A real-time scheduling method and system for a strip mine truck shovel based on multi-source data fusion

Abstract

This invention provides a real-time scheduling method and system for open-pit mine shovels based on multi-source data fusion, comprising: fusing multi-source heterogeneous data from the acquired open-pit mine production system to obtain a fused data base; performing event extraction and state association processing on the fused data base to construct a unified production status data model; generating optimal production task orders for trucks entering an idle state using a multi-dimensional dynamic decision model based on the production status data model; automatically acquiring and recording actual operation data of each stage of loading, transportation, and unloading corresponding to the production task order during execution; processing the fused data base and the full-chain execution data using a graph neural network method to establish a production efficiency evaluation model; evaluating the optimal production task order and its execution process based on the production efficiency evaluation model, and feeding the evaluation results back to the multi-dimensional dynamic decision model to optimize decision parameters, forming a closed-loop control system.

Description

Technical Field

[0001] This invention relates to the field of open-pit mining production technology, specifically to a real-time scheduling method and system for open-pit mine shovels based on multi-source data fusion. Background Technology

[0002] In open-pit mining operations, the dispatching of excavators and shovels is a core aspect affecting production efficiency and operating costs. Currently, the dispatching methods commonly used in the industry mainly include manual dispatching, which relies on human experience, and basic auxiliary dispatching systems based on vehicle positioning. The manual dispatching mode mainly relies on dispatchers communicating with drivers via walkie-talkies and assigning tasks based on their personal experience. The basic auxiliary dispatching system, on the other hand, installs GPS positioning terminals on vehicles to provide dispatchers with visual information on vehicle locations, assisting them in making decisions based on simple rules such as the shortest path or fixed vehicle allocation.

[0003] However, its practical application has several inherent limitations. First, the decision-making process relies solely on single or a few location or simple rule-based information, failing to incorporate multi-dimensional dynamic factors such as the excavator's real-time loading capacity, material loading priority, dynamic transportation time cost, and the current status of the excavator's bucket into a systematic optimization consideration. This results in uneven equipment utilization and makes it difficult to maximize overall production efficiency. Second, for key production links such as loading and unloading, it still heavily relies on manual observation and reporting, making it impossible to achieve automatic, accurate, and non-interference collection of core production data across the entire loading, transportation, and unloading chain, resulting in insufficient data objectivity and timeliness. Furthermore, multi-source data from positioning terminals, video surveillance, sensors, and production planning systems are isolated from each other, forming information barriers and making decision-making lack comprehensive and integrated data support. Therefore, this invention studies and designs a real-time scheduling method and system for open-pit mine truck shovels based on multi-source data fusion. Summary of the Invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the defects of one-sided decision-making basis and inaccurate data collection in the prior art, thereby providing a real-time scheduling method and system for open-pit mine shovels based on multi-source data fusion.

[0005] To address the aforementioned problems, this invention provides a real-time scheduling method for open-pit mine shovels based on multi-source data fusion, comprising the following steps: S1: The multi-source heterogeneous data of the acquired open-pit mine production system are fused to obtain a fused data base. S2: Perform event extraction and state association processing on the fused data base to construct a unified production status data model; S3: Based on the production status data model, use a multi-dimensional dynamic decision-making model to generate the optimal production task order for trucks that have entered the idle state; S4: Automatically acquire and record the actual operation data of each stage of loading, transportation and unloading corresponding to the production task order during the execution process; S5: Use graph neural network methods to process the fused data base and the full-chain execution data to establish a production efficiency evaluation model; S6: Based on the production efficiency evaluation model, evaluate the optimal production task order and its execution process, and feed the evaluation results back to the multidimensional dynamic decision-making model to optimize the decision parameters; The multidimensional dynamic decision-making model is based on a comprehensive decision based at least on excavator demand, dynamic transportation cost, and capacity matching. The excavator demand is determined based on the bucket status of the candidate excavator, the queuing status of waiting trucks, and the priority of the loaded materials. The bucket status indicates whether the excavator bucket is in a state of being raised or suspended, waiting for a truck to enter the loading position. The queuing status of waiting trucks includes at least the number of trucks waiting to be loaded in the candidate excavator's operating area. The dynamic transportation cost is calculated based on the real-time road network status. The capacity matching is determined based on the historical loading data of the candidate excavator and the predicted loading completion time using real-time visual data.

[0006] Preferably, in step S1, before fusing the acquired multi-source heterogeneous data, the acquisition of the multi-source heterogeneous data is performed through: Beidou / GPS positioning terminals installed on trucks and excavators acquire real-time location, speed, and timestamp data; High-definition cameras installed in the excavator's operating area acquire truck identification information, bucket movement trajectory, and cargo box loading status; Ultrasonic sensors installed at the unloading point detect the lifting and resetting status of the truck bed; The data interface, which connects with the mine production planning system, obtains the excavator's theoretical loading capacity, material type, and priority.

[0007] Preferably, the process of fusing the acquired multi-source heterogeneous data to obtain a fused data base is as follows: Spatiotemporal alignment of the spatiotemporal positioning data, visual data, sensor data, and business data is performed based on a unified timestamp. A complex event processing engine is used to correlate and fuse the aligned multi-source data to obtain the fused data base.

[0008] Preferably, the step of generating optimal production task orders for trucks entering an idle state based on the production status data model and utilizing a multi-dimensional dynamic decision model includes: The excavator demand is determined based on the excavator status data and material priority in the production status data model. The dynamic travel time of trucks to each candidate excavator is calculated based on real-time road network conditions and vehicle locations to determine dynamic transportation costs. The loading completion time is predicted based on historical loading data and real-time visual data of the excavator in order to determine the capacity matching degree. The excavator demand, dynamic transportation cost, and capacity matching degree are weighted and comprehensively scored, and the candidate excavator with the highest score is selected to generate the optimal production task order.

[0009] Preferably, the automatic acquisition and recording of actual operational data for each stage of loading, transportation, and unloading during the execution of the production task order includes: During the loading process, the spatial relationship between the bucket and the truck is identified by a target detection model, and the unloading action is identified by a time-series classification model. The loading start time and loading end time are recorded. During the transportation process, the actual transport distance is automatically calculated based on the vehicle's positioning trajectory, and the heavy-load travel time and average speed are recorded. During the unloading process, ultrasonic sensors are used to detect the lifting and resetting status of the carriage, and the unloading start time and unloading end time are recorded.

[0010] Preferably, the construction of the production efficiency evaluation model includes: The equipment status and topology in the production status data model and the time and efficiency indicators in the full-chain execution data are used as inputs; Machine learning methods are used to establish an evaluation model for assessing the performance of a single transportation task and the overall system.

[0011] Preferably, optimizing the decision parameters based on the performance evaluation results includes: The entire chain execution data is bound to the corresponding production task order to generate a transportation record; An error function is constructed based on historical launch records, and the weight parameters of the multidimensional dynamic decision-making model are adjusted using an optimization algorithm.

[0012] Preferably, after generating and executing the optimal production task order, the method further includes: Real-time monitoring of the operating status data of each truck and excavator; Perform fault diagnosis and risk assessment on abnormal state data; Generate fault reports and maintenance recommendations, and send them to the maintenance terminal.

[0013] Preferably, after generating and executing the optimal production task order, the method further includes: Perform threshold analysis on the time consumption of each step in the execution result to identify abnormal steps; Locate abnormal devices based on device topology and geographic location information; The location and diagnostic information of the malfunctioning equipment are pushed to the dispatch terminal and maintenance terminal.

[0014] This invention also provides a real-time scheduling method for open-pit mine shovels based on multi-source data fusion, used in the real-time scheduling system for open-pit mine shovels based on multi-source data fusion described in the preceding claim, comprising: The data fusion module is used to fuse the multi-source heterogeneous data acquired from the open-pit mine production system to obtain a fused data base; the multi-source heterogeneous data includes at least the spatiotemporal positioning data of vehicles and excavators, visual data of the excavator's operating area, the status sensor data of the unloading point's cargo box, and production planning business data. The situation modeling module is used to extract events and associate states with the fused data base to build a unified production situation data model. The intelligent decision-making module is used to generate optimal production task orders for trucks entering an idle state based on the production status data model and using a multi-dimensional dynamic decision-making model; wherein, the multi-dimensional dynamic decision-making model makes comprehensive decisions based at least on excavator demand, dynamic transportation costs, and capacity matching. The process perception module is used to automatically collect the entire chain execution data of loading, transportation and unloading during the execution of the production task order through visual recognition and sensing detection technology; The efficiency evaluation module is used to construct a production efficiency evaluation model based on the production status data model and the full-chain execution data, and to evaluate the efficiency of the optimal production task order and its execution process. The closed-loop optimization module is used to optimize the decision parameters of the multidimensional dynamic decision-making model based on the performance evaluation results.

[0015] The real-time scheduling method and system for open-pit mine shovels based on multi-source data fusion provided by this invention has the following beneficial effects: 1. This invention acquires full data of mine production through multi-source heterogeneous data acquisition equipment, and forms a unified fusion data base through spatiotemporal alignment and complex event processing; then, it constructs a production status data model through event extraction and state association to realize a digital twin of the physical world production process; on this basis, a multi-dimensional dynamic decision-making model comprehensively considers excavator demand, transportation costs and capacity matching to generate the optimal task order for idle trucks. 2. This invention also automatically collects data from the entire chain during task execution through visual recognition and sensor detection; after execution, it uses graph neural networks to evaluate production efficiency and feeds the results back to optimize decision parameters, forming a closed-loop control system. 3. This invention also ensures production safety and reliable equipment operation by enabling the system to perform real-time fault diagnosis and anomaly location. It achieves full-process digitalization and intelligentization of open-pit mine shovel scheduling, multi-source data fusion ensures the comprehensiveness and accuracy of decision-making basis, multi-dimensional dynamic decision-making model significantly improves equipment utilization and production efficiency, full-chain automatic data collection ensures the objectivity and timeliness of production data, graph neural network evaluation model provides scientific performance evaluation methods, and closed-loop optimization mechanism enables the system to have continuous self-improvement capabilities. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system structure of the present invention; Figure 3 This is a schematic diagram illustrating the system dispatching of the present invention; Figure 4 This is a schematic diagram of the truck transport status of the system of the present invention; Figure 5 This is a schematic diagram of the excavator loading process for the system of the present invention. Detailed Implementation

[0017] like Figure 1 As shown, this invention provides a real-time scheduling method for open-pit mine shovels based on multi-source data fusion, which includes the following steps: S1: The multi-source heterogeneous data of the acquired open-pit mine production system are fused to obtain a fused data base. S2: Perform event extraction and state association processing on the fused data base to construct a unified production status data model; S3: Based on the production status data model, use a multi-dimensional dynamic decision-making model to generate the optimal production task order for trucks that have entered the idle state; S4: Automatically acquire and record the actual operation data of each stage of loading, transportation and unloading corresponding to the production task order during the execution process; S5: Use graph neural network methods to process the fused data base and the full-chain execution data to establish a production efficiency evaluation model; S6: Based on the production efficiency evaluation model, evaluate the optimal production task order and its execution process, and feed the evaluation results back to the multidimensional dynamic decision-making model to optimize the decision parameters; The multidimensional dynamic decision-making model is based on a comprehensive decision based at least on excavator demand, dynamic transportation cost, and capacity matching. The excavator demand is determined based on the bucket status of the candidate excavator, the queuing status of waiting trucks, and the priority of the loaded materials. The bucket status indicates whether the excavator bucket is in a state of being raised or suspended, waiting for a truck to enter the loading position. The queuing status of waiting trucks includes at least the number of trucks waiting to be loaded in the candidate excavator's operating area. The dynamic transportation cost is calculated based on the real-time road network status. The capacity matching is determined based on the historical loading data of the candidate excavator and the predicted loading completion time using real-time visual data.

[0018] This application acquires full-volume data from mine production using multi-source heterogeneous data acquisition equipment, and forms a unified fusion data base through spatiotemporal alignment and complex event processing. Then, it constructs a production status data model through event extraction and state correlation, realizing a digital twin of the physical world's production process. Based on this, a multi-dimensional dynamic decision-making model comprehensively considers excavator demand, transportation costs, and capacity matching to generate optimal task orders for idle trucks. During task execution, data from the entire chain is automatically collected through visual recognition and sensor detection. After execution, graph neural networks are used to evaluate production efficiency, and the results are fed back to optimize decision parameters, forming a closed-loop control system. Simultaneously, the system possesses real-time fault diagnosis and anomaly location capabilities, ensuring production safety and reliable equipment operation, realizing full-process digitalization and intelligentization of open-pit mine truck shovel scheduling. Multi-source data fusion ensures the comprehensiveness and accuracy of decision-making basis, the multi-dimensional dynamic decision-making model significantly improves equipment utilization and production efficiency, the full-chain automatic data acquisition ensures the objectivity and timeliness of production data, the graph neural network evaluation model provides scientific performance evaluation methods, and the closed-loop optimization mechanism enables the system to have continuous self-improvement capabilities.

[0019] In some implementations, in step S1, before fusing the acquired multi-source heterogeneous data, the acquisition of the multi-source heterogeneous data is achieved through: obtaining real-time location, speed, and timestamp data from Beidou / GPS positioning terminals installed on trucks and excavators; obtaining truck identification information, bucket movement trajectory, and cargo box loading status from high-definition cameras installed in the excavator's operating area; detecting the lifting and resetting status of the truck cargo box from ultrasonic sensors installed at the unloading point; and obtaining the excavator's theoretical loading capacity, material type, and priority from a data interface connected to the mine production planning system.

[0020] Specifically, before fusing the acquired multi-source heterogeneous data, it is necessary to obtain equipment data from various acquisition systems for the target production system. For the acquisition of multi-source heterogeneous data: spatiotemporal positioning data of vehicles and excavators is collected through Beidou / GPS dual-mode positioning terminals installed in the cabs of trucks and excavators; visual data of the excavator's operating area is collected through high-definition network cameras installed on the excavator boom and the top of the cab, with the installation positions optically calibrated to ensure coverage of the entire operating area within the excavator's turning radius; truck bed status sensing data at the unloading point is collected through an ultrasonic sensor array installed on the edge of the unloading retaining wall or unloading platform to detect the lifting angle and reset status of the truck bed; production planning business data is obtained through a standardized data interface with the mine production planning management system, using the OPC protocol. UA or MQTT, such as mining ERP or MES systems, provides structured data including shift production plans, material type codes, priority levels, and excavator theoretical loading capacity. In practical applications, positioning terminals, cameras, sensors, and system interfaces can be selected from other models and brands that meet technical requirements; this application does not limit this. This application constructs a data acquisition network covering the entire mining, transportation, and unloading process by deploying multiple types of sensors and data interfaces, providing a data foundation for subsequent data fusion and intelligent decision-making, and achieving precise mapping from physical production to digital space. It ensures the integrity and accuracy of multi-source data; Beidou / GPS positioning provides a sub-meter level spatiotemporal reference; high-definition visual data enables refined perception of the operation process; ultrasonic sensor data ensures reliable detection of unloading status; and standardized data interfaces break down information silos, laying a solid data foundation for intelligent scheduling.

[0021] In some implementations, the process of fusing the acquired multi-source heterogeneous data to obtain a fused data base involves: spatiotemporally aligning the spatiotemporal positioning data, visual data, sensor data, and business data based on a unified timestamp; and using a complex event processing engine to associate and fuse the aligned multi-source data to obtain the fused data base.

[0022] Specifically, key business events are extracted from the fused data base, including truck location change events, excavator bucket movement events, truck bed loading status change events, and unloading completion events. Status correlation processing is achieved by establishing topological relationships between equipment, including: the spatial proximity relationship between trucks and excavators, the material flow relationship between excavators and unloading points, and the binding relationship between trucks and tasks. Spatial proximity is determined by calculating an Euclidean distance less than a preset threshold, and the material flow relationship between excavators and unloading points is determined based on the mapping table from mining areas to crushing stations in the production plan. Through these correlations, a unified production status data model is constructed, represented by a graph structure. Nodes include truck nodes, excavator nodes, and unloading point nodes; edges include spatial adjacency edges, task allocation edges, and material flow edges; node attributes include real-time status (idle, driving, loading, unloading), location coordinates, and capacity parameters; edge attributes include distance, estimated travel time, and material type. By unifying the time base and spatial coordinate system, temporal deviations and spatial misalignments of multi-source data are eliminated; the pattern matching and association rules of the complex event processing engine transform discrete multi-source data streams into fused events with business semantics, constructing a structured fused data foundation, achieving high-precision spatiotemporal alignment of multi-source heterogeneous data, and providing reliable data support for real-time production situation awareness.

[0023] In some implementations, generating optimal production task orders for trucks entering an idle state using a multi-dimensional dynamic decision-making model based on the production status data model includes: The excavator demand is determined based on excavator status data and material priority in the production status data model; the dynamic travel time of trucks to each candidate excavator is calculated based on real-time road network status and vehicle location to determine dynamic transportation costs; the loading completion time is predicted based on historical excavator loading data and real-time visual data to determine capacity matching; simultaneously, before scoring candidate excavators, for each candidate excavator, a corresponding local sub-graph (including the candidate excavator, its currently serving truck queue, loading points, unloading points, and road network connections) is extracted from the production status data model to construct a candidate task graph; the candidate task graph is then input into... In the pre-trained graph neural network, which adopts a graph convolutional network architecture, node features include equipment type, location, status, and capability parameters, while edge features include distance and estimated travel time. After passing through multiple layers of graph convolution, a fully connected layer outputs a predicted value, predicting the estimated waiting time, estimated loading time, and estimated transportation time after the truck is assigned to the excavator. Finally, a weighted comprehensive score is applied to the excavator demand, dynamic transportation cost, capability matching degree, and the estimated waiting time, estimated loading time, and estimated transportation time predicted by the graph neural network. The candidate excavator with the highest score is selected to generate the optimal production task order.

[0024] Specifically, The multidimensional dynamic decision-making model takes production status data as input and outputs the optimal production task order. The multidimensional dynamic decision-making model makes comprehensive decisions based at least on excavator demand, dynamic transportation costs, and capacity matching, as detailed below: The calculation of excavator demand comprehensively considers the bucket status of candidate excavators, the queuing status of waiting trucks, and the priority of loaded materials. The bucket status, which characterizes whether the excavator bucket is in a raised or suspended state awaiting a truck to enter the loading position, is obtained through visual data analysis, specifically using Faster... The R-CNN object detection model identifies the spatial position and attitude of the bucket: when the bucket is detected to be in a raised or hovered state and no truck enters the loading position within the excavator's turning radius, it is determined to be in a "waiting ready" state, with a corresponding demand coefficient of 1.0; when the bucket is in a loading or unloading state, it is determined to be in a "working" state, with a corresponding demand coefficient of 0.5; when the material loading in the bucket exceeds a threshold (e.g., full bucket) and no truck receives it, the corresponding demand coefficient is further increased to 1.2; the waiting truck queuing situation includes at least the number of trucks waiting to be loaded in the candidate excavator's working area. This number is calculated by fusing real-time truck positioning data in the data base with the spatial range of the excavator's working area. The more trucks in the queue, the lower the demand of the excavator (demand coefficient multiplied by a queuing decay factor of 0.8); material priority is directly taken from production planning business data. High-priority materials correspond to a demand weighting coefficient of 1.5, ordinary priority materials to 1.0, and low priority materials to 0.7; The calculation of dynamic transportation cost is based on real-time road network status, which is obtained by aggregating the positioning trajectories of all trucks in the fusion data base. Specifically, the Dijkstra algorithm is used to calculate the optimal path from the current location of the truck to each candidate excavator. The travel time estimation comprehensively considers the path length, road gradient, and current road congestion. The dynamic transportation cost takes the estimated travel time as a quantitative indicator. The capacity matching degree is calculated based on the historical loading data of the candidate excavator and the predicted loading completion time using real-time visual data. The historical data includes the average loading time of the excavator over the past 30 days. The real-time visual data is used to detect the current truck bed model and the amount already loaded, and to predict the remaining loading time. The shorter the expected loading completion time, the higher the capacity matching degree. Simultaneously, before scoring candidate excavators, for each candidate excavator, a corresponding local subgraph (including the candidate excavator, its currently serving truck queue, loading point, unloading point, and road network connection relationship) is extracted from the production status data model to construct a candidate task graph. The candidate task graph is then input into a pre-trained graph neural network. This graph neural network adopts a graph convolutional network architecture. Node features include equipment type, location, status, and capability parameters, while edge features include distance and estimated travel time. After passing through multiple graph convolutions, a fully connected layer outputs a predicted value, predicting the estimated waiting time, estimated loading time, and estimated transportation time after the truck is assigned to the excavator. Finally, a weighted comprehensive score is calculated on the excavator demand, dynamic transportation cost (converted to a positive indicator by taking the reciprocal), capacity matching degree, and the expected waiting time, expected loading time, and expected transportation time predicted by the graph neural network. The weight coefficients are obtained by training with historical data. The candidate excavator with the highest score is selected to generate the optimal production task order. The task order includes the target excavator ID, driving route, expected loading time, and material type.

[0025] In some implementations, the automatic acquisition and recording of actual operational data for each stage of loading, transportation, and unloading during the execution of the production task order includes: in the loading stage, identifying the spatial relationship between the bucket and the truck bed using a target detection model and identifying the unloading action using a time-series classification model, and recording the loading start time and loading end time; in the transportation stage, automatically calculating the actual transport distance based on the vehicle positioning trajectory and recording the heavy-load travel time and average speed; and in the unloading stage, detecting the lifting and resetting status of the truck bed using an ultrasonic sensor and recording the unloading start time and unloading end time.

[0026] Specifically, during the loading phase, an edge computing device deployed in the excavator's operating area runs a target detection model to identify the spatial relationship between the bucket and the truck bed in real time. When the bucket is detected to be directly above the truck bed and the distance is less than a preset threshold, loading is considered to have started, and the loading start time is recorded. Unloading actions are identified using a temporal classification model, such as an LSTM-based action recognition network. When the bucket's tilt angle is greater than 45 degrees and the material pixel ratio decreases significantly, loading is considered to have ended, and the loading end time is recorded. During the transportation phase, the actual transport distance is automatically calculated based on the vehicle's positioning trajectory, obtained by accumulating the Euclidean distance of the trajectory point sequence. The heavy-load travel time and average speed are recorded. The loading and unloading travel time is the time difference between the loading end timestamp and the unloading start timestamp. The average speed is the actual transport distance divided by the heavy-load travel time. During the unloading process, ultrasonic sensors detect the lifting and resetting status of the truck bed. When the sensor detects that the distance between the truck bed floor and the sensor rapidly decreases from the reference value, it is determined that lifting has started, which means unloading has begun. When the distance returns to near the reference value and stabilizes after a certain period of time, it is determined that resetting is complete, which means unloading has ended. The unloading start time and unloading end time are recorded respectively. This achieves automated data collection for the entire loading, transportation, and unloading process. It can accurately record the start and end times and operation parameters of each stage without manual intervention, and build a complete, objective, and real-time production execution data chain.

[0027] In some implementations, the construction of the production efficiency evaluation model includes: using the equipment status and topology relationships in the production status data model and the time and efficiency indicators in the full-chain execution data as inputs; and using machine learning methods to establish an evaluation model for evaluating the efficiency of a single transportation task and the overall system.

[0028] Specifically, the production efficiency evaluation model takes the equipment status and topological relationships from the production status data model, along with the time and efficiency indicators from the full-chain execution data, as inputs. Equipment status includes the historical sequence of the operating status of each piece of equipment; topological relationships include the connection graph structure between equipment; time indicators include the single transportation cycle time and the waiting time of each link; and efficiency indicators include hourly output and equipment utilization rate. The single transportation cycle time is the sum of loading time, transportation time, and unloading time, and the equipment utilization rate is the actual operating time divided by the available time. The evaluation model is built using a graph neural network method, specifically adopting a graph convolutional network or graph attention network architecture, with a node embedding dimension of 128 and a layer count of 3. The model learns the mutual influence relationships between equipment by aggregating neighbor node information. The model training uses historical data as samples, with manually labeled efficiency levels. The loss function is cross-entropy loss, and the optimizer is Adam. The model output is the efficiency score of a single transportation task and the overall system efficiency index.

[0029] Building upon this foundation, the present invention further applies graph neural networks to the closed loop of pre-dispatch prediction and post-execution evaluation. Pre-dispatch prediction: a candidate task graph is constructed for each candidate excavator, input into the graph neural network, and the estimated waiting time, estimated loading time, and estimated transportation time are output. Prediction vs. Actual Comparison: after task execution, the actual collected full-chain execution data (waiting time, loading time, transportation time, and unloading time) is compared with the corresponding predicted times output by the graph neural network before dispatch, and the prediction deviations at each stage (such as absolute error and relative error) are calculated. Simultaneously, the task performance score is calculated based on the cycle time (loading + transportation + unloading), waiting time, and equipment utilization rate. Training sample composition: The production status map at the time of task generation is used as the input sample, the corresponding task plan (such as target excavator ID, driving route, etc.) is used as the conditional features, and the actual waiting time, loading time, transportation time, and unloading time obtained after task execution are used as supervision labels to form new training samples; Incremental training: The system periodically (e.g., daily or per shift) uses the accumulated new training samples to incrementally train the graph neural network (using elastic weight consolidation or progressive learning strategies to avoid catastrophic forgetting), updates the network weight parameters, and deploys the updated model to the online inference service for prediction of subsequent candidate solutions, thereby continuously improving the prediction accuracy and the scientific nature of scheduling decisions.

[0030] In some implementations, optimizing decision parameters based on performance evaluation results includes: binding the full-chain execution data with the corresponding production task order to generate a transport record; constructing an error function based on historical transport records; and adjusting the weight parameters of the multidimensional dynamic decision model using an optimization algorithm.

[0031] Specifically, the optimization of decision parameters binds the entire chain of execution data with the corresponding production task orders, generating transport records. These records include fields such as task order ID, truck ID, excavator ID, planned start time, actual start time, planned completion time, actual completion time, and time consumed at each stage. An error function is constructed based on historical transport records, defined as the mean square error between planned and actual times. Optimization algorithms are used to adjust the weight parameters of the multidimensional dynamic decision model. These algorithms employ Bayesian optimization or genetic algorithms, aiming to minimize the error function by iteratively searching for the optimal weight combination. The optimized weight parameters are automatically updated into the decision model, achieving closed-loop optimization. This application quantifies the accuracy of decisions by binding task plans and actual execution data; it measures the deviation between the plan and reality by constructing an error function; and it automatically searches for optimal decision weights using efficient optimization algorithms such as Bayesian optimization, achieving data-driven self-optimization of the decision model and enabling the scheduling system to continuously learn and adapt. The decision model parameters can be automatically adjusted and optimized based on actual production data. Meanwhile, incremental training of graph neural networks is also an important part of closed-loop optimization: by continuously accumulating new production situation maps, task plans, and actual execution time samples, and regularly updating network parameters, the prediction model can adapt to the dynamic evolution of mine production conditions (such as road changes, equipment performance degradation, and material distribution changes), thereby improving the accuracy of pre-dispatch predictions, further optimizing the comprehensive scoring effect of the multi-dimensional dynamic decision-making model, and realizing continuous improvement of the entire closed loop of prediction, execution, evaluation, learning, and optimization.

[0032] In some implementations, after generating and executing the optimal production task order, the method further includes: real-time monitoring of the operating status data of each truck and excavator; fault diagnosis and risk assessment of abnormal status data; generating fault reports and maintenance recommendations, and sending them to the maintenance terminal.

[0033] Specifically, real-time monitoring and fault diagnosis of operational status involves monitoring the operational status data of each truck and excavator in real time, including engine speed, oil temperature, hydraulic pressure, abnormal positioning trajectory, and visual data anomalies. Hydraulic pressure is obtained through the vehicle's OBD interface or CAN bus. Abnormal positioning trajectory includes prolonged stationary states or deviations from the planned route, while visual data anomalies include camera obstruction or blurriness. Fault diagnosis and risk assessment are performed on abnormal status data, using rule-based diagnostic logic (e.g., "engine coolant temperature exceeding 100 degrees Celsius for 5 consecutive minutes" is considered an engine overheating fault) or machine learning-based anomaly detection models, such as the isolated forest algorithm. Fault reports and maintenance suggestions are generated, including equipment ID, fault type, fault level, and suggested handling measures. These reports are sent to maintenance terminals, such as handheld PDAs or fixed workstations, via 4G / 5G networks or a dedicated mining network. By monitoring equipment health status in real time through multi-source data, performing fault diagnosis and risk assessment through rule engines and machine learning models, and achieving rapid scheduling of maintenance resources through automated reporting and message push, passive reactive maintenance is transformed into proactive predictive maintenance.

[0034] In some implementations, after generating and executing the optimal production task order, the method further includes: performing threshold analysis on the time consumption of each step in the execution result to identify abnormal steps; locating abnormal equipment based on equipment topology and geographical location information; and pushing the location and diagnostic information of the abnormal equipment to the scheduling terminal and maintenance terminal.

[0035] Specifically, for the identification and location of abnormal processes, threshold analysis is performed on the time consumption of each step in the execution result. Loading time thresholds, transportation time thresholds (dynamically calculated based on transport distance, threshold = transport distance / minimum permissible speed), and unloading time thresholds are set. Exceeding these thresholds indicates an abnormal process. Abnormal equipment is located based on equipment topology and geographic location information, specifically through equipment association tracing in the production status data model. The location and diagnostic information of abnormal equipment are pushed to the dispatching and maintenance terminals. Abnormal processes are automatically identified by setting dynamic thresholds, and the responsible equipment is quickly located based on equipment topology relationships. Geographic location information is integrated to provide accurate location services, and real-time push notifications through multiple channels ensure that relevant personnel are promptly informed and can handle the abnormality, shortening the abnormality response time.

[0036] like Figure 2-5 As shown, the present invention also provides a real-time scheduling method for open-pit mine shovels based on multi-source data fusion, used in the real-time scheduling system for open-pit mine shovels based on multi-source data fusion described in the preceding claim, comprising: The data fusion module is used to fuse the multi-source heterogeneous data acquired from the open-pit mine production system to obtain a fused data base; the multi-source heterogeneous data includes at least the spatiotemporal positioning data of vehicles and excavators, visual data of the excavator's operating area, the status sensor data of the unloading point's cargo box, and production planning business data. The situation modeling module is used to extract events and associate states with the fused data base to build a unified production situation data model. The intelligent decision-making module is used to generate optimal production task orders for trucks entering an idle state based on the production status data model and using a multi-dimensional dynamic decision-making model; wherein, the multi-dimensional dynamic decision-making model makes comprehensive decisions based at least on excavator demand, dynamic transportation costs, and capacity matching. The process perception module is used to automatically collect the entire chain execution data of loading, transportation and unloading during the execution of the production task order through visual recognition and sensing detection technology; The efficiency evaluation module is used to construct a production efficiency evaluation model based on the production status data model and the full-chain execution data, and to evaluate the efficiency of the optimal production task order and its execution process. The closed-loop optimization module is used to optimize the decision parameters of the multidimensional dynamic decision-making model based on the performance evaluation results. Specifically, the data fusion module utilizes a server cluster deployed in the mine's data center, including data acquisition servers, data processing servers, and time-series database servers. The software architecture adopts a microservice architecture, deployed on a Kubernetes container orchestration platform. Services communicate via RESTful APIs, and the message queue uses Apache Kafka to achieve asynchronous data flow. The situation modeling module deploys a complex event processing engine, configures a business rule base, builds a real-time graph database, stores the graph structure of the production situation data model, and supports querying and updating. The intelligent decision-making module deploys a decision engine service, configures multi-dimensional dynamic decision algorithms, and obtains real-time situation data from the situation modeling module via API. After generating task orders, it pushes them to the vehicle terminal and dispatch terminal via a message queue. The process perception module deploys edge computing nodes at the mine site, with each node covering 2-3 excavators or 1-2 unloading points. It runs a deep learning inference framework, such as TensorRT, to achieve local real-time computation of visual recognition and sensor data processing, reducing network transmission latency. Edge nodes are connected to the data center via a fiber optic ring network or a 5G private network to synchronize data and model updates. The performance evaluation module deploys a model training server for offline training of the graph neural network model; it also deploys an online inference service to load the trained model, receive production data in real time, and output performance scores. The model is periodically incrementally trained and updated based on newly accumulated data. The closed-loop optimization module deploys a parameter optimization service with a Bayesian optimization algorithm library, periodically executes optimization tasks, and synchronizes the optimization results to the intelligent decision-making module through a configuration center. The optimization process is logged, supporting manual auditing and rollback.

[0037] Specifically, the data fusion module outputs to the situation modeling module; the situation modeling module outputs to the intelligent decision-making module and the performance evaluation module; the intelligent decision-making module outputs task orders to the process awareness module as data collection trigger conditions; the process awareness module outputs full-chain execution data to the performance evaluation module; the performance evaluation module outputs evaluation results to the closed-loop optimization module; the closed-loop optimization module updates the parameters of the intelligent decision-making module; the system adopts unified identity authentication and permission management based on the OAuth2.0 protocol to ensure data security and access control; through the collaborative work of the six major functional modules, a complete closed-loop system is constructed from data collection, fusion, situation awareness, intelligent decision-making, process monitoring, performance evaluation to parameter optimization. Each module has clear responsibilities, standardized interfaces, and smooth data flow, realizing the full-process automation and intelligent management of open-pit mine shovel scheduling.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention. The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the protection scope of the present invention.

Claims

1. A real-time scheduling method for open-pit mine shovels based on multi-source data fusion, characterized in that, Includes the following steps: S1: The multi-source heterogeneous data of the acquired open-pit mine production system are fused to obtain a fused data base. S2: Perform event extraction and state association processing on the fused data base to construct a unified production status data model; S3: Based on the production status data model, use a multi-dimensional dynamic decision-making model to generate the optimal production task order for trucks that have entered the idle state; S4: Automatically acquire and record the actual operation data of each stage of loading, transportation and unloading corresponding to the production task order during the execution process; S5: Use graph neural network methods to process the fused data base and the full-chain execution data to establish a production efficiency evaluation model; S6: Based on the production efficiency evaluation model, evaluate the optimal production task order and its execution process, and feed the evaluation results back to the multidimensional dynamic decision-making model to optimize the decision parameters; The multi-source heterogeneous data includes at least the spatiotemporal positioning data of the vehicle and the excavator, the visual data of the excavator's working area, the status sensor data of the truck bed at the unloading point, and the production planning business data. The multidimensional dynamic decision-making model is based on a comprehensive decision based at least on excavator demand, dynamic transportation cost, and capacity matching. The excavator demand is determined based on the bucket status of the candidate excavator, the queuing status of waiting trucks, and the priority of the loaded materials. The bucket status indicates whether the excavator bucket is in a state of being raised or suspended, waiting for a truck to enter the loading position. The queuing status of waiting trucks includes at least the number of trucks waiting to be loaded in the candidate excavator's operating area. The dynamic transportation cost is calculated based on the real-time road network status. The capacity matching is determined based on the historical loading data of the candidate excavator and the predicted loading completion time using real-time visual data.

2. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, In step S1, before fusing the acquired multi-source heterogeneous data, the acquisition of the multi-source heterogeneous data is achieved through: Beidou / GPS positioning terminals installed on trucks and excavators acquire real-time location, speed, and timestamp data; High-definition cameras installed in the excavator's operating area acquire truck identification information, bucket movement trajectory, and cargo box loading status; Ultrasonic sensors installed at the unloading point detect the lifting and resetting status of the truck bed; The data interface, which connects with the mine production planning system, obtains the excavator's theoretical loading capacity, material type, and priority.

3. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, The acquired multi-source heterogeneous data is fused to obtain a fused data base. The fusion process is as follows: Spatiotemporal alignment of the spatiotemporal positioning data, visual data, sensor data, and business data is performed based on a unified timestamp. A complex event processing engine is used to correlate and fuse the aligned multi-source data to obtain the fused data base.

4. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, The step of generating optimal production task orders for trucks entering an idle state based on the production status data model and using a multi-dimensional dynamic decision model includes: The excavator demand is determined based on the excavator status data and material priority in the production status data model. The dynamic travel time of trucks to each candidate excavator is calculated based on real-time road network conditions and vehicle locations to determine dynamic transportation costs. The loading completion time is predicted based on historical loading data and real-time visual data of the excavator in order to determine the capacity matching degree. The excavator demand, dynamic transportation cost, and capacity matching degree are weighted and comprehensively scored, and the candidate excavator with the highest score is selected to generate the optimal production task order.

5. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, The automatic acquisition and recording of actual operational data for each stage of loading, transportation, and unloading during the execution of the production task order includes: During the loading process, the spatial relationship between the bucket and the truck is identified by a target detection model, and the unloading action is identified by a time-series classification model. The loading start time and loading end time are recorded. During the transportation process, the actual transport distance is automatically calculated based on the vehicle's positioning trajectory, and the heavy-load travel time and average speed are recorded. During the unloading process, ultrasonic sensors are used to detect the lifting and resetting status of the carriage, and the unloading start time and unloading end time are recorded.

6. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, The construction of the production efficiency evaluation model includes: The equipment status and topology in the production status data model and the time and efficiency indicators in the full-chain execution data are used as inputs; Machine learning methods are used to establish an evaluation model for assessing the performance of a single transportation task and the overall system.

7. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, The optimization of decision parameters based on performance evaluation results includes: The entire chain execution data is bound to the corresponding production task order to generate a transportation record; An error function is constructed based on historical launch records, and the weight parameters of the multidimensional dynamic decision-making model are adjusted using an optimization algorithm.

8. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, After generating and executing the optimal production task order, the method further includes: Real-time monitoring of the operating status data of each truck and excavator; Perform fault diagnosis and risk assessment on abnormal state data; Generate fault reports and maintenance recommendations, and send them to the maintenance terminal.

9. The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to claim 1, characterized in that, After generating and executing the optimal production task order, the method further includes: Perform threshold analysis on the time consumption of each step in the execution result to identify abnormal steps; Locate abnormal devices based on device topology and geographic location information; The location and diagnostic information of the malfunctioning equipment are pushed to the dispatch terminal and maintenance terminal.

10. A real-time scheduling system for open-pit mine shovels based on multi-source data fusion, characterized in that: The real-time scheduling method for open-pit mine shovels based on multi-source data fusion according to any one of claims 1-9 includes: The data fusion module is used to fuse the multi-source heterogeneous data acquired from the open-pit mine production system to obtain a fused data base; the multi-source heterogeneous data includes at least the spatiotemporal positioning data of vehicles and excavators, visual data of the excavator's operating area, the status sensor data of the unloading point's cargo box, and production planning business data. The situation modeling module is used to extract events and associate states with the fused data base to build a unified production situation data model. The intelligent decision-making module is used to generate optimal production task orders for trucks entering an idle state based on the production status data model and using a multi-dimensional dynamic decision-making model; wherein, the multi-dimensional dynamic decision-making model makes comprehensive decisions based at least on excavator demand, dynamic transportation costs, and capacity matching. The process perception module is used to automatically collect the entire chain execution data of loading, transportation and unloading during the execution of the production task order through visual recognition and sensing detection technology; The efficiency evaluation module is used to construct a production efficiency evaluation model based on the production status data model and the full-chain execution data, and to evaluate the efficiency of the optimal production task order and its execution process. The closed-loop optimization module is used to optimize the decision parameters of the multidimensional dynamic decision-making model based on the performance evaluation results.

Images