Electric power cognition large model research and development tool chain method and related equipment

Abstract

The invention provides an electric power cognition large model research and development tool chain method and related equipment. The method comprises the following steps: acquiring electric power data of a plurality of data sources by utilizing a tool chain; performing correlation processing on the power data by using the tool chain to obtain target power data; and processing the target power data by using the optimized prediction model to obtain prediction data corresponding to the target power data, the prediction model being optimized by using a predetermined algorithm in the tool chain, thereby solving the technical problem in the prior art that the tool chain cannot comprehensively optimize the model, and improving the prediction efficiency. And the generalization ability of the model is improved, so that the result output by the model has accuracy.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a tool chain method and related equipment for developing a large-scale power cognitive model. Background Art

[0002] With the transformation of the global energy structure and the continuous growth of electricity demand, the energy and power industry, as a pillar industry of the national economy, has become particularly important for efficient, safe and intelligent operations. In order to promote the integration of artificial intelligence technology and power grid business and further deepen the support of large-scale power models for the large-scale application of artificial intelligence, it is necessary to create a platform tool chain with unified underlying specifications and efficient industry-data collaboration, covering the sample processing, pre-training, fine-tuning and other links of general models. However, the tool chain in the existing technology has the problem of non-interoperability in the application of the above links, and the tool chain cannot fully support the construction and application of the model, resulting in the tool chain cannot fully optimize the model. For example, in supporting the company in the construction and application of large power models, there are still problems such as difficulty in obtaining multi-scale fine-tuning samples of power and lack of outstanding results in model fine-tuning technology innovation. Summary of the invention

[0003] In view of this, the purpose of this application is to propose a power cognitive large model development tool chain method and related equipment to overcome all or part of the deficiencies in the prior art.

[0004] Based on the above-mentioned purpose, the present application provides a tool chain method for developing a large-scale electric power cognitive model, including: using the tool chain to obtain electric power data from multiple data sources; using the tool chain to perform correlation processing on the electric power data to obtain target electric power data; using an optimized prediction model to process the target electric power data to obtain prediction data corresponding to the target electric power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain.

[0005] Optionally, the use of the tool chain to perform correlation processing on the power data to obtain target power data includes: using preset conditional variables in the conditional generation model in the tool chain to filter the power data to obtain and output the target power data; or, using the thinking chain algorithm in the tool chain to logically adjust the power data to obtain the target power data.

[0006] Optionally, the predetermined algorithm includes a predetermined adjustment strategy algorithm or a low-rank adaptive algorithm; the optimization method of the prediction model includes: optimizing the prediction model using the predetermined adjustment strategy algorithm in the tool chain; or optimizing the prediction model using the low-rank adaptive algorithm in the tool chain.

[0007] Optionally, before processing the target power data using the optimized prediction model, the method includes: packaging the optimized prediction model using the tool chain.

[0008] Optionally, the encapsulating the optimized prediction model by using the tool chain includes: filtering data in the optimized prediction model based on a predetermined distillation strategy; and compressing part of the data in the filtered prediction model based on a predetermined importance adjustment strategy.

[0009] Optionally, the method further includes: monitoring the performance of the optimized prediction model using the tool chain.

[0010] Optionally, before performing correlation processing on the power data using the tool chain to obtain target power data, the method includes: performing preprocessing on the power data using the tool chain.

[0011] Based on the same inventive concept, the present application also provides an electric power cognitive large model research and development tool chain device, including: an acquisition module, configured to use the tool chain to acquire electric power data from multiple data sources; a first processing module, configured to use the tool chain to perform correlation processing on the electric power data to obtain target electric power data; a second processing module, configured to use an optimized prediction model to process the target electric power data to obtain prediction data corresponding to the target electric power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain.

[0012] Based on the same inventive concept, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method as described above when executing the computer program.

[0013] Based on the same inventive concept, the present application also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the method as described above.

[0014] From the above, it can be seen that the power cognitive large model research and development tool chain method and related equipment provided by the present application, the method includes using the tool chain to obtain power data from multiple data sources to achieve the purpose of obtaining comprehensive power data. The power data is processed by the tool chain to obtain target power data, so as to achieve the purpose of screening the power data. The target power data is processed by the optimized prediction model to obtain the prediction data corresponding to the target power data, wherein the prediction model is optimized by the predetermined algorithm in the tool chain to ensure that the model has a high accuracy and generalization ability, so that the prediction data corresponding to the target power data output by the prediction model is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present application or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 A flowchart of a power cognitive large model development tool chain method according to an embodiment of the present application;

[0017] Figure 2 A schematic diagram of a full-stack tool chain for a model according to an embodiment of the present application;

[0018] Figure 3 Schematic diagram of two methods for performing correlation processing on power data according to an embodiment of the present application;

[0019] Figure 4 Schematic diagram of two methods for optimizing a preset test model according to an embodiment of the present application;

[0020] Figure 5 A schematic diagram of a method for packaging and reasoning an optimized prediction model according to an embodiment of the present application;

[0021] Figure 6 A schematic diagram of the structure of a power cognitive large model development tool chain device according to an embodiment of the present application;

[0022] Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0024] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the usual meanings understood by people with ordinary skills in the field to which the present application belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0025] As described in the background technology section, with the transformation of the global energy structure and the continuous growth of electricity demand, the energy and power industry, as a pillar industry of the national economy, has become particularly important for efficient, safe and intelligent operation. In order to promote the integration of artificial intelligence technology and power grid business, further deepen the support of large-scale power models for the large-scale application of artificial intelligence, create a platform tool chain with unified underlying specifications and efficient industry-data collaboration, covering the sample processing, pre-training, fine-tuning, evaluation, deployment and other links of general models. In addition, the power industry has begun to explore and apply artificial intelligence technology, such as power load forecasting, equipment fault diagnosis, and power grid dispatch optimization based on deep learning. However, most of these applications focus on solving a single scenario or task, and lack a unified and efficient model tool chain to support the intelligent operation of the entire power system. The existing technologies mainly include the following solutions: Single model application: By training deep learning models in specific fields, such as convolutional neural networks (CNN) for image recognition and recurrent neural networks (RNN) for time series analysis, specific problems in the power system are solved. Although this method improves processing efficiency to a certain extent, there is a lack of collaboration between models, and it is difficult to form a global optimal solution. Integrated platform: Some power companies have developed application platforms that integrate multiple models, and realize comprehensive applications by integrating modules with different functions. However, these platforms often have problems such as complex interfaces, poor data sharing, and difficulty in updating models, making it difficult to adapt to the rapidly changing needs of power business. Customized solutions: Customized intelligent solutions are developed for specific power companies or projects. Although this method is highly targeted, it is costly, has a long cycle, and is difficult to promote to other scenarios.

[0026] However, the tool chains in the prior art are not interoperable in the application of the above links. The tool chains cannot fully support the construction and application of the model, resulting in the inability of the tool chains to fully optimize the model. For example, in terms of supporting companies in model construction and application, there are still problems such as difficulty in obtaining multi-scale fine-tuning samples for power, lack of outstanding results in model fine-tuning technology innovation, excessive consumption of model storage and computing reasoning resources, lack of a complete platform tool chain to support the rapid implementation of the model, and failure to form an industry model application paradigm. Specifically, they are as follows: (1) It is difficult to obtain multi-scale samples. Due to data of different time, space or complexity, these data may be scattered and difficult to collect uniformly, and require professional equipment and technical support, which increases the difficulty and cost of sample collection. (2) Excessive consumption of model storage and computing resources. Due to the complex structure and huge parameters of large models, they face the challenge of excessive consumption of storage and computing resources in the process of implementation and application in the power field, which greatly limits the agile empowerment of the model in the power scenario. (3) The model reasoning speed is slow and the resource utilization efficiency is low. When the model performs inference and prediction, there are problems such as long inference time and insufficient utilization of computing resources, which cannot meet the needs of real-time monitoring and rapid response in business scenarios such as equipment, safety supervision, and marketing. (4) The model is difficult to implement. The model covers complex text data and multiple business needs, but how to implement strategies such as model automation processing, fine-tuning and adaptation, and lightweight deployment, and how to solve problems such as sample scarcity, efficient resource utilization, and online inference speed have not yet been clarified.

[0027] In view of this, the present application embodiment proposes a power cognitive large model development tool chain method, referring to Figure 1 , including the following steps:

[0028] Step 101: Acquire power data from multiple data sources using the tool chain.

[0029] In this step, toolchains play a vital role in software development and hardware design. They are a collection of tools and software used to complete specific tasks or processes. Toolchains are also indispensable in the construction and application of models, but toolchains cannot fully optimize models. In order to solve the above problems, this application proposes a platform toolchain that supports multiple capabilities such as large model fine-tuning, reasoning, and evaluation, a power cognitive large model toolchain, to promote the deep integration of artificial intelligence technology with production methods and management models, help power companies to transform digitally, and effectively support the construction of new power systems and high-quality development of enterprises. First, using the toolchain to obtain power data from multiple data sources can centrally collect scattered data. The power data from multiple data sources mainly comes from various links such as power generation, transmission, substation, distribution, power consumption, and dispatching, covering power grid operation, equipment detection or monitoring data, power company marketing data, and power company management data. Exemplarily, power data can include power equipment maintenance records, fault reports, operation logs, policy and regulatory documents, etc. This application has a sample management component and an integrated automated text data ingestion system to achieve batch collection and integration of power field documents, reports, logs, etc. from multiple scales in time and space. By acquiring power data from multiple data sources, the purpose of obtaining comprehensive power data is achieved.

[0030] Step 102: Use the tool chain to perform correlation processing on the power data to obtain target power data.

[0031] In this step, the sources of power data from multiple data sources are relatively scattered. If the power data is processed directly, the processing efficiency of the power data will be reduced. Therefore, it is necessary to use the tool chain to process the power data for correlation, integrate the related data, establish the data correlation, and obtain the target power data. The tool chain can automatically process complex data, improve data quality, lay a solid foundation for model training, and achieve the purpose of screening power data.

[0032] Step 103: Process the target power data using the optimized prediction model to obtain prediction data corresponding to the target power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain.

[0033] In this step, the tool chain provides a systematic method and tool that can significantly improve the performance and efficiency of the model. Therefore, the pre-built prediction model is optimized using the predetermined algorithm in the tool chain. This application constructs a unified standard, formulates the data standard, model standard, interface standard, etc. of the model, ensures the interoperability and data sharing between systems, and can use the optimized prediction model to process the target power data, so that the prediction data corresponding to the target power data output by the prediction model is more accurate. The model is trained by methods such as large-scale pre-training and multi-task learning. During the model training process, a variety of evaluation indicators are used to monitor and evaluate the model performance to ensure that the model has a high accuracy and generalization ability, so that it can adapt to different power business scenarios. It should be noted that the target power data comes from multiple data, the target power data is dynamically changing, and the prediction data output by the prediction model is associated with the target power data. For example, when the target power data is data associated with determining the electric energy, the prediction data output by the prediction model is the predicted electric energy; when the target power data is data associated with determining the power equipment failure, the prediction data output by the prediction model is the predicted fault power equipment. It should be noted that the model needs to be continuously optimized and upgraded. According to the actual operation of the power system, user feedback and model operation data are collected to continuously optimize and upgrade the model. Through incremental learning and online learning and other methods, the adaptability and accuracy of the model can be continuously improved.

[0034] In order to solve the problem of implementing large-scale power cognitive models, we will study the construction technology of multi-scale fine-tuning samples in the power industry, provide data support for model fine-tuning and optimization, tackle the adaptive fine-tuning technology of large-scale power cognitive models, improve the accuracy and reliability of large language models in specific tasks in the power field, and study the lightweight and reasoning technology of power cognitive models to shorten the response time of model calculation and deduction. Figure 2 As shown in the figure, a full-stack tool chain for large models is built, and a platform tool chain that supports multiple capabilities such as large-scale model pre-training, fine-tuning, reasoning, and evaluation is constructed to accurately meet the needs of the power industry and enable actual power business scenarios.

[0035] Through the above scheme, the tool chain is used to obtain power data from multiple data sources to achieve the purpose of obtaining comprehensive power data. The tool chain is used to perform correlation processing on the power data to obtain target power data to achieve the purpose of screening the power data. The target power data is processed using the optimized prediction model to obtain the prediction data corresponding to the target power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain to ensure that the model has a high accuracy and generalization ability, so that the prediction data corresponding to the target power data output by the prediction model is more accurate.

[0036] In some embodiments, the use of the tool chain to perform correlation processing on the power data to obtain target power data includes: using preset conditional variables in the conditional generation model in the tool chain to filter the power data to obtain and output the target power data; or, using the thinking chain algorithm in the tool chain to logically adjust the power data to obtain the target power data.

[0037] In this embodiment, if Figure 3 As shown, this embodiment provides two methods for correlation processing of power data, and the user can select one of them to process the power data according to the needs. Exemplarily, when the amount of power data is less than the predetermined amount of data, the conditional generation model can be used to process the power data. The tool chain has a built-in conditional generation model, in which preset conditional variables can be set according to user needs to achieve the purpose of screening power data and screen out data samples that meet the preset conditions. The research on small sample expansion technology of power text based on conditional enhancement is committed to developing text data enhancement methods suitable for the power industry, especially for small sample data sets. By using conditional generation models (such as conditional generation adversarial networks, conditional variational autoencoders, etc.), combined with power business background and domain knowledge, practical and high-quality power text data is generated. The research content includes designing a conditional generation model suitable for the characteristics of power text, defining and extracting conditional variables that affect text generation, and optimizing the generation process to ensure the diversity and reliability of generated text. It should be noted that the application of conditional generation models in power data screening mainly guides the data generation and screening process through preset conditional variables. Specifically, the conditional generation model first learns the characteristics of power data and the relationship between it and the preset conditions, and then evaluates the input power data based on these learned relationships to filter out data samples that meet the preset conditions. In view of the problems of traditional model outputs such as lack of contextual understanding, difficulty in adapting to new tasks, and unfocused output, we study model generation technology based on few-sample prompts to improve the accuracy and reliability of single-round dialogue generation results.

[0038] The toolchain has a built-in Chain-of-Thought (CoT) algorithm that can simulate the human thinking process to make logical adjustments to power data. The focus is on automatically generating fine-tuning samples that conform to business logic by analyzing and modeling power business logic and decision-making processes. The research on automatic generation of fine-tuning samples based on the Chain of Thought aims to develop a method for automatically generating fine-tuning samples for power text data using the Chain of Thought method. This research combines the principles of natural language processing (NLP) and cognitive science, and the research content includes building a Chain of Thought model, defining key reasoning steps in power business, designing data generation strategies, and ensuring the high quality and business relevance of generated data. The multi-round interactive task decomposition technology guided by the Chain of Thought is studied to improve the reasoning ability of large language models for complex power scenario tasks; the intelligent prompt generation algorithm is studied to automatically optimize the prompt engineering content according to user usage habits and scenarios to improve user experience.

[0039] Further supplement, the logical adjustment of power data using the thinking chain algorithm can be achieved through the following steps: Problem decomposition: Decompose the complex power data analysis problem into multiple simple sub-problems. For example, decompose the overall power grid stability analysis into sub-problems such as power load prediction in each region and equipment failure rate assessment. Step-by-step reasoning: For each sub-problem, use the thinking chain algorithm to simulate the human thinking process and gradually reason out the solution. This includes steps such as data collection, preprocessing, feature extraction, model training and prediction. Logical integration: Integrate the solutions to each sub-problem in a logical order to form a complete answer to the original complex problem. During the integration process, pay attention to the correlation and mutual influence between the sub-problems. Feedback and optimization: According to the data analysis results after logical adjustment, feedback and optimization are performed on the thinking chain algorithm to improve the accuracy and efficiency of future data analysis. Through the thinking chain algorithm, power data can be processed more systematically, the logic and accuracy of data analysis can be improved, and strong support can be provided for power grid operation and decision-making.

[0040] This embodiment aims to develop a sample enhancement method for power data, using conditional generation models combined with industry knowledge to generate high-quality data to meet the challenges of small sample data sets. Integrate the power text small sample expansion technology based on conditional enhancement, solve the problem of insufficient samples through data enhancement and synthesis, and improve the coverage and diversity of model training. This embodiment also explores the use of the thinking chain algorithm to simulate the human thinking process and automatically generate fine-tuning samples that conform to the power business logic to improve the quality and relevance of the data. The above method aims to improve the processing efficiency and accuracy of text data in the power field. Integrate the fine-tuning sample automatic generation technology based on the thinking chain algorithm, automatically generate an algorithm for high-quality training samples, and use the thinking chain algorithm to learn professional knowledge in the power field to solve the problems of sample scarcity and imbalance.

[0041] In some embodiments, the predetermined algorithm includes a predetermined adjustment strategy algorithm or a low-rank adaptive algorithm; the optimization method of the prediction model includes: optimizing the prediction model using the predetermined adjustment strategy algorithm in the tool chain; or optimizing the prediction model using the low-rank adaptive algorithm in the tool chain.

[0042] In this embodiment, in order to improve the performance of the prediction model, Figure 4 As shown, this embodiment provides two algorithms to optimize the preset prediction model, and the user can select one of the algorithms to optimize the model according to actual needs. In view of the fact that the traditional large model training fine-tuning relies on a large amount of labeled data to guide the model to learn specific tasks, there are learning bias or overfitting problems. A large model optimization and improvement framework for power professionals based on user feedback reinforcement learning is developed to provide a complete set of tools and methods for data collection, user feedback analysis, reinforcement learning algorithm implementation, etc. The optimal strategy is learned through user feedback such as evaluation, preference or satisfaction of the model generation results to guide model optimization. A preset adjustment strategy algorithm is preset in the tool chain, and the preset adjustment strategy algorithm can optimize the prediction model according to feedback. For example, the feedback includes user feedback such as evaluation, preference or satisfaction of the model generation results. By using user feedback as a reward signal, the model can gradually adjust the generation strategy to better meet user needs, help researchers and developers to more efficiently carry out the research and development of large model optimization for power professionals, and accelerate the improvement of model performance and the implementation of applications.

[0043] In view of the problems of high computational cost, easy overfitting, slow iteration speed, etc. in traditional model fine-tuning, this paper studies the efficient fine-tuning technology of model parameters based on low-rank adaptive algorithm. According to different training difficulties, training methods such as full parameters, partial parameters and LoRA are proposed to ensure the joint improvement of model performance and efficiency. It should be noted that the low-rank adaptive algorithm (LoRA) is a method for fine-tuning large pre-trained language models. It reduces the number of trainable parameters of downstream tasks by introducing low-rank matrix decomposition. LoRA freezes the weights of the pre-trained model and injects the trainable rank decomposition matrix into each layer of the Transformer architecture, thereby significantly reducing the computational cost and memory requirements. This method not only maintains the model performance, but also avoids the increase of inference delay. The advantage of LoRA lies in its high efficiency and flexibility. It can be easily applied to models of different scales and architectures, providing an effective solution for model fine-tuning.

[0044] The research and development of multi-business fine-tuning and resource management components for large power cognitive models aims to achieve multi-business fine-tuning of large cognitive models and effectively manage resources to ensure the efficiency and reliability of the fine-tuning process. This embodiment has a multi-business fine-tuning and resource management component, integrates a feedback mechanism based on a reward model, studies the incremental fine-tuning technology of model parameters, and realizes continuous optimization of model performance. Integrate efficient fine-tuning technology of model parameters based on low-rank adaptive algorithms to adapt to the changing business needs in the power field. Integrate the correction and optimization technology based on prompt word engineering, and improve the generation quality and accuracy of the model in specific business scenarios through customized prompt words. Build a multi-business fine-tuning and resource management platform to realize the automation and intelligence of model training, deployment and resource allocation. In addition, this application also studies the staged instruction fine-tuning technology of the model, divides the instruction samples into three parts: simple, medium and difficult, and designs scientific data matching, data sorting, and training rotation strategies to avoid instruction fine-tuning from damaging the original basic model capabilities and catastrophic forgetting problems. Research model distributed hybrid parallel training technology, and realize hybrid distributed parallel training of large models with tens of billions of parameters through data parallelism, pipeline parallelism, tensor parallelism and other training methods.

[0045] In some embodiments, before processing the target power data using the optimized prediction model, the method includes: packaging the optimized prediction model using the tool chain.

[0046] In this embodiment, the purpose of encapsulating the prediction model is to lightweight the complex cognitive model and provide a simple and easy-to-use interface and encapsulation to facilitate rapid deployment and call in various applications in the power field. This application has a large model lightweight and encapsulation component, encapsulating the lightweight model in an easy-to-deploy format, providing an API interface, simplifying the model integration and application development process, and ensuring the compatibility and stability of the model in different environments.

[0047] The role of encapsulation of the model is mainly reflected in improving the maintainability of the code, reducing coupling, protecting data security and achieving information hiding. Among them, improving code maintainability: encapsulation can avoid direct modification of data, reduce the risk of errors, and make it easier for maintainers to modify the code; reducing coupling: encapsulation realizes the isolation of the class from the outside, and the reduction of coupling between different modules reduces the complexity of the system and improves scalability; protecting data security: encapsulation can control the access rights to data. Only methods defined and authorized by the class can access the data, preventing security issues such as illegal tampering of data; achieving information hiding: encapsulation can effectively protect data and methods from being used by unauthorized people or programs, enhancing the security of the software.

[0048] The use of tool chains to encapsulate the optimized prediction model has the following advantages: (1) Reduce deployment costs: Provide modular and configurable solutions to reduce the cost and time of customized development, while supporting rapid deployment and flexible expansion. (2) Simplify the operation and maintenance process: Provide automated operation and maintenance tools to reduce the difficulty and cost of operation and maintenance and improve the stability and reliability of the system. (3) Application deployment and integration module: Deploy the trained large model into the power system and integrate it with the existing business system. By providing a unified API interface, data interaction and command transmission between the model and the business system can be realized.

[0049] In some embodiments, the packaging of the optimized prediction model using the tool chain includes: filtering the data in the optimized prediction model based on a predetermined distillation strategy; and compressing part of the data in the filtered prediction model based on a predetermined importance adjustment strategy.

[0050] In this embodiment, we focus on the "catastrophic forgetting" problem that occurs when the model updates knowledge, that is, the phenomenon that new learning tasks cause the model to forget old knowledge. In combination with self-distillation technology, we study an innovative predetermined distillation strategy that uses the model's own historical predictions as soft labels to guide the model to gradually refine knowledge representation during incremental learning while retaining key prior knowledge. The research content will cover the design of knowledge distillation algorithms, effective reuse mechanisms for historical knowledge, and adaptive adjustment methods for model structure and learning rate, to reduce forgetting of old tasks or data when the model is learning new tasks or new data. Integrate cognitive model self-distillation technology that mitigates catastrophic forgetting to improve the performance of lightweight models through knowledge transfer.

[0051] In view of the high dimensionality and dynamic characteristics of power system data, we explore scale-aware model quantization technology and build a multi-level model quantization framework. We automatically adjust the quantization granularity according to the importance of different features, use high-precision representation for key features, and implement more substantial quantization compression for non-key features. This ensures the accuracy of prediction and reasoning while significantly reducing model storage requirements and computing costs. We integrate scale-aware quantization technology to reduce the accuracy requirements of model parameters and the size of the model. At the same time, we study quantitative sensitivity analysis and error compensation mechanisms to ensure the robustness and accuracy of quantitative models in various power application scenarios and improve the practicality of the model.

[0052] It should also be added that if Figure 5As shown in the figure, in addition to encapsulating the prediction model, the prediction model is also inferred. In the face of real-time reasoning and decision-making needs in scenarios such as equipment operation and maintenance, safety supervision operations, and scheduling operations, a large model collaborative computing and reasoning framework is studied to optimize the reasoning process of large models through a distributed architecture. Integrate reasoning acceleration technology based on collaborative computing and use distributed computing resources to improve the reasoning speed of the model. The core idea is to build a model segmentation strategy, split the model into multiple computing nodes for parallel processing, and deploy the model in a distributed manner. At the same time, design cross-node data exchange protocols and dynamic scheduling algorithms, and use efficient communication mechanisms between nodes to achieve optimal configuration and load balancing of computing resources, and achieve efficient and stable reasoning in business scenarios such as equipment fault diagnosis, operation safety monitoring, and scheduling auxiliary decision-making.

[0053] In some embodiments, the method further includes: monitoring the performance of the optimized prediction model using the tool chain.

[0054] In this embodiment, the main purpose of using the tool chain model to monitor the optimized prediction model is to ensure the performance of the prediction model. Specifically, the prediction model monitoring aims to detect changes in the performance of the prediction model, promptly discover problems with the prediction model, evaluate the effectiveness of the prediction model in practical applications, and provide a basis for the optimization of the prediction model to ensure that the prediction model maintains stable performance in a changing data environment. The importance of prediction model monitoring is reflected in the following aspects: Monitoring data drift: Detect data drift early to ensure the accuracy and credibility of the machine learning prediction model. Timely detection of errors: Lock in potential risks. Methods to enhance prediction model performance: Optimize prediction model performance through continuous monitoring to gain a competitive advantage.

[0055] In some embodiments, before using the tool chain to perform correlation processing on the power data to obtain target power data, the method includes: using the tool chain to pre-process the power data.

[0056] In this embodiment, the research on power data preprocessing technology mainly focuses on how to effectively acquire and process power-related text data from different sources and at different scales. These data may include power equipment maintenance records, fault reports, operation logs, policy and regulatory documents, etc. The research content covers the steps of automatic collection, cleaning, format conversion, annotation, word segmentation, keyword extraction, semantic analysis, etc. of text data. Considering the terminology, specifications, and data formats unique to the power industry, the research needs to develop targeted methods and tools to ensure the integrity, accuracy, and consistency of the data.

[0057] Focusing on power data collection and preprocessing technology in the power sector, the research focuses on how to effectively collect, clean, convert and annotate power data from multiple sources and scales to ensure data integrity, accuracy and consistency. Integrate text data preprocessing technology, including cleaning, denoising, standardization and formatting, to ensure power data quality.

[0058] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.

[0059] It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0060] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a power cognitive large model research and development tool chain device.

[0061] refer to Figure 6 The power cognitive large model development tool chain device includes:

[0062] An acquisition module 10, configured to acquire power data from multiple data sources using the tool chain;

[0063] A first processing module 20 is configured to perform correlation processing on the power data using the tool chain to obtain target power data;

[0064] The second processing module 30 is configured to process the target power data using the optimized prediction model to obtain prediction data corresponding to the target power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain.

[0065] Through the above-mentioned device, the tool chain is used to obtain power data from multiple data sources to achieve the purpose of obtaining comprehensive power data. The tool chain is used to perform correlation processing on the power data to obtain target power data to achieve the purpose of screening the power data. The target power data is processed using the optimized prediction model to obtain the prediction data corresponding to the target power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain to ensure that the model has a high accuracy and generalization ability, so that the prediction data corresponding to the target power data output by the prediction model is more accurate.

[0066] In some embodiments, the first processing module 20 is further configured to use preset conditional variables in the conditional generation model in the tool chain to filter the power data to obtain and output the target power data; or, use the thinking chain algorithm in the tool chain to logically adjust the power data to obtain the target power data.

[0067] In some embodiments, the second processing module 30 is further configured such that the predetermined algorithm includes a predetermined adjustment strategy algorithm or a low-rank adaptive algorithm; the predetermined adjustment strategy algorithm in the tool chain is used to optimize the prediction model; or the low-rank adaptive algorithm in the tool chain is used to optimize the prediction model.

[0068] In some embodiments, a packaging module is further included, and the packaging module is configured to package the optimized prediction model using the tool chain before processing the target power data using the optimized prediction model.

[0069] In some embodiments, the encapsulation module is further configured to filter the data in the optimized prediction model based on a predetermined distillation strategy; and to compress part of the data in the filtered prediction model based on a predetermined importance adjustment strategy.

[0070] In some embodiments, a monitoring module is further included, wherein the monitoring module is configured to monitor the performance of the optimized prediction model using the tool chain.

[0071] In some embodiments, a preprocessing module is further included, and the preprocessing module is configured to use the tool chain to preprocess the power data before using the tool chain to perform correlation processing on the power data to obtain target power data.

[0072] For the convenience of description, the above device is described in terms of functions divided into various modules. Of course, when implementing the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0073] The device of the above embodiment is used to implement the corresponding power cognitive large model research and development tool chain method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0074] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the power cognitive large model R&D tool chain method as described in any of the above embodiments is implemented.

[0075] Figure 7 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.

[0076] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0077] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.

[0078] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0079] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

[0080] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0081] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0082] The electronic device of the above-mentioned embodiment is used to implement the corresponding power cognitive large model research and development tool chain method in any of the above-mentioned embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0083] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the power cognitive large model R&D tool chain method as described in any of the above embodiments.

[0084] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0085] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the power cognitive large model development tool chain method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0086] Based on the same concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a computer program product, including computer program instructions. When the computer program instructions are run on a computer, the computer executes the power cognitive large model research and development tool chain method as described in any of the above embodiments, which has the beneficial effects of the corresponding method embodiments and will not be repeated here.

[0087] It should be noted that the embodiments of the present application can be further described in the following manner:

[0088] It is understandable that before using the technical solutions of each embodiment of the present disclosure, the type, scope of use, usage scenarios, etc. of the personal information involved will be informed to the user in an appropriate manner, and the user's authorization will be obtained.

[0089] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly remind the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can independently choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0090] As an optional but non-limiting implementation, in response to receiving the user's active request, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0091] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0092] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0093] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present application, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.

[0094] Although the present application has been described in conjunction with specific embodiments of the present application, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0095] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A tool chain method for developing a large power cognitive model, characterized in that: include: Acquiring power data from multiple data sources using the tool chain; Using the tool chain to perform correlation processing on the power data to obtain target power data; The target power data is processed using the optimized prediction model to obtain prediction data corresponding to the target power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain.

2. The method according to claim 1, characterized in that The using the tool chain to perform correlation processing on the power data to obtain target power data includes: Using the preset condition variables in the condition generation model in the tool chain to screen the power data, obtain and output the target power data; or, The power data is logically adjusted using the thinking chain algorithm in the tool chain to obtain the target power data.

3. The method according to claim 1, characterized in that The predetermined algorithm includes a predetermined adjustment strategy algorithm or a low-rank adaptive algorithm; The optimization method of the prediction model comprises: Optimizing the prediction model using a predetermined adjustment strategy algorithm in the tool chain; or, The prediction model is optimized using a low-rank adaptive algorithm in the tool chain.

4. The method according to claim 1, characterized in that Before processing the target power data using the optimized prediction model, the method includes: The optimized prediction model is packaged using the tool chain.

5. The method according to claim 4, characterized in that The step of packaging the optimized prediction model by using the tool chain includes: Based on a predetermined distillation strategy, screening the data in the optimized prediction model; Based on a predetermined importance adjustment strategy, part of the data in the screened prediction model is compressed.

6. The method according to claim 1, characterized in that Also includes: The performance of the optimized prediction model is monitored using the tool chain.

7. The method according to claim 1, characterized in that Before using the tool chain to perform correlation processing on the power data to obtain target power data, the method includes: The power data is preprocessed using the tool chain.

8. A power cognitive large model development tool chain device, characterized in that: include: an acquisition module, configured to acquire power data from multiple data sources using the tool chain; A first processing module is configured to perform correlation processing on the power data using the tool chain to obtain target power data; The second processing module is configured to process the target power data using the optimized prediction model to obtain prediction data corresponding to the target power data, wherein the prediction model is optimized using a predetermined algorithm in the tool chain.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images