A complex stratum oil drilling electric oil power optimization control method and system

Abstract

The application discloses a complex stratum oil drilling electric oil power optimization control method and system, and belongs to the field of drilling safety prevention and control. The method comprises the following steps: collecting multi-source heterogeneous data in the process of oil drilling, performing fusion processing on the collected multi-source heterogeneous data in an edge gateway controller to obtain fused multi-source heterogeneous data flow; processing the fused multi-source heterogeneous data flow through a pre-trained long short-term memory network to obtain a stratum load torque change trend sequence that a drill bit will encounter in a preset future time window; obtaining an optimal power pre-adjustment instruction sequence according to the stratum load torque change trend sequence through a prediction control algorithm; and receiving the instruction sequence by the edge gateway controller and sending the instruction sequence to a drilling fluid pump control system and a heave compensation system of a drilling platform. Through real-time fusion and analysis of multi-source heterogeneous data, the safety and efficiency of drilling operation are improved.

Description

Technical Field

[0001] This invention relates to the field of drilling safety control, and in particular to a method and system for optimizing the power control of electric oil drilling in complex formations. Background Technology

[0002] With the widespread adoption of electric power technology on oil drilling platforms, the use of industrial grid-driven high-power variable frequency motors to replace traditional diesel generator sets has significantly reduced energy costs and environmental pollution in drilling operations. However, existing electric-drive drilling power control systems still face serious technical bottlenecks when dealing with drilling in complex formations. Traditional electric-drive drilling power control systems typically employ a passive adjustment strategy based on real-time operating parameter feedback. This strategy monitors the rotational speed in real time through an edge gateway controller. When the rotational speed decreases due to formation hardening, it attempts to restore the set rotational speed by increasing the inverter's output power through PID algorithms or simple logic judgments. This control method is essentially a reactive measure and a lagging control. Furthermore, in actual drilling operations, the downhole formation structure is often complex and variable, frequently encountering hard rock interlayers, soft mudstone formations, fractured lost circulation formations, or abrupt interfaces between different lithologies. When the drill bit encounters these abrupt formation changes suddenly at high speed, the downhole load can change drastically within milliseconds.

[0003] Furthermore, in traditional drilling platform control architectures, the power control system driving the rotary table or top drive, the pump control system controlling drilling fluid circulation, and the heave compensation system for offshore platforms or specific operating conditions are typically independent control units. These systems lack high-speed, effective information exchange and collaborative decision-making mechanisms. Consequently, a globally optimal control strategy cannot be formed, often resulting in inconsistencies when dealing with complex operating conditions, severely restricting the overall efficiency and intelligence level of drilling operations. Summary of the Invention

[0004] One of the objectives of this invention is to provide an optimized control method for electric power generation in oil drilling in complex formations, in order to solve the problem of drastic fluctuations in rotational speed and torque caused by drastic load changes due to complex formations in the prior art.

[0005] This invention is achieved through the following technical solution: a method for optimizing the power control of oil drilling in complex formations using electric power, comprising the following steps: S100, collecting multi-source heterogeneous data during the oil drilling process, and fusing the collected multi-source heterogeneous data in the edge gateway controller to obtain a fused multi-source heterogeneous data stream; S200, processing the fused multi-source heterogeneous data stream through a pre-trained long short-term memory network to obtain a sequence of formation load torque change trends that the drill bit will encounter within a preset future time window; S300, based on the formation load torque change trend sequence, the predictive control algorithm, with the optimization objective of minimizing grid energy consumption fluctuations, drill bit speed deviations, and torque impacts, obtains the optimal power pre-adjustment command sequence under the condition of satisfying motor safety operation constraints; S400, the edge gateway controller receives the load torque change trend sequence and the power pre-adjustment command sequence, uses the load torque change trend sequence as a collaborative control benchmark, and synchronously sends the power pre-adjustment command sequence to the drilling fluid pump control system and heave compensation system of the drilling platform. The drilling fluid pump control system and heave compensation system execute the command sequence to achieve linkage optimization between the power system and related subsystems.

[0006] Furthermore, the multi-source heterogeneous data includes: prospective logging data containing formation lithology parameters and formation drillability parameters; real-time drilling condition data containing drilling pressure, torque, and rotational speed; and power system electrical parameter data containing the output current, voltage, and power of the main drive frequency converter.

[0007] Furthermore, the fusion process includes: unifying the multi-source heterogeneous data onto a common time base to achieve timestamp alignment; removing outliers and noise interference from the data through outlier detection to clean up outliers; and converting data with different physical units and dimensions into data with the same scale through the Z-Score algorithm to eliminate the influence of dimensions.

[0008] Furthermore, the edge gateway controller can also include a preset space-time mapping model. This model can calculate in real time when the formation features detected by the LWD will reach the drill bit position based on the current drilling rate, thereby achieving precise alignment of the LWD look-ahead data in the time dimension. By accurately and dynamically mapping the discrete formation information measured by LWD in the spatial domain and downhole depth to the time domain required by the controller, at future time points, the edge gateway controller can transform the spatial information of hard rock X meters ahead of the drill bit into the temporal information of a predicted sharp increase in load in Y seconds, which can be used to perform forward-looking control.

[0009] Furthermore, this space-time mapping model can be represented by the following equation:

[0010] ,

[0011] in, The current time indicates the real-time system time at the time the calculation was performed; The depth measured by LWD is the absolute depth at which the LWD sensor measures a specific formation parameter. This indicates the predicted arrival time, which is a predicted future time when the drill bit reaches the depth where the LWD measurement point is located. This is the current time, which is also the real-time system time at the time the calculation was performed; For drill bit depth, This indicates the current drill bit depth, that is, the real-time absolute depth at which the drill bit is located at the current time; For the prediction rate of drilling function, this is a key embedded function that takes the downhole depth (between the current drill bit depth and the LWD measurement depth) as input and outputs the predicted mechanical rate of drilling at that depth. It is a depth infinitesimal element, that is, an infinitesimal depth increment on the integration path.

[0012] Furthermore, the pre-trained Long Short-Term Memory (LSTM) network, with the pre-trained dataset consisting of historical data from the drilled sections of this well and / or historical databases of neighboring drilled wells, also includes online updates. This involves calculating the prediction error by comparing the actual measured torque value with the model's predicted value, and using this error as a correction signal to fine-tune the network weights of the LSM network online through a backpropagation algorithm with a small learning rate.

[0013] Furthermore, the predictive control algorithm includes an optimization objective function for balancing the trade-offs between performance, energy consumption, and device smoothness, as well as constraints for constraining the state of the optimization objective function.

[0014] Furthermore, the objective function is optimized as shown in the following equation:

[0015] ,

[0016] in, Let be the objective function. Let this be the state vector at the current moment. For the motor output torque, This is the matrix transpose symbol; For the entire control sequence within the future control time domain; For a single time step; To predict the time domain, This represents the discrete time step at the current control moment; This refers to the reference value or expected trajectory of the state variable; Here is the weight matrix for state tracking. To control the time domain, This is the control input vector at the current moment; This is the weight matrix for energy consumption control. To control the amount of change in the input, The weight matrix is ​​for smooth control.

[0017] Furthermore, the constraints include:

[0018] System state-space model: ;

[0019] State variable constraints: ;

[0020] Control input constraints: ;

[0021] Input rate of change constraint: ;

[0022] in, Let be the discrete-time state-space matrix of the drilling power system. The discrete-time state-space matrix is ​​the control input of the drilling power system. Let be the discrete-time state-space matrix of the disturbance vector. This is the measurable perturbation vector at the current moment; The minimum value of the state. The maximum value of the state; To control the minimum input value, To control the maximum value of the input; To control the minimum value of the input variation, To control the maximum value of the input variation.

[0023] Furthermore, the coordinated control commands of the drilling fluid pump control system are generated differentially based on the formation lithology corresponding to the load torque change trend sequence: when it is predicted that the system will enter a soft mudstone formation prone to mud pockets, the coordinated command is to increase the drilling fluid discharge rate in advance; when it is predicted that the system will enter a fractured formation prone to leakage, the coordinated command is to appropriately reduce the drilling fluid discharge rate. By embedding a formation lithology-drilling fluid strategy knowledge base in the drilling fluid pump control system, when the load torque change trend sequence indicates that the system will enter a soft mudstone formation prone to mud pockets in the future, the drilling fluid pump control system automatically queries the knowledge base and executes the coordinated command.

[0024] Furthermore, the coordinated control commands of the heave compensation system are generated differentially based on the formation structure corresponding to the load torque change trend sequence. Continuous short-time Fourier transform or fast Fourier transform is performed on real-time torque data. When the analysis results show that the signal energy continuously exceeds a preset threshold within the typical characteristic frequency band of stick-slip vibration (0.1Hz~2Hz), it is determined to be a precursor to stick-slip vibration. Once a precursor to stick-slip vibration is identified, the heave compensation system immediately triggers two parallel suppression actions: a high-frequency, low-amplitude sinusoidal disturbance signal is superimposed on the original speed setpoint and sent to the frequency converter; simultaneously, an instantaneous boost command is sent to the drilling fluid pump control system, increasing its displacement by 5%-10% within 3-5 seconds.

[0025] Another aspect of the present invention provides an electric power optimization control system for oil drilling in complex formations, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the electric power optimization control method for oil drilling in complex formations as described above.

[0026] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0027] 1. This invention achieves real-time monitoring and dynamic analysis of downhole conditions by collecting and fusing multi-source heterogeneous data during the drilling process. It overcomes the limitations of traditional well formation lithology information acquisition methods that rely on fixed data, and improves the accuracy and timeliness of formation condition identification.

[0028] 2. This invention establishes a mapping model from drill bit depth to geological parameters based on LWD prospective data, realizing accurate prediction and dynamic monitoring of complex formation conditions, and effectively solving the problem that existing technologies lack dynamic adaptability to complex formation conditions in well control risk prediction.

[0029] 3. This invention improves the prediction accuracy and reliability of well load torque changes by using a pre-set long short-term memory network combined with an online update mechanism for actual measurement data. It overcomes the problem of poor adaptability of existing prediction models under different ore conditions. Furthermore, based on the load torque change trend sequence as a collaborative control benchmark, and combined with the power pre-adjustment command sequence generated by the predictive control algorithm, it realizes intelligent collaborative control of the drilling fluid pump control system and the heave compensation system, thereby improving the flexibility and response speed of the well control system in response to complex formation conditions.

[0030] 4. This invention establishes a complete downhole condition monitoring and prediction system by real-time fusion and analysis of multi-source heterogeneous data, providing a scientific basis for safety decisions in the oil drilling process, effectively reducing the risk of well control accidents such as blowouts, and improving the safety and efficiency of drilling operations. Attached Figure Description

[0031] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0032] Figure 1 This is a flowchart of the method provided in Embodiment 1 of the present invention.

[0033] Figure 2 This is a timing diagram provided for Embodiment 1 of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0035] Example 1

[0036] This embodiment discloses a method for optimizing the power control of electric oil drilling in complex formations. Figure 1 The overall flowchart of this embodiment is shown. Figure 2 The overall timing diagram of this embodiment is shown, from Figure 1 As can be seen from this embodiment, it includes the following steps:

[0037] Step 1: First, relevant data during the oil drilling process is collected. The edge gateway controller performs real-time fusion of the collected data to obtain a fused data stream. The edge gateway controller then sends this data stream to the backend computer for data processing.

[0038] Specifically, the relevant data includes: forward-looking logging-while-drilling (LWD) data containing formation lithology and drillability parameters; real-time drilling condition data containing drilling pressure, torque, and rotational speed; and power system electrical parameter data containing the output current, voltage, and power of the main drive inverter.

[0039] Real-time fusion of these multi-source heterogeneous data includes: First, unifying multi-source heterogeneous data from different sources and frequencies onto a common time base using interpolation or alignment algorithms to achieve timestamp alignment; then, removing outliers and noise interference from the data through outlier detection, thus cleaning the data, improving data quality, and ensuring the accuracy and reliability of subsequent analysis results; finally, converting data with different physical units and dimensions into data with the same scale using the Z-Score algorithm, thereby eliminating the influence of dimensions and completing the standardization process of the data.

[0040] In this embodiment, a mapping model from drill bit depth (spatial) to geological parameters (inputs for time series prediction) can also be established in the edge gateway controller based on LWD look-ahead data. This model can be a lookup table based on neighboring well data or a dynamic mapping model. This helps subsequent prediction models not only rely on what just happened, but also utilize prior knowledge of what should happen at this depth.

[0041] For LWD look-ahead data, since LWD measures formation information at a certain distance ahead of the drill bit, accurate prediction (look-ahead) requires knowing when these formation features will reach the drill bit. Therefore, a spatial-temporal mapping model between the drill bit position and the LWD measurement depth can be pre-established in the edge gateway controller. Based on the current drilling rate, it can be calculated in real time when the formation features detected by LWD will reach the drill bit position, thus achieving accurate alignment of look-ahead data in the time dimension.

[0042] Specifically, in this implementation, discrete formation information measured by LWD in the spatial domain (downhole depth) is accurately and dynamically mapped to the time domain (future time points) required by the controller. This allows the edge gateway controller to transform the spatial information "there is hard rock X meters ahead of the drill bit" into temporal information "the load is expected to increase sharply in Y seconds," which can be used to execute forward-looking control.

[0043] For example, in this embodiment, the space-time mapping model can be represented by the following equation:

[0044] ,

[0045] in, The current time indicates the real-time system time at the time the calculation was performed; The depth measured by LWD is the absolute depth at which the LWD sensor measures a specific formation parameter. This indicates the predicted arrival time, which is a predicted future time when the drill bit reaches the depth where the LWD measurement point is located. This is the current time, which is also the real-time system time at the time the calculation was performed; For drill bit depth, This indicates the current drill bit depth, that is, the real-time absolute depth at which the drill bit is located at the current time; For the prediction rate of drilling function, this is a key embedded function that takes the downhole depth (between the current drill bit depth and the LWD measurement depth) as input and outputs the predicted mechanical rate of drilling at that depth. It is a depth infinitesimal element, that is, an infinitesimal depth increment on the integration path.

[0046] It should be noted that in the above formula, It is the time base for the entire prediction, the definite integral. This represents the total time required to drill the distance from the current depth to the target depth; the function... This is a dynamic prediction of future drilling speed. As drilling progresses from the current drill bit depth to the measured depth, the drilling speed changes with the formation. This function is determined by formation drillability parameters (such as rock strength and abrasiveness) measured at the depth based on LWD (Low-Drilling Witness) and combined with current drilling operation parameters (such as drilling pressure and rotation speed). The reciprocal of this function, i.e., the reciprocal of the predicted drilling speed, represents the value at depth... The drilling hysteresis rate or time required to drill per unit depth (in seconds per meter) is the drilling hysteresis rate at a given location. If the formation hardens, the predicted drilling rate will decrease, and this value will increase, meaning it will take longer to drill through that layer. The model divides the entire path from the drill bit's current position to the LWD measurement point into numerous tiny depth segments, calculates the time required to drill through each segment, and finally sums all these time values. This sum represents the total time required to drill the entire path.

[0047] Step 2: Based on the fused data stream, calculate and output the sequence of formation load torque changes that the drill bit will encounter within a future preset time window (T) using a preset load prediction model.

[0048] The load prediction model extracts formation lithology parameters, real-time drilling conditions, and power system electrical parameters from the LWD prospective data in the fused data stream, aligns and samples these data according to a preset time step, and constructs a multi-dimensional time-series feature vector.

[0049] Then, the multidimensional temporal feature vector is input into a pre-trained long short-term memory network (LSTM), and the output is the sequence of formation load torque change trends within a future preset time window (T).

[0050] Specifically, the pre-trained Long Short-Term Memory (LSTM) network in this embodiment is trained on a dataset that can be historical data from the drilled sections of the current well or historical databases from neighboring drilled wells. This data contains rich formation information and the correspondence between drilling parameters. LSTM is chosen because drilling is a typical time-series process, and the current load during drilling is related to the formation and operating conditions over a long period of time. The gating mechanism of LSTM can effectively capture such long-term dependencies in time-series data.

[0051] In addition, after actual deployment, the LSTM can be updated online by comparing the actual measured torque value with the model's predicted value, calculating the prediction error, and using this error as a correction signal to fine-tune the network weights of the LSTM model online through a backpropagation algorithm with a small learning rate, so that it can continuously self-optimize and maintain high prediction accuracy.

[0052] Step 3: Based on the output formation load torque change trend sequence, the predictive control algorithm aims to minimize grid energy consumption fluctuations, drill bit speed deviations, and torque impacts, and obtains the optimal power pre-regulation command sequence under the condition of satisfying the motor safe operation constraints.

[0053] The instruction sequence and the formation load torque change trend sequence are sent to the edge gateway controller to adjust its output power in advance, so that it can proactively adapt to the upcoming formation load changes with optimized power output.

[0054] Specifically, in this implementation, the predictive control algorithm transforms the complex multi-objective dynamic control problem into a series of mathematically well-defined quadratic programming problems that are repeatedly solved within a finite time domain, thereby generating proactive and forward-looking power pre-regulation commands. By looking ahead a finite time domain at each control moment and solving a constrained optimization problem, the optimal control input for the current moment is obtained.

[0055] For example, in this embodiment, the optimization objective function can be expressed as follows:

[0056]

[0057] ,

[0058] in, Let be the objective function. For example, the state vector at the current moment. , This refers to the motor speed. For the motor output torque, This is the matrix transpose symbol. For the entire control sequence within the future control time domain; For a single time step; For the prediction time domain, it indicates how far forward the edge gateway controller "looks," that is, how many time steps into the future it predicts; This represents the discrete time step at the current control moment; This refers to the reference value or expected trajectory of the state variable; Here is the weight matrix for state tracking. For the control time domain, it represents how many future control inputs the edge gateway controller calculates at once; generally speaking... . This is the control input vector at the current moment; it is usually a command issued by the edge gateway controller to the frequency converter, such as a torque command or a power command. This is the weight matrix for energy consumption control. To control the amount of change in the input, i.e. , The weight matrix is ​​for smooth control.

[0059] It should be noted that in the above formula, The state tracking term aims to penalize (minimize) the deviation between the predicted and expected values ​​of the system state throughout the entire prediction time domain. The larger the diagonal elements of the weight matrix Q, the more closely the edge gateway controller will track the setpoint of the corresponding state variables (such as rotational speed). This is directly related to drilling efficiency (maintaining optimal rotational speed) and wellbore quality, and is a major driver for achieving drilling process requirements. This is an energy consumption control term designed to penalize (minimize) the size of the control input itself. The larger the diagonal elements of its weight matrix R, the more likely the controller will use smaller control commands (e.g., smaller torque or power commands) to achieve the control objective, helping to avoid unnecessary energy waste and reduce system operating costs. To smooth the control term, it aims to penalize (minimize) the drastic change in control input between two adjacent control moments. The larger the diagonal elements of its weight matrix S, the smoother the sequence of control commands generated by the controller will be, directly suppressing drastic power fluctuations and ensuring a smooth power curve to cope with load shocks. At the same time, smooth commands can also reduce the mechanical shock to the frequency converter and drive system, thereby extending equipment life.

[0060] The objective function for optimization is subject to the following constraints:

[0061] System state-space model: ;

[0062] State variable constraints: ;

[0063] Control input constraints: ;

[0064] Input rate of change constraint: ;

[0065] in, Let be the discrete-time state-space matrix of the drilling power system. The discrete-time state-space matrix is ​​the control input of the drilling power system. The three state-space matrices mentioned above can be constructed based on system identification, and are the discrete-time state-space matrices of the disturbance vector. This is the measurable disturbance vector at the current moment; that is, the sequence of formation load torque change trends from step 2. The minimum value of the state. The maximum value of the state; To control the minimum input value, To control the maximum value of the input; To control the minimum value of the input variation, To control the maximum value of the input variation.

[0066] It should be noted that the predictive control algorithm in this embodiment can balance the trade-offs between performance, energy consumption, and equipment smoothness while satisfying all physical safety constraints. By receiving a sequence of future challenges from the LSTM, it ultimately calculates the optimal execution plan with the highest overall benefit over a future period. The predictive information is transformed into specific, executable, smooth, and safe equipment control commands. For example, in the case of abrupt changes at complex subsurface interfaces, the specific processing flow can be as follows: when LWD look-ahead data predicts that the drill bit is about to enter a low-strength mudstone formation from a high-strength hard rock formation, the LSTM predicts a sharp drop in load torque; this prediction sequence serves as a perturbation. The input is fed into the predictive control algorithm. In order to minimize the state tracking term in the objective function (to prevent the motor speed from far exceeding the set value), the optimal solution of the predictive control algorithm must be to smoothly reduce the control input (i.e., power) in advance before the impact arrives, thereby generating a "pre-load" ramp command. This allows the drill bit to execute a power "pre-load" ramp command in advance before reaching the formation interface, so as to smoothly reduce the output torque and prevent the drill bit from running away with its speed due to excessive unloading when entering the soft formation.

[0067] Step 4: The edge gateway controller receives the load torque change trend sequence and the power pre-adjustment command sequence. It uses the load torque change trend sequence as the collaborative control benchmark and the power pre-adjustment command sequence generated by the predictive control algorithm to synchronously send them to the drilling fluid pump control system and heave compensation system of the drilling platform. The drilling fluid pump control system and heave compensation system execute the command sequence to realize the linkage optimization of the power system and related subsystems.

[0068] Specifically, the coordinated control commands for the drilling fluid pump control system are generated differentially based on the formation lithology corresponding to the load torque change trend sequence: when it is predicted that the system is about to enter a soft mudstone formation prone to mud pockets, the coordinated command is to increase the drilling fluid discharge rate in advance; when it is predicted that the system is about to enter a fractured formation prone to leakage, the coordinated command is to appropriately reduce the drilling fluid discharge rate.

[0069] A formation lithology-drilling fluid strategy knowledge base can be built into the drilling fluid pump control system. When the LSTM predicts that it will enter a soft mudstone formation prone to mud bags, the drilling fluid pump control system will automatically consult the knowledge base and execute coordinated instructions. For example, it may gradually increase the drilling fluid flow rate and / or adjust the drilling fluid properties 5-60 seconds in advance to prevent mud bags. Conversely, if it predicts that it will enter a fractured formation prone to leakage, it will appropriately reduce the flow rate in advance to protect the oil layer.

[0070] The coordinated control commands for the heave compensation system are generated differentially based on the geological structure corresponding to the load torque change trend sequence.

[0071] For example, continuous short-time Fourier transform (STFT) or fast Fourier transform (FFT) can be performed on real-time torque data. When the analysis results show that the signal energy continuously exceeds a preset threshold (e.g., more than 3 standard deviations of normal background fluctuations) within the typical characteristic frequency band of stick-slip vibration, which is 0.1Hz to 2Hz, it is determined to be a precursor to stick-slip vibration.

[0072] Once the warning signs are detected, the heave compensation system immediately triggers two parallel suppression actions: A high-frequency (e.g., 5-10Hz), low-amplitude (e.g., 1%-2% of the set speed) sinusoidal disturbance signal is superimposed on the original speed setpoint and sent to the frequency converter. This disturbance aims to actively disrupt the stable resonance conditions upon which stick-slip vibration depends. Simultaneously, an instantaneous boost command is sent to the drilling fluid pump control system, increasing its displacement by 5%-10% within a short period (e.g., 3-5 seconds). The increased displacement improves the cleanliness and hydraulic parameters at the bottom of the well, thereby altering the frictional characteristics and helping to suppress the further development of torsional vibration.

[0073] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing the power control of electric oil drilling in complex formations, characterized in that, The power optimization control method includes: S100: Acquires multi-source heterogeneous data during the oil drilling process. In the edge gateway controller, the collected multi-source heterogeneous data is fused to obtain a fused multi-source heterogeneous data stream; S200: By processing and fusing multi-source heterogeneous data streams through a pre-trained long short-term memory network, a sequence of formation load torque change trends that the drill bit will encounter within a preset future time window is obtained. S300: The predictive control algorithm, based on the formation load torque variation trend sequence, takes minimizing grid energy consumption fluctuations, drill bit speed deviations, and torque impacts as optimization objectives, and obtains the optimal power pre-regulation command sequence under the condition of satisfying the motor safe operation constraints. The S400 edge gateway controller receives the load torque change trend sequence and the power pre-adjustment command sequence. It uses the load torque change trend sequence as the collaborative control benchmark and synchronously sends the power pre-adjustment command sequence to the drilling fluid pump control system and heave compensation system of the drilling platform. The drilling fluid pump control system and heave compensation system execute the command sequence to realize the linkage optimization of the power system and related subsystems. The coordinated control commands of the drilling fluid pump control system are generated differentially based on the formation lithology corresponding to the load torque change trend sequence. When it is predicted that the well will soon enter a soft mudstone formation prone to mud pockets, the coordinated instruction is to increase the drilling fluid discharge rate in advance. When it is predicted that the well will enter a fractured formation that is prone to leakage, the coordinated instruction is to appropriately reduce the drilling fluid discharge rate. By incorporating a formation lithology-drilling fluid strategy knowledge base into the drilling fluid pump control system, When the load torque change trend sequence indicates that the well will enter a soft mudstone formation prone to mud pockets in the future, the drilling fluid pump control system automatically queries the knowledge base and executes collaborative commands. The coordinated control commands of the heave compensation system are generated differentially based on the geological structure corresponding to the load torque change trend sequence. Continuous short-time Fourier transform or fast Fourier transform is performed using real-time torque data. When the analysis results show that the signal energy continuously exceeds the preset threshold within the typical characteristic frequency band of stick-slip vibration of 0.1Hz to 2Hz, it is determined to be a precursor of stick-slip vibration. Once the early signs of stick-slip vibration are detected, the heave compensation system immediately triggers two parallel suppression actions: a high-frequency, low-amplitude sinusoidal disturbance signal is superimposed on the original speed setpoint and sent to the frequency converter. At the same time, an instantaneous boost command is sent to the drilling fluid pump control system to increase its displacement by 5%-10% within 3-5 seconds.

2. The method for optimizing and controlling the power of electric oil drilling in complex formations according to claim 1, characterized in that, The multi-source heterogeneous data includes: Logging-while-drilling prospective data containing formation lithology parameters and formation drillability parameters; Real-time drilling data including drilling pressure, torque, and rotational speed; Electrical parameter data of the power system, including the output current, voltage, and power of the main drive frequency converter.

3. The method for optimizing and controlling the power of electric oil drilling in complex formations according to claim 1, characterized in that, The fusion process includes: The multi-source heterogeneous data is unified to a common time base to achieve timestamp alignment; Outlier detection removes outliers and noise from the data, thus cleaning up outliers. The Z-Score algorithm is used to convert data with different physical units and dimensions into data with the same scale, thus eliminating the influence of dimensions.

4. The method for optimizing and controlling the power of electric oil drilling in complex formations according to claim 1, characterized in that, The pre-trained long short-term memory network is pre-trained using historical data from the drilled sections of this well and / or historical databases of neighboring drilled wells. The Long Short-Term Memory (LSTM) network also includes online updates. By comparing the actual measured torque value with the model's predicted value, the prediction error is calculated, and this error is used as a correction signal to fine-tune the network weights of the LSM network online through a backpropagation algorithm with a small learning rate.

5. The method for optimizing and controlling the power of electric oil drilling in complex formations according to claim 1, characterized in that, The predictive control algorithm includes an optimization objective function for balancing the trade-offs between performance, energy consumption, and equipment smoothness, and... Constraints are used to constrain the state of the objective function.

6. The method for optimizing and controlling the power of electric oil drilling in complex formations according to claim 5, characterized in that, The optimization objective function is shown in the following equation: ， in, Let be the objective function. Let this be the state vector at the current moment. For the motor output torque, This is the matrix transpose symbol; For the entire control sequence within the future control time domain; For a single time step; To predict the time domain, This represents the discrete time step at the current control moment; This refers to the reference value or expected trajectory of the state variable; Here is the weight matrix for state tracking. To control the time domain, This is the control input vector at the current moment; This is the weight matrix for energy consumption control. To control the amount of change in the input, The weight matrix is ​​for smooth control.

7. The method for optimizing and controlling the power of electric oil drilling in complex formations according to claim 5, characterized in that, The constraints include: System state-space model: ; State variable constraints: ; Control input constraints: ; Input rate of change constraint: ; in, Let be the discrete-time state-space matrix of the drilling power system. The discrete-time state-space matrix is ​​the control input of the drilling power system. Let be the discrete-time state-space matrix of the disturbance vector. This is the measurable perturbation vector at the current moment; The minimum value of the state. The maximum value of the state; To control the minimum input value, To control the maximum value of the input; To control the minimum value of the input variation, To control the maximum value of the input variation.

8. A power optimization control system for electric oil drilling in complex formations, characterized in that, The power optimization control system includes: processor; The memory stores a computer program that, when executed by a processor, implements the method for optimizing the power control of electric oil drilling in complex formations as described in any one of claims 1 to 7.

Images