An AI algorithm deployment system based on edge computing devices

Abstract

This invention discloses an AI algorithm deployment system based on edge computing devices, mainly composed of a site adaptation module, an operating system configuration module, an algorithm execution configuration module, and an operation and maintenance strategy module. The site adaptation module proposes an estimation method for the capacity of energy storage sites supported by edge computing devices, taking into account different edge computing device models selected in practical applications. The operating system configuration module, based on the site adaptation results, selects the edge computing device and provides the lowest-level operating system support for the subsequent installation of the algorithm execution configuration module and the operation and maintenance strategy module. The algorithm execution configuration module supports the deployment of the software environment required for AI algorithm execution. The operation and maintenance strategy module is used for code debugging, software system operation and maintenance, and remote code updates. This invention solves the problem of transmitting battery data to the cloud, reduces data transmission latency, and reduces storage overhead. The edge computing devices used have low power consumption and considerable computing power, freeing up cloud computing resources.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery safety early warning, and in particular to an AI algorithm deployment system based on edge computing devices, which applies to hardware, software deployment, operation and maintenance. Background Technology

[0002] With the introduction of dual-carbon goals and the development of clean energy, the energy storage industry is booming. As the application of residential and commercial energy storage expands rapidly, the demand for operational safety and data governance of energy storage stations is becoming increasingly urgent.

[0003] For existing energy storage sites, the typical approach is to collect multi-dimensional data from numerous internal battery cells, extract the data information needed for AI algorithms, and then process the data uniformly on a cloud server. Clearly, with a large number of battery cells, the AI ​​algorithm only needs key information from a massive amount of cell data, without needing to store all the data. Given widespread data privacy requirements, users generally prefer private deployments where data does not pass through public clouds. Therefore, the limitations of the cloud-based algorithm processing model become more pronounced.

[0004] Furthermore, transmitting massive amounts of battery data to the cloud presents challenges such as data transmission latency, direct economic costs for cloud servers storing large amounts of battery cell data, and technical difficulties. Cloud computing power and data storage capabilities are located far from energy storage sites, resulting in poor economic efficiency and higher requirements for technical stability. Due to the large volume of data, the cloud's response speed is slow, making it difficult to address urgent energy storage safety early warning needs and the governance objectives of massive battery cell data. Summary of the Invention

[0005] To address the aforementioned technical issues, this invention provides an AI algorithm deployment system based on edge computing devices, which can reduce data transmission latency, decrease storage overhead, support private and public cloud deployments, improve data response speed, resolve obstacles for remote debugging by maintenance personnel, and solve application challenges such as the management of massive amounts of battery cell data.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] An AI algorithm deployment system based on edge computing devices mainly consists of a site adaptation module, an operating system configuration module, an algorithm operation configuration module, and an operation and maintenance strategy module.

[0008] The site adaptation module proposes an estimation method for the capacity of energy storage sites supported by edge computing devices based on the different parameters of CPU core count, clock frequency and memory type of edge computing devices. This method is used to determine the adaptation relationship between edge computing devices and energy storage sites, and provides a basis for hardware selection for subsequent software components.

[0009] The operating system configuration module can select an edge computing device based on the site adaptation results and provide the lowest-level operating system support for the installation of the subsequent algorithm running configuration module and operation and maintenance strategy module.

[0010] The algorithm operation configuration module supports the deployment of the software environment required for AI algorithm operation;

[0011] The operation and maintenance strategy module is mainly used for code debugging and software system operation and maintenance, as well as remote code updates, to support engineers in remote technical maintenance of the operating system configuration module and algorithm operation configuration module.

[0012] Furthermore, the site adaptation module provides a basis for selecting edge computing hardware. For different sites, after the user's energy storage site capacity is determined, an estimation formula for matching the energy storage site capacity and the edge computing hardware performance is provided when selecting edge computing hardware.

[0013] Furthermore, the operating system configuration module is a Linux system adapted to 64-bit Arm architecture chips for edge computing hardware, including operating system creation and burning to the hardware.

[0014] Furthermore, the algorithm runtime configuration module mainly refers to the compiler management tools required for the AI ​​algorithm program to run, as well as the libraries that the AI ​​algorithm depends on.

[0015] Furthermore, the operation and maintenance strategy module mainly includes program daemon operation, scheduled tasks, intranet penetration, and Docker container deployment.

[0016] The program's guardian and scheduled tasks ensure that the AI ​​algorithm runs according to a certain execution strategy, from the perspective of stable operation.

[0017] The term "internal network penetration" refers to the ability of hosts located in different network segments to remotely connect to and debug edge computing devices, facilitating code debugging and other maintenance work.

[0018] The Docker container deployment refers to the use of Docker containers to quickly restore the Linux operating system and various software supporting AI algorithms when there is a device software failure or when users expect to quickly deploy the software system.

[0019] The beneficial effects of this invention are:

[0020] The edge computing device used in this invention has low power consumption and considerable computing power. It can transmit specific key feature data to the cloud or transmit calculation results to the cloud, freeing up cloud computing power and significantly reducing cloud storage overhead; it enables the front-end application of AI algorithms with millisecond-level response.

[0021] Furthermore, the single-board computer boasts abundant physical interfaces and strong environmental adaptability. Maintenance engineers can leverage this hardware to improve the cost-effectiveness of remote maintenance and, through the overall hardware and software system, provide a universal model for the deployment of AI algorithms. Attached Figure Description

[0022] Figure 1 This is a block diagram of the system composition of the present invention.

[0023] Figure 2 This is a hardware design model diagram for edge computing involved in the present invention.

[0024] Figure 3 This invention relates to a schematic diagram of the installation port for a mainstream M.2 interface solid-state drive.

[0025] Figure 4 This is a simplified diagram of the overall system modules of the present invention. Detailed Implementation

[0026] This invention pertains to energy storage safety management devices. The main concept is to deploy AI algorithms on low-power edge computing devices, forming a comprehensive deployment solution for the practical commercial application of AI algorithms. For example... Figure 1 As shown, it consists of four modules: site adaptation, operating system configuration, algorithm operation configuration, and operation and maintenance strategy.

[0027] For site adaptation modules, the number of cores and clock speed are selected based on requirements, balancing economic cost and industrial adaptability. Currently, there are many models of edge computing devices, with varying chip clock speeds and memory configurations. Based on experience using other edge computing devices, an empirical formula is summarized: Adapted site capacity = (1MWh × Number of CPU cores × CPU clock speed × DDR type coefficient) / 4.5, where the DDR type coefficient is 0.5 for DDR3 memory and 1 for DDR4 memory; 4.5 is an empirical coefficient for capacity adaptation, selected based on actual production experience. Taking the Rockchip 3588 chip as an example, its CPU core count is 8, clock speed is 2GHz, memory is DDR4, and type coefficient is 1. Its adapted site capacity is: = 3.5WMh (1MWh × 8 × 2 × 1) / 4.5.

[0028] The operating system configuration module includes system creation and flashing. Since the CPU chips in edge computing hardware are primarily Armv7, Armv8, and Armv9 architecture chips, these chips are small, low-power, low-cost, and high-performance. Currently, Armv7 ecosystem support is limited, and Armv9 is mostly used in consumer electronics such as mobile phone chips; therefore, the operating system module only refers to the operating system for Armv8 chips. Figure 2 As shown, this is an edge computing hardware device with various additional peripherals, integrating interfaces such as serial port and USB.

[0029] 1) For system creation, a system image is obtained through cross-compilation. Cross-compilation refers to performing the operating system compilation work for edge devices on a high-performance server. A cross-compilation environment, either natively or in a virtual machine, is established on a Linux server, and the system is compiled based on the Linux system source code provided by the edge computing hardware vendor. Taking an Ubuntu 18.04 server as an example, the cross-compilation environment requires a series of compilation software tools such as g++ and gcc, which are installed using the apt software management tool. The system image is obtained by modifying the configuration files in the edge computing source code.

[0030] 2) For system flashing: A Windows computer acts as the flashing host, executing the system flashing software to flash the Linux system image onto the development board hardware. For example... Figure 3 The rectangular frame represents a common M.2 interface for solid-state drives (SSDs). Different models or manufacturers of edge hardware use different system flashing software. Taking Rockchip as an example, it provides dedicated drivers and the flashing software RKDevTool.exe. Following the operating procedures and connecting the PC to the edge computing device via USB, the system flashing is completed, typically taking less than 10 minutes. Furthermore, for hardware storage expansion, since the hardware device is essentially an independent computer with its own operating system, installing an SSD can provide additional data storage space or serve as a system drive. Figure 2 In practical applications, storage can be flexibly configured according to needs, such as selecting a 128G or 256G hard drive, and in special cases, a 512G solid-state drive, thereby freeing up the cloud's data storage needs.

[0031] like Figure 4 As shown in the diagram, the overall system of this invention is divided into modules, mainly consisting of program daemonization, scheduled tasks, remote operation and maintenance, and Docker container deployment. Regarding algorithm execution configuration, since the mainstream AI algorithm programming language is Python, this module primarily supports the execution of Python programs. Because AI algorithms are manifested as code, it provides the necessary conditions for the code to run.

[0032] Main steps: 1) Establish isolated program execution environments using software such as Mambaforge, Anaconda, and Miniforge, ensuring these environments do not interfere with each other. 2) Install external libraries required by the AI ​​algorithm, primarily machine learning libraries such as Pandas, NumPy, Torch, and TensorFlow. The installation of these software and libraries can be done via online installation after connecting to the public internet, rapid offline deployment, or source code compilation.

[0033] For operation and maintenance management, 1) Supervisor program protection. This refers to situations where the algorithm cannot function properly due to data, program, or equipment issues in an unattended state. Therefore, a reliable Supervisor is used to protect the algorithm program in the background. The Supervisor process consists of four core parts: supervisor.d, a child process for management configuration; supervisorctl, a client command line used to start and stop child processes; WebServer, for viewing and controlling the process status through a web interface; and XML-RPCInterface, used for querying and controlling the management program, etc. Supervisor settings include priority and logging.

[0034] 2) Scheduled Tasks. This part is migrated from the Crontab scheduled task mode of the cloud server. The AI ​​algorithm runs according to the running frequency and the specific task settings are configured.

[0035] 3) Remote Debugging and Maintenance. This invention employs a remote SSH intranet penetration mode. It is not recommended to use traditional remote desktop software such as Rustdesk, Todesk, VNC, or Sunflower, as these have high network bandwidth requirements, high latency, and excessive CPU and RAM consumption. In specific applications, intranet penetration tools such as PeanutShell, Cpolar, and RFP are used to obtain the public IP address, port, and user information. VS Code or PyCharm Professional Edition is used, and the penetrated IP address, port, and user information are entered for debugging AI algorithm code. For remote maintenance, a terminal management tool such as Mobaxterm is used, and the specified intranet penetration information is entered to establish a remote SSH connection, enabling the issuance of maintenance commands with lower bandwidth.

[0036] 4) Docker container deployment. Since Docker supports running on Armv8 architecture, the specific usage can still follow the experience and operation process of Docker on the server side.

[0037] In summary, this invention provides a relatively objective analysis of the current industrial situation and utilizes a comprehensive solution combining edge computing hardware and software to overcome the limitations of cloud-based data transmission and AI algorithm deployment. Edge computing hardware boasts low power consumption, aligning with low-carbon requirements; its cost is low and controllable; furthermore, given the inherent electrical hazards of energy storage sites, remote operation and maintenance reduce costs and improve safety. More significantly, this holistic hardware and software system solution provides a practical model for the implementation of AI algorithms, accelerating their application and effectively safeguarding the safety of energy storage sites.

Claims

1. An AI algorithm deployment system based on edge computing devices, mainly composed of a site adaptation module, an operating system configuration module, an algorithm operation configuration module, and an operation and maintenance strategy module, characterized in that: The site adaptation module proposes an estimation method for the capacity of energy storage sites supported by edge computing devices based on the different parameters of CPU core count, clock speed, and memory type of edge computing devices. This method is used to determine the compatibility between edge computing devices and energy storage sites, providing a basis for hardware selection for subsequent software components. The operating system configuration module can select edge computing devices based on the site adaptation results and provide the lowest-level operating system support for the installation of subsequent algorithm operation configuration modules and operation and maintenance strategy modules. The algorithm execution configuration module supports the deployment of the software environment required for AI algorithm execution; the operation and maintenance strategy module is mainly used for code debugging and software system operation and maintenance, and remote code updates, to support engineers in remote technical maintenance of the operating system configuration module and the algorithm execution configuration module.

2. The AI ​​algorithm deployment system based on edge computing devices according to claim 1, characterized in that: The site adaptation module provides a basis for selecting edge computing hardware. For different sites, after the user's energy storage site capacity is determined, it provides an estimation formula for matching the energy storage site capacity and the edge computing hardware performance when selecting edge computing hardware.

3. The AI ​​algorithm deployment system based on edge computing devices according to claim 1, characterized in that: The operating system configuration module is a Linux system adapted to 64-bit Arm architecture chips for edge computing hardware, including operating system creation and burning to the hardware.

4. The AI ​​algorithm deployment system based on edge computing devices according to claim 1, characterized in that: The algorithm runtime configuration module mainly refers to the compiler management tools needed for running AI algorithm programs and the libraries that the AI ​​algorithm depends on.

5. The AI ​​algorithm deployment system based on edge computing devices according to claim 1, characterized in that: The operation and maintenance strategy module mainly includes program guardian operation, scheduled tasks, intranet penetration, and Docker container deployment.

6. The AI ​​algorithm deployment system based on edge computing devices according to claim 5, characterized in that: The program's guardian and scheduled tasks ensure that the AI ​​algorithm runs according to a certain execution strategy, from the perspective of stable operation.

7. The AI ​​algorithm deployment system based on edge computing devices according to claim 5, characterized in that: The term "internal network penetration" refers to the ability of hosts located in different network segments to remotely connect to and debug edge computing devices, facilitating code debugging and other maintenance work.

8. The AI ​​algorithm deployment system based on edge computing devices according to claim 5, characterized in that: The Docker container deployment refers to the use of Docker containers to quickly restore the Linux operating system and various software supporting AI algorithms when there is a device software failure or when users expect to quickly deploy the software system.

Images