A method for supervising cloud edge collaborative adaptation of multi-heterogeneous systems

Abstract

The present application relates to the field of project supervision information technology, and provides a supervision multi-heterogeneous system cloud edge collaborative adaptation method, which comprises collecting edge device hardware architecture, operating system and other heterogeneous information to generate records; integrating supervision monitoring data and cloud instructions, classifying data according to functions and update frequencies and marking high-frequency units; unifying data formats and adding hash check information; constructing a three-dimensional mapping table in the cloud, combining high-frequency data grouping edge nodes; converting edge device private protocols into general protocols; using an integer linear programming model to allocate tasks with the goal of minimizing transmission volume and energy consumption; dynamically adjusting device energy consumption; and dynamically optimizing configuration based on interface indicators. The method can improve cloud edge interaction efficiency, reduce energy consumption and ensure reliable operation of supervision business.

Description

Technical Field

[0001] This invention relates to the field of information technology for project supervision, specifically a cloud-edge collaborative adaptation method for multiple heterogeneous supervision systems. Background Technology

[0002] In cross-regional project supervision, in order to achieve real-time monitoring and overall management of project sites in various locations, it is necessary to conduct data interaction, status monitoring and instruction issuance through a central cloud platform and distributed electronic devices. This has resulted in a multi-heterogeneous supervision system. Such systems consist of electronic devices and operating systems of various brands and versions at project sites in different locations. The devices cover different hardware architectures and need to work together to support the development of supervision business.

[0003] To address the real-time and global control requirements of cross-regional supervision, cloud-edge collaboration technology has become a key support. Its core logic is to divide and coordinate data processing and task execution between the edge and the cloud: edge devices are close to the data source and are responsible for real-time collection of supervision data and processing of low-latency business needs, reducing data transmission redundancy; the cloud is responsible for global resource scheduling, massive data storage, complex task planning, etc. Through information interaction with the edge, it achieves the control goals of near-source processing and global optimization, which is an important technical architecture for the efficient operation of multi-heterogeneous supervision systems.

[0004] However, existing technologies have failed to establish a collaborative mechanism for heterogeneous data adaptation that ensures standardized formats, compatible protocols, and on-demand transmission. This leads to format conflicts and protocol disconnects during data interaction between different systems, and the full data transmission results in bandwidth waste across regions. In terms of cloud-edge collaborative optimization, the constraints of edge resources are not linked to global goals in the cloud. Some edge devices, due to limited computing power, memory, and network transmission capabilities, cannot adapt to the complex scheduling logic of the cloud, resulting in task execution delays or excessive energy consumption, making it difficult to balance the needs of supervision services with cloud-edge collaborative efficiency.

[0005] In summary, this invention provides a method for supervising cloud-edge collaborative adaptation of heterogeneous systems to solve the above problems. Summary of the Invention

[0006] This invention provides a method for cloud-edge collaborative adaptation of multiple heterogeneous systems. By collecting heterogeneous information of edge devices, unifying data format and verifying integrity, constructing a three-dimensional mapping table in the cloud to group edge nodes, converting private protocols, allocating tasks based on an integer linear programming model, controlling edge energy consumption and dynamically adjusting configuration, this method solves the problems of heterogeneous data interaction conflicts, bandwidth waste, and latency and high energy consumption caused by mismatch between edge resources and cloud scheduling in the prior art.

[0007] The specific technical solution of this invention is as follows:

[0008] A method for cloud-edge collaborative adaptation of heterogeneous systems in supervision includes the following steps:

[0009] S1. Collect heterogeneous device information at the edge device and generate device information records;

[0010] In this invention, edge devices refer to hardware in the supervision scenario that is close to the data generation source, undertakes data collection and preliminary processing functions, and participates in cloud-edge collaboration. Device heterogeneous information refers to the core differentiated information between different edge devices that affects cloud-edge collaboration adaptation. Device information records are a unified carrier formed by structurally integrating the heterogeneous information of a single edge device. Specifically, starting from the global collaboration needs of multiple heterogeneous supervision systems, the scope of edge devices that need to be included in cloud-edge collaboration can be clearly defined to ensure coverage of all hardware involved in supervision data collection and processing to avoid subsequent adaptation gaps. Then, based on the hardware characteristics and functional positioning of the edge devices, a differentiated information collection mechanism is deployed. For devices whose core function is data collection, their hardware operating status, communication interaction rules, and data output format are read. For devices with computing capabilities, their hardware architecture characteristics and software operating environment information are obtained through system-level detection mechanisms. All collected information is transmitted to the edge side for centralized processing and integrated into device information records corresponding to a single device according to a unified structural specification. Finally, the integrity of the generated device information records is verified. If a certain type of heterogeneous information is missing, a targeted supplementary collection mechanism is triggered to complete the information by rereading the device information.

[0011] S2. Based on the equipment information records, integrate the real-time monitoring data of the supervision business with the configuration instructions issued by the cloud to form heterogeneous data. According to the data function and update frequency, divide the heterogeneous data into real-time interactive data, periodic update data and static configuration data units and mark them. Simultaneously record the access frequency of each data unit and mark the high-frequency access data units.

[0012] In this invention, real-time monitoring data for supervision business refers to data collected by edge devices that directly reflects the status of the supervised object; configuration instructions issued by the cloud refer to control instructions issued by the central cloud platform to edge devices or supervision business; heterogeneous data refers to a multi-source, multi-format data set formed by integrating the two types of data based on the aforementioned device information records; and the three types of data units are structured units divided according to data function and update frequency, including real-time interactive data units that focus on real-time control or emergency monitoring functions and have an update frequency of milliseconds, periodic update data units that focus on routine status monitoring functions and have an update frequency of minutes, static configuration data units that focus on basic parameter configuration functions and have a long-term static update frequency, and high-frequency access data units refer to data units whose access frequency exceeds a preset threshold. Specifically, based on the aforementioned equipment information records, the data format specifications and equipment association information are extracted to unify the formats of two types of data. At the same time, the edge device affiliation corresponding to each part of the data is clarified to avoid data failure due to format incompatibility or ambiguous affiliation, and the data is integrated to form heterogeneous data. Then, the heterogeneous data is classified according to data function and update frequency, and a unique identifier is assigned to each type of data unit to ensure data location and traceability. Finally, an access frequency statistics mechanism is deployed to record the number of times each data unit is called or read by edge devices or the cloud in real time. According to the needs of supervision business, an access frequency threshold is set, and data units that exceed the threshold are marked as high-frequency access data units.

[0013] S3. Convert heterogeneous data into a unified format and add integrity verification information to the converted data through hash verification calculation;

[0014] In this invention, "unified format" refers to a standardized data format adapted for cloud-edge collaborative interaction; "hash verification" refers to a verification method that calculates data feature values ​​using a hash algorithm; and "integrity verification information" refers to a unique data feature value generated through hash verification. Specifically, the aforementioned integrated heterogeneous data can be converted into a unified format, completely eliminating format differences between data from different sources to ensure smooth cloud-edge data interaction. Then, a hash verification algorithm is used to calculate the integrity verification information for each piece of converted data, which is then appended to the converted data, forming a binding relationship between the data and the verification information. Subsequently, whether for data transmission or storage, the verification information can be used to confirm whether the data is complete and has not been tampered with.

[0015] S4. In the central cloud platform, a page-level data association logic with shared virtual memory is adopted. Combined with device information records and high-frequency access data units, a three-dimensional mapping table of device identifier, data unit identifier, and virtual address is constructed. According to the distribution characteristics of high-frequency access data units, edge nodes are grouped. Then, the mapping rule fragments that match each edge node group are selected from the three-dimensional mapping table, and the mapping rule fragments are pushed to the corresponding edge nodes respectively.

[0016] In this invention, the page-level data association logic of shared virtual memory refers to a technical solution that associates cloud virtual memory and edge device memory at the page level. The three-dimensional mapping table refers to a structured table recording the correspondence between device identifiers, data unit identifiers, and cloud virtual addresses. Edge node grouping refers to the operation of classifying edge nodes according to the distribution characteristics of frequently accessed data units. Mapping rule fragments refer to address mapping sub-rules selected from the three-dimensional mapping table that match specific edge node groups. Specifically, the page-level data association logic of shared virtual memory can be used on the central cloud platform, combining the aforementioned device information records and frequently accessed data units to construct a three-dimensional mapping table. Then, edge nodes are grouped according to the distribution characteristics of frequently accessed data units, ensuring that edge nodes corresponding to high-frequency data are grouped together to reduce cross-group data access latency. Finally, mapping rule fragments matching each edge node group are selected from the three-dimensional mapping table, and these fragments are pushed to the corresponding edge nodes, enabling the physical addresses of edge devices to be matched with cloud virtual addresses through the mapping rule fragments.

[0017] S5. Based on the heterogeneous information of the devices, identify the private protocol field structure of the edge devices, and convert the private protocol into a general communication protocol through field parsing and format conversion;

[0018] In this invention, the private protocol field structure refers to the data frame structure of the edge device's proprietary communication protocol, including core fields such as frame header, data segment, and checksum. The general communication protocol refers to the standardized communication protocol widely used in cloud-edge collaboration scenarios. Specifically, based on the aforementioned collected heterogeneous device information, communication protocol type-related information can be extracted to deeply identify the private protocol field structure of the edge device and clarify the function, length, and encoding rules of each field. The core data in the private protocol is extracted through field parsing technology, and the data is re-encoded and format-converted according to the format requirements of the general communication protocol, thus converting the private protocol into a general communication protocol.

[0019] S6. According to the supervision business process, sort out the task relationship, construct the task directed acyclic graph, mark the task dependency relationship and hard deadline, and determine the task execution order based on the access frequency of the task-related data unit.

[0020] In this invention, the directed acyclic graph (DAG) refers to a graphical model that reflects the relationships between various tasks in the supervision process. Because it eliminates circular dependencies, it ensures smooth task scheduling logic. Task dependencies refer to the sequential constraints of task execution, and hard deadlines refer to the latest time a task must be completed. Specifically, the relationships between tasks can be analyzed one by one according to the actual workflow of the supervision process, eliminating potential circular dependencies and constructing a DAG. The dependencies and hard deadlines of each task are marked on the graph. Then, considering the access frequency of the associated data units, tasks with frequently accessed data units are prioritized for scheduling, while ensuring that all tasks meet the hard deadline requirements. Finally, a task execution order that balances efficiency and timeliness is determined.

[0021] S7. With the optimization goal of minimizing data transmission volume and total energy consumption of edge devices, construct an integer linear programming model, incorporate task dependency constraints, hard deadline constraints and device resource constraints into the model, and simultaneously add data unit access frequency constraints. Based on the model calculation results, assign the data unit associated tasks to the edge nodes of the corresponding groups and create execution containers.

[0022] In this invention, the integer linear programming model refers to a mathematical model that incorporates multiple constraints with the optimization objective of minimizing data transmission volume and total energy consumption of edge devices. These constraints include task dependency constraints based on task dependencies, hard deadline constraints based on the latest task completion time, device resource constraints based on edge node computing power and memory, and access frequency constraints based on data access frequency. The execution container refers to a lightweight environment that carries task execution, ensuring the consistency and isolation of the task execution environment. Specifically, an integer linear programming model can be constructed and various constraints incorporated into the model. The optimal task allocation scheme can be calculated through the model. Based on the calculation results, data units are associated with tasks and allocated to the corresponding edge nodes after the aforementioned grouping, ensuring that tasks and node resources match and reducing cross-node data transmission. Then, an independent execution container is created for each task allocated to the edge node, providing a dedicated execution environment for the task.

[0023] S8. Based on the computing power requirements of the supervision task, dynamically adjust the voltage and frequency of the edge devices and shut down idle non-core circuits.

[0024] In this invention, dynamically adjusting voltage and frequency refers to adjusting the operating parameters of edge devices according to the real-time computing power requirements of the task, avoiding energy waste caused by excessive computing power. Idle non-core circuits refer to non-core functional circuits in the edge device that are idle when no task is being executed; shutting them down can reduce the static energy consumption of the device. Specifically, the voltage and frequency of the edge device can be dynamically adjusted according to the real-time computing power requirements of the aforementioned assigned supervision tasks. When the task's computing power requirement is high, the voltage and frequency are increased to ensure computing power supply; when the task's computing power requirement is low, the voltage and frequency are decreased to reduce energy consumption. At the same time, the idle state of the edge device is identified and the non-core circuits in the idle state are shut down.

[0025] S9. Set up indicator collection points at the data interface between edge devices and the cloud, and dynamically adjust the data unit division method, the page address range of the three-dimensional mapping table, and the edge node grouping rules based on the collected indicator data.

[0026] In this invention, the indicator collection point refers to the monitoring point set at the data interface between the edge device and the cloud, used to collect key data reflecting the cloud-edge collaboration effect. The indicator data refers to the core data reflecting the adaptation effect obtained through the collection point, such as data transmission latency and protocol adaptation success rate. Specifically, indicator collection points can be set at the data interface between the edge device and the cloud to collect key indicator data in real time. Based on the collected indicator data, the adaptation rationality of the current data unit division method, the page address range of the three-dimensional mapping table, and the edge node grouping rules is analyzed. If the access latency of a certain type of data unit is too high, the data unit division standard is optimized; if the address mapping error frequency increases, the page address range of the three-dimensional mapping table is adjusted; if the node load is unbalanced, the edge node group is re-divided. Through continuous collection, analysis, and adjustment, the system is ensured to maintain a high-efficiency cloud-edge collaboration state in the long term.

[0027] In a preferred embodiment, in step S1, the heterogeneous device information includes the hardware architecture, operating system type, communication protocol type, and data format specification of the edge device.

[0028] In this invention, the hardware architecture reflects the basic characteristics of the edge device's computing hardware, the operating system type reflects the device's software operating environment attributes, the communication protocol type relates to the rules and standards for the device's interaction with the outside world, and the data format specification determines the structural form of the device's output data.

[0029] In a preferred embodiment, in step S4, the computing power, memory, and network transmission capabilities of each edge node are recorded synchronously during grouping, and the edge nodes corresponding to the high-frequency access data units are grouped into the same group.

[0030] In this invention, computing power reflects the computing and processing capabilities of edge nodes, memory determines the temporary data storage capacity of nodes, and network transmission capability affects the data interaction speed between nodes and the cloud and other nodes.

[0031] In a preferred embodiment, in step S3, the marked high-frequency access data units are processed using a lossless compression algorithm, the compressed data is bound and stored with integrity verification information, and the high-frequency access data units are stored in the local cache of the edge node.

[0032] In this invention, the lossless compression algorithm can reduce the storage space occupied by high-frequency data without losing data information. Binding data with integrity verification information can ensure that the integrity of the compressed data can still be verified. Local caching can eliminate the need for high-frequency data to be frequently retrieved from the cloud.

[0033] In a preferred embodiment, in step S4, after the mapping rule fragment is pushed to the edge node, the node's reception status is confirmed through the cluster inner beat detection mechanism.

[0034] In this invention, the cluster heartbeat detection mechanism periodically sends heartbeat signals from edge nodes to designated detection nodes to report their own status. If a detection node does not receive a heartbeat signal from an edge node within a preset time, or if the heartbeat signal does not contain an identifier indicating successful reception of a mapping rule segment, then the node is determined to have failed to receive the signal.

[0035] In a preferred embodiment, in step S5, after converting the private protocol into a general communication protocol, the converted data fragments are extracted and reverse-parsed into the private protocol format according to the general communication protocol format for reverse verification.

[0036] In this invention, reverse parsing refers to restoring the converted general format data to private protocol format data according to the encoding logic of the general communication protocol, and then comparing the restored private protocol data with the original private protocol data for field comparison and semantic verification.

[0037] In a preferred embodiment, during step S7, when creating the execution container, a port detection mechanism and a process monitoring mechanism are configured for the container to monitor the container port response status and process running status in real time.

[0038] In this invention, the port probing mechanism periodically sends probing requests to designated service ports of the container; if the request times out or an error response is received, the port is determined to be abnormal. The process monitoring mechanism reads the running parameters of task processes within the container in real time; if a process disappears or its parameters exceed the normal range, the process is determined to be abnormal.

[0039] In a preferred embodiment, in step S8, the idle state is determined by real-time monitoring of the computing power utilization rate of the edge device, and the energy consumption data before and after the circuit is turned off is recorded, and the energy consumption data is fed back to the integer linear programming model.

[0040] In this invention, the computing power utilization rate is determined by the ratio of the actual computing power used by the edge device per unit time to the maximum available computing power. If the ratio is consistently lower than the preset low load threshold, the device is determined to be in an idle state. Energy consumption data is recorded by collecting data such as power consumption and total energy consumption before and after the circuit is turned off through the energy consumption monitoring module built into the device.

[0041] In a preferred technical solution, in step S9, a time-series data storage component is deployed to store the collected indicator data, and a monitoring interface for the supervision adaptation effect is constructed.

[0042] In this invention, the time-series data storage component can record the index data collected each time in chronological order, and supports data query and trend analysis by time range and index type; the supervision adaptation effect monitoring interface intuitively displays the changing trend of index data and the comparison of adaptation effect before and after strategy adjustment through charts.

[0043] In a preferred embodiment, during step S4, when constructing the three-dimensional mapping table of device identifier, data unit identifier, and virtual address, a continuous page address segment is allocated to the high-frequency access data unit, and the length of the continuous page address segment is determined based on the average data volume of the high-frequency access data unit.

[0044] In this invention, a contiguous page address segment refers to a segment of memory pages with contiguous addresses in the cloud virtual memory, which is specifically used to map high-frequency access data units. The length of the page address segment is determined by calculating the average data volume of historical high-frequency data and combining it with the standard size of virtual memory pages, ensuring that high-frequency data can be fully accommodated without causing excessive memory waste.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. This invention uses a three-dimensional mapping table of virtual addresses to identify device identification data units. It combines the distribution characteristics of high-frequency access data units with the computing power, memory, network, and transmission capabilities of edge nodes to implement two-dimensional grouping. At the same time, it allocates continuous page address segments of length determined by the average data volume to high-frequency access data units. This overcomes the limitation of existing technologies where two-dimensional address mapping cannot associate data requirements and only groups data according to a single dimension of computing power. It not only achieves binding of device data addresses, but also reduces the transmission latency of high-frequency data across nodes through matching within the same group. Furthermore, it reduces the time consumption of memory fragmentation access by using continuous page addresses, reducing the latency of cloud-edge data interaction to the page level. This enables rapid response to real-time needs such as the issuance of control commands for emergency monitoring data transmission devices in supervision operations, ensuring the core operational efficiency of supervision operations.

[0047] 2. This invention constructs an integer linear programming model with the dual objectives of minimizing data transmission volume and total energy consumption of edge devices. The model simultaneously incorporates four types of constraints: task-dependent hard deadlines, device resources, and data access frequency. Furthermore, it feeds back energy consumption data before and after edge device circuit shutdown to the model in real time, dynamically optimizing subsequent task allocation strategies. This solves the problem in existing technologies where task scheduling only considers a single constraint, resulting in static and isolated energy control. It ensures that task allocation meets the hard deadline requirements of the supervision business while avoiding energy waste caused by redundant cross-node transmission and excessive computing power. In scenarios where edge devices in multi-heterogeneous supervision systems are dispersed and resources are limited, it can reduce total edge energy consumption while ensuring business execution, achieving a deep balance between resource utilization and energy consumption control for business needs. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the edge-side device information acquisition and data preprocessing process of the present invention.

[0049] Figure 2 This is a schematic diagram of the cloud mapping construction and edge protocol conversion process of the present invention.

[0050] Figure 3 This is a schematic diagram of the dynamic adjustment process of the task scheduling optimization and adaptation strategy of the present invention. Detailed Implementation

[0051] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0052] Example: Figure 1-3 As shown in this embodiment, an auto parts factory needs to carry out intelligent manufacturing supervision on three production lines in the stamping workshop. The edge devices deployed on site include 12 ARM architecture vibration sensors, 3 x86 architecture edge gateways, and 18 ARM architecture temperature and humidity sensors.

[0053] The staff first collected heterogeneous information from the edge devices. Using laptops (pre-installed with ModbusPoll and PuTTY), they traversed the production line. For vibration / temperature and humidity sensors, they connected to ModbusPoll via RS485 to read the hardware architecture (ARM Cortex-M4 for vibration sensors, ARM Cortex-A7 for temperature and humidity sensors), embedded operating system version (VxWorks 6.9 for both), communication protocol (Modbus-RTU), and data format (binary stream, including device ID, collected value, and timestamp fields). For the edge gateway, they logged in via SSH using PuTTY to view the hardware architecture (x86_64), operating system (Ubuntu Server 20.04), and data output format (JSON). All information was entered into an Excel spreadsheet according to the structure of device ID, hardware, OS, protocol, and data format to form device information records. A Python script was used to verify each record for missing fields. After discovering that the protocol information for two temperature and humidity sensors was missing, it was re-collected using ModbusPoll to complete the data.

[0054] Next, heterogeneous data is integrated and categorized. A Python script is deployed on the edge gateway to integrate vibration / temperature and humidity data collected by sensors (real-time monitoring data from the supervisor) with sensor sampling interval adjustment commands (configuration commands) issued by the cloud platform, according to the data format specifications in the device information records. This is then uniformly converted into JSON data in the format {"device_id":"","data_type":"","value":"","timestamp":""}. Data is categorized by function and update frequency: vibration data (updated every 10 seconds) is classified as real-time interactive data, temperature and humidity data (updated every 5 minutes) as periodically updated data, and gateway IP configuration (unchanging over a long period) as static configuration data. A unique identifier (e.g., REAL-VIB-001, PERI-TEMP-001) is assigned to each data unit. Prometheus is installed on the edge gateway and configured to count the number of data unit accesses every minute. A daily access frequency exceeding 1000 times is set as a high-frequency threshold, and vibration data is marked as a high-frequency access data unit.

[0055] Subsequently, the data format was standardized and integrity was verified. A Java utility class was run on the edge gateway to convert all raw binary streams from the sensors and JSON data from the gateway into the aforementioned standard JSON format. The hash value of each converted data entry was calculated using the SHA-256 algorithm, with the following formula:

[0056]

[0057] In the formula D JSON For standard JSON data, H is the hash value, and H is written as the hash field in the JSON for storage. For high-frequency vibration data, LZ4 compression is used (compression ratio approximately 3:1), and the hash field is retained after compression. The compressed file is then stored in the gateway's local directory / data / high-freq / .

[0058] Address mapping and node grouping were constructed on a factory-level cloud platform (CentOS 8 system). Shared virtual memory was enabled, and cloud and edge memory were linked at the page level. Combining device information records and high-frequency vibration data, a three-dimensional mapping table of device ID, data unit identifier, and virtual address was created in a MySQL database. After calculating the average size of the high-frequency data, four consecutive 4KB page address segments (virtual address range 0x7F001000-0x7F004FFF) were allocated specifically for mapping vibration data. Based on high-frequency data distribution and node resource grouping, a Python script was used to calculate the resource matching degree between edge nodes and groups, using the following formula:

[0059]

[0060] In the formula, N j For edge nodes, Gj To group nodes, CPU represents node computing power and Mem represents node memory. Devices on production lines with a Match value ≥ 0.8 are grouped into one group (3 groups in total). Mapping rule fragments corresponding to each group are selected from the mapping table and pushed to the edge gateways of each group using the Kubernetes kubectl command. A heartbeat packet is configured to be sent every 20 seconds; if the gateway does not respond after 3 attempts, the rule is re-pushed.

[0061] Then, a private protocol conversion is performed. A C language parsing script is written on the edge gateway to extract vibration / temperature and humidity values ​​from the data segments according to the previously collected Modbus-RTU protocol field structure (frame header 0x01, address code 0x02, data segment 0x03-0x06, checksum 0x07). These values ​​are then re-encoded according to the MQTT protocol format (the subject is set to sensor / data, and the payload is standard JSON). After the conversion, 100 MQTT data entries are randomly extracted and reverse-parsed into Modbus-RTU frames according to the MQTT format (e.g., {"vib":0.3} is reverse-parsed into 0x010x020x030x1E0x000x000x9A). These frames are then compared with the original Modbus frames to ensure correctness.

[0062] Subsequently, task scheduling and energy consumption optimization were carried out. First, Visio was used to analyze the relationships between supervision tasks, namely data acquisition, format conversion, hash verification, and cloud upload. After eliminating circular dependencies, a directed acyclic graph of tasks was constructed. Dependencies that could not be hashed if format conversion failed were marked, as well as the hard deadline of uploading vibration data within 15 seconds. Tasks related to vibration data were prioritized for scheduling. An integer linear programming model was constructed on the cloud platform using the Gurobi solver, with the objective function being:

[0063]

[0064] In the formula, D i E represents the amount of data transmitted. j The node energy consumption is represented by a weight of 0.5. The model incorporates dependency constraints such as "task A completion time ≤ task B start time", deadline constraints such as "total time ≤ 15 seconds", and resource constraints such as "CPU utilization ≤ 80%". The calculated task allocation scheme is as follows: vibration data processing tasks are assigned to the corresponding gateway group. An execution container is created on the gateway using Docker, configured to send TCP probe packets to port 8080 of the container every 15 seconds. If there is no response after 3 attempts, the container is automatically restarted. Simultaneously, the `top` command is used to monitor the Python process within the container in real time; if the process disappears, an alarm is triggered and the process is restarted.

[0065] Energy consumption was adjusted based on task computing power requirements. The cpufreq tool was installed on the edge gateway. During vibration data processing, the CPU frequency was increased from 2.0GHz to 2.8GHz, and decreased to 1.2GHz when there was no vibration data. The lm-sensors tool was used to monitor computing power utilization. If the utilization rate remained below 30% for 3 consecutive minutes, the gateway was considered idle, and the idle USB interface circuit was shut down via a shell script. The powerstat tool was used to record the energy consumption before and after the circuit was shut down (85W before shutdown, 52W after shutdown). The data was uploaded to the cloud platform and updated in the integer linear programming model to optimize task allocation for the next day.

[0066] Finally, the adaptation strategy was dynamically adjusted. Metric collection tools were deployed on the MQTT interfaces of the edge gateway and the cloud platform to collect data transmission latency and protocol adaptation success rate metrics. InfluxDB was used to store the metric data (retained for 30 days), and a monitoring interface was built based on Grafana to display metric changes in real time. When a vibration data transmission latency exceeding 80ms was detected, the data unit partitioning standard was modified (the vibration data update frequency was adjusted from 10 seconds to 8 seconds). If address mapping errors occurred more than once per hour, the 3D mapping table page address range was expanded to 8 consecutive pages. If the difference in node CPU utilization exceeded 30%, the resource matching degree was recalculated and node groups were redefined. Through these strategy adjustments twice a week, system collaboration efficiency was ensured.

[0067] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for cloud-edge collaborative adaptation of multi-heterogeneous systems in supervision, characterized in that, Includes the following steps: S1. Collect heterogeneous device information at the edge device and generate device information records; S2. Based on the equipment information records, integrate the real-time monitoring data of the supervision business with the configuration instructions issued by the cloud to form heterogeneous data. According to the data function and update frequency, divide the heterogeneous data into real-time interactive data, periodic update data and static configuration data units and mark them. Simultaneously record the access frequency of each data unit and mark the high-frequency access data units. S3. Convert heterogeneous data into a unified format and add integrity verification information to the converted data through hash verification calculation; S4. In the central cloud platform, a page-level data association logic with shared virtual memory is adopted. Combined with device information records and high-frequency access data units, a three-dimensional mapping table of device identifier, data unit identifier, and virtual address is constructed. According to the distribution characteristics of high-frequency access data units, edge nodes are grouped. Then, the mapping rule fragments that match each edge node group are selected from the three-dimensional mapping table, and the mapping rule fragments are pushed to the corresponding edge nodes respectively. S5. Based on the heterogeneous information of the devices, identify the private protocol field structure of the edge devices, and convert the private protocol into a general communication protocol through field parsing and format conversion; S6. According to the supervision business process, sort out the task relationship, construct the task directed acyclic graph, mark the task dependency relationship and hard deadline, and determine the task execution order based on the access frequency of the task-related data unit. S7. With the optimization goal of minimizing data transmission volume and total energy consumption of edge devices, construct an integer linear programming model, incorporate task dependency constraints, hard deadline constraints and device resource constraints into the model, and simultaneously add data unit access frequency constraints. Based on the model calculation results, assign the data unit associated tasks to the edge nodes of the corresponding groups and create execution containers. S8. Based on the computing power requirements of the supervision task, dynamically adjust the voltage and frequency of the edge devices and shut down idle non-core circuits. S9. Set up indicator collection points at the data interface between edge devices and the cloud, and dynamically adjust the data unit division method, the page address range of the three-dimensional mapping table, and the edge node grouping rules based on the collected indicator data.

2. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S1, the heterogeneous device information includes the edge device's hardware architecture, operating system type, communication protocol type, and data format specifications.

3. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S4, the computing power, memory and network transmission capabilities of each edge node are recorded synchronously during grouping, and the edge nodes corresponding to the high-frequency access data units are grouped into the same group.

4. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S3, the marked high-frequency access data units are processed using a lossless compression algorithm. The compressed data is bound and stored with the integrity verification information, and the high-frequency access data units are stored in the local cache of the edge node.

5. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S4, after the mapping rule fragment is pushed to the edge node, the node's reception status is confirmed through the cluster's internal heartbeat detection mechanism.

6. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S5, after converting the private protocol into a general communication protocol, the converted data fragments are extracted and reverse-parsed into the private protocol format according to the general communication protocol format for reverse verification.

7. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S7, when creating the execution container, a port detection mechanism and a process monitoring mechanism are configured for the container to monitor the container port response status and process running status in real time.

8. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S8, the idle state is determined by real-time monitoring of the computing power utilization rate of the edge device, and the energy consumption data before and after the circuit is turned off is recorded. The energy consumption data is then fed back to the integer linear programming model.

9. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S9, a time-series data storage component is deployed to store the collected indicator data, and a monitoring interface for the supervision adaptation effect is constructed.

10. The cloud-edge collaborative adaptation method for multi-heterogeneous systems in supervision according to claim 1, characterized in that: In step S4, when constructing the three-dimensional mapping table of device identifier, data unit identifier, and virtual address, continuous page address segments are allocated to high-frequency access data units. The length of the continuous page address segments is determined based on the average data volume of the high-frequency access data units.

Images