Neural network adaptive control system and method for engineering gas extraction drilling machine

Abstract

The invention relates to the technical field of coal mine gas extraction, in particular to a neural network self-adaptive control system and method for an engineering gas extraction drilling machine, and the system comprises a data collection module, a dual-network controller module, an execution mechanism module and a monitoring module. The data acquisition module acquires working condition data such as drilling machine rotating speed and torque in real time; the dual-network controller module integrates a fireworks algorithm to optimize a PID controller (FWA-BP-PID) of a BP neural network and an RBF neural network adaptive sliding mode controller (RBF-SMC), and a control strategy is switched through the working condition identification module; the execution mechanism module adjusts the drilling machine rotating speed according to the control signal; and the monitoring module realizes parameter real-time display and fault early warning. According to the method, the problems that a traditional control algorithm is contradictory in complexity and real-time performance and insufficient in multi-working-condition adaptability are solved, and stable and efficient drilling of the drilling machine under the complex coal seam condition is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine gas extraction, and in particular to an engineered gas extraction drill rig neural network adaptive control system and method. Background Art

[0002] Gas extraction drills are core equipment for coal mine gas management, and their control performance directly impacts gas extraction efficiency and operational safety. Traditional control methods, such as PID control, struggle to cope with the nonlinearity of drilling systems, time-varying parameters, and load disturbances under complex operating conditions, resulting in low drilling efficiency and significant equipment loss. While existing neural network control algorithms can improve robustness, they suffer from the following issues: The conflict between algorithm complexity and real-time performance: A single neural network model is computationally intensive when dealing with multivariable coupled systems, making it difficult to meet the real-time control requirements of drilling rigs. For example, traditional BP neural networks require multiple iterations to optimize weights, which can lead to control lags under sudden coal seam changes.

[0003] Inadequate adaptability to multiple operating conditions: During gas extraction, parameters such as coal seam hardness and moisture content change frequently, making a single control strategy unable to balance stable drilling with anti-interference requirements. For example, when drilling in soft coal seams, traditional sliding mode control is prone to chattering; when drilling in hard rock formations, PID control response lags, leading to the risk of drill sticking.

[0004] Limited engineering applications: Existing algorithms are often based on idealized simulation environments and fail to fully consider real-world factors such as strong electromagnetic interference and sensor noise in coal mines, resulting in insufficient controller robustness. For example, sensor noise can distort neural network input data, leading to incorrect adjustment of control parameters.

[0005] To address these issues, existing technologies attempt to improve performance by modifying neural network structures or integrating control strategies. However, these technologies still suffer from drawbacks such as insensitive switching between operating conditions and complex parameter tuning. Therefore, there is an urgent need to design an adaptive control system that balances algorithmic efficiency and engineering practicality to achieve stable and efficient control of gas extraction drills under multiple operating conditions. Summary of the Invention

[0006] The present invention aims to provide an engineered gas extraction drilling rig neural network adaptive control system and method, which is suitable for drilling process control in high-gas mines. Through the fusion of dual network controllers and intelligent working condition switching, it solves the real-time and multi-working condition adaptability problems of traditional control algorithms, and improves the control accuracy and reliability of the drilling rig under complex coal seam conditions.

[0007] The technical solution adopted by the present invention to solve the technical problem is: an engineered gas extraction drilling rig neural network adaptive control system, comprising: The data acquisition module is used to collect real-time operating data such as the rotation speed, torque, hydraulic pressure and coal seam hardness of the drilling rig's rotary system and perform pre-processing; A dual-network controller module, including an FWA-BP-PID controller, an RBF-SMC controller, and an operating condition identification module. The operating condition identification module switches the control strategy based on preset operating condition thresholds (speed fluctuation threshold ±5 rpm, torque change rate threshold ±8 N·m / s); The actuator module includes an electro-hydraulic proportional valve and a hydraulic motor, which is used to receive the voltage signal output by the controller and adjust the drilling rig speed; The monitoring module is used to display system parameters, control curves and fault warning information in real time.

[0008] Specifically, the fireworks algorithm optimization parameters of the FWA-BP-PID controller include: explosion radius adjustment constant A=6-10, explosion spark number adjustment constant M=8-12, mutation spark number=3-7, and iteration number=30-60.

[0009] Specifically, the sliding surface parameter c of the RBF-SMC controller is 0.5-1.2, the saturation function parameter α is 2-5, and the control law parameter η is 0.05-0.2.

[0010] Specifically, the data acquisition module uses high-precision sensors, with a rotation speed sensor accuracy of ±0.1r / s, a torque sensor accuracy of ±1N·m, and a sampling period of 0.005-0.02s.

[0011] The neural network adaptive control method for an engineered gas extraction drilling rig includes the following steps: S1: The data acquisition module collects working condition data in real time and inputs it into the dual network controller module after filtering and noise reduction; S2: The working condition identification module determines the working condition based on the speed fluctuation Δω and the torque change rate ΔT / t. When |Δω|≤3r / s and |ΔT / t|≤5N·m / s, it is determined to be a stable working condition and the FWA-BP-PID controller is activated. When |Δω|>3r / s or |ΔT / t|>5N·m / s, it is determined to be a complex working condition and the RBF-SMC controller is activated. S3: The controller outputs a control signal to the actuator module to adjust the opening of the electro-hydraulic proportional valve to achieve speed tracking; S4: The monitoring module displays the control effect in real time and automatically adjusts the controller parameters if overshoot or oscillation occurs.

[0012] Specifically, the parameter tuning process of the FWA-BP-PID controller includes: optimizing the initial weights of the BP neural network using the fireworks algorithm, and the fitness function is the root mean square error (RMSE), which is expressed as follows: ; Where P is the number of sampling points, r(k) is the reference speed, and y(k) is the actual speed.

[0013] Specifically, the chattering suppression method of the RBF-SMC controller includes: using an inverse tangent saturation function Instead of the traditional sign function, α is the saturation function parameter.

[0014] Specifically, the threshold of the working condition identification module can be dynamically adjusted through on-site debugging, with the adjustment step being ±0.5 r / s for the speed threshold and ±1 N·m / s for the torque change rate threshold.

[0015] Specifically, the dual network controller module supports online learning function, optimizes the neural network weights through historical operating data, and the optimization cycle is 5-10 minutes.

[0016] Specifically, the response time of the electro-hydraulic proportional valve of the actuator module is ≤50ms, the speed adjustment range of the hydraulic motor is 0-10r / s, and the control accuracy is ±0.2r / s.

[0017] Beneficial effects of the present invention: Multi-condition adaptive control: The operating condition identification module switches control strategies in real time. Under stable operating conditions, the FWA-BP-PID controller achieves precise tracking without overshoot (overshoot ≤ 0.1%). Under complex operating conditions, the RBF-SMC controller suppresses chattering (error fluctuation ≤ ±0.5r / s), improving the system's adaptability to coal seam changes.

[0018] Improved real-time and robustness: The Fireworks algorithm reduces the number of iterations for optimizing BP neural network weights by 40% compared to traditional methods. Combined with the fast approximation capability of the RBF neural network, the controller response time is ≤200ms, meeting real-time control requirements. Kalman filtering and anti-interference design enable the system to operate stably even in strong noise environments.

[0019] Parameter self-optimization capability: The dual-network controller supports online learning, dynamically adjusting the neural network weights based on historical data. The optimization cycle can be set to 5-10 minutes to continuously improve control accuracy.

[0020] Perfect engineering design: Integrated sensor noise reduction, fault warning and parameter visualization functions, support remote debugging and maintenance, reduce the difficulty of on-site operation, and meet the high reliability requirements of underground coal mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The present invention will be further described below with reference to the accompanying drawings and examples.

[0022] Figure 1 This is a diagram of the architecture of the neural network adaptive control system for the engineered gas extraction drilling rig provided by the present invention; Figure 2 This is an architectural diagram of the monitoring module in the neural network adaptive control system of the engineered gas extraction drilling rig provided by the present invention; Figure 3 This is a diagram of the architecture of the actuator module in the neural network adaptive control system of the engineered gas extraction drilling rig provided by the present invention; Figure 4 This is a diagram of the architecture of the data acquisition module in the neural network adaptive control system of the engineered gas extraction drilling rig provided by the present invention; Figure 5 This is an architectural diagram of the dual-network control module in the neural network adaptive control system for the engineered gas extraction drilling rig provided by the present invention; Figure 6 This is a flow chart of the neural network adaptive control method for the engineered gas extraction drilling rig provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0024] like Figure 1-5 As shown, the engineered gas extraction drilling rig neural network adaptive control system of the present invention includes: Data acquisition module: A high-precision speed sensor (accuracy ±0.1r / s), torque sensor (accuracy ±1N·m) and hydraulic pressure sensor are used to collect real-time data from the drilling rig's rotation system with a sampling period of 0.01s. The data is then input into the controller after noise reduction using the Kalman filter algorithm.

[0025] Dual network controller modules: FWA-BP-PID controller: Designed for precise control under stable operating conditions, it optimizes the initial weights of the BP neural network using the fireworks algorithm, addressing the problem of traditional BP neural networks easily falling into local optimality. The parameters for the fireworks algorithm are: explosion radius adjustment constant A = 8, explosion spark number adjustment constant M = 10, number of mutation sparks = 5, and number of iterations = 50.

[0026] RBF-SMC controller: Designed for interference-resistant control under complex operating conditions, it uses an inverse tangent saturation function to suppress chattering and an RBF neural network to approximate system uncertainties. The sliding surface parameter c = 0.9, the saturation function parameter α = 3, and the control law parameter η = 0.1.

[0027] Working condition identification module: This module identifies working conditions in real time based on speed fluctuation Δω and torque change rate ΔT / t. The preset stable working condition thresholds are |Δω|≤3r / s and |ΔT / t|≤5N·m / s. A fuzzy logic algorithm is used to achieve smooth switching of control strategies.

[0028] Actuator module: includes an electro-hydraulic proportional valve (response time ≤ 50ms) and a hydraulic motor (speed adjustment range 0-10r / s). It adjusts the drilling rig speed according to the voltage signal (0-5V) output by the controller, with a control accuracy of ±0.2r / s.

[0029] Monitoring module: The human-machine interface is designed based on Kingview software, which displays the speed, torque curve and controller parameters in real time. It integrates fault warning function and triggers sound and light alarm when the speed deviation exceeds ±10%.

[0030] like Figure 6 As shown, the neural network adaptive control method for an engineered gas extraction drilling rig according to the present invention comprises the following steps: S1: The data acquisition module collects working condition data in real time and inputs it into the dual network controller module after filtering and noise reduction; S2: The working condition identification module determines the working condition based on the speed fluctuation Δω and the torque change rate ΔT / t. When |Δω|≤3r / s and |ΔT / t|≤5N·m / s, it is determined to be a stable working condition and the FWA-BP-PID controller is activated. When |Δω|>3r / s or |ΔT / t|>5N·m / s, it is determined to be a complex working condition and the RBF-SMC controller is activated. S3: The controller outputs a control signal to the actuator module to adjust the opening of the electro-hydraulic proportional valve to achieve speed tracking; S4: The monitoring module displays the control effect in real time and automatically adjusts the controller parameters if overshoot or oscillation occurs.

[0031] In S2, the parameter tuning process of the FWA-BP-PID controller includes optimizing the initial weights of the BP neural network using the fireworks algorithm. The fitness function is the root mean square error (RMSE), which is expressed as follows: ; Where P is the number of sampling points, r(k) is the reference speed, and y(k) is the actual speed.

[0032] In S2, the chattering suppression method of the RBF-SMC controller includes: using an inverse tangent saturation function Instead of the traditional sign function, α is the saturation function parameter.

[0033] In S2 , the threshold of the working condition identification module can be dynamically adjusted through on-site debugging, with the adjustment step being ±0.5 r / s for the speed threshold and ±1 N·m / s for the torque change rate threshold.

[0034] In S1, the dual network controller module supports online learning function, optimizes the neural network weights through historical operating data, and the optimization cycle is 5-10 minutes.

[0035] In S3, the response time of the electro-hydraulic proportional valve of the actuator module is ≤50ms, the speed adjustment range of the hydraulic motor is 0-10r / s, and the control accuracy is ±0.2r / s.

[0036] Example 1: FWA-BP-PID control under stable conditions Working background: The drill is drilling in a homogeneous soft coal seam (hardness f=2-3) with a medium gas content. There is no significant load change during the drilling process. Set the reference speed , requiring high control accuracy and small overshoot.

[0037] Controller parameter configuration: Fireworks algorithm optimization parameters: explosion radius adjustment constant A=8, explosion spark number adjustment constant M=10, number of mutation sparks=5, number of iterations=50.

[0038] BP neural network structure: 4 nodes in the input layer (speed error e, error change rate Δe, torque T, hydraulic pressure P), 5 nodes in the hidden layer, and 3 nodes in the output layer (PID parameters ).

[0039] PID parameter initial values: (Traditional PID parameters).

[0040] Optimized PID parameters: .

[0041] Control process: 1. Data collection and preprocessing: The speed sensor collects the actual speed in real time , high-frequency noise is removed by Kalman filtering, and the sampling period is T=0.01s. For example, at a certain moment , calculation error , error change rate .

[0042] 2. FWA-BP-PID control logic: The BP neural network receives the input signal [e, Δe, T, P], calculates the tanh activation function of the hidden layer, outputs the PID parameter increments ΔK_p, ΔK_i, ΔK_d, and updates the current control parameters: ; ; .

[0043] The fireworks algorithm starts a global optimization every 10 sampling periods (i.e. 0.1s), updates the BP network weights through the explosion operator and mutation operator, and the fitness function is the root mean square error (RMSE): .

[0044] 3. Actuator adjustment: The controller outputs a voltage signal u to the electro-hydraulic proportional valve and uses an incremental PID algorithm to calculate the control quantity: ;

[0045] The electro-hydraulic proportional valve adjusts the oil flow according to Δu and drives the hydraulic motor to adjust the speed.

[0046] Control effect: Step response: adjustment time 0.25s, overshoot 0%, steady-state error ≤ 0.01r / s.

[0047] Sine tracking: Input signal ω_d=6+0.5sin(2πt)r / s, the maximum amplitude difference between the tracking curve and the reference signal is ≤0.02r / s, and the phase lag is ≤0.05s.

[0048] Energy consumption performance: The average pressure of the hydraulic system is stable at 8MPa, which is 15% lower than the energy consumption of traditional PID control.

[0049] Example 2: RBF-SMC control under complex working conditions Working background: When the drill rig drills into the coal seam gangue area (hardness f=5-6), significant load disturbance (random torque fluctuation) occurs. ), speed setting value , requiring rapid suppression of disturbances and reduction of chattering.

[0050] Controller parameter configuration: Sliding surface parameter: c=1.2, to ensure that the system converges quickly to the sliding surface.

[0051] RBF neural network structure: 2 nodes in the input layer (speed error e, error change rate Δe), 5 nodes in the hidden layer (Gaussian function center vector c=[-2,-1,0,1,2], width b=3.2), 1 node in the output layer (system uncertainty estimation value ).

[0052] Switching control law parameters: saturation function parameter α=5, control law parameter η=0.2, inverse tangent function is .

[0053] Control process: 1. Working condition identification and switching: The torque sensor detects ΔT / t=8N·m / s (exceeding the threshold of 5N·m / s). The working condition identification module determines it as a complex working condition and triggers the RBF-SMC controller to take over control.

[0054] 2. RBF neural network approximation: The neural network receives [e, Δe] and calculates the hidden layer output , the output layer estimates the system uncertainty: The weight w is updated online through the Lyapunov adaptive law: (γ=0.015 is the adaptive gain).

[0055] 3. Calculation of sliding mode control law: Sliding mode function , the control law is: ; Where g is the hydraulic system gain (g = -4βDmKqKv / (VtJm) = -1.5×10^-6).

[0056] Control effect: Anti-disturbance capability: When a disturbance of T_d=10N·m is applied, the maximum speed deviation is 0.8r / s and the recovery time is ≤0.5s.

[0057] Chattering suppression: The control signal fluctuation range is ±0.3V, which is 80% lower than that of traditional sliding mode control (±1.5V).

[0058] Steady-state accuracy: tracking error ≤ 0.05r / s, meeting the accuracy requirements of hard rock drilling.

[0059] Example 3: Working Condition Switching Performance Test Test scenario: The drill cuts into the hard rock layer (f=4) at a constant speed from the soft coal layer (f=2). , the working condition recognition threshold is adjusted to |Δω|≤2r / s (to improve switching sensitivity).

[0060] Testing process: Soft coal seam stage (0-5s): The FWA-BP-PID controller was enabled, and the speed was stabilized at 5.0±0.05r / s and the torque T=50±2N·m.

[0061] The BP neural network updates the PID parameters every cycle, and the fireworks algorithm optimizes the weights to maintain the RMSE below 0.1.

[0062] Hard rock layer cutting time (5s): The torque suddenly increased to T = 80 N·m, ΔT / t = 6 N·m / s (exceeding the threshold of 5 N·m / s), and the speed dropped sharply to 4.2 r / s (Δω = 0.8 r / s < 2 r / s, and the switching was not triggered).

[0063] FWA-BP-PID controller fast adjustment to 12.0, to 2.5, and the speed returns to 5.0 r / s within 0.3 s.

[0064] Hard rock stabilization stage (5-10s): The torque continues to fluctuate (T = 70-90 N·m), ΔT / t = 4 N·m / s, and the working condition identification module maintains FWA-BP-PID control.

[0065] If the disturbance is artificially increased (such as simulating drill bit jamming, ), Δω=1.5r / s, which still does not reach the switching threshold, the system uses RBF neural network to assist compensation (temporarily enabling the RBF observer).

[0066] Comparative data:

[0067] Example 4: Sensor noise suppression experiment Purpose of the experiment: Verify the Kalman filter's effect on sensor noise suppression by simulating the strong electromagnetic interference environment in a coal mine.

[0068] Experimental setup: Noise injection: Gaussian white noise with a mean of 0 and a variance of σ² = 0.1 is superimposed on the speed sensor signal. The original signal is , the noisy signal is .

[0069] Filtering algorithm: Equation of state: (a=-0.1 is the system damping coefficient, w is the process noise, and the variance Q=0.01).

[0070] Measurement equation: (v is the measurement noise, with variance R=0.1).

[0071] Kalman gain: , where H=1 is the measurement matrix.

[0072] Experimental process: Noisy signal characteristics: When not filtered, the speed curve fluctuates in the range of 4.5-5.5r / s, and the root mean square error RMSE = 0.35r / s.

[0073] Signal characteristics after filtering: After 50 recursive calculations, the noise variance dropped to σ² = 0.01, the curve fluctuation range was 4.9-5.1 r / s, and the RMSE was 0.08 r / s.

[0074] Impact on control: Traditional PID control (unfiltered): Parameter adjustment was falsely triggered by noise, resulting in a speed overshoot of 8% and an adjustment time of 1.5 seconds.

[0075] The system of the present invention (after filtering): the FWA-BP-PID controller operates stably, with an overshoot of 0% and a regulation time of 0.25s.

[0076] Example 5: Online learning function test Test scenario: The drilling rig operated continuously for 12 hours, and the coal seam hardness gradually changed from f=2 (soft coal) to f=5 (hard coal), verifying the system's ability to optimize control parameters through historical data.

[0077] Learning mechanism: Data storage: save a set of working condition data every 10 minutes ( ), forming a training data set.

[0078] Weight update: FWA-BP-PID: The Fireworks algorithm initiates a global optimization search every hour, updates the BP network weights based on the latest 500 sets of data, and adds an energy consumption weight (RMSE × 0.8 + energy consumption × 0.2) to the fitness function.

[0079] RBF-SMC: RBF network weights are updated in batches every 20 minutes using the least squares method to reduce the amount of online computation.

[0080] Performance change trends:

[0081] As the coal seam hardness increases, FWA-BP-PID automatically increases to 15.0 to speed up the response; the α parameter of RBF-SMC is adaptively adjusted to 4.0 to balance the vibration and anti-interference capabilities.

[0082] Drilling efficiency increased by 67%, attributed to parameter optimization that reduced stuck-pipe downtime (from an initial 3 times per hour to 0.5 times per hour).

[0083] Example 6: Engineering Field Application Application scenarios: The 3# coal seam gas extraction project in a coal mine is a high-risk operation area with an average coal seam hardness of f=3.5, three interbedded gangue zones (f=6-7), and a gas emission rate of 15m³ / min.

[0084] System deployment details: Hardware configuration: Data acquisition module: Use explosion-proof speed sensor (Ex d IIC T6 Gb), protection grade IP68, and support -40℃~85℃ working environment.

[0085] Dual network controller: integrated in an explosion-proof control cabinet (explosion-proof mark Ex de IIC T6 Gb), adopts DSP+FPGA architecture, and has a computing speed of up to 200MFLOPS.

[0086] Actuator: Electro-hydraulic proportional valve with built-in pressure compensator, response time ≤30ms, hydraulic motor rated torque 2000N·m.

[0087] Software features: Monitoring module: Developed based on WinCC Flexible, it displays the 3D drilling trajectory, gas concentration warning (threshold ≥ 1%), and hydraulic oil temperature alarm (threshold ≥ 60°C) in real time.

[0088] Remote debugging: Connecting to the ground control center via a 4G wireless module allows engineers to modify controller parameters (such as operating condition thresholds and learning cycles) online.

[0089] Application performance data: Drilling efficiency: The average construction time for a single hole is shortened from 4 hours using traditional technology to 2.5 hours, an increase of 60%.

[0090] Safety: The number of gas over-limit alarms has been reduced from 12 to 2 per day, and there have been no drill jamming accidents.

[0091] Reliability: The system runs continuously for 30 days without any failure, and the mean time between failures (MTBF) is ≥5000 hours.

[0092] Energy consumption: Fuel consumption per unit footage is reduced from 8L / m to 5L / m, meeting the requirements of green mining.

[0093] Typical working condition records: When crossing the gangue band, the RBF-SMC controller automatically activated, suppressing speed fluctuations within 1.2 seconds (from 5 rpm to 4.2 rpm before recovery). The hydraulic system pressure was controlled at 12 MPa ± 0.5 MPa, eliminating the risk of drill pipe breakage. Ground monitoring screens displayed the disturbance waveform and controller adjustment process in real time, providing a basis for on-site decision-making.

[0094] Summary of Examples

[0095] Through six sets of experiments, it was verified that the system of the present invention is superior to traditional control schemes in terms of control accuracy, anti-interference ability, adaptability to working conditions, engineering reliability, etc., especially showing significant advantages under complex coal seam conditions, providing an effective solution for the intelligent and safe operation of gas extraction drilling rigs.

[0096] The following is a comparative example in contrast to the present invention, designed based on a traditional single control algorithm (such as PID control and traditional sliding mode control), used to compare and verify the technical advantages of the present invention: Comparative Example 1: Traditional PID Control System System Architecture: The classic PID controller is used, and the parameters are adjusted by the Ziegler-Nichols method and fixed as 、 、 .

[0097] It only relies on speed feedback and has no working condition recognition and self-adaptation capabilities.

[0098] Control process: Stable working condition test (soft coal seam, f=2-3, ): The step response overshoot is 24.8%, the adjustment time is 0.45s, and the steady-state error is 0.

[0099] During sinusoidal tracking, initial oscillation is obvious, with a tracking error of ±0.06r / s and a lag of 0.2s.

[0100] Complex working condition test (interlayer, f=5-6, ): Under load disturbance, the speed fluctuates by ±1.5r / s, and the recovery time is 1.2s, which can easily cause drill sticking.

[0101] Without vibration suppression measures, the hydraulic system pressure fluctuation reaches ±2MPa, and the equipment wear is significant.

[0102] Defect analysis: Fixed parameters: Unable to adapt to changes in coal seam hardness, large overshoot under soft coal conditions, and delayed response under hard rock conditions.

[0103] Insufficient anti-interference ability: Slow adjustment in the face of random disturbances, relying on manual intervention to adjust parameters.

[0104] Comparative Example 2: Traditional Sliding Mode Control System (SMC) System Architecture: Using traditional sliding mode controller, the sliding surface (c=0.9), switching control law u=-ηsgn(s) (η=0.5).

[0105] There is no neural network to approximate the uncertainty term, and the symbolic function is used directly, resulting in significant chattering.

[0106] Control process: Stable working condition test (soft coal seam, ): There is no overshoot in the step response, but there is high-frequency chattering in the sliding stage (error ±0.8r / s), and the hydraulic system is noisy.

[0107] Complex working condition test (hard rock formation, f=5-6, ): The anti-disturbance recovery time is 0.6s, but the control signal switches violently (±1.5V), resulting in severe wear of the actuator (the life of the electro-hydraulic proportional valve is shortened by 30%).

[0108] Without working condition identification, chattering is aggravated under soft coal conditions, affecting drilling accuracy.

[0109] Defect analysis: Severe chattering: The sign function causes discontinuous control signals, which can easily cause mechanical oscillations in practical applications.

[0110] Strong model dependence: Uncompensated system uncertainties (such as hydraulic oil leakage and coal seam friction changes) lead to a significant decrease in control accuracy as operating conditions worsen.

[0111] Comparative Example 3: Single BP Neural Network PID Control System (BP-PID) System Architecture: BP neural network is used to adjust PID parameters with a network structure of 4-5-3, a learning rate of 0.5, and an inertia factor of 0.05.

[0112] There is no external optimization algorithm, the initial weights are randomly generated, and it is easy to fall into local optimality.

[0113] Control process: Stable working condition test (soft coal seam, ): The step response overshoot is 0.11% and the adjustment time is 0.25s, but the convergence speed is slow (200 iterations are required).

[0114] The sine tracking has a lag of 0.15s and an amplitude error of ±0.04r / s.

[0115] Complex working condition test (gangue layer, f=5-6): Parameter adjustment lags during load disturbance, the maximum speed deviation is 1.2r / s, and the recovery time is 0.8s.

[0116] The neural network falls into a local optimum and cannot adapt to sudden changes in working conditions (such as a sudden increase of 50% in hardness).

[0117] Defect analysis: Insufficient real-time performance: The weight iteration calculation is large, and the control delay is significant under complex working conditions.

[0118] Poor robustness: A single control strategy cannot cope with the differences in multiple working conditions. The error under hard rock conditions is 50% higher than that of the system of the present invention.

[0119] Comparative Example 4: Unoptimized RBF Neural Network Sliding Mode Control System (RBF-SMC) System Architecture: The RBF neural network is used to approximate the uncertain terms, but the adaptive saturation function is not introduced and the traditional symbolic function is still used.

[0120] The working condition recognition threshold is fixed and cannot be adjusted dynamically.

[0121] Control process: Complex working condition test (hard rock formation, f=5-6, ): The jitter amplitude is ±1.0V, which is 33% lower than that of the traditional sliding mode control, but still higher than that of the system of the present invention (±0.3V).

[0122] The neural network approximation error is 0.5 r / s, and the steady-state accuracy is lower than that of the present invention (0.05 r / s).

[0123] Working condition switching test (soft coal → hard rock): The switching delay is 0.5s, and the speed overshoots by 5%. Since the threshold is fixed, it is impossible to predict the working condition changes in advance.

[0124] Defect analysis: Chattering suppression is not complete: the switching control law is not improved, and the sign function still causes high-frequency oscillations.

[0125] Limited adaptability to working conditions: fixed thresholds cannot match dynamically changing coal seam characteristics, and switching logic is rigid.

[0126] Comparative Example 5: Traditional Control System without Noise Suppression System Architecture: PID control is adopted, and the sensor signal is not filtered and is directly input into the controller.

[0127] No anti-interference design, relying on hardware redundancy.

[0128] Test scenario: Gaussian noise (σ²=0.1) is artificially injected into the speed sensor to simulate downhole electromagnetic interference.

[0129] Control effect: The speed curve fluctuated by ±0.8r / s, which the controller mistakenly identified as a load disturbance. Frequent parameter adjustments led to system oscillation.

[0130] The mean time between failures (MTBF) dropped from 5,000 hours to 2,000 hours, and the sensor false alarm rate increased fourfold.

[0131] Defect analysis: Poor reliability: Noise causes control parameter oscillations, which can lead to drilling rig shutdown in severe cases.

[0132] High maintenance cost: Frequent sensor calibration is required, increasing on-site workload.

[0133] Comparative Example 6: Fixed Parameter Control System without Online Learning System Architecture: The FWA-BP-PID controller is used, but the online learning function is turned off and the parameters are optimized only once in the initial stage.

[0134] Long-running tests: After 8 hours of continuous drilling, the coal seam hardness increased from f=2 to f=4, and the speed tracking error increased from 0.01r / s to 0.2r / s.

[0135] The controller was unable to adapt to the time-varying parameters, resulting in a 25% increase in hydraulic system energy consumption and a 12% decrease in drilling efficiency.

[0136] Defect analysis: Insufficient parameter timeliness: Unable to cope with the gradual changes in coal seams during long-term operations, and control performance deteriorates over time.

[0137] Lack of adaptability: Relying on initial optimization results, it is unable to cope with unknown working conditions (such as groundwater intrusion causing coal seams to become soft).

[0138] Comparison of key performances of the present invention and the control example:

[0139] The control examples all have problems such as insufficient adaptability of single control strategies, imbalance between algorithm real-time performance and complexity, and engineering design defects. The present invention significantly improves the control performance of gas extraction drilling rigs in multi-dimensional scenarios through innovations such as dual network fusion, intelligent working condition switching, noise suppression, and online learning.

[0140] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above-described embodiments. The above-described embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The engineered gas extraction drilling rig neural network adaptive control system is characterized by: include: The data acquisition module is used to collect the speed, torque, hydraulic pressure and coal seam hardness working condition data of the drilling rig rotary system in real time and perform pre-processing; A dual-network controller module includes an FWA-BP-PID controller, an RBF-SMC controller, and a working condition identification module. The working condition identification module switches the control strategy according to preset working condition thresholds, including a speed fluctuation threshold of ±5 rpm and a torque change rate threshold of ±8 N·m / s. The actuator module includes an electro-hydraulic proportional valve and a hydraulic motor, which is used to receive the voltage signal output by the controller and adjust the drilling rig speed; The monitoring module is used to display system parameters, control curves and fault warning information in real time.

2. The neural network adaptive control system for an engineered gas extraction drill according to claim 1, characterized in that: The fireworks algorithm optimization parameters of the FWA-BP-PID controller include: explosion radius adjustment constant A=6-10, explosion spark number adjustment constant M=8-12, mutation spark number=3-7, and iteration number=30-60.

3. The neural network adaptive control system and method for an engineered gas extraction drilling rig according to claim 1 is characterized by: The sliding surface parameter c of the RBF-SMC controller is 0.5-1.2, the saturation function parameter α is 2-5, and the control law parameter η is 0.05-0.

2.

4. The neural network adaptive control system and method for an engineered gas extraction drilling rig according to claim 1 is characterized by: The data acquisition module uses high-precision sensors, with a rotation speed sensor accuracy of ±0.1r / s, a torque sensor accuracy of ±1N·m, and a sampling period of 0.005-0.02s.

5. An adaptive control method for an engineered gas extraction drilling rig neural network adaptive control system according to any one of claims 1 to 4, characterized in that: The following steps are involved: S1: The data acquisition module collects working condition data in real time and inputs it into the dual network controller module after filtering and noise reduction; S2: The working condition identification module determines the working condition based on the speed fluctuation Δω and the torque change rate ΔT / t. When |Δω|≤3r / s and |ΔT / t|≤5N·m / s, it is determined to be a stable working condition and the FWA-BP-PID controller is activated. When |Δω|>3r / s or |ΔT / t|>5N·m / s, it is determined to be a complex working condition and the RBF-SMC controller is activated. S3: The controller outputs a control signal to the actuator module to adjust the opening of the electro-hydraulic proportional valve to achieve speed tracking; S4: The monitoring module displays the control effect in real time and automatically adjusts the controller parameters if overshoot or oscillation occurs.

6. The neural network adaptive control method for an engineered gas extraction drill according to claim 5 is characterized in that: The parameter tuning process of the FWA-BP-PID controller includes optimizing the initial weights of the BP neural network using the fireworks algorithm. The fitness function is the root mean square error (RMSE), which is expressed as follows: ; Where P is the number of sampling points, r(k) is the reference speed, and y(k) is the actual speed.

7. The neural network adaptive control method for an engineered gas extraction drill according to claim 5, characterized in that: The chattering suppression method of the RBF-SMC controller includes: using an inverse tangent saturation function Instead of the traditional sign function, α is the saturation function parameter.

8. The neural network adaptive control method for an engineered gas extraction drill according to claim 5, characterized in that: The threshold of the working condition identification module can be dynamically adjusted through on-site debugging, with the adjustment step being ±0.5 r / s for the speed threshold and ±1 N·m / s for the torque change rate threshold.

9. The neural network adaptive control method for an engineered gas extraction drill according to claim 5, characterized in that: The dual network controller module supports online learning function, optimizes the neural network weights through historical operating data, and the optimization cycle is 5-10 minutes.

10. The neural network adaptive control method for an engineered gas extraction drill according to claim 5, characterized in that: The response time of the electro-hydraulic proportional valve of the actuator module is ≤50ms, the speed adjustment range of the hydraulic motor is 0-10r / s, and the control accuracy is ±0.2r / s.

Images