A control method of an internet of things gateway fusing communication and AI computing power

Abstract

The application is particularly a control method of an Internet of Things gateway fusing communication and AI computing power, and relates to the technical field of Internet of Things gateway control, comprising: constructing a machine-understandable structured function identity file and a dynamic digital gene model for a physical device; analyzing structured service requirements and fusing multi-source context information to generate a device and network joint execution strategy meeting QoS requirements through an AI decision model; and dynamically creating a virtual communication end point based on the execution strategy.In the application, the device structured capability declaration and the dynamic digital gene model realize the transformation of the device from passive response to active capability declaration; the device automatically pushes a standardized description file containing function fingerprints, capability vectors and the like after being powered on, so that the gateway can complete identification and adaptation without pre-configuration; in cooperation with the transparent adaptation characteristics of the virtual communication end point, a standard interface is simulated on the cloud and a local agent is simulated for the device, so that the device can be accessed without modifying the firmware.

Description

Technical Field

[0001] This invention relates to the field of IoT gateway control technology, and in particular to a control method for an IoT gateway that integrates communication and AI computing power. Background Technology

[0002] With the rapid development of IoT technology, the number of various terminal devices has surged, exhibiting diversified and heterogeneous characteristics, with significant differences in their communication protocols, data formats, and capability parameters. Existing IoT gateways generally suffer from the following shortcomings: Low device access efficiency: Traditional gateways need to pre-configure protocol drivers for different devices. When upgrading or replacing devices, firmware needs to be redeployed. They cannot achieve proactive declaration and dynamic adaptation of device capabilities. In particular, in scenarios where multiple vendors' devices coexist, compatibility issues are prominent. Resource scheduling is out of sync with business needs: Gateway control logic is mostly statically configured, focusing only on link connectivity without considering all factors such as device power consumption, environmental interference, and business priority for global optimization. This often leads to contradictions such as excessive resource consumption to meet business needs or sacrificing business quality to save resources. Insufficient data transmission reliability: In heterogeneous network environments, link status fluctuates dynamically, and existing gateways lack the ability to dynamically weave paths based on service QoS requirements. In emergency scenarios, data transmission latency is uncontrollable and packet loss rate is high. The integration of AI computing power and communication functions is low: the AI ​​capabilities of most gateways are limited to simple data filtering and are not deeply integrated into core processes such as policy generation, link scheduling, and lifecycle management.

[0003] While Narrowband Internet of Things (NB-IoT) has solved the problems of low power consumption and wide coverage, it still needs to be combined with AI computing power to improve its adaptive capabilities in scenarios such as multi-link collaboration and high-priority business assurance.

[0004] Therefore, a control method for an IoT gateway that integrates communication and AI computing power is proposed to address the aforementioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a control method for an IoT gateway that integrates communication and AI computing power in order to solve the above-mentioned problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A control method for an IoT gateway that integrates communication and AI computing power includes: To build machine-understandable structured functional identity profiles and dynamic digital genetic models for physical devices; The system analyzes structured business requirements and integrates multi-source contextual information, then uses an AI decision model to generate joint execution strategies for devices and networks that meet QoS requirements. Virtual communication endpoints are dynamically created based on execution strategies, and protocol stacks and data processing pipelines are assembled on demand to achieve transparent adaptation and communication between heterogeneous devices and the cloud. It coordinates and schedules heterogeneous links and dynamically weaves the optimal transmission path, ensuring stable data transmission in complex network environments through real-time adaptive adjustments.

[0007] Preferably, the step of constructing a machine-understandable structured functional identity profile and dynamic digital genetic model for physical devices specifically includes: After the device is powered on and initialized, it will start the built-in capability declaration module and broadcast or send the structured functional identity description file through the local communication interface. The file adopts a hierarchical structure, specifically including functional fingerprints, capability vectors, native interface contracts, resource constraints, and extended fields; After receiving the device description file through the multi-mode communication module, the AI ​​gateway processes it. First, invalid files are filtered out through syntax and semantic checks; second, core information is extracted to create a unique digital gene ID for the device; finally, a dynamic device digital gene model is constructed, which includes a three-dimensional structure of basic attribute layer, real-time status layer, and capability evolution layer. The basic attribute layer stores fixed information about the device; the real-time status layer periodically collects and updates the current status of the device through the gateway; and the capability evolution layer records changes in the device's capabilities. The model is ultimately stored in a distributed knowledge base on the gateway's local machine.

[0008] Preferably, the step of parsing structured service requirements and fusing multi-source contextual information to generate a joint device and network execution strategy that meets QoS requirements through an AI decision model specifically includes: The gateway's intent parsing engine performs layered parsing of requests: first, it extracts the core requirements through the NLP module to solve the problem of non-standardized expression; then, it matches QoS constraints through the domain rule engine; and finally, it breaks down the requests into executable sub-goals. After parsing, the engine outputs a list of intent decompositions, which serves as the direct basis for subsequent policy generation, and binds the intent ID to the gateway ID. Three types of contextual information are collected in real time, and redundant and conflicting data are processed through a weighted fusion algorithm to output a unified contextual panoramic view. Network context: Real-time status of heterogeneous links is collected through network diagnostic tools, with specific indicators including: 5G links, Wi-Fi links, and LoRa links; Device context: Obtain the target device status through device heartbeat packets and active queries, including: battery level, current load, mobility status, current operating mode, and historical fault records; Environmental context: obtained by fusing environmental data from the gateway’s built-in environmental sensors with environmental data reported by the device, including physical environment, spatial relationships, and business scenarios.

[0009] Preferably, the method further includes: The gateway's built-in AI decision-making model adopts a hybrid architecture of reinforcement learning and rule constraints. It optimizes the system with multiple objectives, including maximizing intent satisfaction, minimizing resource consumption, and optimizing system stability. It takes as input an intent breakdown list, a device digital gene model, and a contextual panoramic view, and outputs a structured joint communication and computing strategy. After the strategy is generated, its feasibility is verified. If there are conflicts, adjustments are made, and finally an executable strategy file is output.

[0010] Preferably, the step of dynamically creating virtual communication endpoints based on execution strategies and assembling protocol stacks and data processing pipelines on demand to achieve transparent adaptive communication between heterogeneous devices and the cloud specifically includes: After receiving the federated communication and computing policy, the gateway's virtual endpoint management module instantiates a virtual communication endpoint. This endpoint has a unique logical identifier and is bound to a dedicated resource quota. VCE exhibits transparent adaptation characteristics to both upstream and downstream processes: For the cloud, VCE emulation is an ideal device, providing a standard interface and reporting data every 5 seconds; For terminal devices, VCE emulation acts as a local agent, supporting native interaction with the Zigbee protocol and understanding the device's raw ADC data format and calibration rules.

[0011] Preferably, the method further includes: The gateway's protocol stack component library pre-stores various lightweight protocol driver modules. VCE assembles a dedicated protocol stack according to the protocol adaptation requirements in the policy to achieve bidirectional protocol conversion between the device side and the cloud side. Device-side protocol processing: Parse the configuration instructions issued by the gateway, encapsulate them into Zigbee data frames according to the device's native interface contract, and send them to the device through the Zigbee radio frequency module; The system receives Zigbee data frames uploaded by the receiving device, parses out the raw data, calls the data conversion engine, calculates the actual temperature and humidity based on the semantic rules in the device's digital DNA, and finally converts it into intermediate data in standard JSON format. The intermediate data in standard JSON format is converted into CSV format according to the policy requirements, compressed with gzip, and then packaged into an HTTP / 3 request before being sent to the cloud application interface. It receives control commands from the cloud, parses them, converts them into the standard command format within VCE, and then forwards them to the device-side protocol driver module.

[0012] Preferably, the process further includes processing the injection line: VCE internally builds a configurable data processing pipeline. Based on the data processing requirements in the strategy, it selects the corresponding modules from the processing module library and injects them into the pipeline in sequence.

[0013] Preferably, the collaborative scheduling of heterogeneous links and dynamic weaving of optimal transmission paths, ensuring stable data transmission in complex network environments through real-time adaptive adjustments, specifically includes: The gateway's link management module maintains a heterogeneous link status table in real time, recording the metrics of each link. VCE dynamically allocates transmission links based on data priority and QoS requirements, including: High-priority data: Prioritize allocation of 5G links; Medium priority data: Prioritize the allocation of Wi-Fi links; Low-priority data: The data is split and transmitted simultaneously via LoRa and idle Wi-Fi links using link aggregation. When the target device is outside the direct coverage area of ​​the current gateway, the gateway establishes communication with three gateways within a 1-kilometer radius through the edge gateway collaboration protocol to build a gateway collaboration network and realize the dynamic weaving of multiple paths.

[0014] Preferably, the method further includes: Verify the alignment of business intent with feedback data to optimize the AI ​​model, and combine the business lifecycle to achieve control and reclamation of virtual communication endpoints and resources.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention enables devices to transform from passively responding to actively declaring their capabilities through structured capability declarations and dynamic digital gene models. After the device is powered on, it automatically pushes a standardized description file containing functional fingerprints, capability vectors, etc., and the gateway can complete the identification and adaptation without pre-configuration. With the transparent adaptation characteristics of virtual communication endpoints, it simulates standard interfaces on the cloud and local proxies on the device, so the device can access the network without modifying its firmware.

[0016] 2. This invention integrates multiple source contexts through an AI decision model to achieve global coordination of business needs, equipment capabilities, and network resources; resource pooling and lifecycle management mechanisms improve resource utilization, ensuring stringent QoS requirements in scenarios such as industrial monitoring while avoiding resource waste, thus achieving a win-win situation for both business quality and resource efficiency. Attached Figure Description

[0017] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0019] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0020] Example 1

[0021] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0022] Appendix Figure 1 The flowchart of a control method for an IoT gateway that integrates communication and AI computing power, provided in an embodiment of the present invention, illustrates the complete steps from building a machine-understandable structured functional identity profile and dynamic digital gene model for physical devices to dynamically creating virtual communication endpoints based on execution strategies.

[0023] In this embodiment, it includes: To build machine-understandable structured functional identity profiles and dynamic digital gene models for physical devices, enabling devices to transform from passively responding to proactively declaring their capabilities; Specifically, it includes: After the device is powered on and initialized, it will start the built-in capability declaration module (this module is a lightweight firmware, occupying less than 512KB of memory, and is adapted to low-power MCUs), and broadcast or send the structured functional identity description file through local communication interfaces (such as the Class A uplink channel of LoRaWAN, the Beacon frame extension field of Wi-Fi). This document adopts a hierarchical structure of core elements and extended fields, based on common ontologies in the IoT field (such as the W3C's SSN / SOSA ontologies) to ensure cross-platform compatibility, specifically including: Functional fingerprint: A three-dimensional structure consisting of basic categories, industry tags, and custom identifiers; the basic categories clearly define the device type (such as temperature sensors), the industry tags are associated with the semantic standards of the vertical industry, and the custom identifiers include the device manufacturer ID and a unique serial number to avoid functional conflicts.

[0024] Capability vector: It adopts a quantitative description of static parameters and dynamic thresholds; static parameters include fixed performance such as accuracy (±0.1°C), sampling rate (1Hz, supporting dynamic adjustment from 0.1-10Hz), range (-40~125°C), and data bit width (16 bits); dynamic thresholds mark the capability boundaries (e.g., the minimum sampling rate is 0.1Hz in low power mode, and the accuracy becomes ±0.2°C in high temperature environment >85°C).

[0025] Native interface contract: Fully defines the data syntax, semantics, and transmission rules. The syntax part specifies the raw data format (e.g., ADC raw value: 16-bit unsigned integer, little-endian storage, each 2 bytes represents a sample value); the semantic part provides conversion formulas and calibration instructions (e.g., temperature (°C) = ADC value × 0.01 - 20, factory calibration offset + 0.05°C); the transmission rules specify the data encapsulation method (e.g., Zigbee frame type 0x01, endpoint number 0x0A, cluster ID 0x0002) and interaction timing (e.g., wait 100ms after sending data to receive an acknowledgment frame, and retransmit twice if timeout occurs).

[0026] Resource constraints are categorized into three types: communication constraints, computational constraints, and security constraints. Communication constraints include power consumption limits (10mW in active state, 0.1mW in sleep state), bandwidth requirements (40 bytes per frame, 1 frame per second), and communication mode preferences (LoRaWAN preferred, switching to NB-IoT when the signal is weaker than -110dBm). Computational constraints specify local processing capabilities (only simple mean filtering is supported, complex AI inference is not supported). Security constraints specify data processing requirements (sensitive data must be locally encrypted using AES-128, with the key obtained through the security handshake during the device's first network access) and access control rules (only requests from gateway IP segments 192.168.1.0 / 24 are allowed to access).

[0027] Extended fields: Reserve space for industry customization, such as adding anti-interference level (IEC61000-4-3 Level 3) for industrial equipment, and supplementing compliance certification (FDA21CFR Part 11) and other information for medical equipment.

[0028] After receiving the device description file through the multi-mode communication module (integrating LoRa / NB-IoT / Wi-Fi / Ethernet interfaces), the AI ​​gateway performs three steps of processing: First, invalid files are filtered out through syntax validation (verifying the legality of JSON / XML format) and semantic validation (matching industry ontology standards), rejecting devices that do not meet the specifications from accessing the device; second, core information is extracted to create a unique digital gene ID for the device (such as DevGene-2024-001-TempSensor); finally, a dynamic device digital gene model is constructed. This model is not statically stored, but rather a three-dimensional structure containing a basic attribute layer, a real-time state layer, and a capability evolution layer. The basic attribute layer stores fixed information about the device (functional fingerprint, factory capability vector). The real-time status layer periodically collects data (once every 30 seconds) through the gateway to update the current status of the device (such as current battery level 3.2V, current sampling rate 0.5Hz, signal strength -85dBm). The capability evolution layer records changes in device capabilities (e.g., after 6 months of use, the accuracy in high-temperature environments decays to ±0.25°C, and the system automatically updates the capability vector). The model is ultimately stored in the distributed knowledge base on the gateway's local machine (using a hybrid storage of Redis and LevelDB, with Redis caching real-time status and LevelDB storing historical data), and key information is synchronized to the regional device management platform through the edge cloud synchronization interface.

[0029] The system analyzes structured business requirements and integrates multi-source contextual information, then uses an AI decision model to generate joint execution strategies for devices and networks that meet QoS requirements. Specifically, it includes: Demand Intent Analysis: Service requests issued by cloud applications (such as smart park energy management systems and industrial production line monitoring platforms) are no longer simple data collection commands, but structured information that conforms to intent description specifications (using JSON-LD format and supporting semantic association). The gateway's intent parsing engine employs a hybrid architecture combining Natural Language Processing (NLP) and domain rule engines to perform layered parsing of requests. First, the core requirements (such as region A, temperature and humidity data, and 0.2Hz sampling) are extracted using the NLP module to solve the problem of non-standardized representation (such as automatically converting data sampled every 5 seconds to 0.2Hz). Then, the domain rule engine matches QoS constraints (such as latency < 2s corresponding to the latency threshold of the network link, and accuracy ± 0.1°C corresponding to the filtering conditions of the device capability vector). Finally, the executable sub-objectives are broken down (such as screening devices in area A that meet the requirements for temperature and humidity acquisition and have a temperature accuracy of ≥ ±0.1°C, and configuring data transmission link latency < 1.5s, with 0.5s of redundancy reserved).

[0030] After parsing, the engine outputs an intent breakdown list, which serves as the direct basis for subsequent policy generation. At the same time, the intent ID is bound to the gateway ID for easy tracking throughout the entire process. The gateway has a built-in multi-source context acquisition module that collects three types of context information in real time through multiple channels such as hardware sensors, network interfaces, and device interactions. It also processes redundant and conflicting data through a weighted fusion algorithm to output a unified context view. The acquisition frequency is dynamically adjusted according to the intent priority (high-priority intents are acquired once every 100ms, and low-priority intents are acquired once every 1s). Network Context: Real-time status of heterogeneous links is collected using network diagnostic tools (such as ping, traceroute, and SNMP protocol). Specific metrics include: 5G link (bandwidth 200Mbps, latency 30ms, packet loss rate 0.1%, communication cost per MB 0.001 yuan), Wi-Fi link (bandwidth 500Mbps, latency 10ms, packet loss rate 0.05%, free), and LoRa link (bandwidth 300bps, latency 100ms, packet loss rate 1%, free). Link stability is also indicated (e.g., 5G has a medium stability level in the edge area of ​​the factory, while Wi-Fi has a low stability level in densely populated equipment areas).

[0031] Device context: The target device status is obtained through device heartbeat packets (every 30 seconds) and active queries (every minute), including: battery level (3.2V, 70% remaining, supporting 14 days of continuous operation), current load (CPU utilization 15%, memory utilization 20%), movement status (stationary, location coordinates X:120.12, Y:30.15), current operating mode (normal mode, sampling rate 1Hz), and historical fault records (no faults in the last 7 days). For battery-powered devices, additional energy consumption estimates are calculated (e.g., at a sampling rate of 0.2Hz, operation can be extended to 21 days).

[0032] Environmental context: It is obtained by fusing environmental data reported by the device with the environmental sensors built into the gateway, including: physical environment (temperature 25°C, humidity 60%, signal interference intensity -70dBm), spatial relationship (there are 12 temperature and humidity sensors in area A, of which 3 are located near the device heat dissipation vent, and the data may be interfered with), and business scenario (currently in production period, the device is in continuous operation and frequent sleep is not allowed).

[0033] The gateway's built-in AI decision-making model adopts a hybrid architecture of reinforcement learning and rule constraints (based on the TensorFlowLite edge inference framework, model size <2MB, inference latency <50ms). Its multi-objective optimization goals are highest intent satisfaction, lowest resource consumption, and optimal system stability. It takes as input an intent breakdown list, a device digital gene model, and a contextual panoramic view, and outputs a structured joint communication and computation strategy, specifically including the following sub-strategies: Equipment selection sub-strategy: Select 8 sensors in screening area A that meet the requirements of temperature accuracy ±0.1°C and humidity accuracy ±3%RH and are currently in normal condition, and exclude interfering devices near 3 heat dissipation vents; Data acquisition sub-strategy: Configure the sampling rate of the selected device to 0.2Hz, and use batch reporting mode for data transmission (aggregate data from 8 devices every 5 seconds to form a 320-byte data packet). Protocol adaptation strategy: The device side adopts the Zigbee protocol (matching the device's native interface), and the cloud side adopts the HTTP / 3 protocol (to meet the latency requirement of <2s). Link selection sub-policy: Prioritize Wi-Fi link transmission (latency 10ms < 1.5s redundancy threshold, free of charge); if Wi-Fi packet loss rate > 1%, automatically switch to 5G link. Data processing sub-strategy: Perform outlier removal (temperature > 40°C or < 0°C is considered outlier) and data compression (using gzip compression, compression ratio 3:1) on the gateway side; Emergency response sub-strategy: If any device reports a temperature >35°C, immediately increase the sampling rate of that device to 1Hz, switch the transmission link to 5G, and push alarm information to the cloud. After the strategy is generated, its feasibility is verified (e.g., confirming that all selected devices are online and the link bandwidth meets the requirements). If there is a conflict, it is returned for adjustment (e.g., if there is no available Wi-Fi, 5G is selected directly). Finally, an executable strategy file is output.

[0034] QoS is a core technical concept in the field of network communication. It refers to a standardized technical mechanism that precisely defines, allocates, and guarantees the quality indicators of data transmission, ensuring that different business needs (such as industrial monitoring, emergency alarms, and data collection) can obtain the expected transmission performance in complex network environments. Essentially, it is to resolve the contradiction between the limited network resources and the differences in business needs, and to prevent high-priority services from being squeezed out by low-priority services, or critical data from failing to transmit due to network congestion or link fluctuations. Virtual communication endpoints are dynamically created based on execution strategies, and protocol stacks and data processing pipelines are assembled on demand to achieve transparent adaptation and communication between heterogeneous devices and the cloud. Specifically, it includes: After receiving the federated communication and computing policies, the gateway's virtual endpoint management module instantiates a virtual communication endpoint (VCE) using containerization technology (based on DockerEdge lightweight containers, with a startup time of <1s and a memory footprint of <10MB). This endpoint has a unique logical identifier (such as VCE-IntentID-20241202-001) and is bound to a dedicated resource quota (1 CPU core, 64MB of memory, and 10Mbps of network bandwidth). VCE exhibits transparent adaptation characteristics to both upstream and downstream processes: For the cloud, VCE emulation is an ideal device, providing a standard HTTP / 3 interface and reporting temperature and humidity data for region A every 5 seconds in CSV format, which fully meets the cloud's intent requirements. The cloud does not need to concern itself with the protocol differences of the underlying devices. For terminal devices, VCE emulation acts as a local agent, supporting native interaction with the Zigbee protocol. It can understand the device's raw ADC data format and calibration rules, allowing the device to connect without modifying any firmware. Meanwhile, VCE has a built-in status monitoring module that provides real-time feedback on its own operating status (such as CPU utilization and data processing latency) and maintains communication with the gateway's policy management module to support dynamic adaptation after policy adjustments.

[0035] The gateway's protocol stack component library pre-stores various lightweight protocol driver modules (each module is less than 50KB in size and supports hot-swapping). VCE assembles a dedicated protocol stack according to the protocol adaptation requirements in the policy by loading components on demand and dynamically linking them, thereby realizing bidirectional protocol conversion between the device side and the cloud side. Device-side protocol processing (downlink commands combined with uplink data): Load the Zigbee protocol driver module, parse the configuration instructions issued by the gateway (such as adjusting the sampling rate to 0.2Hz), encapsulate them into Zigbee data frames according to the device's native interface contract (frame type 0x01, endpoint number 0x0A, data field 0x020x000x0A represents a sampling rate of 0.2Hz), and send them to the device through the Zigbee radio frequency module; The system receives Zigbee data frames uploaded by the receiving device (containing the raw ADC values ​​of 8 sensors), parses the raw data through the protocol driver module (e.g., the ADC value of sensor 1 is 5200, and the ADC value of sensor 2 is 5180), then calls the data conversion engine to calculate the actual temperature and humidity according to the semantic rules in the device's digital genes, and finally converts it into intermediate data in standard JSON format. Cloud-side protocol processing (uplink data combined with downlink commands): Load the HTTP / 3 protocol driver module, convert the intermediate data in standard JSON format to CSV format according to policy requirements, compress it with gzip and encapsulate it into an HTTP / 3 request (using the QUIC transport layer protocol to ensure low latency), and send it to the cloud application interface; It receives control commands from the cloud (such as extending data collection to 20:00), parses them through the HTTP / 3 protocol driver module, converts them into the standard command format inside VCE, and then forwards them to the device-side protocol driver module to realize cross-protocol transmission of commands.

[0036] After the protocol stack is assembled, VCE will perform protocol connectivity tests (such as sending test frames to verify device response and sending heartbeat packets to verify cloud connection) to ensure the correctness of protocol adaptation. If the test fails, the components will be automatically reloaded or a backup protocol will be switched.

[0037] This also includes processing pipeline injection: VCE internally builds a configurable data processing pipeline. Based on the data processing requirements in the strategy, it selects the corresponding modules from the processing module library (including filtering, aggregation, compression, encryption, AI inference, etc.) and injects them into the pipeline in sequence. This enables edge preprocessing of data during transmission, reducing cloud computing pressure and network transmission volume. Outlier removal: Based on the 3σ criterion, data exceeding the temperature range of 0-40°C and humidity range of 10-90%RH are removed (e.g., if sensor 3 reports a temperature of 45°C, it is determined to be an anomaly, and the data is automatically removed and recorded in the log). Data aggregation: The temperature and humidity data from 8 sensors are aligned by timestamp and aggregated into a single record containing data from all devices, reducing the number of data frames (from 8 frames / 5 seconds to 1 frame / 5 seconds). Data compression: The gzip compression algorithm is used to compress CSV format data, with a compression ratio of 3:1 (320 bytes compressed to 107 bytes), reducing network transmission bandwidth requirements; Lightweight AI Anomaly Detection: Load a pre-trained isolated forest model (model size < 1MB) and perform real-time inference on the processed temperature and humidity data. If a device's data fluctuation is found to be greater than 5°C / minute (e.g., sensor 5 suddenly changes from 32°C to 38°C), it is immediately marked as suspicious data, an alarm flag is added to the reported data, and an emergency response sub-strategy is triggered. The processing pipeline supports dynamic adjustment. If the gateway receives a new policy (such as an increase in data encryption requirements), it can inject an AES-128 encryption module in real time without restarting VCE, ensuring business continuity.

[0038] It coordinates and schedules heterogeneous links and dynamically weaves the optimal transmission path, ensuring stable data transmission in complex network environments through real-time adaptive adjustment; Specifically, it includes: Heterogeneous link collaborative scheduling: The gateway's link management module maintains a heterogeneous link status table in real time (updated every 50ms), recording metrics such as bandwidth, latency, packet loss rate, cost, and stability for each link. Based on data priority and QoS requirements, VCE dynamically allocates transmission links using a link scheduling algorithm (based on weighted round-robin and priority preemption mechanisms), achieving precise matching of link resources. High-priority data (such as emergency alarm data): 5G links are allocated first, and a priority preemption mechanism is adopted (if the 5G link is occupied by low-priority data, the low-priority transmission is immediately interrupted to ensure alarm data transmission). At the same time, link backup is configured (if the 5G latency is >50ms, it automatically switches to the backup industrial Ethernet link). Medium-priority data (such as normal temperature and humidity data): prioritize the allocation of Wi-Fi links (to meet the latency requirement of <2s and with no communication cost), and configure bandwidth reservation (reserve 1Mbps bandwidth for this type of data to avoid it being squeezed out by other data); Low-priority data (such as device log data): The data is split and transmitted simultaneously via LoRa and idle Wi-Fi links using link aggregation to reduce the impact on high-priority links. At the same time, a latency tolerance mechanism (allowing latency <60s) is used to select the time with the best link quality for transmission.

[0039] During the scheduling process, the module monitors link utilization (calculating link occupancy every second). If the utilization of a link exceeds a preset threshold, it automatically restricts access for low-priority data to ensure the transmission quality of high-priority services.

[0040] When the target device is outside the direct coverage area of ​​the current gateway (e.g., the sensor in area B is 500 meters away from the current gateway, and the LoRa signal attenuates to -120dBm), the gateway establishes communication with three gateways within a 1-kilometer radius through the edge gateway collaboration protocol (a lightweight protocol based on MQTT-SN, occupying less than 10kbps of bandwidth), constructing a gateway collaboration network to achieve dynamic weaving of multiple paths, specifically including: Path probing phase: The current gateway sends a path probing request to surrounding gateways, including the target device ID and data QoS requirements. Surrounding gateways respond with the link status between themselves and the target device. Path calculation phase: An improved Dijkstra algorithm is used to calculate the optimal end-to-end path with the goals of minimizing latency and hop count. The final path example is: Area B sensor to Zigbee to gateway B to Wi-Fi Mesh to current gateway to Wi-Fi to cloud. The total latency of this path is 120ms (Zigbee 20ms + Wi-Fi Mesh 50ms + Wi-Fi 50ms), which meets QoS requirements. Path establishment phase: The current gateway and the responding gateway (Gateway B) negotiate to establish a temporary data transmission channel, agree on the data encapsulation format (UDP packets encrypted with IPsec) and transmission timing (data is synchronized every 5 seconds) to ensure the security and reliability of cross-gateway data transmission; Path maintenance phase: The gateway cooperative network monitors the status of each hop link in real time. If the packet loss rate of the Wi-Fi Mesh link of gateway B is greater than the preset threshold, the path is automatically recalculated and a seamless switching mechanism is used (the old path is maintained before the new path is established to avoid data loss). VCE's built-in transmission adaptive module works in conjunction with the link management module to monitor data transmission quality (adding CRC checksums and timestamps to each data frame, with the receiving end providing feedback on transmission status) and perceive transmission performance in real time. If it detects that QoS indicators do not meet requirements, it immediately triggers an adaptive adjustment strategy. Link quality deterioration adjustment: If the Wi-Fi link latency increases from 10ms to 25ms (close to the 2s threshold), immediately initiate a data compression upgrade (switching from gzip compression to LZMA compression, increasing the compression ratio to 5:1) to reduce data transmission volume; if the latency continues to rise to 500ms, automatically switch to the 5G link and send a link quality alarm to the gateway to prompt investigation of Wi-Fi interference issues; Network congestion adjustment: If multiple gateways are detected transmitting data to the cloud simultaneously, causing network congestion (cloud reception latency > 1.8s), a data caching and off-peak transmission mechanism is activated. VCE temporarily stores the data in a local cache (1GB capacity, supports power outage protection) and uploads it in batches during periods of reduced cloud load (such as 0-5 minutes per hour). At the same time, the transmission frequency is adjusted (from 0.2Hz to 0.1Hz to reduce the amount of data generated) to ensure that the latency meets the requirements. Equipment movement and adjustment: If a change in the position of a mobile device (such as a sensor on an inspection robot) is detected, causing the current link signal to weaken, the link pre-switching mechanism is triggered in advance (the link with the new gateway is established before the signal strength drops to -100dBm), so as to achieve zero-interruption transmission.

[0041] Verify the alignment of business intent with feedback data to optimize the AI ​​model, and combine this with the business lifecycle to achieve precise control and efficient recycling of virtual communication endpoints and resources, specifically including: Intent conformity verification: The gateway's intent verification module uses the intent breakdown list as a benchmark and performs full verification of the VCE output from three dimensions: data quality, transmission performance, and service continuity. The verification frequency matches the intent priority (high priority is verified every frame, medium priority is verified every 10 frames, and low priority is verified every 100 frames). Data quality verification: Compare the actual output data with QoS accuracy requirements (such as whether the temperature data meets ±0.1°C). Through error calculation (deviation between actual value and standard value) and consistency verification (correlation of data from different sensors in the same area, such as temperature difference between adjacent sensors should be <2°C), determine whether the data quality meets the standards. If the error of a sensor data is >0.1°C for 3 consecutive times, it is marked as device accuracy decay, and the weight of the device is automatically adjusted (reducing the proportion of its data in the aggregation result). Transmission performance verification: Calculate the data transmission latency (total time from device sampling to cloud reception) by comparing timestamps, and statistically analyze the packet loss rate and retransmission count. If the latency is >2s or the packet loss rate is >0.5%, it is marked as transmission performance failure, triggering the transmission adaptive adjustment strategy. Service continuity verification: Monitor the operating status of VCE and the stability of device connection. If the device is offline for more than 10 seconds or the VCE fails and restarts more than twice per hour, it is marked as a service interruption risk and a backup device is automatically started (supplemented by selecting similar devices from the digital gene bank).

[0042] The verification results generate an intent compliance report in real time, including the compliance rate (e.g., the current compliance rate of medium-priority intents is 99.8%), the reasons for non-compliance (e.g., one instance of latency exceeding the standard was caused by Wi-Fi interference), and improvement suggestions (e.g., adjusting the Wi-Fi channel), and are synchronized to the AI ​​decision-making model of the cloud and the gateway.

[0043] Feedback-based policy evolution: The gateway's AI model optimization module uses intent conformity reports as core feedback data, and continuously evolves its policy generation capabilities through online fine-tuning and offline training. Online fine-tuning: For scenarios that fail to meet standards in real time, the parameters of the AI ​​decision-making model are adjusted immediately (e.g., if latency exceeds the standard due to Wi-Fi interference, the priority weight of the 5G link is automatically increased from 0.3 to 0.5). The fine-tuning process is completed locally on the gateway and does not affect the current business operation. Offline training: Every night from 00:00 to 02:00 (off-peak business period), the intent conformity data, context data, and policy execution data of the whole day are synchronized to the model training platform of the edge cloud. The AI ​​model is updated using incremental training (based on historical model parameters, only training the features corresponding to the new data). The trained model is compressed to <2MB and pushed to the gateway to replace the old model. Knowledge Accumulation: Frequently occurring optimization cases (such as prioritizing 5G link heat dissipation vents in factory edge areas and requiring additional filtering of data from equipment near them) are extracted into a rule base and directly added to the rule constraint layer of the AI ​​decision-making model to improve the model's decision-making efficiency.

[0044] The gateway's lifecycle management module uses the persistence status of business intents as the core trigger condition, combined with context changes, to achieve intelligent management and control of the entire lifecycle of VCE and related resources, avoiding resource waste. Lifecycle startup: When the gateway receives a valid business intent and generates a policy, it automatically starts the VCE instantiation, protocol stack assembly, path establishment and other processes, and at the same time allocates dedicated resources (CPU, memory, network bandwidth) to the VCE and marks the resource status as occupied; Lifecycle operation: Real-time monitoring of the persistence status of business intents (such as whether the cloud sends a termination request) and context changes (such as whether the device is moved out of the coverage area or whether the business times out). If no termination conditions are met, VCE is maintained and resource allocation is dynamically adjusted (such as increasing memory quota when the data volume increases). Lifecycle Termination: When termination conditions are triggered (such as sending a termination command from the cloud, the service duration reaching 18:00 on 2024-12-02, or the device becoming permanently offline), a graceful destruction process is initiated: First, data collection and transmission are stopped, and a service termination report (including data statistics and running status) is sent to the cloud; second, the container resources (CPU, memory) occupied by VCE are released, and temporary protocol stack components and processing pipelines are deleted; finally, the status of the device's digital genetic model is updated (e.g., the device has completed the temperature and humidity collection task in area A and is currently idle), and the VCE's running logs are archived to local storage (retained for 7 days, supporting fault traceability); Resource reclamation optimization: A resource pooling mechanism is adopted, and released resources are uniformly included in the gateway resource pool, which supports rapid allocation to new VCE instances.

[0045] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0046] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the statement "includes a…" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0047] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0048] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0049] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0050] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0051] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0052] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0053] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A control method for an IoT gateway integrating communication and AI computing power, characterized in that, include: To build machine-understandable structured functional identity profiles and dynamic digital genetic models for physical devices, specifically including: After the device is powered on and initialized, it will start the built-in capability declaration module and broadcast or send the structured functional identity description file through the local communication interface. The file uses a hierarchical structure; After receiving the device description file through the multi-mode communication module, the AI ​​gateway processes it. First, invalid files are filtered out through syntax and semantic checks; second, core information is extracted to create a unique digital gene ID for the device; finally, a dynamic device digital gene model is constructed, which includes a three-dimensional structure of basic attribute layer, real-time status layer, and capability evolution layer. The basic attribute layer stores fixed information about the device; the real-time status layer periodically collects and updates the current status of the device through the gateway; and the capability evolution layer records changes in the device's capabilities. The model is ultimately stored in a distributed knowledge base on the gateway's local machine. The system analyzes structured business requirements and integrates multi-source contextual information, then uses an AI decision model to generate joint execution strategies for devices and networks that meet QoS requirements. Virtual communication endpoints are dynamically created based on execution strategies, and protocol stacks and data processing pipelines are assembled on demand to achieve transparent adaptive communication between heterogeneous devices and the cloud, specifically including: After receiving the federated communication and computing policy, the gateway's virtual endpoint management module instantiates a virtual communication endpoint. This endpoint has a unique logical identifier and is bound to a dedicated resource quota. Virtual communication endpoints exhibit transparent adaptation characteristics to upstream and downstream: for the cloud, virtual communication endpoints simulate an ideal device, providing standard interfaces; For terminal devices, the virtual communication endpoint is simulated as a local agent, supporting native interaction with the Zigbee protocol and understanding the device's ADC raw data format and calibration rules; Also includes: The gateway's protocol stack component library pre-stores various lightweight protocol driver modules. Virtual communication endpoints assemble their own protocol stacks according to the protocol adaptation requirements in the policy, enabling bidirectional protocol conversion between the device side and the cloud side. This also includes processing pipeline injection: The virtual communication endpoint internally constructs a configurable data processing pipeline. Based on the data processing requirements in the strategy, the corresponding module is selected from the processing module library and injected into the pipeline in sequence. It coordinates and schedules heterogeneous links and dynamically weaves the optimal transmission path, ensuring stable data transmission in complex network environments through real-time adaptive adjustments.

2. The control method for an IoT gateway integrating communication and AI computing power according to claim 1, characterized in that, The file adopts a hierarchical structure, specifically including functional fingerprints, capability vectors, native interface contracts, resource constraints, and extended fields.

3. The control method for an IoT gateway integrating communication and AI computing power according to claim 1, characterized in that, The system analyzes structured business requirements and integrates multi-source contextual information. It then uses an AI decision model to generate joint execution strategies between devices and the network that meet QoS requirements. Specifically, this includes: The gateway's intent parsing engine performs layered parsing of requests: first, it extracts the core requirements through the NLP module to solve the problem of non-standardized expression; then, it matches QoS constraints through the domain rule engine; and finally, it breaks down the requests into executable sub-goals. After parsing, the engine outputs a list of intent decompositions, which serves as the direct basis for subsequent policy generation, and binds the intent ID to the gateway ID. Three types of contextual information are collected in real time, and redundant and conflicting data are processed through a weighted fusion algorithm to output a unified contextual panoramic view. Network context: Real-time status of heterogeneous links is collected through network diagnostic tools, including 5G links, Wi-Fi links, and LoRa links. Device context: Obtain the target device status through device heartbeat packets and active queries, including: battery level, current load, mobility status, current operating mode, and historical fault records; Environmental context: obtained by fusing environmental data from the gateway’s built-in environmental sensors with environmental data reported by the device, including physical environment, spatial relationships, and business scenarios.

4. The control method for an IoT gateway integrating communication and AI computing power according to claim 3, characterized in that, Also includes: The gateway's built-in AI decision-making model adopts a hybrid architecture of reinforcement learning and rule constraints, with multiple optimization objectives such as intent satisfaction, resource consumption, and system stability. It takes as input an intent breakdown list, a device digital gene model, and a contextual panoramic view, and outputs a structured joint communication and computing strategy. After the strategy is generated, its feasibility is verified. If there are conflicts, adjustments are made, and finally an executable strategy file is output.

5. The control method for an IoT gateway integrating communication and AI computing power according to claim 1, characterized in that, Also includes: Device-side protocol processing: Parse the configuration instructions issued by the gateway, encapsulate them into Zigbee data frames according to the device's native interface contract, and send them to the device through the Zigbee radio frequency module; The system receives Zigbee data frames uploaded by the receiving device, parses out the raw data, calls the data conversion engine, calculates the actual temperature and humidity based on the semantic rules in the device's digital DNA, and finally converts it into intermediate data in standard JSON format. The intermediate data in standard JSON format is converted into CSV format according to the policy requirements, compressed with gzip, and then packaged into an HTTP / 3 request before being sent to the cloud application interface. It receives control commands from the cloud, parses them, converts them into the standard command format inside the virtual communication endpoint, and then forwards them to the device-side protocol driver module.

6. The control method for an IoT gateway integrating communication and AI computing power according to claim 1, characterized in that, Cooperative scheduling of heterogeneous links and dynamic weaving of optimal transmission paths ensures stable data transmission in complex network environments through real-time adaptive adjustments. Specifically, this includes: The gateway's link management module maintains a heterogeneous link status table in real time, records the metrics of each link, and the virtual communication endpoint dynamically allocates transmission links based on data priority and QoS requirements, including selecting the appropriate link based on priority data. When the target device is outside the direct coverage area of ​​the current gateway, the gateway establishes communication with three gateways within a 1-kilometer radius through the edge gateway collaboration protocol to build a gateway collaboration network and realize the dynamic weaving of multiple paths.

7. The control method for an IoT gateway integrating communication and AI computing power according to claim 1, characterized in that, Also includes: Verify the alignment of business intent and provide feedback data to optimize the AI ​​model; combine the business lifecycle to achieve control and reclamation of virtual communication endpoints and resources.

Images