Intelligent management and control decision system for coal preparation plant based on whole-process risk prediction

Abstract

The application discloses a coal preparation plant intelligent management and control decision system based on a whole-process risk prediction, and particularly relates to the technical field of coal preparation intelligent management and control, and integrates heterogeneous data through a whole-process risk knowledge graph construction module, and defines risk transmission edges covering equipment, processes and safety; dynamic data mapping is realized by using a multi-source data sensing and fusion module; a risk prediction module combines a space-time graph neural network, identifies a risk source, and accurately calculates a transmission path, a lag time and an expected risk value; a collaborative decision module generates compensation parameters by using a physical lag time window, and issues the compensation parameters to an executing mechanism to implement feedforward preventive intervention; a feedback graph self-optimization module iterates graph weights in real time according to intervention effects, and the application realizes advanced perception and accurate hedging of risks, and significantly improves the stability and intelligent decision level of a coal preparation plant production system.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for coal preparation, and more specifically, to an intelligent control and decision-making system for coal preparation plants based on full-process risk prediction. Background Technology

[0002] As a fundamental energy source in my country, coal washing and processing is crucial for achieving clean and efficient coal utilization. The intelligent transformation of coal preparation plants has shifted from automation of single-point equipment to intelligent management and control of the entire process. Traditional coal preparation plant management models mainly rely on human experience and separate automated systems. While these methods have improved production efficiency to some extent, they still reveal significant limitations when facing complex production environments and dynamically changing coal qualities.

[0003] Existing control platforms often focus on single-dimensional monitoring, such as simple equipment failure warnings or simple process parameter monitoring. However, coal preparation is a highly coupled continuous process, and there is a deep correlation between equipment health status, process parameter fluctuations, and safety and environmental risks.

[0004] Most current intelligent control logic is based on "feedback" regulation, meaning that alarms and control commands are only triggered when sensors detect deviations in indicators or malfunctions. Due to the significant material transport lag in coal preparation processes, this "post-event processing" mode often results in production fluctuations already spreading, making it difficult to achieve true preventative intervention, leading to waste of reagents, increased energy consumption, and even unplanned shutdowns.

[0005] In addition, the utilization rate of multi-source data is low. Coal preparation plants have accumulated massive amounts of video images, sensor sequences, and offline test data, but due to the lack of effective spatiotemporal correlation analysis methods, these heterogeneous data have failed to be transformed into accurate decision-making suggestions to guide production, resulting in the control system having "data but no intelligence". Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent control and decision-making system for coal preparation plants based on full-process risk prediction, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent control and decision-making system for coal preparation plants based on full-process risk prediction, including a full-process risk knowledge graph construction module, a multi-source data perception and fusion module, a risk prediction module, a collaborative decision-making module, and a feedback-based graph self-optimization module.

[0008] The full-process risk knowledge graph construction module is used to integrate multi-source heterogeneous data from coal preparation plants, construct a heterogeneous knowledge graph covering equipment health risk nodes, process anomaly risk nodes, and safety and environmental risk nodes, and define risk transmission association edges between each risk node based on process flow and physical logic.

[0009] The multi-source data sensing and fusion module is used to collect multi-source data, including equipment vibration, current signals, video images, process parameters and offline test data, and map them to the corresponding nodes of the risk knowledge graph.

[0010] The risk prediction module is used to dynamically analyze the mapped knowledge graph using a spatiotemporal graph neural network, identify the current risk source, and calculate the transmission path, lag time, and expected risk value of the risk between different process nodes.

[0011] The collaborative decision-making module is used to generate compensatory control parameters based on risk transmission weights within the lag time before the predicted risk reaches the target node, and then distribute them to the implementing agency to implement preventive intervention.

[0012] The feedback-based graph self-optimization module is used to monitor the system response after preventive intervention in real time and dynamically update the weight parameters of the associated edges in the risk knowledge graph based on the intervention effect.

[0013] Preferably, as a preferred embodiment of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction described in this invention, it includes a full-process risk knowledge graph construction module for integrating multi-source heterogeneous data from the coal preparation plant, constructing a heterogeneous knowledge graph covering equipment health risk nodes, process anomaly risk nodes, and safety and environmental risk nodes, and defining risk transmission association edges between each risk node based on process flow and physical logic, specifically including the following:

[0014] The risk knowledge graph of the entire coal preparation plant process is represented as a directed heterogeneous graph. Where V is the set of nodes and E is the set of edges. For node type mapping functions, , For edge type mapping functions;

[0015] Device health risk nodes Let the state risk of the d-th device be represented by its feature vector, which is defined as: Process anomaly risk nodes The risk of the p-th process parameter is defined as follows: Safety and environmental risk nodes Represented as the first The risk of a safety monitoring point is defined by its characteristic vector as follows: ;in, As a vibration risk indicator, As a current risk indicator, As a temperature risk indicator, As a wear and tear risk indicator, Instantaneous deviation This is the cumulative deviation. The rate of change of deviation For the deviation integral, For gas concentration risk, For dust concentration risk, To mitigate the risk of intrusion by personnel, Fire risk;

[0016] The risk transmission correlation edges include three types: process transmission edges, causal transmission edges, and coupling transmission edges, which respectively correspond to risk transmission in the material flow direction, process deviations caused by equipment deterioration, and the amplification effect caused by the superposition of multiple risks, including:

[0017] Transmit the process edge This indicates that risk is transmitted from upstream to downstream processes along the material flow direction, and its transmission weight is defined as: ;in, This represents the material flow rate from the process located at node i to the process located at node j. Let i be the set of all downstream process nodes. Let be the sensitivity coefficient of node j to the risk of node i;

[0018] Causal transmission edge This indicates that equipment performance degradation has caused deviations in process parameters, and its propagation weight is defined as: ;in, Let $\frac{i}{j}$ be the conditional probability that node $j$ is at risk given that node $i$ is at risk. Let be the prior probability of node j experiencing risk. The causal strength coefficient;

[0019] Couple the conduction edge This indicates that when multiple risks coexist, a superposition amplification effect occurs, and its coupling amplification coefficient is defined as: ,in, Let these be the risk energy values ​​of the two risk source nodes at the current moment. Based on the fundamental conductivity, This is the single-risk main effect coefficient. Let be the coupling interaction coefficient, when This indicates that there is a risk amplification effect.

[0020] Preferably, as a preferred embodiment of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction described in this invention, it includes the multi-source data sensing and fusion module for collecting multi-source data including equipment vibration, current signals, video images, process parameters, and offline test data, and mapping them to corresponding nodes of the risk knowledge graph, specifically including the following:

[0021] The coal preparation plant acquires all elements of data in real time through an industrial internet platform. Based on the physical attributes and acquisition frequency of the data, a hierarchical acquisition mechanism is established, including: high-frequency time-series data, medium and low-frequency process data, unstructured visual data, and offline testing and manual data.

[0022] The high-frequency time-series data is acquired through piezoelectric vibration sensors deployed on key equipment, with a sampling frequency of up to 10. The above acceleration signals, along with real-time monitoring of the motor's operating current using a current transformer, are mapped to vibration indicators at the equipment's health risk nodes. and current index ;

[0023] The low- and medium-frequency process data is obtained by using the PLC / DCS automatic control system of the coal preparation plant to input real-time process parameters such as liquid level, pressure, flow rate, suspension density, and magnetic content, and mapping them to process anomaly risk nodes.

[0024] The unstructured visual data is obtained through high-definition cameras, capturing images of the conveyor belt surface condition, flotation foam, coal pile conditions in the receiving pit, and video streams of personnel in the work area. These images are then mapped to safety and environmental risk nodes. A convolutional neural network is used to process the video images, transforming the unstructured video into a structured risk feature vector. An edge detection algorithm is used to identify the conveyor belt edge position, calculate its real-time center deviation, and convert it into a standardized conveyor belt misalignment displacement, which serves as input for the safety and environmental risk indicators. Further steps include:

[0025] The acquired belt video images are converted to grayscale and denoised to enhance image features. An edge detection algorithm is then used to process the preprocessed images to obtain left and right edge point sets.

[0026] The least squares method was applied to fit the extracted left and right edge point sets to obtain the linear equations of the edges on both sides of the belt. At the preset detection reference height At this point, calculate the pixel width of the belt. ,in, These are the slopes of the left and right edge lines, respectively. These are the intercepts of the left and right edge lines, respectively; x is the x-coordinate of the image; and y is the y-coordinate of the image.

[0027] Calculate the two edge lines relative to the ideal center axis The mean offset is used to determine the real-time centerline. Introducing physical space calibration coefficients Calculate the final deviation displacement index ,Will As an input to the safety environment risk node, its numerical value is directly related to the risk weight of "belt tear" in the graph;

[0028] In coal preparation processes, the transmission time of risks between different process nodes depends on the material flow rate. This system dynamically adjusts the time-aware window of the spatiotemporal graph convolutional layer by real-time access to the belt scale flow parameter Q. ,in, For the system's rated flow rate, To monitor flow rate in real time, when the feed flow rate increases... At that time, window Automatic shrinkage enables the model to capture the instantaneous characteristics of risk transmission at high flow rates; conversely, it automatically extends the time window at low flow rates.

[0029] The offline testing and manual data are obtained by accessing the lagging test data of raw coal ash content, clean coal moisture content and tailings gangue content through a fully automated laboratory system. The time backtracking algorithm based on the "material tracking model" is used to associate the measured coal quality indicators with the corresponding production parameters when it passes through the washing and beneficiation process.

[0030] For data with different sampling rates, a linear interpolation technique based on "timestamp alignment" is used to unify millisecond-level vibration data and minute-level process data in the same time domain window.

[0031] Preferably, as a preferred embodiment of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction described in this invention, the risk prediction module is used to dynamically analyze the mapped knowledge graph using a spatiotemporal graph neural network, identify the current risk source, and calculate the transmission path, lag time, and expected risk value of the risk between different process nodes. Specifically, it includes the following:

[0032] The spatiotemporal graph neural network model for whole-process risk prediction in coal preparation plants is represented as: ,in, The directed heterogeneous graph output by the full-process risk knowledge graph construction module. Let S be the node feature sequence within a historical time window T, S be the set of identified current risk source nodes, P be the set of predicted risk transmission paths, and A be the expected risk value matrix. This is the prediction lag time matrix between nodes;

[0033] Spatial graph convolutional layers are used to capture lateral penetration and vertical coupling between heterogeneous nodes by calculating the attention score of node i to its neighbor node j. Identify key risk transmission paths and combine them with process transmission weights. Causal transmission weight and coupling amplification factor Calculate the comprehensive transmission probability matrix ;in, Let i be the set of neighboring nodes. This represents traversing each node in the set;

[0034] Starting from the risk source node S, based on Filter out those with a propagation intensity exceeding a preset threshold The directed edge sequence forms a set P;

[0035] Real-time traffic parameters accessed using the multi-source data sensing module Dynamically update the lag time matrix The elements in ,in, For nodes Physical length of the transmission link between them This is a material flow rate correction factor. The inherent process reaction time of the concentration equipment; the lag time matrix Used to determine the alignment bias when the spatiotemporal graph neural network performs feature aggregation in the time dimension;

[0036] The risk evolution trend of nodes is modeled using temporal convolutional layers, and the results are output for each node in the future. The risk state at each step is combined with the heterogeneous graph topology G, the historical feature sequence matrix X, and the real-time corrected hysteresis parameter. 1. Input the belt scale flow rate parameter Q, and output the expected risk value matrix A through the fully connected layer. The specific formula is as follows: ,in, This indicates predicting the future t-th time from the current time t. Risk probability distribution for each process node in the entire plant at a given time step. This represents the spatiotemporal graph convolution operator. This represents the normalized exponential function. When the predicted value in A exceeds the safety red line, it automatically locks the target damaged node on the risk path P, providing the collaborative decision-making module with a precise intervention target.

[0037] Utilizing a dynamic flow adjustment mechanism, the risk prediction module dynamically expands and contracts the time sensing window based on the actual feed load when performing convolution operations. When the system is running at high traffic, reduce This increases the model's response frequency to instantaneous process fluctuations at high flow rates; when the system is operating at low flow rates, it increases... By covering longer historical time-series data, background noise of the sensor is suppressed, ensuring the stability of risk identification results under low-load conditions.

[0038] Preferably, as a preferred embodiment of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction described in this invention, the collaborative decision-making module is used to generate compensatory control parameters based on risk transmission weights within the lag time before the predicted risk reaches the target node, and then distribute them to the implementing agency for preventive intervention. Specifically, it includes the following:

[0039] Based on the transmission path P and expected risk value matrix A output by the risk prediction module, the compensation amount for the target node is calculated using the following formula: ,in, For the target node, Nodes output by the risk prediction module Lag time The expected risk value afterward To start from the risk source node To the target node The conduction path, To control the parameter weight vector, This is the current process state parameter vector. Generates the compensation amount;

[0040] Using prediction lag time By aligning time, a "feedforward control" effect is achieved, with the optimal intervention time being... ,in, For the moment of risk identification, This refers to the physical lag time for risk to be transmitted to the target node. The inherent response time of the actuator from receiving the instruction to completing the action;

[0041] According to the optimal intervention time Determine when to issue the instruction: ,in, The second is the emergency threshold. The minimum compensation threshold is used to determine the waiting time for delayed delivery: ;

[0042] The generated compensation amount is subject to amplitude constraints, rate of change constraints, and risk reduction effectiveness constraints to prevent over-adjustment. The amplitude constraints are as follows: The rate of change constraint is The constraint on the effectiveness of risk reduction is: ,in, The expected risk value after applying compensation. For the target node at time t The generated compensation amount, For the target node Maximum allowable compensation amount per instance This is the compensation amount from the previous moment. For the target node Maximum allowable rate of change of compensation The expected risk value before applying compensation. To minimize the risk reduction;

[0043] And based on the optimal intervention time The generated compensatory control parameters are then sent to the corresponding actuators.

[0044] Preferably, as a preferred embodiment of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction described in this invention, it includes the feedback-based graph self-optimization module for real-time monitoring of the system response after preventive intervention, and dynamically updating the weight parameters of the associated edges in the risk knowledge graph according to the intervention effect, specifically including the following:

[0045] Predictive lag time after collaborative decision-making instructions are issued Within the window, feedback signals from damaged nodes are collected, and intervention scores are calculated. ,in, It is the expected risk value before compensation is applied. It is the actual risk value measured by the sensor after the intervention is implemented; It is the technological cost incurred in performing this intervention. The maximum permissible process cost for a single intervention. It is the performance weighting coefficient, when When the intervention has achieved a positive effect, the reduction in risk outweighs the cost; when When the intervention's effect equals its cost, the marginal benefit is zero; when... When this occurs, it indicates that the intervention was ineffective, and the costs outweighed the risks, reducing the benefits.

[0046] Based on intervention score The gradient ascent method in reinforcement learning is used to analyze the comprehensive propagation probability matrix in the knowledge graph. The correction formula for gradient ascent is as follows: ,in, For learning rate, The baseline score for successful intervention is set. For the updated parameter set, The parameter set before the update, including process propagation weights. Causal transmission weight Coupling interaction coefficient Attention score , The gradient of the intervention score with respect to the parameters;

[0047] Based on the gradient ascent correction formula described above, intervention scores are calculated after each intervention. and compare it with the preset benchmark score. Comparison:

[0048] when When this occurs, it indicates that the intervention has achieved better-than-expected results. Increasing the weight parameters of the corresponding transmission path along the positive gradient direction strengthens the confidence of the risk transmission path in future predictions.

[0049] when When the intervention effect is not as expected, the module reduces the corresponding weight parameters along the negative gradient direction to weaken the influence of the transmission path.

[0050] when At the same time, the parameters remain unchanged, and through the above mechanism, the process transmission weights can be dynamically adjusted. Causal transmission weight Coupling interaction coefficient and attention score This allows the risk knowledge graph to continuously approximate the actual risk transmission patterns in coal preparation processes, enabling adaptive optimization and continuous evolution of the prediction model.

[0051] On the other hand, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements the functional modules of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction as described above.

[0052] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction as described above in the present invention.

[0053] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0054] By integrating all-element data through a heterogeneous knowledge graph, this invention breaks through the limitations of traditional single-point monitoring and achieves in-depth tracing of risk transmission across processes. The introduction of a dynamic lag parameter driven by real-time flow Q enables the spatiotemporal graph neural network prediction model to adapt to changing operating conditions, ensuring the physical consistency between prediction results and material flow rate. The collaborative decision-making module utilizes the "lag window" formed by prediction to implement feedforward intervention, achieving "imperceptible hedging" against process fluctuations and effectively avoiding substandard clean coal quality and unplanned shutdowns. This invention can correct graph weights in real time through a feedback self-optimization module, allowing the control logic to dynamically iterate with equipment aging and coal quality changes. This invention not only improves the foresight and accuracy of risk identification in coal preparation plants but also significantly enhances the stability of the production system and the scientific nature of decision-making. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0056] Figure 1 This is a flowchart of the method of the present invention.

[0057] Figure 2 This is a logical topology diagram of the risk transmission throughout the coal preparation process, as shown in a specific embodiment of the present invention.

[0058] Table 1 is a data recording table of the simulation experiment of this invention. Detailed Implementation

[0059] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0061] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0062] Example 1

[0063] This embodiment provides, for example Figure 1 The intelligent control and decision-making system for coal preparation plants based on full-process risk prediction shown includes a full-process risk knowledge graph construction module, a multi-source data perception and fusion module, a risk prediction module, a collaborative decision-making module, and a feedback-based graph self-optimization module.

[0064] The full-process risk knowledge graph construction module is used to integrate multi-source heterogeneous data from coal preparation plants, construct a heterogeneous knowledge graph covering equipment health risk nodes, process anomaly risk nodes, and safety and environmental risk nodes, and define risk transmission association edges between each risk node based on process flow and physical logic.

[0065] The multi-source data sensing and fusion module is used to collect multi-source data, including equipment vibration, current signals, video images, process parameters and offline test data, and map them to the corresponding nodes of the risk knowledge graph.

[0066] The risk prediction module is used to dynamically analyze the mapped knowledge graph using a spatiotemporal graph neural network, identify the current risk source, and calculate the transmission path, lag time, and expected risk value of the risk between different process nodes.

[0067] The collaborative decision-making module is used to generate compensatory control parameters based on risk transmission weights within the lag time before the predicted risk reaches the target node, and then distribute them to the implementing agency to implement preventive intervention.

[0068] The feedback-based graph self-optimization module is used to monitor the system response after preventive intervention in real time and dynamically update the weight parameters of the associated edges in the risk knowledge graph based on the intervention effect.

[0069] In this embodiment, the specific module for constructing a full-process risk knowledge graph needs to be explained. This module integrates multi-source heterogeneous data from the coal preparation plant to construct a heterogeneous knowledge graph covering equipment health risk nodes, process anomaly risk nodes, and safety and environmental risk nodes. Based on process flow and physical logic, it defines risk transmission association edges between each risk node, specifically including the following:

[0070] The risk knowledge graph of the entire coal preparation plant process is represented as a directed heterogeneous graph. Where V is the set of nodes and E is the set of edges. For node type mapping functions, , For edge type mapping functions;

[0071] Device health risk nodes Let the state risk of the d-th device be represented by its feature vector, which is defined as: Process anomaly risk nodes The risk of the p-th process parameter is defined as follows: Safety and environmental risk nodes Represented as the first The risk of a safety monitoring point is defined by its characteristic vector as follows: ;in, As a vibration risk indicator, This is the root mean square value of the vibration. Vibration threshold This is the load correction factor; As a current risk indicator, For the measured current, Rated current; This is a temperature risk indicator, where T is the measured temperature. For ambient temperature, This is the alarm threshold; As a wear and tear risk indicator, For cumulative running time, Mean time between failures (MTBF) Instantaneous deviation This is the cumulative deviation. The rate of change of deviation For the deviation integral, For gas concentration risk, For dust concentration risk, To mitigate the risk of intrusion by personnel, Fire risk;

[0072] The risk transmission correlation edges include three types: process transmission edges, causal transmission edges, and coupling transmission edges, which respectively correspond to risk transmission in the material flow direction, process deviations caused by equipment deterioration, and the amplification effect caused by the superposition of multiple risks, including:

[0073] Transmit the process edge This indicates that risk is transmitted from upstream to downstream processes along the material flow direction, and its transmission weight is defined as: ;in, This represents the material flow rate from the process located at node i to the process located at node j. Let i be the set of all downstream process nodes. Let be the sensitivity coefficient of node j to the risk of node i. Weighting for process transmission;

[0074] Causal transmission edge This indicates that equipment performance degradation has caused deviations in process parameters, and its propagation weight is defined as: ;in, Let $\frac{i}{j}$ be the conditional probability that node $j$ is at risk given that node $i$ is at risk. Let be the prior probability of node j experiencing risk. The causal strength coefficient For causal transmission weights;

[0075] Couple the conduction edge This indicates that when multiple risks coexist, a superposition amplification effect occurs, and its coupling amplification coefficient is defined as: ,in, Let these be the risk energy values ​​of the two risk source nodes at the current moment. This is the coupling amplification factor. Based on the fundamental conductivity, This is the single-risk main effect coefficient. Let be the coupling interaction coefficient, when This indicates the existence of a risk amplification effect;

[0076] like Figure 2As shown, the diagram illustrates the risk deduction logic for coal preparation based on a knowledge graph. The nodes are divided into equipment, process, and safety layers, and three types are defined: "equipment node," "process node," and "safety node," represented by circles, dashed rectangles, and hexagons, respectively. Solid arrows represent "process edges," dashed arrows represent "causal edges," and dashed dotted arrows represent "coupling edges." The equipment layer includes "raw coal crusher (101)," "mixing pump (102)," and "dewatering vibrating screen (103)." The equipment layer includes process nodes such as "feed gangue content (201)", "heavy medium density fluctuation (202)" and "clean coal ash content quality (203)"; the safety layer includes safety nodes such as "dust / gas risk (301)", "mechanical injury risk (302)" and "fire / smoke risk (303)"; among them, the equipment layer nodes affect the process layer nodes through causal transmission; the process layer nodes transmit risks through the process flow; the risks of different levels converge and amplify in the safety layer through the coupling transmission edge.

[0077] In this embodiment, the multi-source data sensing and fusion module is specifically described. This module collects multi-source data, including equipment vibration, current signals, video images, process parameters, and offline test data, and maps it to the corresponding nodes of the risk knowledge graph. Specifically, it includes the following:

[0078] The coal preparation plant acquires all elements of data in real time through an industrial internet platform. Based on the physical attributes and acquisition frequency of the data, a hierarchical acquisition mechanism is established, including: high-frequency time-series data, medium and low-frequency process data, unstructured visual data, and offline testing and manual data.

[0079] The high-frequency time-series data is acquired through piezoelectric vibration sensors deployed on key equipment, with a sampling frequency of up to 10. The above acceleration signals, along with real-time monitoring of the motor's operating current using a current transformer, are mapped to vibration indicators at the equipment's health risk nodes. and current index ;

[0080] The low- and medium-frequency process data is obtained by using the PLC / DCS automatic control system of the coal preparation plant to input real-time process parameters such as liquid level, pressure, flow rate, suspension density, and magnetic content, and mapping them to process anomaly risk nodes.

[0081] The unstructured visual data is obtained through high-definition cameras, capturing images of the conveyor belt surface condition, flotation foam, coal pile conditions in the receiving pit, and video streams of personnel in the work area. These images are then mapped to safety and environmental risk nodes. A convolutional neural network is used to process the video images, transforming the unstructured video into a structured risk feature vector. An edge detection algorithm is used to identify the conveyor belt edge position, calculate its real-time center deviation, and convert it into a standardized conveyor belt misalignment displacement, which serves as input for the safety and environmental risk indicators. Further steps include:

[0082] The acquired belt video images are converted to grayscale and denoised to enhance image features. An edge detection algorithm is then used to process the preprocessed images to obtain left and right edge point sets.

[0083] The least squares method was applied to fit the extracted left and right edge point sets to obtain the linear equations of the edges on both sides of the belt. At the preset detection reference height At this point, calculate the pixel width of the belt. ,in, These are the slopes of the left and right edge lines, respectively. These are the intercepts of the left and right edge lines, respectively; x is the x-coordinate of the image; and y is the y-coordinate of the image.

[0084] Calculate the two edge lines relative to the ideal center axis The mean offset is used to determine the real-time centerline. Introducing physical space calibration coefficients Calculate the final deviation displacement index ,Will As an input to the safety environment risk node, its numerical value is directly related to the risk weight of "belt tear" in the graph;

[0085] In coal preparation processes, the transmission time of risks between different process nodes depends on the material flow rate. This system dynamically adjusts the time-aware window of the spatiotemporal graph convolutional layer by real-time access to the belt scale flow parameter Q. ,in, For the system's rated flow rate, To monitor flow rate in real time, when the feed flow rate increases... At that time, window Automatic shrinkage enables the model to capture the instantaneous characteristics of risk transmission at high flow rates; conversely, it automatically extends the time window at low flow rates.

[0086] The offline testing and manual data are obtained by accessing the lagging test data of raw coal ash content, clean coal moisture content and tailings gangue content through a fully automated laboratory system. The time backtracking algorithm based on the "material tracking model" is used to associate the measured coal quality indicators with the corresponding production parameters when it passes through the washing and beneficiation process.

[0087] For data with different sampling rates, a linear interpolation technique based on "timestamp alignment" is used to unify millisecond-level vibration data and minute-level process data in the same time domain window.

[0088] In this embodiment, the risk prediction module needs to be specifically described. This module utilizes a spatiotemporal graph neural network to dynamically analyze the mapped knowledge graph, identify current risk sources, and calculate the transmission path, lag time, and expected risk value of the risk across different process nodes. Specifically, it includes the following:

[0089] The spatiotemporal graph neural network model for whole-process risk prediction in coal preparation plants is represented as: ,in, The directed heterogeneous graph output by the full-process risk knowledge graph construction module. Let S be the node feature sequence within a historical time window T, S be the set of identified current risk source nodes, P be the set of predicted risk transmission paths, and A be the expected risk value matrix. This is the prediction lag time matrix between nodes;

[0090] Spatial graph convolutional layers are used to capture lateral penetration and vertical coupling between heterogeneous nodes by calculating the attention score of node i to its neighbor node j. Identify key risk transmission paths and combine them with process transmission weights. Causal transmission weight and coupling amplification factor Calculate the comprehensive transmission probability matrix ;in, Let i be the set of neighboring nodes. This represents traversing each node in the set;

[0091] Starting from the risk source node S, based on Filter out those with a propagation intensity exceeding a preset threshold The directed edge sequence forms a set P;

[0092] Real-time traffic parameters accessed using the multi-source data sensing module Dynamically update the lag time matrix The elements in ,in, For nodes Physical length of the transmission link between them This is a material flow rate correction factor. The inherent process reaction time of the concentration equipment; the lag time matrix Used to determine the alignment bias when the spatiotemporal graph neural network performs feature aggregation in the time dimension;

[0093] The risk evolution trend of nodes is modeled using temporal convolutional layers, and the results are output for each node in the future. The risk state at each step is combined with the heterogeneous graph topology G, the historical feature sequence matrix X, and the real-time corrected hysteresis parameter. 1. Input the belt scale flow rate parameter Q, and output the expected risk value matrix A through the fully connected layer. The specific formula is as follows: ,in, This indicates predicting the future t-th time from the current time t. Risk probability distribution for each process node in the entire plant at a given time step. This represents the spatiotemporal graph convolution operator. This represents the normalized exponential function. When the predicted value in A exceeds the safety red line, it automatically locks the target damaged node on the risk path P, providing the collaborative decision-making module with a precise intervention target.

[0094] Utilizing a dynamic flow adjustment mechanism, the risk prediction module dynamically expands and contracts the time sensing window based on the actual feed load when performing convolution operations. When the system is running at high traffic, reduce This increases the model's response frequency to instantaneous process fluctuations at high flow rates; when the system is operating at low flow rates, it increases... By covering longer historical time-series data, background noise of the sensor is suppressed, ensuring the stability of risk identification results under low-load conditions.

[0095] In this embodiment, the collaborative decision-making module is specifically described. This module generates compensatory control parameters based on risk transmission weights within the lag time before the predicted risk reaches the target node, and then distributes these parameters to the implementing agency for preventative intervention. Specifically, it includes the following:

[0096] Based on the transmission path P and expected risk value matrix A output by the risk prediction module, the compensation amount for the target node is calculated using the following formula: ,in, For the target node, Nodes output by the risk prediction module Lag time The expected risk value afterward To start from the risk source node To the target node The conduction path, To control the parameter weight vector, This is the current process state parameter vector. Generates the compensation amount;

[0097] Using prediction lag time By aligning time, a "feedforward control" effect is achieved, with the optimal intervention time being... ,in, For the moment of risk identification, This refers to the physical lag time for risk to be transmitted to the target node. The inherent response time of the actuator from receiving the instruction to completing the action;

[0098] According to the optimal intervention time Determine when to issue the instruction: ,in, The second is the emergency threshold. The minimum compensation threshold is used to determine the waiting time for delayed delivery: ;

[0099] The generated compensation amount is subject to amplitude constraints, rate of change constraints, and risk reduction effectiveness constraints to prevent over-adjustment. The amplitude constraints are as follows: The rate of change constraint is The constraint on the effectiveness of risk reduction is: ,in, The expected risk value after applying compensation. For the target node at time t The generated compensation amount, For the target node Maximum allowable compensation amount per instance This is the compensation amount from the previous moment. For the target node Maximum allowable rate of change of compensation The expected risk value before applying compensation. To minimize the risk reduction;

[0100] And based on the optimal intervention time The generated compensatory control parameters are then sent to the corresponding actuators.

[0101] Specifically, for abnormal process nodes, the collaborative decision-making module issues control parameters such as the opening degree of the diversion box, the opening degree of the water supply valve, and the density setpoint through the PLC / DCS controller, and the inherent response time of its actuator... 100 500 milliseconds;

[0102] For equipment health nodes, the collaborative decision-making module issues feed rate, motor speed, and load limit control parameters via frequency converter or intelligent protection device, with a response time of 50 seconds. 200 milliseconds;

[0103] For secure environment nodes, the collaborative decision-making module issues warning level and area lock status control commands via audible and visual alarms, with a response time of 10 seconds. 50 milliseconds;

[0104] Through the aforementioned tiered distribution mechanism, control instructions are transmitted and executed before the optimal intervention time, enabling precise preventative intervention before risks materialize.

[0105] In this embodiment, the feedback-based graph self-optimization module is specifically described. This module is used to monitor the system response after preventive intervention in real time and dynamically update the weight parameters of the associated edges in the risk knowledge graph based on the intervention effect. Specifically, it includes the following:

[0106] Predictive lag time after collaborative decision-making instructions are issued Within the window, feedback signals from damaged nodes are collected, and intervention scores are calculated. ,in, It is the expected risk value before compensation is applied. It is the actual risk value measured by the sensor after the intervention is implemented; It is the technological cost incurred in performing this intervention. The maximum permissible process cost for a single intervention. It is the performance weighting coefficient, when When the intervention has achieved a positive effect, the reduction in risk outweighs the cost; when When the intervention's effect equals its cost, the marginal benefit is zero; when... When this occurs, it indicates that the intervention was ineffective, and the costs outweighed the risks, reducing the benefits.

[0107] Based on intervention score The gradient ascent method in reinforcement learning is used to analyze the comprehensive propagation probability matrix in the knowledge graph. The correction formula for gradient ascent is as follows: ,in, For learning rate, The baseline score for successful intervention is set. For the updated parameter set, The parameter set before the update, including process propagation weights. Causal transmission weight Coupling interaction coefficient Attention score , The gradient of the intervention score with respect to the parameters;

[0108] Based on the gradient ascent correction formula described above, intervention scores are calculated after each intervention. and compare it with the preset benchmark score. Comparison:

[0109] when When this occurs, it indicates that the intervention has achieved better-than-expected results. Increasing the weight parameters of the corresponding transmission path along the positive gradient direction strengthens the confidence of the risk transmission path in future predictions.

[0110] when When the intervention effect is not as expected, the module reduces the corresponding weight parameters along the negative gradient direction to weaken the influence of the transmission path.

[0111] when At the same time, the parameters remain basically unchanged, and through the above mechanism, the process transmission weights can be dynamically adjusted. Causal transmission weight Coupling interaction coefficient and attention score This allows the risk knowledge graph to continuously approximate the actual risk transmission patterns in coal preparation processes, enabling adaptive optimization and continuous evolution of the prediction model.

[0112] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0113] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the functional modules of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0114] Example 2

[0115] The following is another embodiment of the present invention, which provides an intelligent control and decision-making system for coal preparation plants based on full-process risk prediction. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0116] This experiment aims to verify the effectiveness of an intelligent control and decision-making system for coal preparation plants based on full-process risk prediction. Through the construction of a full-process risk knowledge graph, multi-source data perception and fusion, spatiotemporal neural network prediction, collaborative decision-making intervention, and feedback-based graph self-optimization, the system enhances early warning and precise control capabilities for risks in complex production environments of coal preparation plants. The experiment uses simulated and actual coal preparation plant production data, including equipment vibration, current, video recognition features, process parameters, and laboratory data. By analyzing the consistency between the system-generated expected risk values ​​and actual process responses, as well as the accuracy of predictions and the effectiveness of interventions, the system's accuracy and robustness in identifying risk evolution, tracing path sources, and closed-loop optimization are verified.

[0117] The simulation experiment steps are implemented according to the content of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction provided in Example 1, and the specific steps include:

[0118] Acquire comprehensive data from the coal preparation plant, with the acquisition frequency set to 10 for high-frequency signals. Low- and medium-frequency process parameters are accessed in real time, and timestamp alignment technology is used to uniformly map heterogeneous data to the corresponding nodes of the risk knowledge graph.

[0119] By integrating multi-source data, a heterogeneous graph is constructed covering three types of risk nodes: equipment health, process anomalies, and safety environment. Based on the flow direction of the washing and beneficiation process, the transmission correlation edges and initial transmission weights between nodes are defined.

[0120] By using a spatiotemporal graph neural network to dynamically analyze the mapped knowledge graph and combining it with the real-time flow Q of the belt scale to dynamically adjust the time sensing window, future predictions can be made. The risk status at each step is analyzed to identify risk sources and calculate the transmission path and lag time. ;

[0121] During the lag time before the risk reaches the target node, the decision-making module generates compensatory control parameters. Automatically sends data to the PLC / frequency converter actuator for preventative process parameter adjustment and safety warnings;

[0122] Real-time monitoring of the system response after intervention and calculation of intervention scores. The gradient ascent method is used to dynamically update the weight parameters and attention scores of graph-related edges based on the risk suppression effect, thereby enabling the continuous evolution of the model.

[0123] The specific data from the above simulation experiment are as follows:

[0124] Time / minute Monitoring nodes / processes Key sensing data (vibration / flow / density) Predicting risk sources Expected risk value A Predicted lag time τ Preventive intervention decision Actual response / prediction consistency Risk management results 0-10 Feeder belt Vibration: 2.1 mm / s; Flow rate: 800 t / h none 0.12 (low) -- Maintain baseline parameters Stable / High performance Monitoring operation 10-20 201# Vibrating Screen Abnormal vibration waveform; current fluctuation 201# sieve 0.82 (High) 120s (transmission to hydrocyclone) Adjust the feed ratio and reduce the load in advance. Risk is being transmitted as expected / extremely high Preventive shutdown for maintenance 20-30 Heavy medium cyclone Density: 1.45 g / cm³; Pressure: 0.18 MPa Feed medium fluctuation 0.65 (Medium) 45s (transfer to the clean coal screen) Generates compensation pressure +0.02MPa Density fluctuation hedging / high Process parameters restored 30-40 Flotation workshop Abnormal visual characteristics of foam Dosing pump fluctuations 0.78 (High) 90s (transmission to tailings) Adjust automatic dosing compensation Indicators stabilize / medium to high Avoid product loss 40-50 Concentrator Increased rake frame torque; increased overflow turbidity Bottom discharge blockage 0.91 (High) 15s (instantaneous evolution) Emergency activation of the emergency circuit for forced material discharge Fast response / high Avoid rake accidents 50-60 The entire factory area Operating parameters regression benchmark none 0.08 (Low) -- Resume normal production mode System self-optimization / high Closed-loop stability

[0125] Table 1

[0126] Experimental Analysis:

[0127] By comparing the consistency between the risk probability matrix predicted by the model and the actual feedback labels from the production site, and analyzing the optimization of weight parameters using a feedback module, the results show that the spatiotemporal graph neural network, by introducing physical lag parameters... The prediction accuracy of the flow rate Q is significantly better than that of traditional single models when dealing with the risks of multi-process coupling. The collaborative decision-making module uses a "lag window" to perform feedforward compensation in advance, successfully nipping multiple process fluctuations in the bud and reducing the instantaneous exceedance of washing and beneficiation indicators. The feedback-based graph self-optimization module enables the graph weights to adapt to the characteristic drift caused by equipment wear through the iteration of intervention scores. Experiments show that the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction has high accuracy, real-time performance and reliability, and can effectively improve the risk control level and production stability of coal preparation plants throughout their entire life cycle.

[0128] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A coal preparation plant intelligent control and decision-making system based on full-process risk prediction, characterized in that: It includes a full-process risk knowledge graph construction module, a multi-source data perception and fusion module, a risk prediction module, a collaborative decision-making module, and a feedback-based graph self-optimization module; The full-process risk knowledge graph construction module is used to integrate multi-source heterogeneous data from coal preparation plants, construct a heterogeneous knowledge graph covering equipment health risk nodes, process anomaly risk nodes, and safety and environmental risk nodes, and define risk transmission association edges between each risk node based on process flow and physical logic. The multi-source data sensing and fusion module is used to collect multi-source data, including equipment vibration, current signals, video images, process parameters and offline test data, and map them to the corresponding nodes of the risk knowledge graph. The risk prediction module is used to dynamically analyze the mapped knowledge graph using a spatiotemporal graph neural network, identify the current risk source, and calculate the transmission path, lag time, and expected risk value of the risk between different process nodes. The collaborative decision-making module is used to generate compensatory control parameters based on risk transmission weights within the lag time before the predicted risk reaches the target node, and then distribute them to the implementing agency to implement preventive intervention. The feedback-based graph self-optimization module is used to monitor the system response after preventive intervention in real time and dynamically update the weight parameters of the associated edges in the risk knowledge graph based on the intervention effect.

2. The intelligent control and decision-making system for coal preparation plants based on full-process risk prediction as described in claim 1, characterized in that: The full-process risk knowledge graph construction module is used to integrate multi-source heterogeneous data from coal preparation plants to construct a heterogeneous knowledge graph covering equipment health risk nodes, process anomaly risk nodes, and safety and environmental risk nodes. Based on process flow and physical logic, it defines risk transmission relationships between each risk node, specifically including the following: The risk knowledge graph of the entire coal preparation plant process is represented as a directed heterogeneous graph. Where V is the set of nodes and E is the set of edges. For node type mapping functions, , For edge type mapping functions; Device health risk nodes Let the state risk of the d-th device be represented by its feature vector, which is defined as: Process anomaly risk nodes The risk of the p-th process parameter is defined as follows: Safety and environmental risk nodes Represented as the first The risk of a safety monitoring point is defined by its characteristic vector as follows: ;in, As a vibration risk indicator, As a current risk indicator, As a temperature risk indicator, As a wear and tear risk indicator, Instantaneous deviation This is the cumulative deviation. The rate of change of deviation For the deviation integral, For gas concentration risk, For dust concentration risk, To mitigate the risk of intrusion by personnel, Fire risk; The risk transmission correlation edges include three types: process transmission edges, causal transmission edges, and coupling transmission edges, which respectively correspond to risk transmission in the material flow direction, process deviations caused by equipment deterioration, and the amplification effect caused by the superposition of multiple risks, including: Transmit the process edge This indicates that risk is transmitted from upstream to downstream processes along the material flow direction, and its transmission weight is defined as: ;in, This represents the material flow rate from the process located at node i to the process located at node j. Let i be the set of all downstream process nodes. Let be the sensitivity coefficient of node j to the risk of node i; Causal transmission edge This indicates that equipment performance degradation has caused deviations in process parameters, and its propagation weight is defined as: ;in, Let $\frac{i}{j}$ be the conditional probability that node $j$ is at risk given that node $i$ is at risk. Let be the prior probability of node j experiencing risk. The causal strength coefficient; Couple the conduction edge This indicates that when multiple risks coexist, a superposition amplification effect occurs, and its coupling amplification coefficient is defined as: ,in, Let these be the risk energy values ​​of the two risk source nodes at the current moment. Based on the fundamental conductivity, This is the single-risk main effect coefficient. Let be the coupling interaction coefficient, when This indicates that there is a risk amplification effect.

3. The intelligent control and decision-making system for coal preparation plants based on full-process risk prediction according to claim 1, characterized in that: The multi-source data sensing and fusion module is used to collect multi-source data, including equipment vibration, current signals, video images, process parameters, and offline test data, and map them to the corresponding nodes of the risk knowledge graph. Specifically, it includes the following: The coal preparation plant acquires all elements of data in real time through an industrial internet platform. Based on the physical attributes and acquisition frequency of the data, a hierarchical acquisition mechanism is established, including: high-frequency time-series data, medium and low-frequency process data, unstructured visual data, and offline testing and manual data. The high-frequency time-series data is acquired through piezoelectric vibration sensors deployed on key equipment, with a sampling frequency of up to 10. The above acceleration signals, along with real-time monitoring of the motor's operating current using a current transformer, are mapped to vibration indicators at the equipment's health risk nodes. and current index ; The low- and medium-frequency process data is obtained by using the PLC / DCS automatic control system of the coal preparation plant to input real-time process parameters such as liquid level, pressure, flow rate, suspension density, and magnetic content, and mapping them to process anomaly risk nodes. The unstructured visual data is obtained by using high-definition cameras to capture the surface condition of the conveyor belt, images of flotation foam, the condition of the coal pile in the receiving pit, and video streams of personnel in the work area, which are then mapped to safety and environmental risk nodes. Convolutional neural networks are used to process the video images, transforming the unstructured video into structured risk feature vectors. The offline testing and manual data are obtained by accessing the lagging test data of raw coal ash content, clean coal moisture content and tailings gangue content through a fully automated laboratory system. The time backtracking algorithm based on the "material tracking model" is used to associate the measured coal quality indicators with the corresponding production parameters when it passes through the washing and beneficiation process. For data with different sampling rates, a linear interpolation technique based on "timestamp alignment" is used to unify millisecond-level vibration data and minute-level process data in the same time domain window.

4. The intelligent control and decision-making system for coal preparation plants based on full-process risk prediction according to claim 3, characterized in that: The process of using a convolutional neural network to process video images, transforming unstructured video into structured risk feature vectors, identifying belt edge positions through edge detection algorithms, calculating real-time center deviation, and converting it into standardized belt misalignment displacement as input for safety environmental risk indicators, further includes: The acquired belt video images are converted to grayscale and denoised to enhance image features. An edge detection algorithm is then used to process the preprocessed images to obtain left and right edge point sets. The least squares method was applied to fit the extracted left and right edge point sets to obtain the linear equations of the edges on both sides of the belt. At the preset detection reference height At this point, calculate the pixel width of the belt. ,in, These are the slopes of the left and right edge lines, respectively. These are the intercepts of the left and right edge lines, respectively; x is the x-coordinate of the image; and y is the y-coordinate of the image. Calculate the two edge lines relative to the ideal center axis The mean offset is used to determine the real-time centerline. Introducing physical space calibration coefficients Calculate the final deviation displacement index ,Will As an input to the safety environment risk node, its numerical value is directly related to the risk weight of "belt tear" in the graph; In coal preparation processes, the transmission time of risks between different process nodes depends on the material flow rate. This system dynamically adjusts the time-aware window of the spatiotemporal graph convolutional layer by real-time access to the belt scale flow parameter Q. ,in, For the system's rated flow rate, To monitor flow rate in real time, when the feed flow rate increases... At that time, window Automatic shrinking enables the model to capture the instantaneous characteristics of risk transmission at high flow rates; conversely, it automatically extends the time window.

5. The intelligent control and decision-making system for coal preparation plants based on full-process risk prediction according to claim 1, characterized in that: The risk prediction module is used to dynamically analyze the mapped knowledge graph using a spatiotemporal graph neural network, identify current risk sources, and calculate the transmission path, lag time, and expected risk value of risks between different process nodes. Specifically, it includes the following: The spatiotemporal graph neural network model for whole-process risk prediction in coal preparation plants is represented as: ,in, The directed heterogeneous graph output by the full-process risk knowledge graph construction module. Let S be the node feature sequence within a historical time window T, S be the set of identified current risk source nodes, P be the set of predicted risk transmission paths, and A be the expected risk value matrix. This is the prediction lag time matrix between nodes; Spatial graph convolutional layers are used to capture lateral penetration and vertical coupling between heterogeneous nodes by calculating the attention score of node i to its neighbor node j. Identify key risk transmission paths and combine them with process transmission weights. Causal transmission weight and coupling amplification factor Calculate the comprehensive transmission probability matrix ;in, Let i be the set of neighboring nodes. This represents traversing each node in the set; Starting from the risk source node S, based on Filter out those with a propagation intensity exceeding a preset threshold The directed edge sequence forms a set P; Real-time traffic parameters accessed using the multi-source data sensing module Dynamically update the lag time matrix The elements in ,in, For nodes Physical length of the transmission link between them This is a material flow rate correction factor. The inherent process reaction time of the concentration equipment; the lag time matrix Used to determine the alignment bias when the spatiotemporal graph neural network performs feature aggregation in the time dimension; The risk evolution trend of nodes is modeled using temporal convolutional layers, and the results are output for each node in the future. The risk state at each step is combined with the heterogeneous graph topology G, the historical feature sequence matrix X, and the real-time corrected hysteresis parameter.

1. Input the belt scale flow rate parameter Q, and output the expected risk value matrix A through the fully connected layer. The specific formula is as follows: ,in, This indicates predicting the future t-th time from the current time t. Risk probability distribution for each process node in the entire plant at a given time step. This represents the spatiotemporal graph convolution operator. This represents the normalized exponential function. When the predicted value in A exceeds the safety red line, it automatically locks the target damaged node on the risk path P, providing the collaborative decision-making module with a precise intervention target. Utilizing a dynamic flow adjustment mechanism, the risk prediction module dynamically expands and contracts the time sensing window based on the actual feed load when performing convolution operations. When the system is running at high traffic, reduce This increases the model's response frequency to instantaneous process fluctuations at high flow rates; when the system is operating at low flow rates, it increases... By covering longer historical time-series data, background noise of the sensor is suppressed, ensuring the stability of risk identification results under low-load conditions.

6. The intelligent control and decision-making system for coal preparation plants based on full-process risk prediction according to claim 1, characterized in that: The collaborative decision-making module is used to generate compensatory control parameters based on risk transmission weights within the lag time before the predicted risk reaches the target node, and then distribute them to the implementing agency for preventive intervention. Specifically, it includes the following: Based on the transmission path P and expected risk value matrix A output by the risk prediction module, the compensation amount for the target node is calculated using the following formula: ,in, For the target node, The target node output by the risk prediction module Lag time The expected risk value afterward To start from the risk source node To the target node The conduction path, To control the parameter weight vector, This is the current process state parameter vector. Generates the compensation amount; By utilizing the predicted lag time, the effect of "feedforward control" is achieved through time alignment, with the optimal intervention time being... ,in, For the moment of risk identification, The inherent response time of the actuator from receiving the instruction to completing the action; According to the optimal intervention time Determine when to issue the instruction: ,in, As an emergency threshold, The minimum compensation threshold is used to determine the waiting time for delayed delivery: ; The generated compensation amount is subject to amplitude constraints, rate of change constraints, and risk reduction effectiveness constraints to prevent over-adjustment. The amplitude constraints are as follows: The rate of change constraint is The constraint on the effectiveness of risk reduction is: ,in, The expected risk value after applying compensation. For the target node at time t The generated compensation amount, For the target node Maximum allowable compensation amount per instance This is the compensation amount from the previous moment. For the target node Maximum allowable rate of change of compensation The expected risk value before applying compensation. To minimize the risk reduction; And based on the optimal intervention time The generated compensatory control parameters are then sent to the corresponding actuators.

7. The intelligent control and decision-making system for coal preparation plants based on full-process risk prediction according to claim 1, characterized in that: The feedback-based graph self-optimization module is used to monitor the system response after preventive intervention in real time, and dynamically update the weight parameters of the associated edges in the risk knowledge graph according to the intervention effect. Specifically, it includes the following: Predictive lag time after collaborative decision-making instructions are issued Within the window, feedback signals from damaged nodes are collected, and intervention scores are calculated. ,in, It is the expected risk value before compensation is applied. It is the actual risk value measured by the sensor after the intervention is implemented; It is the technological cost incurred in performing this intervention. The maximum permissible process cost for a single intervention. It is the performance weighting coefficient, when When the intervention has achieved a positive effect, the reduction in risk outweighs the cost; when When the intervention's effect equals its cost, the marginal benefit is zero; when... When this occurs, it indicates that the intervention was ineffective, and the costs outweighed the risks, reducing the benefits. Based on intervention score The gradient ascent method in reinforcement learning is used to analyze the comprehensive propagation probability matrix in the knowledge graph. The correction formula for gradient ascent is as follows: ,in, For learning rate, The baseline score for successful intervention is set. For the updated parameter set, The parameter set before the update, including process propagation weights. Causal transmission weight Coupling interaction coefficient Attention score , The gradient of the intervention score with respect to the parameters; Based on the gradient ascent correction formula described above, intervention scores are calculated after each intervention. and compare it with the preset benchmark score. Comparison: when When this occurs, it indicates that the intervention has achieved better-than-expected results. Increasing the weight parameters of the corresponding transmission path along the positive gradient direction strengthens the confidence of the risk transmission path in future predictions. when When the intervention effect is not as expected, the module reduces the corresponding weight parameters along the negative gradient direction to weaken the influence of the transmission path. when At the same time, the parameters remain unchanged, and through the above mechanism, the process transmission weights can be dynamically adjusted. Causal transmission weight Coupling interaction coefficient and attention score This allows the risk knowledge graph to continuously approximate the actual risk transmission patterns in coal preparation processes, enabling adaptive optimization and continuous evolution of the prediction model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the functional modules of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the functional modules of the intelligent control and decision-making system for coal preparation plants based on full-process risk prediction as described in any one of claims 1-7.

Images