WO2022227106A1 - 一种基于risc-v指令集的aiot多制式边缘网关通信系统及设备 - Google Patents

一种基于risc-v指令集的aiot多制式边缘网关通信系统及设备 Download PDF

Info

Publication number
WO2022227106A1
WO2022227106A1 PCT/CN2021/092328 CN2021092328W WO2022227106A1 WO 2022227106 A1 WO2022227106 A1 WO 2022227106A1 CN 2021092328 W CN2021092328 W CN 2021092328W WO 2022227106 A1 WO2022227106 A1 WO 2022227106A1
Authority
WO
WIPO (PCT)
Prior art keywords
module
sub
layer
data
management
Prior art date
Application number
PCT/CN2021/092328
Other languages
English (en)
French (fr)
Inventor
郑创杰
陈升东
袁峰
Original Assignee
广州中国科学院软件应用技术研究所
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 广州中国科学院软件应用技术研究所 filed Critical 广州中国科学院软件应用技术研究所
Publication of WO2022227106A1 publication Critical patent/WO2022227106A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0803—Configuration setting
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L63/00—Network architectures or network communication protocols for network security
    • H04L63/04—Network architectures or network communication protocols for network security for providing a confidential data exchange among entities communicating through data packet networks
    • H04L63/0428—Network architectures or network communication protocols for network security for providing a confidential data exchange among entities communicating through data packet networks wherein the data content is protected, e.g. by encrypting or encapsulating the payload
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W12/00—Security arrangements; Authentication; Protecting privacy or anonymity
    • H04W12/03—Protecting confidentiality, e.g. by encryption

Definitions

  • the present invention relates to the technical field of network communication, and more particularly, to an AIOT multi-standard edge gateway communication system and equipment based on a RISC-V instruction set.
  • AIOT is a technical field that combines informatization and artificial intelligence.
  • AIOT is a major upgrade that can more comprehensively serve human production and life, and has huge application prospects.
  • the 5G edge computing gateway is mainly placed on the edge near the data acquisition device to perform local edge computing.
  • the main control processing module communicates with the 5G communication module through USB and is configured to communicate through 5G.
  • the module is connected to the pickup, speaker, and camera for voice collection, voice playback, image collection, and data operation processing; real-time image collection through the camera and transmission to the neural network processor for convolutional neural network convolution calculation to realize vehicle information collection , detection, recognition, face recognition, detection, human posture detection work.
  • the above method only provides third-party libraries that have been compiled and calculated for calling, and the algorithm is relatively closed, which is not conducive to the realization of custom operator development and model optimization in AIOT application scenarios; on the other hand, the compatibility is poor, It can only access 5G base stations, but cannot access edge nodes.
  • the present invention proposes an AIOT multi-standard edge gateway communication system and equipment based on the RISC-V instruction set.
  • the networking communication protocol and multi-dimensional modeling form the functional interaction between the edge gateway and the device, improving work efficiency and data compatibility.
  • an AIOT multi-standard edge gateway communication system based on the RISC-V instruction set is provided.
  • the AIOT multi-standard edge gateway communication system based on the RISC-V instruction set specifically includes:
  • Cloud collaboration platform intelligent algorithm layer, security encryption layer, management layer, gateway adaptation layer, protocol stack layer, device driver layer, embedded operation layer, instruction set processor; wherein, the cloud collaboration platform and the intelligent algorithm
  • the security encryption layer, the management layer, the gateway adaptation layer and the protocol stack layer respectively perform security encryption, data management, protocol conversion, and network transmission for the data to be interacted with.
  • the embedded operation layer is provided with the instruction set processor, the embedded operation layer is configured with the device driver layer, and the device driver layer performs functional interaction with an external transmission device.
  • the full name of the RISC-V instruction set is Reduced Instruction Set Computing V
  • the Chinese name is the fifth-generation RISC reduced instruction set computer.
  • the security encryption layer specifically includes DTLA sub-module, AES sub-module, CA sub-module, SSL sub-module, SHA265 sub-module, HASH sub-module, and MD5 sub-module; wherein, all The DTLA sub-module is used for digital transmission authorization management, the AES sub-module is used for high-level encryption, the CA sub-module is used to manage the issuance of security credentials and encrypted information security keys, and the SSL sub-module is used to provide security Transmission protocol, the SHA265 sub-module is used for secure encryption using cryptographic hashing; the HASH sub-module is used to map an input of any length to a hash value of a fixed length through a hash algorithm; the MD5 sub-module is used for Generates a 16-byte hash value.
  • the protocol stack layer includes MQTT transmission sub-module, CoAP transmission protocol sub-module, TCP/IP transmission protocol sub-module, UDP transmission protocol sub-module, ARP transmission protocol sub-module, DHCP Transmission protocol sub-module, IPv4 transmission protocol sub-module, IPv6 transmission protocol sub-module, NVME transmission protocol sub-module, BLE transmission protocol sub-module; wherein, the MQTT transmission sub-module is used to deliver the transmission data with the message queue telemetry transmission standard
  • the CoAP transmission protocol submodule is used for data forwarding with the CoAP transmission protocol;
  • the TCP/IP transmission protocol submodule is used for forwarding the transmission data between the networks with the TCP/IP transmission protocol, and the IPv4 transmission protocol
  • the submodule and the IPv6 transmission protocol submodule are respectively used for the generation of IP addresses of different levels;
  • the NVME transmission protocol submodule is used to openly collect information in the mobile device and the data center through the non-volatile
  • the management layer includes a Radio management submodule, a Powr management submodule, a Memory management submodule, a Time management submodule, a Client management submodule, a KeyManager management submodule, and an algorithm management submodule.
  • the Radio management sub-module is used to manage the audio module in the embedded system;
  • the Powr management sub-module is used to manage the power management module in the embedded system;
  • the Memory management sub-module is used to manage the embedded system A memory management module in the system;
  • the Time management submodule is used for managing timers and time-triggered task function modules in the embedded system;
  • the Client management submodule is used for the user group management module in the embedded system;
  • the KeyManager management sub-module is used to manage the user key management module in the embedded system;
  • the algorithm management sub-module is used to manage the adopted AI intelligent algorithms, wherein the AI intelligent algorithms include YOLOV3 and MobileNet.
  • the gateway adaptation layer includes access to multiple network interfaces, specifically including WAN, LAN, RS485, Wi-Fi, Zigbee, 4G/5G, LTE-V, BLE4.1, NB-IoT, Sigfox, LoRa, 6LoPWAN.
  • network interfaces specifically including WAN, LAN, RS485, Wi-Fi, Zigbee, 4G/5G, LTE-V, BLE4.1, NB-IoT, Sigfox, LoRa, 6LoPWAN.
  • the cloud collaboration platform includes an edge node and a cloud center
  • the edge node includes a management module, a data operation module, an application service module, and a book storage module
  • the cloud center includes an edge intelligent management suite and an artificial intelligence application service, wherein the artificial intelligence application service cooperates with the intelligent algorithm layer to perform image processing, function operation, network training, stream computing, and big data mining
  • the The Edge Intelligence Management Suite includes program, resource and application management.
  • an AIOT multi-standard edge gateway communication method based on a RISC-V instruction set is provided.
  • the AIOT multi-standard edge gateway communication method based on the RISC-V instruction set specifically includes:
  • the convolution operation is performed through the embedded operation layer and the instruction set processor configured on it to generate processing data, wherein the processing data includes IoT node data, image information, and sound information;
  • the edge hardware computing device perform parameter quantification evaluation to obtain the evaluation accuracy rate, and re-optimize and compress when the evaluation accuracy rate decreases;
  • parameter quantization evaluation is performed to obtain an evaluation accuracy rate, and when the evaluation accuracy rate decreases, optimization and compression are performed again, specifically including:
  • the speed adjustment is performed on the completed calculation, so that different hardware acceleration chips can complete different edge operations and output supporting codes, which specifically includes:
  • a computing scheduling subgraph is generated, and the scheduling configuration, the bmodel file, the C++ reasoning code and the scheduling subgraph are collectively and stored as an edge computing model.
  • an electronic device including a memory and a processor, the memory being used to store one or more computer program instructions, wherein the one or more computer program instructions are processed by the The controller executes to implement the steps described in the first aspect of the embodiment of the present invention.
  • This solution adopts an open source instruction set architecture and uses artificial intelligence to process data, which can realize online access, analysis, processing and calculation of a large number of node data, with high efficiency and strong compatibility;
  • This solution realizes the functions of data-aware computing, data fusion and node broadcasting in different scenarios through various application scenarios such as load vehicle-road coordination, and meets the computing performance requirements of high-load systems.
  • FIG. 1 is a structural diagram of an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • FIG. 2 is a structural diagram of a security encryption layer in an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • FIG. 3 is a structural diagram of a protocol stack layer in an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • FIG. 4 is a structural diagram of a management layer in an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • FIG. 5 is a structural diagram of a cloud collaboration platform in an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • FIG. 6 is a schematic diagram of an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • FIG. 7 is a flowchart of an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set according to an embodiment of the present invention.
  • parameter quantization evaluation is performed to obtain the evaluation accuracy rate, and when the evaluation accuracy rate decreases, the Flowchart for optimization and compression.
  • Fig. 9 is an embodiment of the present invention, in an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set, according to the edge hardware computing device, parameter quantization evaluation is performed to obtain the evaluation accuracy rate, and when the evaluation accuracy rate decreases, the Schematic diagram of optimization and compression.
  • 10 is an embodiment of the present invention, in an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set, the speed adjustment is performed on the completed calculation, so that different hardware acceleration chips can complete different edge operations and output supporting codes. flow chart.
  • FIG. 11 is an embodiment of the present invention, in an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set, speed adjustment is performed on the completed calculation, so that different hardware acceleration chips can complete different edge operations and output supporting codes.
  • schematic diagram in an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set, speed adjustment is performed on the completed calculation, so that different hardware acceleration chips can complete different edge operations and output supporting codes.
  • FIG. 12 is a structural diagram of an electronic device in an embodiment of the present invention.
  • AIOT is a technical field that combines informatization and artificial intelligence.
  • AIOT is a major upgrade that can more comprehensively serve human production and life, and has huge application prospects.
  • the 5G edge computing gateway is mainly placed on the edge near the data acquisition device to perform local edge computing.
  • the main control processing module communicates with the 5G communication module through USB and is configured to communicate through 5G.
  • the module is connected to the pickup, speaker, and camera for voice collection, voice playback, image collection, and data operation processing; real-time image collection through the camera and transmission to the neural network processor for convolutional neural network convolution calculation to realize vehicle information collection , detection, recognition, face recognition, detection, human posture detection work.
  • the above method only provides third-party libraries that have been compiled and calculated for calling, and the algorithm is relatively closed, which is not conducive to the realization of custom operator development and model optimization in AIOT application scenarios; on the other hand, the compatibility is poor, It can only access 5G base stations, but cannot access edge nodes.
  • an AIOT multi-standard edge gateway communication system and device based on the RISC-V instruction set are provided.
  • the solution forms the functional interaction between the edge gateway and the device side, improving work efficiency and data. compatibility.
  • an AIOT multi-standard edge gateway communication system based on the RISC-V instruction set is provided.
  • FIG. 1 is a structural diagram of an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • the AIOT multi-standard edge gateway communication system based on the RISC-V instruction set specifically includes:
  • Cloud collaboration platform 101 intelligent algorithm layer 102, security encryption layer 103, management layer 104, gateway adaptation layer 105, protocol stack layer 106, device driver layer 107, embedded operation layer 108, instruction set processor 109;
  • the cloud collaboration platform 101 performs data interaction with the intelligent algorithm layer 102, and the security encryption layer 103, the management layer 104, the gateway adaptation layer 105 and the protocol stack layer 106 respectively perform data interaction on the interacted data.
  • Security encryption, data management, protocol conversion, network transmission, the embedded operation layer 108 is provided with the instruction set processor 109, the embedded operation layer 108 is configured with the device driver layer 107, so The device driver layer 107 performs functional interaction with external transmission devices.
  • the instruction set refers to RISC-V
  • the full English name of Reduced Instruction Set Computing V is the fifth generation of RISC (reduced instruction set computer);
  • the full English name of IOT/AIOT is Internet of Things/AI+Internet of Things , corresponding to the Chinese name Internet of Things/Artificial Intelligence Internet of Things.
  • artificial intelligence processing data is introduced into the entire data processing architecture through multi-level and multi-type modules, and at the same time, an open source instruction set architecture is used to analyze, process and operate a large amount of node data;
  • a modular design is adopted, which enables different modules to perform different types of functions, simplifies the process of data processing, and increases the possibility and efficiency of external hardware access to the system.
  • FIG. 2 is a structural diagram of a security encryption layer in an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • the security encryption layer 103 specifically includes a DTLA sub-module 201, an AES sub-module 202, a CA sub-module 203, an SSL sub-module 204, and a SHA265 sub-module 205 , HASH sub-module 206, MD5 sub-module 207; wherein, the DTLA sub-module 201 is used for digital transmission authorization management, the AES sub-module 202 is used for high-level encryption, and the CA sub-module 203 is used to manage and issue security credentials.
  • the SSL sub-module 204 is used to provide a secure transmission protocol
  • the SHA265 sub-module 205 is used to perform secure encryption using cryptographic hashing
  • the HASH sub-module 206 is used to An input of any length is mapped to a fixed-length hash value
  • the MD5 sub-module 207 is used to generate a 16-byte hash value.
  • data processing is performed on the input data and information of any length, and the data is encrypted into transmission data that meets the data transmission protocol and data encryption standard, which facilitates the subsequent processing of the data, and The safe operation of the whole system is guaranteed.
  • FIG. 3 is a structural diagram of a protocol stack layer in an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • the protocol stack layer 106 includes an MQTT transmission sub-module 301, a CoAP transmission protocol sub-module 302, a TCP/IP transmission protocol sub-module 303, and a UDP transmission protocol Sub-module 304, ARP transmission protocol sub-module 305, DHCP transmission protocol sub-module 306, IPv4 transmission protocol sub-module 307, IPv6 transmission protocol sub-module 308, NVME transmission protocol sub-module 309, BLE transmission protocol sub-module 310;
  • the MQTT transmission sub-module 301 is used to issue the transmission data with the message queue telemetry transmission standard;
  • the CoAP transmission protocol sub-module 302 is used to forward the data with the CoAP transmission protocol;
  • the TCP/IP transmission protocol sub-module 303 uses To forward the transmission data between the networks with the TCP/IP transmission protocol, the IPv4 transmission protocol sub-module 307 and the IPv6 transmission protocol sub-module 308 are respectively used
  • the actually received multi-type data is forwarded by different types of protocols through various types of data protocols.
  • compatibility with multi-type data in the Internet of Things can be achieved.
  • data conversion is directly performed according to the corresponding module, and data transmission and transfer are performed according to the configuration of different types of data forwarding sub-modules according to the specified transmission data standard.
  • FIG. 4 is a structural diagram of a management layer in an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention.
  • the management layer 104 includes a Radio management submodule 401, a Powr management submodule 402, a Memory management submodule 403, a Time management submodule 404, and a Client management submodule 403.
  • Submodule 405, KeyManager management submodule 406, and algorithm management submodule 407 wherein, the Radio management submodule 401 is used to manage the audio module in the embedded system; the Powr management submodule 402 is used to manage the embedded system
  • the memory management sub-module 403 is used to manage the memory management module in the embedded system; the Time management sub-module 404 is used to manage the timer and the time-triggered task function module in the embedded system;
  • the Client management submodule 405 is used for the user group management module in the embedded system; the KeyManager management submodule 406 is used to manage the user key management module in the embedded system; the algorithm management submodule 407 is used for management
  • the management layer is the core of the entire multi-standard communication. Power is supplied to each sub-module through the management layer, and data storage, audio data storage, time recording and triggering are performed, and through management
  • the KeyManager management sub-module can also encrypt data and manage the user's key through the security encryption layer to ensure system security.
  • the corresponding management of AI intelligent algorithms is also included.
  • intelligent algorithms including at least YOLOV3 and MobileNet
  • fast identification features for different types of input data are realized, and corresponding capabilities are trained according to the features.
  • the AIOT multi-standard edge gateway communication system based on the RISC-V instruction set specifically further includes a device driver, and the device driver includes an Ethernet module, a synchronous transceiver module, an asynchronous transceiver module, a digital-to-analog conversion module, and a security input and output module.
  • USB interface module USB interface module, GPIO interface, I2C interface, SPI interface; wherein, the Ethernet module is used for data transmission in the form of Ethernet, the synchronous transceiver module, the asynchronous transceiver module, the USB interface module, the The GPIO interface, the I2C interface, and the SPI interface are all used as interfaces for data transfer, and the secure input and output module is used for memory data to communicate with the synchronous transceiver module, the asynchronous transceiver module, and the USB interface.
  • Information transfer between modules, the GPIO interface, the I2C interface, and the SPI interface, and the digital-to-analog conversion module is used to convert the collected data between analog and digital.
  • the cooperation at the hardware level and the protocol stack layer at the software level is realized, thereby realizing the multi-type interface. Compatible pass-through of interface data.
  • the gateway adaptation layer includes access to multiple network interfaces, specifically including WAN, LAN, RS485, Wi-Fi, Zigbee, 4G/5G, LTE-V, BLE4.1, NB-IoT, Sigfox, LoRa, 6LoPWAN.
  • network interfaces specifically including WAN, LAN, RS485, Wi-Fi, Zigbee, 4G/5G, LTE-V, BLE4.1, NB-IoT, Sigfox, LoRa, 6LoPWAN.
  • the access of multi-standard network data has many advantages for designing a multi-standard communication protocol compatible architecture, including massive access technologies that can meet long-distance, short-distance, wired, wireless and other communication standards, It is compatible with data transmission of various communication standards such as WAN, LAN, RS485, Wi-Fi, Zigbee, 4G/5G, LTE-V, BLE4.1, NB-IoT, Sigfox, LoRa, 6LoPWAN, etc.
  • various communication standards such as WAN, LAN, RS485, Wi-Fi, Zigbee, 4G/5G, LTE-V, BLE4.1, NB-IoT, Sigfox, LoRa, 6LoPWAN, etc.
  • WAN Wide Area Network
  • LAN Local Area Network
  • 5G 5th generation mobile networks
  • Zigbee Zigbee is a new type of wireless communication technology, suitable for short-range data transmission.
  • 4G 4th generation mobile networks 4th generation cellular communication network
  • LTE-V is specifically Long Term Evolution technology-vehicle communication
  • BLE is called Bluetooth Low Energy, BLE Bluetooth Low Energy
  • NB-IoT is called Narrow Band Internet of Things, NB-IoT stands for Narrow Band Internet of Things
  • Sigfox refers to Sigfox's deployment of low-power wide area networks around the world to provide IoT connection services, and user equipment integration supports Sigfox protocol 6LoPWAN is called IPv6over Low-Power Wireless Personal Area Networks, the full name of LoRa is Long Range, which is a long-distance transmission narrowband Internet of Things. IPv6-based private wireless LAN.
  • FIG. 5 is a structural diagram of a cloud collaboration platform in an AIOT multi-standard edge gateway communication system based on the RISC-V instruction set according to an embodiment of the present invention.
  • the cloud collaboration platform includes an edge end node and a cloud center
  • the edge end node includes a management module, a data operation module, and an application service module , a book storage module
  • the cloud center includes an edge intelligent management suite and an artificial intelligence application service, wherein the artificial intelligence application service cooperates with the intelligent algorithm layer to perform image processing, function operation, network training, stream computing, Big data mining
  • the edge intelligence management suite includes program, resource and application management.
  • YOLO V2, YOLO V3, YOLO V2-tiny, YOLO V3-tiny, MobileNet, MobileNetV2, SqueezeNet, GoogleNet, ShuffleNet, Xception, etc. face recognition, object detection, classification and other fields of intelligent computing, and on this basis, realize the neural network device-cloud collaborative reasoning algorithm model optimization, device-cloud computing collaborative dynamic task division strategy, realize the intelligent computing of edge devices and cloud collaboration, improve Application service efficiency and reliability.
  • YOLO the full name of You Only Look Once, is a neural network model.
  • FIG. 6 is a schematic diagram of an AIOT multi-standard edge gateway communication system based on a RISC-V instruction set according to an embodiment of the present invention. As shown in FIG. 6 , it is a specific implementation manner, wherein the AI algorithm framework includes: YOLOV3, MobileNet and other algorithm models.
  • the IOT gateway adaptation layer includes: Zigbee, LORA and other IoT communication methods; security encryption layer: DTLA (Digital Transmission Authorization Management), AES (Advanced Encryption Standard), CA (Certificate Authority, Certificate Authority) is the management and Network institutions that issue security certificates and encrypted information security keys, SSL (Secure Sockets Layer, Secure Sockets Protocol), Transport Layer Security (Transport Layer Security, TLS) is a kind of security and data integrity for network communication.
  • security encryption layer DTLA (Digital Transmission Authorization Management), AES (Advanced Encryption Standard), CA (Certificate Authority, Certificate Authority) is the management and Network institutions that issue security certificates and encrypted information security keys
  • SSL Secure Sockets Layer, Secure Sockets Protocol
  • Transport Layer Security Transport Layer Security
  • SHA265 Secure Hash Algorithm (English: Secure Hash Algorithm, abbreviated as SHA) is a family of cryptographic hash functions
  • HASH hash, is the input of any length, also known as pre-map
  • MD5 MD5 Message-Digest Algorithm
  • MD5 Message-Digest Algorithm a widely used cryptographic hash function that produces a 128-bit hash Column value used to ensure complete and consistent information transfer.
  • MQTT Message Queue Telemetry Transmission
  • CoAP The Constrained Application Protocol
  • CoAP is the application layer protocol in the 6LowPAN protocol stack
  • TCP/IP Transmission Control Protocol/Internet Protocol
  • UDP protocol User Datagram Protocol
  • URP Address Resolution Protocol, full name Address Resolution Protocol
  • DHCP Dynamic Host Configuration Protocol
  • a network protocol for a local area network which means that a range of IP addresses is controlled by the server, and the client logs in to the server.
  • IPv4 Internet Protocol version 4 (English: Internet Protocol version 4, IPv4), also known as the fourth version of the Internet Communication Protocol, which is the fourth revision in the development process of the Internet Protocol. It is also the first widely deployed version of this protocol.
  • IPv6 full English name Internet Protocol Version 6, Chinese name: Internet Protocol Version 6
  • NVME full name: NVM Express is an open collection of standards and information to fully demonstrate the benefits of non-volatile memory in all types of computing environments, from mobile devices to data centers.
  • Radio refers to the audio module used to manage the embedded system
  • Powr refers to the power management module used to manage the embedded system
  • Memory refers to the management module used to manage the memory in the embedded system
  • Time refers to It is used to manage timers and time-triggered task function modules in embedded systems
  • Client refers to the user group management module used in embedded systems
  • KeyManager refers to the user key management module used to manage embedded systems
  • Radio management sub-module Powr management sub-module, Memory management sub-module, Time management sub-module, Client management sub-module, KeyManager management sub-module
  • Ethernet refers to the Ethernet module on the circuit board
  • UARTs refers to the synchronous and asynchronous transceiver modules on the circuit board, also known as serial ports
  • ADC refers to the analog-to-digital converter on the circuit board, which is used to implement analog circuits Conversion to digital circuits
  • SDIO Secure Digital Input and Output
  • secure digital input and output refers to the peripheral interface on the circuit board, used to realize the communication interface bus with the memory card
  • USB is the universal serial bus, A common communication bus used to realize high-speed communication between peripherals and the main control
  • GPIO is a general-purpose programmable input/output control port, used to realize programmable control of digital circuits
  • I2C is a simple, Bidirectional two-wire synchronous serial bus.
  • SPI is a full-duplex synchronous serial bus developed by Motorola, which is a synchronous serial port for communication between a microprocessor control unit (MCU) and peripheral devices;
  • MCU microprocessor control unit
  • AI Uint is an AI-specific circuit.
  • an AIOT multi-standard edge gateway communication method based on a RISC-V instruction set is provided.
  • FIG. 7 is a flowchart of an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set according to an embodiment of the present invention.
  • the AIOT multi-standard edge gateway communication method based on the RISC-V instruction set specifically includes:
  • processing data includes IoT node data, image information, and sound information
  • S702. Perform data security verification on all data obtained through the device driver layer, and after verification and preprocessing, divide the data into training data and verification data, compress the training data, perform heterogeneous operations, and generate data Analyze the model and store it;
  • S703 according to the edge hardware computing device, perform parameter quantification evaluation to obtain the evaluation accuracy rate, and re-optimize and compress when the evaluation accuracy rate decreases;
  • S704 Adjust the speed of the completed calculation, so that different hardware acceleration chips can complete different edge operations, and output supporting codes.
  • a series of tasks such as data processing, model training and analysis, hardware device configuration, and calculation speed optimization are performed through the embedded operation layer combined with a specific instruction set processor.
  • parameter quantization evaluation is performed to obtain the evaluation accuracy rate, and when the evaluation accuracy rate decreases, the Flowchart for optimization and compression.
  • parameter quantization evaluation is performed to obtain an evaluation accuracy rate, and when the evaluation accuracy rate decreases, optimization and compression are performed again, specifically including: :
  • the model when using the model, it is necessary to first evaluate the accuracy of the pre-trained model.
  • the specific evaluation mainly depends on the determination of pruning and quantization parameters, and the model is based on this.
  • the generation of accuracy when the accuracy of the model is obtained, it is fine-tuned by comparing with the threshold value to realize the online self-adaptive adjustment of the pre-trained model.
  • Fig. 9 is an embodiment of the present invention, in an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set, according to the edge hardware computing device, parameter quantization evaluation is performed to obtain the evaluation accuracy rate, and when the evaluation accuracy rate decreases, the Schematic diagram of optimization and compression.
  • the current model is optimized and compressed according to the deployment hardware.
  • the compression ratio and quantization parameters of the pruning are determined according to the edge hardware computing device.
  • the model is then fine-tuned and quantized, and the accuracy and size are evaluated. If the accuracy rate is greatly reduced after pruning and quantization, the pruning and quantization parameters will be adjusted again for recompression.
  • 10 is an embodiment of the present invention, in an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set, the speed adjustment is performed on the completed calculation, so that different hardware acceleration chips can complete different edge operations and output supporting codes. flow chart.
  • the speed adjustment is performed on the completed calculation, so that different hardware acceleration chips can complete different edge operations and output supporting codes, which specifically includes:
  • S1003 utilize hardware acceleration to calculate scheduling configuration, and generate C++ inference code and bmodel file;
  • the edge computing model needs to be encapsulated before performing edge computing. Therefore, in this way, a total edge computing model is generated for each function type and encapsulated.
  • FIG. 11 is an embodiment of the present invention, in an AIOT multi-standard edge gateway communication method based on the RISC-V instruction set, speed adjustment is performed on the completed calculation, so that different hardware acceleration chips can complete different edge operations and output supporting codes.
  • schematic diagram As shown in Figure 11, for the selected computing scheduling configuration, it is decomposed into CPU computing scheduling configuration and hardware acceleration chip computing scheduling configuration. The five types are all RISC-V chip manufacturers and chip models. The specific back-end computing support codes or models are produced respectively, and they are packaged into the edge computing model together with the sub-graph description.
  • FIG. 12 is a structural diagram of an electronic device in an embodiment of the present invention.
  • the electronic device shown in FIG. 12 is a general multi-standard edge gateway communication device, which includes a general computer hardware structure, which at least includes a processor 1201 and a memory 1202 .
  • the processor 1201 and the memory 1202 are connected by a bus 1203 .
  • the memory 1202 is adapted to store instructions or programs executable by the processor 1201 .
  • the processor 1201 may be an independent microprocessor, or may be a set of one or more microprocessors.
  • the processor 1201 executes the instructions stored in the memory 1202, thereby executing the above-mentioned method flow of the embodiment of the present invention to process data and control other devices.
  • the bus 1203 connects the above-mentioned various components together, while connecting the above-mentioned components to the display controller 1204 and the display device and the input/output (I/O) device 1205 .
  • the input/output (I/O) device 1205 may be a mouse, keyboard, modem, network interface, touch input device, somatosensory input device, printer, and other devices known in the art.
  • input/output devices 1205 are connected to the system through input/output (I/O) controllers 1206 .
  • the technical solutions provided by the embodiments of the present invention are based on the RISCV instruction set and an edge computing controller integrating a CNN hardware accelerator.
  • the CNN hardware accelerator is used to implement convolution processing on data in designated memories such as ROM and RAM, and solves the problem of edge-side CNN. Convolution operation is slow, heavy load, and high overhead.
  • the specific beneficial effects can include:
  • This solution adopts an open source instruction set architecture and uses artificial intelligence to process data, which can realize online access, analysis, processing and calculation of a large number of node data, with high efficiency and strong compatibility;
  • This solution realizes the functions of data-aware computing, data fusion and node broadcasting in different scenarios through various application scenarios such as load vehicle-road coordination, and meets the computing performance requirements of high-load systems.
  • embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, optical storage, and the like.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions
  • the apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Computer Security & Cryptography (AREA)
  • Computing Systems (AREA)
  • Computer Hardware Design (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

本发明提供了一种基于RISC-V指令集的AIOT多制式边缘网关通信系统及设备。该方案包括云端协同平台、智能算法层、安全加密层、管理层、网关适配层、协议栈层、设备驱动层、嵌入式操作层、指令集处理器；其中，云端协同平台与智能算法层进行数据交互，安全加密层、管理层、网关适配层和协议栈层分别对被交互的数据进行安全加密、数据管理、协议转换、网络传输，嵌入式操作层其上设置有指令集处理器，嵌入式操作层其上配置有设备驱动层，设备驱动层进行与外部传输设备的功能交互。该方案通过基于开源指令集架构设计，结合人工智能数据处理算法，利用多种制式物联网通信协议和多维度的组模，形成了边缘网关和设备端的功能交互，提升工作效率和数据兼容性。

Description

一种基于RISC-V指令集的AIOT多制式边缘网关通信系统及设备 技术领域
本发明涉及网络通信技术领域，更具体地，涉及一种基于RISC-V指令集的AIOT多制式边缘网关通信系统及设备。
背景技术
物联网是新一代信息产业的重要组成部分。随着人工智能的出现与人工智能硬件加速器的出现，衍生出了一门新的领域AIOT，AIOT是信息化与人工智能相结合的技术领域。AIOT能更加全面地为人类生产生活服务的重大升级，应用前景巨大。
随着物联网技术的发展，传统云侧的数据中心与上千成万的物联网节点进行数据交互，不仅计算资源开销巨大，而且存在网络拥堵，且在高峰时段更为明显。因此，现各行业均在探索新型“去云端化”的边缘端AI计算，从而提供物联网系统的安全和性能。
目前，现有的物联网技术中，主要通过5G边缘计算网关放置在靠近数据采集设备的边缘，进行本地边缘计算，主控处理模块通过USB与5G通信模组通信，并被配置为通过5G通信模组与拾音器、扬声器、摄像机相连，进行语音采集、语音播放、图像采集、数据运算处理；通过摄像头进行图像实时采集并输送至神经网络处理器进行卷积神经网络卷积计算，实现车辆信息采集、检测、识别，人脸识别、检测、人体姿态检测工作。但是，上述方式一方面，仅提供已编译加算好的第三方库供调用，算法比较封闭，不利于在AIOT应用场景下实现自定义算子开发以及模型优化；另一方面，兼容性较差，仅能够接入5G基站，而无法实现接入边缘节点。
发明内容
鉴于上述问题，本发明提出了一种基于RISC-V指令集的AIOT多制式边缘网关通信系统及设备，通过基于开源指令集架构的设计，结合人工智能数据处理算法，并利用多种制式的物联网通信协议和多维度的组模，形成了边缘网关和设备端的功能交互，提升工作效率和数据兼容性。
根据本发明实施例第一方面，提供一种基于RISC-V指令集的AIOT多制式边缘网关通信系统。
所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统具体包括：
云端协同平台、智能算法层、安全加密层、管理层、网关适配层、协议栈层、设备驱动层、嵌入式操作层、指令集处理器；其中，所述云端协同平台与所述智能算法层进行数据交互，所述安全加密层、所述管理层、所述网关适配层和所述协议栈层分别对被交互的数据进行安全加密、数据管理、协议转换、网络传输，所述嵌入式操作层其上设置有所述指令集处理器，所述嵌入式操作层其上配置有所述设备驱动层，所述设备驱动层进行与外部传输设备的功能交互。
具体的，RISC-V指令集全称为Reduced Instruction Set Computing V，中文名为第五代RISC精简指令集计算机。
在一个或多个实施例中，优选地，所述安全加密层具体包括DTLA子模块、AES子模块、CA子模块、SSL子模块、SHA265子模块、HASH子模块、MD5子模块；其中，所述DTLA子模块用于数字传输授权管理，所述AES子模块用于高级别加密，所述CA子模块用于管理签发安全凭证和加密信息安全密钥，所述SSL子模块用于为提供安全传输协议，所述SHA265子模块用于采用密码散列进行安全加密；所述HASH子模块用于通过散列算法将任意长度的输入映射为固定长度的散列值；所述MD5子模块用于产生一个16字节的散列值。
在一个或多个实施例中，优选地，所述协议栈层包括MQTT传输子模块、CoAP传输协议子模块、TCP/IP传输协议子模块、UDP传输协议子模块、ARP传输协议子模块、DHCP传输协议子模块、IPv4传输协议子模块、IPv6传输协议子模块、NVME传输协议子模块、BLE传输协议子模块；其中，所述MQTT传输子模块用于将传输数据以消息队列遥测传输标准下发；所述CoAP传输协议子模块用于将数据以CoAP传输协议进行数据转发；所述TCP/IP传输协议子模块用于将网络间的传输数据以TCP/IP传输协议转发，所述IPv4传输协议子模块和所述IPv6传输协议子模块分别用于不同级别的IP地址形式生成；所述NVME传输协议子模块用于通过非易失去性存储器在移动设备和数据中心中对信息进行开放收集；所述UDP传输协议子模块用于将数据以UDP传输协议转发；所述ARP传输协议子模块用于将数据以ARP传输协议转发，所述DHCP传输协议子模块用于将数据以DHCP传输协议在局域网内转发。
在一个或多个实施例中，优选地，所述管理层包括Radio管理子模块、Powr管理子模块、Memory管理子模块、Time管理子模块、Client管理子模块、KeyManager管理子模块、算法管理子模块；其中，所述Radio管理子模块用于管理嵌入式系统中的音频模块；所述Powr管理子模块用于管理嵌入式系统中的电源管理模块；所述Memory管理子模块用于管理嵌入式系统中的内存的管理模块；所述Time管理子模块用于管理嵌入式系统中的定时器、时间触发型任务功能模块；所述Client管理子模块用于嵌入式系统中的用户群管理模块；所述KeyManager管理子模块用于管理嵌入式系统中的用户密钥管理模块；所述算法管理子模块用于管理采用的AI智能算法，其中，所述AI智能算法包括YOLOV3和MobileNet。
在一个或多个实施例中，优选地，所述网关适配层包括多种网络接口的接入，具体的包括WAN、LAN、RS485、Wi-Fi、Zigbee、4G/5G、LTE-V、BLE4.1、NB-IoT、Sigfox、LoRa、6LoPWAN。
在一个或多个实施例中，优选地，所述云端协同平台包括边缘端节点和云中心，所述边缘端节点包括管理组模、数据运算组模、应用服务组模、书存储组模，所述云中心包括边缘智能管理套件和人工智能应用服务，其中，所述人工智能应用服务与所述智能算法层配合进行图像处理、函数运算、网络训练、流式计算、大数据挖掘；所述边缘智能管理套件包括程序、资源和应用程序管理。
根据本发明实施例第二方面，提供一种基于RISC-V指令集的AIOT多制式边缘网关通信方法。
所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法具体包括：
通过嵌入式操作层及其上配置的指令集处理器进行卷积运算，生成处理数据，其中处理数据包括物联网节点数据、图像信息、声音信息；
对于通过所述设备驱动层获取的全部数据进行数据安全性校验，并在校验和预处理后将数据分为训练数据和验证数据，压缩训练数据，进行异构运算，生获数据分析模型并存储；
根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩；
对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码。
在一个或多个实施例中，优选地，所述根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩，具体包括：
输入预训练模型；
根据硬件参数和所述预训练模型确定剪枝和量化参数；
根据所述剪枝和量化参数调整所述预训练模型；
对所述预训练模型进行量化生成模型准确度；
当所述模型准确度低于预设正确度门槛值时，重新确定剪枝和量化参数；
将微调后的所述预训练模型保存为压缩模型。
在一个或多个实施例中，优选地，所述对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码，具体包括：
边缘计算模型封装；
获取当前的调度配置信息，并通过CPU进行计算调度配置；
利用硬件加速计算调度配置，并生成C++推理代码和bmodel文件；
生成计算调度子图，并将所述调度配置、所述bmodel文件、所述C++推理代码和所述调度子图共同并存储为边缘计算模型。
根据本发明实施例第三方面，提供一种电子设备，包括存储器和处理器，所述存储器用于存储一条或多条计算机程序指令，其中，所述一条或多条计算机程序指令被所述处理器执行以实现如本发明实施例第一方面中所述的步骤。
本发明的实施例提供的技术方案可以包括以下有益效果：
1)本方案采用了开源指令集架构，利用人工智能处理数据，可以实现大量的节点数据的在线接入、分析、处理和运算，效率高且兼容性强；
2)本方案集成了全部的物联网节点的网关功能，通过多制式、多协议、多功能的设备，提升数据感知和数据处理的速度；
3)本方案通过采用大量的模块化设计，使得外部的硬件组模能够快速接入到系统中，实现便捷的物联网应用；
4)本方案通过负荷车路协调等多种应用场景，实现了不同场景下的数据感知运算、数据融合和节点广播功能，满足了高负载系统下的运算性能需求。
本发明的其它特征和优点将在随后的说明书中阐述，并且，部分地从说明书中变得显而易见，或者通过实施本发明而了解。本发明的目的和其他优点可通过在所写的说明书、权利要求书、以及附图中所特别指出的结构来实现和获得。
下面通过附图和实施例，对本发明的技术方案做进一步的详细描述。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统的结构图。
图2是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统中的安全加密层的结构图。
图3是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统中的协议栈层的结构图。
图4是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统中的管理层的结构图。
图5是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统中的云端协同平台的结构图。
图6是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统的示意图。
图7是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法的流程图。
图8是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩的流程图。
图9是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的根据边缘硬件计算设备，进行参数量化评估获得评估正 确率，当评估正确率降低时，重新进行优化与压缩的示意图。
图10是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码的流程图。
图11是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码的示意图。
图12是本发明一个实施例中一种电子设备的结构图。
具体实施方式
在本发明的说明书和权利要求书及上述附图中的描述的一些流程中，包含了按照特定顺序出现的多个操作，但是应该清楚了解，这些操作可以不按照其在本文中出现的顺序来执行或并行执行，操作的序号如101、102等，仅仅是用于区分开各个不同的操作，序号本身不代表任何的执行顺序。另外，这些流程可以包括更多或更少的操作，并且这些操作可以按顺序执行或并行执行。需要说明的是，本文中的“第一”、“第二”等描述，是用于区分不同的消息、设备、模块等，不代表先后顺序，也不限定“第一”和“第二”是不同的类型。
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
物联网是新一代信息产业的重要组成部分。随着人工智能的出现与人工智能硬件加速器的出现，衍生出了一门新的领域AIOT，AIOT是信息化与人工智能相结合的技术领域。AIOT能更加全面地为人类生产生活服务的重大升级，应用前景巨大。
随着物联网技术的发展，传统云侧的数据中心与上千成万的物联网节点进行数据交互，不仅计算资源开销巨大，而且存在网络拥堵，且在高峰时段更为明显。因此，现各行业均在探索新型“去云端化”的边缘端AI计算，从而提供物联网系统的安全和性能。
目前，现有的物联网技术中，主要通过5G边缘计算网关放置在靠近数据采集设备的边缘，进行本地边缘计算，主控处理模块通过USB与5G通信模组通信，并被配置为通过5G通信模组与拾音器、扬声器、摄像机相连，进行语音采集、语音播放、图像采集、数据运算处理；通过摄像头进行图像实时采集并输送至神经网络处理器进行卷积神经网络卷积计算，实现车辆信息采集、检测、识别，人脸识别、检测、人体姿态检测工作。但是，上述方式一方面，仅提供已编译加算好的第三方库供调用，算法比较封闭，不利于在AIOT应用场景下实现自定义算子开发以及模型优化；另一方面，兼容性较差，仅能够接入5G基站，而无法实现接入边缘节点。
本发明实施例中，提供了一种基于RISC-V指令集的AIOT多制式边缘网关通信系统及设备。该方案通过基于开源指令集架构的设计，结合人工智能数据处理算法，并利用多种制式的物联网通信协议和多维度的组模，形成了边缘网关和设备端的功能交互，提升工作效率和数据兼容性。
根据本发明实施例第一方面，提供一种基于RISC-V指令集的AIOT多制式边缘网关通信系统。
图1是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统的结构图。
如图1所示，所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统具体包括：
云端协同平台101、智能算法层102、安全加密层103、管理层104、网关适配层105、协议栈层106、设备驱动层107、嵌入式操作层108、指令集处理器109；其中，所述云端协同平台101与所述智能算法层102进行数据 交互，所述安全加密层103、所述管理层104、所述网关适配层105和所述协议栈层106分别对被交互的数据进行安全加密、数据管理、协议转换、网络传输，所述嵌入式操作层108其上设置有所述指令集处理器109，所述嵌入式操作层108其上配置有所述设备驱动层107，所述设备驱动层107进行与外部传输设备的功能交互。
具体的，所述指令集指的是RISC-V，英文全称为Reduced Instruction Set Computing V是第五代RISC(精简指令集计算机)；IOT/AIOT的英文全称为Internet of Things/AI+Internet of Things，对应中文名为物联网/人工智能物联网。
在本发明实施例中，一方面，通过多层次、多类型模块将人工智能处理数据引入到了整个数据处理架构中，同时利用了开源指令集架构，将大量的节点数据进行分析、处理和运算；另一方面采用了模块化的设计，使得不同的模块执行不同类型功能，简化的数据处理的流程，增加的外部硬件接入系统的可能性和效率。
图2是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统中的安全加密层的结构图。
如图2所示，在一个或多个实施例中，优选地，所述安全加密层103具体包括DTLA子模块201、AES子模块202、CA子模块203、SSL子模块204、SHA265子模块205、HASH子模块206、MD5子模块207；其中，所述DTLA子模块201用于数字传输授权管理，所述AES子模块202用于高级别加密，所述CA子模块203用于管理签发安全凭证和加密信息安全密钥，所述SSL子模块204用于为提供安全传输协议，所述SHA265子模块205用于采用密码散列进行安全加密；所述HASH子模块206用于通过散列算法将任意长度的输入映射为固定长度的散列值；所述MD5子模块207用于产生一个16字节的散列值。
在本发明实施例中，通过多类型数据加密子模块，对输入的任意长度的 数据和信息进行数据处理，并加密为满足数据传输协议和数据加密标准的传输数据，方便数据的后续处理，并保障了整个系统的安全运行。
图3是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统中的协议栈层的结构图。
如图3所示，在一个或多个实施例中，优选地，所述协议栈层106包括MQTT传输子模块301、CoAP传输协议子模块302、TCP/IP传输协议子模块303、UDP传输协议子模块304、ARP传输协议子模块305、DHCP传输协议子模块306、IPv4传输协议子模块307、IPv6传输协议子模块308、NVME传输协议子模块309、BLE传输协议子模块310；其中，所述MQTT传输子模块301用于将传输数据以消息队列遥测传输标准下发；所述CoAP传输协议子模块302用于将数据以CoAP传输协议进行数据转发；所述TCP/IP传输协议子模块303用于将网络间的传输数据以TCP/IP传输协议转发，所述IPv4传输协议子模块307和所述IPv6传输协议子模块308分别用于不同级别的IP地址形式生成；所述NVME传输协议子模块309用于通过非易失去性存储器在移动设备和数据中心中对信息进行开放收集；所述UDP传输协议子模块304用于将数据以UDP传输协议转发；所述ARP传输协议子模块305用于将数据以ARP传输协议转发，所述DHCP传输协议子模块306用于将数据以DHCP传输协议在局域网内转发。
在本发明实施例中，通过多种类型的数据协议，将实际接收到的多类型的数据进行不同类型协议的数据转发，通过此种方式可以实现对于物联网中多类型数据的兼容，当存在对应类型协议数据时，则直接根据对应的模块进行数据的转换，并根据不同类型数据转发子模块的配置按照指定的传输数据的标准进行数据的发送和转移。
图4是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统中的管理层的结构图。
如图4所示，在一个或多个实施例中，优选地，所述管理层104包括Radio 管理子模块401、Powr管理子模块402、Memory管理子模块403、Time管理子模块404、Client管理子模块405、KeyManager管理子模块406、算法管理子模块407；其中，所述Radio管理子模块401用于管理嵌入式系统中的音频模块；所述Powr管理子模块402用于管理嵌入式系统中的电源管理模块；所述Memory管理子模块403用于管理嵌入式系统中的内存的管理模块；所述Time管理子模块404用于管理嵌入式系统中的定时器、时间触发型任务功能模块；所述Client管理子模块405用于嵌入式系统中的用户群管理模块；所述KeyManager管理子模块406用于管理嵌入式系统中的用户密钥管理模块；所述算法管理子模块407用于管理采用的AI智能算法，其中，所述AI智能算法包括YOLOV3和MobileNet。
在本发明实施例中，所述的管理层是整个多制式通信的核心，通过管理层将电源供给各个子模块，并进行数据的存储、音频数据的存储、时间的记录与触发，并通过管理层进行与用户的交流，所述KeyManager管理子模块还能够通过所述安全加密层进行数据的加密并进行对于的用户的密匙的管理，保证系统安全。
本发明实施例中，对应的还包括了AI智能算法的管理，通过设置至少包括YOLOV3和MobileNet的智能算法，实现对于不同类型的输入数据的快速的识别特征，并根据特征进行相应能力的训练。
所述的基于RISC-V指令集的AIOT多制式边缘网关通信系统具体的还包括设备驱动，所述设备驱动包括以太网模块、同步收发模块、异步收发模块、数模转换模块、安全输入输出模块、USB接口模块、GPIO接口、I2C接口、SPI接口；其中，所述以太网模块用于数据以以太网的方式传递，所述同步收发模块、所述异步收发模块、所述USB接口模块、所述GPIO接口、所述I2C接口和所述SPI接口都用于作为数据传递的接口，所述安全输入输出模块用于进行内存数据与所述同步收发模块、所述异步收发模块、所述USB接口模块、所述GPIO接口、所述I2C接口、所述SPI接口之间的信息 传递，所述数模转换模块用于对收集到的数据进行模拟量与数字量之间的转换。
在本发明实施例中，通过多种类型的接口和接口数据类型的转化的硬件模块之间配合，实现了在硬件层面上与所述协议栈层在软件层面上的配合，进而实现对于多类型接口的数据的兼容传递。
在一个或多个实施例中，优选地，所述网关适配层包括多种网络接口的接入，具体的包括WAN、LAN、RS485、Wi-Fi、Zigbee、4G/5G、LTE-V、BLE4.1、NB-IoT、Sigfox、LoRa、6LoPWAN。
在本发明实施例中，多制式的网络数据的接入对设计多制式通信协议兼容架构有很多优势，包括可满足长距离、短距离，有线、无线等多种通信制式的海量接入技术，并可兼容WAN、LAN、RS485、Wi-Fi、Zigbee、4G/5G、LTE-V、BLE4.1、NB-IoT、Sigfox、LoRa、6LoPWAN等多种通信制式的数据传输。
具体的，WAN全称为Wide Area Network广域网；LAN全称为Local Area Network局域网；5G全称为5th generation mobile networks第5代蜂窝通讯网络；Zigbee是一项新型的无线通信技术，适用于传输范围短数据传输速率低的一系列电子元器件设备之间；4G全称为4th generation mobile networks第4代蜂窝通讯网络；LTE-V具体为长期演进技术-车辆通信；BLE全称为Bluetooth Low Energy，或称Bluetooth LE、BLE低功耗蓝牙；NB-IoT全称为Narrow Band Internet of Things,NB-IoT表示窄带物联网；Sigfox是指Sigfox公司在全球部署低功耗广域网，提供物联网连接服务，用户设备集成支持Sigfox协议的射频模块或者芯片，开通连接服务后，即可连接到Sigfox网络；LoRa全称为Long Range，具体为一种长距离传输的窄带物联网；6LoPWAN全称为IPv6over Low-Power Wireless Personal Area Networks，具体为基于IPv6的私人无线局域网。
图5是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边 缘网关通信系统中的云端协同平台的结构图。
如图5所示，在一个或多个实施例中，优选地，所述云端协同平台包括边缘端节点和云中心，所述边缘端节点包括管理组模、数据运算组模、应用服务组模、书存储组模，所述云中心包括边缘智能管理套件和人工智能应用服务，其中，所述人工智能应用服务与所述智能算法层配合进行图像处理、函数运算、网络训练、流式计算、大数据挖掘；所述边缘智能管理套件包括程序、资源和应用程序管理。
在本发明实施例中，通过端云协同智能计算模型的基于YOLO V2、YOLO V3、YOLO V2-tiny、YOLO V3-tiny、MobileNet、MobileNetV2、SqueezeNet、GoogleNet、ShuffleNet、Xception等神经网络模型的视觉处理、人脸识别、物体检测、分类等领域智能运算，并在此基础上实现神经网络端云协同推理算法模型优化、端云计算协同动态任务划分策略，实现边缘设备与云端协同的智能计算，提高应用服务效率和可靠性。其中，YOLO，全称为You Only Look Once，为一种神经网络模型。
图6是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统的示意图。如图6所示，为一种具体的实施方式，其中，AI算法框架包括：YOLOV3、MobileNet等算法模型。IOT网关适配层包括：Zigbee、LORA等物联通讯方式；安全加密层：DTLA(数字传输授权管理)、AES(Advanced Encryption Standard高级加密标准)、CA(Certificate Authority，证书授权中心)是管理和签发安全凭证和加密信息安全密钥的网络机构、SSL(Secure Sockets Layer，安全套接字协议)，传输层安全协议(Transport Layer Security，TLS)是为网络通信提供安全及数据完整性的一种安全协议、SHA265(安全散列算法(英语：Secure Hash Algorithm，缩写为SHA)是一个密码散列函数家族)、HASH(哈希，是把任意长度的输入，又叫做预映射)通过散列算法变换成固定长度的输出，该输出就是散列值、MD5(MD5信息摘要算法(英语：MD5 Message-Digest Algorithm)，为一种 被广泛使用的密码散列函数，可以产生出一个128位的散列值，用于确保信息传输完整一致。
协议栈层：MQTT(消息队列遥测传输)是ISO标准下基于发布/订阅范式的消息协议、CoAP(The Constrained Application Protocol,CoAP是6LowPAN协议栈中的应用层协议)、TCP/IP((Transmission Control Protocol/Internet Protocol，传输控制协议/网际协议)是指能够在多个不同网络间实现信息传输的协议簇)、UDP协议为用户数据报协议，全称为User Datagram Protocol、ARP(地址解析协议，全称Address Resolution Protocol)，是根据IP地址获取物理地址的一个TCP/IP协议、DHCP(动态主机配置协议)是一个局域网的网络协议，指的是由服务器控制一段IP地址范围，客户机登录服务器时就可以自动获得服务器分配的IP地址和子网掩码、IPv4(网际协议版本4(英语：Internet Protocol version 4，IPv4)，又称互联网通信协议第四版，是网际协议开发过程中的第四个修订版本，也是此协议第一个被广泛部署的版本。)、IPv6(英文全称Internet Protocol Version 6，中文名：互联网协议第6版)是互联网工程任务组设计的用于替代IPv4的下一代IP协议)、NVME(全称：NVM Express是标准和信息的开放收集，以充分展示非易失性存储器在从移动设备到数据中心的所有类型的计算环境中的优势。
管理层：Radio指用于管理嵌入式系统中的音频模块；Powr是指用于管理嵌入式系统中的电源管理模块；Memory是指用于管理嵌入式系统中的内存的管理模块；Time是指用于管理嵌入式系统中的定时器、时间触发型任务功能模块；Client是指用于嵌入式系统中的用户群管理模块；KeyManager是指用于管理嵌入式系统中的用户密钥管理模块；
Radio管理子模块、Powr管理子模块、Memory管理子模块、Time管理子模块、Client管理子模块、KeyManager管理子模块
设备驱动层：Ethernet是指电路板上的以太网模块；UARTs是指电路板上的同步及异步收发器模块，亦称为串口；ADC是指电路板上模数 转化器，用于实现模拟电路至数字电路的转化；SDIO(Secure Digital Input and Output)中文名称：安全数字输入输出，是指电路板上的外设接口，用于实现与内存卡进行通信接口总线；USB即通用串行总线，一种常见的通讯总线，用于实现外设与主控之间的高速通讯的总线；GPIO即通用可编程输入/输出控制口，用于实现对数字电路可编程控制；I2C即一种简单、双向二线制同步串行总线。它只需要两根线即可在连接于总线上的器件之间传送信息。SPI是由摩托罗拉公司开发的全双工同步串行总线，是微处理控制单元(MCU)和外围设备之间进行通信的同步串行端口；AI Uint即AI专用电路。
根据本发明实施例第二方面，提供一种基于RISC-V指令集的AIOT多制式边缘网关通信方法。
图7是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法的流程图。
如图7所示，所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法具体包括：
S701、通过嵌入式操作层及其上配置的指令集处理器进行卷积运算，生成处理数据，其中处理数据包括物联网节点数据、图像信息、声音信息；
S702、对于通过所述设备驱动层获取的全部数据进行数据安全性校验，并在校验和预处理后将数据分为训练数据和验证数据，压缩训练数据，进行异构运算，生获数据分析模型并存储；
S703、根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩；
S704、对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码。
在本发明实施例中，通过嵌入式操作层结合具体的指令集处理器进行了数据的处理，模型的训练分析，硬件设备的配置化和运算速度优化等一系列工作，通过上述的方法，可以充分的利用多制式的边缘网关，实现对于数据 的快速处理、快速分类、快速评估、快速计算，进而有效提高兼容性和运算效率。
图8是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩的流程图。
如图8所示，在一个或多个实施例中，优选地，所述根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩，具体包括：
S801、输入预训练模型；
S802、根据硬件参数和所述预训练模型确定剪枝和量化参数；
S803、根据所述剪枝和量化参数调整所述预训练模型；
S804、对所述预训练模型进行量化生成模型准确度；
S805、当所述模型准确度低于预设正确度门槛值时，重新确定剪枝和量化参数；
S806、将微调后的所述预训练模型保存为压缩模型。
在本发明实施例中，在进行模型使用时，需要先对预训练模型的准确度进行一定的评估，具体的评估主要依靠了对于剪枝和量化参数的确定，并在此基础上进行了模型准确度的生成，当获得模型准确度后通过与门槛值对比的方式，进行微调进而实现对于预训练模型的在线的自适应的调整。
图9是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩的示意图。如图9所示，根据部署硬件情况，对当前模型进行优化和压缩，首先根据边缘硬件计算设备，对剪枝的压缩比例以及量化参数进行确定。然后对模型进行微调和量化，并且评估正确率和大小。如果正确率经过剪枝和量化后大幅降低，将再次调整剪枝和量化参数进行重新压缩。
图10是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码的流程图。
在一个或多个实施例中，优选地，所述对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码，具体包括：
S1001、边缘计算模型封装；
S1002、获取当前的调度配置信息，并通过CPU进行计算调度配置；
S1003、利用硬件加速计算调度配置，并生成C++推理代码和bmodel文件；
S1004、生成计算调度子图，并将所述调度配置、所述bmodel文件、所述C++推理代码和所述调度子图共同并存储为边缘计算模型。
在本发明实施例中，在进行边缘计算前，需要将边缘计算模型进行封装，因此，通过此方式将各功能类型生成总的边缘计算模型，并进行封装。
图11是本发明一个实施例的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法中的对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码的示意图。如图11所示，对于选定的计算调度配置，分解为CPU计算调度配置和硬件加速芯片的计算调度配置，对于堪智K210、绿浪GAP8、算丰BM1880、晶芯N22、蜂鸟E203，以上五种均为RISC-V芯片的厂家及芯片型号，具体的分别产出后端计算支持代码或模型，并且与子图描述一并打包到边缘计算模型。
根据本发明实施例第三方面，提供一种电子设备。图12是本发明一个实施例中一种电子设备的结构图。图12所示的电子设备为通用多制式边缘网关通信装置，其包括通用的计算机硬件结构，其至少包括处理器1201和存储器1202。处理器1201和存储器1202通过总线1203连接。存储器1202适于存储处理器1201可执行的指令或程序。处理器1201可以是独立的微处理器，也可以是一个或者多个微处理器集合。由此，处理器1201通过执行存储器 1202所存储的指令，从而执行如上所述的本发明实施例的方法流程实现对于数据的处理和对于其它装置的控制。总线1203将上述多个组件连接在一起，同时将上述组件连接到显示控制器1204和显示装置以及输入/输出(I/O)装置1205。输入/输出(I/O)装置1205可以是鼠标、键盘、调制解调器、网络接口、触控输入装置、体感输入装置、打印机以及本领域公知的其他装置。典型地，输入/输出装置1205通过输入/输出(I/O)控制器1206与系统相连。
本发明的实施例提供的技术方案基于RISCV指令集以及集成CNN硬件加速器的边缘计算控制器，CNN硬件加速器用于实现对ROM、RAM等指定存储器中的数据进行卷积处理，解决了边缘端CNN卷积运算速度慢、负载重、开销大等问题，具体的有益效果可以包括：
1)本方案采用了开源指令集架构，利用人工智能处理数据，可以实现大量的节点数据的在线接入、分析、处理和运算，效率高且兼容性强；
2)本方案集成了全部的物联网节点的网关功能，通过多制式、多协议、多功能的设备，提升数据感知和数据处理的速度；
3)本方案通过采用大量的模块化设计，使得外部的硬件组模能够快速接入到系统中，实现便捷的物联网应用；
4)本方案通过负荷车路协调等多种应用场景，实现了不同场景下的数据感知运算、数据融合和节点广播功能，满足了高负载系统下的运算性能需求。
本领域内的技术人员应明白，本发明的实施例可提供为方法、系统、或计算机程序产品。因此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器和光学存储器等)上实施的计算机程序产品的形式。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程 图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
显然，本领域的技术人员可以对本发明进行各种改动和变型而不脱离本发明的精神和范围。这样，倘若本发明的这些修改和变型属于本发明权利要求及其等同技术的范围之内，则本发明也意图包含这些改动和变型在内。

Claims (10)

  1. 一种基于RISC-V指令集的AIOT多制式边缘网关通信系统，其特征在于，包括：云端协同平台、智能算法层、安全加密层、管理层、网关适配层、协议栈层、设备驱动层、嵌入式操作层、指令集处理器；其中，所述云端协同平台与所述智能算法层进行数据交互，所述安全加密层、所述管理层、所述网关适配层和所述协议栈层分别对被交互的数据进行安全加密、数据管理、协议转换、网络传输，所述嵌入式操作层其上设置有所述指令集处理器，所述嵌入式操作层其上配置有所述设备驱动层，所述设备驱动层进行与外部传输设备的功能交互。
  2. 如权利要求1所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统，其特征在于，所述安全加密层具体包括DTLA子模块、AES子模块、CA子模块、SSL子模块、SHA265子模块、HASH子模块、MD5子模块；其中，所述DTLA子模块用于数字传输授权管理，所述AES子模块用于高级别加密，所述CA子模块用于管理签发安全凭证和加密信息安全密钥，所述SSL子模块用于为提供安全传输协议，所述SHA265子模块用于采用密码散列进行安全加密；所述HASH子模块用于通过散列算法将任意长度的输入映射为固定长度的散列值；所述MD5子模块用于产生一个16字节的散列值。
  3. 如权利要求1所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统，其特征在于，所述协议栈层包括MQTT传输子模块、CoAP传输协议子模块、TCP/IP传输协议子模块、UDP传输协议子模块、ARP传输协议子模块、DHCP传输协议子模块、IPv4传输协议子模块、IPv6传输协议子模块、NVME传输协议子模块、BLE传输协议子模块；其中，所述MQTT传输子模块用于将传输数据以消息队列遥测传输标准下发；所述CoAP传输协议子模块用于将数据以CoAP传输协议进行数据转发；所述TCP/IP传输协 议子模块用于将网络间的传输数据以TCP/IP传输协议转发，所述IPv4传输协议子模块和所述IPv6传输协议子模块分别用于不同级别的IP地址形式生成；所述NVME传输协议子模块用于通过非易失去性存储器在移动设备和数据中心中对信息进行开放收集；所述UDP传输协议子模块用于将数据以UDP传输协议转发；所述ARP传输协议子模块用于将数据以ARP传输协议转发，所述DHCP传输协议子模块用于将数据以DHCP传输协议在局域网内转发。
  4. 如权利要求1所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统，其特征在于，所述管理层包括Radio管理子模块、Powr管理子模块、Memory管理子模块、Time管理子模块、Client管理子模块、KeyManager管理子模块、算法管理子模块；其中，所述Radio管理子模块用于管理嵌入式系统中的音频模块；所述Powr管理子模块用于管理嵌入式系统中的电源管理模块；所述Memory管理子模块用于管理嵌入式系统中的内存的管理模块；所述Time管理子模块用于管理嵌入式系统中的定时器、时间触发型任务功能模块；所述Client管理子模块用于嵌入式系统中的用户群管理模块；所述KeyManager管理子模块用于管理嵌入式系统中的用户密钥管理模块；所述算法管理子模块用于管理采用的AI智能算法，其中，所述AI智能算法包括YOLOV3和MobileNet。
  5. 如权利要求1所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统，其特征在于，所述网关适配层包括多种网络接口的接入，具体的包括WAN、LAN、RS485、Wi-Fi、Zigbee、4G/5G、LTE-V、BLE4.1、NB-IoT、Sigfox、LoRa、6LoPWAN。
  6. 如权利要求1所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信系统，其特征在于，所述云端协同平台包括边缘端节点和云中心，所述边缘端节点包括管理组模、数据运算组模、应用服务组模、书存储组模，所述云中心包括边缘智能管理套件和人工智能应用服务，其中，所述人工智能应用服务与所述智能算法层配合进行图像处理、函数运算、网络训练、流式 计算、大数据挖掘；所述边缘智能管理套件包括程序、资源和应用程序管理。
  7. 一种基于RISC-V指令集的AIOT多制式边缘网关通信方法，其特征在于，该方法具体包括：
    通过嵌入式操作层及其上配置的指令集处理器进行卷积运算，生成处理数据，其中处理数据包括物联网节点数据、图像信息、声音信息；
    对于通过所述设备驱动层获取的全部数据进行数据安全性校验，并在校验和预处理后将数据分为训练数据和验证数据，压缩训练数据，进行异构运算，生获数据分析模型并存储；
    根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩；
    对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码。
  8. 如权利要求7所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法，其特征在于，所述根据边缘硬件计算设备，进行参数量化评估获得评估正确率，当评估正确率降低时，重新进行优化与压缩，具体包括：
    输入预训练模型；
    根据硬件参数和所述预训练模型确定剪枝和量化参数；
    根据所述剪枝和量化参数调整所述预训练模型；
    对所述预训练模型进行量化生成模型准确度；
    当所述模型准确度低于预设正确度门槛值时，重新确定剪枝和量化参数；
    将微调后的所述预训练模型保存为压缩模型。
  9. 如权利要求7所述的一种基于RISC-V指令集的AIOT多制式边缘网关通信方法，其特征在于，所述对已经完成的计算进行速度调整，使不同硬件加速芯片完成不同边缘运算，并输出支持代码，具体包括：
    边缘计算模型封装；
    获取当前的调度配置信息，并通过CPU进行计算调度配置；
    利用硬件加速计算调度配置，并生成C++推理代码和bmodel文件；
    生成计算调度子图，并将所述调度配置、所述bmodel文件、所述C++推理代码和所述调度子图共同并存储为边缘计算模型。
  10. 一种电子设备，包括存储器和处理器，其特征在于，所述存储器用于存储一条或多条计算机程序指令，其中，所述一条或多条计算机程序指令被所述处理器执行以实现如权利要求7-9任一项所述的步骤。
PCT/CN2021/092328 2021-04-29 2021-05-08 一种基于risc-v指令集的aiot多制式边缘网关通信系统及设备 WO2022227106A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110476307.9 2021-04-29
CN202110476307.9A CN113329358B (zh) 2021-04-29 2021-04-29 一种基于risc-v指令集的aiot多制式边缘网关通信系统及设备

Publications (1)

Publication Number Publication Date
WO2022227106A1 true WO2022227106A1 (zh) 2022-11-03

Family

ID=77414096

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/092328 WO2022227106A1 (zh) 2021-04-29 2021-05-08 一种基于risc-v指令集的aiot多制式边缘网关通信系统及设备

Country Status (2)

Country Link
CN (1) CN113329358B (zh)
WO (1) WO2022227106A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114338281B (zh) * 2021-11-11 2023-04-07 广州软件应用技术研究院 一种基于边缘计算网关的模型分发应用设计方法及系统
CN115270668B (zh) * 2022-07-18 2023-07-07 北京师范大学 一种信息科技教育开源硬件专用芯片

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108092884A (zh) * 2017-11-23 2018-05-29 南京邮电大学 一种无线接入网关系统及应用方法
CN109889592A (zh) * 2019-02-25 2019-06-14 北京邮电大学 一种基于边缘计算的智能制造方法及装置
US20200021586A1 (en) * 2017-02-06 2020-01-16 Pcms Holdings, Inc. Securing communication of devices in the internet of things
CN111163002A (zh) * 2019-12-31 2020-05-15 广州智光电气股份有限公司 一种基于容器的边缘网关系统和能源数据处理方法
CN112260944A (zh) * 2020-10-26 2021-01-22 浙江大学 基于ARM-Linux系统的嵌入式智能边缘计算网关

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102271124B (zh) * 2010-06-01 2015-05-13 富士通株式会社 数据处理设备和数据处理方法
JP6548172B2 (ja) * 2017-06-12 2019-07-24 パナソニックＩｐマネジメント株式会社 端末認証システム、サーバ装置、及び端末認証方法
US11315024B2 (en) * 2018-06-25 2022-04-26 Kyndryl, Inc. Cognitive computing systems and services utilizing internet of things environment
CN109635936A (zh) * 2018-12-29 2019-04-16 杭州国芯科技股份有限公司 一种基于重训练的神经网络剪枝量化方法
CN109815733A (zh) * 2019-01-09 2019-05-28 网宿科技股份有限公司 一种基于边缘计算的智能管理方法和系统
CN110007961B (zh) * 2019-02-01 2023-07-18 中山大学 一种基于risc-v的边缘计算硬件架构
CN109862011B (zh) * 2019-02-01 2021-03-30 华南理工大学 一种基于雾计算的物联网环境实时监测系统
US10742421B1 (en) * 2019-03-08 2020-08-11 Ares Technologies, Inc. Methods and systems for anonymous hardware attestation
CN110913391A (zh) * 2019-10-25 2020-03-24 国网陕西省电力公司信息通信公司 一种多数据源通信加解密系统及方法
CN111324383B (zh) * 2020-02-28 2022-05-10 西安微电子技术研究所 一种基于risc-v指令扩展的安全协处理器结构
CN112199322A (zh) * 2020-09-30 2021-01-08 中国电力科学研究院有限公司 一种电力智能控制终端及其基于risc-v的soc电力芯片架构

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200021586A1 (en) * 2017-02-06 2020-01-16 Pcms Holdings, Inc. Securing communication of devices in the internet of things
CN108092884A (zh) * 2017-11-23 2018-05-29 南京邮电大学 一种无线接入网关系统及应用方法
CN109889592A (zh) * 2019-02-25 2019-06-14 北京邮电大学 一种基于边缘计算的智能制造方法及装置
CN111163002A (zh) * 2019-12-31 2020-05-15 广州智光电气股份有限公司 一种基于容器的边缘网关系统和能源数据处理方法
CN112260944A (zh) * 2020-10-26 2021-01-22 浙江大学 基于ARM-Linux系统的嵌入式智能边缘计算网关

Also Published As

Publication number Publication date
CN113329358B (zh) 2022-10-18
CN113329358A (zh) 2021-08-31

Similar Documents

Publication Publication Date Title
WO2022227106A1 (zh) 一种基于risc-v指令集的aiot多制式边缘网关通信系统及设备
Lin et al. Task offloading for wireless VR-enabled medical treatment with blockchain security using collective reinforcement learning
US20220086052A1 (en) Intelligent management method and system based on edge computing
CN109426574A (zh) 分布式计算系统，分布式计算系统中数据传输方法和装置
CN111556089A (zh) 基于使能区块链移动边缘计算系统的资源联合优化方法
US20220035322A1 (en) Intelligent operation control apparatus and system
CN112217805B (zh) 一种配电物联网多模协议适配方法
CN110515889A (zh) 嵌入式fpga集群智能计算平台硬件框架
CN102932327A (zh) 零终端设备与桌面虚拟机之间的通信方法及系统
US20230137879A1 (en) In-flight incremental processing
CN110928694A (zh) 一种计算机系统
Jiang et al. Design and implementation of smart-home monitoring system with the Internet of Things technology
CN105049485A (zh) 一种面向实时视频处理的负载感知云计算系统
CN104243198B (zh) 一种基于网络配置协议的网络管理方法和系统
CN204231387U (zh) 一种基于fpga实现高速通信的智能微电网控制系统
Min et al. Design and implementation of the multi-channel RS485 IOT gateway
CN117196014B (zh) 基于联邦学习的模型训练方法、装置、计算机设备及介质
CN117312465A (zh) 基于集成学习的大数据分类和聚类方法
Wang et al. An Edge Computing Based Gateway for WIA-PA Networks
CN109753359A (zh) 一种用于构建资源池的fpga板卡、服务器和系统
CN105207946A (zh) 一种网络数据包负载均衡和预解析方法
CN115665218A (zh) 一种物联网设备远程控制方法、系统及相关设备
Wang et al. Research on Design of Internet of Things Gateway for Adaptive Data Acquisition of Power Transmission, Transformation and Distribution Based on Edge Intelligence
CN102694774A (zh) IPv6微型协议栈设计方法
CN114640697B (zh) 配电物联网边缘控制器的数据处理方法、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21938579

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21938579

Country of ref document: EP

Kind code of ref document: A1