Coal storage and transportation intelligent management method and system

Abstract

The invention discloses a coal storage and transportation intelligent management method and system, relates to the technical field of intelligent management systems, and aims to solve the problems of lack of environment monitoring, scheduling decision lag, extensive safety response and serious data island in the prior art. The method comprises the following steps: collecting the physical state, environmental parameters and operation behavior data of a coal storage yard in real time through a distributed sensor network; constructing a three-dimensional dynamic digital twinborn model and inputting the model into a time sequence neural network to predict the risk of spontaneous combustion, collapse or dust explosion in the future; and generating a dynamic scheduling instruction in combination with the task priority and the equipment state, and triggering a multi-level security response mechanism. The system comprises a sensor network module, a data fusion modeling module, a risk prediction module, a scheduling optimization module, and a closed-loop control and safety response module. According to the scheme, safety early warning preposition, scheduling decision intelligent optimization and resource energy efficiency collaborative improvement in the storage and transportation process are achieved, and the response capability and operation reliability of the coal supply chain are remarkably enhanced.

Description

Technical Field

[0001] This invention belongs to the field of intelligent management system technology, specifically relating to an intelligent management method and system for coal storage and transportation. Background Technology

[0002] With the continued widespread use of coal as a basic energy source in heavy industries such as power, metallurgy, and chemicals, intelligent management of its storage and transportation has become a core issue for ensuring supply chain stability and operational safety. Traditional coal storage and transportation systems rely on manual inspections and static scheduling models. Their management logic is based on preset environmental parameters and fixed operating procedures, making it difficult to cope with the dynamically changing physical and chemical properties during storage and transportation. Coal is susceptible to temperature and humidity fluctuations, dust accumulation, and the risk of spontaneous combustion during storage and transshipment. Existing systems generally lack the ability to integrate multi-source environmental sensor data in real time, leading to delayed early warnings, rigid scheduling, and consequently, safety accidents such as localized heat accumulation and dust explosions. At the same time, transportation routes and stockpile allocation strategies are mostly based on historical experience or simple rules, failing to dynamically optimize according to changes in coal quality, equipment status, and weather conditions, resulting in low loading and unloading efficiency, resource misallocation, and increased hidden losses.

[0003] Among them, intelligent coal storage and transportation management, as a typical application scenario of the cross-integration of industrial IoT and intelligent decision-making systems, aims to build a closed-loop control architecture covering the perception, analysis, and execution layers. This architecture needs to have millisecond-level acquisition capabilities for multi-dimensional data such as temperature and humidity, gas concentration, coal pile morphology, and equipment operating conditions at the storage and transportation site, and to achieve local risk prediction and emergency response through edge computing nodes. However, while current mainstream warehouse management systems are mature in general cargo sorting and inventory optimization, they have not designed dedicated perception models and control protocols for coal's unique high volatility, strong hygroscopicity, and spontaneous combustion tendency. This results in a lack of causal relationship modeling between environmental risk factors and scheduling instructions, and safety threshold settings become merely a formality.

[0004] In existing technologies, some systems attempt to introduce IoT sensor networks for basic environmental monitoring, but their data processing remains at the threshold alarm level, failing to establish a collaborative mechanism between coal pile state evolution prediction models and multi-objective scheduling algorithms. For example, when dust concentration exceeds safety limits, the system can only trigger local spray devices, unable to coordinate adjustments to the stacker-reclaimer's operating path or activate negative pressure dust removal systems for area isolation. When coal quality degrades due to long-term storage, the scheduling module still allocates transport vehicles based on initial coal type parameters, resulting in energy efficiency losses for end users. More critically, existing solutions generally neglect redundant control design in high-risk scenarios, such as failing to deploy distributed gas concentration gradient analysis modules in flammable areas or establishing dynamic restricted area delineation rules based on risk probability, resulting in a lack of preemptive prevention capabilities for sudden spontaneous combustion or explosion accidents. These deficiencies are particularly prominent in extreme climates or ultra-large-scale storage and transportation hubs, urgently requiring the construction of an intelligent coal storage and transportation management system that integrates real-time perception, dynamic extrapolation, and flexible decision-making to achieve a dual breakthrough in precise control of safety boundaries and global optimization of resource scheduling. Summary of the Invention

[0005] This invention provides a method and system for intelligent coal storage and transportation management. Its purpose is to construct an intelligent management system covering the entire coal storage and transportation process, achieving multi-dimensional collaborative control of coal storage, transshipment, environmental monitoring, scheduling decisions, and safety early warning. The system deploys a distributed sensor network to collect physical state parameters, environmental parameters, and operational behavior data in real time along coal yards and transportation routes. Combined with dynamic modeling and multi-source data fusion analysis mechanisms, it generates intelligent scheduling instructions for optimizing storage and transportation efficiency and mitigating safety risks. This maximizes the utilization rate of storage and transportation resources, reduces operating costs, and enhances the overall responsiveness of the coal supply chain while ensuring operational safety.

[0006] According to one aspect of the present invention, a method for intelligent management of coal storage and transportation is provided, comprising:

[0007] Through a distributed sensor network deployed in coal storage yards and transportation corridors, physical state data, environmental monitoring data, and operational behavior data are collected in real time during the coal storage and transportation process. The physical state data includes coal pile height, density distribution, surface temperature gradient, and internal temperature rise rate. The environmental monitoring data includes air humidity, dust concentration, oxygen concentration, carbon monoxide concentration, wind speed, and wind direction. The operational behavior data includes stacker-reclaimer operating trajectory, belt conveyor start / stop status, vehicle entry / exit frequency, and loading / unloading operation duration.

[0008] The physical state data, environmental monitoring data, and operational behavior data are spatiotemporally aligned and multi-source heterogeneous data are fused to construct a three-dimensional dynamic digital twin model of coal storage and transportation status. The three-dimensional dynamic digital twin model records the evolution of physical properties of coal, fluctuations of environmental parameters, and correlations of equipment operational behavior in each spatial unit in a time series manner.

[0009] Based on the aforementioned three-dimensional dynamic digital twin model, a sliding time window mechanism is used to extract the state evolution features within five sampling periods before and after the current moment, and input them into a pre-trained storage and transportation risk prediction model. The storage and transportation risk prediction model is a temporal neural network constructed based on a gated recurrent unit and an attention mechanism, and its output is the probability value of spontaneous combustion, collapse or dust explosion of each spatial unit within the next three sampling periods.

[0010] Based on the risk probability value, combined with the current storage and transportation task priority, equipment availability status and transportation route congestion index, dynamic scheduling optimization instructions are generated. The dynamic scheduling optimization instructions include stacker-reclaimer operation area reallocation instructions, belt conveyor running speed adjustment instructions, transportation vehicle route replanning instructions, spray dust suppression system start / stop control instructions, and ventilation equipment air volume adjustment instructions.

[0011] The dynamic scheduling optimization command is sent to the corresponding execution mechanism, and the equipment status and environmental parameters in the three-dimensional dynamic digital twin model are updated synchronously to form a closed-loop feedback control.

[0012] When the risk probability value exceeds the preset safety threshold, a multi-level safety response mechanism is triggered. The multi-level safety response mechanism includes a first-level response: starting local area spray cooling and forced ventilation; a second-level response: suspending operations in high-risk areas and evacuating surrounding equipment; and a third-level response: cutting off the area's power supply and starting the inert gas injection system.

[0013] According to another aspect of the present invention, a coal storage and transportation intelligent management system is provided, comprising:

[0014] A distributed sensor network module is deployed in coal storage yards and transportation corridors to collect real-time physical state data, environmental monitoring data, and operational behavior data during coal storage and transportation. The physical state data includes coal pile height, density distribution, surface temperature gradient, and internal temperature rise rate. The environmental monitoring data includes air humidity, dust concentration, oxygen concentration, carbon monoxide concentration, wind speed, and wind direction. The operational behavior data includes stacker-reclaimer operating trajectory, belt conveyor start / stop status, vehicle entry / exit frequency, and loading / unloading operation duration.

[0015] The multi-source data fusion modeling module is used to perform spatiotemporal alignment and multi-source heterogeneous data fusion on the physical state data, environmental monitoring data and operation behavior data to construct a three-dimensional dynamic digital twin model of coal storage and transportation status. The three-dimensional dynamic digital twin model records the evolution of physical properties of coal, fluctuations of environmental parameters and correlation of equipment operation behavior in each spatial unit in a time series manner.

[0016] The risk prediction and analysis module is used to extract the state evolution features within five sampling periods before and after the current moment based on the three-dimensional dynamic digital twin model using a sliding time window mechanism, and input them into the pre-trained storage and transportation risk prediction model. The storage and transportation risk prediction model is a temporal neural network constructed based on a gated recurrent unit and an attention mechanism. Its output is the probability value of spontaneous combustion, collapse or dust explosion of each spatial unit within the next three sampling periods.

[0017] The dynamic scheduling optimization module is used to generate dynamic scheduling optimization instructions based on the risk probability value, combined with the current storage and transportation task priority, equipment availability status and transportation route congestion index. The dynamic scheduling optimization instructions include stacker-reclaimer operation area reallocation instructions, belt conveyor running speed adjustment instructions, transportation vehicle route replanning instructions, spray dust suppression system start and stop control instructions, and ventilation equipment air volume adjustment instructions.

[0018] The closed-loop control execution module is used to send the dynamic scheduling optimization command to the corresponding execution mechanism and synchronously update the equipment status and environmental parameters in the three-dimensional dynamic digital twin model to form a closed-loop feedback control.

[0019] The multi-level safety response module is used to trigger a multi-level safety response mechanism when the risk probability value exceeds a preset safety threshold. The multi-level safety response mechanism includes a first-level response: starting local area spray cooling and forced ventilation; a second-level response: suspending operations in high-risk areas and evacuating surrounding equipment; and a third-level response: cutting off the area's power supply and starting the inert gas injection system.

[0020] Furthermore, the distributed sensor network module includes fixed sensor nodes and mobile sensor nodes. The fixed sensor nodes are deployed at the boundaries of the stockyard, key structural points of the transfer station, and ventilation openings. They are encapsulated in explosion-proof shells and contain built-in infrared thermal imaging sensors, laser rangefinders, gas concentration sensors, and temperature and humidity sensors. The mobile sensor nodes are mounted on the end of the stacker-reclaimer boom and on the top of the transport vehicle. They collect dynamic operating area data as the equipment moves and communicate with the fixed nodes through a wireless self-organizing network protocol to achieve real-time data transmission and location synchronization.

[0021] Furthermore, in the multi-source data fusion modeling module, spatiotemporal alignment adopts a spatiotemporal reference system based on the fusion of BeiDou satellite positioning and inertial navigation, and all sensor data are mapped according to a unified timestamp and spatial coordinate system; multi-source heterogeneous data fusion adopts a joint modeling method of tensor decomposition and graph neural network, which maps physical state, environmental parameters and operation behavior into three modes of tensor respectively, extracts their coupling correlation features through cross-modal attention mechanism, and finally generates a three-dimensional gridded state representation. Each grid cell corresponds to a spatial coordinate point, and its attribute vector contains the comprehensive state description of the point at the current moment.

[0022] Furthermore, the training data for the storage and transportation risk prediction model comes from historical accident records and simulation datasets. The historical accident records include full-process monitoring data of spontaneous combustion, collapse, and dust explosion events that occurred in large-scale coal storage and transportation bases across the country in the past five years. The simulation dataset is generated jointly through computational fluid dynamics simulation and discrete element particle dynamics simulation, covering the risk evolution path under different stack structures, environmental conditions, and operational intensities. The model training adopts an adversarial generation mechanism to enhance sample diversity, and the loss function is weighted cross-entropy, with the weights dynamically adjusted according to the severity of the accident consequences.

[0023] Furthermore, the dynamic scheduling optimization module adopts a hierarchical decision-making architecture. The upper layer is a global task allocator based on constraint satisfaction problems. The inputs are all the storage and transportation tasks to be executed and their time window constraints, equipment capacity constraints, and safety constraints. The output is the basic operation plan of each device. The lower layer is a local adjuster based on reinforcement learning. It takes the risk probability value, path congestion index, and equipment load rate as state inputs and the scheduling instruction adjustment amount as action outputs. The reward function comprehensively considers task completion efficiency, energy consumption cost, and risk suppression effect, and performs policy iteration through a near-end policy optimization algorithm.

[0024] Furthermore, in the closed-loop control execution module, the actuators include a programmable logic controller, a frequency converter, an electric valve, and a gas release device. All execution commands are transmitted via an industrial Ethernet bus, and the command format follows a unified equipment control protocol, including the target equipment identifier, control parameters, execution time limit, and security check code. The execution result is transmitted back in real time through the equipment status feedback interface. If no confirmation signal is received within a preset time limit, the command retransmission mechanism is triggered. After three failed retransmissions, the system switches to manual intervention mode.

[0025] Furthermore, in the multi-level safety response module, the safety thresholds are dynamically set according to the risk type and historical accident frequency of the space unit. The initial value of the spontaneous combustion risk threshold is 5%, which is increased by 0.5% for each near-miss event, with a maximum of 10%. The initial value of the collapse risk threshold is 3%, which is corrected in real time according to the height and density distribution of the stack. The dust explosion risk threshold is set according to the product of oxygen concentration and dust concentration, and a response is triggered when the product exceeds the critical value of milligrams per cubic meter multiplied by a percentage. Each level of response command has a mandatory priority and can interrupt all current non-safety-related scheduling commands.

[0026] Furthermore, the three-dimensional dynamic digital twin model supports a visual interactive interface, allowing operators to select any spatial unit via a touchscreen to view its historical state evolution curve, current risk score, and associated equipment operation records. The model supports manually marking high-risk areas and locking scheduling strategies, while also providing a "sand table simulation" function, allowing input of hypothetical environmental disturbances or equipment failures to simulate the system response process and risk diffusion path, which can be used for emergency plan drills and pre-verification of scheduling strategies.

[0027] Furthermore, the system also includes an energy efficiency optimization submodule, which, under the premise of controllable risk, dynamically adjusts the storage strategy and transfer priority based on the coal calorific value decay model and the transportation distance cost model, so that high-calorific-value coal is transferred first and low-calorific-value coal is stored first in well-ventilated areas; at the same time, it adjusts the operation plan of large equipment according to the peak and off-peak electricity price periods to avoid peak electricity consumption and reduce the overall energy consumption cost.

[0028] Furthermore, the system also includes a data archiving and knowledge accumulation module, which records each scheduling decision, risk response, equipment action and final result as a structured event log, extracts effective decision rules through a causal reasoning engine, and forms a reusable scheduling strategy template and risk response plan library to support strategy migration and adaptive optimization in new scenarios.

[0029] The beneficial effects of this invention are:

[0030] This invention presents an intelligent coal storage and transportation management method and system that, for the first time, achieves three-dimensional spatiotemporal synchronous modeling of physical state, environmental parameters, and operational behavior, overcoming the limitations of traditional systems that rely solely on single-dimensional data for static scheduling. By introducing a risk prediction model based on gated cyclic units and attention mechanisms, the system can identify potential risks of spontaneous combustion, collapse, and dust explosion three sampling cycles in advance, securing a critical time window for safety response. The dynamic scheduling optimization module employs a hierarchical reinforcement learning architecture, achieving Pareto optimality in storage and transportation efficiency and energy consumption costs while meeting safety constraints. A multi-level safety response mechanism triggers in stages according to risk type and severity, avoiding the resource waste caused by the "one-size-fits-all" shutdown of traditional systems. A closed-loop control execution module ensures precise command implementation, while a data archiving and knowledge accumulation module enables the system to continuously evolve, with scheduling strategies constantly optimized over time. The overall solution addresses core issues in existing technologies such as lack of environmental monitoring, delayed scheduling decisions, coarse safety responses, and severe data silos, significantly improving the safety, efficiency, and intelligence level of coal storage and transportation processes. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall technical architecture of a coal storage and transportation intelligent management method and system proposed in this invention. Detailed Implementation

[0032] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0033] This invention provides a method and system for intelligent management of coal storage and transportation. Its core lies in constructing an intelligent management system covering the entire coal storage and transportation process, achieving multi-dimensional collaborative control of coal storage, transshipment, environmental monitoring, scheduling decisions, and safety early warning. This system deploys a distributed sensor network to collect physical state parameters, environmental parameters, and operational behavior data in real time along coal storage yards and transportation routes. Combined with dynamic modeling and multi-source data fusion analysis mechanisms, it generates intelligent scheduling instructions oriented towards optimizing storage and transportation efficiency and mitigating safety risks. This maximizes the utilization rate of storage and transportation resources, reduces operating costs, and enhances the overall responsiveness of the coal supply chain while ensuring operational safety.

[0034] In this embodiment, the intelligent coal storage and transportation management method includes the following steps: Real-time collection of physical state data, environmental monitoring data, and operational behavior data during coal storage and transportation via a distributed sensor network deployed in coal yards and transportation channels; spatiotemporal alignment and multi-source heterogeneous data fusion of the physical state data, environmental monitoring data, and operational behavior data to construct a three-dimensional dynamic digital twin model of the coal storage and transportation status; based on the three-dimensional dynamic digital twin model, a sliding time window mechanism is used to extract the state evolution features within five sampling periods before and after the current moment, inputting them into a pre-trained storage and transportation risk prediction model, outputting the probability values ​​of spontaneous combustion, collapse, or dust explosion occurring in each spatial unit within the next three sampling periods; generating dynamic scheduling optimization instructions based on the risk probability values, combined with the current storage and transportation task priority, equipment availability, and transportation path congestion index; issuing the dynamic scheduling optimization instructions to the corresponding execution mechanism and synchronously updating the equipment status and environmental parameters in the three-dimensional dynamic digital twin model to form a closed-loop feedback control; triggering a multi-level safety response mechanism when the risk probability value exceeds a preset safety threshold.

[0035] In the aforementioned steps, the distributed sensor network consists of both fixed and mobile sensor nodes. Fixed sensor nodes are deployed at the boundaries of the stockpile, key structural points of the transfer station, and ventilation openings. They are encased in explosion-proof housings and contain built-in infrared thermal imaging sensors, laser rangefinders, gas concentration sensors, and temperature and humidity sensors. The infrared thermal imaging sensor is used for non-contact measurement of the surface temperature distribution of the coal pile. Its measurement range covers -20°C to 500°C, with a spatial resolution of 0.5 meters and a frame rate of 10 frames per second, ensuring high sensitivity in capturing temperature gradient changes. The laser rangefinder is used to scan the stockpile outline in real time, calculating the height and volume changes of the stockpile using triangulation. Its measurement accuracy is ±5 millimeters, and its scanning frequency is 20 times per second, dynamically tracking the evolution of the stockpile morphology caused by the stacker-reclaimer operation. The gas concentration sensor monitors the concentration of dust, oxygen, and carbon monoxide in the air, with detection ranges of 0 to 2000 mg / m³, 0% to 25% of ppm, and 0 to 500 ppm, respectively. Its response time is less than two seconds, and it features automatic calibration and drift compensation functions. The temperature and humidity sensor is used to collect ambient temperature and humidity data. The measurement range is from -40 degrees Celsius to 80 degrees Celsius and from 0% to 100% relative humidity. The accuracy is ±0.5 degrees Celsius and ±3% relative humidity. It supports long-term outdoor deployment.

[0036] Mobile sensing nodes are mounted on the end of the stacker-reclaimer boom and on top of transport vehicles, collecting dynamic data about the operating area as the equipment moves. Each mobile node integrates an inertial measurement unit (IMU), a GPS receiver, a wireless communication module, and an edge computing unit. The IMU senses the equipment's attitude angles and acceleration in real time to assist in positioning corrections. The GPS receiver uses dual-frequency BeiDou-3 signals, achieving positioning accuracy better than one meter, and combines inertial navigation with continuous trajectory calculation even in obstructed environments. The wireless communication module supports self-organizing network protocols, establishing multi-hop communication links with fixed nodes to ensure reliable data transmission in complex stockpile environments. The edge computing unit performs local preprocessing of the raw sensor data, including noise filtering, outlier removal, and timestamp synchronization, reducing the load on the backbone network. Each mobile node collects data once per second and uploads it to the central data aggregation node via wireless link at a rate of 50 kilobits per second, ensuring the real-time nature and integrity of the operational data.

[0037] All data collected by the sensor nodes are aligned to a unified spatiotemporal reference. The time reference is synchronized using a network time protocol, with clock errors for all nodes controlled within ten milliseconds. The spatial reference adopts a three-dimensional Cartesian coordinate system based on the geographic coordinates of the storage yard, with the origin set at the ground reference point in the southwest corner of the storage yard, the X-axis pointing east, the Y-axis pointing north, and the Z-axis pointing vertically upward. The spatial coordinates of fixed nodes are precisely measured and stored using a total station during deployment; the spatial coordinates of mobile nodes are calculated by the fusion of BeiDou positioning and inertial navigation, outputting position and attitude information once per second. All sensor data is immediately tagged with timestamps and spatial coordinate labels after acquisition, forming a raw data stream with spatiotemporal tags, laying the foundation for subsequent multi-source data fusion.

[0038] After completing data acquisition and spatiotemporal alignment, the process moves to the multi-source heterogeneous data fusion stage. The goal of this stage is to fuse physical state data, environmental monitoring data, and operational behavior data into a unified three-dimensional dynamic digital twin model. The fusion process employs a joint modeling method combining tensor decomposition and graph neural networks. First, the stockpile space is divided into cubic grid cells with sides of five meters, each grid cell corresponding to a spatial coordinate point. For each sampling moment, all sensor data falling within that grid cell are aggregated to form the initial state vector of that cell at that moment. The initial state vector includes thirteen dimensions, such as stockpile height, estimated density, average surface temperature, estimated internal temperature rise rate, air humidity, dust concentration, oxygen concentration, carbon monoxide concentration, wind speed, wind direction, whether the stacker-reclaimer is operating in the cell, whether the belt conveyor is running, and whether vehicles have passed through.

[0039] Subsequently, the state vectors of all grid cells over continuous time series are organized into a three-dimensional tensor, with its three modes corresponding to the spatial, temporal, and feature dimensions, respectively. High-order singular value decomposition (SVD) is performed on this tensor to extract its core tensor and factor matrix, achieving data compression and denoising. The core tensor retains the main change patterns of the original data, while the factor matrices represent spatial distribution characteristics, temporal evolution characteristics, and feature coupling characteristics, respectively. In the decomposed low-dimensional space, a graph neural network is introduced for cross-modal correlation modeling. Each grid cell is considered a node in a graph, and connections between nodes are constructed based on spatial proximity and operational correlation. Spatial proximity is defined as the existence of edges between grid cells with an Euclidean distance of less than ten meters; operational correlation is defined as the existence of edges between grid cells visited by the same reclaimer or transport vehicle in adjacent time steps. The graph neural network employs a multi-layer graph convolutional structure, with each layer updating the current node representation by aggregating neighbor node information, ultimately outputting a comprehensive state encoding vector for each grid cell at the current time. This encoding vector has 128 dimensions and contains coupled information of physical state, environmental parameters, and operational behavior, serving as the basic unit of the three-dimensional dynamic digital twin model.

[0040] The 3D dynamic digital twin model is stored in a time-series database, supporting efficient queries by time and spatial range. The model updates every minute, with each update including the fusion results of data from all sampling periods within the past minute. The model supports a visual interactive interface, allowing operators to select any spatial unit via a touchscreen to view its historical state evolution curves, current risk score, and associated equipment operation records. The model also supports a "sand table simulation" function, allowing input of hypothetical environmental disturbances such as sudden strong winds or equipment failures such as stacker-reclaimer shutdowns, to simulate the system response process and risk diffusion paths, used for emergency response drills and pre-verification of scheduling strategies.

[0041] Based on the construction of a three-dimensional dynamic digital twin model, storage and transportation risks are predicted. The prediction model is a temporal neural network built on gated recurrent units and an attention mechanism. The input is the state evolution features extracted by a sliding time window for five sampling periods before and after the current moment, i.e., the comprehensive state encoding vector sequence of each grid cell over the past ten sampling periods. This sequence is first subjected to temporal feature extraction through a bidirectional gated recurrent unit. The forward gated recurrent unit captures the state evolution trend from the past to the present, and the backward gated recurrent unit captures the state backtracking features from the future to the present. The hidden layer dimension of the bidirectional gated recurrent unit is 256 dimensions, and the output is a context-aware state vector for each time step.

[0042] Subsequently, an attention mechanism is introduced to weighted aggregate the context-aware state vectors across ten time steps. The attention weights are calculated from the query vector, key vector, and value vector, where the query vector encodes the current state, and the key and value vectors encode the states at each historical time step. This attention mechanism allows the model to focus on the historical moment most informative for predicting the current risk, such as a sudden increase in the rate of temperature rise or a sharp rise in dust concentration. The weighted aggregated feature vector has a dimension of 256 and is input to a fully connected classification layer, outputting three risk probability values: spontaneous combustion probability, collapse probability, and dust explosion probability. The classification layer uses the Softmax activation function to ensure that the sum of the three probability values ​​is one.

[0043] The training data for the storage and transportation risk prediction model comes from historical accident records and simulation datasets. The historical accident records include full-process monitoring data of spontaneous combustion, collapse, and dust explosion events that occurred at large coal storage and transportation bases nationwide over the past five years, totaling 372 real accident samples. The simulation dataset was generated jointly through computational fluid dynamics simulation and discrete element particle dynamics simulation, covering risk evolution paths under different stack structures, environmental conditions, and operational intensities, generating a total of 12,000 simulation samples. The training employs an adversarial generation mechanism to enhance sample diversity, synthesizing boundary cases through an adversarial network, such as scenarios with slow temperature rise but extremely high dust concentrations. The loss function is weighted cross-entropy, with weights dynamically adjusted according to the severity of the accident consequences. The weight for spontaneous combustion accidents is set to 1.5, for collapse accidents to 1.2, and for dust explosion accidents to 2.0, to reflect their different safety impacts.

[0044] After the risk prediction results are output, the dynamic scheduling optimization phase begins. This phase employs a hierarchical decision-making architecture, with a global task allocator at the upper level based on constraint satisfaction and a local adjuster at the lower level based on reinforcement learning. The global task allocator takes as input all currently pending storage and transportation tasks, along with their time window constraints, equipment capacity constraints, and safety constraints. Its output is the basic work plan for each piece of equipment. Tasks include coal storage, reclaiming, and transfer, with each task specifying the type and quantity of coal, starting location, target location, and latest completion time. Equipment includes stacker-reclaimers, belt conveyors, and transport vehicles, each with its maximum operating capacity, current location, and current status. Safety constraints include prohibitions on operation in high-risk areas and minimum safe distances between equipment. The global task allocator uses a constraint programming solver to generate a basic work plan that satisfies all constraints, with the objective of minimizing the total operation time.

[0045] The local adjuster takes risk probability values, path congestion index, and equipment load rate as state inputs and dispatch command adjustments as action outputs. The state space is Nx3, where N is the total number of grid cells, and each cell corresponds to three risk probability values. The path congestion index is a function of vehicle density and average speed on each transport path. The equipment load rate is the ratio of current workload to maximum capacity for each piece of equipment. The action space includes stacker-reclaimer operating area offset, belt conveyor speed adjustment ratio, transport vehicle path replanning markers, sprinkler system start / stop commands, and ventilation equipment airflow adjustment ratio. The reward function comprehensively considers task completion efficiency, energy consumption cost, and risk mitigation effect, and is defined as a negative weighted sum: task delay penalty weight is 0.6, energy consumption cost weight is 0.3, and risk exposure penalty weight is 0.1. The local adjuster uses a proximal policy optimization algorithm for policy iteration. The policy network is a three-layer fully connected neural network with a hidden layer dimension of 512, outputting an action probability distribution. The value network structure is identical, outputting a state value estimate. During training, experience data is generated by interacting with a 3D dynamic digital twin model. Each simulation runs for 24 hours, collecting state-action-reward sequences for policy updates.

[0046] After the dynamic scheduling optimization instructions are generated, the closed-loop control execution phase begins. The actuators include programmable logic controllers (PLCs), frequency converters, electric valves, and gas release devices. All execution instructions are transmitted via an industrial Ethernet bus, and the instruction format follows a unified equipment control protocol. The protocol frame structure includes the target device identifier, control parameters, execution time limit, and security check code. The target device identifier is a 16-bit device code that uniquely identifies each actuator in the yard; control parameters vary depending on the instruction type, such as percentage values ​​for speed instructions and coordinate sequences for path instructions; the execution time limit is the latest time the instruction takes effect, and the instruction is invalidated if the timeout is exceeded; the security check code is a 32-bit cyclic redundancy check code used to detect transmission errors.

[0047] After the command is issued, the executing mechanism immediately executes it and transmits the execution results back in real time through the equipment status feedback interface. Feedback information includes the command execution status, actual parameter values, and execution timestamp. The central control system starts a timer after sending the command. If no confirmation signal is received within a preset time limit, the command retransmission mechanism is triggered. The retransmission interval is five seconds, with a maximum of three retransmissions. If all three retransmissions fail, the system switches to manual intervention mode, alarms, and locks the relevant equipment, awaiting manual confirmation and handling. Simultaneously, the execution results are used to update the equipment status and environmental parameters in the 3D dynamic digital twin model, such as updating the stacker-reclaimer position, belt speed, and spray area humidity, forming a closed-loop feedback to ensure the model's state is synchronized with the physical world.

[0048] When the risk probability value exceeds the preset safety threshold, a multi-level safety response mechanism is triggered. The safety thresholds are dynamically set based on the risk type of the spatial unit and the frequency of historical accidents. The initial value for the spontaneous combustion risk threshold is 5%, increasing by 0.5% for each near miss, up to a maximum of 10%. The initial value for the collapse risk threshold is 3%, adjusted in real time based on the pile height and density distribution. The adjustment formula is: threshold equals 3% multiplied by the pile height divided by the square root of 10 meters multiplied by the density divided by the square root of 0.8 tons per cubic meter. The dust explosion risk threshold is set based on the product of oxygen concentration and dust concentration. A response is triggered when the product exceeds a critical value of milligrams per cubic meter multiplied by a percentage. The initial critical value is set at 500, adjusted based on historical explosion accidents.

[0049] The multi-level safety response mechanism includes three levels of response. Level 1 involves activating localized sprinkler cooling and forced ventilation, suitable for situations where the risk probability just exceeds the threshold. The sprinkler system covers the high-risk grid cell and its adjacent cells, with the sprinkler intensity dynamically adjusted according to the temperature rise rate, ranging from 0.5 to 2 liters per square meter per minute. Forced ventilation activates the axial flow fans in the corresponding area, with the airflow set to 80% of the maximum value. Level 2 involves suspending operations in the high-risk area and evacuating surrounding equipment, suitable for situations where the risk probability continues to rise or exceeds the threshold by 50%. The system sends an emergency stop command to all equipment entering the high-risk area and guides it to evacuate along a preset safe path to a designated assembly point. Level 3 involves cutting off the area's power supply and activating the inert gas injection system, suitable for situations with extremely high risk probability or where open flames have been detected. Power is cut off via remotely controlled circuit breakers, and the inert gas injection system releases nitrogen into the high-risk area, reducing the oxygen concentration to below 12%. Each level of response command has mandatory priority and can interrupt all current non-safety-related dispatch commands, ensuring that safety measures are executed first.

[0050] In addition, the system includes an energy efficiency optimization submodule and a data archiving and knowledge accumulation module. Under controllable risk conditions, the energy efficiency optimization submodule dynamically adjusts storage strategies and transfer priorities based on a coal calorific value decay model and a transportation distance cost model. The calorific value decay model, based on coal type, storage time, and ambient temperature, predicts the calorific value loss rate under different storage conditions; the transportation distance cost model calculates the distance and energy consumption from each storage point to each transfer point. The optimization objective is to minimize the sum of total calorific value loss and transportation costs, constrained by equipment capacity and safety limitations. The optimization results prioritize the transfer of high-calorific-value coal and prioritize the storage of low-calorific-value coal in well-ventilated areas. Simultaneously, based on peak and off-peak electricity pricing periods, the system adjusts the operation plans of large equipment to avoid peak electricity consumption and reduce overall energy costs.

[0051] The data archiving and knowledge accumulation module records every scheduling decision, risk response, equipment action, and final result as a structured event log. Log fields include timestamp, event type, involved equipment, operating parameters, risk indicators, energy consumption data, and task status. A causal reasoning engine extracts effective decision rules, such as "when dust concentration exceeds 1000 mg / m³ and wind speed is less than 1 m / s, activating the sprinkler system can reduce the explosion risk by 40%." Rules are stored in a strategy template library and a contingency plan library, supporting strategy migration and adaptive optimization in new scenarios. The system periodically evaluates rule effectiveness, eliminates outdated rules, and adds new rules to achieve continuous evolution.

[0052] In summary, the intelligent coal storage and transportation management method described in this embodiment achieves full-element data collection through a distributed sensor network, constructs a three-dimensional dynamic digital twin model through multi-source data fusion, realizes proactive risk prediction through a temporal neural network, achieves dynamic scheduling optimization through hierarchical reinforcement learning, ensures command execution through closed-loop control, guarantees operational safety through multi-level response, reduces operating costs through energy efficiency optimization, and achieves system evolution through knowledge accumulation. This method solves the core problems in existing technologies, such as lack of environmental monitoring, delayed scheduling decisions, crude safety response, and severe data silos, significantly improving the safety, efficiency, and intelligence level of the coal storage and transportation process.

[0053] At the system level, this invention provides an intelligent management system for coal storage and transportation, comprising a distributed sensor network module, a multi-source data fusion modeling module, a risk prediction and analysis module, a dynamic scheduling optimization module, a closed-loop control execution module, and a multi-level safety response module. The distributed sensor network module consists of fixed and mobile sensor nodes, covering the entire stockpile area and enabling synchronous collection of physical states, environmental parameters, and operational behaviors. The multi-source data fusion modeling module employs a combined tensor decomposition and graph neural network approach to construct a three-dimensional dynamic digital twin model, supporting spatiotemporal queries and scenario simulations. The risk prediction and analysis module, based on gated loop units and attention mechanisms, outputs multi-risk probability predictions. The dynamic scheduling optimization module adopts a hierarchical architecture, combining global allocation with local adjustments to achieve a balance between efficiency and safety. The closed-loop control execution module uses industrial Ethernet and a unified protocol to ensure accurate command delivery and status synchronization. The multi-level safety response module triggers in stages based on dynamic thresholds to avoid over-response. The system also includes energy efficiency optimization and knowledge accumulation modules to achieve cost control and continuous evolution. All modules work collaboratively to form a complete intelligent management closed loop, providing a comprehensive, efficient, and safe intelligent solution for the coal storage and transportation industry.

[0054] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. A method for intelligent management of coal storage and transportation, characterized in that, include: Through a distributed sensor network deployed in coal storage yards and transportation corridors, physical state data, environmental monitoring data, and operational behavior data are collected in real time during the coal storage and transportation process. The physical state data includes coal pile height, density distribution, surface temperature gradient, and internal temperature rise rate. The environmental monitoring data includes air humidity, dust concentration, oxygen concentration, carbon monoxide concentration, wind speed, and wind direction. The operational behavior data includes stacker-reclaimer operating trajectory, belt conveyor start / stop status, vehicle entry / exit frequency, and loading / unloading operation duration. The physical state data, environmental monitoring data, and operational behavior data are spatiotemporally aligned and multi-source heterogeneous data are fused to construct a three-dimensional dynamic digital twin model of coal storage and transportation status. The three-dimensional dynamic digital twin model records the evolution of physical properties of coal, fluctuations of environmental parameters, and correlations of equipment operational behavior in each spatial unit in a time series manner. Based on the aforementioned three-dimensional dynamic digital twin model, a sliding time window mechanism is used to extract the state evolution features within five sampling periods before and after the current moment, and input them into a pre-trained storage and transportation risk prediction model. The storage and transportation risk prediction model is a temporal neural network constructed based on a gated recurrent unit and an attention mechanism, and its output is the probability value of spontaneous combustion, collapse or dust explosion of each spatial unit within the next three sampling periods. Based on the risk probability value, combined with the current storage and transportation task priority, equipment availability status and transportation route congestion index, dynamic scheduling optimization instructions are generated. The dynamic scheduling optimization instructions include stacker-reclaimer operation area reallocation instructions, belt conveyor running speed adjustment instructions, transportation vehicle route replanning instructions, spray dust suppression system start / stop control instructions, and ventilation equipment air volume adjustment instructions. The dynamic scheduling optimization command is sent to the corresponding execution mechanism, and the equipment status and environmental parameters in the three-dimensional dynamic digital twin model are updated synchronously to form a closed-loop feedback control. When the risk probability value exceeds a preset safety threshold, a multi-level safety response mechanism is triggered. The multi-level safety response mechanism includes a first-level response: initiating local area spray cooling and forced ventilation. Level 2 Response: Suspend operations in high-risk areas and evacuate surrounding equipment; Level 3 Response: Cut off the power supply to the area and activate the inert gas injection system.

2. The intelligent management method for coal storage and transportation according to claim 1, characterized in that, The physical state data, environmental monitoring data, and operational behavior data are spatiotemporally aligned and multi-source heterogeneous data are fused to construct a three-dimensional dynamic digital twin model of coal storage and transportation status, including: The storage yard space is divided into cubic grid cells with a side length of five meters. Each grid cell corresponds to a spatial coordinate point. All sensor data falling into the grid cell are aggregated to form the initial state vector of the cell at the current moment. The state vectors of all grid cells in the continuous time series are organized into a three-dimensional tensor, with its three modes corresponding to the spatial dimension, time dimension and feature dimension, respectively. The tensor is subjected to high-order singular value decomposition to extract the core tensor and factor matrix. In the decomposed low-dimensional space, a graph neural network is introduced to model cross-modal associations. Each grid cell is regarded as a node in the graph, and the connections between nodes are constructed based on spatial proximity and operational association. The neighbor node information is aggregated through a multi-layer graph convolution structure, and the comprehensive state encoding vector of each grid cell at the current time is output as the basic unit of the three-dimensional dynamic digital twin model.

3. The intelligent management method for coal storage and transportation according to claim 2, characterized in that, Based on the aforementioned three-dimensional dynamic digital twin model, a sliding time window mechanism is used to extract the state evolution features within five sampling periods before and after the current moment, which are then input into a pre-trained storage and transportation risk prediction model, including: The sequence of the comprehensive state encoding vector of each grid cell over the past ten sampling periods is input into the bidirectional gated recurrent unit. The forward gated recurrent unit captures the state evolution trend from the past to the present, and the backward gated recurrent unit captures the state backtracking features from the future to the present, and outputs the context-aware state vector at each time step. An attention mechanism is introduced to perform weighted aggregation of context-aware state vectors at ten time steps. The attention weights are calculated by using the current state code as the query vector and the state codes at each historical time step as the key and value vectors. The weighted aggregated feature vector is input into a fully connected classification layer, which outputs three risk probability values: spontaneous combustion probability, collapse probability, and dust explosion probability. The classification layer uses the Softmax activation function to ensure that the sum of the three probability values ​​is one.

4. The intelligent management method for coal storage and transportation according to claim 3, characterized in that, Based on the aforementioned risk probability value, combined with the current storage and transportation task priority, equipment availability, and transportation route congestion index, dynamic scheduling optimization instructions are generated, including: The upper-level global task allocator takes all pending storage and transportation tasks, their time window constraints, equipment capacity constraints, and safety constraints as input, and uses a constraint programming solver to generate basic operation plans for each device. The lower-level local adjuster takes risk probability values, path congestion index, and equipment load rate as state input, and takes scheduling instruction adjustment amounts as action outputs. It uses a near-end policy optimization algorithm to iterate the policy network and value network. The reward function is the negative weighted sum of task delay penalty, energy consumption cost, and risk exposure penalty, with weights of 0.6, 0.3, and 0.1, respectively.

5. The intelligent management method for coal storage and transportation according to claim 4, characterized in that, The dynamic scheduling optimization command is issued to the corresponding execution mechanism, and the equipment status and environmental parameters in the three-dimensional dynamic digital twin model are updated synchronously, including: The execution instructions are transmitted via the industrial Ethernet bus. The instruction format includes the target device identifier, control parameters, execution time limit, and security check code. The target device identifier is a 16-bit device code, the control parameters are set according to the instruction type, and the security check code is a 32-bit cyclic redundancy check code. Upon receiving the instruction, the actuator immediately executes it and sends back the execution result through the device status feedback interface. If no confirmation signal is received within the preset time limit, the instruction retransmission mechanism is triggered. The retransmission interval is five seconds, and the maximum number of retransmissions is three. After three failures, the system switches to manual intervention mode. The execution results are used to update the device locations, operating status, and environmental parameters in the 3D dynamic digital twin model, ensuring that the model's state is synchronized with the physical world.

6. The intelligent management method for coal storage and transportation according to claim 5, characterized in that, When the risk probability value exceeds a preset safety threshold, a multi-level safety response mechanism is triggered, including: The initial threshold for spontaneous combustion risk is 5%, which is increased by 0.5% for each near-miss incident, up to a maximum of 10%. The initial threshold for collapse risk is 3%, which is adjusted in real time based on the height and density distribution of the pile. The adjustment formula is: threshold equal to 3% multiplied by the square root of the pile height divided by 10 meters multiplied by the square root of the density divided by 0.8 tons per cubic meter. The threshold for dust explosion risk is set based on the product of oxygen concentration and dust concentration, with the initial critical value set at 500. The Level 1 response initiates localized area spray cooling and forced ventilation. The spray intensity is dynamically adjusted according to the rate of temperature rise, ranging from 0.5 to 2 liters per square meter per minute. The forced ventilation volume is set to 80% of the maximum value. A Level 2 response suspends operations in high-risk areas and evacuates surrounding equipment. The system sends an emergency stop command to all equipment that has entered the high-risk area and guides it to evacuate along a preset safe path. A Level 3 response cuts off the regional power supply and activates the inert gas injection system. The release of inert gas reduces the oxygen concentration to below 12 percent. Each level of response command has mandatory priority and can interrupt all current non-safety-related dispatch commands.

7. A coal storage and transportation intelligent management system, characterized in that, include: A distributed sensor network module is deployed in coal storage yards and transportation corridors to collect real-time physical state data, environmental monitoring data, and operational behavior data during coal storage and transportation. The physical state data includes coal pile height, density distribution, surface temperature gradient, and internal temperature rise rate. The environmental monitoring data includes air humidity, dust concentration, oxygen concentration, carbon monoxide concentration, wind speed, and wind direction. The operational behavior data includes stacker-reclaimer operating trajectory, belt conveyor start / stop status, vehicle entry / exit frequency, and loading / unloading operation duration. The multi-source data fusion modeling module is used to perform spatiotemporal alignment and multi-source heterogeneous data fusion on the physical state data, environmental monitoring data and operation behavior data to construct a three-dimensional dynamic digital twin model of coal storage and transportation status. The three-dimensional dynamic digital twin model records the evolution of physical properties of coal, fluctuations of environmental parameters and correlation of equipment operation behavior in each spatial unit in a time series manner. The risk prediction and analysis module is used to extract the state evolution features within five sampling periods before and after the current moment based on the three-dimensional dynamic digital twin model using a sliding time window mechanism, and input them into the pre-trained storage and transportation risk prediction model. The storage and transportation risk prediction model is a temporal neural network constructed based on a gated recurrent unit and an attention mechanism. Its output is the probability value of spontaneous combustion, collapse or dust explosion of each spatial unit within the next three sampling periods. The dynamic scheduling optimization module is used to generate dynamic scheduling optimization instructions based on the risk probability value, combined with the current storage and transportation task priority, equipment availability status, and transportation route congestion index. The dynamic scheduling optimization instructions include instructions for reallocating stacker-reclaimer operating areas, adjusting belt conveyor speeds, replanning transport vehicle routes, controlling the start and stop of the dust suppression spray system, and adjusting ventilation equipment airflow. The closed-loop control execution module is used to send the dynamic scheduling optimization instructions to the corresponding execution mechanisms and simultaneously update the equipment status and environmental parameters in the three-dimensional dynamic digital twin model to form a closed-loop feedback control. A multi-level safety response module is used to trigger a multi-level safety response mechanism when the risk probability value exceeds a preset safety threshold. The multi-level safety response mechanism includes a first-level response: initiating local area sprinkler cooling and forced ventilation. Level 2 Response: Suspend operations in high-risk areas and evacuate surrounding equipment; Level 3 Response: Cut off the power supply to the area and activate the inert gas injection system.

8. The intelligent coal storage and transportation management system according to claim 7, characterized in that, The multi-source data fusion modeling module is used for: The storage yard space is divided into cubic grid cells with a side length of five meters. Each grid cell corresponds to a spatial coordinate point. All sensor data falling into the grid cell are aggregated to form the initial state vector of the cell at the current moment. The state vectors of all grid cells in the continuous time series are organized into a three-dimensional tensor, with its three modes corresponding to the spatial dimension, time dimension and feature dimension, respectively. The tensor is subjected to high-order singular value decomposition to extract the core tensor and factor matrix. In the decomposed low-dimensional space, a graph neural network is introduced to model cross-modal associations. Each grid cell is regarded as a node in the graph, and the connections between nodes are constructed based on spatial proximity and operational association. The neighbor node information is aggregated through a multi-layer graph convolution structure, and the comprehensive state encoding vector of each grid cell at the current time is output as the basic unit of the three-dimensional dynamic digital twin model.

Images