A coal rock gas horizontal well drilling parameter optimization method, system, medium and equipment

Abstract

The present application relates to the technical field of deep coal rock gas horizontal well drilling parameter optimization, in particular to a coal rock gas horizontal well drilling parameter optimization method, system, medium and equipment, the method comprises the following steps: obtaining the while-drilling logging parameter, and constructing a rock breaking energy utilization rate calculation model by using the while-drilling logging parameter to form a sample data set; the sample data set is preprocessed in multiple stages, and a long short-term memory neural network model is constructed; the trained long short-term memory neural network model is embedded into the logging system; based on the real-time predicted real rock breaking energy utilization rate, the optimization objective of maximizing the mechanical drilling speed and minimizing the real rock breaking energy utilization rate is taken, and the drilling pressure, rotating speed and torque parameters are dynamically adjusted; the present application reduces the technical implementation cost without changing the existing drilling and logging process and without adding downhole measurement equipment, and improves the safety and comprehensive benefit of deep coal rock gas horizontal well drilling operation.

Description

Technical Field

[0001] This invention relates to the field of drilling parameter optimization technology for deep coal and rock gas horizontal wells, and in particular to a method, system, medium, and equipment for optimizing drilling parameters for coal and rock gas horizontal wells. Background Technology

[0002] Deep coal-rock gas resources, as important unconventional natural gas resources, generally have geological characteristics such as deep burial, high geostress, complex lithology, and strong heterogeneity. During horizontal well drilling, the reasonable configuration of drilling parameters such as drilling pressure, rotation speed, and torque directly affects the mechanical drilling rate, drill bit life, and wellbore stability. Currently, drilling parameter optimization methods mainly include manual adjustment based on the experience of field engineers, mathematical modeling based on mechanical mechanisms, and intelligent prediction based on machine learning. Among them, manual adjustment relies on individual experience and has a lag in response; mathematical modeling is difficult to adapt to the dynamic changes of complex formations; and machine learning methods based on BP neural networks can use historical data for modeling, but are limited by the gradient vanishing defect inherent in the network structure and cannot effectively capture the long-term time-series dependencies between drilling parameters.

[0003] The aforementioned existing technical solutions have significant limitations in practical applications: First, traditional empirical methods and mathematical modeling methods lack the ability to mine the temporal characteristics of logging data while drilling, and cannot identify complex working conditions such as drill bit stick-slip and torque mutations in real time. This results in parameter adjustments lagging significantly behind formation changes, which can easily induce downhole accidents such as drill bit tooth breakage and wellbore instability. Second, existing optimization methods based on shallow neural networks fail to fully consider the torque transmission loss between the drill string and the drill bit, resulting in large errors in the calculation of rock-breaking energy utilization rate. They may also require additional downhole vibration sensors and well-to-surface communication equipment, significantly increasing operating costs and technical implementation difficulties. Third, existing deep learning models mostly adopt standard feedforward structures, making it difficult to establish a long-term dynamic mapping between drilling parameters and rock-breaking energy efficiency. This cannot meet the engineering requirements of real-time parameter optimization and temporal correlation analysis for deep coal and gas reservoirs. At present, there is a need for a method, system, medium, and equipment for optimizing drilling parameters in horizontal coal and gas wells. Summary of the Invention

[0004] To address the issues of insufficient time-series dependency mining and inadequate real-time identification capability for complex operating conditions in existing methods for optimizing drilling parameters in deep coal and gas horizontal wells, this invention provides a method, system, medium, and equipment for optimizing drilling parameters in coal and gas horizontal wells.

[0005] In a first aspect, the present invention provides a method for optimizing drilling parameters in horizontal coal gas wells, which adopts the following technical solution:

[0006] A method for optimizing drilling parameters in horizontal coal gas wells includes:

[0007] The drilling logging parameters are obtained, including drilling pressure, rotation speed, torque and mechanical drilling speed. The effective rock-breaking torque of the drill bit is obtained based on torque separation. The effective rock-breaking torque of the drill bit is the difference between the real-time torque during normal drilling and the idle torque.

[0008] A calculation model for rock breaking energy utilization rate was constructed using logging-while-drilling parameters and effective rock breaking torque of the drill bit. Characteristic parameters of rock breaking energy utilization rate were determined, and a sample dataset was formed.

[0009] The sample dataset undergoes multi-stage preprocessing, which includes preprocessing based on... Outlier detection, double smoothing of input and output features, and sliding window temporal reconstruction are all based on the criteria.

[0010] A long short-term memory neural network model was constructed, and the long-term temporal dependencies between drilling parameters were captured by a sliding window to establish a nonlinear dynamic mapping between drilling parameters and actual rock-breaking energy utilization.

[0011] The trained long short-term memory neural network model is embedded into the logging system. Based on the real rock-breaking energy utilization rate predicted in real time and combined with the working condition threshold of deep coal and gas reservoir, the drilling pressure, rotation speed and torque parameters are dynamically adjusted with the optimization objectives of maximizing mechanical drilling speed and minimizing real rock-breaking energy utilization rate.

[0012] Furthermore, the acquisition of logging parameters while drilling includes separating the drill string rotation torque and the effective rock-breaking torque of the drill bit based on torque, by acquiring the real-time torque recorded by the logging system when the drill bit is in normal contact with the bottom of the well during rotating drilling. The drilling string rotation torque recorded by the logging system while the drill bit is lifted from the bottom of the well and idling at its original speed. Based on real-time torque With drill string rotation torque The difference calculation is used to separate the effective rock-breaking torque of the drill bit. .

[0013] Furthermore, the method of constructing a rock-breaking energy utilization rate calculation model using drilling logging parameters includes incorporating the effective rock-breaking torque of the drill bit. By substituting the actual drill bit rotation speed into the modified mechanical specific energy formula, the actual rock-breaking energy utilization rate can be calculated. and will As a characteristic parameter of rock-breaking energy utilization rate, the actual drill bit rotation speed is the sum of the top drive rotation speed and the output rotation speed of the downhole power drilling tool. The formula for calculating the actual rock-breaking energy utilization rate is as follows:

[0014] ;

[0015] in, To ensure the true utilization rate of rock-breaking energy, For drilling pressure, The cross-sectional area of ​​the wellbore. The top drive speed is... The output speed of the downhole power drilling tool. This refers to the real-time torque recorded by the logging system during normal drilling. The logging system records the drill string rotation torque when the drill bit is pulled away from the bottom of the well and spinning freely. This refers to the mechanical drilling speed.

[0016] Furthermore, the sample dataset undergoes multi-stage preprocessing, including data cleaning based on the physical feasible region constraints and statistical distribution characteristics of drilling parameters. The cleaned dataset then undergoes double smoothing of input and output features. For the input features, an exponential moving average filter is used to suppress high-frequency random noise, and for the output features, a sliding window averaging method is used to suppress local fluctuations. The smoothed dataset is then converted to a sliding window format, using drilling parameter vectors from multiple consecutive historical time points as input features and the rock-breaking energy utilization rate corresponding to the next time point as the output label. This constructs time-series sample pairs suitable for training a long short-term memory network. The sliding window averaging expression for the output features is:

[0017] ;

[0018] in, Let be the smoothed output feature value at time t. Let i be the smoothed input feature value at time i. The smaller value between the current time sequence number and the sliding window length is taken.

[0019] Furthermore, the construction of the long short-term memory neural network model includes:

[0020] Construct long short-term memory network units containing forget gates, input gates, and output gates, and control the forgetting of historical information and the input of current information through gating mechanisms;

[0021] A time-step attention mechanism is introduced between the long short-term memory network layer and the output layer, and the attention weight of each time step is calculated based on the correlation between the hidden state and the current task.

[0022] The network parameters are iteratively optimized using a batch gradient descent algorithm with L2 regularization, and an early stopping mechanism is implemented to save the optimal model parameters.

[0023] The weight calculation of the time-step attention mechanism incorporates physical prior constraints:

[0024] ;

[0025] in, Let be the attention weight at time t. Let be the hidden state at time t. Let be the cell state at time t. , and This is the weight matrix. This is a physical prior function based on drilling conditions, used to quantify the correlation between historical rock-breaking energy utilization and the current prediction target. As a physical prior influence factor, For time step.

[0026] Furthermore, the threshold values ​​for combined deep coal and gas reservoir operating conditions include:

[0027] The rock-breaking energy utilization rate predicted in real time by the long short-term memory neural network model and the measured mechanical drilling speed are obtained to construct a dual-objective working condition discrimination index.

[0028] The working condition discrimination index is compared with a preset multi-level working condition threshold range to identify the level of the current drilling working condition, wherein:

[0029] When the rock-breaking energy utilization rate is in the range of 80MPa to 120MPa and the mechanical drilling speed is not less than 36m / h, it is determined to be the optimal working condition.

[0030] When the rock-breaking energy utilization rate is in the range of 120MPa to 150MPa and the mechanical drilling speed is in the range of 24m / h to 36m / h, it is determined to be a suboptimal working condition.

[0031] When the rock-breaking energy utilization rate is greater than 150MPa or the mechanical drilling speed is less than 24m / h, it is determined to be a risky working condition.

[0032] When the rock-breaking energy utilization rate is less than 80MPa and the mechanical drilling speed is less than 30m / h, it is judged as an inefficient working condition.

[0033] The working condition discrimination index is constructed based on the synergistic normalization of rock breaking energy efficiency and drilling efficiency:

[0034] ;

[0035] in, As a working condition discrimination index, These are the weighting coefficients. To improve the energy utilization rate of rock breaking, and These are the lower and upper limits of the rock-breaking energy utilization rate threshold, respectively. To measure the mechanical drilling speed, and These are the lower and upper limits of the mechanical drilling rate threshold, respectively.

[0036] Furthermore, the optimization objective of maximizing mechanical drilling speed and minimizing actual rock-breaking energy utilization includes constructing a multi-objective collaborative optimization function based on the identified working condition level, integrating a rock-breaking energy efficiency suppression term, a mechanical drilling speed improvement term, and a parameter adjustment range constraint term, solving the optimal solution of the multi-objective collaborative optimization function, and determining the drilling pressure adjustment amount, rotational speed adjustment amount, and torque adjustment amount to make the drilling parameters converge towards the optimization objective. The drilling pressure adjustment amount ranges from 5 to 10 kN, and the rotational speed adjustment amount ranges from 5 to 10 RPM.

[0037] Secondly, a coal and rock gas horizontal well drilling parameter optimization system includes:

[0038] The rock-breaking energy efficiency calculation module is configured to: acquire logging-while-drilling parameters, construct a rock-breaking energy utilization rate calculation model using the logging-while-drilling parameters, determine the characteristic parameters of rock-breaking energy utilization rate under different well depth conditions, and form a sample dataset. The logging-while-drilling parameters include drilling pressure, rotational speed, torque, and mechanical drilling speed.

[0039] The time series prediction model building module is configured to: perform multi-stage preprocessing on the sample dataset, build a long short-term memory neural network model, capture long-term time series dependencies between drilling parameters through a sliding window method, and establish a nonlinear dynamic mapping between drilling parameters and actual rock breaking energy utilization rate through parameter optimization and training.

[0040] The real-time optimization control module is configured to embed the trained long short-term memory neural network model into the logging system, and dynamically adjust the drilling pressure, rotation speed and torque parameters based on the real rock-breaking energy utilization rate predicted in real time and combined with the working condition threshold of deep coal and gas reservoirs, with the optimization objectives of maximizing mechanical drilling speed and minimizing real rock-breaking energy utilization rate.

[0041] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the method for optimizing drilling parameters in a horizontal well for coal and gas.

[0042] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide the method for optimizing drilling parameters of a coal and gas horizontal well.

[0043] In summary, the present invention has the following beneficial technical effects:

[0044] 1. This invention constructs a rock-breaking energy utilization rate calculation model that integrates drill string torque separation algorithm, accurately identifies the difference between the effective rock-breaking torque of the drill bit and the rotation torque of the drill string, and establishes a rock-breaking energy efficiency evaluation index that conforms to the characteristics of deep coal and gas reservoirs. This solves the technical problem of large calculation error of rock-breaking energy utilization rate in the prior art and difficulty in truly reflecting the rock-breaking state of the drill bit, and significantly improves the physical accuracy and reliability of drilling parameter optimization.

[0045] 2. This invention constructs a long short-term memory neural network model that integrates attention mechanism and uses a sliding window method to capture the long-term temporal dependencies between drilling parameters. It establishes a nonlinear dynamic mapping between drilling parameters and rock breaking energy utilization rate, which solves the technical problems of gradient vanishing and difficulty in mining the temporal evolution law of historical data in traditional BP neural networks. It significantly improves the timeliness and accuracy of drilling parameter prediction under complex working conditions.

[0046] 3. This invention embeds the trained nonlinear dynamic mapping model into the logging system. Based on the real-time predicted rock-breaking energy utilization rate combined with the working condition threshold of deep coal and gas reservoirs, it dynamically adjusts the drilling pressure, rotation speed, and torque parameters with the optimization objectives of maximizing mechanical drilling speed and minimizing rock-breaking energy utilization rate. This solves the technical problems of the traditional empirical method and mathematical model parameter adjustment being lagging and unable to adapt to dynamic geological conditions in real time, and realizes real-time closed-loop optimization of drilling parameters for deep coal and gas horizontal wells.

[0047] 4. This invention, without changing the existing drilling and logging process or adding new downhole measurement equipment, fully utilizes the temporal characteristics and physical prior knowledge of drilling and logging data to construct a data-driven intelligent optimization system for drilling parameters. This solves the technical problem of high costs caused by existing technologies relying on additional sensors and well-to-surface communication equipment, significantly reduces the cost of technology implementation, and improves the safety and overall benefits of drilling operations in deep coal and gas horizontal wells. Attached Figure Description

[0048] Figure 1 This is a flowchart of the method for optimizing drilling parameters in deep coal and rock gas horizontal wells provided in an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of the normal drilling torque and the rotational torque after the drill bit is lifted off the bottom of the well, recorded during the actual drilling process in this embodiment of the invention.

[0050] Figure 3 This is a thermogram showing the correlation between drilling parameters in an embodiment of the present invention;

[0051] Figure 4 This is a graph showing the impact of the time step on the training results of the model in an embodiment of the present invention;

[0052] Figure 5This is a diagram illustrating the impact of the number of hidden layers on the training results of the model in an embodiment of the present invention.

[0053] Figure 6 This is a diagram showing the impact of the number of hidden layer neurons on the training results of the model in an embodiment of the present invention;

[0054] Figure 7 This is a graph showing the impact of the number of iterations on the training results of the model in an embodiment of the present invention;

[0055] Figure 8 This is a fitting graph of the prediction results of the LSTM model in an embodiment of the present invention;

[0056] Figure 9 This is a graph showing the loss variation during the training process of the LSTM model in an embodiment of the present invention. Detailed Implementation

[0057] The present invention will be further described in detail below with reference to the accompanying drawings.

[0058] Example 1

[0059] Reference Figure 1 This embodiment of a method for optimizing drilling parameters in a horizontal coal gas well includes:

[0060] S1. Obtain the logging-while-drilling parameters and use the logging-while-drilling parameters to construct a rock-breaking energy utilization rate calculation model, determine the characteristic parameters of rock-breaking energy utilization rate under different well depth conditions, and form a sample dataset. The logging-while-drilling parameters include drilling pressure, rotational speed, torque, and mechanical drilling speed.

[0061] S2. Perform multi-stage preprocessing on the sample dataset and construct a long short-term memory neural network model. Capture the long-term temporal dependencies between drilling parameters through a sliding window method. Establish a nonlinear dynamic mapping between drilling parameters and actual rock-breaking energy utilization rate through parameter optimization and training.

[0062] S3. Embed the trained long short-term memory neural network model into the logging system. Based on the real rock-breaking energy utilization rate predicted in real time, combined with the working condition threshold of deep coal and gas reservoir, the drilling pressure, rotation speed and torque parameters are dynamically adjusted with the optimization goal of maximizing mechanical drilling speed and minimizing real rock-breaking energy utilization rate.

[0063] Specifically, a method for optimizing drilling parameters in horizontal coal and rock gas wells includes the following:

[0064] like Figure 1As shown in Figure S1, firstly, multi-source parameters during the drilling process of deep coal and gas horizontal wells are collected in real time through the logging-while-drilling system. These multi-source parameters include drilling pressure, rotational speed, torque, and mechanical drilling speed. Drilling pressure is the axial pressure applied by the drill string to the drill bit, measured in Newtons (N). Rotational speed includes the surface top drive speed and the output speed of the downhole power drill. The top drive speed is directly recorded by the surface logging system, while the downhole power drill speed is determined by the displacement-speed curve provided by its manufacturer. Torque is the rotational resistance torque recorded by the logging system. Mechanical drilling speed is the drilling footage per unit time.

[0065] Secondly, based on the above torque parameters, the drill string rotation torque and the effective rock-breaking torque of the drill bit are calculated separately. Specifically, under normal contact bottom hole rotary drilling conditions, the real-time torque value is read and recorded through the logging system. This torque value is the sum of the drill string rotation torque and the effective rock-breaking torque of the drill bit. The drill string is then pulled up to remove the drill bit from the bottom of the well, maintaining the top drive speed at the same rate as during normal drilling. At this point, the drill bit is no longer in contact with the bottom rock, and the torque value recorded by the logging system is read. This value represents the rotational friction torque when the drill string is idling. Since the drill string needs to be moved up and down frequently to remove cuttings during horizontal well drilling, this idling torque can be measured during normal operations without additional downtime. Based on the above two measurement results, the effective rock-breaking torque of the drill bit is obtained by difference calculation. ,Right now ,in, This represents the net torque of the drill bit when it is used to break rocks, expressed in Newton-meters (N·m).

[0066] Then, the effective rock-breaking torque of the drill bit obtained from the above separation is... By substituting the actual drill bit rotation speed into the corrected mechanical specific energy formula, a calculation model for the actual rock-breaking energy utilization rate is constructed. The actual drill bit rotation speed is the top drive rotation speed. Output speed of downhole power drilling tools The sum of, among which This refers to the rotational speed of the ground turntable or top drive, measured in revolutions per second (r / s). The downhole output rotational speed of screw drills or turbine drills is obtained by converting displacement; the actual rock-breaking energy utilization rate is... Calculated using the following formula:

[0067] ;

[0068] in, The true rock-breaking energy utilization rate represents the energy consumed by the drill bit to break a unit volume of rock, and the unit is megapascal (MPa). The drilling pressure applied to the drill bit. The wellbore cross-sectional area is determined by the drill bit diameter and is expressed in square meters. The top drive speed is... The output speed of the downhole power drilling tool. This refers to the real-time torque recorded by the logging system during normal drilling. The logging system records the drill string rotation torque when the drill bit is pulled away from the bottom of the well and spinning freely. This refers to the mechanical drilling speed, also known as the drilling rate, measured in meters per second. The unit conversion factor is used to convert the product of rotational speed and torque into power and divide it by the mechanical drilling speed to obtain the specific energy value.

[0069] because It accurately reflects the net rock-breaking torque of the drill bit, rather than the total torque on the ground, and This corrected formula reflects the actual drill bit rotation speed, not just the surface rotation speed. It overcomes the theoretical shortcomings of traditional mechanical energy specific models that conflate drill string friction torque with drill bit rock-breaking torque and equate surface rotation speed with drill bit rotation speed. It can accurately reflect the actual rock-breaking energy utilization efficiency of the drill bit under different well depth conditions. Finally, the drilling pressure, rotation speed, torque, mechanical drilling speed, and the calculated actual rock-breaking energy utilization rate at each well depth are compared. The data is stored in association to form a sample dataset containing time-series features and energy efficiency labels.

[0070] S2. Perform multi-stage preprocessing on the sample dataset, employing a phased, progressive data cleaning and feature engineering strategy to eliminate noise interference and anomalous disturbances in the original logging-while-drilling data, and construct a high-quality time-series dataset suitable for training long short-term memory neural networks, specifically including:

[0071] The first phase adopts 3 The +IQR criterion employs a two-stage outlier detection method. First, a coarse screening is performed based on the physical feasible region constraints of drilling parameters, eliminating invalid data that exceeds the operating limits of the drilling equipment or violates geological logic. Then, a finer detection is conducted based on statistical deviation metrics, calculating the mean of each parameter sequence. with standard deviation Set three standard deviations as the threshold boundary, and retain values ​​within the boundary. Valid data that conforms to the characteristics of operating condition fluctuations, after 3 The filtered dataset is further processed using the IQR interquartile range criterion to remove local distribution outliers and reduce the impact of distribution skewness on data quality. The upper and lower quartiles and interquartile range of the sequence are calculated. , And IQR, retain within the boundary Valid data that conforms to the characteristics of the working conditions;

[0072] The second stage applies double smoothing to the input and output features of the dataset processed in the first stage. An exponential moving average filter is used to suppress high-frequency random measurement noise in the input feature sequence. The window length is determined based on the sampling frequency and the drill bit's rock-breaking cycle. The exponential moving average at time t is:

[0073] ;

[0074] in, for The exponential moving average at time t. Let be the feature vector at time t, containing drilling pressure, rotational speed, torque, and mechanical drilling speed, initially... The value is equal to the original feature vector. Finally, the smoothed dataset is converted into a sliding window format to extract time-series feature vectors. The time step is set to L, and the drilling parameter vectors from L consecutive historical time points are used... As input feature matrix Using the rock-breaking energy utilization rate at time t+1 as the output label, sample pairs suitable for training a long short-term memory network are constructed, where the temporal feature matrix satisfies:

[0075] ;

[0076] in, Input feature dimensions include drill pressure, rotational speed, torque, and mechanical drilling speed.

[0077] A sliding window average smoothing method is applied to the output feature sequence to further suppress local fluctuations, and the final output is:

[0078] ;

[0079] in, To output the feature vector, For a moment, Let i be the smoothed input feature value at time i. To retrieve the current time sequence number t The smaller value among the sliding window lengths of 4 is used to adaptively handle the situation of insufficient sample size in the initial stage (when t<4), ensuring effective smoothing in the early stage of data accumulation, and averaging the results predicted by the LSTM model through the sliding window to obtain more accurate output results.

[0080] Secondly, a long short-term memory (LSTM) neural network model incorporating an attention mechanism is constructed based on the preprocessed dataset. The network's underlying layer consists of multiple cascaded LSM network units, each of which controls the forgetting of historical information and the input of current information through a gating mechanism, specifically including: a forgetting gate. Based on the current input Hidden state from the previous moment The historical information retention ratio is calculated using the sigmoid activation function, and the input gate... With candidate memory units The collaboration determines the proportion of current input information stored, and the output gate... Regulating the state of memory cells To hidden state The transformation is achieved by using gating collaboration to realize the stable gradient transfer of long-period time-series features.

[0081] A temporal attention mechanism is introduced between the top layer of the network, namely the Long Short-Term Memory (LSTM) network layer and the fully connected output layer. Attention weights are calculated based on the correlation between the hidden states at each historical time step and the current prediction task. Physical prior constraints are introduced into the traditional attention mechanism to strengthen the temporal correlation of drilling conditions, where attention scores... The following is obtained through a nonlinear transformation of the hidden state, cell state, and physical prior function:

[0082] ;

[0083] The attention weight at time t is obtained by normalizing the attention score using the Softmax function. :

[0084] ;

[0085] The context vector is obtained by weighted summation. As input to the fully connected layer:

[0086] ;

[0087] in, Let be the attention weight at time t. Let be the hidden state at time t. Let be the cell state at time t. This is the attention layer weight matrix. and This is the transformation matrix between the hidden state and the cell state. This is a physical prior function based on drilling conditions, used to quantify the correlation between historical rock-breaking energy utilization and the current prediction target. As a physical prior influence factor, For time step.

[0088] Finally, a batch gradient descent algorithm with L2 regularization is used to iteratively optimize the network parameters. The batch size is set to batch_size, and the Adam optimizer is used to adaptively adjust the update step size of each parameter. The loss function L is the sum of the mean square error between the predicted and the true values ​​and the L2 regularization term.

[0089] ;

[0090] in, This refers to the number of samples processed in a batch. This represents the true rock-breaking energy utilization rate of the k-th sample. These are the model's predicted values. The L2 regularization coefficient is... The total number of network layers. Let be the weight matrix of the l-th layer. It uses the Frobenius norm. During training, the validation set loss value is monitored for changes, and an early stopping mechanism is implemented. Early stopping occurs when the validation loss no longer decreases within a consecutive pause of iterations and the early stopping condition is met.

[0091] ;

[0092] Save the current optimal model parameters and terminate training. To verify the set loss, For the current iteration round, To pre-determine the tolerance cycle, a high-precision nonlinear dynamic mapping model between drilling parameters and actual rock-breaking energy utilization rate was finally established.

[0093] S3. The Long Short-Term Memory (LSTM) neural network model is embedded into the existing logging system through a standardized interface to establish a real-time data transmission channel. This allows the model to receive real-time data on drilling pressure, rotational speed, torque, and mechanical drilling speed collected by the logging system, and output a predicted value for rock-breaking energy utilization at the current well depth. Based on the real-time predicted rock-breaking energy utilization and the measured mechanical drilling speed of the logging system, a dual-objective working condition discrimination index is constructed. The current drilling working condition is quantified through the synergistic normalization of rock-breaking energy efficiency and drilling efficiency. The calculation formula for the working condition discrimination index is as follows:

[0094] ;

[0095] in, This is the operating condition discrimination index; the smaller the value, the better the operating condition. Let be the weight coefficient, and satisfy... This is used to balance the relative importance of rock-breaking efficiency and mechanical drilling rate in determining operating conditions. To improve the energy utilization rate of rock breaking, and These are the lower and upper limits of the rock-breaking energy utilization rate threshold, determined based on the rock mechanical properties of the target reservoir and historical data statistics. To measure the mechanical drilling speed, and These are the acceptable lower and upper limits of mechanical drilling speed, determined according to the drilling engineering design.

[0096] Secondly, the calculated working condition discrimination index and rock-breaking energy utilization rate are compared with the preset multi-level working condition threshold range to identify the current drilling working condition level and trigger corresponding parameter adjustment strategies: when the real-time predicted rock-breaking energy utilization rate is in the range of 80MPa to 120MPa and the mechanical drilling rate is not lower than 36m / h, it is judged as the optimal working condition, and the current drilling pressure, rotation speed and torque are kept unchanged; when the rock-breaking energy utilization rate is in the range of 120MPa to 150MPa and the mechanical drilling rate is in the range of 24m / h to 36m / h, it is judged as the suboptimal working condition, and the rock-breaking efficiency needs to be improved; when the rock-breaking energy utilization rate is greater than 150MPa or the mechanical drilling rate is lower than 24m / h, it is judged as a risky working condition, with the risk of drill bit sticking or tooth breakage, and the rock-breaking intensity needs to be reduced immediately; when the rock-breaking energy utilization rate is lower than 80MPa and the mechanical drilling rate is lower than 30m / h, it is judged as an inefficient working condition, the drill bit's rock-breaking energy is not fully utilized but the drilling efficiency is low, and the parameters need to be adjusted to match the formation. The above four-level working condition discrimination mechanism enables real-time identification and classification of drilling working conditions.

[0097] The aforementioned working condition discrimination index aims to solve the engineering challenges of multi-objective conflict identification and heterogeneous parameter synergistic optimization during the drilling of deep coal and gas horizontal wells. It provides a mathematical and quantitative basis for threshold rule tables developed based on engineers' experience, due to the rock-breaking energy utilization rate... The rate of drilling (rDB) is the energy consumed to break a unit volume of rock. Both represent velocity, but their physical dimensions and numerical scales differ. It is usually in the range of 80-150MPa. Typically in the range of 20-40 m / h, the operating condition discrimination index maps the two to a unified dimensionless interval [0,1] through min-max normalization, with the aim of identifying blind spots in operating conditions that cannot be judged by independent thresholds. The overall structure constitutes the comprehensive operating condition loss index, which physically represents the comprehensive deviation of the current operating condition from the Pareto optimal frontier.

[0098] When the rock-breaking energy utilization rate and mechanical drilling rate meet the above threshold judgment conditions, they are assigned to the corresponding working conditions, including optimal working conditions, suboptimal working conditions, risky working conditions, and inefficient working conditions. However, when blind spots exist, for example when... (Second-best interval) and (In the risk range) there will be a decision-making dilemma involving a conflict between strategies to reduce energy consumption and increase drilling speed. The working condition discrimination index will... Inverse normalization This is transformed into a positive indicator (the smaller the value, the greater the efficiency loss), achieving dual-objective optimization in the same direction, thereby transforming multi-objective conflict into a single-objective extreme value problem.

[0099] Specifically, this embodiment uses the rock mechanical properties (uniaxial compressive strength 35-55 MPa, abrasiveness coefficient 0.8-1.2) of the deep coal-gas reservoir in the block where Well X is located and historical drilling data statistics as examples to determine the normalized benchmark parameters: Where 60 MPa corresponds to the lower limit of the theoretical specific energy for fracturing reservoir rocks, and 150 MPa corresponds to the critical specific energy for thermal damage of PDC drill bits. Among them, 15 m / h is the lower limit of the economic drilling speed for the block, and 45 m / h is the historical best drilling speed for the same block; considering the strong abrasiveness of the coal and rock in this block, the following values ​​are taken: To enhance the weighting of rock-breaking energy efficiency, the optimal working condition physical threshold is substituted into the... The calculation formula yields the optimal working condition. ,Right now and Similarly, by substituting the physical thresholds for other operating conditions, we can obtain the corresponding values ​​for suboptimal, risky, and inefficient operating conditions. interval, and then complete Mapping relationship with multi-level operating condition threshold ranges.

[0100] The working condition discrimination index A collaborative criterion system is formed with the rock-breaking energy utilization rate threshold judgment: the physical threshold range of rock-breaking energy utilization rate defines the boundary of geological conditions. Provides a dimensionless quantization standard, which is established by substituting physical thresholds into the formula for calculation. The mapping relationship between the threshold range and the multi-level operating conditions enables continuously changing... Numerical values ​​and discrete working condition levels form a quantitative correspondence, preserving the engineering intuitiveness of physical thresholds while achieving a dual-objective synergistic evaluation through a normalized index. It should be noted that the working condition level is determined first by the rock-breaking energy utilization rate threshold; the working condition discrimination index is only applied when there is a blind spot in the rock-breaking energy utilization rate threshold determination. The determination.

[0101] Finally, a multi-objective collaborative optimization function is constructed based on the identified working condition levels, with the optimization objectives being to maximize the mechanical drilling rate and minimize the rock-breaking energy utilization rate. Simultaneously, a parameter adjustment range constraint term is introduced to prevent downhole instability caused by drastic adjustments. The multi-objective collaborative optimization function is determined by the following formula:

[0102] ;

[0103] in, This represents the value of a multi-objective collaborative optimization function; a smaller value indicates a better optimization effect. , and These are the rock-breaking energy efficiency weighting coefficient, the mechanical drilling speed weighting coefficient, and the adjustment range penalty coefficient, respectively, and they satisfy the following conditions: For example, when the operating condition is suboptimal, the mechanical drilling rate can be prioritized to be increased while suppressing the excessively rapid increase in rock-breaking energy consumption through weighted configuration. The baseline value for rock-breaking energy utilization rate is the statistical average of the optimal rock-breaking energy efficiency of the target reservoir. The benchmark value for mechanical drilling rate is the best historical drilling rate for the same block. This is the drilling pressure adjustment amount. This is the speed adjustment amount. and These represent the maximum allowable adjustment ranges for drilling pressure and rotational speed, with the single adjustment range for drilling pressure being 5-10 kN and the single adjustment range for rotational speed being 5-10 RPM. Torque is adjusted in tandem with changes in drilling pressure and rotational speed. By solving for the minimum value of the above optimization functions, the optimal adjustment amounts for drilling pressure, rotational speed, and torque are determined, and parameter adjustment commands are generated and sent to the drilling control system. This achieves real-time closed-loop optimization of drilling parameters, enabling the drilling operation to converge toward the Pareto optimal frontier, which maximizes mechanical drilling speed and minimizes rock-breaking energy utilization.

[0104] Example 2

[0105] The difference between this embodiment and Embodiment 1 is that this embodiment provides an application example of a coal and rock gas horizontal well drilling parameter optimization method under specific well conditions;

[0106] This embodiment proposes a method for determining the effective drill bit torque by combining drilling testing. The method of obtaining, such as Figure 2 As shown, the normal drilling torque and the rotational torque after the drill bit is lifted off the bottom of the well are recorded during the actual drilling process. The main steps of this method are as follows:

[0107] First, under normal contact with the bottom of the well during rotary drilling, the real-time torque value is recorded using a surface logging system. ,at this time The value represents the sum of the drill string rotation torque and the effective rock-breaking torque of the drill bit (including the output torque of the downhole power drilling tools). In this embodiment, actual drilling data from a deep coal and gas horizontal well, X, is selected. This well maintained normal drilling from 10:30 to 10:32 on July 5, 2025, with a depth ranging from 4760.15m to 4761.42m, a drill pressure maintained between 78.41kN and 81.84kN, a top drive speed of 111RPM, and a displacement of 30L / min. The drilling parameters recorded by the logging system during this period are shown in Table 1.

[0108] Table 1. Some drilling parameters automatically recorded by the logging system during normal drilling of Well X;

[0109]

[0110] Calculations show that the average torque recorded by the logging system during this period is... It is 24.50 kN·m.

[0111] Subsequently, the drill string was raised 5 meters to disengage the drill bit from the bottom of the well. The top drive speed was maintained at 111 RPM and the displacement at 30 L / min. At this point, the pressure on the drill bit dropped to zero, and the drill bit no longer contacted the bottom rock. After the rotational cycle stabilized, the torque value recorded by the logging system was read. This value represents the rotational friction torque when the drill string is idling. In this embodiment, this operation was completed between 10:36 and 10:38, with the well depth maintained at 4761.42m and the drill bit position at 4756.15m. The logging data is shown in Table 2.

[0112] Table 2. Some drilling parameters automatically recorded by the logging system during normal rotation and rock cleaning when the drill bit of well X is lifted from the bottom of the well;

[0113]

[0114] Calculations show that the average torque during the time the drill bit is lifted from the bottom of the well is... It is 18.10 kN·m.

[0115] Based on the above two measurement results, the effective rock-breaking torque of the drill bit is obtained by difference calculation. :

[0116] ;

[0117] In the formula, To achieve effective rock-breaking torque, For torque and, The torque recorded by the logging system; substituting the data from Tables 1 and 2 into the above formula, the average effective rock-breaking torque of the drill bit under this well depth condition is calculated as follows: Since the well depth exceeded 4700m at this point, and the drill bit and drill string only moved up 5m, the torque consumed by the drill string in both periods can be considered equal. The average torque consumed by the drill string rotation during the subsequent 200m drilling depth... Continuing with a value of 18.10 kN·m, the introduced systematic error is approximately 200 / 4700 = 4.3%, which meets the engineering accuracy requirements. Alternatively, it can be established through multiple tests. The functional relationship between the well depth and the well depth is further used to eliminate systematic errors.

[0118] The drill bit is lowered back to the bottom of the well, and the drilling pressure is restored to normal to continue drilling. It should be noted that, since the drill string needs to be moved up and down frequently during horizontal well drilling to improve the rock-carrying efficiency of the horizontal section under the conditions of pump operation and drill string selection, the testing process in step ② above basically does not require special design and can be completed in normal rock clearing operations. Therefore, the method proposed in this embodiment can obtain the drill bit torque without changing the existing drilling process and logging equipment.

[0119] The obtained effective rock-breaking torque and other logging parameters were preprocessed by filling missing values, removing outliers, and normalizing. Correlation analysis was performed on drilling pressure, rotational speed, torque, and mechanical drilling speed, such as... Figure 3 As shown, Figure 3 A thermogram showing the correlation between drilling parameters was generated to confirm the relationship between each parameter and rock-breaking energy utilization rate. It has a significant correlation and is used as an input feature. As output labels, the training and test sets are divided in an 8:2 ratio to construct an LSTM model and optimize its parameters.

[0120] like Figure 4 As shown, the time step is first optimized. With a fixed number of hidden layers (1), a maximum number of iterations (200), 64 hidden layer neurons, and a batch size of 64, the model is trained and tested within a time step range of 1-10, and the output evaluation metrics MAE and _____ are calculated. The comparison revealed that when the time step was 7, the model... The highest value and the lowest MAE result indicate the best fit. Therefore, the optimal time step is determined to be 7, which means using data from the past 7 well depths to predict data from the 8th well depth.

[0121] like Figure 5 As shown, the next step was to optimize the number of hidden layers. With a fixed time step of 7, a maximum number of iterations of 200, 64 neurons in the hidden layers, and a batch size of 64, training and testing were conducted within the range of 1-10 hidden layers. Comparison of evaluation metrics revealed that the model achieved the best fit when the number of hidden layers was 3; therefore, the optimal number of hidden layers was determined to be 3.

[0122] like Figure 6 As shown, the number of hidden layer neurons was fine-tuned again. With a fixed time step of 7, 3 hidden layers, a maximum number of iterations of 200, and a batch size of 64, training and testing were performed with 8, 16, 32, 64, 128, and 256 neurons. The comparison revealed that the model achieved the best prediction performance when the number of hidden layer neurons was 64; therefore, the optimal number of neurons was determined to be 64.

[0123] like Figure 7 As shown, the final iteration count was optimized. With a fixed time step of 7, 3 hidden layers, 64 hidden layer neurons, and a batch size of 64, training and testing were performed within the iteration range of 100-800. The comparison revealed that the model achieved the best convergence performance with 500 iterations; therefore, the optimal iteration count was determined to be 500.

[0124] The optimal parameter combination was adopted as follows: time step 7, number of hidden layers 3, number of hidden layer neurons 64, batch size 64, and number of iterations 500. The Adam algorithm was selected for optimization, with a learning rate of 0.0001, activation functions of sigmoid and tanh, and mean squared error (MSE) as the loss function. The model was then trained and tested. Figure 8 As shown, the model determination coefficients are obtained. A value of 0.92 indicates that the predicted result fits the true value well; for example... Figure 9 As shown, the Loss value decreases rapidly and converges quickly to a low level during training, with small fluctuations and eventually becomes smooth, indicating that the model has stable and efficient performance and is suitable for predicting drilling parameter time series data.

[0125] The trained LSTM model and related parameter thresholds are embedded into the logging system for real-time prediction of subsequent new data. The rock-breaking energy utilization rate is predicted in real-time based on the LSTM model. Based on the measured mechanical drilling speed, we formulated working condition classification thresholds and corresponding parameter adjustment rules.

[0126] In this embodiment, based on the drilling data of Well X, a formula is developed. The working condition classification thresholds and parameter adjustment rules for mechanical drilling speed are shown in Table 3:

[0127] Table 3 Table of working condition classification thresholds and corresponding parameter adjustment rules for mechanical drilling speed;

[0128]

[0129] When the model predicts specific MSEb When the value is set, the current working condition level is identified by combining the real-time mechanical drilling rate and referring to the threshold range shown in Table 3, and the corresponding parameter adjustment rules are executed. For example, when... When the drilling pressure is between 120MPa and 150MPa and the mechanical drilling speed is between 24m / h and 36m / h, it is considered a suboptimal working condition. The system automatically suggests increasing the drilling pressure by 5kN to 10kN, the rotational speed by 5RPM to 8RPM, and the torque is adaptively adjusted according to the drilling pressure to increase the mechanical drilling speed. When the pressure exceeds 150 MPa or the mechanical drilling speed is below 24 m / h, it is considered a risky working condition, with the risk of drill bit sticking or tooth breakage. The system recommends reducing the drilling pressure by 10 kN to 15 kN, the rotation speed by 8 RPM to 12 RPM, and the torque accordingly, in order to reduce the utilization rate of rock breaking energy and avoid downhole accidents.

[0130] The above parameter combination is based on actual drilling data from Well X and can be used for subsequent drilling operations under the same reservoir conditions in the same block. When the drilling location is changed, the threshold parameters are adaptively adjusted according to the specific geological conditions and equipment performance on site, and a threshold parameter rule table applicable to the drilling site is formulated to achieve real-time optimization of drilling parameters for deep coal and gas horizontal wells, improve mechanical drilling speed, reduce rock breaking energy utilization, avoid drill bit sticking, tooth breakage and other accidents, extend drill bit life, shorten drilling cycle and reduce drilling costs.

[0131] The table above is a rule table formulated based on drilling data from Well X. This parameter combination can be used for subsequent drilling operations under the same reservoir conditions in the same block. When changing the drilling location, corresponding adjustments should be made according to the specific site conditions, and a threshold parameter rule table applicable to the drilling site should be formulated.

[0132] Example 3

[0133] This embodiment provides a coal and gas horizontal well drilling parameter optimization system, including:

[0134] The rock-breaking energy efficiency calculation module is configured to: acquire logging-while-drilling parameters, construct a rock-breaking energy utilization rate calculation model using the logging-while-drilling parameters, determine the characteristic parameters of rock-breaking energy utilization rate under different well depth conditions, and form a sample dataset. The logging-while-drilling parameters include drilling pressure, rotational speed, torque, and mechanical drilling speed.

[0135] The time series prediction model building module is configured to: perform multi-stage preprocessing on the sample dataset, build a long short-term memory neural network model, capture long-term time series dependencies between drilling parameters through a sliding window method, and establish a nonlinear dynamic mapping between drilling parameters and actual rock breaking energy utilization rate through parameter optimization and training.

[0136] The real-time optimization control module is configured to embed the trained long short-term memory neural network model into the logging system, and dynamically adjust the drilling pressure, rotation speed and torque parameters based on the real rock-breaking energy utilization rate predicted in real time and combined with the working condition threshold of deep coal and gas reservoirs, with the optimization objectives of maximizing mechanical drilling speed and minimizing real rock-breaking energy utilization rate.

[0137] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the method for optimizing drilling parameters in a horizontal coal gas well.

[0138] A terminal device includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide a method for optimizing drilling parameters in a horizontal coal gas well.

[0139] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing drilling parameters in horizontal coal and rock gas wells, characterized in that, include: The drilling logging parameters are obtained, including drilling pressure, rotation speed, torque and mechanical drilling speed. The effective rock-breaking torque of the drill bit is obtained based on torque separation. The effective rock-breaking torque of the drill bit is the difference between the real-time torque during normal drilling and the idle torque. A calculation model for rock breaking energy utilization rate was constructed using logging-while-drilling parameters and effective rock breaking torque of the drill bit. Characteristic parameters of rock breaking energy utilization rate were determined, and a sample dataset was formed. The sample dataset undergoes multi-stage preprocessing, which includes preprocessing based on... Outlier detection, double smoothing of input and output features, and sliding window temporal reconstruction are all based on the criteria. A long short-term memory neural network model was constructed, and the long-term temporal dependencies between drilling parameters were captured by a sliding window to establish a nonlinear dynamic mapping between drilling parameters and actual rock-breaking energy utilization. The construction of the long short-term memory neural network model includes: Construct long short-term memory network units containing forget gates, input gates, and output gates, cascade multiple long short-term memory network units to form a long short-term memory network layer, and control the forgetting of historical information and the input of current information through a gating mechanism; A fully connected output layer is constructed, and a time-step attention mechanism is introduced between the long short-term memory network layer and the output layer. The attention weights of each time step are calculated based on the correlation between the hidden state and the current task. The network parameters are iteratively optimized using a batch gradient descent algorithm with L2 regularization, and an early stopping mechanism is implemented to save the optimal model parameters. The weight calculation of the time-step attention mechanism incorporates physical prior constraints: ； in, Let be the attention weight at time t. Let be the hidden state at time t. Let be the cell state at time t. , and This is the weight matrix. This is a physical prior function based on drilling conditions, used to quantify the correlation between historical rock-breaking energy utilization and the current prediction target. As a physical prior influence factor, For time step; The trained long short-term memory neural network model is embedded into the logging system. Based on the real rock-breaking energy utilization rate predicted in real time and combined with the working condition threshold of deep coal and gas reservoir, the drilling pressure, rotation speed and torque parameters are dynamically adjusted with the optimization objectives of maximizing mechanical drilling speed and minimizing real rock-breaking energy utilization rate.

2. The method for optimizing drilling parameters in a horizontal coal gas well according to claim 1, characterized in that, The acquisition of logging parameters while drilling includes separating the drill string rotation torque and the effective rock-breaking torque of the drill bit based on torque, and acquiring the real-time torque recorded by the logging system when the drill bit is in normal contact with the bottom of the well during rotating drilling. The drilling string rotation torque recorded by the logging system while the drill bit is lifted from the bottom of the well and idling at its original speed. Based on real-time torque With drill string rotation torque The difference calculation is used to separate the effective rock-breaking torque of the drill bit. .

3. The method for optimizing drilling parameters in a horizontal coal gas well according to claim 2, characterized in that, The method for constructing a rock-breaking energy utilization rate calculation model using drilling logging parameters and the effective rock-breaking torque of the drill bit includes calculating the effective rock-breaking torque of the drill bit. By substituting the actual drill bit rotation speed into the corrected mechanical specific energy formula, the actual rock-breaking energy utilization rate can be calculated. and will As a characteristic parameter of rock-breaking energy utilization rate, the actual drill bit rotation speed is the sum of the top drive rotation speed and the output rotation speed of the downhole power drilling tool. The formula for calculating the actual rock-breaking energy utilization rate is as follows: ； in, To ensure the true utilization rate of rock-breaking energy, For drilling pressure, The cross-sectional area of ​​the wellbore. The top drive speed is... The output speed of the downhole power drilling tool. This refers to the real-time torque recorded by the logging system during normal drilling. The logging system records the drill string rotation torque when the drill bit is pulled away from the bottom of the well and spinning freely. This refers to the mechanical drilling speed.

4. The method for optimizing drilling parameters of a horizontal coal gas well according to claim 1, characterized in that, The sample dataset undergoes multi-stage preprocessing, including data cleaning based on the physical feasible region constraints and statistical distribution characteristics of drilling parameters. The cleaned dataset then undergoes double smoothing of input and output features. For the input features, an exponential moving average filter is used to suppress high-frequency random noise, and for the output features, a sliding window averaging method is used to suppress local fluctuations. The smoothed dataset is then converted to a sliding window format, using drilling parameter vectors from multiple consecutive historical time points as input features and the rock-breaking energy utilization rate corresponding to the next time point as the output label. This constructs time-series sample pairs suitable for training a long short-term memory network. The sliding window averaging expression for the output features is as follows: ； in, Let be the smoothed output feature value at time t. Let i be the smoothed input feature value. The smaller value between the current time sequence number and the sliding window length is taken.

5. The method for optimizing drilling parameters of a horizontal coal gas well according to claim 3, characterized in that, The threshold values ​​for combined deep coal and gas reservoir conditions include: The rock-breaking energy utilization rate predicted in real time by the long short-term memory neural network model and the measured mechanical drilling speed are obtained to construct a dual-objective working condition discrimination index. The working condition discrimination index is compared with a preset multi-level working condition threshold range to identify the level of the current drilling working condition, wherein: When the rock-breaking energy utilization rate is in the range of 80MPa to 120MPa and the mechanical drilling speed is not less than 36m / h, it is determined to be the optimal working condition. When the rock-breaking energy utilization rate is in the range of 120MPa to 150MPa and the mechanical drilling speed is in the range of 24m / h to 36m / h, it is determined to be a suboptimal working condition. When the rock-breaking energy utilization rate is greater than 150MPa or the mechanical drilling speed is less than 24m / h, it is determined to be a risky working condition. When the rock-breaking energy utilization rate is less than 80MPa and the mechanical drilling speed is less than 30m / h, it is judged as an inefficient working condition. The working condition discrimination index is constructed based on the synergistic normalization of rock breaking energy efficiency and drilling efficiency: ； in, As a working condition discrimination index, These are the weighting coefficients. To improve the energy utilization rate of rock breaking, and These are the lower and upper limits of the rock-breaking energy utilization rate threshold, respectively. To measure the mechanical drilling speed, and These are the lower and upper limits of the mechanical drilling rate threshold, respectively.

6. The method for optimizing drilling parameters of a horizontal coal gas well according to claim 5, characterized in that, The optimization objective of maximizing mechanical drilling speed and minimizing actual rock-breaking energy utilization includes constructing a multi-objective collaborative optimization function based on the identified working condition level, integrating rock-breaking energy efficiency suppression, mechanical drilling speed enhancement, and parameter adjustment range constraints; solving the optimal solution of the multi-objective collaborative optimization function; and determining the drilling pressure adjustment, rotational speed adjustment, and torque adjustment to make the drilling parameters converge towards the optimization objective. The drilling pressure adjustment range is 5-10 kN, and the rotational speed adjustment range is 5-10 RPM.

7. A coal and rock gas horizontal well drilling parameter optimization system, executing the method described in claim 1, characterized in that, include: The rock-breaking energy efficiency calculation module is configured to: acquire logging-while-drilling parameters, construct a rock-breaking energy utilization rate calculation model using the logging-while-drilling parameters, determine the characteristic parameters of rock-breaking energy utilization rate under different well depth conditions, and form a sample dataset. The logging-while-drilling parameters include drilling pressure, rotational speed, torque, and mechanical drilling speed. The time series prediction model building module is configured to: perform multi-stage preprocessing on the sample dataset, build a long short-term memory neural network model, capture long-term time series dependencies between drilling parameters through a sliding window method, and establish a nonlinear dynamic mapping between drilling parameters and actual rock breaking energy utilization rate through parameter optimization and training. The real-time optimization control module is configured to embed the trained long short-term memory neural network model into the logging system, and dynamically adjust the drilling pressure, rotation speed and torque parameters based on the real rock-breaking energy utilization rate predicted in real time and combined with the working condition threshold of deep coal and gas reservoirs, with the optimization objectives of maximizing mechanical drilling speed and minimizing real rock-breaking energy utilization rate.

8. A computer-readable storage medium storing a plurality of instructions, characterized in that, The instructions are adapted to be loaded and executed by the processor of the terminal device as described in claim 1, which is a method for optimizing drilling parameters in a horizontal coal gas well.

9. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is configured to implement instructions; and the computer-readable storage medium is configured to store multiple instructions, characterized in that, The instructions are adapted to be loaded by a processor and executed as described in claim 1, a method for optimizing drilling parameters in a horizontal coal gas well.

Images