An optical fiber Internet gateway, energy efficiency regulation method, computer device and medium

Abstract

The present application relates to the technical field of smart grid monitoring, and more particularly to a kind of optical fiber Internet gateway, energy efficiency regulation method, computer equipment and medium;The optical fiber Internet gateway, heterogeneous computing unit, intelligent network switching unit, optical fiber access unit, intelligent power management unit;Heterogeneous computing unit, including main processor and artificial intelligence coprocessor;Intelligent network switching unit, including at least one electric port network port and at least one optical fiber port, intelligent network switching unit and heterogeneous computing unit are communicatively coupled;Optical fiber access unit, including at least one pluggable optical module, pluggable optical module is connected to optical fiber port.Through such a way, it solves the comprehensive technical problem that prior art cannot simultaneously meet the three aspects of high reliable communication, ultra-low power consumption operation and high intelligent integration, provides key equipment foundation for building all-weather, high reliable, intelligent power transmission line monitoring system.

Description

Technical Field

[0001] This invention relates to the field of smart grid monitoring technology, and in particular to a fiber optic Internet of Things (IoT) gateway, an energy efficiency control method, a computer device, and a medium. Background Technology

[0002] With the rapid advancement of ultra-high-voltage power grid construction, the safe and stable operation of transmission lines is crucial to the overall reliability of the power grid. Online monitoring systems for power lines need to collect multi-dimensional data in real time, including conductor temperature, sag, icing, micro-meteorological conditions, and images and videos. These monitoring terminals are typically deployed on harsh outdoor towers, posing significant challenges to their reliable operation. In traditional transmission line monitoring systems, data transmission primarily relies on wireless communication technology, and equipment energy management and system integration are key factors affecting long-term stable operation. However, as monitoring operations evolve from single-data-point acquisition to multi-sensor fusion, high-frequency image and video streaming, and edge intelligent processing, higher demands are placed on the determinism of communication links, the energy efficiency of the equipment itself, and the level of system integration and intelligence.

[0003] In the field of energy-saving technology for transmission line monitoring devices, some explorations have been made in the prior art. Existing technology (application publication number CN102412628A) discloses an ultra-low power intelligent monitoring device for transmission lines, which employs an ultra-low power system CPU. By switching the system between sleep and wake-up states and utilizing software to control the activation and deactivation of peripheral communication and storage modules such as GPRS (General Packet Radio Service) and ZigBee (a low-speed, short-range wireless communication technology based on the IEEE 802.15.4 standard), the corresponding modules are activated only when tasks need to be performed, thereby reducing the average power consumption of the system. This method focuses on achieving energy saving through module start-stop control. However, this technical solution has obvious limitations: First, its communication infrastructure relies on public wireless networks such as GPRS, and its transmission reliability is significantly affected by terrain and weather conditions, resulting in problems such as signal blind spots, bandwidth limitations, and latency jitter. This makes it difficult to meet the stringent requirements of determinism and high reliability for data transmission in critical line monitoring services. Second, although the solution saves energy by shutting down the communication module, the computing, communication, and data acquisition functions in its overall architecture are still separate, with low integration. Equipment stacking leads to complex installation and maintenance and an increase in system failure points. More importantly, its energy-saving strategy is based on a simple passive response mode of "shutting down when the task is completed," lacking the ability to predict and learn from the patterns of business traffic. It cannot make forward-looking dynamic energy consumption adjustments based on actual, time-varying monitoring task requirements, resulting in limited energy efficiency optimization. Specifically, existing technologies suffer from two main problems: First, the reliance on wireless communication has inherent reliability bottlenecks. Switching to more reliable wired fiber optic communication, however, requires tower-mounted fiber optic access equipment (such as industrial switches and protocol converters) with power consumption typically reaching 5-15 watts. In solar-powered outdoor transmission tower scenarios, this results in a bulky and costly power supply system, and exacerbates the risk of power outages during prolonged periods of rain, ultimately compromising system reliability due to power supply challenges. Second, existing solutions have rigid functionalities and energy efficiency management, failing to integrate efficient network switching, deep edge computing, and intelligent energy scheduling. This makes it difficult to meet the comprehensive demands of modern transmission line monitoring for high-reliability communication, ultra-low-power operation, and highly intelligent integration. These intertwined problems make it difficult for existing transmission line monitoring technologies, whether wireless low-power or wired high-reliability solutions, to simultaneously achieve revolutionary improvements in communication reliability and equipment energy efficiency under harsh outdoor power supply conditions. Therefore, existing transmission line monitoring technologies struggle to balance high-reliability communication, ultra-low-power operation, and highly intelligent integration, failing to meet the all-weather, high-reliability intelligent monitoring requirements of ultra-high-voltage lines. Summary of the Invention

[0004] To address the aforementioned shortcomings or deficiencies, this invention provides a fiber optic IoT gateway, energy efficiency control method, computer equipment, and medium, which can solve the comprehensive technical problem that existing technologies cannot simultaneously meet the three requirements of high-reliability communication, ultra-low power consumption operation, and highly intelligent integration.

[0005] This invention provides a fiber optic Internet of Things gateway, comprising a heterogeneous computing unit, an intelligent network switching unit, a fiber optic access unit, and an intelligent power management unit.

[0006] Heterogeneous computing units include a main processor as the control core and an artificial intelligence coprocessor for hardware acceleration.

[0007] The intelligent network switching unit includes at least one electrical network port and at least one optical fiber port. The intelligent network switching unit communicates and couples with the heterogeneous computing unit through a high-speed data bus interface.

[0008] The fiber optic access unit includes at least one pluggable optical module, which is pluggable to the fiber optic port of the intelligent network switching unit.

[0009] The intelligent power management unit includes a digital programmable power management integrated circuit. The control interface of the intelligent power management unit is connected to the heterogeneous computing unit, and multiple independent power supply output terminals of the intelligent power management unit are respectively connected to the power input terminals of the heterogeneous computing unit, the intelligent network switching unit, and the fiber optic access unit.

[0010] The main processor is configured to execute a three-level collaborative energy-saving management process, which includes a business forecasting and learning phase, an intelligent mode decision-making phase, and a hardware collaborative execution phase.

[0011] The intelligent power management unit is configured to perform independent switching control or dynamic voltage regulation on multiple independent power supply outputs in response to power management commands from the heterogeneous computing unit.

[0012] According to a second aspect, this invention provides an energy efficiency control method based on a fiber optic IoT gateway. This method, based on the fiber optic IoT gateway described in the above embodiments, is applied to a main processor and includes: Perform time-series analysis and feature extraction on historical business data and real-time sensor data, and train and deploy a business volume prediction model.

[0013] Based on the forecast output of the traffic volume prediction model, the working mode of the fiber optic IoT gateway is determined. The working modes include active mode, intelligent standby mode and deep sleep mode.

[0014] The hardware coordination steps include chip-level coordination steps, link-level coordination steps, and system-level coordination steps.

[0015] The chip-level collaboration steps include: adjusting the operating voltage and core frequency of the main processor through a dynamic voltage and frequency adjustment strategy based on the real-time load indicators of the fiber optic IoT gateway, and offloading computationally intensive data processing tasks to the artificial intelligence coprocessor for execution.

[0016] The link-level coordination steps include: controlling the intelligent network switching unit to enable the energy-efficient Ethernet protocol at the data link layer, and performing power gating operations on the physical layer circuits of designated network ports that are idle on the intelligent network switching unit according to the current decision operating mode.

[0017] The system-level coordination steps include: sending power management commands to the intelligent power management unit to perform time-sharing and zone-based power supply scheduling for multiple independent power supply domains corresponding to multiple independent power supply output terminals of the intelligent power management unit.

[0018] According to a third aspect, the present invention provides a computer device comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform any of the fiber optic IoT gateway control methods in the embodiments of the present invention.

[0019] According to another aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute any of the fiber optic Internet of Things gateway control methods in the embodiments of the present invention.

[0020] The present invention provides a fiber optic IoT gateway. This gateway achieves system-level energy efficiency optimization through a collaborative architecture of four core hardware units: a heterogeneous computing unit, an intelligent network switching unit, an optical fiber access unit, and an intelligent power management unit, along with a three-level collaborative energy-saving management process executed by the main processor. Specifically, the heterogeneous computing unit integrates the main processor and an artificial intelligence coprocessor, providing a heterogeneous computing foundation for intelligent energy efficiency management; the intelligent network switching unit is coupled to the computing unit via a high-speed data bus and has both electrical and optical fiber ports, forming a flexible internal and external data exchange hub; the optical fiber access unit connects to the optical fiber port via pluggable optical modules, establishing a deterministic wired communication backbone based on power optical cables; and the intelligent power management unit receives instructions through a digitally programmable power management integrated circuit and performs fine-grained control over multiple independent power outputs, constituting the execution end of energy consumption regulation. The three-level collaborative energy-saving management process covers three stages: business prediction and learning, intelligent mode decision-making, and hardware collaborative execution, linking intelligent decision-making and hardware control in a closed loop.

[0021] In its overall technical solution, this invention addresses the comprehensive technical problems described in the background art, including insufficient reliability due to reliance on wireless communication, increased power supply pressure due to excessive power consumption of access devices, and low integration and lack of intelligence due to functional separation. By introducing an optical fiber access unit based on pluggable optical modules, it directly utilizes existing optical ground wire (OPGW) or all-dielectric self-supporting optical cables. Self-Supporting (ADSS) utilizes redundant fiber cores to construct communication links, physically avoiding terrain and weather interference and solving the core bottleneck of high-reliability communication. Addressing the issue of high power consumption, an intelligent power management unit performs independent switching control or dynamic voltage regulation on multiple independent power outputs. Combined with a three-level collaborative energy-saving management process, it achieves refined management of power supply heads by region and time, providing a hardware foundation for significantly reducing the system's average power consumption. Addressing the issues of functional dispersion and lack of intelligence, it deeply integrates heterogeneous computing, network switching, fiber optic access, and power management into a single gateway device. The main processor uniformly executes a collaborative management process of prediction, decision-making, and execution, improving integration, reducing maintenance complexity, and endowing the gateway with the intelligent capability to adaptively adjust based on service status. Therefore, the technical solution of the present invention, through the collaborative design of the above-mentioned hardware architecture and management process, simultaneously addresses the comprehensive technical problem in the background technology that the existing technology cannot simultaneously meet the three requirements of high-reliability communication, ultra-low power consumption operation and highly intelligent integration, by simultaneously exerting efforts from three levels: communication medium, energy consumption management and system integration. It provides a key equipment foundation for building an all-weather, highly reliable and intelligent transmission line monitoring system. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the specific fiber optic IoT gateway hardware implementation architecture and its energy and data flow coordination in one embodiment of the present invention; Figure 2 This is a schematic flowchart of an energy efficiency control method based on a fiber optic Internet of Things gateway according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the networking architecture and service flow of a multi-point collaborative intelligent monitoring system based on the optical fiber IoT gateway of the present invention in one embodiment; Figure 4 This is a schematic diagram of the business logic and data flow based on the three-level collaborative energy-saving management concept of the present invention in one embodiment; Figure 5 This is a block diagram of a computer device for implementing embodiments of the present invention. Detailed Implementation

[0023] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0024] During the development of this invention, researchers, through extensive experiments and data analysis, revealed a profound correlation between device power consumption, communication reliability, and the level of system integration intelligence. In harsh outdoor scenarios reliant on solar power, pursuing a single indicator (such as solely pursuing high reliability or solely pursuing low power consumption) leads to a "weakest link" effect in overall system performance, making it difficult to meet the comprehensive requirements of modern power transmission line monitoring for energy efficiency, reliability, and intelligence. Based on this discovery, this invention innovatively proposes this technical solution, the core of which lies in constructing a three-level collaborative hardware energy efficiency architecture and its intelligent management process at the "chip-level-link-system-level." Through the collaborative work of heterogeneous computing units, highly reliable network switching and fiber optic access units, and finely scheduled power management units, and by executing a closed-loop management process encompassing service prediction, mode decision-making, and hardware collaboration, the average system power consumption is reduced while ensuring communication determinism and high reliability. This solution embodies the core inventive concepts of "collaborative hardware architecture, predictive energy management, and closed-loop system control."

[0025] Specifically, through long-term comparative testing of existing wireless and high-power wired solutions on ultra-high voltage lines, the invention team discovered that traditional solutions suffer from a technical deficiency where it is difficult to simultaneously achieve high communication reliability, equipment energy efficiency, and functional integration. This invention, by introducing fiber optic access units based on pluggable optical modules, directly utilizes power OPGW / ADSS optical cables to construct a deterministic backbone network, fundamentally circumventing the inherent reliability bottleneck of wireless communication by using wired media. Furthermore, by deeply integrating heterogeneous computing, intelligent network switching, fiber optic access, and digital programmable power management into a single gateway, it physically merges communication, computing, data acquisition, and power management functions. It also integrates intelligent traffic prediction, dynamic decision-making on operating modes, and energy-saving execution of various hardware units (processors, network ports, power supply domains) into a closed-loop process, improving the overall energy efficiency and intelligent adaptive level of the system. This achieves a unity of high-reliability communication, ultra-low-power operation, and highly intelligent integration, solving the technical problem mentioned in the background art where existing transmission line monitoring technologies cannot simultaneously meet the requirements of all-weather high-reliability monitoring and ultra-low-power long-term operation under limited power supply conditions in the field.

[0026] Therefore, according to a first aspect, the present invention provides a fiber optic Internet of Things gateway, including a heterogeneous computing unit, an intelligent network switching unit, a fiber optic access unit, and an intelligent power management unit.

[0027] Heterogeneous computing units include a main processor as the control core and an artificial intelligence coprocessor for hardware acceleration.

[0028] The main processor is the core computing device responsible for system control, protocol processing, and scheduling decisions; the artificial intelligence coprocessor is a hardware acceleration unit designed specifically for executing machine learning algorithms such as neural networks, used to improve the energy efficiency of specific computing tasks.

[0029] Specifically, the main processor and the AI ​​coprocessor interact with each other via a high-speed on-chip interconnect bus, forming a heterogeneous computing architecture. The main processor runs a lightweight real-time operating system, handling communication protocol stacks, sensor data aggregation, and energy management decisions; the AI ​​coprocessor loads a quantized and compiled lightweight neural network model, dedicated to handling computationally intensive tasks such as image recognition and data anomaly detection. For example, the heterogeneous computing unit can specifically use Huada Semiconductor's HC32F4A0 series microcontroller (based on the ARM Cortex-M4 core) as the main processor and integrate Rockchip's RK1808 Neural Processing Unit (NPU) as the AI ​​coprocessor. When the gateway receives a video stream from a high-definition camera, the main processor schedules the video frame data to the NPU's dedicated memory via Direct Memory Access (DMA). The NPU executes a pre-built MobileNetV2 model to identify wire icing or abnormal hanging objects, uploading only structured alarm information such as "ice detected" or "foreign object detected" and keyframe indexes, thus completing intelligent analysis locally.

[0030] The specific chip models and algorithm models described above are merely examples. Those skilled in the art can select other processors, switching chips, power management chips, or machine learning models with similar functions to implement this invention according to actual needs, without departing from the scope of protection of this invention.

[0031] The intelligent network switching unit includes at least one electrical network port and at least one optical fiber port. The intelligent network switching unit communicates and couples with the heterogeneous computing unit through a high-speed data bus interface.

[0032] The electrical network port is an Ethernet interface conforming to the IEEE 802.3 standard, used to connect various sensors (such as conductor temperature measurement, micro-meteorology, and image sensors) deployed on the tower; the optical fiber port is an optical Ethernet interface that supports small form-factor pluggable (SFP) optical modules, used to build backbone optical fiber communication links.

[0033] Specifically, this unit is a managed industrial Ethernet switch chip. Its data plane connects to the main processor's Media Access Control (MAC) layer via high-speed parallel buses such as Media Independent Interface (MII) / Reduced Gigabit Media Independent Interface (RGMII). Its control plane receives configuration and management from the main processor via a Serial Management Interface (SMI) or Management Data Input / Output (MDIO) bus, enabling Virtual Local Area Network (VLAN) segmentation, Quality of Service (QoS) policies, and port energy efficiency control. For example, Motorcom's YT9218 series switch chip can be used. This chip provides eight 10 / 100 / 1000M (megabits per second) adaptive electrical ports for connecting sensors and two SFP+ fiber optic ports for uplink. The main processor configures the switching chip via the SMI bus, assigns the port where the video sensor is located to a high-priority VLAN, and enables Energy Efficient Ethernet (EEE) based on IEEE 802.3az, which automatically reduces the power consumption of the port's physical layer (PHY) when there is no data transmission.

[0034] The fiber optic access unit includes at least one pluggable optical module, which is pluggable to the fiber optic port of the intelligent network switching unit.

[0035] Among them, a pluggable optical module is an optoelectronic device that converts electrical signals into optical signals and transmits them. It complies with the SFP or SFP+ Multi-Source Agreement (MSA) standard and has hot-swappable characteristics.

[0036] Specifically, this unit utilizes redundant fiber cores in existing overhead ground wire composite optical cables (OPGW) or all-dielectric self-supporting optical cables (ADSS) to convert the photoelectric signals of the intelligent network switching unit into optical signals of specific wavelengths (such as 1310 nm or 1550 nm) through the insertion of optical modules, enabling long-distance, interference-resistant data transmission. The main processor can access the digital diagnostic monitoring (DDM) interface of the optical module via the integrated circuit bus (I2C) to monitor parameters such as optical power and operating temperature in real time. For example, a single-fiber bidirectional (BiDi) SFP optical module with a center wavelength of 1550 nm can be used. This module can achieve data transmission and reception through a single optical fiber, saving half of the optical fiber resources. On a 25-kilometer-long OPGW optical cable link, the received optical power monitoring value of this module is... (decibels and milliwatts), within the normal range. to The value within ) indicates that the optical link is in good condition.

[0037] The intelligent power management unit includes a digital programmable power management integrated circuit. The control interface of the intelligent power management unit is connected to the heterogeneous computing unit, and multiple independent power supply output terminals of the intelligent power management unit are respectively connected to the power input terminals of the heterogeneous computing unit, the intelligent network switching unit, and the fiber optic access unit.

[0038] Among them, digital programmable power management integrated circuits are power management chips that can receive instructions through digital interfaces such as integrated circuit bus (I2C) or serial peripheral interface (SPI) and dynamically adjust the voltage, current or switching state of each output channel.

[0039] Specifically, this unit divides the power supply of the entire gateway system into multiple independent and controllable voltage domains, such as a core voltage domain, an input / output (I / O) voltage domain, a physical layer (PHY) voltage domain, and an optical module voltage domain. The main processor sends commands to this chip via the I2C bus based on the system's operating status to achieve independent power-on, power-off, or dynamic voltage adjustment for different voltage domains. For example, Texas Instruments' (TI) TPS65263 digital multiplexer power management integrated circuit (PMIC) is used. The main processor is configured with this PMIC to provide 1.2 volts (V) to the main processor core operating at 240 MHz (core voltage domain) and 3.3V to the physical layer circuitry of the electrical network ports connecting sensors (PHY voltage domain). When it is predicted that there will be no video transmission service within the next 30 minutes, the main processor sends a command to shut down the 3.3V output of the optical module voltage domain, thereby reducing the optical module's power consumption to 0 watts (W).

[0040] The main processor is configured to execute a three-level collaborative energy-saving management process, which includes a business forecasting and learning phase, an intelligent mode decision-making phase, and a hardware collaborative execution phase.

[0041] Specifically, in the business forecasting and learning phase, the main processor analyzes historical traffic data, trains a time series forecasting model based on Long Short-Term Memory (LSTM) networks, and combines it with real-time meteorological data to predict future business load. In the intelligent mode decision-making phase, the main processor, based on the forecast results, current time, and system instructions, decides whether the gateway should enter one of three operating modes: "Active," "Intelligent Standby," or "Deep Sleep." In the hardware collaborative execution phase, the main processor translates the mode decision into specific hardware control instructions and sends them to the heterogeneous computing unit itself, the intelligent network switching unit, and the intelligent power management unit for execution. For example, during the quiet period from 0:00 to 6:00 AM, the decision is made to enter "Deep Sleep Mode." In the execution phase, the main processor adjusts its core voltage from 1.2V to 0.9V and its frequency from 240MHz to 50MHz (chip-level collaboration); it controls the switching chip to shut down the physical layer power supply to all electrical network ports not connected to sensors (link-level collaboration); simultaneously, it instructs the intelligent power management unit to shut down the power supply output of the optical module voltage domain, retaining only the minimum sustaining power supply of the core voltage domain (system-level collaboration).

[0042] The intelligent power management unit is configured to perform independent switching control or dynamic voltage regulation on multiple independent power supply outputs in response to power management commands from the heterogeneous computing unit.

[0043] Independent switch control refers to the ability to individually connect or disconnect the power supply to a specific hardware module or circuit domain; dynamic voltage regulation refers to adjusting the power supply voltage value in real time according to the load conditions.

[0044] Specifically, the power management command sent by the main processor includes the target power output terminal number and the desired operation (such as "on", "off", or "set voltage value XX volts"). After parsing the command, the intelligent power management unit executes the corresponding operation through its internal switch array and DC-DC converter, and may feed back the execution status to the main processor. For example, when the main processor determines based on a predictive model that video data transmission will occur in 5 seconds, it will send a command to the intelligent power management unit in advance, requesting that the output voltage of the "optical module voltage domain" be powered up from 0V to 3.3V. Upon receiving the command, the intelligent power management unit completes the power-up and stabilization process of this voltage domain within 5 milliseconds (ms), ensuring that the optical module is ready before the data arrives, achieving a balance between "zero wait" for services and "zero waste" of energy consumption.

[0045] In another embodiment, such as Figure 1This illustration showcases a specific fiber optic IoT gateway hardware architecture and its energy and data flow coordination diagram for a field power transmission line monitoring scenario. Using a monitoring point located on a power transmission tower in a mountainous area as an application background, this example fully demonstrates the hardware foundation from solar power input to fiber optic data uplink, and deeply integrates three-level collaborative energy-saving management. In the illustrated architecture, the power supply and energy management domains constitute the system's energy base. The solar panels (18V open-circuit voltage, 5A short-circuit current) serve as the primary energy source, achieving maximum power point tracking via a high-efficiency MPPT charging management module (such as TI's BQ24650) to charge the downstream system and backup lithium batteries. The intelligent power management unit (PMIC) (such as the TPS65263) receives instructions from the main control CPU and generates and manages four key independent power outputs: an interface voltage domain (3.3V) for powering sensors and interfaces, a PHY voltage domain (1.0V / 2.5V switchable) for powering network physical layer chips, an SFP voltage domain (3.3V, managed by a controllable switch) for powering optical modules, and the most crucial core voltage domain (0.9V~1.2V adjustable). This PMIC directly responds to the CPU's dynamic voltage and frequency adjustment and power gating control instructions, realizing system-level time-sharing and zone-based power supply scheduling. The heterogeneous computing domain is the intelligent decision-making and control center. The main control CPU uses Huada Semiconductor's HC32F4A0 (based on an ARM Cortex-M4 core, operating at 240MHz), which integrates a hardware artificial intelligence accelerator, constituting the "main processor" and "artificial intelligence coprocessor" as described in the claims. This CPU is configured to execute the aforementioned three-level collaborative energy-saving management process. For example, during the business forecasting and learning phase, the CPU runs lightweight MobilNet or 1D-CNN models to perform localized analysis of sensor data. During the decision-making and execution phase, the CPU implements chip-level DVFS (Dynamic Voltage and Frequency Scaling), link-level gating, and system-level power scheduling by configuring the PMIC and switching chip. The intelligent interface domain enables reliable data aggregation and transmission. The core is Motorcom's YT9218 series managed gigabit industrial Ethernet switching chip. The CPU is coupled to the MAC layer of the switching chip via the RMII / SGMII interface and configured and managed through the SMI interface (MDIO / MDC bus). This switching chip provides multiple electrical ports for connecting field sensors and connects to two SFP+ optical module interfaces via an internal high-speed SerDes (serializer / deserializer) channel. After the optical modules are inserted, the existing OPGW fiber (phase A cores 1 and 2) is used to construct the uplink primary channel and protection link, achieving 1+1 redundant communication. The switching chip supports Energy Efficient Ethernet (EEE) and port-level PHY power gating, making it a key hardware component for implementing link-level collaborative energy saving.The specific workflow of this embodiment is as follows: During the day, with sufficient solar power, the gateway is in active mode. The CPU dynamically adjusts its core voltage to 1.2V based on the image recognition task load. When the nighttime period is predicted to be a quiet period for business operations, the CPU decides to enter deep sleep mode. During the execution phase, the CPU first reduces its core voltage to 0.9V and its frequency to 50MHz (chip-level coordination); then, it shuts down the PHY power supply to all idle sensor ports via the SMI bus (link-level coordination); finally, it sequentially shuts down the power supply to non-critical sensors in the SFP voltage domain (disconnecting the optical module power supply) and the interface voltage domain via the PMIC, retaining only the minimum maintenance voltage in the core voltage domain (system-level coordination). The overall gateway power consumption drops from approximately 1.8W during the day to 0.15W at night. When the vibration sensor triggers a conductor galloping warning in the early morning, the gateway is urgently woken up, restores full functionality within milliseconds, and reliably uploads high-priority warning data via the OPGW optical path.

[0046] Therefore, this embodiment is achieved through Figure 1 The specific hardware architecture and interconnection relationships shown clearly and concretely reveal how each unit of the fiber optic IoT gateway described in the claims of this invention is specifically implemented, how it is connected, and how it works collaboratively in real-world scenarios to achieve the goals of ultra-low power consumption and high-reliability communication. This provides intuitive and powerful support for examiners and the public to understand this invention.

[0047] Therefore, by utilizing the aforementioned fiber optic IoT gateway, users can achieve the transmission of monitoring data back through a highly reliable wired fiber optic network in complex field environments. Based on intelligent prediction of business load, the gateway's internal computing, communication, and power supply modules can be managed in a three-level collaborative and refined manner at the chip, link, and system levels. This reduces the average power consumption of the equipment during long-term operation, alleviates the pressure on the solar power supply system, and improves the overall deployment feasibility and operational stability of the online monitoring system for power transmission lines, while ensuring the reliability of monitoring functions and communication.

[0048] Next, according to a second aspect, the present invention provides an energy efficiency control method based on a fiber optic IoT gateway. This method is based on the fiber optic IoT gateway described in the above embodiments and applied to a main processor, which can be integrated onto the hardware board of the fiber optic IoT gateway. The main processor can execute program instructions stored in its internal or external memory to perform analysis and learning of historical and real-time service data, intelligent decision-making regarding gateway operating modes, and coordinated control of chip-level, link-level, and system-level energy-saving operations. Specifically, the hardware device of the main processor includes, but is not limited to, a high-performance microcontroller based on a specific instruction set architecture such as the ARM Cortex-M series or RISC-V.

[0049] like Figure 2 As shown, the method may include: Step S110: Perform time-series analysis and feature extraction on historical business data and real-time sensor data, and train and deploy the business volume prediction model.

[0050] Historical business data refers to the traffic and timestamp information of each sensor data upload recorded by the fiber optic IoT gateway in the past time period; real-time sensor data refers to the raw measurement values ​​collected from the sensors (such as micro-meteorological, image, and vibration sensors) connected to the fiber optic IoT gateway at the current moment; and the traffic prediction model refers to a mathematical model based on machine learning algorithms used to predict the gateway data traffic load in a specific future time period.

[0051] Specifically, the system can use a data analysis engine deployed on the main processor to aggregate and feature-engineer historical business data by hour, day, week, etc., extract periodic patterns and trends, and use the LSTM algorithm to train a time-business volume correlation model. At the same time, the engine combines real-time acquired meteorological sensor data such as temperature, humidity, wind speed, and light intensity, and uses the Gradient Boosting Decision Tree (GBDT) algorithm to train a weather-business volume prediction model. The outputs of the two models are then weighted and fused to form the final traffic prediction value.

[0052] For example, the system trains an LSTM model using historical data from the past 30 days, with a time granularity of 1 hour. This model uses 24 hours and 7 days of the week as periodic features and takes the traffic flow sequence of the previous 6 hours as input to predict the total data flow for the next 30 minutes, with a prediction error of less than 5 kilobits per second (kbps). When the real-time wind speed exceeds 10 meters per second (m / s), the weather-traffic volume prediction model outputs a traffic gain coefficient (e.g., 1.5) to indicate a potential increase in video data volume due to conductor galloping monitoring.

[0053] Step S120: Based on the collected environmental status data, combined with preset priority rules and judgment conditions, generate control commands.

[0054] In this step, environmental status data specifically refers to high-priority event data used to trigger specific control actions, such as conductor galloping events determined by vibration sensors or smoke and flame events identified by wildfire monitoring cameras; the preset priority rules refer to the response levels set for different types of events, and high-priority events can interrupt or adjust the established energy-saving scheduling plan; the judgment conditions refer to the quantification threshold or logical criteria for triggering control commands.

[0055] Specifically, the system can analyze video streams in real time and identify features such as smoke and flames using a lightweight convolutional neural network (CNN) model running on the AI ​​coprocessor; or it can process the spectral data from vibration sensors through the main processor to determine whether wire swaying has occurred. Once a high-priority event is identified, the system will generate control commands based on rules, such as "immediately switch to active mode and increase the image acquisition frame rate to 30 frames per second (fps)".

[0056] For example, the system deploys a 1D CNN vibration analysis model to monitor conductor acceleration data in real time. When a low-frequency, high-amplitude oscillation with a frequency range of 0.1Hz to 1Hz is detected and its duration exceeds 10 seconds, it is determined to be a conductor galloping event. The system then generates a high-priority control command, the content of which is: (1) Control the gateway to immediately switch from smart standby mode to active mode; (2) Send a command to the connected video surveillance sensor to increase its acquisition frame rate from 1fps to 10fps in order to record the dance pattern in detail.

[0057] Step S130: Based on the forecast output of the traffic volume prediction model, determine the working mode of the fiber optic IoT gateway.

[0058] The operating modes include activation mode, intelligent standby mode, and deep sleep mode. Activation mode means that all functional units of the gateway are in full-speed operation to process high-frequency service data. Intelligent standby mode means that the gateway maintains only core communication and monitoring functions based on prediction results, while non-critical modules are in a low-power state and can be precisely woken up before service arrives. Deep sleep mode means that during preset periods without service, the gateway shuts down most functions and maintains only the most basic timed wake-up and network heartbeat keep-alive capabilities with extremely low power consumption.

[0059] Specifically, the system can use the decision logic unit built into the main processor to compare multiple inputs, such as the traffic volume prediction value obtained in step S110, the current system time, and whether there are high-priority instructions, with preset threshold rules to determine the appropriate working mode. The decision logic follows the rule of "high-priority events > predicted traffic volume > silent period".

[0060] For example, the system is configured with the following rules: if the predicted average traffic for the next 30 minutes is consistently higher than 50kbps, or if a high-priority event occurs, the system will enter active mode; if the predicted traffic is lower than 50kbps but higher than 1kbps, and the current time is not within a quiet period (e.g., 0:00 to 6:00), the system will enter intelligent standby mode; if the current time is within a quiet period and there are no high-priority events, the system will enter deep sleep mode regardless of the predicted traffic. The system automatically decides to enter deep sleep mode at 23:55 daily based on these time rules. Alternatively, the main processor's built-in decision module maintains a priority lookup table: if a high-priority event exists, the system will force entry into active mode; otherwise, it will determine if the current time is within a quiet period, and if so, enter deep sleep mode; otherwise, it will compare the predicted traffic volume with a threshold to determine whether to enter active mode or intelligent standby mode.

[0061] Step S140: Perform hardware coordination steps, which include chip-level coordination steps, link-level coordination steps, and system-level coordination steps.

[0062] The chip-level coordination steps include: adjusting the main processor's operating voltage and core frequency using a dynamic voltage and frequency adjustment strategy based on the real-time load indicators of the fiber optic IoT gateway, and offloading computationally intensive data processing tasks to the AI ​​coprocessor. The link-level coordination steps include: controlling the intelligent network switching unit to enable the energy-efficient Ethernet protocol at the data link layer, and performing power gating operations on the physical layer circuits of designated idle network ports on the intelligent network switching unit according to the current decision-making operating mode. The system-level coordination steps include: sending power management commands to the intelligent power management unit to perform time-sharing and zone-based power supply scheduling for multiple independent power domains corresponding to multiple independent power outputs of the intelligent power management unit.

[0063] Specifically, the system can use the main processor's co-schedule to parse the working mode decided in step S130 into a set of low-level hardware operation instructions that can be executed concurrently or sequentially, and send them to the corresponding hardware units. Chip-level operations are implemented by writing to the processor performance status (P-State) register and the task scheduler; link-level operations are implemented by configuring the registers of the switching chip through the Serial Management Interface (SMI) bus; system-level operations are implemented by sending commands to the power management integrated circuit through the Inter-Integrated Circuit (I2C) bus.

[0064] For example, when the system decides to enter intelligent standby mode: During the chip-level coordination process, the main processor detected that its CPU utilization was below 20%, and subsequently adjusted the core voltage from 1.2V to 0.9V and the core frequency from 240MHz to 50MHz. Simultaneously, it processed the newly arrived frame... The pixel-wise JPEG (Joint Photographic Experts Group, a widely used lossy compression standard for still images) image recognition task is scheduled to be processed by the artificial intelligence coprocessor via Direct Memory Access (DMA).

[0065] In the link-level coordination step, the main processor sends a command to the intelligent network switching unit via the SMI bus to enable the energy-efficient Ethernet function of all active ports. Subsequently, based on the prediction that the port connected to the micro-weather sensor will have no data for the next 2 hours, the main processor sends a command to cut off the power supply to the physical layer transmit and receive circuits of that port.

[0066] In the system-level coordination step, the main processor sends instructions to the intelligent power management unit via the I2C bus: (1) Maintain power supply to the core voltage domain (0.9V) and the physical layer voltage domain; (2) Turn off the power supply to the voltage domain of the pluggable optical module (because there is no data transmission requirement); (3) Maintain the power supply to the vibration sensor in the interface voltage domain (for event monitoring), but disconnect the power supply to the HD camera interface. Through the above three-level coordination, the measured power consumption of the gateway in intelligent standby mode is reduced from 1.5W in active mode to 0.6W.

[0067] In another embodiment, such as Figure 3 This illustration showcases a network architecture and service flow diagram of a multi-point collaborative intelligent monitoring system based on the fiber optic IoT gateway of this invention, applied to ultra-high voltage transmission line corridors. This embodiment fully presents a three-tiered distributed system from the "regional control center" to the "metropolitan area edge" and then to the "transmission tower" site, and clarifies the core hub position of the fiber optic IoT gateway (i.e., the "tower-level gateway" in the diagram) within the entire system and the rich monitoring services it carries.

[0068] In the illustrated architecture, the cloud control and core network layer serves as the decision-making and command center. The "Regional Control Center" is responsible for aggregating, storing, analyzing, and providing a panoramic view of the monitoring data for the entire line, and issuing control strategies. The "Core Communication Network" is typically built based on power SDH / MSTP or SPN, providing highly reliable, high-bandwidth backbone communication capabilities. The "QKD / MDS Optical Transceiver" represents an optical transmission device that integrates quantum key distribution and multi-service access, providing high physical layer security for the system. After protocol conversion, preliminary edge computing screening, and routing distribution by the "Metropolitan Area Edge Gateway," the data enters the dedicated power communication channel. The field access and sensing layer is the cornerstone of the system's intelligent monitoring, consisting of multiple fiber optic IoT gateways (i.e., "Tower-level Gateways #1, #2, and #3") deployed on different towers and based on the present invention. Each gateway connects to the "Metropolitan Area Edge Gateway" in a point-to-point or serial manner via the "Fiber Optic Access Unit" described in the claims, utilizing redundant fiber cores in the existing "OPGW / ADSS Fiber," thus constructing a highly deterministic and reliable wired backhaul network. Each gateway can flexibly connect to various monitoring sensors deployed on and around the tower through multiple electrical network ports provided by its "intelligent network switching unit", realizing plug-and-play and flexible expansion of monitoring terminals.

[0069] Among them, SDH (Synchronous Digital Hierarchy) is a standardized digital signal transmission technology for optical fiber communication; MSTP (Multi-Service Transport Platform) is an evolution technology based on SDH, which enhances the effective carrying and switching capabilities of data services such as Ethernet on the basis of traditional time-division multiplexing; SPN (Slicing Packet Network) is a new generation of packet transport network technology for the new era of services such as 5G, which integrates the flexibility of packet switching and the reliability of hard-pipe isolation; QKD (Quantum Key Distribution) is a technology that realizes secure distribution of cryptographic keys based on the principles of quantum mechanics, which can provide encryption protection for information theory security of communication systems; MDS (Multi-Service) here specifically refers to the ability of optical transmission equipment to simultaneously carry and switch multiple types of services (such as TDM, Ethernet, storage, etc.).

[0070] This embodiment specifically demonstrates the differentiated and professional monitoring configurations of three tower gateways, revealing the broad adaptability of the gateway of this invention: Tower-level gateway #1: Primarily connects to "video surveillance" terminals, utilizing the "artificial intelligence coprocessor" within the gateway to perform real-time analysis of video streams, enabling visual intelligent identification of foreign objects in wires, bird nests on the tower, and personnel intrusion, while only uploading alarm information and key images, greatly saving uplink bandwidth.

[0071] Tower-level gateway #2: Integrates with multiple sensors, including those for "tower tilt," "vibration monitoring," and "video surveillance." The gateway's "main processor" can perform real-time fusion analysis of tilt and vibration data. When an anomaly is detected (such as slight movement of the tower base caused by precursors to a landslide), it can immediately trigger the video sensors to turn to preset positions for capture and confirmation, and report the data through the "high-priority service response channel," demonstrating local multi-service collaboration and intelligent linkage capabilities.

[0072] Tower-level gateway #3: Connects to advanced diagnostic equipment such as "wildfire monitoring" (multispectral or infrared) and "distributed fault diagnosis". For such large-volume, complex analysis and calculation tasks, the gateway's "heterogeneous computing unit" plays a crucial role. It can perform initial fire point identification on multispectral images locally, or preprocess traveling wave fault data, effectively reducing the pressure on the backend analysis and improving the real-time performance of early warnings.

[0073] therefore, Figure 3 The system architecture embodiment shown demonstrates that the fiber optic IoT gateway provided by this invention is not an isolated device, but rather a standardized and intelligent edge node constituting a modern intelligent power transmission line monitoring network. It solves reliable communication issues through unified wired fiber optic access, achieves edge intelligence through heterogeneous computing, and meets stringent power supply requirements in the field through three-level collaborative energy-saving management. Ultimately, it successfully integrates diverse monitoring terminals, complex local computing, and highly reliable uplink transmission, providing a crucial hardware platform and implementation plan for building a new generation of intelligent power transmission line inspection system characterized by "panoramic perception, intelligent handling, and efficient operation and maintenance."

[0074] In another embodiment, such as Figure 4This diagram illustrates the business logic and data flow of a fire protection and security system for substations or large power facilities, based on the three-level collaborative energy-saving management concept of this invention. Using a "three-level fire protection and energy-saving management system" as a specific application context, the diagram intuitively reveals the entire closed-loop process from data perception, intelligent analysis, pattern decision-making to execution control. It is a concrete mapping and instantiation of the three stages described in this invention—"business prediction and learning, intelligent pattern decision-making, and hardware collaborative execution"—in the vertical scenario of fire protection and energy saving. In the illustrated process, the system first enters the data perception and feature extraction stage. This stage corresponds to the "historical flowchart learning," "time flow analysis," and "weather system monitoring" modules in the diagram. This is equivalent to the "business prediction and learning stage" in this invention. Specifically, the system collects historical fire equipment inspection data and alarm logs ("historical flowchart learning"), analyzes the temporal patterns of equipment energy consumption and event occurrences ("time flow analysis"), and integrates real-time meteorological data ("weather system monitoring") to jointly construct multi-dimensional features for predicting fire risks and equipment operating loads. For example, the system might learn that during a hot, dry summer afternoon, the fire risk level in a certain area increases, and the expected activation probability of equipment such as smoke exhaust fans also increases. Subsequently, the system enters the intelligent analysis and mode decision-making stage, corresponding to the "Event-Based Processing Assessment," "Mode Flowchart," and "Mode Decision Tree" modules in the diagram. This embodies the "Intelligent Mode Decision-Making Stage" of this invention. In this stage, the system performs "Event-Based Processing Assessment" on real-time sensed data such as smoke concentration, temperature, and video images to determine whether a real fire has occurred or a false alarm. Simultaneously, based on the features extracted and prediction results from the first stage, the system calls upon pre-set "Mode Flowcharts" and "Mode Decision Trees" to determine the optimal system operating mode. For example, the decision tree might decide that the system enters an "Enhanced Alert Mode" based on conditions such as "Risk Level = High," "Weather = Dry," and "Wind Speed ​​= Low." This mode corresponds to the higher-load "Activation Mode" in this invention and includes pre-set equipment activation plans. Finally, the system enters the instruction generation and collaborative execution stage, corresponding to the "Fire Protection System Flowchart," "System Flowchart," and other modules on the right side of the diagram, along with arrows pointing to the final "target." This corresponds to the "Hardware Collaborative Execution Stage" of this invention. At this stage, the system translates pattern decisions into specific, executable instruction sequences. It controls terminal equipment such as smoke exhaust valves, fire pumps, and emergency lighting through a "fire protection system flowchart," and schedules the fiber optic IoT gateway of this invention through a "system flowchart" to coordinate the control of related communication links, computing units, and power supply modules. For example, in "enhanced alert mode," the system instructs the fiber optic IoT gateway to maintain its operating mode in "active mode," ensuring the video stream analysis AI coprocessor runs at full speed and maintaining uninterrupted physical layer power supply to all fire sensor backhaul links. Simultaneously, it implements "time-sharing and zone-based" scheduling of power supply for general lighting in non-critical areas to save energy.

[0075] therefore, Figure 4 The illustrated business process diagram demonstrates that the three-level collaborative energy-saving management architecture proposed in this invention possesses high universality and scalability. It is not only applicable to transmission line monitoring, but its core "prediction-decision-execution" logic can also empower the intelligent and energy-saving transformation of other key power system scenarios, such as substation fire protection. Through a unified framework, it achieves a closed loop from data to knowledge, and then from knowledge to automated action, fully validating the practical value and broad application prospects of the methodology of this invention.

[0076] In summary, based on the above implementation method, the system first achieves system-level energy efficiency optimization through the collaborative architecture of four core hardware units: a heterogeneous computing unit, an intelligent network switching unit, a fiber optic access unit, and an intelligent power management unit, and a three-level collaborative energy-saving management process executed by the main processor. Specifically, the heterogeneous computing unit integrates the main processor and an artificial intelligence coprocessor, providing a heterogeneous computing foundation for intelligent energy efficiency management; the intelligent network switching unit is coupled to the computing unit via a high-speed data bus and has both electrical and fiber optic ports, forming a flexible internal and external data exchange hub; the fiber optic access unit connects to the fiber optic port via pluggable optical modules, establishing a deterministic wired communication backbone based on power fiber optic cables; and the intelligent power management unit receives instructions through a digitally programmable power management integrated circuit and performs fine-grained control over multiple independent power outputs, forming the execution end of energy consumption regulation. The three-level collaborative energy-saving management process covers three stages: business prediction and learning, intelligent mode decision-making, and hardware collaborative execution, linking intelligent decision-making and hardware control in a closed loop.

[0077] Throughout the process, this embodiment addresses the comprehensive technical problems described in the background art, including insufficient reliability due to reliance on wireless communication, increased power supply pressure due to excessive power consumption of access devices, and low integration and lack of intelligence due to functional separation. By introducing a fiber optic access unit based on pluggable optical modules, it directly utilizes redundant fiber cores in existing overhead ground wire composite optical cables (OPGW) or all-dielectric self-supporting optical cables (ADSS) to construct communication links, thus physically avoiding terrain and weather interference and solving the core bottleneck of high-reliability communication. Regarding the high power consumption of the equipment, an intelligent power management unit performs independent switching control or dynamic voltage regulation on multiple independent power output terminals. Combined with the decision-making of a three-level collaborative energy-saving management process, it achieves refined management of power supply heads by region and time, providing a hardware foundation for significantly reducing the average power consumption of the system. Addressing the issues of functional dispersion and lack of intelligence, it deeply integrates four major functions—heterogeneous computing, network switching, fiber optic access, and power management—into a single gateway device. The main processor uniformly executes a collaborative management process of prediction, decision-making, and execution, improving integration, reducing maintenance complexity, and endowing the gateway with the intelligent capability of adaptive adjustment based on service status. Therefore, the technical solution of this embodiment, through the collaborative design of the above-mentioned hardware architecture and management process, simultaneously addresses the comprehensive technical problem in the background technology that the existing technology cannot simultaneously meet the three requirements of high-reliability communication, ultra-low power consumption operation and highly intelligent integration, by simultaneously exerting efforts from three levels: communication medium, energy consumption management and system integration. This provides a key equipment foundation for building an all-weather, highly reliable and intelligent transmission line monitoring system.

[0078] In some embodiments, time-series analysis and feature extraction are performed on historical business data and real-time sensor data to train and deploy a business volume prediction model, including: Build and train a time series forecasting model to learn and output time-volume correlation forecasting results based on the periodic patterns and trend characteristics of historical business data.

[0079] Among them, time series forecasting models are statistical or machine learning models specifically designed to process data sequences arranged in chronological order and predict values ​​at future points in time.

[0080] Specifically, the system can construct this model using a Gated Recurrent Unit (GRU) network. The model's input is a fixed-length historical traffic sequence, such as 24 data points from the past 24 hours, spaced at 1-hour intervals. Each data point represents the total data transmission volume (in kbps) within that hour. The model's output is a predicted traffic volume for one or more future time points. During training, Mean Squared Error (MSE) is used as the loss function, and the model parameters are optimized using backpropagation. The goal is to enable the model to capture the patterns and long-term trends of traffic changes over daily and weekly periods. For example, the system collects historical hourly traffic data from a gateway over the past 90 days. The first 60 days of data are used as the training set, and the last 30 days as the test set. The constructed GRU model contains two hidden layers, each with 128 neurons. The traffic sequence from the past 24 hours is used... To predict traffic flow in the next 6 hours After training, the model achieved a mean absolute percentage error (MAPE) of 8.5% in predicting traffic flow for the next hour on the test set.

[0081] Construct and train a multi-feature fusion prediction model to learn and output weather-business volume correlation prediction results based on meteorological feature parameters in real-time sensor data and historical business data.

[0082] Among them, the multi-feature fusion prediction model is a machine learning model that can accept and jointly process features from different data sources with different scales, and is used to explore the complex nonlinear relationship between multiple factors and the target variable.

[0083] Specifically, the system can use the eXtreme Gradient Boosting (XGBoost) algorithm to build the model. This means the system trains the model offline using historical data via the main processor, deploys the trained lightweight model in the gateway storage, and loads it at runtime for online prediction. The model's input feature vector includes: real-time meteorological sensor data (such as current temperature, humidity, wind speed, and light intensity), time features (such as hour, day of the week, and holiday status), and statistical features of recent historical traffic volume (such as the average of the past 3 hours and the maximum of the past 24 hours). The model output is a prediction of the deviation coefficient or increment of traffic volume relative to the baseline level in the next short period (such as the next hour). By ranking the importance of features, the influence of different meteorological factors on traffic volume can be analyzed. For example, the input feature vector has 15 dimensions, including: temperature (… The data included relative humidity (%), wind speed (m / s), daylight intensity (lux), hourly rate (0-23), day of the week (0-6), and average traffic flow over the past 3 hours. After model training, feature importance analysis showed that "wind speed" and "daylight intensity" were the two most critical meteorological features affecting traffic volume (especially video traffic). When wind speed exceeded 8 m / s, the model typically predicted an increase in traffic volume of 15% to 40%, which aligns with the reality that strong winds can trigger conductor galloping monitoring, thereby increasing video data uploads.

[0084] Configure high-priority business response channels for preset emergency event types, wherein the emergency event is generated by real-time sensor data flow through real-time identification model determination, or triggered by the prediction output of at least one of the time series prediction model and multi-feature fusion prediction model.

[0085] The high-priority service response channel is a predefined, low-latency processing link within the system. When specific conditions are met, service flows and corresponding control commands will be prioritized for processing through this channel to ensure rapid response to emergencies. The preset emergency types include, but are not limited to, conductor galloping, icing, wildfires, and external damage. Furthermore, at the software level, the high-priority service response channel is manifested as a task or interrupt service routine with the highest scheduling priority; at the hardware level, it can be configured to directly connect the corresponding sensor interrupt signal to the high-priority interrupt pin of the main processor. When the real-time identification model or prediction model triggers an event, this channel can interrupt the current energy-saving management process, directly schedule resources, and ensure that the response latency is below the order of hundreds of milliseconds.

[0086] Specifically, the system maintains an "event-action" mapping table at the software level. When a real-time recognition model (such as a lightweight YOLO model running on an AI coprocessor) detects smoke or flames in the video stream and generates a "suspected wildfire event" alarm, the event is immediately sent to a high-priority response channel. This channel executes multiple actions in parallel: (1) Send an interrupt to the main processor to force the gateway to switch to active mode; (2) Increase the frequency of data acquisition and reporting from event-related sensors; (3) Send the highest-level alarm information to the remote monitoring center. In addition, if both the time series prediction model and the multi-feature fusion prediction model output extremely high traffic prediction values ​​(e.g., exceeding the historical 99th percentile), this "prediction anomaly" itself can also trigger a high-priority channel as a synthetic event to initiate preventive resource allocation. For example, the system presets a "wire dancing" event. The raw data from the vibration sensor is analyzed in real time by a 1D convolutional neural network (1D-CNN) on the artificial intelligence coprocessor. If a feature with a frequency between 0.5Hz and 1.5Hz and an amplitude exceeding 0.5 times the gravitational acceleration (g) is identified and lasts for more than 5 seconds, the event is determined to have occurred. This event triggers a high-priority response channel, which completes the following operations within 100ms: wakes up the network port connected to the dancing camera from the energy-efficient Ethernet low-power state; instructs the camera to switch to high-speed capture mode (10 frames per second); and sends the initial dancing data packet with an "emergency" tag to the backend via the fiber optic link.

[0087] Therefore, based on the above implementation method, the system can comprehensively utilize historical patterns, real-time environmental factors, and intelligent event detection to construct a multi-layered, adaptive business volume perception and prediction system. This not only provides accurate decision-making basis for subsequent energy efficiency regulation but also ensures that the monitoring system can quickly switch to a high-protection working state in the event of a sudden emergency, thereby optimizing the overall balance between energy consumption and performance and realizing intelligent and forward-looking management of the power transmission line monitoring network.

[0088] In some embodiments, the operating mode of the fiber optic IoT gateway is determined based on the forecast output of the traffic volume forecasting model, including: When the traffic forecasting model's forecast output indicates that future traffic will continue to be lower than the preset low traffic threshold, and the fiber optic IoT gateway is currently in active mode, the decision-making fiber optic IoT gateway's operating mode switches from active mode to intelligent standby mode.

[0089] "Continuously below" means that the output value of the traffic volume prediction model for multiple consecutive prediction periods is lower than the low traffic threshold. This indicates that the traffic has entered and may remain in a low load state for a period of time, rather than a momentary fluctuation.

[0090] Specifically, the system can set a continuous judgment window, such as the next three prediction periods (each period being 5 minutes). The main processor compares the traffic prediction model's predictions for these three periods. Only when all predictions are below a low traffic threshold (e.g., 10kbps), and the gateway is currently in active mode, will a switching decision to intelligent standby mode be triggered. This mechanism avoids frequent mode switching caused by instantaneous traffic fluctuations. For example, setting the low traffic threshold to 10kbps, the traffic prediction model outputs a sequence of predicted values ​​for the next 15 minutes (3 periods) as follows: Since all predicted values ​​are below 10kbps and the gateway is currently in active mode, the system immediately switches from active mode to smart standby mode.

[0091] When the system clock enters the preset silent period, the decision fiber optic IoT gateway enters deep sleep mode.

[0092] Among them, the preset silent period refers to a period of time with extremely low business volume or no need for active monitoring, which is pre-set according to historical business patterns, such as the late night period.

[0093] Specifically, the system defines one or more silent period intervals in the configuration file. The real-time clock module built into the main processor continuously provides the current time. When the clock time falls within any of the preset silent period intervals and no high-priority events are activated, the system will decide to enter deep sleep mode regardless of the current traffic forecast. This is a scheduling strategy based on strong time rules that guarantees the lowest power consumption. For example, the preset silent period is from 1:00 AM to 5:00 AM daily. When the system clock reaches 1:00:00 AM, the main processor immediately decides to switch the gateway's operating mode to deep sleep mode, regardless of whether the gateway is currently in active mode or intelligent standby mode.

[0094] In deep sleep mode, the working mode of the fiber optic IoT switch is woken up from deep sleep mode to intelligent standby mode at preset intervals, and network heartbeat messages are sent in intelligent standby mode.

[0095] The preset period refers to the fixed time interval between two wake-up operations in deep sleep mode; the network heartbeat message is a short data packet used to announce the online status of the device to the network peer and maintain the network connection from being interrupted by timeout.

[0096] Specifically, in deep sleep mode, most of the main processor's circuitry is shut down, but a separate low-power timer remains active. This timer generates a wake-up interrupt every preset period (e.g., 300 seconds), waking the main processor from deep sleep. Upon waking, the main processor first restores its operating mode to intelligent standby mode, then controls the network stack to generate a short connection-keeping message (e.g., a TCP-based keepalive message or a custom heartbeat packet based on the U protocol), sends it through the intelligent network switching unit, and immediately re-enters deep sleep mode, awaiting the next scheduled wake-up. For example, the wake-up period can be set to 300 seconds (5 minutes). In deep sleep mode, the gateway triggers a wake-up every 300 seconds using its low-power timer. After waking, the main processor returns to intelligent standby mode within 50ms, then constructs and sends a 64-byte heartbeat message within 100ms, and reconfigures and re-enters deep sleep mode within 200ms. The average power consumption during this process is only a brief increase from the base power consumption of deep sleep mode.

[0097] When the traffic forecasting model indicates that traffic data has arrived, or when a wake-up command is received from the remote network management system, the working mode of the decision fiber optic IoT gateway switches from intelligent standby mode to active mode while in intelligent standby mode.

[0098] "Business data arrival" means that the output value of the business volume prediction model for the next one or several prediction periods exceeds the trigger threshold for switching from smart standby mode to active mode (this threshold may be higher than the low traffic threshold for entering smart standby mode); the wake-up command from the remote network management system is a data packet received over the network that carries a specific wake-up command code.

[0099] Specifically, in intelligent standby mode, the main processor periodically (e.g., every 10 seconds) calls the traffic prediction model to forecast short-term traffic. If the predicted value exceeds a preset activation threshold (e.g., 30kbps), the system immediately switches to active mode to prepare resources in advance. Simultaneously, the main processor listens to the network port. If it receives a remote wake-up command with a matching format (e.g., a UDP packet containing a specific "Magic Packet" or command code), it forces a switch to active mode regardless of the current predicted traffic volume to respond to remote emergency control or debugging needs. For example, the activation threshold is set to 30kbps. In intelligent standby mode, the traffic prediction model predicts that video data will begin uploading within the next 2 minutes, with a predicted traffic of 45kbps. Since 45kbps > 30kbps, the system immediately switches from intelligent standby mode to active mode and pre-activates the optical module power supply. Meanwhile, if the remote network management center sends a predefined wake-up command (hexadecimal string 0xAA55FF00) to the gateway's IP address and port for emergency inspection, the gateway will immediately decide to switch to active mode after receiving the command in smart standby mode.

[0100] Therefore, according to the above implementation method, the system can automatically, reliably, and with low latency transition between different working modes (active, intelligent standby, deep sleep) based on precise prediction rules, a defined schedule, and external instructions. This multi-condition triggered mode decision-making mechanism ensures that the gateway can operate in an extremely low-power state during periods of no or low service, while also ensuring that it can be promptly and accurately woken up to full-function working state before service arrives or upon receiving instructions. Thus, while meeting real-time monitoring requirements, it maximizes the optimization of the system's overall energy efficiency.

[0101] In some embodiments, the step of adjusting the operating voltage and core frequency of the main processor using a dynamic voltage-frequency adjustment strategy based on the real-time load indicators of the fiber optic IoT gateway includes: Periodically sample the CPU utilization of the main processor.

[0102] Among them, CPU utilization refers to the percentage of the main processor's non-idle time in a sampling period, which is a key indicator for measuring the computing load.

[0103] Specifically, the system reads the active time slice count of the main processor at fixed time intervals (sampling period) through the performance monitoring interface of the embedded real-time operating system. The utilization rate is calculated using the following formula: Utilization Rate Active time slice count / Total time slice count The main processor stores this periodically sampled utilization value in a circular buffer for use by subsequent decision-making logic. For example, the system is set to a sampling period of 1 second. At the end of each 1-second period, the operating system reports that the main processor cores were in a running state (not idle) for 300ms in the past second. The CPU utilization for that period is then calculated. This 30% figure was recorded.

[0104] When the CPU utilization rate is consistently higher than the preset high load threshold, the main processor's operating performance state will be configured to the highest performance state.

[0105] "Sustained above" means that the utilization value is higher than the high load threshold for multiple consecutive sampling periods, indicating that the computing load has entered and may remain in a high-intensity state. The highest performance state refers to the highest combination of operating voltage and core frequency of the main processor within the allowable reliable operating range, designed to provide the strongest instantaneous computing power.

[0106] Specifically, the system sets a continuous judgment window, such as three consecutive sampling cycles. The main processor checks the utilization values ​​of the three most recent sampling points. If all three values ​​are higher than the high load threshold (e.g., 80%), a state switch is triggered. The main processor adjusts the core operating voltage to the highest level (e.g., 1.2V) and locks the core frequency to the highest value (e.g., 240MHz) by writing to the internal performance status control register. At this time, the main processor's power consumption increases accordingly, but the computational throughput is maximized. For example, the high load threshold is set to 80%. The utilization rates of the three most recent sampling cycles are 85%, 82%, and 88%, respectively. Since the three consecutive values ​​are all greater than 80%, the main processor switches its operating state from the current balanced energy efficiency state (e.g., 1.0V, 120MHz) to the highest performance state (1.2V, 240MHz).

[0107] When the CPU utilization is below the high load threshold and remains above the preset low load threshold, the CPU's operating performance state is configured to a balanced energy efficiency state. The operating voltage and core frequency corresponding to the balanced energy efficiency state are lower than those of the highest performance state.

[0108] Among them, the balanced energy efficiency state refers to an operating state that achieves a better balance between power consumption and performance by appropriately reducing voltage and frequency while meeting the current computing performance requirements.

[0109] Specifically, the system switches to this state when it detects that the CPU utilization has decreased from a high level but has not dropped to an extremely low level. For example, when the utilization values ​​of three consecutive sampling points are all below the high load threshold (80%) but above the low load threshold (e.g., 20%), the main processor adjusts the voltage to a middle level (e.g., 1.0V) and the frequency to a middle value (e.g., 120MHz). This state is the main state for the gateway to handle regular, medium load services. For example, the low load threshold is set to 20%. After switching from the highest performance state, the system monitors the utilization rates for the next three cycles at 25%, 18%, and 30%. Since the value of 18% in the second cycle is lower than 20%, it does not meet the condition of "consistently above" the low load threshold, so the state remains unchanged. When the subsequent utilization rates are 25%, 28%, and 22%, the condition is met, and the main processor switches from the highest performance state to the balanced energy efficiency state (1.0V, 120MHz).

[0110] When the CPU utilization rate is consistently below the low load threshold, the main processor's operating performance state is configured to a low power state, where the operating voltage and core frequency are lower than those of the balanced energy efficiency state.

[0111] The low-power state refers to a state set to cope with extremely low computing loads, in which voltage and frequency are reduced to a low level to significantly save dynamic power consumption.

[0112] Specifically, the system switches to this state when it detects a very light computational load. The criterion is that the utilization value for several consecutive sampling periods (e.g., 3) is below the low-load threshold (20%). The main processor then adjusts the voltage to a low level (e.g., 0.9V) and the frequency to a low value (e.g., 50MHz). This state is suitable for handling lightweight tasks such as background tasks and network protocol stack maintenance. For example, when the system is only maintaining basic communication and not processing large amounts of data, the utilization sequence obtained by continuous sampling might be: 10%, 8%, 12%. Since all values ​​are below 20%, the main processor switches from the balanced energy efficiency state to the low-power state (0.9V, 50MHz).

[0113] When the fiber optic IoT gateway is decided to enter intelligent standby mode or deep sleep mode, the main processor's operating performance state is configured to the preset deep sleep state.

[0114] Deep sleep is the lowest voltage and frequency operating state, which is tied to the gateway's specific low-power operating mode (smart standby, deep sleep). In this state, most clock domains of the main processor are turned off or significantly reduced in frequency, maintaining only the most basic interrupt response and state retention capabilities.

[0115] Specifically, configuring deep sleep mode is an accompanying action of mode switching. Regardless of the current performance state of the main processor, once the system decides (e.g., based on time rules) to enter intelligent standby or deep sleep mode, the main processor will first adjust its voltage to the minimum holding voltage (e.g., 0.8V) and reduce its core frequency to the minimum (e.g., 10MHz) before performing other operations for mode switching, or switch to a special sleep power mode. In deep sleep mode, the main processor may be periodically woken up by a low-power timer. After waking up, it must first restore to a suitable performance state (e.g., a low-power state) before it can execute tasks. For example, when the gateway decides to enter deep sleep mode due to reaching a silent period, before performing operations such as disabling peripherals and configuring wake-up timers, the main processor will reduce its core voltage from the current 1.0V to 0.8V, its core frequency from 120MHz to 10MHz, and enter a hardware-supported "standby" state. At this time, the dynamic power consumption of the main processor core drops to the microwatt (µW) level.

[0116] Therefore, according to the above implementation method, the system can dynamically and precisely adjust the voltage and frequency of the main processor based on real-time changes in computing load, automatically switching between various predefined performance states (maximum performance, balanced energy efficiency, low power consumption, and deep sleep). This dynamic voltage and frequency adjustment strategy, which strictly matches the load, avoids energy waste by preventing the processor from operating in a high-power state under low load, and also ensures sufficient computing power during high-load bursts, thereby achieving optimal matching of computing performance and energy consumption at the chip level.

[0117] In some embodiments, controlling the intelligent network switching unit to enable the energy-efficient Ethernet protocol at the data link layer, and performing power gating operations on the physical layer circuitry of designated network ports that are idle on the intelligent network switching unit according to the currently decided operating mode, including: During the link auto-negotiation phase, if it is detected that the peer device of the intelligent network switching unit supports the energy-efficient Ethernet protocol, the energy-efficient Ethernet function negotiation is initiated, so that the link established between the intelligent network switching unit and the peer device enters a dynamic energy-saving state.

[0118] The link auto-negotiation phase refers to the process by which the communicating devices automatically exchange capability parameters to determine the optimal operating mode (such as speed and duplex mode) after the Ethernet physical link is established. Dynamic power saving mode is an operating mode based on the energy-efficient Ethernet protocol. In this mode, when there is no valid data transmission on the link, the physical layer circuitry of the link can automatically enter a low-power state and quickly recover when data transmission is needed.

[0119] Specifically, after a port connection is established, the intelligent network switching unit interacts with the peer device via link layer discovery protocol messages to confirm whether the peer supports the energy-efficient Ethernet function defined by the IEEE 802.3az standard. If both parties support it, the media access control sublayer inside the switching chip coordinates with the peer to enable a low-power idle mode. In this mode, when there is no data frame transmission, the physical layer circuitry can enter a power-saving state within microseconds, maintaining only the necessary link detection signals. For example, after establishing a gigabit link on the Ethernet port connecting a high-definition camera, the switching chip detects that the camera also supports energy-efficient Ethernet. Both parties negotiate to enter a low-power idle mode. During the interval when there is no video stream transmission, the link automatically enters a dynamic power-saving state, at which time the power consumption of the port's physical layer can be reduced from approximately 350 milliwatts (mW) to approximately 150 mW.

[0120] In dynamic energy-saving mode, when there is no service data frame transmission on the link, the physical layer transmit / receive circuit corresponding to the designated network port in the control intelligent network switching unit switches to a low-power idle state.

[0121] Low-power idle state is a specific operating mode under dynamic power saving, which means that the physical layer transmitter and receiver circuits are partially or completely shut down during idle periods and the bias voltage is reduced to achieve the minimum maintenance power consumption.

[0122] Specifically, the energy-efficient Ethernet control logic within the intelligent network switching unit continuously monitors the transmit and receive queues of the port. When there is no data transmitted or received for a continuous period (e.g., several microseconds), this logic module generates a control signal to shut down the high-speed serializer / deserializer clock in the physical layer transmit / receive circuit and adjusts the bias current of the analog front-end circuit, putting it into deep sleep mode. Once a data frame arrives, the circuit can be quickly woken up and resume full-speed operation within the time specified in the protocol (e.g., several microseconds). For example, in dynamic energy-saving mode, when the link is detected to be idle for more than 5 microseconds, the switching chip automatically switches the physical layer transmit / receive circuit of the corresponding port to a low-power idle state. At this time, the power consumption of this part of the circuit further decreases from 150mW to approximately 50mW. When the camera generates a new frame of image data, the circuit is woken up within 3µs, returning to normal operating level, ensuring no additional delay in data transmission.

[0123] Based on the current decision-making working mode of the fiber optic IoT gateway and the predicted output of the traffic prediction model, the service activity of sensor devices connected to a specified network port is evaluated.

[0124] Among them, business activity is a comprehensive indicator used to quantify the likelihood of sensor devices generating data in the foreseeable future. Its value is subject to mandatory constraints on the gateway's operating mode and is calibrated based on fine-grained predictions from the business volume prediction model.

[0125] Specifically, the evaluation logic is based on the gateway's operating mode: if the gateway is in deep sleep mode, the service activity evaluation result for all sensor ports is forced to be "inactive"; if it is in active mode, a detailed evaluation is performed based on the prediction model. During the evaluation, an activity score is calculated using a weighted scoring algorithm, combining the predicted data traffic output of the service volume prediction model for the sensor's category, the port's historical data generation pattern, and the current time, and compared with a preset activity threshold. For example, the gateway is currently in smart standby mode. The service volume prediction model outputs that the predicted data traffic of the micro-weather sensor connected to port 3 will average 2kbps over the next 30 minutes, and historical data shows that the sensor only reports data once every 5 minutes at night. The evaluation algorithm calculates its activity score as 15 (out of 100), which is lower than the set activity threshold of 50. Therefore, the evaluation result is that the sensor device is in a "low activity" or "inactive" state.

[0126] If the assessment result indicates that the sensor device is in the predicted service quiet period, a port-level power gating command is issued through the serial management interface of the intelligent network switching unit to disconnect the power supply line to the physical layer circuit of the specified network port.

[0127] Port-level power gating commands are digital commands sent through the management interface of the switching chip to directly control the power supply switch of the physical layer circuit of a specific port. The service quiet period refers to the assessed time during which sensor devices will not generate valid service data for a relatively long period (such as tens of minutes or hours).

[0128] Specifically, when the evaluation logic determines that a sensor connected to a certain port will enter a service silence period, the main processor accesses the internal management register of the intelligent network switching unit through the serial management interface. The main processor writes a specific control value to this register, corresponding to the operation of "disabling the physical layer power supply to the specified port". After receiving the instruction, the switching chip physically disconnects the power supply to the physical layer transmit / receive circuit of that port through its internal power switch array, thereby achieving near-zero standby power consumption. Before the end of the silence period or upon receiving a wake-up instruction, the power supply is restored through the reverse register write operation. For example, the evaluation results show that the port connected to the LiDAR will have no data for the next 2 hours. The main processor writes data 0x00 (corresponding to disabling PHY power supply) to the management register (address 0x0A) of the switching chip through the serial management interface. The switching chip executes this instruction, cutting off the 3.3V power supply to the physical layer circuit of that port. At this time, the power consumption of that port drops from approximately 50mW to 0mW. Two hours later, based on the predictive model, the main processor writes data 0x01 to the same register 50ms before the expected service arrives, restores power, and ensures that the link is ready when the radar sensor goes online.

[0129] Therefore, according to the above implementation method, the system can combine the dynamic energy-saving protocol at the link layer with static power shutdown based on service prediction to form a dual energy-saving mechanism of "dynamic + static". During short idle periods, dynamic energy saving with microsecond-level response is achieved using the energy-efficient Ethernet protocol; during long silent periods, thorough port-level power gating is implemented through predictive evaluation to completely eliminate standby power consumption. This approach not only ensures the real-time requirements of service response but also maximizes the energy-saving potential at the link level, which is one of the key aspects of achieving ultra-low power operation in this invention.

[0130] In some embodiments, by sending power management commands to the intelligent power management unit, time-sharing and zone-based power supply scheduling is performed on multiple independent power supply domains corresponding to multiple independent power supply output terminals of the intelligent power management unit, including: Based on the real-time operating performance status of the main processor, a first power management instruction is generated and sent to the intelligent power management unit to perform dynamic voltage regulation on the supply voltage of the core voltage regulation domain, which supplies power to the core logic circuits of the heterogeneous computing unit and the intelligent network switching unit.

[0131] The core voltage regulation domain is a power supply zone that provides adjustable voltage for high-speed digital logic circuits in the system that are sensitive to voltage and frequency (such as the main processor core and the switching engine of the switching chip). Dynamic voltage regulation refers to the process of changing the output voltage value of this power supply zone in real time through digital interface commands.

[0132] Specifically, the main processor monitors its own operating performance status in real time (such as maximum performance, balanced efficiency, and low power consumption), which corresponds to a set of preset voltage-frequency combinations. When the status changes, the main processor sends a first power management command to the intelligent power management unit via the Inter-Integrated Circuit (I2C) bus. This command contains the target voltage value parameter. After receiving the command, the intelligent power management unit adjusts the output of its internal DC-DC converter corresponding to the core voltage regulation domain, realizing dynamic voltage switching. The switching process typically takes tens to hundreds of microseconds to maintain stability. For example, when the main processor switches from the balanced efficiency state (1.0V, 120MHz) to the maximum performance state, the main processor immediately sends a command via the I2C bus to the TPS65263 power management integrated circuit to adjust the output voltage of the core voltage regulation domain from 1.0V to 1.2V. The power management integrated circuit completes the voltage regulation within 200µs, providing a stable voltage guarantee for the main processor to operate at 240MHz.

[0133] Based on the current decision of the working mode of the fiber optic IoT gateway and the physical layer power gating state of the designated network port, a second power management command is generated and sent to the intelligent power management unit to perform switching control on the power supply of the physical layer voltage domain, which supplies power to the physical layer interface circuit of the intelligent network switching unit.

[0134] The physical layer voltage domain is an independent partition that supplies power to the physical layer transmitter and receiver circuits of all network ports on the intelligent network switching unit. The switch control refers to turning the power output to this entire power supply partition on or off.

[0135] Specifically, the main processor makes a comprehensive judgment based on the gateway's current operating mode (active, smart standby, deep sleep) and the physical layer power gating status of each network port obtained through the Serial Management Interface (SMI). When the gateway enters deep sleep mode, or when the physical layer circuits of all network ports have been individually gated off, the main processor determines that the entire physical layer voltage domain can be safely powered down. Subsequently, the main processor sends a second power management command via the I2C bus, instructing the intelligent power management unit to shut down the power supply output of the physical layer voltage domain. For example, if the gateway decides to enter deep sleep mode, the main processor first shuts down the physical layer power of all electrical and optical network ports via the SMI bus. After confirming that all ports are powered down, the main processor sends a command to the intelligent power management unit via the I2C bus to shut down the 3.3V power supply output of the physical layer voltage domain. At this time, the power consumption of the entire switching chip's physical layer circuit drops to 0W. When the gateway is woken up, the main processor first sends a command to turn on the physical layer voltage domain power supply, and after the voltage stabilizes, it restores the physical layer functions of the required ports one by one via the SMI bus.

[0136] Based on the data transmission schedule determined by the forecast output of the traffic prediction model, a third power management command is generated and sent to the intelligent power management unit to complete the power-on timing of the pluggable optical module voltage domain before the data transmission window, and to trigger the power-off timing of the pluggable optical module voltage domain after the data transmission is completed. The pluggable optical module voltage domain supplies power to the pluggable optical modules of the fiber optic access unit.

[0137] Among them, data transmission scheduling refers to the precise communication window planned based on the estimation of uplink data transmission time of the optical fiber link according to the traffic prediction model. The power-on sequence and power-off sequence refer to the power-on and power-off operations for powering the optical module and the necessary stabilization waiting time.

[0138] Specifically, the main processor uses a traffic prediction model to estimate the future time when data needs to be sent to the uplink fiber optic link next. .exist A previous fixed lead time (e.g., 50ms) The main processor sends a third power management command (power-on command) to the intelligent power management unit via the I2C bus to initiate the power-on process of the pluggable optical module's voltage domain. Power restoration and self-test completion for the optical module, as well as optical transmitter stabilization, typically take several milliseconds to tens of milliseconds. After data transmission is complete, following a brief idle monitoring period to confirm no further data transmission, the main processor sends another third power management command (power-off command) to cut off power to that voltage domain. For example, a traffic prediction model indicates that video data will begin uploading in the next 300 seconds and is expected to last 30 seconds. The main processor is set... The timeframe is 50ms. At 299.95 seconds, the main processor sends a command to enable the 3.3V power supply to the pluggable optical module's voltage domain. The optical module completes initialization within 10ms. At 300 seconds, data transmission begins normally. At 330.5 seconds (transmission end plus 0.5 seconds for idle monitoring), the main processor sends a command to disable the optical module's power supply. This method avoids unnecessary power consumption by the optical module during idle periods lasting several minutes.

[0139] Based on the assessment results of the service activity of the sensor devices, a fourth power management command is generated and sent to the intelligent power management unit to perform on-demand switching control of the power supply to the interface voltage domain, which supplies power to the sensor devices connected to the specified network port.

[0140] The interface voltage domain is a separate partition that provides remote power to network ports connected to external sensors (such as via Power over Ethernet, PoE) or to power dedicated interface circuits for sensors. The on-demand switching control refers to determining whether to power the sensor based on a predictive assessment of the sensor's operational activities, rather than simply the port link status.

[0141] Specifically, the main processor determines whether to power the interface voltage domain based on the sensor's activity level assessment (active, low-activity, silent) and the gateway's operating mode. For example, for a sensor assessed as entering a long silent period, the main processor will send a fourth power management command to shut down the interface voltage domain output that powers it, even if its physical connection exists. Power is then restored in advance when its service is predicted to resume or an external wake-up signal is received. This works in conjunction with the physical layer power gating at the switch chip port level, but applies to the power rail that powers the sensor itself. For example, the system assesses that the LiDAR sensor connected to port 5 has no inspection tasks between 0:00 and 6:00 AM, placing it in a silent period. At 0:00, after controlling the switch chip to shut down the physical layer power supply to that port, the main processor further sends a command via the I2C bus to shut down the interface voltage domain output that provides 24V DC power to the LiDAR. At 5:55 AM, based on timing rules, the main processor sends a command to restore the interface voltage domain power supply, enabling the LiDAR to complete self-start and warm-up before the 6:00 AM inspection task begins.

[0142] Therefore, according to the above implementation method, the system can implement precise and differentiated "time-sharing" and "zone-sharing" power supply scheduling for the four key power supply domains—core voltage, physical layer, optical module, and sensor interface—based on multi-dimensional information such as chip computing status, network operating mode, service communication scheduling, and sensor activity patterns. This system-level refined management strictly distributes energy to different hardware modules on demand and eliminates all power supply overhead during unnecessary periods in the time dimension. Thus, in conjunction with chip-level and link-level energy-saving measures, it reduces the overall energy consumption of the gateway to an ultra-low level that is difficult to achieve with traditional solutions.

[0143] In some embodiments, the step of offloading computationally intensive data processing tasks to an artificial intelligence coprocessor includes: The system performs online sensing and classification of real-time sensor data streams input to fiber optic IoT gateways to obtain the data stream type and metadata.

[0144] Online sensing and classification refers to the process of quickly identifying the sensor type (such as video, audio, vibration) and key attributes (such as resolution, sampling rate, and data format) of the data source by analyzing data packet header information, data characteristics, or preceding data fragments in real time without interrupting the continuity of the data stream. The metadata describes the attributes of the data stream itself.

[0145] Specifically, the system sets up a lightweight packet parser at the data link layer or network layer. This parser examines the source Media Access Control (MAC) address, Ethernet type field, or User Datagram Protocol (UDP) port number of each data frame entering the gateway. For video streams, it may further parse the Real-Time Streaming Protocol (RTSP) header to obtain encoding format and resolution information. This information is extracted and encapsulated into structured metadata and passed to the task scheduler. For example, a packet arrives at the gateway. The parser checks that the source MAC address of the packet corresponds to a known IP camera A, and the destination port number is 554 (the default port for RSP). The parser thus determines that the data stream is an H.264 encoded video stream and further extracts the video resolution from the Session Description Protocol (STP) information. Pixels, frame rate 25 frames per second. The metadata structure is recorded as follows: .

[0146] When the data stream is perceived to be a visual sensing data stream, and the data throughput or single-frame data volume of the visual sensing data stream reaches the computational unloading trigger threshold, the corresponding computer vision inference task is instantiated as a task to be unloaded.

[0147] The computational offload trigger threshold is a preset threshold used to determine whether the computational load of the current data stream has reached a level requiring hardware acceleration using a dedicated AI coprocessor. Instantiation refers to the process of creating a schedulable and executable task object based on specific input data and a predefined algorithm model.

[0148] Specifically, the system continuously monitors the real-time throughput (e.g., average data rate, in megabits per second, Mbps) and / or keyframe size (in kilobytes, KB) of the visual sensing data streams. When the monitored values ​​exceed a preset trigger threshold, the system task manager selects a matching computer vision model (e.g., the object detection model YOLOv5s) from the task template library and binds the current video frame buffer address, model parameter address, and output result buffer address to create a specific "unloadable task" instance, which is then placed in the processing queue. For example, the system sets the video stream calculation unloading trigger threshold as follows: a continuous data rate higher than 2 Mbps, or a single I-frame (keyframe) size exceeding 100 KB. When the average rate of the video stream from a certain camera reaches 3.5 Mbps, the system determines that the unloading conditions are met. The task manager then instantiates a "wire foreign object detection" task, which contains a pointer to the memory address of the latest JPEG image frame (approximately 120 KB in size) and the configured YOLOv5s-tiny model parameters.

[0149] By using the shared memory controller or high-speed on-chip interconnect bus within the heterogeneous computing unit, direct memory access transfer is initiated from the system main memory accessed by the main processor to the local memory of the artificial intelligence coprocessor, in order to move the computation graph and input tensor corresponding to the task to be unloaded.

[0150] Direct memory access (DMI) is a technology that moves data directly between memory and external devices (in this case, AI coprocessors) without intervention from the main processor's CPU core. This significantly reduces the load on the main processor and improves transmission efficiency. The computation graph describes the computational logic and data flow of the neural network model. The input tensor is a multidimensional data array that conforms to the requirements of the computation graph's input nodes.

[0151] Specifically, after the task scheduler on the main processor side prepares the tasks to be unloaded, it configures the Direct Memory Access Controller (DMI). The DMI source address is set to the physical address of the system main memory storing the input tensors (such as image pixel data), the destination address is set to a specific region address in the local memory of the AI ​​coprocessor, and the total amount of data to be transferred is set. DMI transfer is then initiated. Simultaneously, the weight parameters and computation graph structure of the neural network model may have been pre-loaded into the coprocessor memory or transferred as part of the task. For example, the JPEG image to be processed (120KB) has been decompressed into an RGB bitmap and stored at address 0x8000_0000 in the system main memory, with a size of... Bytes ≈ 6MB. The task scheduler configures the direct memory access controller to transfer 6MB of data from address 0x8000_0000 to address 0x1_0000 in the AI ​​coprocessor's local memory. After the transfer is complete, the direct memory access controller generates an interrupt to notify the main processor.

[0152] It calls and executes the pre-built hardware acceleration kernel in the artificial intelligence coprocessor, which corresponds to the computer vision reasoning task, to process the input tensor and write back the structured semantic labels represented by the output tensor to the specified result buffer in the system main memory.

[0153] The hardware-accelerated kernel is firmware or microcode optimized for specific neural network operators (such as convolution and pooling), and can be executed efficiently on the dedicated computing unit of the AI ​​coprocessor. The structured semantic labels are a machine-readable description of the inference results, such as the detected target category, confidence level, and bounding box coordinates.

[0154] Specifically, after receiving the input tensor, the AI ​​coprocessor loads a hardware-accelerated kernel sequence corresponding to the task model via its internal task scheduling unit. This kernel sequence drives the tensor processing unit array to perform a series of operations on the input tensor, such as convolution and nonlinear activation. After the computation is complete, an output tensor (e.g., a tensor with a shape of...) is obtained. An array, where N is the number of detected targets, and 6 represents [category, confidence level]. The coprocessor then initiates another direct memory access transfer (or is read by the main processor) to write the output tensor back to the result buffer that the main processor has pre-allocated in the system main memory. For example, the AI ​​coprocessor loads a dedicated accelerated kernel for YOLOv5s-tiny. After processing the input tensor, the kernel produces an output array: [ The array indicates that two targets were detected: the first is category 0 ("bird"), with 85% confidence, located in the center of the image; the second is category 2 ("plastic bag"), with 90% confidence, located slightly to the right of the image. This output array is passed back to the system's main memory at address 0x9000_0000 via direct memory access.

[0155] Therefore, according to the above implementation method, the system can dynamically offload computationally intensive and power-intensive tasks such as video analysis from the general-purpose main processor to a dedicated artificial intelligence coprocessor through intelligent perception and decision-making. This approach not only significantly improves the execution efficiency and energy efficiency of intelligent analysis tasks such as image recognition and anomaly detection, but also frees up the main processor's computing resources, allowing it to focus more on core control tasks such as system scheduling and communication protocol processing. This optimizes the overall utilization of the gateway's heterogeneous computing resources and achieves the best balance between performance and power consumption.

[0156] The specific functions and examples of each module and submodule of the device in this embodiment of the invention can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0157] According to a fourth aspect, the present invention provides a computer device comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform any of the fiber optic IoT gateway control methods in the embodiments of the present invention.

[0158] According to another aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute any of the fiber optic Internet of Things gateway control methods in the embodiments of the present invention.

[0159] Figure 5 A schematic block diagram of an example computer device 600 that can be used to implement embodiments of the present invention is shown. The computer device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The computer device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0160] like Figure 5As shown, the computer device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 may also store various programs and data required for the operation of the computer device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0161] Multiple components in computer device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows computer device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0162] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a fiber optic IoT gateway control method. For example, in some embodiments, a fiber optic IoT gateway control method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the computer device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the fiber optic IoT gateway control method described above can be performed. Alternatively, in other embodiments, the computing unit 601 can be configured to perform a fiber optic IoT gateway control method by any other suitable means (e.g., by means of firmware).

[0163] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0164] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0165] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0166] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual, auditory, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0167] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0168] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0169] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0170] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A fiber optic Internet of Things (IoT) gateway, characterized in that, It includes heterogeneous computing units, intelligent network switching units, fiber optic access units, and intelligent power management units; The heterogeneous computing unit includes a main processor as the main control core and an artificial intelligence coprocessor for hardware acceleration. The intelligent network switching unit includes at least one electrical network port and at least one optical fiber port. The intelligent network switching unit is communicatively coupled to the heterogeneous computing unit through a high-speed data bus interface. The fiber optic access unit includes at least one pluggable optical module, which is pluggable to the fiber optic port of the intelligent network switching unit. The intelligent power management unit includes a digital programmable power management integrated circuit. The control interface of the intelligent power management unit is connected to the heterogeneous computing unit. The multiple independent power supply output terminals of the intelligent power management unit are respectively connected to the power input terminals of the heterogeneous computing unit, the intelligent network switching unit, and the fiber optic access unit. The main processor is configured to execute a three-level collaborative energy-saving management process, which includes a business forecasting and learning phase, an intelligent mode decision-making phase, and a hardware collaborative execution phase. The intelligent power management unit is configured to perform independent switching control or dynamic voltage regulation on the plurality of independent power supply output terminals in response to power management commands from the heterogeneous computing unit.

2. An energy efficiency control method based on fiber optic IoT gateways, characterized in that, The method, based on the fiber optic IoT gateway as described in claim 1, is applied to the main processor and includes: Perform time-series analysis and feature extraction on historical business data and real-time sensor data, and train and deploy a business volume prediction model. Based on the predicted output of the traffic volume prediction model, the working mode of the fiber optic IoT gateway is determined, including an active mode, an intelligent standby mode, and a deep sleep mode. The hardware coordination steps include chip-level coordination steps, link-level coordination steps, and system-level coordination steps. The chip-level collaboration step includes: adjusting the operating voltage and core frequency of the main processor according to the real-time load index of the fiber optic IoT gateway through a dynamic voltage and frequency adjustment strategy, and offloading computationally intensive data processing tasks to the artificial intelligence coprocessor for execution; The link-level coordination steps include: controlling the intelligent network switching unit to enable the energy-efficient Ethernet protocol at the data link layer, and performing power gating operations on the physical layer circuits of the designated network ports that are idle on the intelligent network switching unit according to the currently decided working mode. The system-level collaborative steps include: sending power management commands to the intelligent power management unit to perform time-sharing and zone-based power supply scheduling for multiple independent power supply domains corresponding to multiple independent power supply output terminals of the intelligent power management unit.

3. The method according to claim 2, characterized in that, The process of performing time-series analysis and feature extraction on historical business data and real-time sensor data, and training and deploying a business volume prediction model, includes: A time series prediction model is constructed and trained to learn and output time-volume correlation prediction results based on the periodic patterns and trend characteristics of the historical business data. Construct and train a multi-feature fusion prediction model to learn and output weather-business volume correlation prediction results based on meteorological feature parameters in the real-time sensing data and the historical business data; Configure a high-priority service response channel for a preset type of emergency, wherein the emergency is generated by the real-time sensing data stream through the real-time identification model, or triggered by the prediction output of at least one of the time series prediction model and the multi-feature fusion prediction model.

4. The method according to claim 2, characterized in that, The decision on the operating mode of the fiber optic IoT gateway based on the prediction output of the traffic volume prediction model includes: When the predicted output of the traffic volume prediction model indicates that the future traffic volume will continue to be lower than the preset low traffic threshold, and the fiber optic IoT gateway is currently in active mode, the decision is made to switch the working mode of the fiber optic IoT gateway from the active mode to the intelligent standby mode. When the system clock enters a preset silent period, the working mode of the fiber optic IoT gateway is determined to enter the deep sleep mode. In the deep sleep mode, the working mode of the fiber optic IoT gateway is woken up from the deep sleep mode to the intelligent standby mode at preset intervals, and network heartbeat messages are sent in the intelligent standby mode. When the traffic prediction model indicates that traffic data has arrived in the intelligent standby mode, or when a wake-up command is received from the remote network management system, the working mode of the fiber optic IoT gateway is switched from the intelligent standby mode to the active mode.

5. The method according to claim 2, characterized in that, The step of adjusting the operating voltage and core frequency of the main processor according to the real-time load indicators of the fiber optic IoT gateway through a dynamic voltage and frequency adjustment strategy includes: Periodically sample the CPU utilization of the main processor; When the utilization rate of the central processing unit is continuously higher than the preset high load threshold, the operating performance state of the main processor is configured to the highest performance state. When the utilization rate of the central processing unit is lower than the high load threshold and remains higher than the preset low load threshold, the operating performance state of the main processor is configured to a balanced energy efficiency state, and the operating voltage and core frequency corresponding to the balanced energy efficiency state are lower than the highest performance state. When the utilization rate of the central processing unit is consistently lower than the low load threshold, the operating performance state of the main processor is configured to a low power state, and the operating voltage and core frequency corresponding to the low power state are lower than the balanced energy efficiency state. When the fiber optic IoT gateway is determined to enter the intelligent standby mode or the deep sleep mode, the operating performance state of the main processor is configured to a preset deep sleep state.

6. The method according to claim 2, characterized in that, The control of the intelligent network switching unit to enable the energy-efficient Ethernet protocol at the data link layer, and the execution of power gating operations on the physical layer circuits of designated network ports that are idle on the intelligent network switching unit according to the currently decided operating mode, including: During the link auto-negotiation phase, if it is detected that the peer device of the link of the intelligent network switching unit supports the energy-efficient Ethernet protocol, the energy-efficient Ethernet function negotiation is initiated, so that the link established between the intelligent network switching unit and the peer device enters a dynamic energy-saving state. In the dynamic energy-saving state, when there is no service data frame transmission on the link, the physical layer transmit / receive circuit corresponding to the designated network port in the intelligent network switching unit is controlled to switch to a low-power idle state. Based on the current decision, the operating mode of the fiber optic IoT gateway, and the prediction output of the traffic prediction model, the service activity of the sensor device connected to the designated network port is evaluated. If the evaluation result indicates that the sensor device is in the predicted service quiet period, a port-level power gating command is issued through the serial management interface of the intelligent network switching unit to disconnect the power supply line to the physical layer circuit of the specified network port.

7. The method according to claim 2, characterized in that, The step of sending power management commands to the intelligent power management unit to perform time-sharing and zone-based power supply scheduling for multiple independent power supply domains corresponding to multiple independent power supply output terminals of the intelligent power management unit includes: Based on the real-time operating performance status of the main processor, a first power management instruction is generated and sent to the intelligent power management unit to perform dynamic voltage adjustment on the power supply voltage of the core voltage adjustment domain, which supplies power to the core logic circuits of the heterogeneous computing unit and the intelligent network switching unit. Based on the current decision of the working mode of the fiber optic IoT gateway and the physical layer power gating state of the designated network port, a second power management command is generated and sent to the intelligent power management unit to perform switching control on the power supply of the physical layer voltage domain, which supplies power to the physical layer interface circuit of the intelligent network switching unit. Based on the data transmission schedule determined by the predicted output of the traffic prediction model, a third power management instruction is generated and sent to the intelligent power management unit to complete the power-on timing of the pluggable optical module voltage domain before the data transmission window, and to trigger the power-off timing of the pluggable optical module voltage domain after the data transmission is completed. The pluggable optical module voltage domain supplies power to the pluggable optical module of the optical fiber access unit. Based on the service activity assessment results of the sensor device, a fourth power management command is generated and sent to the intelligent power management unit to perform on-demand switching control of the power supply to the interface voltage domain, which supplies power to the sensor device connected to the designated network port.

8. The method according to claim 2, characterized in that, The step of offloading computationally intensive data processing tasks to the artificial intelligence coprocessor includes: The real-time sensing data stream input to the fiber optic IoT gateway is sensed and classified online to obtain the data stream type and metadata; When it is sensed that the data stream is a visual sensing data stream, and the data throughput or single-frame data volume of the visual sensing data stream reaches the computational unloading trigger threshold, the corresponding computer vision inference task is instantiated as a task to be unloaded. Through the shared memory controller or high-speed on-chip interconnect bus inside the heterogeneous computing unit, a direct memory access transfer is initiated from the system main memory accessed by the main processor to the local memory of the artificial intelligence coprocessor, so as to move the computation graph and input tensor corresponding to the task to be unloaded. The system invokes and executes a pre-built hardware acceleration kernel in the artificial intelligence coprocessor that corresponds to the computer vision reasoning task, to process the input tensor and writes back the structured semantic labels represented by the output tensor to the designated result buffer in the system main memory.

9. A computer device, characterized in that, include: At least one processor; and a memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 2-8.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, in, Computer instructions are used to cause a computer to perform the method according to any one of claims 2-8.

Images