Scraper conveyer speed regulation control method fusing perception and self-evolution strategy

Abstract

The invention discloses a scraper conveyor speed regulation control method fusing perception and a self-evolution strategy. The scraper conveyor speed regulation control method comprises the following steps: S1, data acquisition; s2, constructing a multi-dimensional collaborative state vector for the collected data; s3, action decision generation: inputting a reinforcement learning strategy network, and outputting a chain speed regulation instruction; s4, verifying the legality of the boundary; s5, performing speed regulation execution and feedback acquisition: sending the speed regulation instruction subjected to boundary verification to a variable frequency driving system through an edge controller; a lag compensation mechanism is introduced to predict and correct delay between strategy action and feedback; s6, strategy optimization is carried out on the reinforcement learning strategy network every fixed step number; and S7, activating a safety protection mechanism. The states of multiple devices such as the coal machine, the support and the scraper conveyor are fused, chain speed control is achieved, system-level closed-loop cooperative speed regulation logic is formed, it is ensured that chain stress and motor loads run within the controllable range, and therefore the impact and damage risks are reduced.

Description

Technical Field

[0001] This invention relates to a speed control method for scraper conveyors that integrates sensing and self-evolution strategies. Background Technology

[0002] With the rapid development of intelligent coal mining faces, scraper conveyors, as core equipment for coal transport, are becoming increasingly important for the stability and safety of the entire production system due to their operational efficiency and control precision. Currently, the common operating modes of scraper conveyors mainly include the following:

[0003] (1) Fixed chain speed operation;

[0004] (2) Variable frequency drive operation in different speed ranges;

[0005] (3) Simple feedback speed regulation based on coal level sensor or motor current.

[0006] These control methods generally have the following shortcomings:

[0007] (1) It relies on a single sensor and lacks system-level state awareness;

[0008] (2) The control strategy relies on preset rules or static PID parameters and cannot adapt to changes in operating conditions;

[0009] (3) Lack of tension and load boundary control during speed regulation can easily lead to chain impact, overload and chain breakage;

[0010] (4) The inability to achieve automatic strategy optimization limits the operating efficiency and energy consumption control capabilities.

[0011] Therefore, there is an urgent need for a scraper conveyor speed control system with system collaborative perception, operational safety protection, and intelligent adaptive optimization capabilities. Summary of the Invention

[0012] The purpose of this invention is to overcome the shortcomings of the prior art and provide a scraper conveyor speed control method that integrates sensing and self-evolution strategies. This method integrates the states of multiple devices such as coal mining equipment, supports, and scraper conveyors to achieve chain speed control and ensure that the chain force and motor load operate within a controllable range.

[0013] The objective of this invention is achieved through the following technical solution: a speed control method for scraper conveyors that integrates sensing and self-evolution strategies, comprising the following steps:

[0014] S1. Data Acquisition: The control system connects to the industrial control interfaces of each device in the fully mechanized mining face through an edge gateway, and communicates with each device using the Modbus TCP / UDP protocol to acquire data.

[0015] S2. Construct a multi-dimensional collaborative state vector from the collected data. : = [Support pushing rhythm, coal mining machine advance speed, motor current, real-time chain speed, chain tension, coal flow rate, push rod stroke change rate, automatic coal cutting status, scraper conveyor motor temperature];

[0016] S3, Action Decision Generation: This involves generating the state vector... Input the reinforcement learning policy network to obtain the state-action mapping function, and output the chain speed adjustment instruction. As an action;

[0017] S4. Boundary Validity Verification: Set three safety boundaries: upper limit for tension change rate, safe power range for motor, and upper limit for chain speed change rate; output actions in the strategy network. Subsequently, if the rate of change of tension exceeds the upper limit of the rate of change of tension, or the motor power exceeds the safe power range of the motor, or the rate of change of chain speed exceeds the upper limit of the rate of change of chain speed, the chain speed adjustment command is scaled proportionally to meet the safety boundary; when multiple constraints are violated at the same time, the strictest principle is adopted to find the safe command that meets all safety boundaries and is closest to the chain speed adjustment command.

[0018] S5. Speed ​​regulation execution and feedback acquisition: The speed regulation command, after boundary verification, is sent to the frequency converter drive system through the edge controller to control the scraper chain to achieve dynamic adjustment of chain speed;

[0019] A lag compensation mechanism is introduced to predict and correct the delay between policy action and feedback. The specific implementation steps are as follows:

[0020] S51. The system maintains a fixed-length action-state queue in memory, storing historical states in chronological order. and actions already performed ;

[0021] S52. Based on the total delay of the control loop, set a fixed delay step d;

[0022] S53. When the system feedback status is collected at time step t. At that time, this state is a historical action. The result after execution;

[0023] S54. The system retrieves the state d steps ago from the queue. and actions Compare it with the current state and calculated rewards Combined, they form a complete delayed transfer sample. , , , ), and store it in the experience replay buffer;

[0024] S55. The reinforcement learning policy network is trained using delayed transition samples;

[0025] S6. Policy Adaptive Optimization: The reinforcement learning policy network is optimized once every fixed number of steps using stochastic gradient descent.

[0026] S7. Security Protection Mechanism Activation: Set a multi-level security protection policy independent of the reinforcement learning policy network. This policy has the highest execution priority. Once the following conditions are met, it will unconditionally overwrite the current chain speed instruction and trigger the corresponding protection:

[0027] Over-tension speed reduction protection: If the instantaneous value of the chain tension force collected in real time is greater than the rated upper limit, the speed reduction protection will be triggered immediately, reducing the chain speed to a fixed proportion of the current speed or a preset safe speed.

[0028] Over-power shutdown protection: If the instantaneous value of the motor power collected in real time exceeds 110% of the rated power, the emergency shutdown protection will be triggered immediately to cut off the motor power.

[0029] Sudden change in rhythm and chain deceleration: If the stent pushing rhythm calculated based on a preset very short time window exceeds the safety threshold, it is determined that the pushing rhythm has changed suddenly, and a chain deceleration is immediately triggered, forcibly setting the chain speed to a preset low-speed crawling state.

[0030] After the aforementioned protection mechanism is activated, the system will temporarily disconnect from the speed adjustment mode of the reinforcement learning policy network and enter a safety protection mode; once the risk indicators return to normal and are confirmed by the operator, it will switch back to the speed adjustment mode of the reinforcement learning policy network. The beneficial effects of this invention are:

[0031] 1. Strong intelligent collaborative sensing capability: It integrates the status of multiple equipment such as coal mining machines, supports, and scraper conveyors to achieve chain speed control and form a system-level closed-loop collaborative speed regulation logic;

[0032] 2. The strategy has self-evolution capability: speed control does not rely on fixed rules, but achieves multi-objective performance optimization (capacity, energy consumption) through deep reinforcement learning.

[0033] 3. Safety: A dynamic constraint model is constructed to ensure that the chain force and motor load operate within a controllable range, thereby reducing the risk of impact and damage;

[0034] 4. High adaptability: The system has continuous learning capabilities and can adapt to complex working conditions such as different coal seam thicknesses, advance rhythms, and coal flow fluctuations;

[0035] 5. Reduced energy consumption: By dynamically matching load prediction and chain speed, the energy consumption of the motor during redundant operation under low load is effectively reduced, achieving energy saving and consumption reduction.

[0036] 6. Compatibility: The control system adopts a modular and standardized interface design, which makes it compatible with different models of frequency converters and PLC systems, and has good portability and integration convenience. Attached Figure Description

[0037] Figure 1 This is a flowchart of the scraper conveyor speed control method of the present invention. Detailed Implementation

[0038] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0039] like Figure 1 As shown, the present invention provides a speed control method for a scraper conveyor that integrates sensing and self-evolution strategies, comprising the following steps:

[0040] S1. Data Acquisition: The control system accesses the industrial control interface of each piece of equipment in the fully mechanized mining face through the edge gateway. In this embodiment, the system mainly collects data from three key pieces of equipment: the coal mining machine, the support, and the scraper conveyor. The control system uses the Modbus TCP / UDP protocol to communicate with each piece of equipment and performs high-frequency (1Hz) data acquisition.

[0041] To improve data stability, the following data preprocessing mechanism is designed for the collected data in this embodiment: if the sensor fault flag is 1, then the item is set to missing; peaks are removed using the 3σ criterion or the sliding window median method; for intermittent sampling, forward padding or 0 value replacement is used to ensure the integrity of the state vector.

[0042] S2. Construct a multi-dimensional collaborative state vector from the collected data. This system incorporates multi-source operational data from three types of equipment: coal mining machines, supports, and scraper conveyors, to construct a multi-dimensional collaborative state vector. The structure is as follows: = [Support pushing rhythm, coal mining machine advance speed, motor current, real-time chain speed, chain tension, coal flow rate, push rod stroke change rate, automatic coal cutting status, scraper conveyor motor temperature].

[0043] The construction of the multidimensional collaborative state vector is based on sliding window aggregation and feature extraction of various raw data streams. The system processes the collected data using a fixed-duration sliding window, calculating corresponding statistical features for the raw data within each window, thereby transforming high-frequency time-series data into stable parameters characterizing the short-term operating state of the system. The specific methods for obtaining each state parameter are as follows:

[0044] (1) Stent pushing rhythm: Estimated using the sliding window frequency statistics method. A fixed-length sliding window (e.g., 30 seconds) is set, and the number of times the "push-slip flag = 1" of a certain stent or the number of times the push rod action state changes within each sliding window is counted. The number of times is divided by the width of the time window to obtain the action frequency per unit time, which is the stent pushing rhythm. The pushing rhythm can be further divided into "high / medium / low rhythm" state labels for input state encoding of the speed regulation model. If there is a pushing distance field, the average pushing speed can also be derived as an auxiliary feature.

[0045] (2) Coal mining machine advance speed and motor current: Real-time traction speed signal of the coal mining machine and three-phase current value of the motor are collected from the coal mining machine PLC or monitoring system through industrial protocol interfaces such as OPC UA or Modbus TCP. The real-time traction speed signal of the coal mining machine is processed through a sliding window, and the average value of the instantaneous speed in each sliding window is calculated as the coal mining machine advance speed;

[0046] The three-phase current values ​​of the motor are processed by a sliding window, and the average current value within each sliding window is calculated as a parameter reflecting the average load. The maximum current value within each sliding window is then used as a parameter for overload detection.

[0047] (3) Real-time chain speed; First, the instantaneous chain speed v is calculated using the following formula:

[0048] v = π * D * P * n / L

[0049] Where D is the chain diameter, P is the chain pitch, n is the current motor speed, and L is the chain length; this formula dynamically estimates the chain linear velocity (m / s) based on the geometric relationship of the chain drive. After calculating the instantaneous chain speed, it is processed through a sliding window (e.g., 10 seconds), and the average value of the instantaneous chain speed within each sliding window is calculated as the real-time chain speed; calculating the average value can eliminate fluctuations, and the obtained stable value is used as the real-time chain speed in the state vector.

[0050] (4) Chain tension: This is obtained by installing a pressure sensor in the piston chamber of the hydraulic cylinder of the tensioning device at the tail of the scraper conveyor (i.e., the side that drives the sprocket). The sensor's standard 4-20mA or 0-10V analog signal is connected to the PLC or edge computing module via an IO module or directly via an RS-485 interface, and transmitted to the control system using the Modbus RTU / TCP protocol. The collected raw pressure data is processed through a sliding window, and the average pressure within each sliding window is calculated as the chain tension in the state vector.

[0051] (5) Coal Flow Rate (Dynamic Conveying Section Weighing Method): The coal flow rate during scraper conveying is estimated by combining chain speed with the weighing section. The method structure is as follows:

[0052] Structural components:

[0053] Weighing middle section: Select 2 to 3 sections of the trough in the middle of the scraper conveyor and install high-precision weighing sensors (such as strain gauge sensors).

[0054] Data acquisition and synchronization module: Weighing sensor data is input to PLC or edge device, and chain speed is obtained in real time in conjunction with frequency converter or speed sensor;

[0055] Fusion processing module: The edge-side or central processing unit calculates the coal flow rate in real time. The coal flow rate estimation formula is: Q = G * v / L', where Q is the coal flow rate (kg / s or t / h), G is the weight of the weighing section, v is the instantaneous chain speed, and L' is the length of the weighing section.

[0056] (6) Push rod stroke change rate: Real-time stroke data of the push rod is obtained by a displacement sensor (such as a rope displacement sensor or a magnetostrictive displacement sensor) installed on the hydraulic support push cylinder.

[0057] Calculation method: For the collected raw stroke data S, set a sliding window (e.g., 10 seconds), and calculate the linear fitting slope of the stroke value within the sliding window (unit: mm / s). This slope is the "push rod stroke change rate", which directly represents the speed of the push action. The calculation formula can be simplified to ΔS / Δt, where ΔS is the stroke change within the window, and Δt is the window time length.

[0058] (7) Automatic coal cutting status: Read directly from the coal mining machine PLC or the fully mechanized mining face automation control system via industrial communication protocols (such as Modbus TCP, OPC UA, PROFINET). Access a specific status register or data block inside the coal mining machine control system. This register usually has a specific bit or integer value (e.g., 0 represents manual mode, 1 represents automatic coal cutting mode, 2 represents maintenance mode, etc.) to indicate the current working mode. The control system polls or subscribes to this status value periodically through the communication interface.

[0059] (8) Scraper conveyor motor temperature: Data from temperature sensors embedded in the three-phase windings of the motor are obtained via fieldbus (Modbus / CAN, etc.). These sensors are usually built into large and medium-sized variable frequency motors.

[0060] S3, Action Decision Generation: This involves generating the state vector... Input the reinforcement learning policy network to obtain the state-action mapping function. Output chain speed adjustment command As an action;

[0061] The reinforcement learning policy network employs the Deep Deterministic Policy Gradient (DDPG) algorithm framework, and its technical principle is as follows:

[0062] Actor Network (μ(s|θ^μ)): This serves as the policy network, with its input being an 8-dimensional multidimensional cooperative state vector. The output is a continuous action. (i.e., chain speed adjustment). The network structure is a fully connected neural network with two hidden layers containing 256 and 128 neurons, respectively, using the ReLU activation function. The output layer uses the Tanh activation function to constrain the action within the range of [-1, 1], and then linearly maps it to the actual chain speed adjustment range (e.g., ±0.2 m / s).

[0063] Critic network (Q(s, a|θ^Q)): As a value network, its input is the state. and actions The output is the predicted cumulative reward Q for that state-action pair. The network structure is also a fully connected neural network, with 256 and 128 neurons in the two hidden layers, respectively, using the ReLU activation function.

[0064] Training mechanism: The system maintains an Experience Replay Buffer to store transfer samples. , , , ),in, To enhance the reward function of the learning policy network, Let be the state vector for the next time step. During training, a mini-batch of samples is randomly sampled, and the Critic network parameters are updated by minimizing the temporal difference error of the Critic network; the Actor network parameters are updated using a gradient ascent strategy. A soft update method is used to synchronize the target network parameters, with the update coefficient τ set to 0.01 to improve training stability.

[0065] During training, the model constructs a composite reward function with coal flow conveying efficiency, motor load fluctuations, and tension changes as objectives, and progressively optimizes the policy network weights. The output speed regulation actions are converted into inverter frequency setpoints and sent to the scraper conveyor to achieve chain speed control.

[0066] Model input: state vector ;

[0067] Model output: Target chain speed / frequency conversion ;

[0068] The chain speed adjustment command output by the reinforcement learning policy network is essentially a target chain speed value. This value is converted into a frequency converter setpoint using a fixed conversion factor k before being sent to the variable frequency drive system. The conversion factor k is determined by the mechanical transmission parameters of the scraper conveyor, satisfying the formula v = k * f, where v is the chain speed (m / s) and f is the frequency converter output frequency (Hz). Therefore, the "target chain speed" and "frequency converter frequency" commands mentioned in this paper are equivalent in control logic.

[0069] S4. Boundary Validation: Output actions in the policy network. Then, its validity needs to be verified by the dynamic boundary modeling module. This module predicts and executes chain actions based on the current system state. The resulting solution establishes the following three core security boundaries:

[0070] Tension change limit: To prevent chain breakage due to impact, an upper limit ΔF is set for the rate of change of tension. max = 2.0MPa / s. This value is determined based on the maximum statistical value of hydraulic system safety regulations and historical normal operation data.

[0071] Motor safe power range: To protect the motor, the motor safe power range is set to [0, 110% * P_rated] based on the specific operating parameters of the scraper conveyor, where P_rated is the rated power of the motor.

[0072] Chain speed change rate limit: Action The required chain speed change amplitude and rate must not exceed the threshold that the system's mechanical inertia can withstand. The upper limit of the chain speed change rate V_rate_max needs to be calculated and determined in combination with the maximum torque of the drive motor and the total rotational inertia of the system, and finally verified during on-site debugging. In this embodiment, it is set to 0.15 m / s².

[0073] Output chain speed adjustment instructions in the policy network. Subsequently, the system employs an instruction pruning mechanism, using a reverse solving mechanism to automatically adjust the instructions to ensure that the adjusted instructions meet the safety boundaries. Specifically, this includes:

[0074] (1) Constraint violation detection: Output chain speed adjustment instructions in the policy network. Afterwards, the system detects the rate of change of tension, motor power, and chain speed after executing the command.

[0075] (2) Command adjustment algorithm: If the rate of change of tension exceeds the upper limit of the rate of change of tension ΔF maxIf the motor power exceeds the motor's safe power range, or the chain speed change rate exceeds the upper limit of the chain speed change rate V_rate_max, then the chain speed adjustment command will be scaled proportionally to meet the safety boundary.

[0076] (3) Multi-constraint coordination: When multiple constraints are violated at the same time, the strictest principle is adopted to ensure that the adjusted instructions meet all safety requirements and to find the safe instructions that meet all safety boundaries and are closest to the chain speed adjustment instructions.

[0077] (4) Smooth transition: Smooth the instructions before and after the adjustment to avoid sudden changes in control instructions and maintain system stability.

[0078] This mechanism ensures that AI decision-making, while fully leveraging optimized performance, always operates within preset safety boundaries.

[0079] S5. Speed ​​Regulation Execution and Feedback Acquisition: The speed regulation command, after boundary verification, is sent to the frequency converter drive system through the edge controller to control the scraper chain to achieve dynamic adjustment of chain speed; and the following feedback information is recorded in real time: deviation between actual chain speed and target chain speed, chain tension, current, and voltage changes;

[0080] Introducing a hysteresis compensation mechanism to predict and correct the delay between policy action and feedback: An action-state queue is introduced into the control logic, with a set delay step size. ; to the current action and Perform binding training, which represents "delayed response";

[0081] The specific implementation steps of the aforementioned delayed compensation mechanism are as follows:

[0082] S51. The system maintains a fixed-length action-state queue in memory, storing historical states in chronological order. and executed chain speed adjustment commands ;

[0083] S52. Based on the total delay of the control loop (including communication, actuator response, and physical system inertia), set a fixed delay step d (e.g., d = 5, corresponding to a delay of approximately 5 seconds).

[0084] S53. When the system feedback status is collected at time step t. At that time, this state is a historical action. The result after execution;

[0085] S54. The system retrieves the state d steps ago from the queue. and actions Compare it with the current state and calculated rewards Combined, they form a complete delayed transfer sample. , , , ), and store it in the experience replay buffer;

[0086] To enhance the system's multi-objective control capability, the composite reward function of the reinforcement learning policy network in this invention is defined as follows:

[0087]

[0088] In the above functions, Q, P, ΔF, The new state data is obtained by adjusting the chain speed based on the target chain speed / variable frequency output from the reinforcement learning policy network. , , , All are weighting coefficients; this function design, while considering transportation efficiency, strengthens the constraint optimization of safety boundaries and equipment rhythm synchronization, which is significantly better than the traditional strategy model that only targets prediction error.

[0089] in, The coal flow rate per unit time (in tons per minute) is calculated as follows:

[0090] Q = G * v / L'

[0091] Where G is the weight of the weighing section, v is the instantaneous chain speed, and L' is the length of the weighing section;

[0092] To determine the severity of motor current overload, the calculation method is as follows: Motor current is sampled at a fixed frequency (e.g., 1Hz). If any motor current exceeds 110% of the rated current, then that sampling point is considered overloaded. A sliding time window is set to perform a weighted evaluation based on both the frequency and duration of the overload.

[0093]

[0094] Where N is the total number of overload events occurring within the statistical window, and T is the total time spent in an overload state within the cumulative window. and These are weighting coefficients used to adjust the penalty intensity for overload frequency and duration, respectively. The specific initial values ​​of the weighting coefficients (e.g., ...) = 0.1, = 0.15) is not fixed and can be adjusted within a certain range of experience after system deployment, based on the motor characteristics and safety requirements of the specific working surface (e.g., / The optimal value can be manually fine-tuned within [1, 10] or automatically found through hyperparameter optimization algorithms to achieve the best balance between safety and efficiency.

[0095] The chain tension change rate (in MPa / s) is calculated as follows:

[0096]

[0097] , Let be the chain tension at time t and time t-1, respectively. The time difference between two moments;

[0098] The penalty for mismatch in the operating rhythm among the support frame, coal mining machine, and scraper conveyor is calculated as follows:

[0099] (1) Extracting core rhythm parameters: Obtaining the following parameters in each control cycle:

[0100] Coal mining machine rhythm (i.e., its traction speed, m / min);

[0101] support movement rhythm (times / min);

[0102] Equivalent rhythm of scraper conveyor (i.e., chain speed, m / min);

[0103] (2) Normalization: To avoid dimensional differences, firstly normalize the units of measurement. , , The three rhythm parameters are normalized to a similar numerical range. For example, min-max normalization is performed using the maximum and minimum values ​​from recent history (e.g., the past 10 minutes), calculated as follows:

[0104]

[0105] , These are the results after normalization and before normalization, respectively. , or )data, , These are the maximum and minimum values ​​in the data, respectively.

[0106] (3) Calculate the mismatch degree:

[0107]

[0108] , , These are the normalized coal mining machine rhythm, support pushing rhythm, and scraper conveyor equivalent rhythm, respectively.

[0109] (4) Generate penalty terms:

[0110] .

[0111] Wherein, k is a preset mismatch tolerance threshold. The value of k can be slightly adjusted according to actual operation feedback to achieve the best balance between production smoothness and equipment protection. In this embodiment, k=0.2.

[0112] Initial weights are set to =1、 =3、 =2、 =1. The system can adaptively adjust the above weights based on operational feedback during long-term training.

[0113] S55. The reinforcement learning policy network is trained using delayed transfer samples to learn the mapping relationship between actions and delayed consequences, thereby achieving predictive compensation for system delays.

[0114] S6. Policy Adaptive Optimization: A "train-as-you-go" policy update approach is adopted, optimizing the reinforcement learning policy network once every fixed number of steps (e.g., 5000 iterations); stochastic gradient descent is used for policy optimization. Specifically, this involves the update process of two networks:

[0115] Critic network update: By minimizing the mean squared error loss between the predicted value Q and the target value Q, the gradient of this loss with respect to the Critic network parameters is calculated, and the Critic network is updated using the backpropagation algorithm to make its predictions more accurate.

[0116] Actor network update: Using the chain rule, the gradient of the Critic network output Q with respect to the Actor network output action, and the gradient of the action with respect to the Actor network parameters, are calculated. These two are then combined to obtain the gradient of the Q value with respect to the Actor network parameters. The Actor network parameters are updated by ascending along the gradient direction (opposite to descending, since the goal is to maximize the Q value), enabling it to output better action commands.

[0117] Unlike traditional LSTM based on static prediction, this method has continuous learning capabilities and can dynamically adapt to changes in coal seam rhythm.

[0118] S7. Security Protection Mechanism Activated: The system sets up a multi-level security protection strategy independent of the reinforcement learning policy network. This strategy has the highest execution priority. It directly monitors the raw sensor data or the derived quantities calculated at extremely high frequencies. Once the following conditions are met, it unconditionally overrides the current chain speed instruction and triggers the corresponding protection:

[0119] Over-tension speed reduction protection: If the instantaneous value of the chain tension force collected in real time is greater than the rated upper limit, the speed reduction protection will be triggered immediately, reducing the chain speed to a fixed proportion of the current speed (such as 70%) or a preset safe speed.

[0120] Over-power shutdown protection: If the instantaneous value of the motor power collected in real time exceeds 110% of the rated power, the emergency shutdown protection will be triggered immediately to cut off the motor power.

[0121] Sudden change in rhythm and chain deceleration: If the stent pushing rhythm calculated based on a preset very short time window (such as 5 seconds) exceeds the safety threshold, it is determined that the pushing rhythm has changed suddenly, and a chain deceleration is immediately triggered, forcibly setting the chain speed to a preset low-speed crawling state.

[0122] Once the above protection mechanism is activated, the system will temporarily disconnect from the speed adjustment mode of the reinforcement learning policy network and enter the security protection mode; after the risk indicators return to normal and are confirmed by the operator, it will switch back to the speed adjustment mode of the reinforcement learning policy network.

[0123] The scraper conveyor speed control system of the present invention consists of the following functional modules:

[0124] (1) Multi-source state perception module; This module is responsible for collecting and integrating the operating data of various equipment in the fully mechanized mining face, mainly including: coal mining machine advance speed; support pushing rhythm; scraper chain speed; motor current; coal flow rate; chain tension. High-frequency data acquisition, fusion and structured encoding are realized through edge computing devices or industrial data buses to form a state vector for speed regulation decision.

[0125] (2) Intelligent decision-making module for speed regulation strategy; This module constructs a State-Action-Reward (SAR) model based on the Deep Reinforcement Learning (DDPG) algorithm, and automatically generates the optimal speed regulation command according to the current running state. This model has self-learning capability and can continuously optimize the control strategy through running feedback.

[0126] (3) Dynamic operation boundary modeling module; The operation safety boundary model to be constructed is: chain tension change rate limit; motor safe power range; scraper chain speed change rate limit. Before executing control commands, the boundary legality of the action is judged to ensure that the speed regulation process operates within the safe range.

[0127] (4) Execution control and closed-loop feedback module; the speed regulation command after boundary filtering is transmitted to the frequency converter or speed regulation unit to control the dynamic change of chain speed; the execution result is fed back to the learning module to participate in the next round of strategy optimization, forming an intelligent closed loop of perception-decision-execution-feedback.

[0128] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

[0129] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A speed control method for scraper conveyors that integrates sensing and self-evolution strategies, characterized in that, Includes the following steps: S1. Data Acquisition: The control system connects to the industrial control interfaces of each device in the fully mechanized mining face through an edge gateway, and communicates with each device using the Modbus TCP / UDP protocol to acquire data. S2. Construct a multi-dimensional collaborative state vector from the collected data. : = [Support pushing rhythm, coal mining machine advance speed, motor current, real-time chain speed, chain tension, coal flow rate, push rod stroke change rate, automatic coal cutting status, scraper conveyor motor temperature]; S3, Action Decision Generation: This involves generating the state vector... Input the reinforcement learning policy network to obtain the state-action mapping function, and output the chain speed adjustment instruction. As an action; S4. Boundary Validity Verification: Set three safety boundaries: upper limit for tension change rate, safe power range for motor, and upper limit for chain speed change rate; output actions in the strategy network. Subsequently, if the rate of change of tension exceeds the upper limit of the rate of change of tension, or the motor power exceeds the safe power range of the motor, or the rate of change of chain speed exceeds the upper limit of the rate of change of chain speed, the chain speed adjustment command is scaled proportionally to meet the safety boundary; when multiple constraints are violated at the same time, the strictest principle is adopted to find the safe command that meets all safety boundaries and is closest to the chain speed adjustment command. S5. Speed ​​regulation execution and feedback acquisition: The speed regulation command, after boundary verification, is sent to the frequency converter drive system through the edge controller to control the scraper chain to achieve dynamic adjustment of chain speed; A lag compensation mechanism is introduced to predict and correct the delay between policy action and feedback. The specific implementation steps are as follows: S51. The system maintains a fixed-length action-state queue in memory, storing historical states in chronological order. and actions already performed ; S52. Based on the total delay of the control loop, set a fixed delay step d; S53. When the system feedback status is collected at time step t. At that time, this state is a historical action. The result after execution; S54. The system retrieves the state d steps ago from the queue. and actions Compare it with the current state and calculated rewards Combined, they form a complete delayed transfer sample. , , , ), and store it in the experience replay buffer; S55. The reinforcement learning policy network is trained using delayed transition samples; S6. Policy Adaptive Optimization: The reinforcement learning policy network is optimized once every fixed number of steps using stochastic gradient descent. S7. Security Protection Mechanism Activation: Set a multi-level security protection policy independent of the reinforcement learning policy network. This policy has the highest execution priority. Once the following conditions are met, it will unconditionally overwrite the current chain speed instruction and trigger the corresponding protection: Over-tension speed reduction protection: If the instantaneous value of the chain tension force collected in real time is greater than the rated upper limit, the speed reduction protection will be triggered immediately, reducing the chain speed to a fixed proportion of the current speed or a preset safe speed. Over-power shutdown protection: If the instantaneous value of the motor power collected in real time exceeds 110% of the rated power, the emergency shutdown protection will be triggered immediately to cut off the motor power. Sudden change in rhythm and chain deceleration: If the stent pushing rhythm calculated based on a preset very short time window exceeds the safety threshold, it is determined that the pushing rhythm has changed suddenly, and a chain deceleration is immediately triggered, forcibly setting the chain speed to a preset low-speed crawling state. Once the above protection mechanism is activated, the system will temporarily disconnect from the speed adjustment mode of the reinforcement learning policy network and enter the security protection mode; after the risk indicators return to normal and are confirmed by the operator, it will switch back to the speed adjustment mode of the reinforcement learning policy network.

2. The method for speed control of a scraper conveyor integrating sensing and self-evolution strategies according to claim 1, characterized in that, In step S2, the specific methods for obtaining each state parameter are as follows: (1) Stent pushing rhythm: The sliding window frequency statistics method is used to estimate the number of times a certain stent "push slip mark = 1" or the number of times the push rod action state changes within each sliding window. The number of times is divided by the width of the time window to obtain the action frequency per unit time, which is the stent pushing rhythm. (2) Coal mining machine advance speed and motor current: The real-time traction speed signal of the coal mining machine and the three-phase current value of the motor are collected from the coal mining machine PLC or monitoring system; the real-time traction speed signal of the coal mining machine is processed through a sliding window, and the average value of the instantaneous speed in each sliding window is calculated as the coal mining machine advance speed. The three-phase current values ​​of the motor are processed by a sliding window, and the average current value within each sliding window is calculated as a parameter reflecting the average load. The maximum current value within each sliding window is then used as a parameter for overload detection. (3) Real-time chain speed; First, the instantaneous chain speed v is calculated using the following formula: v = π * D * P * n / L; Where D is the chain diameter, P is the chain pitch, n is the motor speed at the current moment, and L is the chain length; after calculating the instantaneous chain speed, the instantaneous chain speed is processed through a sliding window to become the real-time chain speed; (4) Chain tension: obtained by installing a pressure sensor in the piston chamber of the hydraulic cylinder of the tensioning device at the tail of the scraper conveyor; (5) Coal flow rate: The coal flow rate during the scraper conveying process is estimated by combining the chain speed and the weighing section. The estimation formula is: Q = G * v / L', where Q is the coal flow rate, G is the weight of the weighing section, v is the instantaneous chain speed, and L' is the length of the weighing section. (6) Push rod stroke change rate: Real-time stroke data of the push rod is obtained by a displacement sensor installed on the hydraulic support pushing cylinder; for the collected original stroke data S, a sliding window is set, and the linear fitting slope of the stroke value within the sliding window is calculated, which is the "push rod stroke change rate"; (7) Automatic coal cutting status: read from the coal mining machine PLC or the fully mechanized mining face automation control system; (8) Scraper motor temperature: The temperature sensor data embedded in the three-phase winding of the motor is obtained through the fieldbus.

3. The method for speed control of a scraper conveyor integrating sensing and self-evolution strategies according to claim 1, characterized in that, In step S54, the reward function of the reinforcement learning policy network is defined as: ； In the above functions, Q, P, ΔF, It is based on the action output from the reinforcement learning policy network. After adjusting the chain speed, new state data is obtained; , , , All are weighting coefficients; in, This refers to the coal flow rate per unit time. The severity of motor current overload is calculated as follows: Motor current is sampled at a fixed frequency; if any motor current exceeds 110% of the rated current, the sampling point is considered overloaded. A sliding time window is set to perform a weighted evaluation based on both the frequency and duration of the overload. ； Where N is the total number of overload events occurring within the statistical window, and T is the total time spent in an overload state within the cumulative window. and These are the weighting coefficients; The rate of change of chain tension is calculated as follows: ； , Let be the chain tension at time t and time t-1, respectively. The time difference between two moments; This is a penalty item for the degree of mismatch in the operating steps among the support frame, coal mining machine, and scraper conveyor.

Images