A method and system for early warning of abnormal operating states of a strip mine shovel-truck system
By constructing a multimodal probability distribution model and relative entropy quantization, combined with a sliding time window and entropy weight method, the problem of early warning in open-pit mine shovel and transportation systems is solved. This achieves keen detection and millisecond-level response during the fault latency period, avoids false-order missed reports, and provides reliable predictive maintenance support.
Patent Information
- Application Number
- CN202610505854.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-10
AI Technical Summary
Existing technologies are insufficient for real-time and accurate monitoring and early warning of operational status in open-pit mine shovel and transport systems. In particular, they lack early warning capabilities in dynamically changing field environments, and traditional methods are prone to false-ordering omissions and delayed responses.
By constructing a multimodal probability distribution model, the deviation of the system state is quantified using relative entropy, and weighted fusion is performed by combining a sliding time window and entropy weighting method to monitor the rate of change of the weighted total entropy of the system in real time and trigger an early warning.
It achieves keen detection during the fault latency period, avoids false-order missed reports, has millisecond-level response capability, strong noise resistance and robustness, and provides reliable technical support for predictive maintenance of open-pit mine shovel and transport systems.
Smart Images
Figure CN122367658A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mine information technology and intelligent monitoring, and in particular to a method and system for early warning of abnormal operating conditions in open-pit mine shovel and transport systems. Background Technology
[0002] The open-pit mine shovel and transport system is a core component of mine production, and its operational status directly affects the mine's production efficiency and safety. Because this system involves multiple processes such as electric shovel excavation, truck transportation, and ore unloading, these processes are closely coupled and operate in a complex and variable environment. An anomaly in any one of these processes can easily trigger a cascading failure, leading to a decrease in the efficiency of the entire system or even a shutdown. Therefore, real-time and accurate monitoring and early warning of the shovel and transport system's operational status are of significant engineering importance.
[0003] Currently, industrial sites commonly use fixed physical threshold methods (such as current thresholds and speed thresholds) based on engineering experience to determine equipment anomalies. However, this method is difficult to adapt to the dynamically changing environment of open-pit mines, and alarms are usually only triggered after the fault has completely occurred, lacking early warning capabilities. Some studies have attempted to apply Shannon entropy to industrial diagnostics, but when equipment experiences deterministic faults such as "crashes" or sensors remaining at zero, the system state distribution is extremely homogeneous, and the Shannon entropy tends to approach zero, exhibiting a "pseudo-ordered" characteristic, leading to missed fault detection. Traditional statistical indicators such as Overall Equipment Effectiveness (OEE) are affected by the smoothing effect of cumulative averaging calculations, resulting in a lag in response to minor faults and difficulty in capturing transient impacts from sudden disturbances. Summary of the Invention
[0004] The purpose of this invention is to provide an early warning method for abnormal operating conditions of open-pit mine shovel and transport systems. This method can capture probability drift during the fault latency period to achieve early warning, is noise-resistant and robust, interpretable, avoids false-order missed reports, and guides early intervention.
[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system, comprising the following steps: Multi-source operational data of the open-pit mine shovel and transport system are acquired according to a preset sampling frequency, and feature data representing the operation of the logistics and transportation subsystem, equipment condition subsystem, and production operation subsystem are extracted from them. For the feature data of each subsystem, a multimodal probability distribution model is constructed to fit its distribution characteristics in the historical healthy period, determine multiple potential operating states and corresponding distribution parameters, and based on the distribution parameters and the feature data of each subsystem at the current sampling time, determine the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time. Based on multi-source operational data during historical health periods, the average posterior probability vector of each subsystem under each potential operational state is determined as the health baseline distribution. Relative entropy is used to measure the information distance between the posterior probability vector of each subsystem and the corresponding health baseline distribution at the current sampling time, and the dynamic state deviation of each subsystem is obtained. A sliding time window is established, and the state deviation sequence of each subsystem within the time window is extracted. An objective weight is assigned to each subsystem based on the degree of variation of each sequence using the entropy weight method. Within the time window, the state deviation of each subsystem at each sampling time is weighted and fused based on the objective weight to obtain the total weighted entropy of the system at each sampling time. The rate of change of the weighted total entropy of the system is determined in real time. When the rate of change exceeds the extreme value of fluctuations during historical healthy periods and continues to exceed the tolerance time, an early warning is triggered.
[0006] In some optional embodiments, the method further includes determining the contribution of each subsystem to the change in the weighted total entropy of the system after triggering the early warning. The specific steps are as follows: The formula for determining the contribution of each subsystem to the weighted total entropy change of the system is as follows: ; In the formula, For the first The contribution of each subsystem to its corresponding link. These correspond to the logistics and transportation subsystem, the equipment operating condition subsystem, or the production operation subsystem, respectively. Assigning entropy weights to the first The objective weights of each subsystem; The first time within the current sliding time window The change in the state deviation of each subsystem; The change in the weighted total entropy of the system within the same time window; The subsystem with the largest contribution is identified as the core causative node that triggers the cascading failure, and an intervention command is output.
[0007] In some optional embodiments, the steps of constructing a multimodal probability distribution model for the feature data of each subsystem to fit its distribution characteristics during historical healthy periods, determining multiple potential operating states and corresponding distribution parameters, and determining the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time based on the distribution parameters and the feature data of each subsystem at the current sampling time are as follows: The number of potential operating states is pre-defined based on historical operating condition types, and the mean, covariance, and mixing coefficient of the corresponding number of Gaussian components are initialized. The model is trained using feature data samples from each subsystem during historical health periods. The expectation-maximization algorithm is used for iterative computation. Each iteration includes determining the posterior probability of each historical sample belonging to each Gaussian component based on the current Gaussian component parameters, and updating the mean, covariance, and mixing coefficient of each component based on the posterior probability. The iteration is repeated until the parameters converge to obtain the final distribution parameters of each Gaussian component. Each Gaussian component is semantically mapped to the physical conditions of the open-pit mine shovel and transport system, so that each Gaussian component corresponds to a potential operating state with a clear physical meaning. For the feature data of each subsystem at the current sampling time, the posterior probability of each subsystem belonging to each Gaussian component is determined by Bayes' formula based on the final distribution parameters, and the posterior probability vector is output.
[0008] In some optional embodiments, the relative entropy is used to measure the information distance between the posterior probability vector of each subsystem at the current sampling time and the corresponding health baseline distribution, to obtain the dynamic state deviation of each subsystem. The specific formula is as follows: ; In the formula, For a certain subsystem at the current sampling time t The degree of deviation of the dynamic state; For this subsystem at the current sampling time t The real-time posterior probability vector, ; This is the health baseline distribution vector of the subsystem. ; K The total number of potential operating states preset for this subsystem; The current sampling time t Subsystem belongs to the first k Real-time posterior probabilities of a potential operating state; For the subsystem to be in the first position under the health baseline distribution k The average posterior probability of a potential operating state.
[0009] In some alternative embodiments, the length of the sliding time window is matched to the complete cycle of a single "loading-transporting-unloading" operation of the mine car.
[0010] In some optional embodiments, the method of assigning objective weights to each subsystem based on the degree of variation of each sequence using the entropy weight method involves the following specific steps: Extraction length is m An evaluation matrix is constructed from the state deviation sequences of each subsystem within a sliding time window, and the matrix data is then subjected to range standardization to obtain standardized values. ; Calculate the first j Information entropy of subsystem indicators The formula is as follows: ; In the formula, , and when ,Regulation .
[0011] Based on the information entropy calculation, it is assigned to the first... j Objective weights of each subsystem The formula is as follows: ; In the formula, the denominator contains k This is an index item for the subsystem, corresponding to the logistics and transportation subsystem, equipment operating condition subsystem, and production operation subsystem. 3 Subsystems.
[0012] In some optional embodiments, the rate of change of the system's weighted total entropy is a first-order rate of change.
[0013] In some optional embodiments, the tolerance time is a preset integer number of consecutive time windows.
[0014] Embodiments of the present invention also provide an early warning system for abnormal operating conditions of an open-pit mine shovel and transport system, characterized in that the system comprises: The data perception and classification module is used to acquire multi-source operational data of the open-pit mine shovel and transportation system at a preset sampling frequency, and extract feature data that respectively characterizes the operation of the logistics transportation subsystem, equipment condition subsystem and production operation subsystem. The probability state space reconstruction module constructs a multimodal probability distribution model for the feature data of each subsystem to fit its distribution characteristics during the historical healthy period, determines multiple potential operating states and corresponding distribution parameters, and determines the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time based on the distribution parameters and the feature data of each subsystem at the current sampling time. The dynamic relative entropy measurement module is used to determine the average posterior probability vector of each subsystem under each potential operating state based on multi-source operation data during historical health periods, and to use the relative entropy to measure the information distance between the posterior probability vector of each subsystem at the current sampling time and the corresponding health benchmark distribution, thereby obtaining the dynamic state deviation of each subsystem. The objective entropy weight and total deviation fusion module is used to establish a sliding time window, extract the state deviation sequence of each subsystem within the time window, and assign objective weights to each subsystem according to the degree of variation of each sequence using the entropy weight method. Within this time window, the state deviation of each subsystem at each sampling time is weighted and fused based on the objective weights to obtain the system weighted total entropy at each sampling time. The cascaded fault diagnosis and early warning module is used to determine the rate of change of the weighted total entropy of the system in real time. When the rate of change exceeds the extreme value of fluctuation during the historical healthy period and continues to exceed the tolerance time, an early warning is triggered.
[0015] The method for early warning of abnormal operating conditions in open-pit mine shovel and transport systems provided by this invention has at least the following beneficial effects: The method for early warning of abnormal operation of open-pit mine shovel and transportation system provided by the present invention decouples multi-source operation data into feature data representing three subsystems: logistics and transportation, equipment condition and production operation. Multimodal probability distribution models are established to characterize the normal state distribution of each subsystem. The relative entropy is used to quantify the deviation of the current state from the health benchmark in real time. Then, the entropy weight method within the sliding time window is used to adaptively fuse the system to obtain the weighted total entropy. Finally, the dynamic fluctuation extreme value of the total entropy change rate is used to determine the triggering of early warning. This method does not rely on fixed physical thresholds and can keenly capture early drifts in the probability distribution of system states during the fault latency period, achieving millisecond-level early warning. Through probabilistic modeling and relative entropy measurement, it effectively avoids the "pseudo-order" underreporting problem of traditional Shannon entropy under deterministic faults such as equipment crashes. At the same time, the sliding time window and the judgment mechanism for continuous exceedance of tolerance time can filter out instantaneous disturbances under normal operating conditions, and has strong noise resistance and robustness. In addition, the entire method, from data decoupling to entropy flow analysis to early warning output, has clear logic and strong interpretability, providing reliable technical support for predictive maintenance and lean management of open-pit mine shovel and transport systems. Attached Figure Description
[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:
[0017] Figure 1 This is a flowchart of an early warning method for abnormal operating conditions of an open-pit mine shovel and transport system according to an embodiment of the present invention; Figure 2 This is a structural diagram of an early warning system for abnormal operation of an open-pit mine shovel and transport system according to an embodiment of the present invention; Figure 3 This is a time series diagram of multi-source monitoring data of multiple subsystems of open-pit mine shovel and transport system, which is provided according to an embodiment of the present invention. Figure (a) is the GPS speed time series curve of the logistics and transportation subsystem, Figure (b) is the load current time series curve of the electric shovel of the equipment operating condition subsystem, and Figure (c) is the scatter plot of the unloading flow interval of the production operation subsystem. Figure 4 This is a time-series diagram of the anomaly detection comparison results of an open-pit mine shovel and transport system based on multimodal entropy flow, according to an embodiment of the present invention. Figure 5 This is a pie chart showing the percentage of entropy increase contribution of each subsystem under abnormal operating conditions of an open-pit mine shovel and transport system according to an embodiment of the present invention. Figure 6 This is a timing diagram comparing the KL divergence advance warning and fixed threshold alarm of an open-pit mine shovel and transport system according to an embodiment of the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0019] One embodiment of the present invention relates to an early warning method for abnormal operating conditions of an open-pit mine shovel and transportation system. The implementation details of the early warning method for abnormal operating conditions of an open-pit mine shovel and transportation system in this embodiment are described in detail below. The following implementation details are provided for ease of understanding and are not necessary for implementing this solution.
[0020] The specific process of the early warning method for abnormal operation of the open-pit mine shovel and transport system in this embodiment can be described as follows: Figure 1 As shown, it includes: Step 101: Acquire multi-source operation data of the open-pit mine shovel and transport system according to the preset sampling frequency, and extract feature data representing the operation of the logistics and transportation subsystem, equipment condition subsystem and production operation subsystem respectively. Among them, the characteristic data of the logistics and transportation subsystem corresponds to GPS trajectory speed and acceleration, representing logistics characteristics; the characteristic data of the equipment operating condition subsystem corresponds to CAN bus data such as electric shovel motor load current, representing equipment operating conditions; and the characteristic data of the production operation subsystem corresponds to unloading interval time Δt, representing production continuity.
[0021] Step 102: For the feature data of each subsystem, construct a multimodal probability distribution model to fit its distribution characteristics in the historical healthy period, determine multiple potential operating states and corresponding distribution parameters, and based on the distribution parameters and the feature data of each subsystem at the current sampling time, determine the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time. The specific steps are as follows: The number of potential operating states is pre-defined based on historical operating condition types, and the mean, covariance, and mixing coefficients of the corresponding number of Gaussian components are initialized. In this embodiment, taking the logistics transportation subsystem as an example, the following settings are made: K=4Each Gaussian component is mapped to: efficient running state, job waiting state, unnecessary creeping state, and abnormally long stopping state.
[0022] The model is trained using feature data samples from each subsystem during historical health periods. The expectation-maximization algorithm is used for iterative computation. Each iteration includes determining the posterior probability of each historical sample belonging to each Gaussian component based on the current Gaussian component parameters, and updating the mean, covariance, and mixing coefficient of each component based on the posterior probability. The iteration is repeated until the parameters converge to obtain the final distribution parameters of each Gaussian component. Each Gaussian component is semantically mapped to the physical conditions of the open-pit mine shovel and transport system, so that each Gaussian component corresponds to a potential operating state with a clear physical meaning. For the feature data of each subsystem at the current sampling time, the posterior probability of each subsystem belonging to each Gaussian component is determined by Bayes' formula based on the final distribution parameters, and the posterior probability vector is output.
[0023] Step 103: Based on the multi-source operation data of historical health periods, determine the average posterior probability vector of each subsystem under each potential operation state as the health benchmark distribution; use relative entropy to measure the information distance between the posterior probability vector of each subsystem at the current sampling time and the corresponding health benchmark distribution to obtain the dynamic state deviation of each subsystem. Information distance is obtained by measuring the current posterior probability vector and the baseline distribution using the KL divergence formula. The measurement result is the dynamic state deviation of each subsystem.
[0024] Specifically, the formula for calculating the KL divergence is: ; In the formula, For a certain subsystem at the current sampling time t The degree of deviation of the dynamic state; For this subsystem at the current sampling time t The real-time posterior probability vector, ; This is the health baseline distribution vector of the subsystem. ; K The total number of potential operating states preset for this subsystem; The current sampling time t Subsystem belongs to the first k Real-time posterior probabilities of a potential operating state; For the subsystem to be in the first position under the health baseline distribution k The average posterior probability of a potential operating state.
[0025] Step 104: Establish a sliding time window, extract the state deviation sequence of each subsystem within the time window, and assign objective weights to each subsystem according to the degree of variation of each sequence using the entropy weight method; within the time window, perform weighted fusion of the state deviation of each subsystem at each sampling time based on the objective weights to obtain the system weighted total entropy at each sampling time. The length of the sliding time window is not arbitrarily set, but is matched with the complete cycle of a single mining truck's "loading-transporting-unloading" (e.g., 30 minutes), ensuring that the entropy value sequence extracted each time can completely contain the cascading interaction process of the system, avoiding feature fragmentation caused by an excessively small window or smoothing out transient abnormal features by an excessively large window.
[0026] The entropy weight method is used to assign objective weights to each subsystem based on the degree of variation of each sequence. The specific steps are as follows: Extraction length is m An evaluation matrix is constructed from the state deviation sequences of each subsystem within a sliding time window, and the matrix data is then subjected to range standardization to obtain standardized values. ; Calculate the first j Information entropy of subsystem indicators The formula is as follows: ; In the formula, , and when ,Regulation .
[0027] Based on the information entropy calculation, it is assigned to the first... j Objective weights of each subsystem The formula is as follows: ; In the formula, the denominator contains k This is an index item for the subsystem, corresponding to the logistics and transportation subsystem, equipment operating condition subsystem, and production operation subsystem. 3 Subsystems.
[0028] Step 105: Determine the rate of change of the weighted total entropy of the system in real time. When the rate of change exceeds the extreme value of fluctuations during historical healthy periods and continues to exceed the tolerance time, trigger an early warning.
[0029] The rate of change of the system's weighted total entropy is the first-order rate of change; the tolerance time is a preset integer number of continuous time windows.
[0030] This also includes determining the contribution of each subsystem to the change in the system's weighted total entropy after triggering the early warning system. The specific steps are as follows: The specific steps to determine the contribution of each subsystem to the change in the weighted total entropy of the system are as follows: The formula for determining the contribution of each subsystem to the weighted total entropy change of the system is as follows: ; In the formula, For the first The contribution of each subsystem to its corresponding link. These correspond to the logistics and transportation subsystem, the equipment operating condition subsystem, or the production operation subsystem, respectively. Assigning entropy weights to the first The objective weights of each subsystem; The first time within the current sliding time window The change in the state deviation of each subsystem; The change in the weighted total entropy of the system within the same time window; The subsystem with the largest contribution is identified as the core causative node that triggers the cascading failure, and an intervention command is output.
[0031] Another embodiment of the present invention relates to an early warning system for abnormal operation of an open-pit mine shovel and transportation system. The implementation details of the early warning system for abnormal operation of the open-pit mine shovel and transportation system in this embodiment are described in detail below. The following implementation details are provided for ease of understanding and are not necessary for implementing this solution.
[0032] Specifically, the data perception and classification module is used to acquire multi-source operational data of the open-pit mine shovel and transportation system at a preset sampling frequency, and extract feature data from it that respectively characterize the operation of the logistics transportation subsystem, equipment condition subsystem and production operation subsystem. The probability state space reconstruction module constructs a multimodal probability distribution model for the feature data of each subsystem to fit its distribution characteristics during the historical healthy period, determines multiple potential operating states and corresponding distribution parameters, and determines the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time based on the distribution parameters and the feature data of each subsystem at the current sampling time. The dynamic relative entropy measurement module is used to determine the average posterior probability vector of each subsystem under each potential operating state based on multi-source operation data during historical health periods, and to use the relative entropy to measure the information distance between the posterior probability vector of each subsystem at the current sampling time and the corresponding health benchmark distribution, thereby obtaining the dynamic state deviation of each subsystem. The objective entropy weight and total deviation fusion module is used to establish a sliding time window, extract the state deviation sequence of each subsystem within the time window, and assign objective weights to each subsystem according to the degree of variation of each sequence using the entropy weight method. Within this time window, the state deviation of each subsystem at each sampling time is weighted and fused based on the objective weights to obtain the system weighted total entropy at each sampling time. The cascaded fault diagnosis and early warning module is used to determine the rate of change of the weighted total entropy of the system in real time. When the rate of change exceeds the extreme value of fluctuation during the historical healthy period and continues to exceed the tolerance time, an early warning is triggered.
[0033] The structure diagram of the early warning system for abnormal operation of the open-pit mine shovel and transport system is as follows: Figure 2 As shown, the present invention provides an early warning system for abnormal operating conditions of an open-pit mine shovel and transport system. Logically, it mainly includes a data perception layer, a state space reconstruction layer, and an entropy flow calculation and diagnosis layer. The specific connection relationships and functions of each layer and its internal modules are described below:
[0034] The data perception layer is used to collect multimodal operational data of the open-pit mine's shovel and transportation system in real time. Specifically, it includes: a trajectory data acquisition unit for acquiring GPS trajectory data characterizing the spatiotemporal characteristics of logistics, including but not limited to truck speed and acceleration; a working condition data acquisition unit for acquiring CAN bus data characterizing the operating load of key equipment, such as electric shovel motor current data; and a command data acquisition unit for acquiring scheduling command data characterizing production cycle time, such as unloading interval time series.
[0035] The state space reconstruction layer, unidirectionally connected to the data perception layer, receives multi-dimensional temporal features and mines their probability distribution characteristics using a Gaussian mixture model (GMM). This layer specifically has three parallel processing channels: a transportation subsystem processing channel, comprising a kinematic feature joint probability modeling unit and a transportation state mapping unit connected in sequence, used to fit trajectory data and classify it into different physical mapping states such as efficient operation, work waiting, unnecessary creeping, or abnormal long stoppage; an equipment subsystem processing channel, comprising a load multimodal distribution fitting unit and an equipment state mapping unit connected in sequence, used to fit operating condition data and classify it into mapping states such as smooth operation, idling / waiting for materials, or overload / impact; and a production subsystem processing channel, comprising a logistics continuity characterization unit and a production state mapping unit connected in sequence, used to fit scheduling instruction data and classify it into mapping states such as balanced stable production, flow interruption / waiting for materials, or accumulation / overflow.
[0036] The entropy flow calculation and diagnosis layer is unidirectionally connected to the state space reconstruction layer and is used to quantify the degree of system state deviation and perform fault tracing. This layer further includes: a real-time posterior probability statistics unit, used to summarize the posterior probability vectors of the current observation samples belonging to a specific operating state, received from each state mapping unit; a relative entropy flow calculation unit, which uses the relative entropy (KL divergence) algorithm to quantify the information distance between the real-time probability distribution and the preset health benchmark distribution, and outputs the state deviation of each subsystem (i.e., transportation entropy, equipment entropy, and production entropy); an objective weighting unit, which introduces a sliding time window and uses the entropy weight method to calculate the degree of variation of the state deviation sequence of each subsystem, thereby adaptively allocating objective weights; and a system total entropy evaluation unit, which performs weighted fusion calculation on the state deviation of each subsystem according to the objective weights to obtain the system weighted total entropy; the output of the system total entropy evaluation unit is connected to the system state deviation output unit and the cascaded fault cause localization unit, respectively. The former is used to generate and output the overall deviation curve of the system to achieve early warning of the fault latency period; the latter is used to calculate the contribution ratio of each subsystem to the total entropy increase of the system, so as to accurately determine and output the core causal node that causes the cascading failure of the system.
[0037] A specific example: A typical open-pit mine stripping operation scenario was constructed using Python. To closely resemble real-world conditions, random disturbances conforming to a normal distribution were introduced into the simulation model, and parameter settings referenced the historical statistical characteristics of a certain open-pit mine's SCADA system. Parameter settings included: 3 electric shovel operating points, 1 spoil heap, and 12 mining trucks. The time window was set to 30 minutes, and the sampling frequency to 1Hz. Two experimental groups were set up: Condition A (normal state, equipment availability 98%, unobstructed roads); Condition B (sudden failure state, simulating a sudden failure of electric shovel No. 2 at t=60min, with current dropping to 0, lasting 30 minutes). After applying the above steps using this method, in a cascading failure scenario of "transportation congestion leading to shutdown," an advance warning of approximately 12.9 minutes was achieved compared to the traditional fixed physical threshold method.
[0038] The time series diagram of multi-source monitoring data of multiple subsystems of open-pit mine shovel and haulage system (transportation, equipment, production) is shown below. Figure 3As shown in the figure, Figure (a) is the GPS speed time-series curve of the logistics transportation subsystem, which intuitively presents the speed changes of the mining trucks during the monitoring period; Figure (b) is the load current time-series curve of the electric shovel of the equipment operating condition subsystem, which reflects the load condition of the electric shovel during operation in real time; Figure (c) is the scatter plot of the unloading flow interval of the production operation subsystem, which clearly shows the continuity of the unloading operation cycle. The three sub-graphs together cover the complete monitoring period of 120 minutes, clearly showing the abnormal states that occur simultaneously in the three subsystems within the 60-90 minute interval: the mining truck transportation speed drops sharply, the electric shovel load current returns to zero and stops, and the unloading flow interval deviates significantly from the ideal cycle range. This kind of synchronous abnormal data of multiple subsystems provides core raw data support for the dynamic status assessment, early warning and fault tracing of the shovel transportation system driven by multimodal entropy flow.
[0039] The time series diagram of the comparison results of anomaly detection in open-pit mine shovel and transport systems based on multimodal entropy flow is shown below. Figure 4 As shown in the figure, the solid black line represents the normalized anomaly score curve of our method (DKM + KL divergence), the dashed black line is the anomaly score curve of the comparison method (isolated forest), and the gray dashed line delineates the threshold range for anomaly detection. The curves clearly demonstrate the anomaly evolution characteristics of the shovel and transport system within a 120-minute monitoring period. In the 60-90 minute range, the anomaly score of our method significantly increases and exceeds the threshold, accurately capturing the outbreak stage of synchronous anomalies in multiple subsystems. Furthermore, the fluctuation characteristics during the subsequent anomaly decline are more consistent with actual working conditions. In contrast, the comparison method exhibits problems such as inflated anomaly scores and delayed response. This fully verifies the accuracy and superiority of our method in dynamic status assessment, early warning, and fault tracing of open-pit mine shovel and transport systems.
[0040] The pie chart showing the percentage contribution of entropy increase of each subsystem to abnormal operating conditions of the open-pit mine shovel and transport system is as follows: Figure 5 As shown in the figure, the entropy increase contribution ratio of the three subsystems—transportation entropy (HT), equipment entropy (HE), and production entropy (HP)—during the abnormal period is intuitively presented. Among them, production entropy (HP) accounts for as high as 67.9%, which is the core cause of this system anomaly. Equipment entropy (HE) and transportation entropy (HT) account for 17.3% and 14.8%, respectively. By quantifying the entropy increase contribution of each subsystem, the source of the anomaly was accurately located as the production system, providing an intuitive quantitative basis for tracing the faults of the multimodal entropy flow-driven shovel and transport system.
[0041] The time sequence diagram comparing the KL divergence early warning and fixed threshold alarm of the open-pit mine shovel and haulage system is as follows: Figure 6As shown in the figure, the left vertical axis represents the numerical changes of physical monitoring indicators such as speed and current, while the right vertical axis represents the alarm status. The black solid line marks the early warning curve of KL divergence (this method), and the black dashed line represents the fixed threshold alarm curve. It is also noted that this method completes the early warning 12.9 minutes earlier than the fixed threshold alarm. The chart clearly shows that within the 120-minute monitoring period, this method has already triggered the early warning signal before the system anomaly occurs, accurately capturing the early signs of the anomaly. In contrast, the fixed threshold alarm is only triggered after the anomaly appears. This fully demonstrates that this method, based on KL divergence analysis of multimodal entropy flow, has significant timeliness and superiority in early warning of faults in open-pit mine shovel and transportation systems, providing key technical support for intelligent monitoring of shovel and transportation systems.
[0042] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of this invention. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, without changing the core design of the algorithm and process, are also within the protection scope of this invention.
[0043] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.
[0044] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0045] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing the present invention, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present invention.
Claims
1. A method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system, characterized in that, The method includes: Multi-source operational data of the open-pit mine shovel and transport system are acquired according to a preset sampling frequency, and feature data representing the operation of the logistics and transportation subsystem, equipment condition subsystem, and production operation subsystem are extracted from them. For the feature data of each subsystem, a multimodal probability distribution model is constructed to fit its distribution characteristics in the historical healthy period, determine multiple potential operating states and corresponding distribution parameters, and based on the distribution parameters and the feature data of each subsystem at the current sampling time, determine the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time. Based on multi-source operational data during historical health periods, the average posterior probability vector of each subsystem under each potential operational state is determined as the health baseline distribution. Relative entropy is used to measure the information distance between the posterior probability vector of each subsystem and the corresponding health baseline distribution at the current sampling time, and the dynamic state deviation of each subsystem is obtained. A sliding time window is established, and the state deviation sequence of each subsystem within the time window is extracted. An objective weight is assigned to each subsystem based on the degree of variation of each sequence using the entropy weight method. Within the time window, the state deviation of each subsystem at each sampling time is weighted and fused based on the objective weight to obtain the total weighted entropy of the system at each sampling time. The rate of change of the weighted total entropy of the system is determined in real time. When the rate of change exceeds the extreme value of fluctuations during historical healthy periods and continues to exceed the tolerance time, an early warning is triggered.
2. The method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system as described in claim 1, characterized in that, This also includes determining the contribution of each subsystem to the change in the system's weighted total entropy after triggering the early warning system. The specific steps are as follows: The formula for determining the contribution of each subsystem to the weighted total entropy change of the system is as follows: ； In the formula, For the first The contribution of each subsystem to its corresponding link. These correspond to the logistics and transportation subsystem, the equipment operating condition subsystem, or the production operation subsystem, respectively. Assigning entropy weights to the first The objective weights of each subsystem; The first time within the current sliding time window The change in the state deviation of each subsystem; The change in the weighted total entropy of the system within the same time window; The subsystem with the largest contribution is identified as the core causative node that triggers the cascading failure, and an intervention command is output.
3. The method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system as described in claim 1, characterized in that, The process involves constructing a multimodal probability distribution model for the feature data of each subsystem to fit its distribution characteristics during historical healthy periods, determining multiple potential operating states and corresponding distribution parameters, and based on the distribution parameters and the feature data of each subsystem at the current sampling time, determining the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time. The specific steps are as follows: The number of potential operating states is pre-defined based on historical operating condition types, and the mean, covariance, and mixing coefficient of the corresponding number of Gaussian components are initialized. The model is trained using feature data samples from each subsystem during historical health periods. The expectation-maximization algorithm is used for iterative computation. Each iteration includes determining the posterior probability of each historical sample belonging to each Gaussian component based on the current Gaussian component parameters, and updating the mean, covariance, and mixing coefficient of each component based on the posterior probability. The iteration is repeated until the parameters converge to obtain the final distribution parameters of each Gaussian component. Each Gaussian component is semantically mapped to the physical conditions of the open-pit mine shovel and transport system, so that each Gaussian component corresponds to a potential operating state with a clear physical meaning. For the feature data of each subsystem at the current sampling time, the posterior probability of each subsystem belonging to each Gaussian component is determined by Bayes' formula based on the final distribution parameters, and the posterior probability vector is output.
4. The method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system as described in claim 1, characterized in that, The relative entropy is used to measure the information distance between the posterior probability vector of each subsystem at the current sampling time and the corresponding health baseline distribution, thereby obtaining the dynamic state deviation of each subsystem. The specific formula is as follows: ； In the formula, For a certain subsystem at the current sampling time t The degree of deviation of the dynamic state; For this subsystem at the current sampling time t The real-time posterior probability vector, ; This is the health baseline distribution vector of the subsystem. ; K The total number of potential operating states preset for this subsystem; The current sampling time t Subsystem belongs to the first k Real-time posterior probabilities of a potential operating state; For the subsystem to be in the first position under the health baseline distribution k The average posterior probability of a potential operating state.
5. The method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system as described in claim 1, characterized in that, The length of the sliding time window is matched with the complete cycle of a single "loading-transporting-unloading" operation of the mine car.
6. The method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system as described in claim 1, characterized in that, The method of assigning objective weights to each subsystem based on the degree of variation of each sequence using the entropy weight method involves the following specific steps: Extraction length is m An evaluation matrix is constructed from the state deviation sequences of each subsystem within a sliding time window, and the matrix data is then subjected to range standardization to obtain standardized values. ; Calculate the first j Information entropy of subsystem indicators The formula is as follows: ； In the formula, , and when ,Regulation ; Based on the information entropy calculation, it is assigned to the first... j Objective weights of each subsystem The formula is as follows: ； In the formula, the denominator contains k This is an index item for the subsystem, corresponding to the logistics and transportation subsystem, equipment operating condition subsystem, and production operation subsystem. 3 Subsystems.
7. The method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system as described in claim 1, characterized in that, The rate of change of the weighted total entropy of the system is a first-order rate of change.
8. The method for early warning of abnormal operating conditions in an open-pit mine shovel and transport system as described in claim 1, characterized in that, The tolerance time is a preset integer number of consecutive time windows.
9. A pre-warning system for abnormal operating conditions of an open-pit mine shovel and transport system, characterized in that, The system includes: The data perception and classification module is used to acquire multi-source operational data of the open-pit mine shovel and transportation system at a preset sampling frequency, and extract feature data that respectively characterizes the operation of the logistics transportation subsystem, equipment condition subsystem and production operation subsystem. The probability state space reconstruction module constructs a multimodal probability distribution model for the feature data of each subsystem to fit its distribution characteristics during the historical healthy period, determines multiple potential operating states and corresponding distribution parameters, and determines the posterior probability vector of each subsystem belonging to each potential operating state at the current sampling time based on the distribution parameters and the feature data of each subsystem at the current sampling time. The dynamic relative entropy measurement module is used to determine the average posterior probability vector of each subsystem under each potential operating state based on multi-source operation data during historical health periods, and to use the relative entropy to measure the information distance between the posterior probability vector of each subsystem at the current sampling time and the corresponding health benchmark distribution, thereby obtaining the dynamic state deviation of each subsystem. The objective entropy weight and total deviation fusion module is used to establish a sliding time window, extract the state deviation sequence of each subsystem within the time window, and assign objective weights to each subsystem according to the degree of variation of each sequence using the entropy weight method. Within this time window, the state deviation of each subsystem at each sampling time is weighted and fused based on the objective weights to obtain the system weighted total entropy at each sampling time. The cascaded fault diagnosis and early warning module is used to determine the rate of change of the weighted total entropy of the system in real time. When the rate of change exceeds the extreme value of fluctuation during the historical healthy period and continues to exceed the tolerance time, an early warning is triggered.