A scheduling method for multi-machine cooperative control of crushing stations based on edge computing

Abstract

A kind of scheduling method of crushing station multi-machine cooperative control based on edge computing belongs to the technical field of mine mechanical equipment automation control.Each device in the crushing station is regarded as an edge node;The edge node communicates with other edge nodes to share the results and carry out multi-machine cooperative control, and the running state and fault of the device are predicted.The invention can make the multi-machine equipment of the crushing station maintain efficient operation under different working conditions, realize adaptive and cooperative work, realize dynamic scheduling and continuous improvement of control strategy, and ensure the global optimization of equipment start-stop and speed optimization.At the same time, by using historical data and real-time data to the greatest extent through reinforcement learning, the continuous improvement of equipment fault prediction and state maintenance is realized, and the unplanned downtime is reduced.The innovation and practicality of the invention in the method of multi-machine cooperative control and scheduling of mine crushing station significantly improve the production efficiency and economic benefit of the system.

Description

Technical Field

[0001] This invention belongs to the field of automated control technology for mining machinery and equipment, and specifically relates to a scheduling method for multi-machine collaborative control of a crushing station based on edge computing. Background Technology

[0002] In traditional mining crushing plants, equipment typically operates independently, resulting in low efficiency in coordination and collaboration between devices. This makes it impossible to dynamically adjust equipment operation based on actual production needs. Traditional control methods rely primarily on manual experience and simple control tools such as PLCs. This approach allows each piece of equipment to operate independently, failing to achieve load balancing and dynamic adjustments to production rhythm. When equipment failures or changes in production demands occur, the response speed of scheduling and control is slow, impacting production efficiency. Furthermore, data between devices is not shared, requiring significant manual intervention and monitoring, and lacking unified monitoring and analysis capabilities.

[0003] Existing methods lack intelligence; traditional control methods lack the ability to analyze and make decisions based on real-time data, and cannot dynamically adjust according to actual operating conditions. Collaborative control capabilities are poor; there is a lack of coordination mechanisms between devices, resulting in each operating independently and overall low efficiency. There is a lack of prediction and optimization; the operating status and malfunctions of equipment cannot be predicted, hindering effective preventative maintenance and production optimization. Data utilization is insufficient; the operational data from each device is not fully utilized for analysis and optimization, resulting in low information utilization. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a scheduling method for multi-machine collaborative control of a crushing plant based on edge computing. This method uses edge computing technology to achieve coordinated control of multiple devices (feeder, crusher, vibrating screen, belt conveyor) within the crushing plant, thereby improving the overall working efficiency and resource utilization of the crushing plant.

[0005] The technical solution adopted in this invention is: a scheduling method for multi-machine collaborative control of a crushing plant based on edge computing. The key technical points include the following steps: taking each piece of equipment in the crushing plant as an edge node, cleaning and detecting anomalies in the operating parameters collected by each edge node; the edge nodes communicate with other edge nodes to share the results and send them to a remote controller for multi-machine collaborative control, predicting the operating status and faults of the equipment, causing the faulty equipment in the crushing plant to stop operating, while other equipment continues to operate according to the predicted operating status.

[0006] In the above scheme, the data cleaning refers to setting upper and lower thresholds for each parameter, and then standardizing the feedback values ​​of each parameter in terms of format and unit before packaging them into a dataset according to function type.

[0007] In the above scheme, the data anomaly detection refers to compensating for the drift amount when the parameter exceeds the lower limit due to drift, and determining the data invalid and reporting a fault when the parameter exceeds the upper limit due to equipment damage or disconnection.

[0008] In the above scheme, the communication between edge nodes and other edge nodes to share results and perform multi-machine collaborative control includes the following steps:

[0009] Each edge node derives positive feedback from the global node through a Markov decision process (MDP) based on real-time data from neighboring nodes, including current, voltage, speed, temperature, vibration, and tension, control commands from current nodes, including start, stop, emergency stop, acceleration, and deceleration, and its own state transition probability.

[0010] In the above scheme, during the process of predicting the operating status and faults of the equipment, if a fault is detected, the control commands of the global node are corrected by the MPC algorithm, and the corrected control commands are then sent to each edge computing node. After receiving the control commands, each edge computing node uses the MILP model to perform mixed solutions on continuous variables including current, voltage, and speed, and discrete variables including start-stop status, to derive the optimal start-stop strategy and speed setting for each global node device, and then executes them.

[0011] The beneficial effects of this invention are as follows: This edge computing-based multi-machine collaborative control scheduling method for crushing plants treats each piece of equipment within the crushing plant as an edge node. The operating parameters collected by each edge node are cleaned and anomaly detected. Edge nodes communicate with other edge nodes to share results and perform multi-machine collaborative control, predicting the operating status and faults of the equipment. This causes faulty equipment within the crushing plant to stop operating, while other equipment operates according to the predicted operating status. This invention enables multi-machine equipment in the crushing plant to maintain efficient operation under different working conditions, achieving adaptive and collaborative work, and continuously improving dynamic scheduling and control strategies, ensuring global optimization of equipment start-up, shutdown, and speed. Simultaneously, by maximizing the use of historical and real-time data through reinforcement learning, it achieves continuous improvement in equipment fault prediction and condition maintenance, reducing unplanned downtime. The innovation and practicality of this invention in multi-machine collaborative control and scheduling methods for mining crushing plants significantly improve the system's production efficiency and economic benefits. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of the scheduling method for multi-machine collaborative control of a crushing station based on edge computing in this embodiment;

[0014] Figure 2 This is a flowchart of the MPC algorithm in this embodiment;

[0015] Figure 3 This is a flowchart of the MILP model in this embodiment. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the following description is provided in conjunction with the accompanying drawings. Figures 1-3 The present invention will be further described in detail below with reference to specific embodiments.

[0017] This embodiment employs a scheduling method for multi-machine collaborative control of a crushing plant based on edge computing, comprising the following steps:

[0018] Step 1: Edge computing node deployment: The key equipment of the crushing plant is used as edge computing nodes, including the feeder, crusher, vibrating screen and belt conveyor, to collect and process the real-time operating data of the equipment.

[0019] Step 2: Data Acquisition and Processing: Edge computing nodes collect operating parameters of the equipment, including current, voltage, speed, temperature, vibration, and tension, and perform preliminary data cleaning and anomaly detection. Data cleaning involves defining upper and lower thresholds for each parameter, standardizing the feedback values ​​of each parameter in format and unit, and then packaging them into datasets according to function type. Data anomaly detection refers to compensating for drift when a parameter exceeds the lower limit due to drift, and determining the data as invalid and reporting a fault when a parameter exceeds the upper limit due to equipment damage or disconnection.

[0020] Step 3: Multi-machine Collaborative Control: Collaborative control between devices is achieved through communication between edge computing nodes. Each edge node determines its own state based on real-time data (current, voltage, speed, temperature, vibration, tension) from neighboring nodes (upstream and downstream process equipment), current control commands from each node (start, stop, emergency stop, acceleration, deceleration), and its own state transition probability, and sends this state to the remote controller. The remote controller derives the positive feedback of the global nodes through a Markov Decision Process (MDP). This positive feedback corrects the control commands of the global nodes using an MPC algorithm, and the corrected control commands are then sent to the global edge computing nodes. For example, when the crusher experiences an abnormal temperature or vibration alarm, the remote controller combines the received information from each edge computing node to perform global control. This includes ensuring that the upstream feeder decelerates or stops according to the crusher's alarm level; while the downstream vibrating screen and belt conveyor maintain normal operation until the existing material has been screened and conveyed before decelerating or stopping.

[0021] By interconnecting edge computing nodes, collaborative work among multiple devices is achieved. Each edge node not only collects operational data from its local equipment but also shares key data from its node, such as the current overall load, current, voltage, and fault status of the crushing plant, with other nodes. The algorithm employs a distributed decision-making model, achieving a balance between local and global decision-making through collaborative computation among multiple edge nodes. The specific steps and algorithm are as follows:

[0022] Step 3.1: Establish a distributed multi-agent system model.

[0023] In the multi-agent system model, the behavior of each agent (feeder, crusher, vibrating screen, belt conveyor) can be described as a Markov decision process (MDP), defined as a quintuple (s, a, P, R, γ).

[0024] The goal of this step is to maximize the accumulated positive feedback G. t G t The model formula is as follows:

[0025]

[0026] In the formula, s is the state space, representing the device state (current, voltage, speed, temperature, vibration, tension) of each edge node; a is the action space, representing the control command set of the device (start, stop, emergency stop, acceleration, deceleration); P(s′|s,a) is the state transition probability, representing the probability of transitioning to state s′ after performing action a in state s; R(s,a) is the feedback function, representing the feedback obtained by performing action a in state s; γ is the coefficient factor, representing the weight of future feedback (0≤γ≤10).

[0027] Agents share state and action information and use policy π(a|s) to determine the best action. The policy can be learned and optimized using Q-network reinforcement learning algorithm.

[0028] Step 3.2: Use the MPC algorithm to predict and optimize the operating status of the equipment.

[0029] In multi-machine collaborative control, the MPC algorithm is used to predict and optimize the operating status of equipment. By making real-time predictions on the system's dynamic model, MPC can optimize control commands based on the current state and future operating conditions, ensuring that each piece of equipment operates under optimal conditions. The steps are broken down as follows: establishing dynamic models of each piece of equipment, including the feeder, crusher, vibrating screen, and belt conveyor, using historical and real-time data; obtaining the control sequence for the future time period by solving the optimization problem within each time step; actually executing the first step of the optimal sequence; and recalculating the optimal sequence in the next step based on the new state.

[0030] The control input model after MPC optimization is as follows:

[0031]

[0032] In the formula, N is the length of the prediction time domain; x k Let x be the state vector of the system at time step k; ref For the target state (such as the desired device state); u k R is the control input for time step k; Q is the weight matrix of the state error; and R is the weight matrix of the control input. That is, if the edge computing node reports that the remote controller cannot complete the instruction, the remote controller optimizes the issued instruction to ensure that the edge computing node in normal working condition continues to work, while the edge computing node with a problem notifies the user for handling.

[0033] Step 4: Dynamic Scheduling Optimization: Based on the production plan and actual operating conditions, a MILP model is used to solve for a mixture of continuous variables (current, voltage, speed) and discrete variables (start-stop status) to derive the optimal start-stop strategy and speed settings for all nodes globally, ensuring the continuity and stability of production. For example, when the crusher experiences an abnormal temperature or vibration alarm, the upstream and downstream equipment have already completed coordinated control according to the control logic, continuously monitoring the abnormal alarm status of the crusher. If the alarm has been downgraded or resolved, the dynamic feedback control logic restores the operating speed of the upstream and downstream equipment. This process is a closed-loop monitoring system used to reduce unnecessary downtime.

[0034] In scheduling optimization, the MILP model is applied to optimize equipment start-up, shutdown, and operating speed. MILP achieves the global optimal solution for scheduling optimization by solving a mixture of continuous variables (current, voltage, speed) and discrete variables (start-up and shutdown states). When the production plan of the crushing plant changes, the MILP model can quickly solve for the optimal start-up and shutdown strategies and speed settings for each piece of equipment, ensuring the balance of material flow and the continuity of production. The steps are broken down as follows:

[0035] Real-time data collected from edge nodes serves as the input to the environmental status; the RL algorithm learns the optimal scheduling strategy through interaction with the environment and adjusts the operating mode of the equipment. Based on multi-dimensional evaluation indicators such as production efficiency, energy consumption, and equipment health, a feedback function is designed to guide the scheduling strategy towards convergence towards the optimal direction.

[0036] The feedback mechanism adopts the MILP model, and the feedback function model is as follows:

[0037]

[0038] In the formula, T is the length of the scheduling time domain; xt is the continuous variable vector (equipment current, voltage, speed) at time step t; yt is the integer variable vector (equipment start / stop status) at time step t; Ct is the cost vector of the continuous variables; and Ft is the cost vector of the integer variables.

[0039] Step 5: Remote Monitoring and Feedback: Upload the processed data from the edge nodes to the cloud for remote monitoring and global optimization analysis. Managers can view production status in real time through the SCADA / HMI interface and perform remote control and adjustments.

[0040] Step 6: Fault Prediction and Maintenance: Based on historical and real-time data, Q-network reinforcement learning is used to provide technical support for equipment health status assessment and predicting equipment fault maintenance cycles, reducing unplanned downtime. The specific implementation process is as follows: Collect historical data (equipment operating parameters and fault records, etc.) and real-time data (current, voltage, temperature, vibration, etc.), clean and standardize the data, and define the current state of the equipment. The current state includes three categories: normal, minor fault, and severe fault. Input the current state of the equipment into the constructed Q-network learning model to obtain the Q-values ​​for different maintenance strategies. Based on the Q-values, predict the probability of equipment failure and the optimal maintenance time (early or late), and generate maintenance cycle suggestions (updating the original maintenance plan).

[0041] The Q-function is used to perform reinforcement learning on historical and real-time data to achieve continuous improvement in equipment failure prediction and condition-based maintenance, thereby reducing unplanned downtime. The learning model is as follows:

[0042]

[0043] In the formula, Q(s,a) is the Q-value of performing action a in state s; α is the learning rate; r is the immediate feedback; γ is the coefficient factor; max a′ Q(s′,a′) is the optimal Q value for the next state s′.

[0044] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A scheduling method for multi-machine collaborative control of a crushing plant based on edge computing, characterized in that, Includes the following steps: Each piece of equipment in the crushing plant is treated as an edge node. The operating parameters collected by each edge node are cleaned and anomalies are detected. The edge nodes communicate with other edge nodes to share the results and send them to the remote controller for multi-machine collaborative control. The operating status and faults of the equipment are predicted, so that the faulty equipment in the crushing plant stops running, while the other equipment continues to run according to the predicted operating status. The process of the edge nodes communicating with other edge nodes to share results and perform multi-machine collaborative control includes the following steps: Each edge node derives positive feedback from the global node through a Markov decision process based on real-time data from neighboring nodes, including current, voltage, speed, temperature, vibration, and tension, control commands from current nodes, including start, stop, emergency stop, acceleration, and deceleration, and its own state transition probability. In the process of predicting the operating status and faults of the equipment, if a fault is detected, the control commands of the global node are corrected by the MPC algorithm, and the corrected control commands are then sent to each edge computing node. After receiving the control commands, each edge computing node uses the MILP model to perform mixed solutions on continuous variables including current, voltage, and speed and discrete variables including start-stop status, and derives the optimal start-stop strategy and speed of each global node device before continuing to execute.

2. The scheduling method for multi-machine collaborative control of a crushing station based on edge computing as described in claim 1, characterized in that, The data cleaning process involves setting upper and lower thresholds for each parameter, standardizing the format and units of the feedback values ​​of each parameter, and then encapsulating them into datasets according to their functional types.

3. The scheduling method for multi-machine collaborative control of a crushing station based on edge computing as described in claim 1, characterized in that, The aforementioned data anomaly detection refers to compensating for the drift amount when the parameter exceeds the lower limit due to drift, and determining the data as invalid and reporting a fault when the parameter exceeds the upper limit due to equipment damage or disconnection.

Images