A power system operation decision method, system and terminal device

Abstract

The application discloses a power system operation decision method, system and terminal equipment, and belongs to the cross technical field of artificial intelligence and energy systems. The method is as follows: obtaining standardized prompt words based on user selection operation; inputting power characteristic data and the standardized prompt words into a generative pre-training conversion model to obtain a standardized prompt word template and call a frozen calculation step, calling corresponding professional power data calculation tools based on the standardized prompt word template and the power characteristic data to execute the frozen calculation step, and then obtaining a power system decision scheme based on obtained future period energy prediction values, visual charts, feature importance sorting and feature influence natural language explanation. The application discloses a power system operation decision method, system and terminal equipment, which can solve the technical problems that traditional power system decision relies on expert programming modeling and has high technical threshold, and effectively improves the real-time performance and reliability of power system decision.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and energy systems, and in particular to a power system operation decision-making method, system and terminal equipment. Background Technology

[0002] Since the Second Industrial Revolution, over-reliance on fossil fuels has led to global warming and severe environmental pollution. Improving energy efficiency and sustainability is crucial to reducing carbon emissions. Therefore, as a key sector for reducing carbon emissions, the energy industry must continuously evolve and improve its efficiency to reduce its environmental impact. Traditional decision-making problems in the energy industry often require adherence to specific norms, necessitating programming skills and domain knowledge. Emerging artificial intelligence technologies, utilizing generative pre-trained Transformer 4 (GPT-4) technology, can generate prompts rather than code, demonstrating human-like performance in professional activities and showing the potential to transform the energy industry in terms of efficiency and sustainability.

[0003] However, the power system industry currently faces a major pain point: with the exponential growth in the amount of electricity consumption data generated from multiple sources across all sectors, understanding the value of such massive amounts of data and programming it with specialized coding skills is crucial for real-time decision-making in integrating distributed energy resources in grid operations. However, traditional artificial intelligence technologies are unable to extract valuable knowledge from such vast amounts of data and struggle to learn from it. For example, grid integration of distributed energy resources requires real-time analysis of multi-source data (weather, load), and traditional AI relies on expert programming modeling, resulting in slow response and poor flexibility. Traditional optimization methods (such as MATPOWER for solving Optimal Power Flow (OPF)) require complete modeling and programming, leading to slow response and computational lag. Furthermore, rule engines cannot adapt to changing environments and exhibit poor generalization capabilities. Summary of the Invention

[0004] This invention provides a power system operation decision-making method, system, and terminal equipment, which can solve the technical problems of traditional power system decision-making relying on expert programming modeling and having high technical thresholds. It realizes the standardization and professionalization of the entire process from data input to operation adjustment, and effectively improves the real-time performance and reliability of power system decision-making.

[0005] This invention provides a power system operation decision-making method, comprising: Obtain standardized prompts based on user selections; The pre-acquired power feature data and the standardized prompt words are input into the pre-constructed generative pre-trained transformation model to obtain the standardized prompt word template and call the freeze calculation step bound to the standardized prompt words. Based on the standardized prompt word template and the power feature data, the corresponding professional power data calculation tool is called to execute the freeze calculation step. The freeze calculation steps include: Based on the standardized prompt word template and professional power data calculation tools, numerical calculations are performed on the power characteristic data to obtain energy prediction values ​​for future periods. Visual charts are generated based on the energy forecast values ​​for the future period. Based on the energy forecast values ​​for the future time period, power characteristic data, and professional power data calculation tools, the importance of features is ranked, and the natural language interpretation of the impact of features is obtained based on the ranking of feature importance. Based on the energy forecast values ​​for the future period, visualization charts, feature importance ranking, and natural language interpretation of feature impact, a power system decision-making scheme is obtained, so as to realize the operation adjustment of the power system based on the power system decision-making scheme.

[0006] This invention provides a power system operation decision-making method. It obtains standardized prompt words through user selection, and uses a generative pre-trained transformation model to retrieve pre-bound frozen calculation steps, avoiding logical deviations caused by the model's independent operation. Simultaneously, it calls upon professional power data calculation tools to perform numerical processing, compensating for the shortcomings of the generative pre-trained transformation model in power professional numerical calculations and ensuring the accuracy of data calculation and processing. Then, through the frozen calculation steps, it sequentially completes the prompt word template triggering, energy prediction numerical acquisition, visualization chart generation, feature importance ranking, and natural language interpretation acquisition, ultimately integrating these to form a decision-making scheme. This achieves standardization of the entire process from data input to operational adjustment, effectively reducing reliance on professional programming skills while improving the real-time performance and reliability of power system decision-making.

[0007] Furthermore, in the process of inputting the pre-acquired power feature data and the standardized prompt words into the pre-built generative pre-trained transformation model, obtaining the standardized prompt word template and calling the freeze calculation step bound to the standardized prompt words, and then calling the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word template and the power feature data, the pre-construction process of the generative pre-trained transformation model includes: Construct an energy knowledge base that includes professional knowledge data on power systems; Based on the energy knowledge base, a prompting project is carried out to construct a standardized prompting word template library that corresponds one-to-one with any standardized prompting word. Define a library of frozen calculation steps that corresponds one-to-one with the standardized prompt word templates in the standardized prompt word template library; Based on the energy knowledge base, standardized prompt word template library, and frozen calculation step library, an original generative pre-trained transformation model is constructed. Input the example data into the original generative pre-trained transformation model to obtain prompt word result pairs; The prompt word results are input into the original generative pre-trained transformation model, and context learning training and few-shot fine-tuning training operations are performed to obtain the generative pre-trained transformation model.

[0008] In the above scheme, an energy knowledge base containing professional power system knowledge data is constructed to provide professional data support for subsequent prompt word template construction and frozen calculation step definition. Then, prompt engineering based on the knowledge base ensures the professionalism and adaptability of standardized prompt words and binds them one by one with the frozen calculation steps to ensure process stability. Finally, the original generative pre-trained transformation model is subjected to context learning and few-sample fine-tuning based on the prompt word results. It can adapt to the power scenario without retraining the core parameters of the model. While reducing training costs, it enables the generative pre-trained transformation model to accurately understand the needs of the power field, thereby further improving the professionalism and adaptability of the final generated decision solution.

[0009] Furthermore, in the step of inputting the pre-acquired power feature data and the standardized prompt words into a pre-constructed generative pre-trained transformation model, obtaining a standardized prompt word template, and calling the freeze calculation step bound to the standardized prompt words, so as to call the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word template and the power feature data, the pre-acquired process of the power feature data includes: Acquire environmental and power data from monitoring points provided by the user; The environmental data and power data of the monitoring points are standardized to obtain normalized environmental data and normalized power data; The normalized environmental data and normalized power data are processed in matrix format to obtain power characteristic data.

[0010] In the above scheme, the environmental and power data of the monitoring points are first standardized and then converted into power characteristic data in matrix format. This eliminates the dimensional differences between different types of data and makes them conform to the input requirements of the model and professional calculation tools, thus avoiding calculation errors caused by the direct use of raw data.

[0011] Furthermore, the step of performing numerical calculations on the power characteristic data based on the standardized prompt word template and professional power data calculation tools to obtain energy forecast values ​​for future periods includes: Based on the standardized prompt word template and professional power data calculation tools, numerical calculations are performed on the power characteristic data to obtain power prediction time series and power generation prediction time series. Energy forecast values ​​for future periods are obtained based on the power forecast time series and the power generation forecast time series.

[0012] In the above scheme, the numerical calculation and processing results of power characteristic data are refined into power prediction time series and power generation prediction time series. The two types of time series present the future energy situation from the dimensions of real-time power output and total power generation, respectively, making the dimensions of future energy prediction values ​​more comprehensive and the description more accurate. This provides more specific basis for the subsequent formulation of renewable energy dispatch schemes and the generation of visualization charts, and helps to make decision-making schemes more targeted.

[0013] Furthermore, the generation of visualization charts based on the energy forecast values ​​for the future time period includes: An initial chart is generated based on the energy forecast values ​​for the future period. Retrieve user formatting instructions; The initial chart is formatted according to the user's formatting instructions to obtain a visual chart.

[0014] The above scheme allows users to issue format adjustment commands and user data adjustment commands based on the initial chart and complete the corresponding adjustments, meeting users' personalized needs for chart display formats, making the visualization charts more intuitive and easy to understand, facilitating users to quickly identify key operating condition information, providing a more convenient reference for the formulation and optimization of decision-making schemes, and improving the flexibility and practicality of human-computer interaction; it also enables the optimization of prediction results based on the latest actual data or experience, making the energy prediction values ​​for future periods more consistent with the actual scenario, thereby improving the feasibility and adaptability of the decision-making schemes formulated based on these values, and enhancing the flexibility of the final operational decision-making scheme.

[0015] Furthermore, the step of obtaining a power system decision-making scheme based on the future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and then adjusting the operation of the power system based on the power system decision-making scheme, includes: A renewable energy dispatch plan is formulated based on the energy forecast values ​​for the future period. Scheduling priorities are obtained based on the importance of the aforementioned features; Based on the renewable energy dispatch scheme, dispatch priority, and natural language interpretation of the influence of features, a power system decision scheme is obtained, so as to realize the operation adjustment of the power system based on the power system decision scheme.

[0016] The above scheme formulates a renewable energy dispatch plan based on future energy forecast values ​​to ensure that the dispatch strategy matches the expected energy supply; it determines the dispatch priority by combining the importance of features to tilt dispatch resources toward key factors that affect the forecast results; and it integrates the dispatch plan, priorities and natural language interpretation to form a decision plan, thereby achieving the precision and interpretability of the dispatch strategy, effectively improving the efficiency of renewable energy consumption, and ensuring the economy and stability of the power system operation.

[0017] Furthermore, before the step of obtaining a power system decision-making scheme based on the future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and then implementing the power system operation adjustment step based on the power system decision-making scheme, the following is also included: Based on the preset evaluation index data, the energy forecast values ​​for the future period are evaluated and calculated to obtain various error data. A future time period error report is generated based on the aforementioned error data; Based on the future time period error report, a prediction risk assessment is performed, and the prediction risk assessment result is obtained. When the predicted risk assessment result exceeds the preset reasonable risk threshold, the current prediction reliability is marked as insufficient, and the re-prediction process is triggered. The process returns to the numerical calculation of the power characteristic data based on the standardized prompt word template and professional power data calculation tool to obtain the energy prediction value for the future period.

[0018] In the above scheme, the reliability of the prediction results is quantified by calculating the error through preset evaluation index data, generating an error report, and conducting risk assessment. When the risk assessment result exceeds the threshold, the re-prediction process is triggered. The prediction deviation is corrected by repeatedly freezing the calculation steps, avoiding the formulation of decision-making schemes based on unreliable prediction results, and further improving the credibility of decision-making schemes and the safety of power system operation and adjustment.

[0019] The present invention also provides a power system operation decision-making system, including a standard word matching module and a decision calculation and processing module, wherein the decision calculation and processing module includes a decision submodule and a freeze calculation submodule, wherein: The standard word matching module is used to obtain the corresponding standardized prompt words based on the user's selection operation; The decision submodule is used to input the pre-acquired power feature data and the standardized prompt words into the pre-constructed generative pre-trained transformation model, obtain the standardized prompt word template, and call the freeze calculation step bound to the standardized prompt words, so that the freeze calculation submodule can call the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word template and the power feature data. The freeze calculation submodule is used to execute the freeze calculation steps: performing numerical calculations on the power feature data based on the standardized prompt word template and professional power data calculation tools to obtain future energy forecast values; generating visualization charts based on the future energy forecast values; obtaining feature importance rankings based on the future energy forecast values, power feature data, and professional power data calculation tools, and obtaining natural language explanations of feature impacts based on the feature importance rankings; and obtaining power system decision-making schemes based on the future energy forecast values, visualization charts, feature importance rankings, and natural language explanations of feature impacts, so as to achieve operational adjustments to the power system based on the power system decision-making schemes.

[0020] Furthermore, it also includes a risk assessment module, which, before the frozen calculation submodule executes the step of obtaining a power system decision scheme based on the future time period energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and then implementing the power system operation adjustment step based on the power system decision scheme, is used for: Based on preset evaluation index data, the energy forecast values ​​for the future period are evaluated and calculated to obtain various error data; a future period error report is generated based on the various error data; a prediction risk assessment is performed based on the future period error report to obtain the prediction risk assessment result; when the prediction risk assessment result exceeds a preset reasonable risk threshold, the current prediction reliability is marked as insufficient, and a re-prediction process is triggered, returning to the process of performing numerical calculations on the power characteristic data based on the standardized prompt word template and professional power data calculation tools to obtain the energy forecast values ​​for the future period.

[0021] The present invention provides a power system operation decision-making system. Through the division of labor and cooperation between the standard word matching module and the decision calculation and processing module, the process steps in the method are solidified into executable system modules, realizing the modularization and automation of power system operation decision-making, reducing manual intervention, improving the efficiency and consistency of decision execution, and ensuring the practicality and stability of the method in real applications.

[0022] The present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the power system operation decision method as described above.

[0023] The present invention provides a terminal device that executes a corresponding computer program through the processor of the terminal device, enabling the power system operation decision-making method of the present invention to be implemented on hardware devices. This breaks through the theoretical limitations of the method, allowing standardized and precise power system operation decisions to be quickly deployed and executed in real-world scenarios. It provides hardware support for the real-time optimization and adjustment of the power system, ensuring the practicality and operability of the technical solution.

[0024] This invention provides a power system operation decision-making method, system, and terminal equipment. By constructing an energy knowledge base, a standardized prompt word template library, and freezing calculation steps, and combining the logical adaptability of generative pre-trained transformation models with the numerical processing advantages of professional power data calculation tools, a complete closed-loop solution is formed, encompassing data preprocessing, predictive calculation, feature analysis, decision generation, and operational adjustment. This eliminates the need for specialized programming skills, effectively reducing expert reliance and lowering the technical threshold. Furthermore, error assessment and re-prediction mechanisms correct deviations, improving the reliability of prediction results. Feature importance analysis and natural language interpretation enhance the interpretability of decisions. Modular system design and terminal equipment implementation automate and efficiently execute the decision-making process. This effectively improves the power system's efficiency in absorbing renewable energy, ensures the safety, economy, and stability of grid operation, and adapts to the real-time decision-making needs of power systems in the context of distributed energy grid integration. Attached Figure Description

[0025] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of a power system operation decision-making method provided in this embodiment; Figure 2 This is a schematic diagram of a power system operation decision system provided in this embodiment. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0029] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0030] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0032] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0033] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0034] Example 1: This embodiment provides a power system operation decision-making method, such as... Figure 1 As shown, it includes: S1. Obtain the corresponding standardized prompt words based on the user's selection operation; S2. Input the pre-acquired power feature data and the standardized prompt words into the pre-constructed generative pre-trained transformation model, obtain the standardized prompt word template and call the freeze calculation step bound to the standardized prompt words, so as to call the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word template and power feature data; The freeze calculation steps include: S21. Based on the standardized prompt word template and professional power data calculation tool, perform numerical calculation processing on the power characteristic data to obtain energy prediction values ​​for future periods; S22. Generate a visualization chart based on the energy forecast values ​​for the future period; S23. Based on the energy forecast values ​​for the future period, power characteristic data, and professional power data calculation tools, obtain the feature importance ranking, and based on the feature importance ranking, obtain the natural language interpretation of the feature impact; S24. Based on the energy forecast values ​​for the future period, visualization charts, feature importance ranking, and natural language interpretation of feature impact, obtain a power system decision scheme, and adjust the operation of the power system based on the power system decision scheme.

[0035] In terms of power system operation, this paper analyzes the auxiliary role of GPT at the user, electronic equipment, power system, and energy service market levels to clarify how GPT can serve as an energy efficiency optimizer and sustainability promoter. To integrate more distributed energy resources into grid operation, efficient decision-making in areas such as power dispatch and frequency control ancillary services (FCAS) largely relies on a deep understanding of massive amounts of supply and demand data. The large model library GPT can provide these interpretations with hints, without requiring expert programming skills.

[0036] In practical implementation, with the rapid development of digital technology, smart energy systems are being designed to improve energy efficiency and enhance environmental sustainability. To further enhance the intelligence of energy systems, generative pre-trained transform models (GPT) can be introduced to provide personalized decision-making strategies, schedule efficient energy consumption tasks, monitor energy usage, and generate decision prompts. Ultimately, this can improve energy efficiency and identify and reduce carbon footprint. The steps are as follows: After inputting grid topology data, a standardized prompt is generated; then, the frozen optimization equation solution process is invoked; the GPT generative matrix laboratory (MATLAB, MATrix LABoratory) code is called; finally, the optimal power generation plan is output.

[0037] In practical applications, the generative pre-trained transformation model used in this embodiment can help optimize the operation of the energy system from four levels: users, devices, systems, and energy services. The specific implementation method is as follows: At the user level: The generative pre-trained transformation model can access diverse energy data from multiple sources and adjust according to the user's environmental conditions. Therefore, it can formulate energy management strategies by combining energy prices, load demand, and weather conditions, thereby reducing energy consumption and saving costs. Furthermore, the model can utilize historical user data to track energy footprints, estimate energy consumption status, and predict future energy consumption, providing more data information for power system operation decisions. Simultaneously, leveraging its large-scale model characteristics, the model can make in-depth interpretations based on weather conditions, different types of equipment, and user preferences, better understanding the user's actual needs. Therefore, it can provide real-time energy-saving strategy decisions integrating multi-source information by generating standardized prompts, supporting the generation and selection of standardized prompts.

[0038] At the equipment level: Addressing the issue of large-scale equipment deployment in energy systems, generative pre-trained transformation models can analyze massive amounts of monitoring data, combining equipment characteristics and environmental conditions to determine the optimal deployment locations and usage methods for different equipment. By optimizing equipment layout, they extend equipment lifespan, improve operating efficiency, and thus save energy and reduce waste. Furthermore, the model can use real-time equipment monitoring data to estimate and predict abnormal or irregular operating behaviors, providing a basis for equipment protection and efficiency improvement decisions. Addressing the maintenance challenges brought about by the increased number of devices, the model can generate maintenance recommendation prompts based on a vast maintenance knowledge base, providing real-time equipment maintenance strategies. Under harsh environmental conditions such as storms, extreme cold, and frequent load switching, technicians can perform equipment maintenance operations based on these recommendations, avoiding equipment damage and potential environmental pollution. Ultimately, it provides real-time maintenance decisions for inefficient equipment based on environmental conditions, achieving the goals of reducing energy consumption and improving sustainability. Simultaneously, it provides effective data support at the equipment level for acquiring power characteristic data, ensuring the comprehensiveness and accuracy of power characteristic data.

[0039] At the system level: As a vast and complex system encompassing various devices and millions of power grid nodes, the feasible method for optimizing the energy system based on the generative pre-trained transformation model proposed in this scheme is to construct a comprehensive energy knowledge base. This base relies on prior expert knowledge in the power system field to define and freeze the prompt formats and solution steps corresponding to standardized prompts—that is, the frozen calculation steps bound to the standardized prompts. Simultaneously, recommended prompts and solution steps are provided for solving the optimization problem. The model can output the determined optimal decision result according to the recommended standardized prompts and the frozen calculation steps. Therefore, it can obtain the optimal solution for power system optimization by leveraging prior expert knowledge, helping power system operators manage complex energy systems without requiring specialized programming skills. In the future, it is possible to construct a dedicated generative pre-trained transformation model for the energy sector with a parameter scale exceeding trillions. For optimal power flow problems and optimal decision-making required for frequency control ancillary services, under the performance challenges brought by millions of grid nodes, this large-parameter model can gradually decompose and track the problem through generated standardized prompts, and output optimal decision results with in-depth interpretation. This will effectively integrate distributed energy resources, improve the operating efficiency of the energy system, and promote the sustainable development of the energy system. This is consistent with the core logic of achieving power system optimization decision-making by freezing the calculation steps in this example.

[0040] At the energy service level: Generative pre-trained transformation models can provide real-time, 24 / 7 answers to user questions. Leveraging human-like dialogue and interaction capabilities, they enhance user engagement and can respond to various real-time questions with insightful interpretations, facilitating real-time decision-making in energy consumption strategies. For example, regarding the charging needs of electric vehicle owners, the model can use historical data on driving behavior and electricity prices, combined with questions raised by drivers about destination weather conditions, to infer and recommend the best charging station. Furthermore, the model can use real-time user queries to predict the power load at various grid nodes, thereby making decisions on power allocation during grid operation and ultimately improving the overall operational efficiency of the energy system. Simultaneously, it provides multi-dimensional demand references and data support for obtaining future energy forecast values ​​and formulating power system decision-making schemes as described in claim 1, ensuring that decision-making schemes match actual energy service needs.

[0041] In practical applications, the power system decision scheme generated by this approach is the optimal power flow decision. The Operating Power Flow (OPF) is used to optimize grid power allocation in real time to minimize generation costs and meet load demand, while integrating distributed energy resources (such as solar and wind power) to improve system efficiency and sustainability. Traditional OPF solutions require expert programming skills; this embodiment simplifies this process through GPT technology, enabling real-time decision-making.

[0042] In the specific implementation process, step S1 is used to trigger standardized prompt words for model execution. During this process, the user does not need to devise a question themselves, but selects or inputs keywords through the interface (i.e., through the user selection operation), thereby triggering the corresponding standardized prompt word template in the energy knowledge base when GPT is input. The placeholders in the standardized prompt word template (such as [prediction method]) will be automatically filled by the actual parameters in the power characteristic data.

[0043] Optionally, in step S2, the pre-construction process of the generative pre-trained transformation model includes: Construct an energy knowledge base that includes professional knowledge data on power systems; Based on the energy knowledge base, a prompting project is carried out to construct a standardized prompting word template library that corresponds one-to-one with any standardized prompting word. Define a library of frozen calculation steps that corresponds one-to-one with the standardized prompt word templates in the standardized prompt word template library; Based on the energy knowledge base, standardized prompt word template library, and frozen calculation step library, an original generative pre-trained transformation model is constructed. Input the example data into the original generative pre-trained transformation model to obtain prompt word result pairs; The prompt word results are input into the original generative pre-trained transformation model, and context learning training and few-shot fine-tuning training operations are performed to obtain the generative pre-trained transformation model.

[0044] In this implementation, the GPT model used is a pre-trained generative transformation model trained on a large-scale dataset. To adapt to the OPF scenario, this embodiment does not retrain the core GPT model. Instead, it customizes the model behavior by constructing an energy knowledge base and performing prompt engineering. The energy knowledge base, based on historical operational data and expert domain knowledge (such as power system parameters, constraints, and optimization objectives), is used to define specific prompt formats and freeze computation steps. The power system professional knowledge data in the energy knowledge base includes power system parameters (such as node data, generator data, branch data, etc.), constraints (such as power balance, voltage limits, etc.), and optimization objectives (such as minimizing costs, etc.). The energy knowledge base data originates from multi-source historical data, including weather information and load demand. Prompt engineering is manifested in generating standardized prompts based on the energy knowledge base, and then constructing a standardized prompt template library that corresponds one-to-one with any standardized prompt. Prompts are used to guide GPT to understand energy domain problems, avoiding direct programming, and ensuring input-output consistency by freezing the prompt format. The input is guaranteed to be standardized system baseline data (such as megavolt-ampere values, matrix-formatted node data, generator data, and branch data). The output is the optimization decision result (such as branch power flow, generator scheduling, etc.). The training process focuses on knowledge integration rather than model retraining, enabling GPT to handle energy-specific tasks through prompts.

[0045] The core function of standardized prompts is to constrain the model's output space, ensuring that GPT does not produce irrelevant or random responses, but rather strictly adheres to domain requirements in its calculations and analyses. The standardized prompts have a strong "instruction-response" relationship with the GPT output. The accuracy of the standardized prompts directly determines the interpretability and usability of the output results. The prompts explicitly specify which feature matrix GPT needs to process (e.g., "Please process the attached data_pv.csv"), thus linking natural language instructions with structured data. Ultimately, the input to GPT consists of standardized prompts and power characteristic data.

[0046] The standardized prompt template library consists of a set of structured, reusable natural language instruction templates predefined by domain experts. Experts write templates containing placeholders based on the objectives of each step. Frozen computation steps are crucial for stable system operation; they are defined as a pre-defined, immutable, sequential sequence of computation and analysis steps. These standardized prompt templates in the standardized prompt template library correspond one-to-one with the frozen computation steps in the frozen computation step library. For example: The template for S21 in the frozen calculation step is: "Please use the [prediction method] model to predict the data from the [data source]. The prediction target is the [target variable], the prediction duration is the [time range], and the input features include the [feature list]. Please output the predicted values ​​and calculate MAE (Mean Absolute Error), (Mean Squared Error), and RMSE (Root Mean Squared Error)." The template for S23 in the freeze calculation step is: "Please perform feature importance analysis on the above prediction results, using [analysis methods, such as permutation importance method], and explain why [key features, such as temperature] are important features."

[0047] In the specific implementation process, the core of the generative pre-trained transformation model training in this embodiment lies in building a large language model that can understand and accurately execute specific tasks in the field of energy forecasting. This training process is not a traditional model parameter retraining, but a model preparation process based on energy knowledge base prompting engineering and context learning. At the same time, it combines the method of freezing the calculation steps to ensure that the model can produce reliable and interpretable output results in the application stage of power system operation decision-making. This embodiment first constructs an energy knowledge base including power system professional knowledge data. This energy knowledge base is the foundation for the entire model training and subsequent operation of the power system decision-making system. It consists of two core parts: a domain knowledge base and a standardized prompt word template library. The domain knowledge base stores power system professional knowledge directly related to energy prediction, specifically including the following four parts: data source definition, which clarifies the source and specifications of the input data. For example, the solar irradiance, temperature, humidity, and historical load data specifically come from real-time or historical data of a specific solar power plant monitoring point or inverter equipment (such as photovoltaic power plant A) provided by a user. The data format is time series (such as data_pv.csv), ensuring data consistency and traceability; prediction model library, which pre-sets various classic prediction methods and their parameter ranges, such as linear regression (LR), back propagation neural network (BPNN), and gradient boosting regression. Regression (GBR), etc.; an evaluation index library for defining standard evaluation indicators, such as mean absolute error, mean square error, and root mean square error; domain rules, such as the physical relationship between environmental variables (temperature, humidity) and power generation efficiency, typical load curves for different seasons, and other power system professional rules. After constructing the energy knowledge base, a prompting engineering process is performed based on the energy knowledge base to build a standardized prompting word template library that corresponds one-to-one with any standardized prompting word. Then, a frozen calculation step library is constructed based on the power system domain expert knowledge definitions and the standardized prompting word templates in the standardized prompting word template library. Subsequently, an original generative pre-trained transformation model is constructed based on the completed energy knowledge base, standardized prompting word template library, and frozen calculation step library. Example data is input into the original generative pre-trained transformation model, which processes the data by combining the contents of the energy knowledge base, standardized prompting word template library, and frozen calculation step library to obtain prompting word result pairs. Finally, the prompting word result pairs are input back into the original generative pre-trained transformation model, allowing the model to perform context learning training and few-sample fine-tuning training operations based on the prompting word result pairs, completing the customized preparation of the model, and ultimately obtaining a generative pre-trained transformation model adapted to the power system operation decision-making task.

[0048] Optionally, in step S2, the pre-acquisition process of the power characteristic data includes: Acquire environmental and power data from monitoring points provided by the user; The environmental data and power data of the monitoring points are standardized to obtain normalized environmental data and normalized power data; The normalized environmental data and normalized power data are processed in matrix format to obtain power characteristic data.

[0049] In the specific implementation process, the raw data is collected in real time from power system monitoring equipment (such as smart meters and sensors), including power data at monitoring points, such as load demand, generator output, branch power flow, etc., as well as environmental data at monitoring points, such as weather information (such as temperature and light intensity).

[0050] In the specific implementation process, the raw data obtained above undergoes data preprocessing to convert it into a specific format that GPT can process. First, it is standardized according to a pre-defined data standard format to obtain normalized environmental data and normalized power data. Then, it is processed into a matrix format and organized into a matrix to obtain power characteristic data. For example, node data, generator data, and branch data are organized into a matrix format and incorporating pre-generated standardized prompts to obtain the power characteristic data. This preprocessing process ensures that the power characteristic data finally input into GPT conforms to the frozen calculation steps, avoiding errors when GPT directly solves the problem.

[0051] Next, the preprocessed power characteristic data is input into GPT. GPT generates a calculation process or code (i.e., the corresponding frozen calculation steps) based on the prompts (i.e., the standardized prompt word templates). For example, it generates MATLAB code to solve for OPF. The input instructions generated based on the standardized prompt word templates must follow the steps defined in the energy knowledge base.

[0052] GPT outputs and parses the results: GPT outputs frozen calculation steps or application programming interface (API) call instructions (such as calling the MATLAB optimization toolbox, included in the standardized prompt word template). Based on the API call instructions, it invokes the corresponding professional power data calculation tool (via an external API), and then, based on the standardized prompt word template and power characteristic data, calls the professional power data calculation tool to execute the frozen calculation steps, obtaining a power system decision scheme. Finally, the power system decision scheme (such as a generator dispatch plan) is transmitted to the grid controller, automatically adjusting the operating status of equipment in the power system to reduce energy consumption.

[0053] This embodiment ensures logical correctness by freezing the process. First, it checks and corrects each step in the GPT generation process. Then, it calls an external API to perform numerical calculations, solving the technical problem of traditional OPF solutions relying on expert coding. This embodiment, through GPT's hint engineering and API calls, requires no specialized programming skills, lowering the application threshold. Simultaneously, it addresses the unreliability of direct GPT solutions through a knowledge base and frozen steps. This enables real-time, efficient OPF decision-making, enhances the grid's integration capability with distributed energy resources, and promotes low-carbon emission goals.

[0054] Optionally, step S21 includes: performing numerical calculations on the power characteristic data based on the standardized prompt word template and professional power data calculation tools to obtain a power prediction time series and a power generation prediction time series; and obtaining energy prediction values ​​for future periods based on the power prediction time series and the power generation prediction time series.

[0055] Optionally, step S22 includes: generating an initial chart based on the energy forecast values ​​for the future period; obtaining a user format adjustment instruction; adjusting the format of the initial chart based on the user format adjustment instruction to obtain a visual chart.

[0056] Optionally, step S24 includes: formulating a renewable energy dispatching scheme based on the energy forecast values ​​for the future period; obtaining dispatching priorities based on the importance ranking of features; obtaining a power system decision scheme based on the renewable energy dispatching scheme, dispatching priorities, and natural language interpretation of feature impacts, so as to realize the operation adjustment of the power system based on the power system decision scheme.

[0057] In its implementation, GPT will strictly follow the frozen calculation steps bound to the standardized prompts: In step S21, GPT, based on the standardized prompts, calls upon professional power data calculation tools and uses a specified algorithm to perform numerical calculations on the power characteristic data (feature matrix) to obtain future energy forecast values. Based on preset evaluation indicators (such as root mean square error (RMSE), it generates a future time period error report to quantify the reliability of the forecast results. In step S22, GPT generates an initial visualization chart based on the future energy forecast values ​​and responds to user formatting or data adjustment commands, adjusting the initial chart accordingly to output a visualization chart that meets user needs, facilitating an intuitive presentation of energy forecast trends. In step S23, GPT analyzes the impact of each power characteristic based on the future energy forecast values, power characteristic data, and professional power data calculation tools, obtaining a ranking of feature importance. Based on this ranking, generative pre-training transforms GPT to generate corresponding natural language explanations of the feature impacts, clearly explaining the mechanism by which each feature affects the forecast results, thus achieving interpretable AI. When executing step S24, GPT integrates the future energy forecast values ​​and error reports from step S21, the visualization charts from step S22, and the feature importance ranking and natural language interpretation from step S23 to provide decision support, obtain power system decision schemes, and provide a basis for subsequent power system operation adjustments.

[0058] The final output of GPT (Power System Decision Program) is a comprehensive and interpretable result package that includes future time-period energy forecast values, visualization charts, interpretable analysis reports (feature importance ranking and natural language interpretation), and power system decision schemes. It is specifically applied to power allocation optimization and renewable energy integration assessment. When optimizing power allocation, the future time-period energy forecast values ​​from step S21 are input into the grid dispatch system. For example, if it is predicted that solar power generation will reach its peak at noon tomorrow, a power system decision scheme is automatically generated: "It is recommended to reduce the output of thermal power unit B by 50MW from 12:00 to 14:00, prioritizing the consumption of photovoltaic power." This achieves economic dispatch and reduces fossil fuel consumption. When assessing renewable energy integration, the future time-period error report (such as the RMSE index) generated in step S21 can be used to assess the reliability of the forecast, thereby evaluating the grid connection stability of the photovoltaic power station. Through the feature importance analysis in step S23, it can be found that humidity is a key factor affecting the forecast, thus recommending strengthened humidity monitoring of the power station environment in the power system decision scheme, providing data support for improving the integration quality of renewable energy and grid planning.

[0059] Optionally, before the step of obtaining a power system decision scheme based on the future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and then implementing the power system operation adjustment step based on the power system decision scheme, the method further includes: Based on the preset evaluation index data, the energy forecast values ​​for the future period are evaluated and calculated to obtain various error data. A future time period error report is generated based on the aforementioned error data; Based on the future time period error report, a prediction risk assessment is performed, and the prediction risk assessment result is obtained. When the predicted risk assessment result exceeds the preset reasonable risk threshold, the current prediction reliability is marked as insufficient, and the re-prediction process is triggered. The process returns to the numerical calculation of the power characteristic data based on the standardized prompt word template and professional power data calculation tool to obtain the energy prediction value for the future period.

[0060] In practical applications, when the predicted risk assessment result does not exceed the preset reasonable risk threshold, the current prediction reliability is marked as good, and step S24 is executed directly.

[0061] This embodiment provides a power system operation decision-making method that offers significant improvements compared to directly applying general-purpose large language models like ChatGPT. It specifically addresses the adaptation deficiencies of general-purpose large language models in power system decision-making scenarios: by freezing computation steps through standardized prompt words, the originally open and complex power energy prediction and decision-making task is decomposed into a standardized and repeatable pipeline: numerical calculation to obtain predicted values, generating visualization charts, obtaining feature importance and natural language interpretations, and generating decision schemes. This standardizes and solidifies the task process, completely avoiding the arbitrary processes and output instability caused by the free-flowing nature of general-purpose large language models. Furthermore, by constructing an energy prediction knowledge base containing professional power system knowledge data and combining it with prompt engineering to build standardized prompt word models... The board library deeply integrates expert knowledge in the power field into the entire model interaction process. Through knowledge base constraints and prompts, it ensures that the model output always aligns with the professional needs of the power system, achieving controllable knowledge guidance and output. This solves the problem of the lack of professionalism and accuracy in the output of general-purpose large language models, while providing professional knowledge support for the orderly execution of the frozen calculation steps. In the frozen calculation steps and decision-making process, it extends the pure energy prediction capability to feature importance analysis, natural language interpretation of feature impact, and power system decision support. This forms a closed-loop system from power feature data input and numerical calculation prediction to feature insight, final output of decision-making schemes, and guidance for operational adjustments. It achieves deep integration of prediction and decision-making, breaking through the limitation of general-purpose large language models that can only provide prediction results and cannot connect with actual decision-making.

[0062] This embodiment has the following significant advantages: High interpretability: By acquiring the feature importance ranking and the natural language interpretation of feature influence, users can not only obtain future energy forecast values ​​through step one, but also clearly understand the basis for the forecast results, clarifying why such a forecast was made. Simultaneously, through the generated power system decision-making scheme, they can obtain specific guidance on what to do next, realizing the application of interpretable AI in power decision-making scenarios. High reliability: Relying on the energy forecast knowledge base and standardized prompt word template library, as well as the fixed frozen calculation steps, it ensures that when different users execute the same power decision-making task at different times, consistent results are output according to a unified standard process, avoiding the randomness of the output of general-purpose large language models and improving the reliability of decision results. Effectively improved efficiency: The design of obtaining standardized prompt words through user selection operations allows users to trigger standardized decision-making processes without professional programming skills or prompt word debugging experience, significantly lowering the threshold for using AI models in power system scenarios. It also eliminates the need for repeated debugging of prompt words and verification of result accuracy, effectively improving the efficiency of power system operation decisions and adapting to the needs of scenarios such as real-time power system dispatching and renewable energy integration assessment.

[0063] This embodiment provides a power system operation decision-making method. By freezing calculation steps, using prompt engineering, and API calls, it constructs a collaborative decision-making system based on GPT (Geometry, Physics, and Power Parameters) that features real-time response and cross-scenario applicability, enabling real-time optimization of the power system. In terms of power system operation, it analyzes GPT's ability to address core issues such as energy efficiency optimization. It addresses the exponential growth in the amount of electricity consumption data from multiple sources across all industry sectors, emphasizing the crucial application need for understanding the value of such massive amounts of data and using professional coding skills for real-time decision-making in grid operation by integrating distributed energy resources. This solves the technical problem that traditional artificial intelligence technologies struggle to discover valuable knowledge from such large amounts of data and are difficult to learn from.

[0064] Compared with existing technologies, this embodiment has significant advantages. The generative pre-trained transformation model uses user selection to obtain standardized prompt words and constructs a standardized prompt word template library based on an energy prediction knowledge base and prompt engineering. This design eliminates the excessive reliance on domain experts in power system decision-making in existing technologies, eliminating the need for expert involvement in programming modeling, prompt word debugging, and process design. This significantly lowers the technical threshold for power system decision-making. Simultaneously, relying on fixed frozen calculation steps and rapid access to professional power data calculation tools, it achieves rapid response in the decision-making process, effectively improving the efficiency of power system operation decision-making. Furthermore, by converting multi-source heterogeneous data such as meteorological, load, and equipment status into standardized P... The rompt format effectively solves the technical challenge of fusing multi-source heterogeneous data in existing technologies, providing support for accurate input of GPT; it solidifies the knowledge of power industry experts into standardized modules that can be directly called, avoiding decision-making errors caused by the unrestrained development of generative pre-trained transformation models, and ensuring the reliability of decision results; through a human-machine intelligent collaborative task decomposition mechanism, it breaks down the complex task of power system decision-making into a logical design layer (such as freezing calculation step definitions and knowledge base construction) that engineers are good at and an execution layer (such as numerical calculation, feature analysis, and natural language interpretation generation) that generative pre-trained transformation models are good at, achieving complementary advantages between humans and machines, and further improving the accuracy and efficiency of decision-making.

[0065] Example 2: like Figure 2 As shown, this embodiment provides a power system operation decision-making system, including a standard word matching module and a decision calculation and processing module. The decision calculation and processing module includes a decision submodule and a freeze calculation submodule, wherein: The standard word matching module is used to obtain the corresponding standardized prompt words based on the user's selection operation; The decision submodule is used to input the pre-acquired power feature data and the standardized prompt words into the pre-constructed generative pre-trained transformation model, obtain the standardized prompt word template, and call the freeze calculation step bound to the standardized prompt words, so that the freeze calculation submodule can call the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word template and the power feature data. The freeze calculation submodule is used to execute the freeze calculation steps: performing numerical calculations on the power feature data based on the standardized prompt word template and professional power data calculation tools to obtain future energy forecast values; generating visualization charts based on the future energy forecast values; obtaining feature importance rankings based on the future energy forecast values, power feature data, and professional power data calculation tools, and obtaining natural language explanations of feature impacts based on the feature importance rankings; and obtaining power system decision-making schemes based on the future energy forecast values, visualization charts, feature importance rankings, and natural language explanations of feature impacts, so as to achieve operational adjustments to the power system based on the power system decision-making schemes.

[0066] Optional, such as Figure 2 As shown, the system also includes a model building module, which is used to: construct an energy knowledge base including power system professional knowledge data; perform prompting engineering based on the energy knowledge base to construct a standardized prompting word template library corresponding one-to-one with any standardized prompting word; define a frozen calculation step library corresponding one-to-one with the standardized prompting word templates in the standardized prompting word template library; construct an original generative pre-trained transformation model based on the energy knowledge base, the standardized prompting word template library, and the frozen calculation step library; input example data into the original generative pre-trained transformation model to obtain prompting word result pairs; input the prompting word result pairs into the original generative pre-trained transformation model, perform context learning training operations and few-sample fine-tuning training operations, and obtain a generative pre-trained transformation model.

[0067] Optional, such as Figure 2 As shown, the system also includes a data acquisition module, which is used to: acquire environmental data and power data of the monitoring points provided by the user; standardize the environmental data and power data of the monitoring points to obtain normalized environmental data and normalized power data; and process the normalized environmental data and normalized power data in a matrix format to obtain power characteristic data.

[0068] Optional, such as Figure 2 As shown, the frozen calculation submodule includes an energy prediction unit. This energy prediction unit performs numerical calculations on the power characteristic data based on the standardized prompt word template and a professional power data calculation tool to obtain energy prediction values ​​for future periods. Specifically, it performs numerical calculations on the power characteristic data based on the standardized prompt word template and the professional power data calculation tool to obtain a power prediction time series and a power generation prediction time series; and obtains energy prediction values ​​for future periods based on the power prediction time series and the power generation prediction time series.

[0069] Optional, such as Figure 2As shown, the frozen calculation submodule includes a visualization unit, which is used to generate visualization charts based on the energy forecast values ​​for the future time period.

[0070] Optionally, the visualization unit is used to: generate an initial chart based on the energy forecast values ​​for the future period; obtain user format adjustment instructions; adjust the format of the initial chart based on the user format adjustment instructions, and obtain a visualization chart.

[0071] Optional, such as Figure 2 As shown, the frozen calculation submodule includes a natural language interpretation unit, which is used to generate a visualization chart based on the energy forecast values ​​for the future period; obtain a feature importance ranking based on the energy forecast values ​​for the future period, power characteristic data, and professional power data calculation tools; and obtain a natural language interpretation of the feature impact based on the feature importance ranking.

[0072] Optional, such as Figure 2 As shown, the frozen calculation submodule includes a decision processing unit. This unit is used to obtain a power system decision scheme based on the future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact. The power system operation is adjusted based on this decision scheme. Specifically, this involves: formulating a renewable energy dispatch scheme based on the future energy forecast values; obtaining dispatch priorities based on the feature importance ranking; and obtaining a power system decision scheme based on the renewable energy dispatch scheme, dispatch priorities, and natural language interpretation of feature impact. The power system operation is then adjusted based on this decision scheme.

[0073] Optional, such as Figure 2As shown, the system also includes a risk assessment module. Before the frozen calculation submodule executes the step of obtaining a power system decision-making scheme based on the future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and then adjusting the power system operation based on the power system decision-making scheme, this risk assessment module performs the following: It performs index assessment calculations on the future energy forecast values ​​based on preset assessment index data to obtain various error data; it generates a future time period error report based on the various error data; it performs a prediction risk assessment based on the future time period error report to obtain the prediction risk assessment result; when the prediction risk assessment result exceeds a preset reasonable risk threshold, it marks the current prediction reliability as insufficient and triggers a re-prediction process, returning to the step of performing numerical calculations on the power feature data based on the standardized prompt word template and professional power data calculation tools to obtain the future energy forecast values. When the prediction risk assessment result does not exceed the preset reasonable risk threshold, it marks the current prediction reliability as good, at which point the decision processing unit continues to execute the corresponding steps.

[0074] Example 3: Based on the power system operation decision-making method described in Embodiment 1, this embodiment provides a step for performing a freezing calculation in a solar energy forecasting application scenario, specifically: The goal of solar energy prediction is to generate point-based predictions of future energy values ​​based on historical data. This involves obtaining power characteristic data through a pre-defined method. Specifically, user-provided environmental data (solar irradiance, temperature, humidity) and power data (historical load, etc.) from specific monitoring points are standardized to obtain normalized environmental and power data. These are then processed into a time series matrix using matrix format. This step utilizes standardized prompt templates and professional power data calculation tools to perform numerical calculations on the power characteristic data, outputting a power prediction time series and a power generation prediction time series for a specific future period (e.g., the next 24 hours). These two data series together generate the future energy prediction value (numerical time series). Simultaneously, based on pre-defined evaluation index data (e.g., MAE, MSE, RMSE), the predicted value is evaluated, and various error data are obtained to generate a future period error report. A prediction risk assessment is also completed concurrently. If the risk assessment result exceeds a pre-defined reasonable risk threshold, the prediction reliability is flagged as insufficient, triggering a re-prediction process and returning to the previous frozen calculation step. Adjusting the forecast results aims to visualize and calibrate the preliminary forecast results: First, an initial chart is generated based on the energy forecast values ​​for the future period. Then, according to the user's format adjustment instructions, the elements of the initial chart are adjusted to make the results clearer. Alternatively, according to the user's data adjustment instructions, the forecast values ​​are fine-tuned by combining a small amount of the latest actual data to finally obtain a visualization chart that meets the user's needs. The goal of feature importance analysis is to provide interpretability of prediction results: the generative pre-trained transformation model analyzes and outputs the degree of influence of each input feature (such as temperature and humidity) on the prediction results based on future energy prediction values, power characteristic data and professional power data calculation tools, and generates corresponding natural language explanations of feature influence based on the importance of the feature, clearly explaining the cause (such as rising temperature usually leads to a decrease in photovoltaic panel efficiency, but the effect is nonlinear at extremely low temperatures), thus achieving interpretable AI; Decision support based on forecast results aims to transform forecast information into actionable decisions. Based on the output of the aforementioned three steps, a renewable energy (solar) dispatch scheme is first formulated based on the energy forecast values ​​for the future period. Then, the dispatch priority is obtained based on the importance ranking of features. Finally, the renewable energy dispatch scheme, dispatch priority, and natural language interpretation of feature impact are integrated to obtain a power system decision scheme. Based on this decision scheme, the operation adjustment of power grid distribution and load management is realized, accurately supporting the power grid decision-making needs.

[0075] The training process of the generative pre-trained transformation model in this embodiment is as follows: First, the system constructs an energy knowledge base including power system professional knowledge data (covering data source definitions, prediction model libraries, evaluation index libraries, and domain rules related to solar energy prediction). Then, based on this energy knowledge base, prompt engineering is performed to construct a standardized prompt word template library that corresponds one-to-one with any standardized prompt word. At the same time, based on power system domain expert knowledge, a frozen calculation step library (i.e., the four specific frozen calculation steps in the above solar energy prediction scenario) is defined one-to-one with the standardized prompt word templates. Subsequently, based on this energy knowledge base, the standardized prompt word template library, and the frozen calculation step library, an original generative pre-trained transformation model is constructed. Example data related to solar energy prediction is input into the original model to obtain prompt word result pairs. Finally, the prompt word result pairs are input again into the original generative pre-trained transformation model to perform context learning training operations and few-sample fine-tuning training operations, ultimately obtaining a generative pre-trained transformation model adapted to the solar energy prediction scenario. Through this training process, the model learns to call the correct professional power data calculation tools (such as code interpreters to perform linear regression) and generate expected, structured outputs when it receives standardized prompts that conform to specific steps (i.e., standardized prompts obtained by the user based on the selection operation). This ensures the orderly and accurate execution of subsequent frozen calculation steps, further guaranteeing the reliability of solar energy forecasting and the professionalism of decision-making schemes.

[0076] In practical applications, energy forecasting is a critical time series problem. Traditional methods require complex modeling and coding skills. However, this embodiment uses generative pre-trained transformation model technology, combined with frozen calculation steps and energy knowledge base design, to achieve interpretable solar energy forecasting, effectively improving grid decision-making efficiency and energy system sustainability.

[0077] Example 4: Based on the above embodiments of the power system operation decision-making method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the power system operation decision-making method of any embodiment of the present invention.

[0078] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0079] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0080] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0081] Example 5: Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the power system operation decision method described in any of the above-described method embodiments of the present invention.

[0082] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0083] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A power system operation decision-making method, characterized in that, include: Obtain standardized prompts based on user selections; The pre-acquired power feature data and the standardized prompt words are input into the pre-constructed generative pre-trained transformation model to obtain the standardized prompt word template and call the freeze calculation step bound to the standardized prompt words. Based on the standardized prompt word template and the power feature data, the corresponding professional power data calculation tool is called to execute the freeze calculation step. The freeze calculation steps include: Based on the standardized prompt word template and professional power data calculation tools, numerical calculations are performed on the power characteristic data to obtain energy prediction values ​​for future periods. Visual charts are generated based on the energy forecast values ​​for the future period. Based on the energy forecast values ​​for the future time period, power characteristic data, and professional power data calculation tools, the importance of features is ranked, and the natural language interpretation of the impact of features is obtained based on the ranking of feature importance. Based on the energy forecast values ​​for the future period, visualization charts, feature importance ranking, and natural language interpretation of feature impact, a power system decision-making scheme is obtained, so as to realize the operation adjustment of the power system based on the power system decision-making scheme.

2. The power system operation decision-making method as described in claim 1, characterized in that, The pre-construction process of the generative pre-trained transformation model includes the following steps: inputting pre-acquired power feature data and standardized prompt words into a pre-built generative pre-trained transformation model; obtaining standardized prompt word templates and calling the freeze calculation step bound to the standardized prompt words; and calling the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word templates and power feature data. Construct an energy knowledge base that includes professional knowledge data on power systems; Based on the energy knowledge base, a prompting project is carried out to construct a standardized prompting word template library that corresponds one-to-one with any standardized prompting word. Define a library of frozen calculation steps that corresponds one-to-one with the standardized prompt word templates in the standardized prompt word template library; Based on the energy knowledge base, standardized prompt word template library, and frozen calculation step library, an original generative pre-trained transformation model is constructed. Input the example data into the original generative pre-trained transformation model to obtain prompt word result pairs; The prompt word results are input into the original generative pre-trained transformation model, and context learning training and few-shot fine-tuning training operations are performed to obtain the generative pre-trained transformation model.

3. The power system operation decision-making method as described in claim 1, characterized in that, In the process of inputting the pre-acquired power feature data and the standardized prompt words into a pre-constructed generative pre-trained transformation model, obtaining the standardized prompt word template, and calling the freeze calculation step bound to the standardized prompt words, and then calling the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word template and the power feature data, the pre-acquired power feature data process includes: Acquire environmental and power data from monitoring points provided by the user; The environmental data and power data of the monitoring points are standardized to obtain normalized environmental data and normalized power data; The normalized environmental data and normalized power data are processed in matrix format to obtain power characteristic data.

4. The power system operation decision-making method as described in claim 1, characterized in that, The process of numerically calculating and processing the power characteristic data based on the standardized prompt word template and professional power data calculation tools to obtain energy forecast values ​​for future periods includes: Based on the standardized prompt word template and professional power data calculation tools, numerical calculations are performed on the power characteristic data to obtain power prediction time series and power generation prediction time series. Energy forecast values ​​for future periods are obtained based on the power forecast time series and the power generation forecast time series.

5. The power system operation decision-making method as described in claim 1, characterized in that, The generation of visualization charts based on the energy forecast values ​​for the future time period includes: An initial chart is generated based on the energy forecast values ​​for the future period. Retrieve user formatting instructions; The initial chart is formatted according to the user's formatting instructions to obtain a visual chart.

6. The power system operation decision-making method as described in claim 1, characterized in that, The process of obtaining power system decision-making schemes based on future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and then adjusting the operation of the power system based on these decision-making schemes, includes: A renewable energy dispatch plan is formulated based on the energy forecast values ​​for the future period. Scheduling priorities are obtained based on the importance of the aforementioned features; Based on the renewable energy dispatch scheme, dispatch priority, and natural language interpretation of the influence of features, a power system decision scheme is obtained, so as to realize the operation adjustment of the power system based on the power system decision scheme.

7. The power system operation decision-making method as described in claim 1, characterized in that, Before the step of obtaining a power system decision-making scheme based on the future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and then implementing the power system operation adjustment based on the power system decision-making scheme, the following steps are included: Based on the preset evaluation index data, the energy forecast values ​​for the future period are evaluated and calculated to obtain various error data. A future time period error report is generated based on the aforementioned error data; Based on the future time period error report, a prediction risk assessment is performed, and the prediction risk assessment result is obtained. When the predicted risk assessment result exceeds the preset reasonable risk threshold, the current prediction reliability is marked as insufficient, and the re-prediction process is triggered. The process returns to the numerical calculation of the power characteristic data based on the standardized prompt word template and professional power data calculation tool to obtain the energy prediction value for the future period.

8. A power system operation decision-making system, characterized in that, It includes a standard word matching module and a decision calculation and processing module. The decision calculation and processing module includes a decision submodule and a freeze calculation submodule, wherein: The standard word matching module is used to obtain the corresponding standardized prompt words based on the user's selection operation; The decision submodule is used to input the pre-acquired power feature data and the standardized prompt words into the pre-constructed generative pre-trained transformation model, obtain the standardized prompt word template, and call the freeze calculation step bound to the standardized prompt words, so that the freeze calculation submodule can call the corresponding professional power data calculation tool to execute the freeze calculation step based on the standardized prompt word template and the power feature data. The freeze calculation submodule is used to execute the freeze calculation steps: performing numerical calculations on the power feature data based on the standardized prompt word template and professional power data calculation tools to obtain future energy forecast values; generating visualization charts based on the future energy forecast values; obtaining feature importance rankings based on the future energy forecast values, power feature data, and professional power data calculation tools, and obtaining natural language explanations of feature impacts based on the feature importance rankings; and obtaining power system decision-making schemes based on the future energy forecast values, visualization charts, feature importance rankings, and natural language explanations of feature impacts, so as to achieve operational adjustments to the power system based on the power system decision-making schemes.

9. A power system operation decision-making system as described in claim 8, characterized in that, It also includes a risk assessment module, which, before the frozen calculation submodule executes the process of obtaining a power system decision scheme based on the future energy forecast values, visualization charts, feature importance ranking, and natural language interpretation of feature impact, and before implementing the power system operation adjustment steps based on the power system decision scheme, is used for: Based on preset evaluation index data, the energy forecast values ​​for the future period are evaluated and calculated to obtain various error data; a future period error report is generated based on the various error data. Based on the future time period error report, a prediction risk assessment is performed, and the prediction risk assessment result is obtained. When the predicted risk assessment result exceeds the preset reasonable risk threshold, the current prediction reliability is marked as insufficient, and the re-prediction process is triggered. The process returns to the numerical calculation of the power characteristic data based on the standardized prompt word template and professional power data calculation tool to obtain the energy prediction value for the future period.

10. A terminal device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the power system operation decision method as described in any one of claims 1-7.

Images