Mine equipment group intelligent control method and system based on cloud edge collaboration

Abstract

The application provides a mine equipment group intelligent control method and system based on cloud edge collaboration, relates to the technical field of cloud edge collaboration, and comprises the following steps: collecting mine equipment operation data through an edge node, performing time sequence decomposition, and constructing a mapping relationship from a state vector to a control output; uploading local state vector statistical characteristics and an effective range of the mapping relationship to the cloud; identifying the coupling relationship between the equipment in the cloud and constructing a resource allocation constraint set to generate a resource quota and a priority sequence; selecting an optimal control sequence based on the edge node and executing the optimal control sequence; and updating the mapping relationship in real time. The application realizes intelligent collaborative control of the mine equipment group, improves resource utilization efficiency, and improves system operation stability.

Description

Technical Field

[0001] This invention relates to cloud-edge collaboration technology, and more particularly to a method and system for intelligent control of mine equipment clusters based on cloud-edge collaboration. Background Technology

[0002] With the continuous improvement of intelligence and automation in the mining industry, intelligent control of mine equipment groups has become a key technology for improving mine production efficiency and safety. However, existing control methods often overlook the complex coupling relationships between mine equipment, and independent control of each piece of equipment leads to low overall resource utilization efficiency. The operation of equipment often affects each other. For example, there are obvious process dependencies between coal mining machines, conveyors, and support equipment, but existing control systems cannot effectively identify and utilize these relationships for global optimization. Summary of the Invention

[0003] This invention provides a cloud-edge collaborative intelligent control method and system for mine equipment clusters, which can solve the problems in the prior art.

[0004] A first aspect of this invention provides an intelligent control method for a group of mine equipment based on cloud-edge collaboration, comprising:

[0005] The operational status data of the mine equipment group is collected through edge nodes;

[0006] The operating state data is decomposed into time series at the edge layer. Periodic and transient components are separated by frequency domain transformation. The amplitude features of the periodic components and the rate of change features of the transient components are extracted as local state vectors, and a mapping relationship from the local state vectors to the control output is constructed. The statistical features of the local state vectors and the effective range of the mapping relationship are uploaded to the cloud. The cloud identifies the coupling relationship between devices based on the statistical features of multiple nodes and constructs a global constraint set for resource allocation based on the effective range.

[0007] The cloud platform solves the resource allocation scheme based on the global constraint set, generating resource quotas and priority sequences for each edge node. The edge nodes use the resource quotas and priority sequences as constraints, and use the mapping relationship to predict the state trajectory of the mine equipment. Among the trajectories that satisfy the constraints, they select the control sequence that minimizes the control cost to generate execution instructions.

[0008] The edge node controls the corresponding mining equipment based on the execution command and collects feedback data to update the mapping relationship; when the feedback data deviates from the effective range, it triggers cloud reporting and corrects the global constraint set.

[0009] The steps of performing time-series decomposition on the operating state data at the edge layer, extracting local state vectors, and constructing a mapping relationship between local state vectors and control outputs include:

[0010] The operating status data is divided into multiple time segments according to a time window. Frequency domain transformation is performed on each time segment to separate periodic components and transient components. The amplitude characteristics of the periodic components and the rate of change characteristics of the transient components are extracted as local state vectors.

[0011] Based on the historical sequence of the local state vectors, a causal dependency rule between the local state vectors and historical control outputs is established through state-space mapping, forming a mapping relationship;

[0012] The sensitivity weights of each dimension component of the local state vector to the control output are calculated according to the causal dependency rule, and the range of state vector values ​​whose sensitivity weights exceed the preset stability interval are identified as the effective range of the mapping relationship.

[0013] The steps of uploading the statistical features and effective range of the mapping relationship of the local state vector to the cloud; identifying the coupling relationship between devices based on the statistical features of multiple nodes in the cloud, and constructing a global constraint set for resource allocation in combination with the effective range include:

[0014] Statistical analysis is performed on the distribution of the local state vector within a preset time window, and the mean offset and variance fluctuation amplitude are calculated as statistical features. The effective range of the statistical features and the mapping relationship is then encapsulated and uploaded to the cloud.

[0015] After receiving the statistical features uploaded by multiple edge nodes, the cloud calculates the cross-correlation coefficient between the mean offsets of different edge nodes, identifies node pairs whose cross-correlation coefficient exceeds the independence threshold, and marks the mining equipment corresponding to the node pairs as having a coupling relationship;

[0016] For the effective range uploaded by each edge node, the boundary values ​​of each dimension are extracted and projected in the global coordinate system. The intersection of the feasible regions of the effective range of each edge node in the resource configuration space is calculated. Based on the intersection of the feasible regions and the coupling relationship, a global constraint set describing the resource competition conflict between devices is generated.

[0017] The steps for the cloud to solve the resource allocation scheme based on the global constraint set and generate resource quotas and priority sequences for each edge node include:

[0018] The cloud platform determines the feasible solution space for resource allocation based on the intersection of feasible regions in the global constraint set. Within this feasible solution space, an objective function is established with the optimization goal of minimizing the energy consumption of the mine equipment group and maximizing the global resource utilization rate. The weight coefficients of the weighted combination are dynamically adjusted according to the urgency of the current mine production task and the health status of the equipment. Iterative search is performed to find the solution that achieves the optimal value of the objective function while satisfying all constraints in the global constraint set, which is then used as the resource allocation scheme.

[0019] Based on the resource allocation scheme and the dependency hierarchy of each edge node control device in the inter-device coupling relationship, calculate the resource quota for each edge node; construct a dependency graph structure based on the inter-device coupling relationship and perform topological sorting to generate a priority sequence for each edge node;

[0020] For the resource quota of each edge node, the negotiable space is determined according to the sensitivity of the coupling component of the corresponding device in the device coupling relationship; the resource quota, the priority sequence and the negotiable space are encapsulated and sent to the corresponding edge node.

[0021] For the resource quotas of each edge node, the step of determining the negotiable space based on the sensitivity of the coupling component of the device corresponding to the edge node in the inter-device coupling relationship includes:

[0022] Extract all directed coupling relationships when the device corresponding to the edge node acts as the response device from the inter-device coupling relationships. Calculate the product of the coupling strength and time delay of each directed coupling relationship as the coupling influence factor. Normalize and sum all coupling influence factors to obtain the coupling component sensitivity of the edge node.

[0023] The expansion range of the negotiable space is set according to the sensitivity of the coupling component. If the sensitivity of the coupling component is lower than the weak coupling threshold, the negotiable space is set to a fixed proportion of the resource quota expansion; if the sensitivity of the coupling component is higher than the strong coupling threshold, the negotiable space is set to an adaptive expansion range dynamically calculated based on the coupling influence factor.

[0024] For edge nodes that adopt adaptive expansion amplitude, the frequency and magnitude of resource quota breaches of the edge node in the historical control cycle are extracted. The frequency and magnitude are used as correction coefficients to weight and adjust the coupling influence factor, thereby generating a negotiable spatial boundary that takes into account both coupling dependency and historical execution deviation.

[0025] The edge node uses the resource quota and priority sequence as constraints, and uses the mapping relationship to predict the state trajectory of the mine equipment. The step of selecting the control sequence that minimizes the control cost from the trajectory that satisfies the constraints to generate the execution instruction includes:

[0026] The edge node receives the resource quota, priority sequence, and negotiable space from the cloud, converts the resource quota into the upper limit of the control variable, and converts the priority sequence into a temporal execution order constraint;

[0027] Using the mapping relationship, forward extrapolation is performed starting from the current state of the mine equipment to generate multiple candidate state trajectories that satisfy the resource quota limit and priority sequence time constraints within a preset prediction time domain. The validity of each candidate state trajectory is verified, and candidate state trajectories that fall within the effective range of the mapping relationship are marked as feasible trajectories.

[0028] For each feasible trajectory, the cumulative value of control quantities required to drive the mine equipment along the trajectory is calculated as the control cost. The trajectory with the minimum control cost is selected from the feasible trajectories, and the corresponding control quantity time series is extracted as the control sequence.

[0029] If the set of feasible trajectories is empty, the upper limit of the resource quota is extended to the boundary of the negotiable space, candidate state trajectories are regenerated, and the trajectory with the lowest control cost is selected as the control sequence. An adjustment feedback carrying the quota extension is generated and reported to the cloud; the control sequence is converted into an execution instruction.

[0030] The edge node controls the corresponding mining equipment based on the execution command and collects feedback data to update the mapping relationship; the step of triggering cloud reporting and correcting the global constraint set when the feedback data deviates from the effective range includes:

[0031] The edge node converts the execution command into control signals for the mining equipment and sends them to the corresponding mining equipment, collecting the actual operating status of the mining equipment after executing the control signals as feedback data;

[0032] Calculate the deviation between the feedback data and the predicted state corresponding to the execution instruction, and use the deviation to correct the mapping relationship;

[0033] Determine whether the feedback data falls within the effective range of the corrected mapping relationship. If it does, continue collecting feedback data at the next moment. If it deviates, extract the dimensions and magnitude of the feedback data that exceed the effective range.

[0034] The deviation type is determined based on the deviation magnitude. If the deviation magnitude is lower than the abnormal threshold, the feedback data and the corresponding deviation dimension are reported to the cloud. The cloud locates the affected constraints in the global constraint set based on the deviation dimension and performs boundary expansion correction. If the deviation magnitude reaches the abnormal threshold, the cloud is triggered to re-identify the coupling relationship between the devices and reconstruct the global constraint set.

[0035] A second aspect of the present invention provides an intelligent control system for a mine equipment cluster based on cloud-edge collaboration, comprising:

[0036] The data acquisition module is used to collect operational status data of the mine equipment group through edge nodes;

[0037] The edge analysis and reporting module is used to perform time-series decomposition on the operating status data at the edge layer, separate periodic components and transient components through frequency domain transformation, extract the amplitude characteristics of the periodic components and the rate of change characteristics of the transient components as local state vectors, construct the mapping relationship between the local state vectors and the control output, and upload the statistical characteristics of the local state vectors and the effective range of the mapping relationship to the cloud.

[0038] The cloud-based collaborative optimization module is used to identify the coupling relationship between devices based on the statistical characteristics of multiple nodes, construct a global constraint set for resource allocation based on the effective range, solve the resource allocation scheme according to the global constraint set, and generate resource quotas and priority sequences for each edge node.

[0039] The edge control decision module is used to use the resource quota and priority sequence as constraints, predict the state trajectory of the mine equipment using the mapping relationship, and select the control sequence that minimizes the control cost from the trajectory that satisfies the constraints to generate execution instructions;

[0040] The feedback correction module is used to control the corresponding mine equipment based on the execution command and collect feedback data to update the mapping relationship; when the feedback data deviates from the effective range, it triggers cloud reporting and corrects the global constraint set.

[0041] A third aspect of the present invention provides an electronic device, comprising:

[0042] processor;

[0043] Memory used to store processor-executable instructions;

[0044] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0046] This invention utilizes edge nodes for equipment data acquisition and local control, while the cloud handles global resource coordination. This fully leverages the real-time capabilities of edge computing and the global optimization capabilities of cloud computing, improving the overall response speed and efficiency of the control system. By constructing a global constraint set and solving for resource allocation schemes based on this set, the optimal configuration of mine equipment resources is achieved, avoiding resource waste and equipment conflicts. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the intelligent control method for mine equipment clusters based on cloud-edge collaboration, as described in an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes will not be repeated in some embodiments.

[0050] Figure 1 This is a flowchart illustrating the intelligent control method for mine equipment clusters based on cloud-edge collaboration according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0051] The operational status data of the mine equipment group is collected through edge nodes;

[0052] The operating state data is decomposed into time series at the edge layer. Periodic and transient components are separated by frequency domain transformation. The amplitude features of the periodic components and the rate of change features of the transient components are extracted as local state vectors, and a mapping relationship from the local state vectors to the control output is constructed. The statistical features of the local state vectors and the effective range of the mapping relationship are uploaded to the cloud. The cloud identifies the coupling relationship between devices based on the statistical features of multiple nodes and constructs a global constraint set for resource allocation based on the effective range.

[0053] The cloud platform solves the resource allocation scheme based on the global constraint set, generating resource quotas and priority sequences for each edge node. The edge nodes use the resource quotas and priority sequences as constraints, and use the mapping relationship to predict the state trajectory of the mine equipment. Among the trajectories that satisfy the constraints, they select the control sequence that minimizes the control cost to generate execution instructions.

[0054] The edge node controls the corresponding mining equipment based on the execution command and collects feedback data to update the mapping relationship; when the feedback data deviates from the effective range, it triggers cloud reporting and corrects the global constraint set.

[0055] In one optional implementation, the steps of performing temporal decomposition on the operating state data at the edge layer, extracting local state vectors, and constructing a mapping relationship between the local state vectors and the control output include:

[0056] The operating status data is divided into multiple time segments according to a time window. Frequency domain transformation is performed on each time segment to separate periodic components and transient components. The amplitude characteristics of the periodic components and the rate of change characteristics of the transient components are extracted as local state vectors.

[0057] Based on the historical sequence of the local state vectors, a causal dependency rule between the local state vectors and historical control outputs is established through state-space mapping, forming a mapping relationship;

[0058] The sensitivity weights of each dimension component of the local state vector to the control output are calculated according to the causal dependency rule, and the range of state vector values ​​whose sensitivity weights exceed the preset stability interval are identified as the effective range of the mapping relationship.

[0059] For example, the data acquisition module deployed on the edge node connects to the sensor network of the mine equipment group via an industrial Ethernet interface. The acquisition cycle is set to 100 milliseconds. Each acquisition covers physical quantities such as current, voltage, rotational speed, temperature, and vibration acceleration of the equipment. The data format uses a structured record of timestamps and multidimensional numerical pairs, with timestamp accuracy at the microsecond level to ensure the accuracy of subsequent time-series analysis. The acquired operating status data is stored in a circular buffer on the edge node, with a buffer capacity set to hold the most recent 10 minutes of historical data. When the buffer is full, old data is discarded according to a first-in-first-out (FIFO) strategy. During data acquisition, a dual mechanism of hardware interrupt and software watchdog is set. If no valid data is received for three consecutive acquisition cycles, a sensor fault flag is triggered, and a backup sensor channel or an interpolation completion strategy is activated. The interpolation uses a linear fitting method and only allows completion of no more than five consecutive missing points. If the number exceeds this range, it is marked as an invalid segment and removed in subsequent processing.

[0060] After receiving the running status data in the buffer, the time-series decomposition module divides the data into segments with a fixed time window length of 30 seconds. Adjacent segments overlap by 50% to avoid information loss due to boundary effects. Each time segment contains 3000 sampling points. A Fast Fourier Transform (FFT) is performed on each segment to convert the time-domain signal to the frequency domain. Before the transform, a Hanning window function is applied to smooth the segment edges to suppress spectral leakage. After the frequency domain transformation, periodic and transient components are identified through frequency energy distribution. Specifically, the power spectral density of each frequency point is calculated, and frequencies with a power spectral density exceeding twice the mean are marked as the dominant frequency of the periodic component, and their corresponding amplitudes are extracted as the amplitude features of the periodic component. Transient component identification targets frequency bands with a power spectral density less than 0.5 times the mean. The rate of change of energy over time within this frequency band is calculated. This rate of change is obtained by dividing the energy difference between adjacent time windows by the time interval, and this value serves as the rate of change feature of the transient component. The periodic component amplitude features and the transient component rate of change features together constitute the local state vector. The vector dimension is dynamically determined according to the equipment type. Under typical configurations, 6-dimensional features are extracted for fan equipment, 8-dimensional features are extracted for water pump equipment, and 10-dimensional features are extracted for transportation equipment.

[0061] The mapping relationship construction module maintains a historical state vector database, which stores local state vectors and their corresponding control output records for the most recent 24 hours using a time-series index. Control outputs include discrete or continuous control quantities such as valve opening, motor speed setpoint, and start / stop commands. The state-space mapping employs a nonlinear regression model. The model input is a sequence of local state vectors for the current time and the previous five time steps, and the output is the predicted control output for the next time step. Model training uses a sliding window strategy, triggering incremental training whenever the database accumulates 1000 new records. The training process uses a gradient descent optimization algorithm with a learning rate of 0.001. The maximum number of iterations is 200, or the process terminates early when the loss function converges to a change in magnitude less than 0.0001. The causal dependency rule is established by analyzing the time delay correlation between each dimension component of the state vector and the control output. Specifically, the cross-correlation function between the state vector component and the control output is calculated at different time offsets. The time offset corresponding to the peak of the cross-correlation coefficient is the causal delay of that component to the control output. Components with an absolute value of cross-correlation coefficient greater than 0.6 are identified as having significant causal dependencies. These causal dependencies are stored in a directed graph structure to form the topological basis of the mapping relationship.

[0062] The sensitivity weight calculation module quantifies the influence of each dimension component of the local state vector on the control output for each causal dependency rule in the mapping relationship using a perturbation analysis method. During the perturbation analysis, other dimension components are kept constant, and a small perturbation of ±5% is applied to the target dimension component. The corresponding change in control output is observed, and the ratio of the control output change to the input perturbation amplitude is defined as the sensitivity weight of that dimension component. After the sensitivity weight calculation is completed, the sensitivity weight distribution of that dimension component at all historical moments is statistically analyzed, and the mean and standard deviation of the distribution are calculated. A preset stable interval is defined as the range of the mean plus or minus two standard deviations. If a state vector value causes its sensitivity weight to fall outside this stable interval, then this range is identified as the effective boundary of the mapping relationship. The effective range is expressed as the upper and lower limits of each dimension component. For example, the effective range of the periodic component amplitude characteristic of a wind turbine is 0.8 to 1.2 times the rated value, and the effective range of the transient component rate of change characteristic is no more than 0.3 times the rated value per second. The effective range update cycle is set to be executed once every 6 hours. During the update, the sensitivity weight distribution of the most recent 24 hours needs to be recalculated to adapt to equipment aging or changes in operating conditions. The updated effective range and statistical features are packaged together and uploaded to the cloud for global collaborative use.

[0063] By adopting this method, edge nodes can quickly identify abnormal equipment status locally and accurately locate the applicable boundaries of the mapping relationship, providing reliable local feature input for global resource collaboration in the cloud, and significantly improving the real-time performance and robustness of mine equipment group control.

[0064] In one optional implementation, the effective range of the statistical features and mapping relationship of the local state vector is uploaded to the cloud; the cloud identifies the coupling relationship between devices based on the statistical features of multiple nodes, and constructs a global constraint set for resource allocation in combination with the effective range, including the following steps:

[0065] Statistical analysis is performed on the distribution of the local state vector within a preset time window, and the mean offset and variance fluctuation amplitude are calculated as statistical features. The effective range of the statistical features and the mapping relationship is then encapsulated and uploaded to the cloud.

[0066] After receiving the statistical features uploaded by multiple edge nodes, the cloud calculates the cross-correlation coefficient between the mean offsets of different edge nodes, identifies node pairs whose cross-correlation coefficient exceeds the independence threshold, and marks the mining equipment corresponding to the node pairs as having a coupling relationship;

[0067] For the effective range uploaded by each edge node, the boundary values ​​of each dimension are extracted and projected in the global coordinate system. The intersection of the feasible regions of the effective range of each edge node in the resource configuration space is calculated. Based on the intersection of the feasible regions and the coupling relationship, a global constraint set describing the resource competition conflict between devices is generated.

[0068] For example, after completing the extraction and mapping of local state vectors, the edge nodes quantify the distribution characteristics of local state vectors within a preset time window. The preset time window length is set to 5 minutes, containing 3000 local state vector samples. Statistical analysis is performed independently for each dimension component. The mean offset is calculated by subtracting the arithmetic mean of each dimension component within the current time window from the historical baseline mean of that dimension component. The historical baseline mean is the weighted average of the mean values ​​of all time windows within the most recent 24 hours, with weights decaying over time; windows closer to the current time have greater weights, and the decay coefficient is set to 0.95. The variance fluctuation amplitude is calculated by extracting the sample variance of each dimension component within the current time window and comparing it to the historical baseline variance. A ratio greater than 1 indicates increased volatility, while a ratio less than 1 indicates decreased volatility. The historical baseline variance also uses a 24-hour weighted average strategy. The statistical feature data structure includes edge node identifiers, device type codes, timestamps, mean offset arrays and variance fluctuation amplitude arrays for each dimension component. During encapsulation, the effective range data of the mapping relationship is simultaneously carried, encoded as upper and lower limit pairs for each dimension component. After data encapsulation, it is uploaded to the cloud via an encrypted channel.

[0069] After receiving statistical feature data from multiple edge nodes, the cloud stores the data in a distributed time-series database aligned with device type and time window. The database employs a partitioned indexing strategy, with the primary key being a combination of the edge node identifier and timestamp, and secondary indexes containing device type and dimensional component numbers. The mean offset data of all edge nodes within the same time window is extracted from the database. For any pair of edge nodes, the cross-correlation coefficient between the mean offset sequences of each dimensional component is calculated. The cross-correlation coefficient is calculated using the Pearson correlation coefficient method, taking the product of the covariance of the mean offsets of the corresponding dimensional components of two nodes over 10 consecutive time windows and their respective standard deviations. The result ranges from -1 to +1, with a stronger correlation indicating a closer absolute value to 1. An independence threshold is set to 0.7. When the absolute value of the cross-correlation coefficient for any dimensional component of a node pair exceeds this threshold, the node pair is identified as having a coupling relationship. The coupling relationship record includes the identifiers of the two nodes, the coupling dimension number, the cross-correlation coefficient value, and the discovery timestamp. If multiple dimensions of a node pair simultaneously exceed the independence threshold, the dimension with the largest absolute value of the cross-correlation count is selected as the primary coupling dimension, and the remaining dimensions are recorded as secondary coupling dimensions. Coupling relationships are stored in a directed graph data structure, where nodes represent edge nodes, edges represent coupling relationships, and edge weights are set to the absolute value of the cross-correlation count. The graph structure supports dynamic updates; when the cross-correlation count of a node pair is below half the independence threshold for three consecutive time windows, the corresponding coupling relationship edge is deleted to avoid interference from outdated information.

[0070] The boundary values ​​of each dimension component are extracted from the valid range data uploaded by each edge node. These boundary values ​​include upper and lower limits. The global coordinate system is defined as a multi-dimensional Euclidean space, where the number of dimensions equals the total number of dimension components involved in all edge nodes. The valid range of each edge node is represented as a hyperrectangular region in this space. The projection operation maps the valid range of each edge node from local coordinates to global coordinates. The mapping rule is to convert local dimension numbers into global dimension indices, and the conversion table is pre-configured according to the device type and physical quantity type. The feasible region intersection calculation compares the boundary values ​​of the valid ranges of all edge nodes dimension by dimension, extracting the maximum value of all lower limits in each dimension as the lower intersection limit and the minimum value of all upper limits as the upper intersection limit. If the lower intersection limit is greater than the upper intersection limit, then there is no feasible region for that dimension, and it is marked as a resource conflict dimension. The resource configuration space is defined as a multi-dimensional space with shared resources such as power supply capacity, ventilation flow, and transportation bandwidth as coordinate axes. The valid range of each edge node is converted and projected into the resource configuration space using parameters such as device rated power consumption, airflow demand, and transportation volume. The conversion factor is determined based on device nameplate parameters and historical operating data statistics. When generating the global constraint set, the boundary conditions of the feasible region intersection are converted into linear inequality constraints. Two constraints with corresponding upper and lower limits are generated for each dimension. The coupling relationship is transformed into dependency constraints between nodes, meaning that if a certain dimension component of node A exceeds the baseline value, the corresponding dimension component of node B must be adjusted accordingly to maintain the cross-correlation relationship. The coefficient of the dependency constraint is equal to the cross-correlation coefficient. The global constraint set is stored in the form of a constraint matrix. The rows of the matrix represent constraint entries, the columns represent control variables of each edge node, the matrix elements are constraint coefficients, and the right-hand side of the matrix represents the constraint boundary values. The constraint set is updated every minute, and historical versions are retained during updates to support rollback operations.

[0071] This invention accurately captures implicit coupling relationships through correlation analysis of multi-node statistical features, and achieves automated construction of global resource constraints by combining effective range projection, which significantly improves the safety and resource utilization efficiency of collaborative control of mine equipment groups.

[0072] In one optional implementation, the step of generating resource quotas and priority sequences for each edge node by solving the resource allocation scheme based on the global constraint set in the cloud includes:

[0073] The cloud platform determines the feasible solution space for resource allocation based on the intersection of feasible regions in the global constraint set. Within this feasible solution space, an objective function is established with the optimization goal of minimizing the energy consumption of the mine equipment group and maximizing the global resource utilization rate. The weight coefficients of the weighted combination are dynamically adjusted according to the urgency of the current mine production task and the health status of the equipment. Iterative search is performed to find the solution that achieves the optimal value of the objective function while satisfying all constraints in the global constraint set, which is then used as the resource allocation scheme.

[0074] Based on the resource allocation scheme and the dependency hierarchy of each edge node control device in the inter-device coupling relationship, calculate the resource quota for each edge node; construct a dependency graph structure based on the inter-device coupling relationship and perform topological sorting to generate a priority sequence for each edge node;

[0075] For the resource quota of each edge node, the negotiable space is determined according to the sensitivity of the coupling component of the corresponding device in the device coupling relationship; the resource quota, the priority sequence and the negotiable space are encapsulated and sent to the corresponding edge node.

[0076] For example, the cloud extracts boundary conditions of the intersection of feasible regions from the global constraint set, and combines the lower and upper limits of the intersection of resources in each dimension to form a feasible solution space for resource allocation. The feasible solution space is represented in the form of a multi-dimensional hypercube, where each dimension corresponds to a type of shared resource, and the range of dimension values ​​is the allocatable interval of that resource. When establishing the optimization objective function, the energy consumption minimization objective is calculated by accumulating the power consumption of all edge node control devices in the prediction time domain. The power consumption is estimated based on the rated power of the device × the running time, and the prediction time domain is set to the next 15 minutes. The resource utilization maximization objective is obtained by calculating the arithmetic mean of the utilization rates of all resource types by dividing the actual usage time of each shared resource by the total available time. In the weighted combination, the sum of the weight coefficients for minimizing energy consumption and maximizing resource utilization is always 1. The weight coefficients are dynamically adjusted according to the urgency of the current mine production task and the health status of the equipment. The urgency is quantified by weighted summation of the queue length of the pending tasks and the task priority. The task priority is divided into three levels: high, medium, and low, with corresponding weight values ​​of 1.0, 0.6, and 0.3. The urgency score is obtained by summing the priority weight values ​​of each task in the queue and dividing by the queue length. The score range is from 0.3 to 1.0. The resource utilization weight = 0.5 + (urgency score - 0.65), and the energy consumption weight = 1 - resource utilization weight. Equipment health status is assessed by collecting data on equipment temperature rise, vibration amplitude, and operating time. Temperature rise exceeding the rated value by 80% is scored as 0.3 points, vibration exceeding the standard by 60% is scored as 0.3 points, and continuous operation exceeding 8 hours is scored as 0.4 points. The health score is calculated by adding these three values ​​and subtracting the sum from 1. When the health score is below 0.7, the energy consumption weight is adjusted. The adjusted energy consumption weight is calculated as: original energy consumption weight + (1 - health score) × 0.5. The adjusted resource utilization weight is then adjusted to 1 minus the adjusted energy consumption weight to ensure that the weight sum remains 1.

[0077] The iterative search employs a particle swarm optimization algorithm to find the optimal solution for the objective function within the feasible solution space. The number of particles is set to 50, with each particle representing a resource allocation scheme. The components of the particle's position vector correspond to the resource allocation amount for each edge node. During iteration, each particle updates its velocity and position based on its historical best position and the global best position. The inertia weight for velocity updates is set to 0.6, and the cognitive and social coefficients are both set to 1.5. Position updates require verification of all constraints in the global constraint set, including feasible region boundary constraints and inter-device coupling dependency constraints. Particles that do not meet the constraints are projected to the nearest feasible region boundary point. The iteration terminates after 200 iterations or 20 consecutive iterations where the change in the global optimal value is less than 0.001. After the iteration ends, the position vector of the globally optimal particle represents the resource allocation scheme. The resource allocation scheme is stored in structured data format, containing the identifiers of each edge node and their corresponding resource allocation amounts. The unit of resource allocation is determined based on the resource type.

[0078] The resource quota for each edge node is calculated based on the dependency hierarchy in the resource allocation scheme and the coupling relationship between devices. The dependency hierarchy is determined by analyzing the in-degree and out-degree of nodes in the directed graph of the coupling relationship. Nodes with an in-degree of 0 are identified as independent devices with a hierarchy of 0. The hierarchy of other nodes is equal to the maximum value of the hierarchy of all their predecessor nodes plus 1. When calculating the resource quota, nodes with higher hierarchies receive the resource amount specified in the resource allocation scheme first. Nodes with the same hierarchy are allocated resources according to the weighted proportion of their cross-relationship coefficients. If the allocation amount for a node in the resource allocation scheme is 100 kWh, and the node's hierarchy contains 3 nodes with cross-relationship coefficients of 0.8, 0.6, and 0.5 respectively, then the actual quota for that node is 100 × 0.8 ÷ 1.9 = 42.1 kWh. The calculation result is rounded to one decimal place. The dependency graph structure is stored in the form of an adjacency list. The node table entries contain edge node identifiers and level numbers, and the edge table entries contain source nodes, target nodes, and the number of mutual relations. The topology sorting adopts a depth-first traversal algorithm. The traversal starts from all nodes with an in-degree of 0. During the traversal, the node visit order is recorded. This order is the priority sequence, and the nodes that are earlier in the sequence have higher priority.

[0079] For each edge node's resource quota, all directed coupling edges of that node as a responding device are extracted from the inter-device coupling relationships. The product of the cross-correlation coefficient of each coupling edge and the time delay is calculated as the coupling influence factor. The time delay is determined by the time offset corresponding to the peak position of the cross-correlation function, in seconds. The coupling component sensitivity of the edge node is obtained by normalizing and summing all coupling influence factors. Normalization is achieved by dividing each coupling influence factor by the sum of all factors. If the coupling component sensitivity is lower than the weak coupling threshold of 0.3, the negotiable space is set as a fixed percentage expansion of the resource quota, with an expansion ratio of 15%, i.e., the resource quota multiplied by 1.15, as the upper limit of the negotiable space. If the sensitivity of the coupled component exceeds the strong coupling threshold of 0.7, the negotiable space adopts adaptive expansion. The expansion magnitude is calculated by multiplying the weighted average of the coupling influence factors by a dynamic adjustment coefficient. The dynamic adjustment coefficient is adjusted based on the frequency and magnitude of resource quota violations in the historical control cycles of the node. The frequency is the number of times the quota was exceeded in the last 24 hours divided by the total number of control cycles, and the magnitude is the average amount of quota violations divided by the resource quota. The adjustment coefficient is equal to 1 plus the product of the frequency and magnitude. The resource quota, priority sequence, and negotiable space are encapsulated into a binary message packet. The message header contains the edge node identifier, message type code, and timestamp. The message body contains the resource quota value, priority sequence position index, and negotiable space upper limit value. Message verification uses a cyclic redundancy check code. After encapsulation, the message is sent to the corresponding edge node through an encrypted channel. The confirmation timeout is 2 seconds. If no confirmation is received, the message is resent up to 3 times.

[0080] This invention achieves a balance between energy consumption and production capacity through dual-objective optimization and dynamic weight adjustment. Combined with a negotiated space design that is sensitive to coupling, it provides edge nodes with flexible execution leeway, significantly improving the global optimality and local adaptive capability of resource allocation.

[0081] In one optional implementation, the step of determining the negotiable space based on the sensitivity of the coupling component of the device corresponding to the edge node in the inter-device coupling relationship, for the resource quota of each edge node, includes:

[0082] Extract all directed coupling relationships when the device corresponding to the edge node acts as the response device from the inter-device coupling relationships. Calculate the product of the coupling strength and time delay of each directed coupling relationship as the coupling influence factor. Normalize and sum all coupling influence factors to obtain the coupling component sensitivity of the edge node.

[0083] The expansion range of the negotiable space is set according to the sensitivity of the coupling component. If the sensitivity of the coupling component is lower than the weak coupling threshold, the negotiable space is set to a fixed proportion of the resource quota expansion; if the sensitivity of the coupling component is higher than the strong coupling threshold, the negotiable space is set to an adaptive expansion range dynamically calculated based on the coupling influence factor.

[0084] For edge nodes that adopt adaptive expansion amplitude, the frequency and magnitude of resource quota breaches of the edge node in the historical control cycle are extracted. The frequency and magnitude are used as correction coefficients to weight and adjust the coupling influence factor, thereby generating a negotiable spatial boundary that takes into account both coupling dependency and historical execution deviation.

[0085] For example, the cloud executes a negotiated space determination process for resource quotas of each edge node. It queries the directed graph data structure of device coupling relationships to find all directed edges where the edge node is the target node. Each directed edge represents the coupling relationship between the corresponding device and a certain driving device when the node acts as a response device. The data record of the directed coupling relationship includes a source node identifier, a target node identifier, a coupling strength value, and a time delay value. The coupling strength is the absolute value of the aforementioned cross-correlation coefficient, ranging from 0 to 1. The time delay is calculated by performing a cross-correlation function on the state vector sequences of the two nodes; the time offset corresponding to the peak of the cross-correlation function is the time delay, in seconds, with a precision of 0.1 seconds. When calculating the coupling influence factor of each directed coupling relationship, the coupling strength value is multiplied by the time delay value to obtain the original coupling influence factor. If the time delay is 0, a minimum delay value of 0.1 seconds is set for calculation to avoid division by zero errors. After extracting the coupling influence factors for all directed coupling relationships of a given edge node, the sum of all coupling influence factors is calculated as the normalization base. Each coupling influence factor is then divided by the normalization base to obtain the normalized coupling influence factor. The sum of all normalized coupling influence factors equals 1. The sensitivity of the coupling component of this edge node is defined as the maximum value of all normalized coupling influence factors, ranging from 0 to 1. A larger value indicates a higher degree of influence of a single driving device on the node.

[0086] The expansion range of the negotiable space is set based on the sensitivity of the coupled components. The weak coupling threshold is set to 0.3, and the strong coupling threshold is set to 0.7. The range between these two thresholds is the medium coupling range. If the sensitivity of the coupled component is lower than the weak coupling threshold of 0.3, the edge node is less affected by other nodes, and the negotiable space adopts a fixed-ratio expansion strategy. The fixed expansion ratio is set to resource quota × 1.15, meaning the upper limit of the negotiable space is resource quota + resource quota × 0.15, and the expansion space is 15% of the resource quota. The adjustment range of the expansion ratio is 10% to 20%, with a default value of 15% suitable for general operating conditions. If the sensitivity of the coupled component is in the medium coupling range of 0.3 to 0.7, a linear interpolation strategy is used to determine the expansion ratio: expansion ratio = 0.15 + (coupled component sensitivity - 0.3) × 0.5. When the sensitivity is 0.3, the expansion ratio is 15%, and when the sensitivity is 0.7, the expansion ratio is 35%. The linear transition ensures that the expansion range matches the degree of coupling. If the sensitivity of the coupling component is higher than the strong coupling threshold of 0.7, the edge node is significantly affected by other nodes. The negotiable space adopts an adaptive expansion strategy, and the expansion range is calculated by combining the coupling influence factor and the historical execution deviation.

[0087] For edge nodes using adaptive scaling, records of resource quota violations for these nodes within the most recent 24-hour control period are extracted from the historical database. Each control period lasts 5 minutes, totaling 288 control periods within 24 hours. Resource quota violation is defined as the node's actual resource consumption exceeding its allocated resource quota. Violation records include a timestamp, quota value, actual consumption value, and the extent of the violation. The extent of the violation is calculated as (actual consumption value - quota value) ÷ quota value × 100%, quantified as a percentage. The violation frequency is calculated as the number of periods exceeding the quota within the 288 control periods. The frequency normalization value is calculated as the number of violation periods ÷ 288, ranging from 0 to 1. The average violation extent is calculated by summing the extent of violations from all records and dividing by the number of violation periods. This value reflects the severity of the quota violation, typically ranging from 5% to 50%. A correction factor is calculated by multiplying the frequency normalization value by the average violation extent. The correction factor ranges from 0 to 0.5; a larger value indicates a more significant historical execution deviation. When weighting the coupling influence factors of each directed coupling relationship, the adjusted coupling influence factor = original coupling influence factor × (1 + correction coefficient). The correction coefficient, as a magnification factor, reflects the reinforcing effect of historical deviations on coupling dependencies. The total adjusted influence factor is obtained by summing all adjusted coupling influence factors. The adaptive expansion amplitude = adjusted total influence factor ÷ original total influence factor - 1. This value represents the expansion ratio relative to the baseline quota and works synergistically with the sensitivity of the coupling components to generate a negotiable spatial boundary.

[0088] When calculating the negotiable space boundary, for edge nodes using a fixed-ratio expansion, the lower limit of the negotiable space equals the resource quota, and the upper limit equals the resource quota × 1.15. The boundary is stored in the form of a closed interval. For edge nodes using adaptive expansion, the lower limit of the negotiable space is also the resource quota, and the upper limit equals the resource quota × (1 + adaptive expansion magnitude). The upper limit of the adaptive expansion magnitude is constrained to 50% to prevent over-allocation of resources. If the calculated expansion magnitude exceeds 50%, it is truncated to 50%. The negotiable space data structure includes the edge node identifier, resource quota value, negotiable lower limit value, negotiable upper limit value, expansion strategy type identifier, and calculation timestamp. The data precision is retained to two decimal places and stored in a distributed cache system for subsequent scheduling decisions. The cache expiration time is set to 10 minutes, and recalculation is performed after expiration to adapt to dynamic changes. The validity of the negotiable space is verified by comparing the difference between the upper and lower limits. If the difference is less than 5% of the resource quota, it is determined to be an invalid negotiation space, triggering an alarm and reverting to a fixed 15% expansion strategy to avoid scheduling failure due to calculation anomalies.

[0089] This invention combines sensitivity grading of coupled components with historical deviation correction to accurately characterize the resource elasticity requirements of edge nodes. The generated negotiable space satisfies both coupled dependencies and takes into account actual execution characteristics, effectively improving the flexibility and robustness of resource scheduling.

[0090] In one optional implementation, the step of edge nodes using the resource quota and priority sequence as constraints, predicting the state trajectory of mine equipment using the mapping relationship, and selecting the control sequence that minimizes the control cost from the trajectories that satisfy the constraints to generate execution instructions includes:

[0091] The edge node receives the resource quota, priority sequence, and negotiable space from the cloud, converts the resource quota into the upper limit of the control variable, and converts the priority sequence into a temporal execution order constraint;

[0092] Using the mapping relationship, forward extrapolation is performed starting from the current state of the mine equipment to generate multiple candidate state trajectories that satisfy the resource quota limit and priority sequence time constraints within a preset prediction time domain. The validity of each candidate state trajectory is verified, and candidate state trajectories that fall within the effective range of the mapping relationship are marked as feasible trajectories.

[0093] For each feasible trajectory, the cumulative value of control quantities required to drive the mine equipment along the trajectory is calculated as the control cost. The trajectory with the minimum control cost is selected from the feasible trajectories, and the corresponding control quantity time series is extracted as the control sequence.

[0094] If the set of feasible trajectories is empty, the upper limit of the resource quota is extended to the boundary of the negotiable space, candidate state trajectories are regenerated, and the trajectory with the lowest control cost is selected as the control sequence. An adjustment feedback carrying the quota extension is generated and reported to the cloud; the control sequence is converted into an execution instruction.

[0095] For example, edge nodes receive resource quotas, priority sequences, and negotiable space data packets from the cloud via an encrypted channel. After decrypting the data packets, the resource quota values, priority sequence position indexes, and negotiable space upper limit values ​​are extracted and stored in the local cache. The process of converting resource quotas into upper limits for control variables is achieved by querying a conversion table between control variables and resource types. The conversion table pre-stores the rated parameters of the equipment and resource consumption coefficients. Taking a wind turbine as an example, the upper limit of the current control variable = resource quota ÷ rated voltage ÷ prediction time domain length. The voltage uses the nominal value on the equipment nameplate, and the prediction time domain length is set to 15 minutes, or 900 seconds. The resource quota unit is kilowatt-hours, which need to be converted to joules before participating in the calculation. 1 kilowatt-hour equals 3,600,000 joules. When converting a priority sequence into a temporal execution order constraint, the position index of the current edge node in the sequence is extracted. The smaller the position index, the higher the execution priority. The constraint means that the trigger time of the control action must be later than the control completion time of all nodes with higher priority. The temporal constraint is stored in the form of a timestamp interval. The lower limit of the interval is the maximum control completion time of the preceding node, and the upper limit of the interval is the end time of the predicted time domain. The time precision is 0.1 seconds.

[0096] Candidate state trajectories are generated by forward extrapolation starting from the current state of the mine equipment using a mapping relationship. The current state is represented by a state vector constructed from physical quantities such as equipment current, rotational speed, and temperature collected by sensors. The acquisition period is 100 milliseconds, and the dimension of the state vector is equal to the dimension of the mapping relationship input. The forward extrapolation adopts a discrete-time stepping method with a time step size of 1 second. The prediction time domain is 15 minutes, corresponding to 900 time steps. At each time step, the state vector for the next moment is calculated according to the mapping relationship. The input of the mapping relationship function is the current state vector and the control variable, and the output is the state vector for the next moment. During the generation of candidate state trajectories, the control variable is sampled within the range from 0 to the upper limit of the control variable value. The sampling strategy adopts a uniform discretization method, dividing the value range into 20 equally spaced sampling points. Each sampling point corresponds to a control strategy, and 20 candidate state trajectories are obtained from different sampling points. The resource quota upper limit constraint is verified by accumulating the resource consumption at each time step. The resource consumption = control variable × rated voltage × time step size ÷ 3,600,000, converted to kilowatt-hours. If the accumulated value does not exceed the resource quota, the upper limit constraint is satisfied. Priority sequence timing constraint verification is performed by comparing the trigger time of the control action with the lower limit of the timing constraint interval. If the trigger time is later than the lower limit, the timing constraint is satisfied. The trigger time is defined as the time step corresponding to the first non-zero time step of the control variable.

[0097] The validity of each candidate state trajectory is verified by determining whether all state vectors in the trajectory fall within the valid range of the mapping relationship. The valid range is stored in the form of upper and lower limit pairs for each dimension component. During verification, each dimension component of the state vector is extracted step by step and compared with the corresponding upper and lower limits. If any dimension component exceeds the upper or lower limit, the state at that time step is invalid. Trajectories containing invalid states are marked as infeasible trajectories, and trajectories where all time step states are valid are marked as feasible trajectories. The set of feasible trajectories is stored as a trajectory index list, with each element being a candidate trajectory number. The control cost is calculated for each feasible trajectory. The control cost is defined as the cumulative value of the control quantity required for the drive equipment to run along the trajectory. The cumulative value of the control quantity is obtained by summing the absolute values ​​of the control variables at all time steps in the trajectory. The larger the cumulative value, the larger the amplitude of the control action or the longer the duration, resulting in a corresponding increase in energy consumption and equipment wear. The control cost is calculated by traversing each trajectory in the set of feasible trajectories. The minimum control cost and the corresponding trajectory number are recorded. The control quantity time series of the trajectory with the minimum cost is the control sequence. The control sequence is stored in the form of an array. The array length is equal to the prediction time domain time step number 900. The array elements are the control variable values ​​at each time step. The numerical precision is retained to three decimal places.

[0098] An empty feasible trajectory set occurs when all candidate trajectories contain state vectors outside the valid range. In this case, the edge node expands the resource quota upper limit to the negotiable space boundary. After expansion, the upper limit of control variable values ​​is calculated as: Negotiable space upper limit ÷ Rated voltage ÷ Prediction time domain length. The control variable sampling and forward extrapolation process is then re-executed to generate new candidate state trajectories. The number of candidate trajectories remains 20 after expansion, with sampling points covering the expanded control variable value range. Validity checks and control cost calculations are performed again, and the trajectory with the lowest control cost among the expanded feasible trajectories is selected as the control sequence. The quota expansion amount is calculated as the negotiable space upper limit minus the resource quota, with the expansion amount in kilowatt-hours. This expansion amount is encapsulated as an adjustment feedback message, containing the edge node identifier, the original resource quota value, the expanded quota value, the expansion amount value, and a trigger timestamp. This message is reported to the cloud via an encrypted channel with a 2-second timeout. If reporting fails, the expansion event is recorded in the local log for subsequent auditing, without blocking the execution of the control sequence.

[0099] The process of converting control sequences into execution instructions is achieved by querying a mapping table between control variables and actuator instructions. This mapping table stores the numerical range of control variables and their corresponding actuator instruction codes. Taking a wind turbine as an example, the control variable is the current setpoint, the actuator is a frequency converter, and the instruction code is the frequency setpoint. The mapping relationship is: Frequency = Control Variable × Speed ​​Coefficient. The speed coefficient is configured according to the equipment nameplate parameters, with a typical value of 50 Hz corresponding to the rated current. The execution instruction data structure includes the actuator address, instruction type code, instruction parameter value, and execution time. The execution time is converted from the control sequence array index to the actual timestamp, using the conversion rule: current time + index × time step, with a time step of 1 second. The execution instructions are arranged in ascending order of time and stored in the instruction queue. The instruction queue adopts a priority queue data structure, and the dequeue priority is determined by the execution time. The execution module polls the queue every 10 milliseconds. When the execution time of the first instruction in the queue is less than or equal to the current time, the instruction is retrieved and sent to the actuator via the industrial bus. The transmission protocol adopts Modbus RTU, the baud rate is set to 9600 bits per second, the data bits are 8 bits, the stop bits are 1 bit, and there are no parity bits. After the instruction is sent, it waits for the actuator's response. The response timeout is 500 milliseconds. If the timeout occurs, it will resend up to 3 times. After 3 failures, a fault alarm is triggered and the instruction is skipped to continue executing subsequent instructions, ensuring the continuity of the control process.

[0100] This invention achieves autonomous trajectory planning and optimal control sequence selection for edge nodes by combining forward deduction of mapping relationships with constraint satisfaction verification.

[0101] In one optional implementation, the edge node controls the corresponding mining equipment based on the execution command and collects feedback data to update the mapping relationship; the step of triggering cloud reporting and correcting the global constraint set when the feedback data deviates from the effective range includes:

[0102] The edge node converts the execution command into control signals for the mining equipment and sends them to the corresponding mining equipment, collecting the actual operating status of the mining equipment after executing the control signals as feedback data;

[0103] Calculate the deviation between the feedback data and the predicted state corresponding to the execution instruction, and use the deviation to correct the mapping relationship;

[0104] Determine whether the feedback data falls within the effective range of the corrected mapping relationship. If it does, continue collecting feedback data at the next moment. If it deviates, extract the dimensions and magnitude of the feedback data that exceed the effective range.

[0105] The deviation type is determined based on the deviation magnitude. If the deviation magnitude is lower than the abnormal threshold, the feedback data and the corresponding deviation dimension are reported to the cloud. The cloud locates the affected constraints in the global constraint set based on the deviation dimension and performs boundary expansion correction. If the deviation magnitude reaches the abnormal threshold, the cloud is triggered to re-identify the coupling relationship between the devices and reconstruct the global constraint set.

[0106] For example, the edge node retrieves the instruction to be executed from the instruction queue, extracts the actuator address, instruction type code, and instruction parameter value from the instruction. The actuator address is encoded using a combination of the device number and channel number, with the device number occupying the high 8 bits and the channel number occupying the low 8 bits. The instruction type code identifies the control action type, and the parameter value is the specific control quantity value. The process of converting the execution instruction into a mine equipment control signal is achieved by accessing the actuator drive interface. The drive interface selects the corresponding protocol encapsulation strategy according to the actuator type. The frequency converter uses the Modbus RTU protocol. Function code 0x06 is written to a single register. The register address is mapped according to the instruction type code; frequency setting corresponds to register address 0x1000, current limiting corresponds to 0x1001, and parameter values ​​are encoded as 16-bit integers in big-endian byte order and written to the register data field. The control signal is sent to the mine equipment via the RS485 bus. The bus arbitration adopts a master-slave mode, with the edge node initiating communication as the master and the mine equipment responding to the request as the slave. The typical duration of a single communication is 20 milliseconds, including 8 milliseconds for transmission, 10 milliseconds for equipment processing, and 2 milliseconds for response transmission. After the control signal is issued, the edge node enters the status acquisition waiting state. The waiting time is set to the control signal activation delay plus the sensor sampling period. The activation delay is determined by the actuator response characteristics. The inverter output frequency stabilization time is about 300 milliseconds, the sensor sampling period is 100 milliseconds, and the total waiting time is set to 400 milliseconds.

[0107] The process of collecting actual operating status data of mine equipment as feedback data is achieved through polling the sensor interface. The sensor interface supports various types of sensors, such as current sensors, speed sensors, and temperature sensors. Each sensor is connected to the edge node through an analog input module or a digital input module. Analog input is achieved using a 12-bit ADC conversion, with the input voltage range of 0 to 10 volts mapped to a digital value of 0 to 4095. The digital value is converted into a physical quantity through linear calibration. The calibration coefficient of the current sensor is 0.11 amperes per digital unit. The output pulse frequency of the speed sensor is measured by a counter module. The cumulative number of pulses within a 100-millisecond counting period is multiplied by 600 to convert to revolutions per minute. The temperature sensor uses a PT100 resistance temperature detector (RTD), and the resistance value is converted into a voltage signal through a bridge circuit before being acquired by the ADC. Feedback data is organized in the form of state vectors. The number of vector dimensions is consistent with the number of dimensions of the mapping relationship output. Each dimension component corresponds to the measured value of a physical quantity. The state vector is appended with a timestamp to identify the acquisition time. The timestamp accuracy is 1 millisecond. The state vector is stored in a local time series database. The database adopts a circular buffer structure with a capacity of the acquisition records of the most recent hour. The size of a single record is about 64 bytes, and the capacity of 1 hour is about 2.3 million bytes. When the buffer is full, it overwrites the earliest record.

[0108] When calculating the deviation between the feedback data and the predicted state corresponding to the executed command, the predicted state vector corresponding to the trigger time of the executed command is extracted from the control sequence extrapolation record. The predicted state vector has been calculated and cached during the forward extrapolation process. The cache is a hash table indexed by the time step, and the lookup time complexity is constant. The deviation is calculated by subtracting the corresponding dimension component of the predicted state vector from the feedback data state vector, resulting in deviation components for each dimension. The deviation components are signed numerical values; positive values ​​indicate that the actual value is higher than the predicted value, and negative values ​​indicate that the actual value is lower than the predicted value. The strategy of correcting the mapping relationship using the deviation adopts an incremental learning method. The mapping relationship function is implemented using a neural network model with a three-layer fully connected network structure. The number of neurons in the input layer is equal to the dimension of the state vector plus the dimension of the control variables. The number of neurons in the hidden layer is set to 32, and the number of neurons in the output layer is equal to the dimension of the state vector. The ReLU function is used as the activation function. The correction process takes the current state vector and control variables as inputs, the predicted state vector plus the deviation as the correction target, and updates the network weights through the backpropagation algorithm. The learning rate is set to 0.001, each correction iteration is 1 epoch, the batch size is 1, the loss function is the mean squared error, and the mapping relationship after the weight update is used for subsequent predictions.

[0109] The test to determine whether the feedback data falls within the valid range of the corrected mapping relationship is achieved through dimension-by-dimensional comparison. The valid range data structure stores the lower and upper limits of each dimension component. These limits are determined statistically based on historical data during the mapping relationship construction phase. Typically, the lower limit is set to the mean minus three standard deviations, and the upper limit is set to the mean plus three standard deviations, covering 99.7% of normally distributed data. During the test, each dimension component of the feedback data is extracted and compared with the corresponding upper and lower limits of the valid range. If the component value is greater than or equal to the lower limit and less than or equal to the upper limit, the dimension falls within the valid range. If all dimensions fall within the valid range, the feedback data is considered valid, and the next feedback data is collected, with the collection period remaining constant at 100 milliseconds. If any dimension component exceeds the upper or lower limit, the feedback data is considered to deviate from the valid range. The dimension index and deviation magnitude of the deviating dimension are extracted. The deviation magnitude is calculated as the component value minus the upper limit (if the component value is greater than the upper limit), or the lower limit minus the component value (if the component value is less than the lower limit). The absolute value of the deviation magnitude represents the severity of the deviation, with the unit consistent with the corresponding physical quantity.

[0110] When determining the deviation type based on the exceedance magnitude, an anomaly threshold is set to 20% of the effective range width. The effective range width equals the upper limit minus the lower limit. The anomaly threshold is calculated by multiplying the width by 0.2. The adjustable range of the threshold is 10% to 30%, with a default value of 20% balancing sensitivity and fault tolerance. If the exceedance magnitude is less than the anomaly threshold, the deviation type is identified as boundary drift. The feedback data is encapsulated as a status reporting message. The message includes an edge node identifier, a feedback data status vector, an exceedance dimension index array, an exceedance magnitude array, and a collection timestamp. The message size is approximately 128 bytes. It is reported to the cloud through an encrypted channel. The reporting adopts an asynchronous non-blocking method, continuing the control flow without waiting for a cloud response. If the reporting fails, the message is stored in a local queue to be reported. The queue capacity is 100 messages. When the queue is full, the oldest message is discarded. After receiving the status reporting message, the cloud parses the exceedance dimension index to locate the affected constraints in the global constraint set. The constraints are stored in the form of linear inequalities. The row index of the constraint matrix corresponds to the constraint number, and the column index corresponds to the combination of edge nodes and dimensions. All constraint rows containing exceedance dimensions are retrieved by column index. Boundary expansion correction is achieved by adjusting the right-hand side of the constraints. If the excess magnitude of the out-of-dimension is positive, it means that the actual value exceeds the upper limit. In this case, the upper limit value of the right-hand side of the corresponding constraint is increased by 1.1 times the excess magnitude. The 1.1 times coefficient provides a 10% safety margin. If the excess magnitude is negative, it means that the actual value is lower than the lower limit. In this case, the lower limit value of the right-hand side is reduced by 1.1 times the excess magnitude. The corrected constraint set is updated, stored, and marked with a version number. The version number is incremented to ensure traceability.

[0111] If the deviation exceeds or equals the abnormal threshold, the deviation type is identified as abnormal fluctuation, triggering the cloud to re-identify the coupling relationship between devices. The edge node encapsulates the feedback data and the historical state vector sequence within the past 5 minutes into an abnormal reporting message. The historical sequence is extracted from the local time-series database, with a sequence length of 3000 sampling points corresponding to 5 minutes, and a message size of approximately 192 kilobytes, compressed to approximately 48 kilobytes. It is reported through a fragmented transmission mechanism, with a single fragment size of 4 kilobytes. The fragment sequence number and the total number of fragments are encoded in the message header. The cloud receives all fragments and reassembles the complete message. After receiving the abnormal reporting message, the cloud extracts the historical state vector sequence and the state vector sequences of other edge nodes within the same time window, and recalculates the cross-correlation matrix. The cross-correlation calculation method is consistent with the initial coupling relationship identification, using the Pearson correlation coefficient, and the calculation window is the received 5-minute historical sequence. Node pairs with an absolute cross-correlation value exceeding the independence threshold of 0.7 are identified, the directed graph of device coupling relationships is updated, old coupling edges with a cross-correlation value below 0.35 are deleted, and newly identified coupling edges are added, with the edge weights updated to the newly calculated cross-correlation value. When reconstructing the global constraint set, dependency constraints are regenerated based on the updated directed graph of coupling relationships. The dependency constraint coefficient is equal to the weight of the coupling edge. The right-hand side of the constraint is calculated based on the coupling strength and time delay between nodes. The calculation method is the same as that used to generate the initial constraint set. The reconstructed constraint set replaces the old version and triggers the recalculation of resource quotas for all edge nodes. The recalculated quotas and priority sequences are broadcast to each edge node via a broadcast message. The confirmation timeout is 5 seconds. Unconfirmed nodes are recorded in the exception list for manual intervention.

[0112] This invention achieves online adaptive correction of the mapping relationship through incremental learning of feedback deviation. Combined with a graded processing strategy for deviation magnitude, it ensures both rapid response to boundary drift and systematic reconstruction of abnormal fluctuations, effectively improving the self-healing capability and long-term stability of the control system.

[0113] A second aspect of the present invention provides an intelligent control system for a mine equipment cluster based on cloud-edge collaboration, comprising:

[0114] The data acquisition module is used to collect operational status data of the mine equipment group through edge nodes;

[0115] The edge analysis and reporting module is used to perform time-series decomposition on the operating status data at the edge layer, separate periodic components and transient components through frequency domain transformation, extract the amplitude characteristics of the periodic components and the rate of change characteristics of the transient components as local state vectors, construct the mapping relationship between the local state vectors and the control output, and upload the statistical characteristics of the local state vectors and the effective range of the mapping relationship to the cloud.

[0116] The cloud-based collaborative optimization module is used to identify the coupling relationship between devices based on the statistical characteristics of multiple nodes, construct a global constraint set for resource allocation based on the effective range, solve the resource allocation scheme according to the global constraint set, and generate resource quotas and priority sequences for each edge node.

[0117] The edge control decision module is used to use the resource quota and priority sequence as constraints, predict the state trajectory of the mine equipment using the mapping relationship, and select the control sequence that minimizes the control cost from the trajectory that satisfies the constraints to generate execution instructions;

[0118] The feedback correction module is used to control the corresponding mine equipment based on the execution command and collect feedback data to update the mapping relationship; when the feedback data deviates from the effective range, it triggers cloud reporting and corrects the global constraint set.

[0119] A third aspect of the present invention provides an electronic device, comprising:

[0120] processor;

[0121] Memory used to store processor-executable instructions;

[0122] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0123] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0124] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cloud-edge collaborative intelligent control method for mine equipment clusters, characterized in that, include: The operational status data of the mine equipment group is collected through edge nodes; The operational status data is decomposed temporally at the edge layer. Periodic and transient components are separated using frequency domain transformation. The amplitude features of the periodic components and the rate of change features of the transient components are extracted as local state vectors. A mapping relationship between the local state vectors and the control output is constructed. Specifically, this includes: dividing the operational status data into multiple time segments according to a time window; performing frequency domain transformation on each time segment to separate periodic and transient components; extracting the amplitude features of the periodic components and the rate of change features of the transient components as local state vectors; establishing causal dependency rules between the local state vectors and historical control outputs through state space mapping, forming a mapping relationship; calculating the sensitivity weights of each dimension component of the local state vectors to the control output according to the causal dependency rules; identifying the range of state vector values ​​whose sensitivity weights exceed a preset stability interval as the effective range of the mapping relationship; uploading the statistical features of the local state vectors and the effective range of the mapping relationship to the cloud; and using the statistical features of multiple nodes to identify the coupling relationships between devices and constructing a global constraint set for resource allocation based on the effective range. The cloud-based system solves for resource allocation schemes based on the global constraint set, generating resource quotas and priority sequences for each edge node. The edge nodes use the resource quotas and priority sequences as constraints, predict the state trajectory of the mine equipment using the mapping relationship, and select the control sequence that minimizes the control cost from the trajectories that satisfy the constraints to generate execution instructions. Edge nodes control corresponding mining equipment based on the execution instructions and collect feedback data to update the mapping relationship. When the feedback data deviates from the effective range, the cloud is triggered to report and correct the global constraint set. Specifically, the edge nodes convert the execution instructions into control signals for the mining equipment and send them to the corresponding mining equipment. The actual operating status of the mining equipment after executing the control signals is collected as feedback data. The deviation between the feedback data and the predicted state corresponding to the execution instructions is calculated, and the mapping relationship is corrected using the deviation. It is determined whether the feedback data falls within the effective range of the corrected mapping relationship. If it does, the feedback data is collected at the next moment. If it deviates, the dimension and magnitude of the feedback data exceeding the effective range are extracted. The deviation type is determined based on the magnitude of the deviation. If the magnitude of the deviation is lower than the abnormal threshold, the feedback data and the corresponding excess dimension are reported to the cloud. The cloud locates the affected constraints in the global constraint set based on the excess dimension and performs boundary expansion correction. If the magnitude of the deviation reaches the abnormal threshold, the cloud is triggered to re-identify the coupling relationship between the devices and reconstruct the global constraint set.

2. The method according to claim 1, characterized in that, The steps of uploading the statistical features and effective range of the mapping relationship of the local state vector to the cloud, and the cloud identifying the coupling relationship between devices based on the statistical features of multiple nodes and constructing a global constraint set for resource allocation based on the effective range, include: Statistical analysis is performed on the distribution of the local state vector within a preset time window, and the mean offset and variance fluctuation amplitude are calculated as statistical features. The effective range of the statistical features and the mapping relationship is then encapsulated and uploaded to the cloud. After receiving the statistical features uploaded by multiple edge nodes, the cloud calculates the cross-correlation coefficient between the mean offsets of different edge nodes, identifies node pairs whose cross-correlation coefficient exceeds the independence threshold, and marks the mining equipment corresponding to the node pairs as having a coupling relationship. For the effective range uploaded by each edge node, the boundary values ​​of each dimension are extracted and projected in the global coordinate system. The intersection of the feasible regions of the effective range of each edge node in the resource configuration space is calculated. Based on the intersection of the feasible regions and the coupling relationship, a global constraint set describing the resource competition conflict between devices is generated.

3. The method according to claim 1, characterized in that, The steps by which the cloud solves the resource allocation scheme based on the global constraint set and generates resource quotas and priority sequences for each edge node include: The cloud platform determines the feasible solution space for resource allocation based on the intersection of feasible regions in the global constraint set. Within this feasible solution space, an objective function is established with the optimization goal of minimizing the energy consumption of the mine equipment group and maximizing the global resource utilization rate. The weight coefficients of the weighted combination are dynamically adjusted according to the urgency of the current mine production task and the health status of the equipment. Iterative search is performed to find the solution that achieves the optimal value of the objective function while satisfying all constraints in the global constraint set, which is then used as the resource allocation scheme. Based on the resource allocation scheme and the dependency hierarchy of each edge node control device in the inter-device coupling relationship, calculate the resource quota for each edge node; construct a dependency graph structure based on the inter-device coupling relationship and perform topological sorting to generate a priority sequence for each edge node; For the resource quota of each edge node, the negotiable space is determined according to the sensitivity of the coupling component of the corresponding device in the device coupling relationship; the resource quota, the priority sequence and the negotiable space are encapsulated and sent to the corresponding edge node.

4. The method according to claim 3, characterized in that, For the resource quotas of each edge node, the step of determining the negotiable space based on the sensitivity of the coupling component of the corresponding device in the inter-device coupling relationship includes: Extract all directed coupling relationships when the edge node corresponds to the device as a response device from the inter-device coupling relationships. Calculate the product of the coupling strength and time delay of each directed coupling relationship as a coupling influence factor. Normalize and sum all coupling influence factors to obtain the coupling component sensitivity of the edge node. The expansion range of the negotiable space is set according to the sensitivity of the coupling component. If the sensitivity of the coupling component is lower than the weak coupling threshold, the negotiable space is set to a fixed proportion of the resource quota expansion; if the sensitivity of the coupling component is higher than the strong coupling threshold, the negotiable space is set to an adaptive expansion range dynamically calculated based on the coupling influence factor. For edge nodes that adopt adaptive expansion amplitude, the frequency and magnitude of resource quota breaches of the edge node in the historical control cycle are extracted. The frequency and magnitude are used as correction coefficients to weight and adjust the coupling influence factor, thereby generating a negotiable spatial boundary that takes into account both coupling dependency and historical execution deviation.

5. The method according to claim 3, characterized in that, The edge node uses the resource quota and priority sequence as constraints, and uses the mapping relationship to predict the state trajectory of the mine equipment. The step of selecting the control sequence that minimizes the control cost from the trajectory that satisfies the constraints to generate the execution instruction includes: The edge node receives the resource quota, the priority sequence, and the negotiable space from the cloud, converts the resource quota into the upper limit of the control variable, and converts the priority sequence into the order constraint of execution. Using the mapping relationship, forward extrapolation is performed starting from the current state of the mine equipment to generate multiple candidate state trajectories that satisfy the resource quota limit and priority sequence time constraints within a preset prediction time domain. The validity of each candidate state trajectory is verified, and candidate state trajectories that fall within the effective range of the mapping relationship are marked as feasible trajectories. For each feasible trajectory, the cumulative value of control quantities required to drive the mine equipment along the trajectory is calculated as the control cost. The trajectory with the minimum control cost is selected from the feasible trajectories, and the corresponding control quantity time series is extracted as the control sequence. If the set of feasible trajectories is empty, the upper limit of the resource quota is extended to the boundary of the negotiable space, candidate state trajectories are regenerated, and the trajectory with the lowest control cost is selected as the control sequence. An adjustment feedback carrying the quota extension is generated and reported to the cloud. The control sequence is converted into an execution instruction.

6. A cloud-edge collaborative intelligent control system for mine equipment clusters, used to implement the method of any one of claims 1-5, characterized in that, include: The data acquisition module is used to collect operational status data of the mine equipment group through edge nodes; The edge analysis and reporting module is used to perform time-series decomposition on the operating status data at the edge layer. It separates periodic and transient components through frequency domain transformation, extracts the amplitude characteristics of the periodic components and the rate of change characteristics of the transient components as local state vectors, and constructs a mapping relationship between the local state vectors and the control output. Specifically, this includes: dividing the operating status data into multiple time-series segments according to a time window; performing frequency domain transformation on each time-series segment to separate periodic and transient components; extracting the amplitude characteristics of the periodic components and the rate of change characteristics of the transient components as local state vectors; establishing a causal dependency rule between the local state vectors and historical control outputs through state-space mapping based on the historical sequence of the local state vectors, forming a mapping relationship; calculating the sensitivity weights of each dimension component of the local state vectors to the control output according to the causal dependency rules; identifying the range of state vector values ​​whose sensitivity weights exceed a preset stability interval as the valid range of the mapping relationship; and uploading the statistical characteristics of the local state vectors and the valid range of the mapping relationship to the cloud. The cloud-based collaborative optimization module is used to identify the coupling relationship between devices based on the statistical characteristics of multiple nodes, construct a global constraint set for resource allocation based on the effective range, solve the resource allocation scheme according to the global constraint set, and generate resource quotas and priority sequences for each edge node. The edge control decision module is used to use the resource quota and priority sequence as constraints, predict the state trajectory of the mine equipment using the mapping relationship, and select the control sequence that minimizes the control cost from the trajectory that satisfies the constraints to generate execution instructions; The feedback correction module is used to control the corresponding mine equipment based on the execution command and collect feedback data to update the mapping relationship. When the feedback data deviates from the effective range, it triggers cloud reporting and correction of the global constraint set. Specifically, it includes: the edge node converts the execution command into a control signal for the mine equipment and sends it to the corresponding mine equipment; collecting the actual operating state of the mine equipment after executing the control signal as feedback data; calculating the deviation between the feedback data and the predicted state corresponding to the execution command; using the deviation to correct the mapping relationship; determining whether the feedback data falls within the effective range of the corrected mapping relationship; if it falls within, continuing to collect feedback data at the next moment; if it deviates, extracting the dimension and magnitude of the feedback data exceeding the effective range; determining the deviation type based on the magnitude of ...

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images