Intelligent feeding and distribution method for underground multi-source gangue solid filling based on digital twinning

Abstract

The application provides a kind of digital twin-based underground multi-source gangue solid filling intelligent feeding and distribution method, the real-time acquisition of filling working face raw coal production and multi-source gangue production and storage state is obtained through underground intelligent sensing network, and the production-storage-transportation-filling integrated digital twin model is constructed by edge computing fusion;Based on the model, the gangue demand prediction value of the future period is output by using the prediction model combined with the design filling mining quality ratio;Further, a multi-objective optimization model is constructed, and the optimal strategy is obtained by virtual simulation and pre-rehearsal of the feeding and distribution strategy in the digital twin space, and finally it is converted into control instructions such as feeding frequency and conveying speed to execute;At the same time, through online comparison of the residual error between the physical side filling effect and the virtual theoretical value, the prediction and optimization model are iteratively self-corrected by reinforcement learning.The application realizes real-time dynamic matching of multi-source gangue supply and filling demand, significantly improves warehouse utilization, reduces transportation energy consumption and guarantees the continuity of filling operation.

Description

Technical Field

[0001] This invention relates to the field of green mining and intelligent control technology in coal mines, specifically to an intelligent feeding method for underground multi-source gangue solid filling based on digital twins. Background Technology

[0002] In solid backfilling mining of coal mines, gangue generated underground needs to be used as backfill material in the goaf. To ensure the compaction of the backfill and the effectiveness of roof control, coal mining and backfilling operations must follow the designed backfill-to-mining mass ratio, that is, the mass ratio of gangue required to backfill for each unit mass of raw coal mined. However, the sources of gangue underground are scattered, the production volume fluctuates greatly, the advance speed of the backfilling face varies, and there are constraints such as the storage capacity of each gangue bin and the capacity of conveying equipment. How to dynamically allocate gangue from multiple sources based on real-time raw coal production and the designed backfill-to-mining mass ratio to avoid underfilling or overflow is a technical problem that urgently needs to be solved.

[0003] Existing technologies largely rely on manual experience or simple rule-based control, making it difficult to simultaneously consider storage utilization, transportation energy consumption, and filling continuity, and they cannot adapt to rapid changes in operating conditions. The development of digital twin and intelligent optimization technologies has made it possible to achieve dynamic allocation of multi-source gangue, but there is still no complete method for intelligent dynamic allocation of the entire process of production-storage-transportation-filling of multi-source gangue downhole, oriented towards the production-filling quality ratio. Summary of the Invention

[0004] To address the aforementioned technical shortcomings, the purpose of this invention is to provide an intelligent distribution method for multi-source gangue solid filling in downhole based on digital twins, in order to solve the problems of untimely multi-source gangue distribution, difficulty in ensuring the filling-to-production quality ratio, and low operating efficiency in the existing technology.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A digital twin-based intelligent feeding method for multi-source gangue solid backfilling in wells includes the following steps:

[0007] Step S1: Real-time data on raw coal production q at the filling face and production and storage status of multi-source gangue in the mine are obtained through the underground intelligent sensing network, and the operating status parameters of the feeder, conveyor and coal mining machine are collected simultaneously.

[0008] Step S2: Perform edge computing fusion processing on the multi-source heterogeneous data collected by the downhole intelligent sensing network to construct a digital twin model integrating downhole multi-source gangue production, storage, transportation and charging;

[0009] Step S3: Based on the historical production data and real-time operation data output by the digital twin model, combined with the designed filling-mining quality ratio γʹ, the prediction model is used to predict the demand for gangue in the filling face within the future time window, and the predicted gangue demand value Q0ʹ is output.

[0010] Step S4: Construct a multi-objective intelligent optimization model with the goals of optimal warehouse utilization, minimum energy consumption of the conveying system, and optimal filling density. Within the digital twin model, conduct virtual simulations in parallel on multiple alternative allocation strategies that follow preset dynamic allocation rules for gangue, and evaluate to obtain the optimal strategy.

[0011] Step S5: Convert the optimal strategy into control commands for the feeding frequency and conveyor belt speed of each gangue bin feeder, and send them to the corresponding equipment controller in real time to realize intelligent dynamic gangue feeding.

[0012] Step S6: Construct a self-learning mechanism based on feedback bias self-correction to improve the prediction accuracy of the prediction model and the decision quality of the multi-objective intelligent optimization model.

[0013] Preferably, the underground intelligent sensing network includes conveyor belt pressure gauges, gangue bin level gauges, equipment operation status sensors, and filling face monitoring sensors. The conveyor belt pressure gauges are used to measure the output of multi-source gangue underground, the gangue bin level gauges are used to measure the material level in each gangue bin, and the equipment operation status sensors are used to collect the current, frequency, and start / stop parameters of the feeder, conveyor, and coal mining machine. The sensor data is cleaned, timestamp aligned, and feature-level fused by edge computing nodes deployed in the underground roadway to form a unified data stream.

[0014] Preferably, the underground multi-source gangue includes gangue from rock tunnel excavation and gangue from underground washing and processing; the output of gangue from rock tunnel excavation includes the amount of gangue G1 generated by the shield tunneling machine, the amount of gangue G2 generated by the fully mechanized tunneling, the amount of gangue G3 generated by the blasting tunneling, and the amount of gangue G4 generated by the repair; the output of gangue from underground washing and processing is calculated by the gangue content of the coal seam, the amount of coal coming from the coal mining face M1, the amount of coal coming from the coal / semi-coal rock tunnel M2, and the amount of gangue screened and processed s.

[0015] Preferably, in step S2, the digital twin model includes a gangue generation sub-model, a storage sub-model, a conveying sub-model, and a filling working face sub-model. The gangue generation sub-model simulates the generation rate of gangue and washed gangue in the rock tunnel based on the tunneling footage and washing operation rate. The storage sub-model updates the material level and inlet / outlet flow rate of each gangue bin in real time. The conveying sub-model reflects the relationship between the conveying capacity, belt speed, and energy consumption of the belt conveyor. The filling working face sub-model calculates the required amount of filling gangue and the filling rate based on the coal mining machine cutting depth, traction speed, and designed filling-mining mass ratio. The gangue generation sub-model, storage sub-model, conveying sub-model, and filling working face sub-model achieve real-time bidirectional mapping and updates with physical entities through the OPC UA protocol, forming a virtual-real consistent dynamic operation system for full-process strategy virtual pre-simulation.

[0016] Preferably, the prediction model is constructed using methods such as time series regression or long short-term memory neural networks. Its inputs are the historical coal mining volume sequence, the working face advance speed sequence, and the current real-time raw coal production q provided by the digital twin model. The output is the predicted raw coal production value qʹ within the future time window. The predicted gangue demand value Q0ʹ is obtained by multiplying the designed filling-mining quality ratio γʹ by the predicted raw coal production qʹ, i.e.: .

[0017] Preferably, in step S4, the virtual simulation pre-run is as follows: the digital twin model, based on the predicted value of gangue demand, performs parallel simulations of various combinations of feeding frequencies of each gangue bin feeder and operating parameters of the conveying system, following the dynamic allocation rules of gangue. It then comprehensively evaluates the performance of each combination under three objectives: storage utilization rate, conveying system energy consumption, and filling density. From these, the parameter combination that achieves the optimal overall multi-objective outcome is selected as the optimal strategy.

[0018] The multi-objective intelligent optimization model aims to optimize storage utilization, minimize conveying system energy consumption, and achieve optimal filling continuity. Storage utilization is characterized by the deviation function of each bin's material level from its safety upper and lower limits; conveying energy consumption is calculated based on a model of conveyor belt speed, carrying capacity, and power; filling continuity is measured by penalties for filling rate fluctuations and interruptions. Constraints include the frequency range of each feeder, conveyor belt speed limits, and the balance of material inflow and outflow from the gangue bins. In the digital twin model, parallel simulations are performed for each alternative feeding strategy, and the optimal strategy with the highest overall benefit score is obtained based on weighted summation or Pareto optimization.

[0019] Preferably, in step S4, the dynamic allocation rules for gangue include:

[0020] Prioritize consuming the gangue in the tunneling gangue bin. When the storage capacity of the tunneling gangue bin is Q... G When the amount of gangue exceeds the predicted demand value Q0ʹ, control the tunneling gangue feeder to supply material only, and stop feeding the washing gangue feeder; when the tunneling gangue storage capacity Q G When the amount of gangue is less than the predicted demand value Q0ʹ, the tunneling gangue feeder supplies material at its rated capacity, and the remaining demand is supplemented by the washing gangue feeder. Based on the dynamic deviation between the total feed amount and the predicted gangue demand value Q0ʹ, the feeding frequency of each feeder is adjusted to make the total feed amount approach the predicted gangue demand value Q0ʹ.

[0021] The condition for the washing gangue feeder to supplement the remaining demand is: the washing gangue storage capacity QX meets the supplementary demand for the remaining demand; if QX is insufficient, an alarm is triggered and priority is given to ensuring the supply of gangue for tunneling, and the insufficient amount is supplemented in subsequent cycles.

[0022] Preferably, the dynamic allocation rules for gangue also include: the gangue discharged by the tunneling gangue feeder must be screened by a screening machine, the oversize material is collected into the underground screening system, and the undersize material is transported to the filling face by a belt conveyor.

[0023] Preferably, when the level gauge of any gangue silo reaches a preset full-silo alarm threshold, the digital twin model, based on the prediction of future storage conditions, generates a decision in advance to transfer some of the gangue in the silo to the ground for temporary storage, and calculates the amount of gangue to be transferred based on the future time window gangue production and storage capacity output by the prediction model; based on the decision and the amount of gangue to be transferred, a control command is generated and issued to adjust the corresponding source gangue transportation path to execute the transfer.

[0024] Preferably, the self-learning mechanism compares the residuals between the physically measured filling effect data and the virtual theoretical standard value calculated by the digital twin model online, constructs a reward function, and iteratively optimizes the prediction model and the multi-objective intelligent optimization model online through reinforcement learning. The key parameters of the prediction model and the device execution benchmark parameters in the optimization model are used as reinforcement learning actions, and the model is updated based on residual feedback, thereby gradually improving prediction accuracy and decision quality.

[0025] The filling effect data includes the actual filling volume and the compaction of the filling body. The virtual theoretical standard value is obtained by the digital twin model running virtually under the same working conditions.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] By using a digital twin model to accurately mirror the entire process of multi-source gangue generation, storage, transportation, and filling, and combining a predictive model with the mining-to-filling quality ratio, future gangue demand is calculated in real time. Through multi-objective optimization and parallel simulation in virtual space, the optimal supply strategy is quickly derived and used for real-time control. Finally, through a self-learning mechanism, continuous adaptive optimization is achieved, realizing long-term accurate matching between multi-source gangue supply and filling demand. This significantly reduces human intervention, avoids gangue overflow or shortage, and ensures safe and efficient filling operations. Attached Figure Description

[0028] Figure 1 This is a flowchart of an intelligent distribution process for multi-source gangue solid filling in underground mines based on digital twins.

[0029] Figure 2 This is a logic diagram for the dynamic control of gangue. Detailed Implementation

[0030] The invention will now be further described with reference to the accompanying drawings.

[0031] like Figure 1 , Figure 2As shown in the figure. This embodiment takes a solid backfilling mining face in a certain mine as the implementation object. The designed mining-backfilling mass ratio γʹ of this working face is 1.25, that is, 1.25 tons of gangue need to be backfilled for every 1 ton of raw coal mined.

[0032] Step S1: Data Acquisition and Sensing

[0033] Underground, conveyor belt pressure gauges are deployed on the conveyor belts used for rock tunnel excavation and the waste rock washing and discharging belts; waste rock bin level gauges, equipment operation status sensors, and backfilling face monitoring sensors are also installed. These sensors are connected via industrial Ethernet to an edge computing node located in the mining area substation. The edge computing node performs protocol conversion, outlier removal, timestamp alignment, and feature extraction on the data, providing real-time data on the raw coal production (q) at the backfilling face, the instantaneous production of waste rock from various sources, the material levels in each waste rock bin, and the operating status of the feeder, conveyor, and coal mining machine.

[0034] Step S2: Digital Twin Model Construction

[0035] Edge computing nodes upload the fused data to the downhole monitoring center server, using platforms such as Unity or AnsysTwin Builder to build an integrated digital twin model of production, storage, transportation, and filling. The model includes sub-models for gangue production, storage, transportation, and filling face. Each model is driven by real-time data and undergoes bidirectional calibration updates with the physical entity every second to achieve consistency between the virtual and physical models.

[0036] Step S3: Gangue Demand Forecast

[0037] A prediction model based on LSTM was trained using historical coal mining volume and face advance speed data from the past 30 days. During online operation, the model is input as the current shift's mined coal volume sequence and advance speed to predict the raw coal production qʹ for the next 15 minutes. Based on the designed filling-to-mining ratio γʹ = 1.25, the predicted gangue demand is calculated. .

[0038] If the predicted value of qʹ is 100 tons, then the value of Q0ʹ is 125 tons.

[0039] Step S4: Multi-objective optimization and virtual pre-simulation

[0040] A multi-objective optimization model is constructed, with the objective functions being optimal warehouse utilization, minimum conveying energy consumption, and optimal filling density. Constraints include that the feed rate to each warehouse equals the production rate, and the total output rate equals the demand Q0ʹ.

[0041] The generation of alternative allocation strategies follows the following dynamic allocation rules for gangue:

[0042] Priority is given to consuming the gangue in the tunneling gangue bin. When the storage capacity of the tunneling gangue bin QG is greater than Q0ʹ, the tunneling gangue feeder supplies material alone, and the washing gangue feeder stops feeding. When QG is less than Q0ʹ, the tunneling gangue feeder supplies material at its rated capacity, and the remaining material is supplemented by the washing gangue feeder. If the storage capacity of the washing gangue bin QX is insufficient, an alarm is triggered, and priority is given to ensuring the supply of tunneling gangue. The insufficient amount is replenished in the next cycle.

[0043] The gangue discharged by the tunneling gangue feeder must be screened by a screening machine. The oversize material is collected into the underground screening system, and the undersize material is transported to the filling face by a belt conveyor.

[0044] Based on the digital twin environment, multiple sets of allocation strategies are generated in parallel in the background. The NSGA-II algorithm is used for virtual simulation and pre-run. Each simulation step is 1 minute, and the rolling optimization window is 15 minutes. The Pareto front is obtained and the comprehensive optimal strategy is selected according to subjective preferences.

[0045] Step S5: Issuance of control commands

[0046] The optimal strategy is interpreted into specific control parameters and sent to the corresponding frequency converter and PLC via OPC UA. The system performs rolling optimization and instruction update every 5 minutes to achieve dynamic feeding. During the feeding process, the actual total feed rate is monitored in real time, and the feeding frequency of each feeder is adjusted based on its dynamic deviation from Q0ʹ.

[0047] Furthermore, when the level gauge of any gangue silo reaches the preset full-silo alarm threshold, the digital twin model, based on the prediction of future storage conditions, generates a decision in advance to transfer some of the gangue in the silo to the ground for temporary storage. Based on the gangue production and storage capacity in the future time window output by the prediction model, the amount of gangue to be transferred is calculated. Based on the decision and the amount of gangue to be transferred, a control command is generated and issued to adjust the corresponding source gangue transportation path to execute the transfer, thus upgrading from passive alarm handling to proactive predictive transfer.

[0048] Step S6: Self-learning iterative optimization

[0049] During the filling process, filling effect data is measured using filling density sensors and inventory changes. A digital twin model synchronously calculates a virtual theoretical standard value and calculates residuals online. For example, if the actual filling rate is lower than the theoretical value, leading to a rise in material level, the residual is positive. Using this as a state, the reinforcement learning agent outputs actions to adjust the LSTM forget gate bias parameters in the prediction model and the material gradation correction coefficient in the optimization model. The reward function is designed to be a negative value of the absolute residual, and a deep Q-network algorithm is used for online iterative learning, allowing the prediction and optimization models to gradually adapt to equipment wear and geological changes, maintaining long-term high-precision operation.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital twin-based intelligent feeding method for multi-source gangue solid filling in wells, characterized in that, Includes the following steps: Step S1: Real-time data on raw coal production q at the filling face and production and storage status of multi-source gangue in the mine are obtained through the underground intelligent sensing network, and the operating status parameters of the feeder, conveyor and coal mining machine are collected simultaneously. Step S2: Perform edge computing fusion processing on the multi-source heterogeneous data collected by the downhole intelligent sensing network to construct a digital twin model integrating downhole multi-source gangue production, storage, transportation and charging; Step S3: Based on the historical production data and real-time operation data output by the digital twin model, combined with the designed filling-mining quality ratio γʹ, the prediction model is used to predict the demand for gangue in the filling face within the future time window, and the predicted gangue demand value Q0ʹ is output. Step S4: Construct a multi-objective intelligent optimization model with the goals of optimal warehouse utilization, minimum energy consumption of the conveying system, and optimal filling density. Within the digital twin model, conduct virtual simulations in parallel on multiple alternative allocation strategies that follow preset dynamic allocation rules for gangue, and evaluate to obtain the optimal strategy. Step S5: Convert the optimal strategy into control commands for the feeding frequency and conveyor belt speed of each gangue bin feeder, and send them to the corresponding equipment controller in real time to realize intelligent dynamic gangue feeding. Step S6: Construct a self-learning mechanism based on feedback bias self-correction to improve the prediction accuracy of the prediction model and the decision quality of the multi-objective intelligent optimization model.

2. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, In step S1, the underground intelligent sensing network includes a conveyor belt pressure gauge, a gangue bin level gauge, an equipment operation status sensor, and a filling face monitoring sensor; data preprocessing and fusion are performed through edge computing nodes; the conveyor belt pressure gauge is used to measure the underground multi-source gangue production, and the gangue bin level gauge is used to measure the underground multi-source gangue storage status data.

3. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, In step S1, the underground multi-source gangue includes gangue from rock tunnel excavation and gangue from underground washing and processing; the output of gangue from rock tunnel excavation includes the amount of gangue G1 generated by the shield tunneling machine, the amount of gangue G2 generated by the fully mechanized tunneling, the amount of gangue G3 generated by the blasting tunneling, and the amount of gangue G4 generated by the repair; the output of gangue from underground washing and processing is calculated by the gangue content of the coal seam, the amount of coal coming from the coal mining face M1, the amount of coal coming from the coal / semi-coal rock tunnel M2, and the amount of gangue screened and processed s.

4. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, In step S2, the digital twin model includes a gangue generation sub-model, a storage sub-model, a conveying sub-model, and a filling working face sub-model; the gangue generation sub-model simulates the generation rate of gangue and washed gangue in the rock tunnel based on the tunneling progress and the washing and beneficiation operation rate; the storage sub-model updates the material level and inlet / outlet flow rate of each gangue bin in real time; The conveying sub-model reflects the relationship between the conveying capacity, belt speed, and energy consumption of the belt conveyor; the filling working face sub-model calculates the required amount of backfill gangue and the filling rate based on the coal mining machine's cutting depth, traction speed, and designed backfilling-mining mass ratio; the gangue generation sub-model, storage sub-model, conveying sub-model, and filling working face sub-model achieve real-time bidirectional mapping and updates with physical entities through the OPC UA protocol, forming a virtual-real consistent dynamic operation system for full-process strategy virtual pre-simulation.

5. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, In step S3, the prediction model predicts the raw coal output within a future time window based on historical coal mining volume, working face advance speed, and current real-time raw coal production, outputting a predicted raw coal output value qʹ, which is used to output the predicted gangue demand value for the future time window; the predicted gangue demand value is calculated using the following formula: .

6. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, In step S4, the virtual simulation pre-run is as follows: the digital twin model, based on the predicted value of gangue demand, performs parallel simulations on various combinations of feeding frequency of each gangue bin feeder and operation parameters of the conveying system, which follow the dynamic allocation rules of gangue. It also comprehensively evaluates the performance of each combination under the three objectives of storage utilization rate, energy consumption of the conveying system and filling density, and selects the parameter combination that makes the multi-objective comprehensive optimal as the optimal strategy.

7. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, In step S4, the dynamic allocation rules for gangue include: Prioritize consuming the gangue in the tunneling gangue bin. When the storage capacity of the tunneling gangue bin is Q... G When the amount of gangue exceeds the predicted demand value Q0ʹ, control the tunneling gangue feeder to supply material only, and stop feeding the washing gangue feeder; when the tunneling gangue storage capacity Q G When the amount of gangue is less than the predicted demand value Q0ʹ, the tunneling gangue feeder supplies material at its rated capacity, and the remaining demand is supplemented by the washing gangue feeder. Based on the dynamic deviation between the total feed amount and the predicted gangue demand value Q0ʹ, the feeding frequency of each feeder is adjusted to make the total feed amount approach the predicted gangue demand value Q0ʹ. The condition for the washing gangue feeder to supplement the remaining demand is: the washing gangue storage capacity QX meets the supplementary demand for the remaining demand; if QX is insufficient, an alarm is triggered and priority is given to ensuring the supply of gangue for tunneling, and the insufficient amount is supplemented in subsequent cycles.

8. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 7, characterized in that, The dynamic allocation rules for gangue also include: the gangue discharged by the tunneling gangue feeder must be screened by a screening machine, the oversize material is collected into the underground screening system, and the undersize material is transported to the filling face by a belt conveyor.

9. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, When the level gauge of any gangue silo reaches the preset full silo alarm threshold, the digital twin model generates a decision to transfer some of the gangue in the silo to the ground for temporary storage based on the prediction of the future storage status. The model also calculates the amount of gangue to be transferred based on the gangue production and storage balance in the future time window output by the prediction model. Based on the decision and the amount of gangue to be transferred, control commands are generated and issued to adjust the corresponding source gangue transport path in order to execute the transfer.

10. The intelligent feeding method for downhole multi-source gangue solid filling based on digital twin as described in claim 1, characterized in that, The self-learning mechanism compares the residuals between the physical side's measured filling effect data and the virtual theoretical standard value calculated by the digital twin model online, and performs online iterative optimization of the prediction model and the multi-objective intelligent optimization model through reinforcement learning; The filling effect data includes the actual filling volume and the compaction of the filling body. The virtual theoretical standard value is obtained by the digital twin model running virtually under the same working conditions.

Images