078] In the embodiment, the output of the attitude angle of the rotary steering system according to the rotation speed compensation based on the attitude estimation value includes the following steps:

[0079] The attitude estimation value is obtained, and the optimal covariance matrix is updated before the attitude solution to output the attitude angle of the rotary steering system, wherein the attitude angle includes the inclination angle and the tool face angle.

[0080] It can be understood that based on the attitude estimation value output by the Kalman filter, the optimal covariance matrix is first updated based on the current measurement noise covariance, for example, the inverse matrix of the covariance matrix is weighted to eliminate the gyro drift error caused by high-speed rotation; then the attitude estimation value is converted from the rotating coordinate system of the drilling tool to the fixed geographic coordinate system through the quaternion differential equation or the direction cosine matrix, for example, the accelerometer and three-axis magnetic flux gate data are fused to solve the direction of the gravity vector; finally, the inclination angle is realized by calculating the included angle between the tool axis and the gravity vector, and the tool face angle is solved by projecting the rotation angle to the horizontal plane, for example, the arctangent function is used to calculate the included angle between the magnetic north direction and the projection direction of the drilling tool. Through dynamic updating of the covariance matrix, the noise interference in the attitude solution process is effectively suppressed, and the solution accuracy of the inclination angle and the tool face angle is improved to provide reliable input for subsequent fault compensation of the radial basis function neural network.

[0081] In the embodiment, the output of the control instruction for the attitude angle after fault compensation by the radial basis function neural network includes the following steps:

[0082] The fault dynamic function is obtained, the attitude estimation value is mapped to a high-dimensional space through a nonlinear activation function, and then multiplied by a weight vector to obtain an estimation value of the fault signal;

[0083] The observation error is obtained, and the observation error and the integral of the observation error are added to obtain an integral sliding surface;

[0084] The fault-tolerant control law is obtained to obtain a compensation force, and the control instruction is generated after the attitude estimation value of the rotary steering system is converged to the integral sliding surface.

[0085] It can be understood that after the attitude estimation value is input into the radial basis function neural network, nonlinear space mapping is performed through a Gaussian kernel function, wherein a kernel function center point is generated according to historical fault data clustering, and a kernel width parameter is optimized through a gradient descent method. The mapped high-dimensional feature vector is multiplied by a dynamically adjusted weight vector, and amplitude and phase estimation values of a fault signal are output. An observation error is calculated through a difference between a current attitude angle and a reference trajectory, and an integral term is sequentially accumulated through a discrete time integrator. An integral sliding mode surface superimposes the error and the integral term according to a preset proportion to form a convergence condition with anti-interference capability. A fault-tolerant control law generates a compensation force according to a sliding mode surface state, and the compensation force is superimposed with a nominal control instruction in a time domain, wherein a compensation force gain coefficient is determined through Lyapunov function stability analysis. By adjusting the weight vector and the integral sliding mode surface parameters in real time, it is ensured that the control instruction can accurately track the preset trajectory when sensor noise and actuator faults exist at the same time.

[0086] It can also be understood that based on the above process, faults in the rotary steering system can be effectively detected and compensated, and the robustness and control accuracy of the system are improved, so that the system can maintain stable operation in a complex geological environment and reduce downtime and economic losses caused by faults. In summary, the method can estimate fault signals in real time through a radial basis function neural network, combine an integral sliding mode control strategy, quickly respond to and compensate various unknown faults and disturbances, and thus ensure accurate control of the drilling trajectory.

[0087] In the embodiment, the obtaining of the fault-tolerant control law to generate a control instruction after the attitude estimation value of the rotary steering system is converged to an integral sliding mode surface by the compensation force includes the following steps:

[0088] Obtaining a nominal control law;

[0089] After the fault signal is detected and estimated by the radial basis function neural network, a fault-tolerant control law is generated, and the fault-tolerant control law is used to generate a compensation force to offset the influence of the fault signal on the rotary steering system;

[0090] Superimposing the nominal control law and the fault-tolerant control law to output a control instruction.

[0091] It should be noted that the nominal control law is designed as an inverse model based on the rotational steering system dynamics model, and the output thereof contains the basic components of the pushing force and the biasing force. When the radial basis function neural network detects a decrease in actuator efficiency or a sensor offset fault, the network extracts fault features through the nonlinear mapping of the hidden layer, and the output layer weight matrix is updated online at a learning rate of 0.01-0.1 to generate a compensation force matching the fault amplitude. The integral term of the fault-tolerant control law is used to eliminate the steady-state error, and the integral action time window is limited to 10-50 milliseconds to avoid overshoot. The superposition process of the nominal control law and the fault-tolerant control law is completed synchronously within the control cycle, and the output components of the two are linearly superimposed in the force / torque dimension to form a composite control instruction containing the basic control amount and the fault compensation amount. The composite control instruction is applied to the steering pushing block through the actuator, so that the tracking error of the actual attitude angle and the desired trajectory converges to within ±0.5°.

[0092] It should also be noted that through the above technical solutions, the present application realizes real-time detection and compensation of faults of the rotational steering system, the radial basis function neural network can effectively estimate unknown fault signals, the fault-tolerant control law can generate accurate compensation forces to offset the effects of faults, and the superposition of the nominal control and the fault-tolerant control guarantees the control performance of the system in normal and fault states, thereby improving the reliability and robustness of the rotational steering system and ensuring the trajectory control accuracy under complex geological conditions.

[0093] In the embodiment, the compensation of the execution time delay by the extreme learning machine according to the filtered signal and the derivative of the filtered signal, and output of the control signal of the rotational steering system comprises the following steps:

[0094] Obtaining the approximation function of the filtered signal and the derivative of the filtered signal under the extreme learning, and modeling the execution time delay;

[0095] Obtaining the adaptive law of the extreme learning machine to update the network weight, and obtaining the compensation signal to offset the execution time delay;

[0096] Combining the filtered signal, the derivative, and the compensation signal to obtain the control signal, the control signal being used to drive the rotational steering system.

[0097] It should be noted that the filtered signal and its derivative are input into the extreme learning machine, which is mapped to a high-dimensional space through randomly generated hidden layer parameters, and the output layer weight is dynamically adjusted according to the real-time error, so that the function is approximated with exponential convergence speed. The actual time delay model. The adaptive law updates the weight matrix by gradient descent of the error signal, and the compensation signal and the filtered signal derivative are linearly superimposed to offset the phase lag caused by the time delay. In the control signal generation process, the single hidden layer structure of the extreme learning machine avoids the traditional neural network from falling into local optimum, and its millisecond-level training speed meets the real-time control demand of the downhole. After compensation, the time delay of the control signal is reduced to within 5 milliseconds, effectively improving the dynamic response performance of the rotary steering system.

[0098] It should also be noted that the above process can effectively compensate for the execution time delay and improve the control accuracy and response speed of the rotary steering system. The introduction of the extreme learning machine enables the system to have adaptive ability and can cope with complex and variable downhole environments. At the same time, through the smoothing processing of the control instruction and the use of derivative information, the robustness and stability of the system are further enhanced. This data-driven intelligent control method overcomes the dependence on accurate models of traditional control methods and provides a new solution for high-precision control of the rotary steering system.

[0099] Embodiment 2:

[0100] A control system is provided in this embodiment, which comprises:

[0101] The acquisition module is configured to obtain the original attitude data of the rotary steering system, adjust the filtering parameters through the Kalman filtering algorithm, and output the attitude estimation value;

[0102] The compensation module is configured to output the attitude angle of the rotary steering system through speed compensation according to the attitude estimation value;

[0103] The control module is configured to output the control instruction compensated for faults through the radial basis function neural network according to the attitude angle;

[0104] The control module is further configured to smooth the control instruction and obtain the filtered signal and the derivative of the filtered signal of the control instruction;

[0105] The control module is further configured to compensate for the execution time delay through the extreme learning machine according to the filtered signal and the derivative of the filtered signal, and output the control signal of the rotary steering system.

[0106] It can be understood that when the traditional rotary steering system operates in the complex downhole environment, there is an execution time delay in the process of transmitting the control signal to the actuator, which causes a deviation between the control signal and the actual state of the system, affecting the steering accuracy.

[0107] The embodiment further provides a rotary steering system, comprising a collection module, a compensation module and a control module. The collection module acquires original attitude data and outputs an attitude estimation value through a Kalman filtering algorithm; the compensation module outputs an attitude angle after speed compensation based on the attitude estimation value; and the control module generates a control instruction after fault compensation through a radial basis function neural network, smoothes the control instruction to obtain a filtered signal and a derivative of the filtered signal, and then compensates for execution time lag through a extreme learning machine and outputs a final control signal.

[0108] The control module comprises a command filter, a radial basis function neural network and an extreme learning machine. The command filter performs low-pass filtering on the control instruction to suppress high-frequency noise and extract a signal derivative. The extreme learning machine adopts a single-hidden-layer feedforward neural network structure, the hidden layer node activation function is a Gaussian function, the input layer receives the filtered signal and the derivative of the filtered signal, and the output layer generates a compensation signal. The network weight is updated online through an adaptive law, and the control signal is generated by superimposing the compensation signal and the filtered signal.

[0109] More specifically, after the control instruction is processed by the command filter, the filtered signal retains the low-frequency component of the instruction, and the derivative reflects the signal trend. The extreme learning machine establishes an approximation model of the execution time lag based on the filtered signal and the derivative of the filtered signal, adjusts the network weight through an adaptive law, estimates the time lag in real time and generates a compensation signal. After the compensation signal is superimposed with the filtered signal, the phase deviation caused by the response delay of the execution mechanism is offset. For example, the number of hidden layer nodes is set to 10-20, and the learning rate of the adaptive law is selected from the range of 0.01-0.1, which ensures the convergence speed while avoiding weight oscillation. When the final control signal drives the steering mechanism, the compensated signal is synchronized with the actual state of the system, and the dynamic response accuracy of the rotary steering system is improved.

[0110] As a preferred embodiment, the preferred implementation process of the embodiment scheme is as follows:

[0111] The collection module acquires original attitude data of the rotary steering system, adjusts filtering parameters through a Kalman filtering algorithm, outputs an attitude estimation value, the compensation module acquires the attitude estimation value, outputs an attitude angle of the rotary steering system through speed compensation, the control module acquires the attitude angle, outputs a control instruction after fault compensation through a radial basis function neural network, the control module further acquires the control instruction, smoothes the control instruction, acquires a filtered signal of the control instruction and a derivative of the filtered signal, and the control module further acquires the filtered signal and the derivative of the filtered signal, compensates for execution time lag through an extreme learning machine, and outputs a control signal of the rotary steering system.

[0112] The acquisition module can include an accelerometer and a three-axis fluxgate for measuring acceleration and magnetic field intensity of the rotary steerable system, the compensation module can include a rotational speed sensor and a compensation algorithm for measuring the rotational speed of the drilling tool and compensating the attitude estimation value, and the control module can include an observer, a command filter and an extreme learning machine, the observer being embedded with a radial basis function neural network for detecting an estimated fault signal. The command filter is used to obtain a filtered signal and a derivative of the filtered signal. The extreme learning machine is used to obtain an approximation function of the filtered signal and the derivative of the filtered signal under extreme learning, model an execution time delay, obtain an adaptive law to update network weights, obtain a compensation signal to offset the execution time delay, and combine the filtered signal, the derivative and the compensation signal to obtain a control signal.

[0113] Through the above technical solution, the strong nonlinearity, time variation, hysteresis and uncertainty problems in the complex downhole environment are effectively handled. The acquisition module improves the accuracy of the attitude estimation through the Kalman filtering algorithm. The compensation module improves the accuracy of the attitude angle through the rotational speed compensation. The control module realizes fault compensation through the radial basis function neural network and compensates the execution time delay through the extreme learning machine.

[0114] In the embodiment, the control module includes an observer embedded with a radial basis function neural network for detecting an estimated fault signal.

[0115] It can be understood that, in the operation process of the observer, first, the attitude estimation value is input into the input layer of the radial basis function neural network. The Gaussian function of the hidden layer performs nonlinear transformation on the input. The output layer generates a fault estimation value through the multiplication of the weight vector and the output of the hidden layer. The center vector and the width parameter of the hidden layer are determined in advance through sample training. The weight vector is adjusted online according to the real-time observation error. When the downhole vibration causes the sensor signal to be abnormal, the neural network separates the fault component through high-dimensional space mapping and superimposes the component as a compensation signal into the control command. Through dynamic adjustment of the weight vector, the neural network can adaptively track the change of the fault signal, thereby maintaining the robustness of fault detection in a strong noise environment. The structure effectively solves the problem of insufficient fault separation capability of the traditional observer under nonlinear interference, and improves the fault tolerance performance of the control system.

[0116] In the embodiment, the control module further includes a command filter for obtaining a filtered signal and a derivative of the filtered signal.

[0117] In some preferred embodiments, the command filter is implemented by a second-order Butterworth low-pass filter, and the transfer function of the filter is determined by the system control bandwidth. When the control command is input into the command filter, the cut-off frequency of the filter is adaptively adjusted according to the current attitude angle error, specifically by dynamically configuring the filter parameters after online identification of the system response frequency. During the filtering process, the high-frequency noise components of the control command are filtered out to generate a smooth filtered signal; at the same time, the filtered signal is subjected to first-order derivative calculation through a numerical differentiation algorithm to obtain the signal change rate. The filtered signal and the derivative signal are transmitted through parallel transmission channels to the extreme learning machine module for constructing a compensation model of the execution time delay.

[0118] In the embodiment, the control module further comprises an extreme learning machine, which is configured to obtain an approximation function of the filtered signal and the derivative of the filtered signal under extreme learning, and model the execution time delay;

[0119] The extreme learning machine is further configured to obtain an adaptive law to update the network weights and obtain a compensation signal to offset the execution time delay;

[0120] The extreme learning machine is further configured to combine the filtered signal, the derivative and the compensation signal to obtain a control signal, which is used to drive the rotary steering system.

[0121] It should be noted that the extreme learning machine first constructs a time delay dynamic model through the filtered signal and its derivative of the input layer, maps the input signal to a high-dimensional space by using a Gaussian kernel function, and fits the nonlinear characteristics of the time delay through the hidden layer nodes; the adaptive law adjusts the network weights in real time according to the error between the control signal and the expected response, so that the compensation signal can offset the phase lag caused by the time delay; finally, the compensation signal is superimposed with the filtered signal and the derivative according to a preset proportion to generate a control signal that eliminates the influence of the time delay. For example, when the actuator has a time delay of 0.5 seconds, the extreme learning machine predicts the system state change in the future 0.5 seconds through the approximation function, and generates a corresponding compensation signal, so that the control signal acts on the actuator 0.5 seconds in advance, thereby eliminating the tracking error caused by the time delay.

[0122] Embodiment 3:

[0123] As a preferred embodiment, the preferred implementation process of the embodiment scheme is as follows:

[0124] In the stage of obtaining the original attitude data of the rotary steering system, an inertial measurement unit composed of a high-precision three-axis accelerometer and a three-axis gyroscope is used to collect the original attitude data at a sampling frequency of 200 Hz. After the original data is subjected to 16-bit analog-to-digital conversion, it is transmitted to the control processor through the CAN bus.

[0125] The Kalman filter algorithm adopts a 9-dimensional state vector, including three-axis angles, angular velocities and angular accelerations. In the prediction stage, the fourth-order Runge-Kutta method is used to solve the system state equation, and in the update stage, an adaptively adjusted measurement noise covariance matrix is adopted, and the dynamic adjustment of the filter parameters is realized by analyzing the statistical characteristics of the innovation sequence, and the process noise covariance matrix is updated every 100 ms.

[0126] The rotation speed compensation adopts a quaternion-based attitude solution algorithm, which compensates the gyroscope measurement value in combination with the drilling tool rotation speed information, and the attitude angle solution frequency is 10 Hz, and the well inclination angle and tool face angle are output.

[0127] The radial basis function neural network adopts Gaussian radial basis function, the number of hidden layer nodes is 50, the network input is the attitude angle and its rate of change, and the output is the fault compensation amount, and the network weight is updated online through the recursive least square method, and the learning rate is 0.01.

[0128] The control instruction smoothing processing adopts a second-order Butterworth low-pass filter, and the cut-off frequency is 5 Hz, and the filtered signal is calculated by the central difference method to calculate the first and second derivatives.

[0129] The extreme learning machine adopts a sigmoid activation function, the number of hidden layer nodes is 100, the input includes the filtered control instruction and its derivative, and the output is the time delay compensation amount, and the network parameters are updated through the online sequence extreme value learning algorithm, and the initial learning rate is 0.1.

[0130] The final control signal is generated by superimposing the original control instruction, the fault compensation amount and the time delay compensation amount, and is output to the servo driver of the actuator at a frequency of 50 Hz.

[0131] Embodiment 4:

[0132] In order to make the technical scheme of the application more clear, the acquisition process of the original attitude data and the attitude estimation value is specifically described; specifically:

[0133] The acquisition module acquires the analog output signals of the seven-axis sensor (three-axis accelerometer, three-axis fluxgate, and in some embodiments, also including angular rate gyroscope), which are transmitted to the main control chip of the control module through the communication interface (such as SPI) after being digitized.

[0134] In the main control chip, instead of using a fixed noise covariance matrix, the system adaptively adjusts the values of the process noise covariance matrix and the measurement noise covariance matrix according to the real-time collected drilling tool data, so as to more accurately reflect the noise level of the current environment. Under severe vibration, the system will increase and value, so that the filtering process relies more on the model prediction, thus avoiding being disturbed by abnormal noise, then the system updates the optimal pose estimation and the system dynamics model to predict the state at the current time.

[0135] The specific expression satisfies:

[0136]

[0137] where, is the predicted state at the current time, i.e., the optimal pose estimation; is the state transition matrix of the system from time k 1 to k; is the pose estimation value at the last time step k 1; is the input transition matrix of the system; is the control input vector of the system at time k; is the process noise of the system.

[0138] It can be understood that the above expression is the state prediction part in the time update equation of Kalman filtering, which predicts the state at the current time by multiplying the optimal state at the last time step by the state transition matrix and adding the influence of external input , the purpose of which is to give a priori estimation based on the system dynamics model before new measurement data arrives.

[0139] Subsequently, after predicting the state, the system updates the prediction covariance matrix , which incorporates the adaptive adjustment , the expression of which satisfies:

[0140]

[0141] where, is the state transition matrix of , and the transpose matrix is obtained by interchanging the rows and columns of the original matrix;

[0142] is the prior covariance matrix at time step k;

[0143] is the posterior covariance matrix at time step k 1, which represents the uncertainty of the optimal estimation of the system state at the last time, and this value is obtained by fusing the predicted value and the actual measurement value in the last filtering cycle.

[0144] ​​ is the process noise covariance matrix, representing the uncertainty of the system model itself.

[0145] For measurement update and output: the system acquires sensor measurements and computes the Kalman gain, then uses the Kalman gain to correct the predicted state, and outputs the attitude estimate, which satisfies the expression:

[0146] ;

[0147] where, is the Kalman gain;

[0148] is the measurement matrix, describing how the system state vector maps to the measurement vector, which can be understood as converting internal states (such as inclination angle, angular velocity) to physical quantities that sensors can measure, and this matrix is pre-set according to the relationship between the sensor and the system model;

[0149] is the transpose of the measurement matrix.

[0150] For the attitude estimate, the expression satisfies:

[0151] ;

[0152] is the posterior state estimate, which is the final output of the Kalman filter, representing the optimal estimate of the system state at time step k, combining both prediction and measurement information, and thus more accurate than any single information source;

[0153] is the prior state estimate, which is the output of the previous time update, representing the system state based on model prediction;

[0154] is the measurement vector;

[0155] is the measurement residual, measuring the difference between the actual measurement and the measurement based on the predicted model, which is the basis for the filter to correct.

[0156] For the attitude angle, the following steps are taken:

[0157] Obtain the attitude estimate output from the previous step ;

[0158] Update the optimal covariance matrix , which satisfies the expression:

[0159] ; where, denotes the unit matrix, in the above expression, the unit matrix The purpose is to perform matrix subtraction operation, ensure the mathematical integrity of the formula, by subtracting A correction factor can be obtained, which will act on the prior covariance matrix So as to obtain a more accurate posterior covariance matrix The correction process is essentially to reduce uncertainty to obtain more accurate information through new measurement data.

[0160] The rotation speed data measured by the angular rate gyroscope is used to compensate the errors caused by centrifugal force and tangential force generated by high-speed rotation of the drilling tool according to the compensation formula;

[0161] The compensated signal is calculated into the inclination angle θ and the tool face angle The expression satisfies:

[0162] ;

[0163] For the compensation formula, the expression satisfies:

[0164] ;

[0165] The expression also satisfies:

[0166] ;

[0167] It can be understood that, The original output signals of the three-axis accelerometer respectively, containing additional acceleration generated by rotation;

[0168] The signal value corresponding to the gravity acceleration measured by the accelerometer in the static state is used as a reference to convert the acceleration value into actual physical units;

[0169] R is the distance from the accelerometer to the center of the drilling tool rotation, the installation position of the sensor determines the size of the centrifugal force and tangential force it feels;

[0170] ω is the angular speed of the drilling tool measured by the angular rate gyroscope;

[0171] The angular acceleration of the drilling tool, i.e. the rate of change of angular velocity, when rotating at a constant speed, this value is zero;

[0172] The compensated signals are respectively.

[0173] ​In this embodiment, for the specific compensation process, during the drilling process, the angular rate gyroscope continuously measures the angular velocity ω and angular acceleration of the drilling tool in real time ; at the same time, the triaxial accelerometer measures and outputs the original signal containing the rotation error ; the system substitutes the real-time obtained angular velocity, angular acceleration, and the preset sensor installation parameters (such as distance R) into the above compensation formula. For the X-axis (tangential) acceleration, the compensation formula subtracts the error term proportional to the angular acceleration; for the Y-axis (centrifugal) acceleration, the compensation formula subtracts the error term proportional to the square of the angular velocity; after obtaining the compensated signal, the system performs the final attitude calculation with the filtered signal to calculate the accurate inclination angle θ and tool face angle .

[0174] Embodiment 5:

[0175] For the acquisition process of the control instruction:

[0176] It can be understood that in the actual drilling process, not all system states can be directly measured, therefore, by designing a full-dimensional state observer with an embedded control module, the observer uses the measurable output data in combination with the dynamics model of the system to perform real-time estimation of all system states, and outputs the state estimation value , which ensures that even in the case of partial state loss, the controller can obtain complete state information.

[0177] Inside the state observer, by embedding a radial basis function neural network, the main function of which is to online approximate and reconstruct the unknown fault signal, the radial basis function neural network takes the system state estimation value output by the state observer as input, and through its powerful nonlinear fitting capability, outputs the real-time reconstruction value of the fault dynamics , the expression satisfies:

[0178] ;

[0179] wherein, is the estimation value of the fault signal at time k;

[0180] is the online learning weight vector of the radial basis function neural network, which is updated in real time through an adaptive law;

[0181] is the Gaussian activation function vector of the radial basis function neural network;

[0182] is the system state estimation value output by the state observer.

[0183] After state observer and radial basis function neural network reconstruction, an estimated value containing fault dynamic information is obtained .

[0184] It can be understood that in some embodiments, the control module further comprises a fault-tolerant controller, which is used to generate a control instruction capable of actively compensating for the influence of the fault after obtaining the estimated value of the fault signal, so as to ensure the continuous and stable tracking of the wellbore trajectory. Specifically, the state estimation value and the fault signal estimation are input into the fault-tolerant controller.

[0185] The core task of the fault-tolerant controller is to adjust the control instruction in real time according to the fault estimation to offset the influence of the fault. The fault-tolerant control law of the fault-tolerant controller is composed of two parts: a nominal control law and a fault compensation control law . Among them, the design of the fault compensation control law directly depends on the reconstructed fault signal in the foregoing step , so as to ensure that the system can be compensated immediately when the fault occurs. The process satisfies the expression:

[0186] ;

[0187] Among them, represents the final control instruction at time k;

[0188] represents the nominal control law for trajectory tracking in the absence of faults;

[0189] represents the fault compensation control law, the size and direction of which depend on the fault signal estimated by the radial basis function neural network.

[0190] Embodiment 6:

[0191] In some other embodiments, an integral sliding mode based fault-tolerant control law can also be selected. Specifically, an integral sliding mode surface is introduced to weight and sum the observation error of the system and the cumulative amount (i.e., integral) of the observation error with respect to the dimensionless distance ξ. It can be understood that the observation error input into the fault-tolerant controller in this embodiment and the observation error are essentially the state estimation value and the fault signal estimation , and only differ in the symbolic meaning expression. The expression satisfies:

[0192] ;

[0193] It can be understood that the above expression defines the integral sliding surface The current observation error of the system and the cumulative of the history observation error are combined and weighted by gain matrixes c1 and c2, the purpose of which is to make the system converge to a preset sliding surface in the state space, and when the system state reaches this surface, it can be kept running smoothly on the surface even if there is disturbance, so as to achieve robust control of the fault.

[0194] In the formula, is the integral sliding surface, which is a vector representing the distance between the system state and the sliding surface;

[0195] Both are positive sliding gain matrixes, which are pre-designed parameters for adjusting the characteristics and convergence speed of the sliding surface, and are constants set by the controller according to system performance indicators and stability requirements;

[0196] is the observation error of the system state, which can be expressed by the aforementioned fault signal estimate, and in other embodiments, it can also be the difference between the output of the state observer and the actual system state;

[0197] is the integral of the observation error, which represents the cumulative of the observation error in time (or dimensionless distance), reflecting the average deviation of the system in the past period of time. τ is the integral variable, which is used to traverse the distance from 0 to ξ, and is derived from the real-time integration of the observation error during operation.

[0198] It can also be understood that after the integral sliding surface is defined, this step will generate a fault-tolerant control law, the purpose of which is to generate a compensation force to force the system state to converge to the sliding surface and keep it on the surface.

[0199] The designed fault-tolerant control law is to force the system state to converge to the sliding surface and keep it on the surface, and this control law is composed of multiple parts, each of which serves a specific control goal.

[0200] The expression of this process satisfies:

[0201] ;

[0202] In the formula, is the fault-tolerant control law, which is derived from the compensation force calculated by the controller according to the system state, the observation error and the fault estimate;

[0203] G is a control gain matrix, which is a pre-set constant matrix, used to map the output of the control law to the input of the system, and is derived from the constant set by the controller according to the system requirements.

[0204] B is a system input matrix, which describes how the control input affects the system state, and is derived from the dynamics model of the rotary steerable system.

[0205] is a control parameter of the approaching law, which is a positive constant, used to adjust the convergence speed and chattering size, and is derived from the constant set by the controller according to the system performance requirements.

[0206] is a hyperbolic tangent function, which is a nonlinear function, used to smooth the approaching process and reduce chattering.

[0207] are system matrices, which describe the evolution relationship of the system state under different time delays, and are derived from the dynamics model of the rotary steerable system.

[0208] is an observation error, including the error at the current time and the time delay time.

[0209] F is a fault input matrix, which describes how the fault affects the system state.

[0210] is a fault signal reconstructed by the radial basis function neural network, derived from the output of the aforementioned radial basis function neural network.

[0211] Finally, the system superimposes the fault-tolerant control law with the normal nominal control law to form the final control command.

[0212] Example 7:

[0213] For the command filter, in the control method of the rotary steerable system, the control command output from the radial basis function neural network is usually used to drive the actuator, however, in actual application, since the output of the neural network may have high-frequency chattering, direct use will wear the actuator and may cause system instability, therefore, the control command needs to be smoothed.

[0214] In the traditional high-order nonlinear system backstepping control design, it is necessary to derive the virtual control signal step by step, which may lead to a multi-modal data processing problem, that is, as the order of the system increases, the expression of the control law will become extremely complex and lengthy, making it difficult to implement the controller, in this embodiment, the command filter approximates the virtual control signal and its derivative through a dynamic system, thereby avoiding the cumbersome analytical derivation process, greatly simplifying the controller design.

[0215] its expression satisfies:

[0216] ;

[0217] In the formula, u represents the control instruction from the radial basis function neural network, that is, the input signal of the filter;

[0218] u c1 represents the state variable of the filter, representing the smoothed control signal, that is, the filtered signal;

[0219] u c2 represents another state variable of the filter, representing the derivative of the filtered signal;

[0220] ω n represents the natural frequency, used to adjust the bandwidth of the filter, which is usually designed according to the dynamic response requirements of the system. The larger the natural frequency, the faster the filter response, but the smoother the effect may be; otherwise, the slower the response, the better the smoothing effect.

[0221] ζ represents the damping ratio, used to adjust the response characteristics of the filter to prevent overshoot and oscillation.

[0222] Example 8:

[0223] For extreme learning machines, actuator response, signal transmission, and computing processes in the rotary steering system may introduce execution time delay, which is a nonlinear characteristic of system dynamics. It will make the output signal of the controller lag behind the actual demand of the system, thereby causing the control precision to decrease, and even causing system oscillation and instability in severe cases.

[0224] It can be understood that the extreme learning machine is a kind of feedforward neural network, and its significant advantage lies in that the input layer weight and hidden layer bias are randomly generated, and no iterative training is required. Only one step calculation can obtain the output weight, which makes the training speed extremely fast, and is very suitable for online control application.

[0225] Suppose there is an unknown function caused by execution time delay, the extreme learning machine can be used to approximate the unknown function, and the expression satisfies:

[0226] ;

[0227] In the formula, is the unknown function to be approximated, where represents the system state at t-τ, which embodies the time delay characteristic;

[0228] represents the estimated value of the extreme learning machine to ;

[0229] L represents the number of hidden layer nodes;

[0230] represents the hidden layer activation function, such as the Sigmoid function, etc.

[0231] represents the input layer to the hidden layer weight vector, randomly generated and fixed;

[0232] represents the hidden layer neuron bias, randomly generated and fixed;

[0233] represents the hidden layer to the output layer weight, which needs to be determined by learning;

[0234] represents the hidden layer output matrix;

[0235] represents the estimated value of the extreme learning machine output weight, which is a vector that needs to be adjusted online.

[0236] In some embodiments, in order to ensure that the extreme learning machine can effectively compensate for the time delay and guarantee the stability of the entire closed-loop system, an adaptive law is needed to update the weight estimate in real time, and the derivation of this adaptive law preferably relies on the Lyapunov-Krasovskii functional.

[0237] The final control signal will be a superposition of multiple terms, including the control command after smoothing processing by the command filter and the time delay compensation term generated by the extreme learning machine. This control signal is transmitted to the actuator to drive the rotary steering system, so as to realize accurate tracking of the desired trajectory and at the same time guarantee the stability and robustness of the system.

[0238] It should be noted that in this paper, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or system. Without more limitations, the element defined by the statement "includes a" does not exclude the presence of another identical element in the process, method, article or system that includes the element.

[0239] The above-mentioned embodiment number of the present application is only for description, not representing the pros and cons of the embodiments.

[0240] Those skilled in the art can clearly understand the above-mentioned example method can be realized by means of software and the necessary general hardware platform, of course, also can be through hardware, but many cases the former is the better implementation. Based on such understanding, the technical solutions of the present application essentially or say the part of the contribution to the prior art can be embodied in the form of software products, the computer software product is stored in a storage medium (such as read-only memory / random access memory, disk, optical disc), including a number of instructions to make a multimedia terminal device (may be a mobile phone, computer, television receiver, or network equipment, etc.) executes the method described in various embodiments of the present application.

[0241] The above is only the preferred embodiment of the present application, not therefore limit the patent scope of the present application, any equivalent structure or equivalent process transformation using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A control method of a rotary steering system, characterized by, The control method comprises the following steps: Obtain the original attitude data of the rotary steering system, adjust the filtering parameters through the Kalman filtering algorithm, and output the attitude estimation value; According to the attitude estimation value, output the attitude angle of the rotary steering system through speed compensation; Output the control instruction of the attitude angle after fault compensation through the radial basis function neural network, including: obtaining the fault dynamic function, mapping the attitude estimation value to a high-dimensional space through a nonlinear activation function, and multiplying it by a weight vector to obtain the estimation value of the fault signal; Obtain the observation error, add the observation error and the integral of the observation error to obtain an integral sliding surface, the observation error is calculated by the difference between the current attitude angle and the reference trajectory, and the integral of the observation error is accumulated by a discrete time integrator; Obtain the fault-tolerant control law to obtain the compensation force, and superimpose the control instruction after the attitude estimation value of the rotary steering system converges to the integral sliding surface; Obtain the nominal control law, the nominal control law is based on the inverse model of the dynamic model of the rotary steering system, and the output of the nominal control law contains the basic components of the pushing force and the bias force; and generate the fault-tolerant control law after detecting and estimating the fault signal through the radial basis function neural network, the fault-tolerant control law is used to generate a compensation force to offset the influence of the fault signal on the rotary steering system; Superimpose the nominal control law and the fault-tolerant control law to output the composite control instruction, wherein the superimposition process of the nominal control law and the fault-tolerant control law is completed synchronously in the control period, and the output components of the nominal control law and the fault-tolerant control law are linearly superimposed in the force or torque dimension to form a composite control instruction containing a basic control quantity and a fault compensation quantity; Smooth the composite control instruction, and obtain the filtered signal and the derivative of the filtered signal of the composite control instruction; According to the filtered signal and the derivative of the filtered signal, compensate for the execution time lag through the extreme learning machine, and output the control signal of the rotary steering system.

2. The control method of a rotary steering system according to claim 1, wherein, The method comprises the following steps: Obtain the analog output signal of the acquisition module, and transfer it to the control module through the communication interface after digitization; Obtain the dynamic model of the rotary steering system, and use the dynamic model to transfer the optimal covariance at the last time to the current, add the process noise covariance to output the updated mean square difference matrix; According to the mean square difference matrix, modify the predicted state through the Kalman gain to output the attitude estimation value.

3. The method of claim 1, wherein, According to the attitude estimation value, output the attitude angle of the rotary steering system through speed compensation, which comprises the following steps: Obtain the attitude estimation value, update the optimal covariance matrix, and perform attitude calculation to output the attitude angle of the rotary steering system, wherein the attitude angle includes the inclination angle and the tool face angle.

4. The method of claim 1, wherein, According to the filtered signal and the derivative of the filtered signal, compensate for the execution time lag through the extreme learning machine, and output the control signal of the rotary steering system, which comprises the following steps: Obtain the approximation function of the filtered signal and the derivative of the filtered signal under the limit learning, and model the execution time lag; Obtain the adaptive law of the extreme learning machine to update the network weight and obtain the compensation signal to offset the execution time lag; The control signal is obtained by superimposing the filtered signal, the derivative of the filtered signal, and the compensation signal in a preset ratio, and the control signal is used to drive the rotary steering system.

5. A control system characterized by, The control system is used to execute the control method of the rotary steering system of claim 4, and the control system comprises: The acquisition module is configured to obtain original attitude data of the rotary steering system, adjust filter parameters by using a Kalman filtering algorithm, and output an attitude estimation value. The compensation module is configured to output an attitude angle of the rotary steering system by using a rotation speed compensation according to the attitude estimation value. The control module is configured to output a control instruction for the attitude angle after fault compensation by using a radial basis function neural network, and the control instruction comprises: obtaining a fault dynamic function, mapping the attitude estimation value to a high-dimensional space by using a nonlinear activation function, and multiplying the attitude estimation value by a weight vector to obtain an estimated value of a fault signal. An observation error is obtained, and the observation error and an integral of the observation error are added to obtain an integral sliding surface, the observation error is calculated by using a difference between a current attitude angle and a reference trajectory, and the integral of the observation error is sequentially accumulated by using a discrete-time integrator. The fault-tolerant control law is obtained to generate the control instruction after the attitude estimation value of the rotary steering system is converged to the integral sliding surface by using a compensation force. The control module is further configured to obtain a nominal control law, the nominal control law is an inverse model based on a dynamic model of the rotary steering system, an output of the nominal control law comprises a basic component of a pushing force and a biasing force, and the fault-tolerant control law is generated after the fault signal is detected and estimated by using the radial basis function neural network, and the fault-tolerant control law is used to generate a compensation force to offset an influence of the fault signal on the rotary steering system. The nominal control law and the fault-tolerant control law are superimposed to output a composite control instruction, wherein the superimposition process of the nominal control law and the fault-tolerant control law is completed synchronously in a control period, output components of the nominal control law and the fault-tolerant control law are linearly superimposed in a force or torque dimension, and the composite control instruction comprising a basic control amount and a fault compensation amount is formed. The control module further comprises an observer, the observer is embedded with a radial basis function neural network, and the radial basis function neural network is used to detect and estimate a fault signal.

6. A control system as claimed in claim 5, characterised in that, The control module further comprises a command filter, the command filter is used to obtain a filtered signal and a derivative of the filtered signal.

7. A control system as claimed in claim 5, characterised in that, The control module further comprises a limit learning machine, the limit learning machine is used to obtain an approximation function of the filtered signal and the derivative of the filtered signal under limit learning, and an execution time lag is modeled.

8. A control system as claimed in claim 5 or 7, characterised in that, The limit learning machine is further configured to obtain an adaptive law to update network weights and obtain a compensation signal to offset the execution time lag. The limit learning machine is further configured to obtain a control signal by combining the filtered signal, the derivative of the filtered signal, and the compensation signal, and the control signal is used to drive the rotary steering system. ​

Images