Intelligent anchor rod support system for phosphate mine based on multi-source data fusion

Abstract

This invention provides an intelligent anchor bolt support system for phosphate mines based on multi-source data fusion, belonging to the field of intelligent anchor bolt support technology for phosphate mines. The intelligent anchor bolt support system includes: a data acquisition module that collects multi-source monitoring signals to obtain basic data sources; a data preprocessing module that extracts core feature parameters to generate a standardized feature dataset; a data fusion module that constructs a Bayesian weighted fusion model, performs probabilistic inference, and obtains a comprehensive evaluation result; a decision generation module that constructs a multi-agent reinforcement learning model to generate an optimal distributed control instruction set; an execution control module that performs instruction parsing and target value calibration, and uses a magnetorheological damper to drive anchor bolt stress adjustment; and a feedback iteration module that optimizes the parameters corresponding to the Bayesian weighted fusion model and the multi-agent reinforcement learning model based on the evaluated control effect. This invention significantly reduces the misjudgment rate of rock mass stability levels, achieves global collaborative bearing capacity of the anchor bolt group, and enables the control strategy to adapt to dynamic changes in phosphate mine geology.

Description

Technical Field

[0001] This invention relates to the field of intelligent anchor bolt support technology for phosphate mines, and more specifically to an intelligent anchor bolt support system for phosphate mines based on multi-source data fusion. Background Technology

[0002] As mining progresses and land resources become increasingly scarce, promoting underground space development has become a strategic priority for my country. As an important underground support structure, anchor bolts, with their excellent adaptability to geological conditions and outstanding economic advantages, are now widely used in phosphate mine support systems, playing a crucial role in ensuring the safety and economy of underground space development.

[0003] Existing intelligent anchor bolt support systems for phosphate mines suffer from several drawbacks during use. These include interference from multiple monitoring signals in complex geological conditions and the limited effectiveness of assessing rock mass stability with a single signal, leading to misjudgments in support warnings. Furthermore, traditional anchor bolt support systems often rely on passive, single-point control, lacking the collaborative bearing capacity of anchor bolt groups, which can easily trigger single-point overloads and cascading collapses. Additionally, the lack of closed-loop feedback and model iteration in anchor bolt prestress control makes it difficult to adapt to changes in geological conditions over long-term use, resulting in a tendency for the support effect to deteriorate. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent anchor bolt support system for phosphate mines based on multi-source data fusion, in order to solve the technical problems in the existing technology where signal interference in complex geological conditions of phosphate mines leads to misjudgment of support early warning, which can easily cause single-point overload and chain collapse risks, as well as the lack of closed-loop feedback and model iteration in anchor bolt prestress control.

[0005] To achieve the above objectives, this invention provides an intelligent anchor bolt support system for phosphate mines based on multi-source data fusion. The intelligent anchor bolt support system includes: a data acquisition module for acquiring multi-source monitoring signals by deploying multiple types of sensors to obtain a basic data source; a data preprocessing module for performing interference identification and noise reduction on the basic data source, extracting core feature parameters, and generating a standardized feature dataset; a data fusion module for constructing a Bayesian weighted fusion model, initializing model parameters and allocating weights, and performing probabilistic inference based on the standardized feature dataset to obtain a comprehensive evaluation result; a decision generation module for constructing a multi-agent reinforcement learning model, modeling agents and setting learning parameters, and performing iterative reinforcement learning training based on the comprehensive evaluation result to generate an optimal distributed control instruction set; and an execution control module for using an anchor bolt prestress dynamic compensation actuator to perform instruction parsing and target value calibration according to the optimal distributed control instruction set, and using a magnetorheological damper to drive anchor bolt stress adjustment, achieving closed-loop correction and dynamic compensation of anchor bolt prestress.

[0006] Optionally, the step of deploying multiple types of sensors to collect multi-source monitoring signals and obtain basic data sources includes: determining the monitoring points and deployment locations of multiple types of sensors based on the geological conditions of the target phosphate mine; performing on-site calibration after the deployment of multiple types of sensors; and connecting the multiple types of sensors to a data acquisition terminal, which adopts a dual-link design and communicates with the data processing center to collect multi-source monitoring signals.

[0007] Optionally, the step of performing interference identification and noise reduction processing on the basic data source, and extracting core feature parameters to generate a standardized feature dataset includes: comparing the collected multi-source monitoring signals with a preset interference signal feature library to identify the type of interference signal; using differentiated noise reduction algorithms to classify and denoise the interference signals according to different types of interference signals; filtering the denoised multi-source monitoring signals, removing abnormal signals, and extracting the core features of the multi-source monitoring signals to generate a standardized feature dataset.

[0008] Optionally, the construction of the Bayesian weighted fusion model and the initialization and weight allocation of model parameters include: constructing a Bayesian weighted fusion model based on the rock mechanical characteristics of the target phosphate mine; the Bayesian weighted fusion model includes a feature input layer, a weight allocation layer, a probabilistic inference layer, and a result output layer; initializing the Bayesian weighted fusion model and importing the initial weight parameters obtained from training with historical support data to ensure that the model is adapted to the geological conditions of the target phosphate mine.

[0009] Optionally, the step of performing probabilistic inference based on the standardized feature dataset to obtain a comprehensive evaluation result includes: importing the standardized feature dataset into the feature input layer according to the category of the standardized feature dataset; weighting the feature values ​​of the standardized feature dataset by a weight allocation layer according to preset prior weights; calling Bayes' theorem by a probabilistic inference layer and combining the weighted feature values ​​to calculate the posterior probability of the target event; determining each evaluation index based on the maximum value of the posterior probability, generating an evaluation report containing each evaluation index, and performing cross-validation to obtain a comprehensive evaluation result.

[0010] Optionally, the construction of the multi-agent reinforcement learning model, including agent modeling and learning parameter setting, includes: constructing an agent reinforcement learning model with the target phosphate mine's mining support area as the boundary, modeling each anchor bolt and the surrounding rock within its target range as a corresponding anchor bolt agent; setting the maximization of the force balance of the anchor bolt group in the target area as the learning objective of the agent reinforcement learning model; setting the minimization of the sum of squared deviations between the force values ​​of all anchor bolts in the target area and the average force value as the objective function, designing a reward and punishment mechanism, and setting reinforcement learning parameters.

[0011] Optionally, the step of performing reinforcement learning iterative training based on the comprehensive evaluation results to generate the optimal distributed control instruction set includes: converting the comprehensive evaluation results into the environmental state input of the agent reinforcement learning model; after each anchor bolt agent obtains the environmental state input, it initially generates control actions; after the intelligent anchor bolt support system executes the control actions, each anchor bolt agent performs learning corrections; repeating the above steps until the number of iterations reaches a preset value to obtain the optimal distributed control instruction set.

[0012] Optionally, the step of using an anchor bolt prestress dynamic compensation actuator to perform instruction parsing and target value calibration based on the optimal distributed control instruction set includes: receiving the optimal distributed control instruction set via Ethernet for the target phosphate mine; using the anchor bolt prestress dynamic compensation actuator to decode and verify the optimal distributed control instruction set to determine the integrity of the instructions; comparing the target prestress value extracted from the optimal distributed control instruction set with the actual prestress value, and calculating the adjustment difference and driving parameters.

[0013] Optionally, the step of using a magnetorheological damper to drive anchor stress adjustment to achieve closed-loop correction and dynamic compensation of anchor prestress includes: outputting a corresponding current signal to the magnetorheological damper according to the driving parameters to adjust the prestress; when the actual prestress value during the prestress adjustment process reaches the target prestress value and remains stable for a preset time, the magnetorheological damper maintains its current state, completing a single dynamic compensation of anchor prestress.

[0014] Optionally, the intelligent anchor bolt support system further includes: a feedback iteration module, used to collect data from the execution control module, perform data preprocessing and data fusion, compare and evaluate the control effect, and optimize the parameters corresponding to the Bayesian weighted fusion model and the multi-agent reinforcement learning model based on the evaluated control effect, so as to improve the performance of the intelligent anchor bolt support system.

[0015] Through the above technical solutions, by constructing a Bayesian weighted fusion model, the one-sidedness of single signal evaluation is broken, realizing the perception of the entire rock mass state from deep to shallow to the support body. The misjudgment rate of rock mass stability level is significantly reduced, providing accurate state basis for subsequent support control. Through the construction of a multi-agent reinforcement learning model and anchor bolt agent modeling, the global collaborative bearing of the anchor bolt group is realized, avoiding the failure of a single anchor bolt due to overload. The control strategy can adapt to the dynamic changes of phosphate mine geology, the regional anchor bolt force balance is greatly improved, and the overall risk resistance of the support system is significantly enhanced. Through the precise execution of control by magnetorheological dampers, the precise and rapid adjustment of anchor bolt prestress is realized, ensuring the effective implementation of decision commands. The entire closed-loop control process of acquisition-processing-fusion-decision-execution-feedback is realized. The system can adapt to the dynamic changes of phosphate mine geological conditions, and the long-term support effect remains stable.

[0016] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the intelligent anchor bolt support system for phosphate mines based on multi-source data fusion, as described in this invention. Figure 1 ; Figure 2 This is a schematic diagram of the process for acquiring multi-source monitoring signals in this invention; Figure 3 This is a schematic diagram of the process for generating a standardized feature dataset in this invention; Figure 4 This is a schematic diagram of the process of constructing the Bayesian weighted fusion model in this invention; Figure 5 This is a flowchart illustrating the process of obtaining comprehensive evaluation results in this invention; Figure 6 This is a schematic diagram of the process of constructing a multi-agent reinforcement learning model in this invention; Figure 7 This is a schematic diagram of the process for generating the optimal distributed control instruction set in this invention; Figure 8 This is a flowchart illustrating the instruction parsing and target value calibration process in this invention; Figure 9 This is a flowchart illustrating the process of generating a comprehensive evaluation report after regulation in this invention; Figure 10 This is a flowchart illustrating the intelligent anchor bolt support system for phosphate mines based on multi-source data fusion, as described in this invention. Figure 2 . Detailed Implementation

[0018] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0019] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0020] Please refer to Figure 1This invention provides an intelligent anchor bolt support system for phosphate mines based on multi-source data fusion. The intelligent anchor bolt support system may include: a data acquisition module for acquiring multi-source monitoring signals by deploying multiple types of sensors to obtain a basic data source; a data preprocessing module for identifying and reducing noise in the basic data source, extracting core feature parameters, and generating a standardized feature dataset; a data fusion module for constructing a Bayesian weighted fusion model, initializing model parameters and assigning weights, and performing probabilistic inference based on the standardized feature dataset to obtain a comprehensive evaluation result; and a decision generation module for constructing a multi-agent reinforcement learning model to model the agents. Based on the learning parameter settings and comprehensive evaluation results, reinforcement learning iterative training is performed to generate the optimal distributed control instruction set. The execution control module can be used to perform instruction parsing and target value calibration using the anchor bolt prestress dynamic compensation actuator according to the optimal distributed control instruction set, and to drive anchor bolt stress adjustment using a magnetorheological damper to achieve closed-loop correction and anchor bolt prestress dynamic compensation. The feedback iteration module can be used to collect data from the execution control module, perform data preprocessing and data fusion, compare and evaluate the control effect, and optimize the parameters corresponding to the Bayesian weighted fusion model and the multi-agent reinforcement learning model based on the evaluated control effect, thereby improving the performance of the intelligent anchor bolt support system.

[0021] Combination Figure 10 In this embodiment of the invention, the data acquisition module can be used to acquire multi-source monitoring signals by deploying multiple types of sensors to obtain basic data sources.

[0022] Please refer to Figure 2 In this embodiment of the invention, by deploying multiple types of sensors to collect multi-source monitoring signals and obtaining the basic data source, the data source may include: Step S100: Determine the monitoring points and deployment locations of various types of sensors based on the geological conditions of the target phosphate mine (e.g., lithological distribution, burial depth, joint development, etc.).

[0023] In a preferred embodiment of the present invention, microseismic sensor groups can be arranged in a cross-shaped pattern in the roof area of ​​the phosphate mine, with three sensors in each group arranged in an equilateral triangle to achieve three-dimensional positioning of the seismic source; along the anchor bolt axis in the side area, a flat mounting surface is pre-set on the surrounding rock surface at the top of each anchor bolt for attaching the ground sound sensor; during the anchor bolt processing stage, the anchor bolt stress sensor is integrated into the stress-sensitive area in the middle of the bolt body to ensure that the sensor and the anchor bolt are stressed synchronously, and at the same time, special mining fasteners are used to fix the sensor during the deployment process to avoid displacement caused by mine vibration.

[0024] Step S110: After deploying multiple types of sensors, perform on-site calibration.

[0025] In a preferred embodiment of the present invention, the microseismic sensor can simulate rock fracture signals of different energy levels using a standard vibration source, record the sensor response amplitude and frequency, and establish a calibration curve; the ground sound sensor uses a pulse signal generator to output a standard pulse, and adjusts the sensor gain to control the signal acquisition error within the allowable range; the anchor bolt stress sensor applies a standard load to the anchor bolt through a hydraulic loading device, corrects the deviation between the sensor output value and the actual force, and ensures the accuracy of the acquired data.

[0026] Step S120: Connect multiple types of sensors to the data acquisition terminal. The data acquisition terminal adopts a dual-link design and communicates with the data processing center to collect multi-source monitoring signals.

[0027] In a preferred embodiment of the invention, the sensors are connected to a field data acquisition terminal via a mine-use explosion-proof industrial bus. The terminal employs a dual-link design (e.g., fiber optic as primary and wireless as backup) to communicate with the ground data processing center. When setting sampling parameters, the sampling frequency of the microseismic sensor is adapted to the frequency band of rock fracture signals, the sampling frequency of the ground acoustic sensor covers the range of fracture expansion pulse signals, and the sampling frequency of the anchor stress sensor meets the actual stress change monitoring requirements. After data acquisition is initiated, the terminal acquires signals from each sensor according to a preset cycle, temporarily stores the data using a cyclic buffering mechanism, and transmits the data to the processing center using anti-interference coding technology to ensure continuous and unlossable data transmission.

[0028] For example, in a phosphate mine roadway (e.g., with apatite as the main lithology, deep burial, and well-developed joints), a set of microseismic sensors is installed at fixed intervals on the roof. Each set of three sensors is installed at the center of the roadway roof and above the side walls, arranged in an equilateral triangle. After the surrounding rock surface at the top of each anchor bolt on the side walls is ground smooth, a ground acoustic sensor is attached and fixed with epoxy resin. During anchor bolt processing, a stress sensor is embedded in the middle of the bolt body and sealed with sealant for waterproofing. During the calibration phase: a vibration source is used to simulate a microseismic signal with increasing energy to correct the microseismic sensor response curve; a pulse generator outputs a standard pulse of 100 times / minute to adjust the ground acoustic sensor gain; a hydraulic device is used to apply a gradient load to the anchor bolt to calibrate the stress sensor error. The transmission link uses mining fiber optic cable to connect the field terminal to the ground center, with a backup 5G industrial module. The sampling frequencies of the microseismic, ground acoustic, and stress sensors are set to corresponding values ​​to achieve 24-hour uninterrupted data acquisition. The sensors maintain stable data transmission even under interference environments such as mine car traffic and blasting operations.

[0029] In this embodiment of the invention, the data preprocessing module can be used to perform interference identification and noise reduction on the basic data source, and extract core feature parameters to generate a standardized feature dataset.

[0030] Please refer to Figure 3In this embodiment of the invention, interference identification and noise reduction are performed on the basic data source, and core feature parameters are extracted to generate a standardized feature dataset, which may include: Step S200: Compare the collected multi-source monitoring signals with a preset interference signal feature library to identify the type of interference signal.

[0031] In a preferred embodiment of the present invention, the main interference of the microseismic signal is the periodic vibration of equipment such as tunneling machines and crushers, which manifests as a continuous signal with a fixed frequency; the main interference of the ground sound signal is the instantaneous impact of mine cars passing by, which manifests as short-duration high-amplitude pulses; and the main interference of the anchor bolt stress signal is the drift caused by changes in ambient temperature, which manifests as a slowly linearly changing baseline offset. The identified interference signals are marked to provide a basis for targeted noise reduction.

[0032] Step S210: Based on different types of interference signals, use differentiated noise reduction algorithms to classify and reduce noise in the interference signals.

[0033] In a preferred embodiment of the present invention, the microseismic signal can be decomposed into multiple scales using a wavelet transform algorithm. After decomposition, the wavelet coefficients corresponding to the interference frequencies of the equipment are removed, and then the effective signal after noise reduction is obtained by inverse transform reconstruction. The ground sound signal is filtered using a moving average filtering algorithm, and a filtering window matching the duration of the impact interference is set to smooth the instantaneous impact noise and retain the continuous pulse signal of crack propagation. The anchor stress signal is filtered using a Kalman filtering algorithm to establish the state equation and the observation equation. The baseline offset caused by temperature drift is corrected by prediction-update iteration, and the true stress signal is output.

[0034] Step S220: Filter the noise-reduced multi-source monitoring signals, remove abnormal signals, extract the core features of the multi-source monitoring signals, and generate a standardized feature dataset.

[0035] In a preferred embodiment of the present invention, the noise-reduced signals can be screened for effectiveness, eliminating abnormal signal segments under extreme conditions such as blasting operations and equipment start-up and shutdown (e.g., determined by signal amplitude threshold and duration). Subsequently, core features of each signal are extracted: for microseismic signals, the three-dimensional coordinates of the seismic source are extracted (e.g., through multi-sensor time-difference positioning), signal energy (e.g., integral calculation), and dominant frequency (e.g., power spectrum analysis); for ground sound signals, the number of pulses per unit time, peak and average pulse amplitude, and pulse duration are extracted; for anchor bolt stress signals, the actual force value, stress change rate (e.g., difference between adjacent sampling points), and stress stability coefficient (e.g., fluctuation amplitude of continuous sampling points) are extracted. All feature parameters are organized in the format of "sensor number-acquisition time-feature type-feature value" to form a standardized feature dataset.

[0036] For example, in the aforementioned phosphate mine data processing, the signal analysis module identified fixed-frequency interference from the tunneling machine in the microseismic signal, instantaneous impact interference from mine car passage in the ground sound signal, and temperature drift interference in the stress signal. The microseismic signal was decomposed into 6 layers of wavelet, and the wavelet coefficients corresponding to the interference frequencies were removed before reconstruction. The ground sound signal was filtered using a 10-second window moving average to remove instantaneous impacts. The stress signal was corrected for temperature drift using Kalman filtering, resulting in a reduced baseline fluctuation amplitude. After filtering out 5-minute abnormal signal segments during blasting operations, data for a specific time period was extracted: the microseismic signal source coordinates were (X1, Y1, Z1), energy was E1, and dominant frequency was F1; the ground sound signal had N1 pulse counts and A1 peak value per hour; and the anchor bolt stress signal had S1 actual value, V1 rate of change, and K1 stability coefficient. This data was then organized according to a standard format and stored in the feature database.

[0037] In this embodiment of the invention, the data fusion module can be used to construct a Bayesian weighted fusion model, initialize model parameters and assign weights, and perform probabilistic inference based on a standardized feature dataset to obtain a comprehensive evaluation result.

[0038] Please refer to Figure 4 In this embodiment of the invention, constructing a Bayesian weighted fusion model and initializing model parameters and assigning weights may include: Step S300: Construct a Bayesian weighted fusion model based on the rock mechanics characteristics of the target phosphate mine.

[0039] Step S310: The Bayesian weighted fusion model includes a feature input layer, a weight allocation layer, a probability inference layer, and a result output layer.

[0040] In a preferred embodiment of the present invention, the feature input layer can receive a standardized feature dataset; the weight allocation layer presets prior weight values ​​based on the degree of influence of different features on rock mass stability assessment (for example, microseismic energy reflects deep fractures, so the weight is set to a higher value; the number of ground sound pulses reflects shallow fractures, so the weight is set to a medium value; the rate of change of anchor stress reflects the support bearing state, so the weight is set to a higher value); the probabilistic inference layer embeds Bayes' formula to calculate the posterior probability; the result output layer defines a three-level rock mass stability level of "stable-critically stable-unstable", the coordinate range of the stress concentration area, and the anchor stress balance index (for example, a value of 0-1, the closer to 1, the more balanced).

[0041] Step S320: Initialize the Bayesian weighted fusion model and import the initial weight parameters obtained from training historical support data to ensure that the model is adapted to the geological conditions of the target phosphate mine.

[0042] Please refer to Figure 5In this embodiment of the invention, performing probabilistic reasoning based on a standardized feature dataset to obtain a comprehensive evaluation result may include: Step S301: Import the standardized feature dataset into the feature input layer according to the category of the standardized feature dataset.

[0043] Step S302: According to the preset prior weights, the weight allocation layer performs weighted processing on each feature value of the standardized feature dataset.

[0044] Step S303: The probabilistic inference layer calls Bayes' theorem and, in combination with the weighted feature values, calculates the posterior probability of the target event.

[0045] In a preferred embodiment of the present invention, when the probabilistic inference layer invokes the Bayesian formula, events such as "rock mass stability level," "stress concentration," and "stress equilibrium" can be used as the events to be inferred. Combined with weighted feature data, the posterior probability of each event is calculated. For example, when microseismic energy exceeds a threshold, the number of ground sound pulses increases, and the rate of change of anchor bolt stress accelerates, the posterior probability of the "rock mass instability" event significantly increases; when the stress values ​​of multiple anchor bolts in a certain area far exceed the average level, the posterior probability of the "stress concentration in this area" event increases. During the calculation process, the model automatically records the contribution of each feature to the posterior probability, providing a basis for subsequent weight optimization.

[0046] Step S304: Determine each evaluation index based on the maximum value of the posterior probability, generate an evaluation report containing each evaluation index, and perform cross-validation to obtain a comprehensive evaluation result.

[0047] In a preferred embodiment of the present invention, the rock mass stability level is taken as the level with the highest posterior probability; the stress concentration area is determined by locating the coordinates of the "stress concentration" area with the highest posterior probability, combined with the sensor distribution range; the anchor bolt force balance is converted into a balance index of 0-1 by calculating the standard deviation of the force values ​​of all anchor bolts in the area. After generating a preliminary assessment report containing the above indicators, it is cross-validated with on-site geological inspection records (e.g., roof crack observation, anchor bolt deformation inspection). If there is a deviation between the assessment result and the on-site condition that exceeds the allowable range, the process returns to the weight allocation layer to fine-tune the prior weights, recalculates, and outputs the final comprehensive assessment report.

[0048] For example, based on the aforementioned phosphate mine mining characteristic data, a Bayesian weighted fusion model is constructed, with preset weights for microseismic energy (W1), ground sound pulse frequency (W2), and anchor stress change rate (W3) (W1=W3>W2). After inputting the characteristic data, the weight allocation layer completes the weighting process, and the probabilistic inference layer calculates the following: the posterior probability of the "critical stability" event is 0.75 (e.g., much higher than "stable" 0.20 and "unstable" 0.05); the posterior probability of the "stress concentration in the left sidewall area of ​​the roadway" event is 0.80; and the anchor stress balance index in this area is 0.45 (e.g., lower than the acceptable threshold of 0.6). After the preliminary assessment report is generated, it is compared with the on-site inspection records: new fractures were indeed observed to develop in the left sidewall, and the anchors showed slight bending deformation, verifying the reliability of the assessment results. The final report is output: "The rock mass stability level is critically stable, the stress concentration area is located in the left sidewall of the roadway (e.g., coordinate range X2-X3, Y2-Y3, Z2-Z3), and the anchor stress in this area is unbalanced."

[0049] In this embodiment of the invention, the decision generation module can be used to construct a multi-agent reinforcement learning model, perform agent modeling and learning parameter setting, and perform reinforcement learning iterative training based on the comprehensive evaluation results to generate the optimal distributed control instruction set.

[0050] Please refer to Figure 6 In this embodiment of the invention, constructing a multi-agent reinforcement learning model and setting agent modeling and learning parameters may include: Step S400: Using the target phosphate mine's mining support area as the boundary, construct an agent reinforcement learning model, modeling each anchor bolt and the surrounding rock within its target range as the corresponding anchor bolt agent.

[0051] In a preferred embodiment of the present invention, all anchor bolt agents within the system can interact in terms of state through a communication interface. Each anchor bolt agent includes a state perception module, a decision output module, and a learning update module: the state perception module is used to acquire its own stress characteristics and the state information of neighboring agents; the decision output module is used to generate prestress control action commands; and the learning update module is used to adjust the decision strategy based on the control effect. Simultaneously, a system global coordination module is set up to summarize the states of each agent and allocate global rewards.

[0052] Step S410: Maximize the force balance of the anchor bolt group in the target area as the learning objective of the agent reinforcement learning model.

[0053] Step S420: Minimize the sum of squared deviations between the force values ​​of all anchor bolts in the target area and the average force value as the objective function, design a reward and punishment mechanism, and set reinforcement learning parameters.

[0054] In a preferred embodiment of the present invention, when designing the reward and punishment mechanism, a global positive reward is given when the regional force balance index increases (e.g., the reward value is positively correlated with the increase); a global negative penalty is given when the balance index decreases or the anchor bolt force exceeds the threshold (e.g., the penalty value is positively correlated with the decrease or the degree of exceeding the threshold). When setting reinforcement learning parameters, the Q-learning algorithm can be used as the learning core, the learning rate can be set to a value that adapts to the dynamic changes in the mining area, the discount factor can be set to a value that emphasizes the current control effect, and the exploration rate can be gradually reduced as the number of iterations increases (from exploration-oriented to utilization-oriented), ensuring that the optimal strategy is explored in the early stage and reliable decisions are output stably in the later stage.

[0055] Please refer to Figure 7 In this embodiment of the invention, based on the comprehensive evaluation results, reinforcement learning iterative training is performed to generate the optimal distributed control instruction set, which may include: Step S401: Transform the comprehensive evaluation results into the environmental state input of the agent reinforcement learning model.

[0056] Step S402: After each anchor bolt intelligent agent obtains the environmental status input, it initially generates control actions.

[0057] In a preferred embodiment of the present invention, the coordinates of the stress concentration area can be mapped to a "dangerous area marker", the anchor bolt force balance index can be used as an "initial environmental score", and the actual force value of each anchor bolt can be used as the "initial state value" of the corresponding intelligent agent. After each anchor bolt intelligent agent obtains its own and neighboring intelligent agents' states through the state perception module, the decision output module initially generates control actions (e.g., "increase prestress", "decrease prestress", "remain unchanged", etc.).

[0058] Step S403: After the intelligent anchor bolt support system executes the control action, each anchor bolt intelligent agent learns and corrects itself.

[0059] Step S404: Repeat the above steps until the number of iterations reaches the preset value to obtain the optimal distributed control instruction set.

[0060] In a preferred embodiment of the present invention, after the system executes an action, the global coordination module calculates reward and penalty values ​​based on the new force balance index, and each agent corrects its own Q-value table through the learning update module. The process of "state perception - decision output - reward and penalty calculation - strategy update" is repeated until the number of iterations reaches a preset value or the reward and penalty values ​​stabilize in the positive range, and the final set of prestress control instructions for each anchor bolt is output.

[0061] For example, in the critically stable region of the aforementioned phosphate mine, a multi-agent system with 25 anchor bolts is constructed, with one agent corresponding to each anchor bolt, and the communication range of each agent covering three adjacent anchor bolts. The objective function is set as "minimizing the sum of squares of the force deviations of the 25 anchor bolts," with positive rewards proportional to the increase in the equilibrium index, and negative penalties being twice the difference between the excess force value and the threshold. The Q-learning algorithm's learning rate is set to α, the discount factor to γ, and the exploration rate is gradually reduced from 0.8 to 0.1. Input environment state: the 5 anchor bolts on the left side are marked as "dangerous area agents," the initial equilibrium index is 0.45, and the initial force values ​​of each agent are S1-S25. After 100 iterations: the first 20 rounds explored different control combinations, and after the 20th round, the balance index increased to 0.65 and obtained a positive reward; in the 100th round, the balance index stabilized at 0.85, and the control instruction set was output: the prestress of the 5 "danger zone agents" on the left was reduced by ΔS1, the prestress of the 10 surrounding agents was increased by ΔS2, and the remaining 10 agents remained unchanged.

[0062] In this embodiment of the invention, the execution control module can be used to perform instruction parsing and target value calibration according to the optimal distributed control instruction set, using the anchor bolt prestress dynamic compensation execution mechanism, and using the magnetorheological damper to drive anchor bolt stress adjustment, thereby realizing closed-loop correction and anchor bolt prestress dynamic compensation.

[0063] Please refer to Figure 8 In this embodiment of the invention, based on the optimal distributed control instruction set, an anchor bolt prestress dynamic compensation actuator is used to perform instruction parsing and target value calibration, which may include: Step S500: Receive the optimal distributed control instruction set via Ethernet for the target phosphate mine.

[0064] Step S510: Use the anchor bolt prestress dynamic compensation actuator to decode and verify the optimal distributed control command set to determine the integrity of the commands.

[0065] In a preferred embodiment of the present invention, the anchor bolt prestress dynamic compensation actuator (integrated on the top of each anchor bolt) can receive the control instruction set output from the decision generation stage via a mining industrial Ethernet network. The instructions include information such as "anchor bolt number - target prestress value - adjustment duration". The control module of the actuator decodes and verifies the instructions, confirming the integrity of the instructions and the matching of the anchor bolt number. If an instruction is lost or the number is incorrect, it immediately reports to the data processing center and requests a resend. If the instruction is valid, it extracts the target prestress value of the corresponding anchor bolt and retrieves the historical stress data of the anchor bolt (e.g., initial prestress, previous adjustment records, etc.).

[0066] Step S520: Compare the target prestress value extracted from the optimal distributed control command set with the actual prestress value, and calculate the adjustment difference and driving parameters.

[0067] In a preferred embodiment of the present invention, the extracted target prestress value can be compared with the current prestress value actually fed back by the anchor stress sensor, and the adjustment difference can be calculated using the following formula. :

[0068] Among them, the driving parameters of the actuator are calculated by combining the mechanical property parameters of the anchor bolt (e.g., elastic modulus, anchorage length, etc.) with historical adjustment efficiency data: If >0 (prestress needs to be increased), calculate the required positive current intensity I1 (and Positive correlation) and loading time T1 (to ensure a smooth stress increase); if <0 (prestress needs to be reduced), calculate the required reverse current intensity I2 (and | |Positive correlation) and uninstallation time T2; if =0, outputs the "maintain current state" command, and the drive module does not operate. At the same time, a safety threshold is set, and when the rate of stress change exceeds the allowable value during the adjustment process, deceleration adjustment is automatically triggered.

[0069] In this embodiment of the invention, using a magnetorheological damper to drive anchor stress adjustment to achieve closed-loop correction and dynamic compensation of anchor prestress may include: Step S501: Based on the driving parameters, output the corresponding current signal to the magnetorheological damper to adjust the prestress.

[0070] Step S502: When the actual prestress value during the prestress adjustment process reaches the target prestress value and remains stable for a preset time, the magnetorheological damper maintains its current state, completing the dynamic compensation of the anchor bolt prestress for a single operation.

[0071] In a preferred embodiment of the invention, the forward current increases the magnetic field strength of the damper, increasing the damping force and causing the anchor rod to contract, thus gradually increasing the prestress. Conversely, the reverse current weakens the magnetic field strength, reducing the damping force, causing the anchor rod to elongate under rock stress, and gradually decreasing the prestress. During the adjustment process, the anchor stress sensor collects actual stress data and feeds it back to the control module. The control module compares the deviation between the actual stress value and the target value: if the deviation exceeds the allowable range, the current intensity is fine-tuned; if the stress change rate exceeds a threshold, the adjustment time is automatically extended. When the actual stress value reaches the target value and remains stable for a preset time, the control module sends an adjustment completion signal, the drive module stops outputting current, the damper maintains its current state, and a single dynamic prestress compensation is completed.

[0072] For example, in the aforementioned phosphate mine, the control command for a certain anchor bolt (e.g., numbered M1) on the left side slope is "Target prestress value target = 150kN, adjustment duration 60s". After decoding, the actuator control module confirms the command is valid, extracts the historical initial prestress of 180kN and the current real-time stress of 170kN for the anchor bolt, and calculates... =-20kN (needs to be reduced). Based on the anchor bolt's elastic modulus E, anchorage length L, and historical data, the reverse current intensity I2 is calculated to be 2A, the unloading time T2 to be 60s, and the safe rate threshold to be 5kN / s. The drive module outputs a 2A reverse current to the magnetorheological damper, reducing the damping force from F1 to F2. The anchor bolt slowly elongates, and the stress sensor provides real-time feedback: stress 160kN at 30s, 155kN at 45s, and 150kN at 55s. After the control module detects that the stress has stabilized for 10s, it sends an "adjustment complete" signal, the drive module is powered off, and the damper maintains its current state. The maximum stress change rate during this anchor bolt adjustment process is 3kN / s, which does not exceed the safe threshold.

[0073] In this embodiment of the invention, the feedback iteration module can be used to collect data from the execution control module, perform data preprocessing and data fusion, compare and evaluate the control effect, and optimize the parameters corresponding to the Bayesian weighted fusion model and the multi-agent reinforcement learning model based on the evaluated control effect, thereby improving the performance of the intelligent anchor bolt support system.

[0074] Please refer to Figure 9 In this embodiment of the invention, data from the execution control module is collected, preprocessed, and fused. The control effect is compared and evaluated, and the parameters of the Bayesian weighted fusion model and the multi-agent reinforcement learning model are optimized based on the evaluated control effect. This may include: Step S600: After completing the dynamic compensation of anchor bolt prestress, collect signals such as micro-seismic activity, ground sound, and anchor bolt stress within a preset time after adjustment, and repeat data preprocessing and data fusion.

[0075] In a preferred embodiment of the present invention, after the prestressed dynamic compensation is completed, the data acquisition system can be kept running continuously to collect microseismic, ground sound, and anchor stress signals within a preset time period after regulation. During the repeated data preprocessing process, temporary interference signals generated when the equipment resumes operation after regulation can be identified and removed. The same noise reduction algorithm is used to process the three types of signals, extracting core feature parameters consistent with those before regulation, forming a standardized feature dataset after regulation. This ensures the consistency of the processing flow and parameters, thereby guaranteeing the comparability of data before and after regulation.

[0076] Step S610: Input the standardized feature dataset after regulation into the Bayesian weighted fusion model to generate a comprehensive evaluation report after regulation.

[0077] In a preferred embodiment of the present invention, the effects before and after regulation can be compared from three dimensions. In terms of the rock mass stability dimension, compare the change in the stability level (for example, whether it is improved from "critically stable" to "stable"); in terms of the stress concentration dimension, compare the reduction ratio of the stress concentration area range and the decrease amplitude of the microseismic energy within the area; in terms of the bolt force dimension, compare the improvement amplitude of the bolt force balance index in the area and the reduction ratio of the number of bolts exceeding the threshold. Set the standard for achieving the effects, including the improvement or maintenance of the stability level, the reduction of the stress concentration area, and the improvement of the balance index exceeding the preset amplitude. If all three indicators meet the standards, it is determined that the regulation effect is qualified; if any one does not meet the standards, analyze the reasons (such as unreasonable model weights, improper learning parameter settings, etc.).

[0078] Step S620: For the regulation effect evaluation results in the comprehensive evaluation report, optimize the parameters of the multi-agent reinforcement learning model and the Bayesian fusion model respectively.

[0079] In a preferred embodiment of the present invention, for the multi-agent reinforcement learning model, use the qualified "environmental state - regulation instruction - effect data" as positive samples and add them to the model experience library. Fine-tune the learning rate and discount factor of the Q-learning algorithm to make the model converge faster in similar scenarios; for the Bayesian fusion model, according to the matching degree between the feature data after regulation and the evaluation results, correct the prior values of the weights of each feature parameter (for example, if the ground sound signal has insufficient evaluation accuracy for shallow cracks, increase the weight of the ground sound feature). The optimized model is applied to the technical process of the next cycle to implement a continuous improvement mechanism of "one regulation, one optimization".

[0080] For example, 2 hours after the regulation of the above phosphate mine stope is completed, collect the monitoring data within 1 hour. After preprocessing, extract the features: the microseismic energy drops to E2 (E2 < E1), and the main frequency returns to the normal range; the number of ground sound pulses within 1 hour drops to N2 (N2 < N1); the bolt stress balance index increases to 0.85. After inputting into the fusion model, the evaluation report shows that the rock mass stability level is improved to "stable", the stress concentration area range is reduced by 50%, and the number of bolts exceeding the threshold drops from 5 to 0. It is determined that the regulation effect is qualified. Take the data of this "critically stable state - distributed regulation instruction - stable effect" as positive samples and add them to the multi-agent experience library. Adjust the Q-learning learning rate from α to α' (for example, α' < α to improve the convergence stability); increase the weight of the ground sound pulse count in the Bayesian model from W2 to W2' to enhance the evaluation accuracy of shallow cracks. The optimized model is used for the support regulation of the next stope area, and the decision convergence time is shortened by 30%.

[0081] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0083] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0085] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0086] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0087] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0088] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0089] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A smart anchor bolt support system for phosphate mines based on multi-source data fusion, characterized in that, The intelligent anchor bolt support system includes: The data acquisition module is used to collect multi-source monitoring signals by deploying multiple types of sensors to obtain basic data sources; The data preprocessing module is used to perform interference identification and noise reduction on the basic data source, and extract core feature parameters to generate a standardized feature dataset. The data fusion module is used to construct a Bayesian weighted fusion model, initialize model parameters and assign weights, and perform probabilistic inference based on the standardized feature dataset to obtain a comprehensive evaluation result. The decision generation module is used to construct a multi-agent reinforcement learning model, perform agent modeling and learning parameter setting, and perform reinforcement learning iterative training based on the comprehensive evaluation results to generate the optimal distributed control instruction set. The execution control module is used to perform command parsing and target value calibration according to the optimal distributed control command set, using the anchor bolt prestress dynamic compensation execution mechanism, and using the magnetorheological damper to drive the anchor bolt stress adjustment to achieve closed-loop correction and anchor bolt prestress dynamic compensation.

2. The intelligent anchor bolt support system according to claim 1, characterized in that, The process involves deploying multiple types of sensors to collect multi-source monitoring signals, thereby obtaining basic data sources, including: Based on the geological conditions of the target phosphate mine, determine the monitoring points and the deployment locations of various types of sensors; After deploying multiple types of sensors, on-site calibration is performed. Multiple types of sensors are connected to the data acquisition terminal, which adopts a dual-link design and communicates with the data processing center to collect multi-source monitoring signals.

3. The intelligent anchor bolt support system according to claim 1, characterized in that, The process of performing interference identification and noise reduction on the basic data source, and extracting core feature parameters to generate a standardized feature dataset includes: The collected multi-source monitoring signals are compared with a preset interference signal feature library to identify the type of interference signal; Differentiated noise reduction algorithms are used to classify and reduce noise in different types of interference signals. The denoised multi-source monitoring signals are screened to remove abnormal signals, and the core features of the multi-source monitoring signals are extracted to generate a standardized feature dataset.

4. The intelligent anchor bolt support system according to claim 1, characterized in that, The construction of the Bayesian weighted fusion model, including model parameter initialization and weight allocation, includes: Based on the rock mechanics characteristics of the target phosphate mine, a Bayesian weighted fusion model is constructed. The Bayesian weighted fusion model comprises a feature input layer, a weight allocation layer, a probability inference layer, and a result output layer. Initialize the Bayesian weighted fusion model and import the initial weight parameters obtained from training historical support data to ensure that the model is adapted to the geological conditions of the target phosphate mine.

5. The intelligent anchor bolt support system according to claim 4, characterized in that, The step of performing probabilistic inference based on the standardized feature dataset to obtain a comprehensive evaluation result includes: According to the category of the standardized feature dataset, import the standardized feature dataset into the feature input layer; Based on the preset prior weights, the weight allocation layer performs weighted processing on each feature value of the standardized feature dataset; The probabilistic inference layer calls Bayes' theorem and combines the weighted feature values ​​to calculate the posterior probability of the target event. Each evaluation index is determined based on the maximum posterior probability, an evaluation report containing each evaluation index is generated, and cross-validation is performed to obtain a comprehensive evaluation result.

6. The intelligent anchor bolt support system according to claim 1, characterized in that, The construction of the multi-agent reinforcement learning model, including agent modeling and learning parameter setting, includes: Using the target phosphate mine's mining support area as the boundary, an agent reinforcement learning model is constructed, and each anchor bolt and the surrounding rock within its target range are modeled as the corresponding anchor bolt agent. Maximizing the force balance of the anchor bolt group in the target area is set as the learning objective of the agent reinforcement learning model; The objective function is set as minimizing the sum of squared deviations between the force values ​​of all anchor bolts within the target area and the average force value. A reward and punishment mechanism is designed, and reinforcement learning parameters are set.

7. The intelligent anchor bolt support system according to claim 6, characterized in that, The step of performing reinforcement learning iterative training based on the comprehensive evaluation results to generate the optimal distributed control instruction set includes: The comprehensive evaluation results are then converted into the environmental state input for the agent reinforcement learning model. After acquiring the environmental status input, each anchor bolt intelligent agent initially generates control actions; After the intelligent anchor bolt support system executes the control action, each anchor bolt intelligent agent learns and corrects itself. Repeat the above steps until the number of iterations reaches the preset value, and obtain the optimal distributed control instruction set.

8. The intelligent anchor bolt support system according to claim 1, characterized in that, The step of using an anchor bolt prestress dynamic compensation actuator to perform command parsing and target value calibration based on the optimal distributed control command set includes: Receive the optimal distributed control instruction set via Ethernet for the target phosphate mine; The anchor bolt prestress dynamic compensation actuator is used to decode and verify the optimal distributed control command set to determine the integrity of the command. The target prestress value extracted from the optimal distributed control command set is compared with the actual prestress value, and the adjustment difference and driving parameters are calculated.

9. The intelligent anchor bolt support system according to claim 8, characterized in that, The method of using a magnetorheological damper to drive anchor stress adjustment to achieve closed-loop correction and dynamic compensation of anchor prestress includes: Based on the driving parameters, a corresponding current signal is output to the magnetorheological damper to adjust the prestress. When the actual prestress value reaches the target prestress value and remains stable for a preset time during the prestress adjustment process, the magnetorheological damper maintains its current state, completing the dynamic compensation of the anchor bolt prestress for a single operation.

10. The intelligent anchor bolt support system according to claim 1, characterized in that, The intelligent anchor bolt support system also includes: The feedback iteration module is used to collect data from the execution control module, perform data preprocessing and data fusion, compare and evaluate the control effect, and optimize the parameters of the Bayesian weighted fusion model and the multi-agent reinforcement learning model based on the evaluation effect, so as to improve the performance of the intelligent anchor bolt support system.

Images