Wind power generation type large model space intelligent agent intelligent integrated operation and maintenance system
By designing a wind power generation large-model space intelligent integrated operation and maintenance system, using the domestic deep learning framework and the system-wide reasoning architecture, the problem of dependence on foreign high-end GPUs in the existing technology is solved, efficient reasoning and cost reduction are achieved, and the system's autonomous controllability and security are improved.
Patent Information
- Application Number
- CN202510205052.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-24
- Publication Date
- 2025-05-13
AI Technical Summary
Existing large-model inference solutions rely on hardware such as high-end GPUs abroad, resulting in high usage costs, poor autonomy and controllability, and information security and data security problems.
A wind power generation large-scale model space intelligent integrated operation and maintenance system was designed, adopting a domestic deep learning framework to support the inference tasks of a variety of large-scale models, combining the system-wide inference architecture and storage-based conversion technology to reduce the demand for computing power, and through the model management module and task scheduling module, we ensure the efficient operation of the large model and the reasonable allocation of resources.
It has achieved efficient reasoning, met the high requirements of real-time and accuracy in the new energy industry, greatly reduced the cost of implementing large models, improved the economic benefits of enterprises, and improved the system's independence and controllability through domestic hardware and software, and ensured information security and data security.
Smart Images
Figure CN119990677A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of the new energy industry, and specifically is a wind power generation large-model space intelligent body intelligent integrated operation and maintenance system. Background Art
[0002] Generative large models refer to artificial intelligence models with a large number of parameters. They can generate new content, such as text, pictures, music, etc. Their technical principles are mainly based on the neural network architecture in deep learning, such as Transformer. These models learn the distribution characteristics of data through a large amount of training data, so that they have the ability to generate new content, that is, by calculating the vector relationship between different data to represent the probabilistic relationship between data, and then generate content that conforms to the current probability based on these relationships.
[0003] With the continuous development of big model technology, its application in the new energy industry is becoming more and more extensive. For example, in the operation and maintenance management of wind farms, big model technology can perform energy forecasting, fault diagnosis, real-time monitoring, intelligent inspection, etc. However, the existing big model reasoning solutions mostly rely on foreign high-end GPUs and other hardware. Although they can perform reasoning, foreign high-end GPUs and other hardware have high usage costs, poor autonomous controllability, and information security and data security issues. Therefore, improvements are needed. Summary of the invention
[0004] The purpose of the present invention is to provide a wind power generation large-scale model space intelligent body intelligent integrated operation and maintenance system to solve the problems raised in the above-mentioned background technology.
[0005] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: a wind power generation type large model space intelligent body intelligent integrated operation and maintenance system, including: a data acquisition module, a data processing module, a data storage module, a reasoning framework, a task scheduling module and a model management module, wherein:
[0006] The data acquisition module is used to collect various data of the new energy power station;
[0007] The data processing module is divided into data preprocessing and postprocessing, which processes the data collected by the data collection module and post-processes the reasoning results of the reasoning framework;
[0008] The data storage module is used to store preprocessed data and reasoning results of the reasoning framework;
[0009] The inference framework is developed based on a domestic deep learning framework, supports inference tasks of multiple large models, and is used to perform inference calculations on processed data;
[0010] The task scheduling module: reasonably allocates computing resources according to the priority and resource requirements of tasks to improve the concurrent processing capability of the system;
[0011] The model management module is responsible for loading, updating and optimizing the large model to ensure the efficient operation of the large model.
[0012] As a preferred technical solution of the present invention, the model management module uses a deep learning framework to train large models based on the collected massive data for different application scenarios and hardware environments, improves training efficiency through distributed training technology, and compresses and optimizes the model.
[0013] As a preferred technical solution of the present invention, the inference framework adopts a full-system inference architecture, coordinating multiple devices such as storage, CPU, GPU, and NPU, and releasing storage power as a supplement to computing power through the "storage-for-computing" technology, thereby reducing the demand for computing power. At the same time, it adopts the idea of "heterogeneous collaboration" to closely link HBM / DRAM / SSD and CPU / GPU / NPU full-system heterogeneous devices.
[0014] As a preferred technical solution of the present invention, the data storage module caches some commonly used data and intermediate calculation results when storing data.
[0015] As a preferred technical solution of the present invention, the all-in-one hardware of the intelligent integrated operation and maintenance system includes a computing unit, a storage unit, a network unit, a power supply and cooling unit, and a management control unit. The open API interface is provided to support local deployment of tens of billions of first-line large models.
[0016] As a preferred technical solution of the present invention, the computing unit includes a CPU, a GPU and a memory, the CPU is connected to other components through a system bus, the GPU is connected to the motherboard through a high-speed PCIe bus, and the memory is directly connected to the CPU through a memory bus.
[0017] As a preferred technical solution of the present invention, the storage unit includes an SSD and a HDD, the SSD is connected to the mainboard via a high-speed NVMe interface or a SATA interface, and the HDD is connected to the mainboard via a SATA interface.
[0018] As a preferred technical solution of the present invention, the network unit includes NIC and Switch, the NIC is connected to the mainboard via a PCIe interface and is responsible for connecting the all-in-one machine to an external network, and the Switch is responsible for forwarding data between different devices.
[0019] As a preferred technical solution of the present invention, the management control unit adopts a BMC, which is connected to various components on the mainboard through a dedicated management bus and is responsible for monitoring the status of the hardware devices.
[0020] The beneficial effects of the present invention are as follows:
[0021] The present invention realizes efficient reasoning through the whole system reasoning architecture and storage-based computing technology, meets the high requirements of the new energy industry for real-time and accuracy, greatly reduces the implementation cost of large models, and improves the economic benefits of enterprises. At the same time, it adopts domestically produced hardware and software to improve the autonomy and controllability of the system, reduce dependence on foreign technology, and ensure the information security and data security of the new energy industry. In addition, it also supports local deployment and customized development, and can be flexibly configured and optimized according to different business scenarios and needs of the new energy industry, thereby improving the applicability and flexibility of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 This is a system framework diagram of the present invention. DETAILED DESCRIPTION
[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0024] like Figure 1 As shown, the embodiment of the present invention provides a wind power generation type large model space intelligent body intelligent integrated operation and maintenance system, including: a data acquisition module, a data processing module, a data storage module, a reasoning framework, a task scheduling module and a model management module, wherein:
[0025] The data acquisition module is used to collect various data of new energy power stations;
[0026] The data processing module is divided into data preprocessing and postprocessing. It processes the data collected by the data acquisition module and post-processes the reasoning results of the reasoning framework.
[0027] The data storage module is used to store the preprocessed data and the inference results of the inference framework;
[0028] The reasoning framework is developed based on the domestic deep learning framework, supports reasoning tasks of multiple large models, and is used to perform reasoning calculations on processed data;
[0029] Task scheduling module: reasonably allocate computing resources according to task priority and resource requirements to improve the concurrent processing capability of the system;
[0030] The model management module is responsible for loading, updating and optimizing large models to ensure efficient operation of large models.
[0031] After the data acquisition module collects the power generation, equipment status, weather and other data of the new energy power station, it is transmitted to the data processing module for preprocessing such as data analysis and format conversion to ensure the accuracy and consistency of the data. The preprocessed data is then transmitted to the data storage module for storage. Subsequently, the inference framework reads the preprocessed data in the data storage module through the CPU and performs inference calculations based on the loaded large model parameters. The calculation results are then transmitted to the data processing module for post-processing, converted into a format understandable to the business, and finally output to the business system or user. In the process of inference calculation, the task scheduling module is used to reasonably allocate computing resources to achieve parallel computing. At the same time, the model management module trains, updates and optimizes the large model to ensure the efficient operation of the large model. In actual operation, by receiving meteorological data and equipment status data in real time, power generation is predicted, and the prediction results are output to the dispatching system of the power plant, which can improve the dispatching efficiency and power generation benefits of the power plant; by real-time monitoring of equipment status data, fault diagnosis is performed, and the diagnosis results are output to the enterprise's operation and maintenance system, equipment failures can be warned in advance, reducing equipment downtime and maintenance costs.
[0032] Among them, the model management module uses a deep learning framework to train large models based on the massive data collected, targeting different application scenarios and hardware environments, improves training efficiency through distributed training technology, and compresses and optimizes the model.
[0033] Through regular training and optimization of large models, the model size can be reduced, the reasoning speed can be improved, and the deployment cost can be reduced. At the same time, the large model has a certain self-learning ability, and can continuously optimize and update its own parameters and knowledge base according to new data and feedback, so as to better adapt to the dynamic changes of new energy power stations and improve the accuracy and reliability of reasoning.
[0034] Among them, the inference framework adopts a full-system inference architecture, coordinating multiple devices such as storage, CPU, GPU, and NPU. Through the "storage-for-computing" technology, it releases storage power as a supplement to computing power, reducing the demand for computing power. At the same time, it adopts the idea of "heterogeneous collaboration" to closely link HBM / DRAM / SSD and CPU / GPU / NPU full-system heterogeneous devices.
[0035] The "heterogeneous collaboration" approach can break through the limitations of video memory capacity and fully unleash the storage and computing power of the entire system, increasing the inference throughput by more than 10 times and significantly reducing the implementation cost of large models. By utilizing the "storage-for-computing" technology, storage resources are used as a supplement to computing power. In the RAG scenario, response latency is reduced by 20 times and performance is improved by 10 times. This fusion of storage and computing power can extract reusable content from historical related information even when facing new problems, and perform online fusion calculations with on-site information, reducing the amount of calculations.
[0036] Among them, the data storage module caches some commonly used data and intermediate calculation results when storing data.
[0037] When a similar reasoning request is encountered again, data can be directly obtained from the cache to avoid repeated calculations. At the same time, by integrating reasoning technology, reusable content can be extracted from historical related information to further reduce the amount of calculations.
[0038] Among them, the all-in-one hardware of the intelligent integrated operation and maintenance system includes computing units, storage units, network units, power and cooling units, and management control units, providing open API interfaces to support local deployment of tens of billions of first-line large models.
[0039] The open API interface facilitates flexible calls by third parties, and localized deployment can meet the high requirements of the new energy industry for data security and avoid risks such as data leakage. Users can customize the enterprise intelligent assistant according to their needs. At the same time, the all-in-one machine integrates domestic hardware and software, including domestic GPU, CPU, storage devices and corresponding reasoning frameworks and algorithms. Through deep collaborative optimization of software and hardware, the overall performance and stability of the system are improved, and power consumption and costs are reduced.
[0040] Among them, the computing unit includes CPU, GPU and memory. The CPU is connected to other components through the system bus, the GPU is connected to the motherboard through the high-speed PCIe bus, and the memory is directly connected to the CPU through the memory bus.
[0041] CPU, or central processing unit, is the control and management center of the entire system. The system bus includes data bus, address bus and control bus. CPU communicates with memory, storage devices and various I / O devices through these buses; GPU, or graphics processing unit, PCIe bus provides a high-bandwidth data transmission channel to ensure that GPU can quickly interact with other hardware. At the same time, in some high-end systems, GPUs may be directly connected through high-speed interconnection technologies such as NVLink to achieve more efficient data exchange when multiple GPUs are computing in parallel; Memory, or memory, is the high-speed cache area when the CPU processes data.
[0042] Among them, the storage unit includes SSD and HDD. The SSD is connected to the motherboard through a high-speed NVMe interface or SATA interface, and is connected to the motherboard through a SATA interface.
[0043] SSDs with NVMe interfaces have higher bandwidth and lower latency, and are suitable for scenarios that require fast reading and writing of large amounts of data. Data stored in SSDs can be transferred to memory via the bus on the motherboard and then processed by the CPU or GPU. HDDs have larger capacities but relatively slow read and write speeds and are often used for long-term data storage. When data in the HDD needs to be accessed, the data is first transferred to memory and then used by the computing unit.
[0044] Among them, the network unit includes NIC and Switch. NIC is connected to the mainboard through the PCIe interface and is responsible for connecting the all-in-one machine to the external network. Switch is responsible for forwarding data between different devices.
[0045] After the network data enters the all-in-one machine through the NIC, it is transmitted to the memory through the bus on the motherboard, and then processed by the computing unit. The processed results are also returned to the NIC through the same path and sent to the external network. When the all-in-one machine needs to communicate with multiple devices over the network, it is usually connected to a switch. The all-in-one machine and the switch are connected by a network cable or optical fiber. The data is forwarded in the switch according to the MA address to achieve communication with other devices.
[0046] Among them, the management control unit adopts BMC, which is connected to various components on the mainboard through a dedicated management bus and is responsible for monitoring the status of hardware devices.
[0047] BMC can run independently of the CPU and communicate with the external management system through the network interface. Administrators can obtain the hardware status information of the all-in-one machine in real time through remote management tools and perform corresponding management and maintenance operations.
[0048] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0049] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A wind power generation large model space intelligent integrated operation and maintenance system, characterized in that: It includes: data acquisition module, data processing module, data storage module, reasoning framework, task scheduling module and model management module, among which, The data acquisition module is used to collect various data of the new energy power station; The data processing module is divided into data preprocessing and postprocessing, which processes the data collected by the data collection module and post-processes the reasoning results of the reasoning framework; The data storage module is used to store preprocessed data and reasoning results of the reasoning framework; The inference framework is developed based on a domestic deep learning framework, supports inference tasks of multiple large models, and is used to perform inference calculations on processed data; The task scheduling module: reasonably allocates computing resources according to the priority and resource requirements of tasks to improve the concurrent processing capability of the system; The model management module is responsible for loading, updating and optimizing the large model to ensure the efficient operation of the large model.
2. According to claim 1, a wind power generation type large model space intelligent body intelligent integrated operation and maintenance system is characterized by: Based on the massive data collected, the model management module uses a deep learning framework to train large models for different application scenarios and hardware environments, improves training efficiency through distributed training technology, and compresses and optimizes the model.
3. According to claim 1, a wind power generation type large model space intelligent body intelligent integrated operation and maintenance system is characterized by: The inference framework adopts a full-system inference architecture, coordinating multiple devices such as storage, CPU, GPU, and NPU. It releases storage power as a supplement to computing power through the "storage-for-computing" technology, reducing the demand for computing power. At the same time, it adopts the idea of "heterogeneous collaboration" to closely link HBM / DRAM / SSD and CPU / GPU / NPU full-system heterogeneous devices.
4. According to claim 1, a wind power generation type large model space intelligent body intelligent integrated operation and maintenance system is characterized by: The data storage module caches some commonly used data and intermediate calculation results when storing data.
5. According to claim 1, a wind power generation type large model space intelligent body intelligent integrated operation and maintenance system is characterized by: The all-in-one hardware of the intelligent integrated operation and maintenance system includes a computing unit, a storage unit, a network unit, a power supply and cooling unit, and a management control unit. The system provides an open API interface to support local deployment of tens of billions of first-line large models.
6. A wind power generation type large model space intelligent integrated operation and maintenance system according to claim 5, characterized in that: The computing unit includes a CPU, a GPU and a memory. The CPU is connected to other components via a system bus, the GPU is connected to a mainboard via a high-speed PCIe bus, and the memory is directly connected to the CPU via a memory bus.
7. According to claim 5, a wind power generation type large model space intelligent body intelligent integrated operation and maintenance system is characterized by: The storage unit includes an SSD and an HDD, the SSD is connected to the mainboard via a high-speed NVMe interface or a SATA interface, and the HDD is connected to the mainboard via a SATA interface.
8. The wind power generation type large model space intelligent agent intelligent integrated operation and maintenance system according to claim 5 is characterized by: The network unit includes a NIC and a Switch. The NIC is connected to the mainboard via a PCIe interface and is responsible for connecting the integrated machine to an external network. The Switch is responsible for forwarding data between different devices.
9. The wind power generation type large model space intelligent agent intelligent integrated operation and maintenance system according to claim 5 is characterized by: The management control unit adopts BMC, which is connected to various components on the mainboard through a dedicated management bus and is responsible for monitoring the status of hardware devices.