Metadata interaction method and system for smart community based on edge computing framework

Abstract

The present application relates to the technical field of wisdom community data management, in particular to a wisdom community metadata interaction method and system based on an edge computing framework, comprising: dynamic storage of multidimensional data using hierarchical rolling cache; generation of a uniform resource vector through time sequence progressive mapping; addition of community semantic labels to the uniform resource vector through extensible label slots, and establishment of connections according to association rules to build a resource semantic graph oriented to community services; reception of resident task requests, construction of a multi-factor weighted scoring engine based on the resource semantic graph, and generation of a task data packet with a priority label; construction of a dual-channel hybrid scheduling network, output of resource availability and node correlation degree in the first channel, and output of time sequence characteristics and load prediction parameters in the second channel; and fusion of the dual-channel outputs to obtain task-node correlation and generate an optimal matching strategy. The present application realizes task collaboration and resource adaptive management through semantic resource modeling and hybrid scheduling.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart community data management, in particular to a smart community metadata interaction method and system based on an edge computing framework. BACKGROUND

[0002] The integration and intelligent scheduling of multi-source computing power data are core technologies in modern distributed computing systems, and their efficiency directly affects system performance and resource utilization.

[0003] Currently, mainstream solutions mostly use a unified resource abstraction layer to standardize the management of heterogeneous computing power and use machine learning algorithms to design an intelligent scheduling engine for task allocation. However, existing resource abstraction methods are prone to scheduling delays and lack flexibility when facing high-dynamic and large-scale computing power demands. Meanwhile, intelligent scheduling engines have difficulty in accurately establishing a dynamic mapping model of task demands and multi-dimensional real-time resource states, resulting in low computing power matching accuracy, and the problems of resource waste and processing bottlenecks are still prominent.

[0004] Therefore, in the cloud, edge, and end collaborative computing scenario, the overall scheduling efficiency and resource utilization of existing solutions still have a lot of room for improvement.

[0005] To this end, the present application proposes a smart community metadata interaction method and system based on an edge computing framework. SUMMARY

[0006] The purpose of the present application is to propose a smart community metadata interaction method and system based on an edge computing framework, which generates a unified resource vector through hierarchical cache management of end, edge, and cloud data and time sequence mapping and constructs a resource semantic graph, accurately matches resources and tasks through a dual-channel hybrid scheduling network, reduces overhead, and improves service efficiency.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] The smart community metadata interaction method based on the edge computing framework comprises:

[0009] Multi-dimensional data of smart community end, edge, and cloud nodes are acquired and dynamically stored using hierarchical rolling cache based on community life rules; the multi-dimensional data is converted into a unified resource vector through time sequence progressive mapping; the edge controller adds community semantic labels to the unified resource vector through extensible label slots and establishes connections according to association rules to construct a resource semantic graph for community services;

[0010] A resident task access unit is set and receives resident task requests, a multi-factor weighted scoring engine is constructed based on the resource semantic graph to quantitatively evaluate service priorities, and a task data packet with priority labels is generated;

[0011] A double-channel hybrid scheduling network is constructed, a first channel performs graph neural network inference based on a resource semantic graph to output resource availability and node association degree, and a second channel outputs timing characteristics and load prediction parameters based on task data packets; the task-node correlation is obtained by fusing the resource availability, the node association degree, the timing characteristics and the load prediction parameters, and the optimal matching strategy is generated.

[0012] Preferably, the process of acquiring multi-dimensional data of the wisdom community end, edge and cloud nodes is: collecting multi-dimensional data of wisdom community terminal equipment state, multi-dimensional data of edge gateway load and multi-dimensional data of cloud server resources, the multi-dimensional data including device identification, timestamp, location information, energy consumption index and service type.

[0013] Preferably, the process of dynamic storage using hierarchical rolling cache based on community life rules is: establishing a hierarchical rolling cache architecture according to the work and rest rules of community residents and the device use cycle, the hierarchical rolling cache architecture including: a hot data cache layer storing recent high-frequency access data and maintaining a short period time window; a warm data cache layer storing medium-frequency access data and maintaining a medium period time window, retaining the last valid state; a cold data cache layer storing low-frequency access and archiving data and maintaining a long period time window, with simplified index and source identification.

[0014] Preferably, the process of converting multi-dimensional data into a unified resource vector through time sequence progressive mapping is: performing time sequence progressive mapping on multi-dimensional data, extracting data feature sequences through a sliding time window, and converting discrete features into continuous vector representation using embedding units; adopting normalization processing to unify data features of different dimensions to the same numerical range, and merging device identification, timestamp, location information, energy consumption index and service type into a fixed-dimensional unified resource vector through a feature fusion algorithm.

[0015] Preferably, the process of constructing a resource semantic graph for community services is: dynamically adding community semantic labels to the unified resource vector through an extensible label slot in the edge controller; calculating the association rules between the unified resource vectors based on cosine similarity and mutual information theory to determine the connection strength between nodes; establishing the connection relationship between nodes using a graph construction algorithm, and using sparse adjacency representation to record only the connected edges and edge index information.

[0016] Preferably, the extensible label slot includes: a location semantic label slot identifying the physical location and logical area to which the resource belongs; a function semantic label slot describing the service type and processing capacity of the resource; a state semantic label slot recording the real-time state and availability information of the resource; and a priority semantic label slot identifying the importance and scheduling priority of the resource in community services.

[0017] Preferably, the specific implementation process of constructing a multi-factor weighted scoring engine based on the resource semantic graph includes: extracting task-related node features from the resource semantic graph, including clustering coefficient and shortest path length; using the clustering coefficient and the shortest path length to establish a multi-factor scoring matrix, including resource availability factor, service matching degree factor and geographical location factor; using the analytic hierarchy process to determine the weight coefficients of each factor, and calculating the comprehensive score by weighted summation; prioritizing the tasks according to the scoring results, and generating task data packets with priority labels.

[0018] Preferably, the double-channel hybrid scheduling network includes: a first channel adopts a graph neural network, taking the resource semantic graph as input, learning node embedding representation through multi-layer graph convolution operation, and outputting resource availability and node correlation degree of each node; the second channel adopts a long short-term memory network, taking the task data packet as input, generating time sequence features and load prediction parameters; based on the load prediction parameters, the load allocation scheme is output by the scheduling optimizer.

[0019] Preferably, the process of fusing resource availability, node correlation degree, time sequence features and load prediction parameters to obtain task-node correlation and generate optimal matching strategy is: constructing a task-node correlation calculation matrix, normalizing the scores of the four dimensions of resource availability, node correlation degree, time sequence features and load prediction parameters; using a multi-objective optimization algorithm to calculate the matching correlation score between tasks and nodes, and establishing constraint conditions including resource capacity limit, service quality requirement and response time threshold; under the constraint conditions, the minimum cost flow algorithm is used to solve the optimal allocation scheme and generate the optimal matching strategy.

[0020] The smart community metadata interaction system based on the edge computing framework comprises:

[0021] The data caching module: acquires multi-dimensional data of the smart community end, edge and cloud nodes, and dynamically stores the data by using hierarchical rolling caching based on community life rules;

[0022] The semantic mapping module: converts the multi-dimensional data into a unified resource vector through time sequence progressive mapping; the edge controller adds community semantic labels to the unified resource vector through extensible label slots, and establishes connections according to association rules, thereby constructing a resource semantic graph for community services;

[0023] The task evaluation module: sets a resident task access unit and receives resident task requests, constructs a multi-factor weighted scoring engine based on the resource semantic graph, quantitatively evaluates the service priority, and generates task data packets with priority labels;

[0024] The matching strategy module: a double-channel hybrid scheduling network is constructed, a first channel performs graph neural network inference based on a resource semantic graph to output resource availability and node correlation degree, and a second channel outputs timing characteristics and load prediction parameters based on task data packets; resource availability, node correlation degree, timing characteristics and load prediction parameters are fused to obtain task-node correlation and generate an optimal matching strategy.

[0025] Compared with the prior art, the present application has the following beneficial effects:

[0026] 1. By constructing a hierarchical rolling cache architecture based on community living rules, including three-layer cache mechanisms of hot data, warm data and cold data, dynamic storage is performed according to resident work and rest rules and device use cycles, network overhead and access delay are greatly reduced, and multi-dimensional data is efficiently circulated among end, edge and cloud nodes.

[0027] 2. By time sequence progressive mapping, heterogeneous multi-dimensional data is converted into a unified resource vector, community semantic labels are dynamically added in combination with an extensible label slot, an association rule is established based on cosine similarity and mutual information theory, a resource semantic graph for community services is constructed, data heterogeneity is eliminated and accurate resource description is provided.

[0028] 3. A resident task access unit is set and a multi-factor weighted scoring engine is constructed, node features are extracted and a scoring matrix including resource availability, service matching degree, geographic location and other factors is established, an analytic hierarchy process is used for quantitative evaluation, resident task requests are standardized into task data packets with priority labels, and the pertinence and efficiency of task processing are improved.

[0029] 4. A double-channel hybrid scheduling network is used to process resource semantic graphs and task data packets, a first channel outputs resource availability and node correlation degree through a graph neural network, a second channel outputs timing characteristics and load prediction parameters through a long short-term memory network, four-dimensional information is fused to generate an optimal matching strategy, and matching deviation caused by single-dimensional information is avoided. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 A flowchart of a smart community metadata interaction method based on an edge computing framework provided by the embodiment of the present application is shown;

[0031] Figure 2 A structural diagram of an extensible label slot provided by the embodiment of the present application is shown;

[0032] Figure 3 A flowchart of a multi-factor weighted scoring engine constructed based on a resource semantic graph provided by the embodiment of the present application is shown;

[0033] Figure 4A structural schematic diagram of a smart community metadata interaction system based on an edge computing framework is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0035] The present application proposes a smart community metadata interaction method and system based on an edge computing framework. A layered cache management end, edge, cloud data, and time sequence mapping are used to generate a uniform resource vector and construct a resource semantic graph. A double-channel hybrid scheduling network precisely matches resources and tasks, reduces overhead, and improves service efficiency. The effectiveness of the present application will be illustrated from two embodiments below.

[0036] Embodiment one:

[0037] The smart community metadata interaction method based on the edge computing framework has a specific process as shown in Figure 1 , which includes:

[0038] Multi-dimensional data of a smart community end, edge, and cloud node are acquired, and are dynamically stored by using layered rolling cache based on community life rules. The multi-dimensional data is converted into a uniform resource vector by time sequence progressive mapping. A community semantic label is added to the uniform resource vector by an extensible label slot in an edge controller, and a connection is established according to an association rule to construct a resource semantic graph for community services.

[0039] A resident task access unit is set and receives a resident task request. A multi-factor weighted scoring engine is constructed based on the resource semantic graph to quantitatively evaluate the service priority and generate a task data packet with a priority label.

[0040] A double-channel hybrid scheduling network is constructed. The first channel performs graph neural network inference based on the resource semantic graph to output resource availability and node correlation degree. The second channel outputs time sequence features and load prediction parameters based on the task data packet. The task-node correlation is obtained by fusing the resource availability, node correlation degree, time sequence features, and load prediction parameters to generate an optimal matching strategy.

[0041] Further, the process of acquiring multi-dimensional data of a smart community end, edge, and cloud node is as follows: multi-dimensional data of a smart community terminal device state, multi-dimensional data of an edge gateway load, and multi-dimensional data of a cloud server resource are collected. The multi-dimensional data includes device identification, time stamp, location information, energy consumption index, and service type.

[0042] Specifically, the device identifier adopts a globally unique identifier format to distinguish different terminal devices, edge gateways, and cloud server nodes; the timestamp adopts a high-precision time format to record the specific time of data generation, ensuring the synchronization of data timing among multiple nodes; the location information contains the physical coordinates and the logical area identifier of the device, which is used to identify the spatial distribution of resources; the energy consumption index reflects the real-time power consumption state and energy efficiency level of each node, including instantaneous power and cumulative energy consumption; the service type identifies the specific service functions provided by each node, such as data processing, storage service, network forwarding, etc., providing service capability description for subsequent resource matching.

[0043] By collecting five-dimensional data including device identifier, timestamp, location information, energy consumption index, and service type, a comprehensive data portrait of the smart community end, edge, and cloud nodes is established. The global uniqueness of the device identifier ensures accurate identification and tracking of multiple-level nodes; the high-precision timestamp ensures the consistency of data timing in a distributed environment, providing a reliable foundation for subsequent timing analysis; the physical coordinates and logical area dual identification of the location information support near-resource scheduling and regional service optimization based on geographic location; the real-time monitoring of the energy consumption index provides data support for green energy-saving scheduling decisions, effectively reducing the overall energy consumption of the community; the explicit identification of the service type avoids scheduling errors caused by mismatched resource capabilities, improving the success rate of task execution and service quality.

[0044] Further, the process of dynamic storage using hierarchical rolling cache based on community life rules is as follows: a hierarchical rolling cache architecture is established according to the work and rest rules of community residents and the device usage cycle, which includes: a hot data cache layer that stores recent high-frequency access data and maintains a short period of time window; a warm data cache layer that stores medium-frequency access data and maintains a medium period of time window, retaining the last valid state; a cold data cache layer that stores low-frequency access and archival data and maintains a long period of time window, with simplified indexing and source identification.

[0045] Specifically, the hot data cache layer sets a short period of 1-4 hours, mainly stores the data of intelligent access control, elevator control, parking management and other devices frequently used by community residents during morning and evening peak hours, and real-time security monitoring, environmental monitoring and other service requests, and uses the least recently used replacement strategy in the edge node to improve the cache hit probability; In view of the influence of network link and device interface conditions on end-to-end response time, this scheme is suitable for scenarios where local access is dominant and link is controllable, so as to reduce access latency; The warm data cache layer sets a medium period of 4-24 hours, stores device operation and maintenance data, resident life trajectory information and periodic property service requests with medium frequency access in community daily operation, and avoids repeated storage by retaining the last effective operation state of the device and service; The cold data cache layer sets a long period of more than 24 hours, stores low-frequency access archived information such as community historical energy consumption statistics, device maintenance records, and resident satisfaction survey, and realizes efficient cross-period retrieval through a simplified index structure and data source identifier. According to the work and rest rules of community residents and the frequency of device use, intelligent data migration is carried out between each cache level, the hot data cache is automatically expanded during the resident activity peak period, and the data is sunk to the warm and cold cache layers during the night valley period, so as to ensure the optimal configuration of storage resources.

[0046] By establishing a hierarchical rolling cache architecture based on community life rules, intelligent management of smart community data storage is realized. The hot data cache layer is aimed at the high-frequency access demand of residents during morning and evening peak hours, and keeps the data of key devices such as intelligent access control, elevator control and parking management within a limited response range, which helps to improve the real-time performance of community services and user experience; The warm data cache layer avoids repeated storage of routine operation and maintenance data through a medium period time window and an effective state retention mechanism, reduces storage overhead while ensuring access efficiency; The cold data cache layer uses long-term management and a simplified index structure to provide a cost-effective solution for long-term storage and statistical analysis of community historical data.

[0047] Further, the process of converting multi-dimensional data into a unified resource vector through time sequence progressive mapping is as follows: time sequence progressive mapping is performed on multi-dimensional data, data feature sequences are extracted through sliding time windows, and discrete features are converted into continuous vector representations using embedding units; Normalize the data features of different dimensions to the same numerical range, and merge device identifiers, timestamps, location information, energy consumption indicators and service types into fixed-dimensional unified resource vectors through feature fusion algorithms.

[0048] Specifically, the time series progressive mapping adopts a multi-scale sliding time window for segmented processing according to the multi-level data characteristics of the smart community. The short time window captures the real-time device state changes in the community, the medium time window extracts the daily activity patterns of the residents, and the long time window mines the seasonal rules of the community operation. The window sizes are set to 15 minutes, 2 hours, and 24 hours, respectively, and the time series feature sequence is gradually constructed through progressive steps. In the embedding unit processing, the smart community device identifier is generated as a unique vector representation through multi-layer hash coding, distinguishing different device types such as smart home, public facilities, and security monitoring. The timestamp is encoded using sine and cosine periodicity to capture the hourly and weekly community life rules. The location information is double-encoded using the community geographic coordinate system and building unit logic division. The energy consumption index is quantitatively encoded according to the community green energy-saving standards. The service type is hierarchically one-hot encoded based on the smart community service classification system. The normalization processing uses the min-max normalization method to unify the data features of different dimensions to the value range of zero to one, eliminating the dimensional differences of heterogeneous data such as device energy consumption power, geographic distance, and time span. In the feature fusion stage, an attention weighted splicing mechanism is used to dynamically adjust the weight distribution of the five dimensions according to different application scenarios in the smart community. The security scene emphasizes location and time features, the energy-saving scene highlights energy consumption and device features, and the convenience service scene balances the weight of each dimension. Finally, a 512-dimensional standardized unified resource vector is generated, ensuring the comparability and operability of all heterogeneous resources in the community in the same vector space.

[0049] Through time series progressive mapping and unified resource vector conversion, the key technical problems of heterogeneous data fusion in smart communities are effectively solved. The multi-scale sliding time window mechanism can capture the real-time state changes, daily activity patterns, and long-term operation rules of community devices, providing a full-time domain data basis for fine management of smart communities. The multi-dimensional encoding processing of the embedding unit eliminates the data heterogeneity between device identifiers, timestamps, location information, energy consumption indexes, and service types, realizes the semantic preservation conversion from discrete symbols to continuous vectors, and enables quantitative comparison and correlation analysis of different types of community resources in a unified mathematical space.

[0050] Further, the process of constructing a resource semantic graph for community services is as follows: the edge controller dynamically adds community semantic labels to the unified resource vector through an extensible label slot; based on cosine similarity and mutual information theory, the association rules between unified resource vectors are calculated to determine the connection strength between nodes; the connection relationship between nodes is established using a graph construction algorithm, and a sparse adjacency representation is used to record only the connected edges and edge index information.

[0051] Specifically, for any two resource vectors with attached semantic labels, the similarity calculation method is used to analyze their correlation degree in multiple dimensions such as function, location, state, etc. The similarity degree of two resources in the feature space is evaluated by cosine calculation, and the closer the numerical value is to 1, the higher the similarity is; then the information correlation between the two resources is analyzed, and the co-occurrence frequency and mutual influence degree of them in the historical data are calculated by statistical analysis method. The two calculation results are fused according to the preset weight ratio to obtain the comprehensive correlation score, which directly reflects the necessity and connection strength of establishing connection between the two resource nodes.

[0052] After obtaining the correlation score, the connection relationship between nodes is established by using a graph construction algorithm. All resource nodes are initialized as the vertices of the graph, all node pairs are traversed, and it is judged whether to establish an edge connection between the two nodes according to the correlation score calculated in the foregoing. When the correlation score exceeds the preset threshold value, an edge is created between the corresponding two nodes, and the weight of the edge is equal to the correlation score value. In order to avoid the graph structure being too dense, an adaptive threshold adjustment mechanism is used to dynamically adjust the connection threshold according to the statistical distribution of all correlation scores, so as to ensure the connectivity of the graph while avoiding excessive redundant connections.

[0053] After determining the graph structure, the topology of the graph is stored in the form of sparse adjacency representation. Sparse adjacency representation is a two-dimensional array, and the rows and columns correspond to the nodes in the graph respectively. Each element in the matrix represents whether there is a connection between the corresponding two nodes and the weight of the connection. Considering the characteristics of large number of resource nodes and relatively sparse actual connections in the smart community, only the non-zero elements and their position information are stored, which greatly reduces the storage space occupation.

[0054] By dynamically adjusting the adaptive threshold with the score distribution, the necessary connectivity is ensured while the redundant edges are suppressed, reducing the subsequent computational overhead; the compressed storage of sparse adjacency representation significantly reduces the memory and I / O pressure, facilitating large-scale deployment on the edge side; the incremental update mechanism only affects the affected local nodes and edges, avoiding full reconstruction, shortening the update link and maintaining the topology consistency; overall, an interpretable, scalable, low-overhead and online evolving resource semantic graph is formed, providing a stable and accurate graph data basis for subsequent multi-factor scoring and double-channel scheduling.

[0055] Further, with reference to Figure 2 , the expandable label slot includes: a location semantic label slot identifying the physical location and logical area to which the resource belongs; a function semantic label slot describing the service type and processing capacity of the resource; a state semantic label slot recording the real-time state and availability information of the resource; and a priority semantic label slot identifying the importance and scheduling priority of the resource in the community service.

[0056] Specifically, the location semantic label slot records the specific location information of the device, including the building number and room number, and labels the functional area belonging to the location, such as the security control area, the public service area, or the private residential area. The functional semantic label slot classifies the service type of the device, divides the device into security monitoring, environmental control, energy management, and communication network, and further subdivides the specific function sub-class. The state semantic label slot tracks the running state of the device in real time, records the key information including the device online condition, the workload degree, the device health condition, and whether there is a fault. The priority semantic label slot grades the resources according to the importance of community services, and the emergency resources such as fire alarm and access control system are marked as the highest priority, and the general environmental monitoring device is marked as a lower priority.

[0057] The four types of extensible labels of location, function, state, and priority are used to abstract heterogeneous devices into a unified description, reduce matching ambiguity caused by field differences, facilitate mapping, retrieval, and cross-module reuse; dual labeling of location and logical area makes the topology and business boundary visible at the same time, supporting quick filtering and local update by building / floor / area; function-subfunction hierarchical coding strengthens the same aggregation and different classification; state and availability are real-time mapped, and offline or fault nodes can be directly filtered in the scheduling stage; priority is embedded in the resource side, and no additional weight inference is needed for evaluation and scheduling; slot extension maintains backward compatibility when adding new labels, reducing maintenance cost and modification range.

[0058] Further, with reference to Figure 3 , the specific implementation process of constructing a multi-factor weighted scoring engine based on the resource semantic graph includes: extracting task-related node features from the resource semantic graph, including clustering coefficient and shortest path length; using the clustering coefficient and the shortest path length to establish a multi-factor scoring matrix, including resource availability factor, service matching degree factor, and geographic location factor; using the analytic hierarchy process to determine the weight coefficients of each factor, and calculating the comprehensive score by weighted summation; according to the scoring results, the tasks are prioritized, and the task data packet with priority label is generated.

[0059] Specifically, the clustering coefficient is used to measure the connection tightness of the node in the local network, for example, the interconnection density of the fire control center node and its adjacent smoke detector, sprinkler equipment, evacuation indicator light, etc. The higher the value, the stronger the coordination ability of the node in emergency response. The shortest path length reflects the cooperation convenience between nodes. When a resident requests maintenance service, the shortest connection path between the maintenance tool library and the faulty device is calculated, and the shorter the path, the higher the response efficiency of resource scheduling.

[0060] In establishing the multi-factor score matrix, the clustering coefficient is an important part of the resource availability factor, reflecting the cooperative response ability of the device; the shortest path length directly affects the calculation of the geographical location factor, and the shorter the path, the more obvious the geographical advantage; the service matching degree factor is formed by combining the functional attributes of the device and the task demand fit degree. The resource availability factor comprehensively considers the current load condition, health status, online state and cooperative ability of the device, and evaluates the running load, maintenance state, fault record and linkage with related devices of the elevator system in the smart community; the service matching degree factor evaluates the fit degree of resource capacity and task demand, and when residents need to adjust the indoor temperature, the air conditioning equipment with corresponding refrigeration and heating capacity is matched; the geographical location factor is evaluated based on the shortest path length combined with physical distance and network delay, and the device resource closest to the task initiation point and with the shortest connection path is preferentially selected.

[0061] According to the service characteristics of the smart community, the judgment matrix is established to determine the weight coefficient, for the emergency safety task, the weight of the resource availability factor is the highest, to ensure that the device can respond immediately at the critical moment; for the daily convenience service, the weight of the service matching degree factor is larger, to ensure the accuracy of service quality; for the task with higher real-time requirement, the weight of the geographical location factor is increased. The relative importance between factors is determined through expert experience and historical data analysis, and the standardized weight coefficient vector is calculated.

[0062] In the comprehensive score calculation process, the scores of the three factors are weighted and summed according to the determined weight coefficient. For example, when the residents report the elevator failure, the availability of the standby elevator is 0.9, the service matching degree is 0.8, and the geographical location advantage is 0.95, the comprehensive score is calculated combined with the weight coefficient, which is used as the decision basis for task scheduling.

[0063] According to the score result, the tasks are prioritized, the task with the highest score is marked as the highest priority, and the priority decreases in turn. The task data packet with priority label is generated, including task identification, priority label, score value, time stamp, task demand description and recommended resource list and other information, high priority task is processed first, and tasks with the same priority are arranged according to the time stamp in chronological order, to ensure the fairness and orderliness of task processing.

[0064] In the present application, the clustering coefficient and the shortest path length and other graph features are extracted from the resource semantic graph, the clustering coefficient is used to reflect the cooperative response ability of the device, the shortest path length is used to reflect the geographical advantage, and the multi-factor score matrix is formed by combining the service matching degree, the weight coefficients of each factor are scientifically determined by the analytic hierarchy process, so that the task priority calculation has quantitative basis and interpretability; the comprehensive score is calculated by weighted summation, to ensure the objectivity and consistency of the score result, to provide a scientific and reasonable priority quantification method for resource scheduling in the smart community, and to significantly improve the orderliness of task processing and the accuracy of resource allocation.

[0065] Further, the dual-channel hybrid scheduling network comprises: a first channel adopts a graph neural network to take a resource semantic graph as input, learns node embedding representation through multi-layer graph convolution operation, and outputs resource availability and node correlation degree of each node; a second channel adopts a long short-term memory network to take a task data packet as input, generates time sequence features and load prediction parameters; and based on the load prediction parameters, a load allocation scheme is output by a scheduling optimizer.

[0066] Specifically, the graph neural network of the first channel receives the previously constructed resource semantic graph as input data. The network first initializes the encoding of each node in the graph, converting the semantic label information of the node into a numerical vector representation. Then, through multi-layer graph convolution operation, the deep features of the nodes are gradually learned, and each layer of convolution aggregates the information of the node and its neighbor nodes, so that the node representation can integrate the local and global context information in the graph. For example, the intelligent access control node integrates the state information of the monitoring camera, elevator controller and floor lighting system connected thereto during the convolution process, forming a comprehensive node embedding representation. After multi-layer processing, the network outputs two types of key information: resource availability of each node, reflecting the current service capability and load status of each device; and node correlation degree, quantifying the dependency relationship between any two devices.

[0067] The long short-term memory network of the second channel takes a task data packet as input to process structured data containing task identification, priority label, timestamp, demand description and other information. The network encodes the sequence of task data packets within a historical time window, and learns the operation rules and resident behavior patterns of the smart community. For example, the system can identify the pattern of increased demand for elevator use in the morning of weekdays, frequent access to underground garages, the rule of centralized activation of access control systems, elevators and lighting devices after residents go home in the afternoon, and the feature of increased use of fitness facilities and entertainment area devices during weekends. Through the continuous updating of the hidden layer state, the network remembers the long-term dependency relationship of community device use in different seasons and during holidays, such as higher load of air conditioning systems in summer, longer operation time of heating devices in winter, and increased demand for security monitoring during the Spring Festival. The gating mechanism helps the network selectively retain key information when processing complex time series data of the smart community. The forget gate filters out occasional abnormal task requests, such as device false alarms or test operations; the input gate focuses on representative service patterns, such as residents' daily travel, regular maintenance of devices, and emergency response; and the output gate ensures that the prediction result focuses on time sequence features that are most valuable for resource scheduling.

[0068] Through the dual-channel design, both the spatial dimension resource analysis using the graph structure information and the time dimension demand prediction based on the time sequence pattern are realized. The first channel deeply mines the correlation between devices through the graph neural network, which can identify the implicit collaboration mode that is difficult to be found by traditional methods, and improves the resource matching accuracy. The second channel captures the time regularity of community services through the long short-term memory network, accurately predicts the resource demand in peak period, avoids the blindness of resource allocation, and realizes the efficient utilization of community resources and the continuous stability of service quality.

[0069] Further, the process of obtaining task-node correlation and generating optimal matching strategy by fusing resource availability, node correlation degree, time sequence feature and load prediction parameter is as follows: a task-node correlation calculation matrix is constructed, and the scores in four dimensions of resource availability, node correlation degree, time sequence feature and load prediction parameter are normalized; a multi-objective optimization algorithm is used to calculate the matching correlation score between tasks and nodes, and constraint conditions including resource capacity limit, service quality requirement and response time threshold are established; under the constraint conditions, a minimum cost flow algorithm is used to solve the optimal allocation scheme and generate the optimal matching strategy.

[0070] Specifically, a task-node correlation calculation matrix is constructed, and the rows of the matrix represent the tasks to be processed, and the columns represent the available resource nodes. Taking the elevator maintenance task as an example, four-dimensional score data of each elevator are extracted: the resource availability score reflects the current running state and fault degree of the elevator, the node correlation degree score reflects the connection tightness between the elevator and the maintenance tool library and the spare parts warehouse, the time sequence prediction score is based on historical maintenance records to estimate the time required for this maintenance, and the load prediction parameter score considers the current use of other elevators; the minimum-maximum standardization method is used to convert the original scores in four dimensions to the interval of 0-1, and the calculation formula is that the standardized value is equal to the original value minus the minimum value and then divided by the difference between the maximum value and the minimum value, so that different dimension indicators can participate in subsequent calculation fairly.

[0071] The multi-objective optimization algorithm comprehensively considers the three core targets of service efficiency, resource utilization rate and user satisfaction, adopts the Pareto frontier search strategy, generates multiple candidate solutions through mutation and crossover operations of genetic algorithm, and each solution represents a task allocation scheme; three types of key constraints are established for the smart community scene: the capacity constraint limits the number of tasks that a single elevator can undertake at the same time to be less than its processing capacity upper limit, the quality constraint requires that the monitoring task must be allocated to a camera device with a pixel density meeting the requirements, and the time constraint stipulates that the response delay of emergency tasks such as fire alarm cannot exceed the preset threshold. The algorithm integrates the constraint conditions into the objective function through the Lagrange multiplier method, forming a single cost representation that can be used for subsequent network flow modeling.

[0072] In the assignment solving stage, a directed network of "source-task-resource-sink" is constructed: tasks are modeled as nodes with single assignment demand, resources are modeled as nodes with capacity upper limit; candidate edges between tasks and resources are set with only single assignment allowed, whose cost is taken from the aforementioned single cost matrix; combinations not meeting quality or time window are disabled or given high cost. A minimum cost flow algorithm is used on this network to obtain the least cost feasible assignment. To form primary and backup schemes, after obtaining the baseline solution, exclusive constraints are imposed on the selected edges or a slight perturbation is imposed on the related cost, and the solution is repeated to obtain a set of suboptimal solutions sorted by cost, which constitutes the candidate resource set for each task for failure switching and load redistribution.

[0073] By the optimal matching strategy of fusion four-dimensional score generation method, high-precision task-resource pairing and globally optimized resource allocation are realized; the construction of task-node correlation matrix and normalization eliminates the dimensional difference between different dimension scores, so that resource availability, correlation degree and load prediction parameters can participate in the decision-making process fairly; the multi-objective optimization algorithm effectively balances service efficiency, resource utilization and user satisfaction, and the global optimality of the allocation scheme is ensured through the Pareto front search. On this basis, a network flow model is constructed and a minimum cost flow algorithm is used to solve it, and a feasible assignment scheme with smaller cost is obtained under the constraints of capacity, quality and time, and the primary and backup sets sorted by cost can be generated to support switching and redistribution.

[0074] The application breaks through the technical bottleneck of traditional centralized scheduling in data heterogeneity and real-time response by establishing a hierarchical rolling cache system based on community life rules, mapping and converting heterogeneous node data of edge, cloud and end into a unified resource vector through time sequence progression, realizing intelligent matching of tasks and resources by using a double-channel hybrid scheduling network composed of a graph neural network and a long short-term memory network, generating an optimal allocation strategy by combining a multi-factor weighted scoring engine and a multi-objective optimization algorithm. The scheme enables the smart community to maintain high-precision task-resource matching, efficient graph structure retrieval and rapid dynamic adjustment capability under the scenario of dynamic device state change and task load fluctuation, and realizes device collaborative response and automatic failure switching through deep fusion of semantic labels and association rules, which helps to improve the overall utilization efficiency, service response speed and system reliability of community resources, and provides a solid technical guarantee for the continuous and stable operation of the smart community.

[0075] Embodiment two:

[0076] To further verify the effectiveness of the application, a smart community metadata interaction system based on an edge computing framework is also proposed in the embodiments of the application as a specific implementation manner of the method of the application. Referring to Figure 4 , the structural diagram of the system of the application.

[0077] data caching module: obtain multi-dimensional data of smart community end, edge and cloud nodes, and store the data dynamically by using hierarchical and rolling caching based on community life rules;

[0078] semantic mapping module: map the multi-dimensional data to uniform resource vectors through time sequence progressive mapping; add community semantic labels to the uniform resource vectors by an extensible label slot in the edge controller, and establish connections according to association rules to build a resource semantic graph for community services;

[0079] Further, the process of building a resource semantic graph for community services is as follows: dynamically adding community semantic labels to the uniform resource vectors by an extensible label slot in the edge controller; calculating the association rules between the uniform resource vectors based on cosine similarity and mutual information theory to determine the connection strength between nodes; establishing the connection relationship between nodes by using a graph construction algorithm, and using sparse adjacency representation to record only the edge and edge index information that exist.

[0080] When establishing the connection relationship between nodes by using a graph construction algorithm, an incremental update mechanism is introduced:

[0081] When a new device node is added to the smart community, the change detector identifies the semantic label type of the new node, calculates the association degree score between the new node and the existing nodes, and only establishes an edge connection when the association degree exceeds a preset threshold; when the state of an existing node changes, the system recalculates the semantic vector of the node and its one-hop neighbor nodes, and updates the edge weight values involved; a node change buffer is established to collect all node change requests within a predetermined time window, and the corresponding rows and columns of the sparse adjacency representation are updated synchronously in a batch processing manner.

[0082] Through the incremental update mechanism, the calculation efficiency and resource consumption of the traditional graph construction algorithm are optimized. The traditional method needs to recalculate the sparse adjacency representation and node features of the entire graph structure when the node state changes, with a calculation complexity of O(n²), while the incremental update mechanism only needs to recalculate the local structure of the changed node and its neighbors, with a complexity reduced to O(k), where k is the number of affected nodes, which is usually much smaller than the total number of nodes n. This mechanism meets the scene characteristics of dynamic access of smart community devices, and when a new resident activates a smart home device, it can quickly access the new node without affecting the continuity of existing services. The batch processing strategy effectively deals with the peak period of concentrated online and offline of devices, and improves the stability of the smart community resource management system.

[0083] task evaluation module: set a resident task access unit and receive resident task requests, build a multi-factor weighted scoring engine based on the resource semantic graph, quantitatively evaluate the service priority, and generate a task data packet with a priority label;

[0084] Further, the specific implementation process of constructing a multi-factor weighted scoring engine based on the resource semantic graph includes: extracting task-related node features from the resource semantic graph, including clustering coefficient and shortest path length; using the clustering coefficient and the shortest path length to establish a multi-factor scoring matrix, including resource availability factor, service matching degree factor and geographical location factor; determining the weight coefficients of each factor by using the analytic hierarchy process, calculating the comprehensive score by weighted summation; prioritizing the tasks according to the scoring results, and generating task data packets with priority labels.

[0085] When determining the weight coefficients of each factor by using the analytic hierarchy process, a dynamic weight adjustment mechanism is added:

[0086] The three-factor weight distribution ratio is adjusted in real time. When handling emergency tasks, the resource availability factor weight is increased to ensure that online available devices are selected first. When solving routine convenience services, the service matching degree factor weight is increased accordingly to ensure service accuracy. According to different time periods in the community, the geographical location factor weight is adjusted, and the factor weight is increased during the traffic peak period to reduce the response distance. A weight adjustment rule library is established to store weight configuration schemes in different scenarios, match the current task characteristics and environmental state, and apply the corresponding weight combination.

[0087] By adding this dynamic weight adjustment mechanism, the performance of the traditional fixed weight configuration multi-factor scoring method is improved. The analytic hierarchy process in the past usually uses static weight distribution, which cannot adapt well to the dynamic changes of the smart community service scene, often leading to delays in the scheduling of critical resources due to unreasonable weight configuration in emergency situations. The dynamic weight adjustment mechanism can optimize weight distribution in real time according to the specific task type and environmental state, especially in emergency scenarios such as fire alarms and medical emergencies, by increasing the weight of the resource availability factor, effectively shortening the average response time. This mechanism also exhibits good scene adaptability. In routine convenience services, it will strengthen the service matching degree weight to ensure service quality, and in traffic peak periods, it will adjust the geographical location weight to reduce service distance, and during device maintenance period, it will balance the weight distribution to realize the rational use of resources.

[0088] Matching strategy module: construct a dual-channel hybrid scheduling network, the first channel performs graph neural network inference based on the resource semantic graph, outputs resource availability and node association degree, the second channel outputs time sequence features and load prediction parameters based on the task data packet; fuse resource availability, node association degree, time sequence features and load prediction parameters to obtain task-node correlation and generate the optimal matching strategy.

[0089] The double-channel hybrid scheduling network comprises: a first channel adopts a graph neural network, takes a resource semantic graph as input, learns node embedding representation through multi-layer graph convolution operation, and outputs resource availability and node correlation degree of each node; and a second channel adopts a long short-term memory network, takes a task data packet as input, performs time series modeling through hidden layer state update and gating mechanism, and outputs time series features and load prediction parameters.

[0090] In the first channel graph neural network, a layer number adjustment mechanism is introduced.

[0091] The optimal number of layers of graph convolution is dynamically determined according to the node size and connection density of the resource semantic graph. When the number of graph nodes is small and the connection is sparse, the number of convolution layers is automatically reduced; when the number of graph nodes is large and the connection is dense, the number of convolution layers is increased to capture deep association information of multi-hop neighbors; an evaluation index of the number of layers is established, the complexity of the graph structure is quantified and mapped to the most suitable number of convolution layers by calculating the diameter, clustering coefficient and node degree distribution characteristics of the graph; a layer number adjustment trigger condition is set, when the graph structure changes significantly or the quality of node embedding decreases, the number of convolution layers is re-evaluated and adjusted.

[0092] Through the layer number adjustment mechanism, the performance of the traditional fixed layer number graph neural network on different scale resource semantic graphs is optimized. The traditional graph neural network adopts a fixed three-layer or four-layer convolution structure, which cannot adapt to the differences in graph structure from small regional device groups to large-scale community networks in smart communities, resulting in overfitting of small-scale graphs and underfitting of large-scale graphs. The layer number adjustment mechanism adaptively selects the optimal number of layers according to the graph features, improving the quality of node embedding representation. This mechanism exhibits scene adaptability under different community scales, reduces to two layers of convolution to avoid information over-smoothing and maintain the discriminability of node features in a small community containing 100 nodes; increases to five layers of convolution to fully capture long-range dependencies and mine deep collaboration patterns between devices in a large community containing 1000 nodes.

[0093] By constructing the metadata interaction system of smart community based on edge computing framework, the whole process management from data collection to resource scheduling is realized, which significantly improves the overall service level and operation efficiency of smart community. The data cache module adapts to the community life by hierarchical and rolling cache, greatly reducing the storage cost and access delay; the semantic mapping module eliminates the data difference of end, edge and cloud, and constructs the semantic graph which accurately reflects the resource association; the task evaluation module realizes the scientific quantification of task priority based on graph features and multi-factor scoring, ensuring the priority response of emergency events; the matching strategy module realizes the forward-looking resource allocation decision by fusing spatial correlation and time series prediction through double-channel network. The four modules work together to systematically solve the problems of data island, response lag and resource waste in traditional community management. The system has good scalability and fault tolerance, and can adapt to the demand of community scale expansion and equipment type increase, providing strong technical support for the sustainable development of smart community.

[0094] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A smart community metadata interaction method based on an edge computing framework, characterized in that, The method comprises the following steps: acquiring multi-dimensional data of smart community end, edge and cloud nodes, and dynamically storing the multi-dimensional data by using hierarchical rolling cache based on community life rules; mapping and converting the multi-dimensional data into a unified resource vector through time sequence progressive mapping; adding community semantic labels to the unified resource vector by an extensible label slot in an edge controller, and establishing a connection according to an association rule to construct a resource semantic graph for community services; setting a resident task access unit and receiving a resident task request, constructing a multi-factor weighted scoring engine based on the resource semantic graph, quantitatively evaluating service priorities, and generating a task data packet with priority labels; constructing a double-channel hybrid scheduling network, wherein a first channel is based on the resource semantic graph to perform graph neural network inference, and outputs resource availability and node correlation degree, and a second channel is based on the task data packet to output time sequence features and load prediction parameters; the double-channel hybrid scheduling network comprises: the first channel adopts a graph neural network, takes the resource semantic graph as input, learns node embedding representation through multi-layer graph convolution operation, and outputs resource availability and node correlation degree of each node; the second channel adopts a long short-term memory network, takes the task data packet as input, generates time sequence features and load prediction parameters; and based on the load prediction parameters, a load distribution scheme is output by a scheduling optimizer; fusing resource availability, node correlation degree, time sequence features and load prediction parameters to obtain task-node correlation and generate an optimal matching strategy. 2.The smart community metadata interaction method based on the edge computing framework according to claim 1, characterized in that, The process of acquiring multi-dimensional data of smart community end, edge and cloud nodes is: collecting multi-dimensional data of smart community terminal device state, edge gateway load and cloud server resource, which includes device identification, timestamp, location information, energy consumption index and service type. 3.The smart community metadata interaction method based on the edge computing framework according to claim 1, characterized in that, The process of dynamically storing the multi-dimensional data by using hierarchical rolling cache based on community life rules is: establishing a hierarchical rolling cache architecture according to community resident work and rest rules and device use cycle, which comprises: a hot data cache layer storing recent high-frequency access data and maintaining a short period time window; a warm data cache layer storing medium-frequency access data and maintaining a medium period time window, retaining the last valid state; and a cold data cache layer storing low-frequency access and archiving data and maintaining a long period time window, with simplified index and source identification. 4.The smart community metadata interaction method based on the edge computing framework according to claim 1, characterized in that, The process of mapping and converting the multi-dimensional data into a unified resource vector through time sequence progressive mapping is: performing time sequence progressive mapping on the multi-dimensional data, extracting data feature sequences through a sliding time window, converting discrete features into continuous vector representation by using embedding units; normalizing different dimensional data features to the same numerical range, and merging device identification, timestamp, location information, energy consumption index and service type into a fixed-dimensional unified resource vector through a feature fusion algorithm.

5. The smart community metadata interaction method based on the edge computing framework according to claim 1, characterized in that, The process of constructing the resource semantic graph for community services is: the edge controller dynamically adds community semantic tags to the uniform resource vector through the extensible tag slot; the connection strength between nodes is determined based on the cosine similarity and the mutual information theory; the connection relationship between nodes is established by using a graph construction algorithm, and a sparse adjacency representation is adopted to record only the connected edges and edge index information. 6.The smart community metadata interaction method based on the edge computing framework according to claim 5, characterized in that, The extensible tag slot includes: a location semantic tag slot, which identifies the physical location and logical area to which the resource belongs; a function semantic tag slot, which describes the service type and processing capacity of the resource; a state semantic tag slot, which records the real-time state and availability information of the resource; and a priority semantic tag slot, which identifies the importance and scheduling priority of the resource in community services. 7.The smart community metadata interaction method based on the edge computing framework according to claim 1, characterized in that, The specific implementation process of constructing a multi-factor weighted scoring engine based on the resource semantic graph includes: extracting task-related node features from the resource semantic graph, including clustering coefficients and shortest path lengths; establishing a multi-factor scoring matrix using the clustering coefficients and shortest path lengths, including resource availability factors, service matching degree factors, and geographic location factors; determining the weight coefficients of each factor by using the analytic hierarchy process, and calculating the comprehensive score by weighted summation; prioritizing the tasks according to the scoring results, and generating task data packets with priority labels. 8.The smart community metadata interaction method based on the edge computing framework of claim 1, wherein, The process of obtaining task-node correlation and generating an optimal matching strategy by fusing resource availability, node correlation, timing characteristics, and load prediction parameters is: constructing a task-node correlation calculation matrix, normalizing the scores of the four dimensions of resource availability, node correlation, timing characteristics, and load prediction parameters; calculating the matching correlation score between tasks and nodes using a multi-objective optimization algorithm, and establishing constraint conditions including resource capacity limits, service quality requirements, and response time thresholds; under the constraint conditions, the minimum cost flow algorithm is used to solve the optimal allocation scheme and generate the optimal matching strategy. 9.A smart community metadata interaction system based on an edge computing framework, characterized in that, It includes: a data caching module that acquires multi-dimensional data of smart community edge, node, and cloud, and dynamically stores the data using hierarchical rolling caching based on community life rules; a semantic mapping module that maps the multi-dimensional data to uniform resource vectors through time sequence progressive mapping; the edge controller adds community semantic tags to the uniform resource vectors through the extensible tag slot, and establishes connections according to the association rules to construct a resource semantic graph for community services; a task evaluation module that sets up a resident task access unit and receives resident task requests, constructs a multi-factor weighted scoring engine based on the resource semantic graph, quantitatively evaluates service priorities, and generates task data packets with priority labels; a matching strategy module that constructs a dual-channel hybrid scheduling network, the first channel performs graph neural network inference based on the resource semantic graph to output resource availability and node correlation, and the second channel outputs timing characteristics and load prediction parameters based on the task data packets; The double-channel hybrid scheduling network comprises: a first channel adopting a graph neural network, taking a resource semantic graph as input, learning node embedding representation through multi-layer graph convolution operation, and outputting resource availability and node correlation degree of each node; a second channel adopting a long short-term memory network, taking a task data packet as input, generating timing characteristics and load prediction parameters; based on the load prediction parameters, a load distribution scheme is output by a scheduling optimizer; and task-node correlation is obtained by fusing resource availability, node correlation degree, timing characteristics and load prediction parameters, and an optimal matching strategy is generated.

Images