Power grid dispatching instruction determination method and device, storage medium and electronic equipment

Abstract

The invention discloses a scheduling instruction determination method and device of a power grid, a storage medium and electronic equipment. The method comprises the following steps: collecting current operation data and current meteorological data of a target power grid; determining a target instruction template of the target power grid based on the current operation data; and based on the current operation data, the current meteorological data and the target instruction template, adopting a target scheduling model to determine a target scheduling instruction of the target power grid. The technical problem that the scheduling instruction determination result of the power grid is inaccurate in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of power systems, and more specifically, to a method, apparatus, storage medium, and electronic device for determining power grid dispatch instructions. Background Technology

[0002] With the development of smart grids and the widespread application of renewable energy, the operating status of power grids has become increasingly complex. The accuracy and timeliness of power grid dispatch instructions directly affect the stability, security, and efficiency of the power grid. Therefore, accurately determining the dispatch instructions for the target power grid is not only a basic requirement for daily power grid management but also crucial for coping with complex operating environments and improving the intelligence level of the power grid. Current technologies mainly rely on dispatchers manually analyzing current power grid operating data and weather information, and using their experience and expertise to formulate power grid dispatch instructions. This method is not only time-consuming and unable to respond in real-time to changes in the real-time operating status of the power grid, but also the formulation of dispatch instructions is subject to the personal experience of dispatchers, making it highly subjective and prone to bias. Therefore, current technologies suffer from the technical problem of inaccurate determination of power grid dispatch instructions.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method, apparatus, storage medium, and electronic device for determining power grid dispatch instructions, in order to at least solve the technical problem of inaccurate determination results of power grid dispatch instructions in related technologies.

[0005] According to one aspect of the embodiments of this application, a method for determining dispatch instructions for a power grid is provided, comprising: collecting current operating data and current meteorological data of a target power grid; determining a target instruction template for the target power grid based on the current operating data; and determining a target dispatch instruction for the target power grid by adopting a target dispatch model based on the current operating data, current meteorological data, and target instruction template.

[0006] According to another aspect of the embodiments of this application, a power grid dispatch instruction determination device is provided, comprising: a data acquisition module for acquiring current operating data and current meteorological data of a target power grid; a first determination module for determining a target instruction template of the target power grid based on the current operating data; and a second determination module for determining a target dispatch instruction of the target power grid based on the current operating data, the current meteorological data, and the target instruction template, using a target dispatch model.

[0007] According to another aspect of the embodiments of this application, a non-volatile storage medium is provided, which stores multiple instructions, the instructions being adapted for a power grid scheduling instruction determination method to be loaded by a processor and any one of them executed.

[0008] According to another aspect of the embodiments of this application, an electronic device is provided, including: one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the power grid dispatch instruction determination methods.

[0009] According to another aspect of the embodiments of this application, a computer program product is provided, which, when executed on a data processing device, is adapted to perform the steps of a method for determining dispatch instructions for a power grid.

[0010] In this embodiment, current operating data and meteorological data of the target power grid are collected; a target instruction template for the target power grid is determined based on the current operating data; and a target scheduling model is used to determine the target scheduling instruction for the target power grid based on the current operating data, meteorological data, and target instruction template. This achieves the goal of obtaining the target scheduling instruction for the target power grid by collecting current operating data and meteorological data and using a target scheduling model, thereby improving the accuracy of the target scheduling instruction determination result and solving the technical problem of inaccurate power grid scheduling instruction determination results in related technologies. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0012] Figure 1 This is a flowchart of a method for determining power grid dispatch instructions according to an embodiment of this application;

[0013] Figure 2 This is a flowchart of an optional power grid dispatch instruction determination method provided according to an embodiment of this application;

[0014] Figure 3 This is a schematic diagram of an optional power grid dispatch instruction determination device according to an embodiment of this application. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0018] Low-rank adaptation instruction technology is a method that simplifies and optimizes instruction sets by reducing the dimensionality of high-dimensional data to remove redundant information, retain key features, and thus improve system performance.

[0019] TensorBoard is a visualization tool used to monitor and understand the training process of deep learning models. It allows users to intuitively view the training progress of a model, evaluate its performance, debug its structure, and optimize its hyperparameters through a graphical interface.

[0020] According to an embodiment of this application, a method embodiment for determining dispatch instructions for a power grid is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0021] Figure 1 This is a flowchart of a method for determining power grid dispatch instructions according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0022] Step S102: Collect the current operating data and current meteorological data of the target power grid;

[0023] It is understandable that real-time data collection from the target power grid yields its current operational and meteorological data. The current meteorological data provides crucial information about the target power grid's operating environment. Combined with the current operational data, this allows for more comprehensive data support for target power grid dispatch instructions, thereby improving the accuracy of the dispatch instruction determination.

[0024] Step S104: Based on the current operating data, determine the target instruction template for the target power grid;

[0025] It is understandable that analyzing the current operating data of the target power grid helps determine the target instruction template. This template refers to a pre-defined format and logical framework of a series of operational instructions based on the current operating status and specific needs of the target power grid, used to guide the generation of target dispatch instructions. The target instruction template determined by analyzing the current operating data of the target power grid can specifically address the problems currently faced by the target power grid, such as overload and low-frequency oscillations, thereby improving the accuracy and effectiveness of the obtained target dispatch instructions.

[0026] In one optional embodiment, determining the target instruction template for the target power grid based on current operating data includes: determining the current operating status of the target power grid based on current operating data; determining the risk identification result of the target power grid based on the current operating status; and determining the target instruction template based on the risk identification result.

[0027] It is understandable that, based on the current operating data of the target power grid, the current operating status of the target power grid is determined, including key indicators such as equipment status, load conditions, and power balance. Based on this current operating status, the risk identification results of the target power grid are determined, such as equipment overload, voltage exceeding limits, and frequency anomalies. Based on the risk identification results of the target power grid, the target instruction template used to generate target dispatch instructions is determined. By analyzing the current operating data, the current operating status and risk identification results of the target power grid can be accurately determined, thereby generating a more accurate dispatch instruction template and improving the accuracy of the target dispatch instruction determination results.

[0028] Optionally, a real-time status assessment module and a risk warning module can be developed within the target scheduling model to determine the current operating status and risk identification results of the target power grid. The real-time status assessment module reads real-time data (i.e., current operating data) from the target power grid, performs data preprocessing and feature extraction, and utilizes the real-time status assessment module of the target scheduling model for real-time analysis. It outputs the current operating status of the target power grid, including key indicators such as equipment status, load conditions, and power balance, helping dispatchers to understand the target power grid's operating condition in a timely manner. The risk warning module, based on the target power grid's current operating data and status, and combined with a risk assessment algorithm, outputs the risk identification results of the target power grid. It quickly identifies and warns of potential risks in the target power grid's operation, such as equipment overload, voltage exceeding limits, and frequency anomalies, promptly reminding dispatchers to take appropriate measures.

[0029] Optionally, the target scheduling model may also include a decision suggestion generation module and a historical data analysis module. The decision suggestion generation module selects an appropriate target instruction template based on the current operating status of the target power grid and the risk identification results. It then calls the large-scale power grid scheduling model (i.e., the target scheduling model) and combines current operating data and current meteorological data to perform decision reasoning, generating reasonable scheduling decision suggestions (i.e., target scheduling instructions), such as adjusting generation plans, optimizing load allocation, and switching equipment status, providing scientific decision support for dispatchers. The historical data analysis module performs in-depth mining and analysis of the historical operating data of the target power grid, extracting key features and patterns to provide rich historical experience support for the target scheduling model, helping it better understand the operating trends of the target power grid.

[0030] Step S106: Based on the current operating data, current meteorological data, and target instruction template, the target scheduling model is used to determine the target scheduling instruction for the target power grid.

[0031] It is understandable that, based on current operational and meteorological data, a target scheduling model is used to generate target scheduling instructions for the target power grid according to the target instruction template. The generation of target scheduling instructions takes into account current operational and meteorological data, enabling the received instructions to reflect the current state of the target power grid and changes in the external environment in real time, thus improving the accuracy of the instructions. Furthermore, by utilizing the target instruction template and the target scheduling model, more refined and specific target scheduling instructions can be generated, addressing problems or potential risks in the target power grid in a targeted manner, thereby improving the relevance and accuracy of the target scheduling instructions.

[0032] In an optional embodiment, before determining the target scheduling instruction for the target power grid using a target scheduling model based on current operating data, current meteorological data, and the target instruction template, the method further includes: acquiring an initial historical dataset of the target power grid, wherein the initial historical dataset includes historical operating data, historical scheduling instructions, and historical meteorological data of the target power grid over a predetermined historical period; preprocessing the initial historical dataset to obtain a target historical dataset of the target power grid; dividing the target historical dataset to obtain a training dataset and a test dataset; using the training dataset to adjust the parameter values ​​of the first target parameter of the initial scheduling model to obtain a training scheduling model; using the test dataset to test the training scheduling model to obtain a test scheduling model; and injecting knowledge into the test scheduling model to obtain the target scheduling model.

[0033] The target scheduling model for the target power grid is determined as follows: First, historical operating data, historical scheduling instructions, and historical meteorological data of the target power grid over a predetermined historical period are acquired to form the initial historical dataset. Next, the initial historical dataset is preprocessed, including data cleaning, standardization, transformation, feature extraction, and dimensionality reduction, to obtain the target historical dataset for the target power grid. Then, the target historical dataset is divided according to a certain ratio, such as 8:2 or 7:3, to obtain a training dataset and a test dataset. Next, the parameters of the first objective parameter of the initial scheduling model are adjusted using the data from the training dataset to obtain the training scheduling model. The training scheduling model is then tested using the test dataset to obtain the test scheduling model. Finally, to improve the model's understanding and decision-making ability regarding the target power grid, knowledge is injected into the test scheduling model, such as professional knowledge about target power grid equipment parameters, operating constraints, and physical principles, resulting in the target scheduling model for the target power grid. By capturing patterns and trends in the target historical dataset and combining this knowledge injection, the determination of the target scheduling model is based on both rich historical experience and professional knowledge of power grid operation, thereby improving the reliability and accuracy of the determined target scheduling instructions.

[0034] Optionally, target power grid scheduling involves massive amounts of data and complex decision-making, which traditional methods struggle to handle efficiently. However, large-scale power grid scheduling models can integrate historical and real-time data from the target power grid, deeply learn its operational patterns, and accurately predict future states. By combining low-rank adaptation instructions, these models can quickly generate effective target scheduling instructions for different scheduling scenarios, improving scheduling efficiency and accuracy, and contributing to the stable operation of the target power grid.

[0035] Optionally, by surveying the architecture and performance of mainstream large language models, we can analyze their performance and application cases in natural language processing tasks. Based on the characteristics of the target power grid scheduling task, we evaluate the model's ability to understand technical terms, perform logical reasoning, and process historical data to ensure that the model can adapt to the complex needs of the target power grid domain. Simultaneously, we consider the model's scalability and flexibility to facilitate subsequent parameter updates and structural expansions based on changes in the target power grid's scheduling requirements. Finally, we select a model with high compatibility and adaptability with the target power grid domain as the base model (i.e., the initial scheduling model), laying a solid foundation for subsequent model fine-tuning and application.

[0036] Optionally, the above preprocessing may also include: organizing power grid dispatching experts to annotate the data, clarifying the meaning and purpose of the data to improve data quality, and converting the annotated data into a format suitable for training the initial dispatching model to ensure that the data can be effectively read and processed by the initial dispatching model to obtain the target historical dataset of the target power grid.

[0037] Optionally, evaluation metrics such as accuracy, recall, F1 score, and inference speed can be set according to the scheduling task requirements of the target power grid to comprehensively measure the performance of the trained scheduling model. The trained scheduling model is evaluated on a test dataset, and the values ​​of the above metrics are calculated to understand the generalization ability and predictive performance of the trained scheduling model. After the predictive performance and generalization ability of the trained scheduling model meet the requirements, a test scheduling model for the target power grid is obtained. The test scheduling model is compared and analyzed with the initial scheduling model and other power grid scheduling models to identify its strengths and weaknesses. Finally, an evaluation report is written, detailing the evaluation process, metric results, problem analysis, and optimization suggestions, providing a basis for continuous improvement of the scheduling model and ensuring that the scheduling model can meet the actual application needs of the target power grid.

[0038] In one optional embodiment, a training dataset is used to adjust the parameter values ​​of the first target parameter of the initial scheduling model to obtain a trained scheduling model. This includes: determining a training instruction template corresponding to the historical training scheduling instruction set in the training dataset, wherein the training instruction template is used to guide the initial scheduling model to generate scheduling instructions that conform to the target power grid operation logic by learning the historical training scheduling instruction set; determining the first target parameter based on the training instruction template; and adjusting the parameter values ​​of the first target parameter using the historical training operation dataset and the historical training meteorological dataset in the training dataset to obtain a trained scheduling model.

[0039] It is understandable that the training instruction template corresponding to the historical training scheduling instruction set in the training dataset is determined, and based on this template, the first target parameter that needs parameter adjustment in the initial scheduling model is determined, such as weights, biases, and thresholds. The parameter values ​​of the first target parameter are adjusted using the historical training operation dataset and historical training meteorological dataset in the training dataset to obtain the training scheduling model of the target power grid. Guided by the training instruction template, not only can the number of model parameters adjusted be reduced, but the model can also more explicitly learn the decision logic behind historical scheduling instructions, avoiding blind training, improving the model's learning efficiency and decision quality, and increasing the accuracy of the target scheduling instruction determination results.

[0040] Optionally, key parameters for fine-tuning training, such as learning rate, batch size, and number of training epochs, can be determined based on the training instruction template corresponding to the historical training scheduling instruction set in the training dataset. The learning rate is typically set between 1e-5 and 1e-4, and the batch size is chosen from 32 to 128 depending on the available GPU memory and the initial scheduling model size. Tools such as TensorBoard are used to monitor the loss function value, accuracy, and other metrics in real time during training to promptly identify and address overfitting or underfitting issues. Then, low-rank adaptation techniques are introduced to optimize traditional fine-tuning methods. Based on the training instruction template corresponding to the historical training scheduling instruction set in the training dataset, the first target parameters of the initial scheduling model are determined, such as weights, biases, and thresholds. This achieves the goal of training only a small number of first target parameters, reducing the number of parameters adjusted during training and lowering computational and storage costs. Detailed records of parameter settings, training time, and changes in loss values ​​during training are maintained to ensure the efficiency and traceability of the initial scheduling model training, providing a comprehensive record for the optimization and evaluation of the initial scheduling model.

[0041] Optionally, when the training instruction templates corresponding to the historical training scheduling instruction set are not unique, the initial scheduling model can be trained in the following way. First, multiple training instruction templates can be merged to extract their common parts and construct a comprehensive template. This comprehensive template should contain all key elements and decision points, ensuring that the decision logic learned by the initial scheduling model is comprehensive and universal. Then, based on this comprehensive template, the parameters of the initial scheduling model are adjusted to train it into a training scheduling model capable of handling multiple scheduling instruction patterns. If the importance or frequency of application of multiple training instruction templates differs, a template-weighted training strategy can also be used. During training, more common or more important training instruction templates are given higher weights, allowing the initial scheduling model to focus more on the decision logic behind these training instruction templates during learning. In this way, the scheduling model can generate more accurate target scheduling instructions when facing similar scenarios.

[0042] Optionally, when the training instruction templates corresponding to the historical training scheduling instruction set are not unique, the initial scheduling model can be trained in the following way: First, select a training instruction template as a guide for the initial training phase, and adjust the primary objective parameters of the initial scheduling model, such as the learning rate and batch size. After the initial scheduling model has initially learned the decision logic of a training instruction template, other training instruction templates are gradually introduced, and multiple rounds of training and parameter adjustment are conducted to enable the model to gradually adapt to and master the generation rules of scheduling instructions under different templates.

[0043] In one optional embodiment, determining the training instruction template corresponding to the historical training scheduling instruction set in the training dataset includes: for any historical training scheduling instruction in the historical training scheduling instruction set, decomposing the instruction into multiple components; based on these components, using natural language processing techniques to determine the semantic features corresponding to each component; based on the semantic features corresponding to each component, determining any initial instruction template for any historical training scheduling instruction; based on the historical scheduling experience of the target power grid, modifying the initial instruction template to obtain any modified instruction template for any historical training scheduling instruction; determining the modified instruction templates corresponding to multiple historical training scheduling instructions in the historical training scheduling instruction set by using the method of determining any modified instruction template; and merging the modified instruction templates corresponding to multiple historical training scheduling instructions to obtain the training instruction template.

[0044] The following approach is used to determine the training instruction template corresponding to the historical training scheduling instruction set: For any historical training scheduling instruction in the historical training scheduling instruction set, firstly, the instruction is decomposed to obtain multiple components, such as the operation object (e.g., generator, transformer), operation type (e.g., start-up, shutdown, adjustment), operation conditions (e.g., time, status), and expected result (e.g., power adjustment, voltage stabilization). Next, natural language processing techniques are used to extract the semantics of these components, obtaining the semantic features corresponding to each component. Then, based on the semantic features corresponding to each component, an initial instruction template for any historical training scheduling instruction is determined. Next, the initial instruction template is modified using the historical scheduling experience of the target power grid, such as professional knowledge and expert experience in the target power grid scheduling field, to obtain a modified instruction template for any historical training scheduling instruction. By determining the modified instruction template, the modified instruction templates corresponding to multiple historical training scheduling instructions in the historical training scheduling instruction set are determined. Pattern recognition is performed on the modified instruction templates corresponding to multiple historical training scheduling instructions to identify recurring instruction structures and common semantic feature combinations. Based on the pattern recognition results, similar modified instruction templates are merged and simplified to form a general instruction template set covering multiple scenarios. The merged instruction template set is further optimized to obtain the training instruction template, ensuring its simplicity, universality, and readability for easy model understanding and learning. The determination and optimization of the training instruction template standardizes and structures the target scheduling instructions, enabling even complex instructions to be clearly recognized and imitated by the model, reducing comprehension errors caused by unclear instruction descriptions or inconsistent formats. Simultaneously, through the analysis and templating of various types of scheduling instructions in the historical training instruction set, the training scheduling model can learn more diverse scheduling logic, enhancing its adaptability and generalization ability to unknown or similar scenarios.

[0045] Optionally, considering the characteristics of the target power grid's scheduling tasks, an instruction method that efficiently combines domain knowledge with the scheduling model can be designed. This can be achieved by deeply studying the principles of low-rank adaptation technology, analyzing the language structure and semantic characteristics of the target power grid's scheduling instructions, and developing suitable instruction templates. This allows for a significant reduction in the number of training parameters (i.e., the first target parameters) while maintaining the performance of the scheduling model, improving the training efficiency and adaptability of the scheduling model. This enables the scheduling model to more accurately understand the scheduling needs of the target power grid, providing technical support for subsequent scheduling decisions.

[0046] Optionally, the collected instruction samples (i.e., historical dispatch instructions) can be decomposed to identify key components, including the operation object (e.g., generator, transformer), operation type (e.g., start / stop, adjustment), operation conditions (e.g., time, status), and expected result (e.g., power adjustment, voltage stabilization). Decomposition clarifies the syntactic structure and logical relationships of the instructions. Natural language processing techniques are then used to extract the semantic features of the instruction components, understanding the specific meaning and role of each component in the target power grid dispatch. For example, identifying the operation object (generator), operation type (increasing active power), and expected result (improving power grid supply capacity) in the instruction "increase generator active power".

[0047] Optionally, based on the characteristics of historical power grid dispatch instructions obtained from analysis, the structure of the initial instruction template can be defined, including fixed parts (such as operation verbs and object types) and variable parts (such as specific values ​​and time). The parameters in the initial instruction template are determined; these parameters can be adjusted according to specific circumstances to generate instructions for a specific task. For example, parameters may include time, equipment name, parameter type (such as power and voltage), and target value. Professional knowledge and experience in the target power grid dispatching field are integrated into the initial instruction template design (i.e., the initial instruction template is modified based on the historical dispatching experience of the target power grid) to obtain a modified instruction template, ensuring that the generated target dispatch instructions conform to the physical laws and operational procedures of the target power grid. For example, ensuring that the operation sequence in the target dispatch instructions conforms to the target power grid's safe operating procedures and that the parameter range is within the allowable range of the equipment.

[0048] In one optional embodiment, knowledge injection is performed on the test scheduling model to obtain the target scheduling model, including: constructing an initial knowledge graph of the target power grid; and injecting the initial knowledge graph into the test scheduling model to obtain the target scheduling model.

[0049] This process involves acquiring professional knowledge about the target power grid, including its equipment parameters, operational constraints, and physical principles, and constructing an initial knowledge graph of the target power grid. This initial knowledge graph is then injected into the test scheduling model to obtain the target scheduling model for the target power grid. The injection of the initial knowledge graph endows the target scheduling model with a certain level of "power grid common sense," enabling it to make more reasonable scheduling decisions based on operating rules and equipment characteristics, and avoiding the generation of target scheduling instructions that violate basic physical laws or operational principles.

[0050] Optionally, an initial knowledge graph can be constructed, and specialized knowledge in the power grid domain can be structured and integrated into the test scheduling model. First, a comprehensive collection of specialized knowledge regarding the target power grid, including equipment parameters, operational constraints, physical principles, and scheduling procedures, as well as historical scheduling cases and expert experience, is gathered. This information serves as the nodes and relationships of the initial knowledge graph. Next, the entities (such as generators, transformers, and lines) and relationships (such as "connection," "impact," and "belong to") of the initial knowledge graph are defined, constructing its framework. Then, the collected knowledge is populated into the initial knowledge graph, establishing connections between entities to form a structured knowledge network of the target power grid. Finally, the initial knowledge graph is regularly updated and maintained to ensure it reflects the latest developments and changes in the target power grid, providing the target scheduling model with up-to-date domain knowledge support.

[0051] In one optional embodiment, injecting an initial knowledge graph into a test scheduling model to obtain a target scheduling model includes: aligning the initial knowledge graph with the test scheduling model to obtain a target knowledge graph; adjusting the structure of the test scheduling model and the parameter values ​​of the second target parameter based on the target knowledge graph to obtain an adjusted scheduling model; and training the adjusted scheduling model using the target knowledge graph to obtain the target scheduling model.

[0052] Understandably, to match the professional knowledge in the initial knowledge graph with the internal representation of the test scheduling model, knowledge alignment is performed between the initial knowledge graph and the test scheduling model. This involves aligning the entities and relationships in the initial knowledge graph with the parameters and structure in the test scheduling model to obtain the target knowledge graph. The target knowledge graph is then used to adjust the structure of the test scheduling model and the parameter values ​​of the second objective parameter, resulting in an adjusted scheduling model. Simultaneously, the target knowledge graph is used to train the adjusted scheduling model, yielding the target scheduling model for the target power grid. Through knowledge alignment and structural parameter adjustment, the target scheduling model can directly utilize the logical rules in the target knowledge graph for decision-making, enhancing the decision-making logic of the target scheduling model and making the generated template scheduling instructions more reasonable and accurate.

[0053] Optionally, an effective connection can be established between the initial knowledge graph and the test scheduling model by designing a knowledge injection mechanism. First, a dedicated interface is designed in the test scheduling model to receive and process knowledge from the initial knowledge graph. Then, a mapping algorithm is developed to map entities and relationships in the initial knowledge graph to the parameters or activation functions of the test scheduling model, enabling the model to understand and utilize this knowledge. Next, a specific knowledge injection process is defined, including knowledge extraction, transformation, and loading steps, ensuring that the knowledge from the initial knowledge graph can be smoothly integrated into the test scheduling model. Finally, parameters during the injection process, such as the learning rate and knowledge weights, are adjusted to optimize the knowledge injection effect, allowing the test scheduling model to better absorb and apply domain knowledge.

[0054] Optionally, knowledge alignment can be achieved as follows: First, align the entities and relations in the initial knowledge graph with the representations within the test scheduling model to ensure that the test scheduling model can correctly identify and process this knowledge, thus obtaining the target knowledge graph. Next, based on the structure of the target knowledge graph, adjust the parameters (i.e., the second target parameters) or architecture of the test scheduling model to obtain an adjusted scheduling model, ensuring that the representation of the adjusted scheduling model matches the target knowledge graph. Then, train the adjusted scheduling model using the aligned data (i.e., the target knowledge graph) to enhance its understanding and application capabilities of the target knowledge graph, thus obtaining the target scheduling model. Finally, verify the effectiveness of knowledge alignment through experiments to ensure that the target scheduling model can accurately combine the knowledge in the target knowledge graph with its internal representations, thereby improving the scheduling accuracy and interpretability of the target scheduling model.

[0055] Optionally, the effects of knowledge injection and alignment can be evaluated as follows: First, design a verification experiment, including experimental data, evaluation metrics, and procedures, to verify the performance of the target scheduling model after knowledge fusion. Then, compare the target scheduling model with the unfused knowledge to evaluate the performance improvement, focusing on the model's accuracy and interpretability in the target power grid scheduling task. Next, invite power grid scheduling experts to evaluate the output of the target scheduling model, collect their feedback, and understand its performance in practical applications. Finally, based on the verification results and expert feedback, analyze the problems in the knowledge fusion process, optimize and adjust the target scheduling model and knowledge injection mechanism to ensure that knowledge fusion effectively improves the performance and practicality of the test scheduling model.

[0056] Optionally, adjusting the parameter values ​​of the test scheduling model is a key step in injecting the initial knowledge graph into the test scheduling model. During the knowledge alignment phase, parameter adjustment primarily focuses on matching the input, output, and internal structure of the test scheduling model with the target knowledge graph. For example, if the target knowledge graph contains attribute sets for specific devices, the test scheduling model may need to adjust the size and format of its input layer to handle this attribute data. Furthermore, some parameters of the test scheduling model may need to be initialized or modified so that the model can correctly "interpret" the meaning of entities and relationships in the target knowledge graph. This can be achieved by converting entities in the target knowledge graph into vector representations using embedding methods. This process typically involves parameter adjustment to ensure that the entity vectors reflect their semantics and relationships within the target knowledge graph. During the training phase, the goal of parameter adjustment is to optimize the performance of the scheduling model, enabling it to learn the best scheduling strategies from historical operational and meteorological data, while integrating the rules and logic of the target knowledge graph into the decision-making process. Parameter tuning in this stage is typically based on a backpropagation algorithm using a loss function (such as cross-entropy loss, mean squared error, etc.). This involves dynamically adjusting parameters such as scheduling model weights and biases to minimize the difference between the model-predicted scheduling commands and the actual scheduling commands. Through training in conjunction with a target knowledge graph, the parameter tuning of the scheduling model also considers how to better utilize the information in the target knowledge graph, ensuring that the generated target scheduling commands are not only based on historical data patterns but also follow the physical rules of power grid operation and professional knowledge such as expert knowledge.

[0057] Through the above steps S102 to S106, the goal of obtaining the target dispatch command of the target power grid by collecting the current operation data and current meteorological data of the target power grid and adopting the target dispatch model can be achieved. This improves the accuracy of the target dispatch command determination result of the target power grid and solves the technical problem of inaccurate power grid dispatch command determination result in related technologies.

[0058] Based on the above embodiments and optional embodiments, this application proposes an implementation method for determining dispatch instructions for a power grid. This implementation method can be understood as a power grid dispatch operation auxiliary decision-making method based on low-rank adaptation instructions, to solve the problems of low model adaptation efficiency, insufficient integration of domain knowledge, and poor interpretability of dispatch decision results faced by the target power grid in dispatch decision-making. This method, by introducing low-rank adaptation instruction technology, improves the target dispatch model's understanding and application ability of the target power grid dispatch domain knowledge, enhances the accuracy and interpretability of the generated target dispatch instructions, and provides a more efficient, intelligent, and reliable solution for the target power grid's dispatch decision-making. It plays a crucial role in promoting the intelligent transformation and upgrading of power grid dispatch and improving the operational efficiency and decision-making quality of complex power grid dispatch.

[0059] Figure 2 This is a flowchart of an optional power grid dispatch instruction determination method provided according to an embodiment of this application, such as... Figure 2 As shown, the steps of this method include:

[0060] Step S1: Requirements Analysis and Problem Definition.

[0061] By investigating the current dispatching status and business needs of the target power grid, the pain points and difficulties in the dispatching work of the target power grid were accurately identified, pointing out the direction and goals for subsequent research. For example, it was clarified that there are problems such as low efficiency in dispatching model adaptation and insufficient integration of domain knowledge in the dispatching of the target power grid, providing a direct basis for designing a low-rank adaptation instruction method. At the same time, the business processes and requirements were analyzed to clarify the application scenarios and functional objectives of the dispatching decision-making method, ensuring that the developed method closely aligns with actual business needs and improves the practicality and effectiveness of the method. Finally, the research objectives and scope were determined, providing clear planning and guidance for the entire research process, rationally allocating research resources, and ensuring the efficient conduct of the research work. The above steps laid a solid foundation for the entire research process, ensuring that the finally developed dispatching decision-making method can effectively solve practical problems and meet the dispatching business needs of the target power grid.

[0062] Step S11: Investigate the current status of power grid dispatching;

[0063] Collect dispatch data from the target power grid to understand its dispatching workflow and operational needs. Collect operational data from the target power grid, including grid topology, equipment parameters, historical load data, and fault records, to analyze the current status and existing problems of target power grid dispatching. Review relevant literature and reports to analyze current development trends and application cases of power grid dispatching technologies, providing theoretical support for target power grid dispatching research. Observe the actual operation process of target power grid dispatching, recording the dispatchers' workflow and operating habits to provide a reference for subsequent research.

[0064] Step S12: Analyze business processes and requirements;

[0065] A detailed analysis of the target power grid dispatching business processes was conducted, including real-time monitoring, fault prediction, load allocation, and dispatching plan formulation, clarifying the specific tasks and operational steps of each stage. Discussions were held with dispatching experts and staff to determine the requirements for dispatching decision-making functions at each stage, such as real-time data processing, forecast accuracy requirements, and response time for dispatching recommendations. Historical data and case studies were analyzed to understand the target power grid dispatching business needs and decision-making patterns in different scenarios, providing a reference for functional design. Based on the existing dispatching system's functions, its shortcomings in the business processes were evaluated, and directions for improving dispatching decision-making methods were identified.

[0066] Step S13: Determine the research objectives and scope.

[0067] Based on the survey results and business needs analysis, the research objectives are clearly defined, such as improving the intelligence level of the target power grid dispatch, enhancing the adaptability efficiency of the target dispatch model, and strengthening the integration of domain knowledge. In conjunction with the research objectives, the research scope is determined, including specific research content, technical routes, and application scenarios. A research plan is developed, clarifying the tasks and timelines for each stage to ensure the orderly progress of the research work.

[0068] Step S2, data collection and preprocessing.

[0069] By integrating historical, real-time, and multi-source heterogeneous data of the target power grid, comprehensive materials are provided for the training of the target scheduling model. After preprocessing such as denoising and filling missing values, the integrity and accuracy of the multi-source heterogeneous data are ensured, improving the stability and generalization ability of the scheduling model training, and providing high-quality data support for subsequent research. This ensures that the target scheduling model can accurately learn the operating rules of the target power grid and provides a reliable basis for the intelligent scheduling of the target power grid.

[0070] Step S21: Determine the data source;

[0071] We identified various data nodes, including power grid topology, equipment parameters, historical operation records, real-time monitoring data, and meteorological data. We established a data source list, recording detailed information for each source. Simultaneously, we evaluated the relevance and reliability of each data source to the research objectives, initially screening out suitable data sources for scheduling model training and testing. We also reached acquisition agreements with relevant data providers to ensure the legality and feasibility of the data sources, laying the foundation for subsequent data collection.

[0072] Step S22: Collect historical and real-time data;

[0073] Historical datasets (i.e., initial historical datasets) covering multiple dimensions such as equipment status, load changes, fault records, and meteorological conditions are extracted from the historical database of the target power grid dispatching system. The datasets should cover as wide a time span as possible to fully reflect the operating characteristics of the target power grid under different operating conditions. Real-time monitoring tools such as SCADA systems (Supervisory Control and Data Acquisition Systems) are used to acquire real-time operating data of the target power grid (i.e., current operating data), including real-time measurement data and equipment status monitoring data, ensuring the timeliness and accuracy of the data. Multi-source heterogeneous data are integrated, and data fusion technology is used to eliminate inconsistencies between data, ensuring data integrity and consistency. The collected data is labeled and classified according to characteristics such as data type, timestamp, and geographical information to facilitate subsequent processing and analysis, providing a clear and rich dataset for training the dispatching model.

[0074] Step S23: Cleaning and preprocessing the data;

[0075] Professional data cleaning tools and algorithms are used to comprehensively clean the collected multi-source heterogeneous data of the target power grid. This includes checking data consistency through pivot tables, identifying and processing duplicate data, correcting erroneous data, and filling in missing values; analyzing data distribution using box plots to remove outliers that significantly deviate from the normal range, ensuring data accuracy and completeness. The cleaned data is then standardized using Z-score standardization or Min-Max standardization methods to transform data of different dimensions and ranges to a uniform scale, making them comparable and consistent. According to the input requirements of the scheduling model, the data is converted into a format recognizable by the scheduling model, such as converting time-series data into input-output pairs required for scheduling model training. Key features are extracted from the cleaned and standardized multi-source heterogeneous data using feature selection and feature extraction methods. Dimensionality reduction techniques such as Principal Component Analysis (PCA) are used to reduce the dimensionality of the multi-source heterogeneous data, highlighting key information and providing refined and efficient data support for scheduling model training and decision analysis.

[0076] Step S24, Data storage and management.

[0077] Based on the characteristics of the target power grid data and research needs, and considering factors such as data type, scale, access frequency, and storage cost, relational databases (such as MySQL and PostgreSQL) are selected to store structured data, such as power grid equipment parameters and historical operating data; non-relational databases (such as MongoDB and Redis) are selected to store semi-structured or unstructured data, such as log files and image data. A data storage architecture is constructed, adopting a hierarchical storage strategy. Hot data is stored on high-performance storage devices to meet the needs of rapid access; cold data is stored on low-cost, high-capacity storage devices to reduce storage costs. Simultaneously, a data indexing mechanism is designed to improve data query efficiency. A data management strategy is formulated, establishing a comprehensive data backup and recovery mechanism, employing multi-copy backup and off-site disaster recovery backup to ensure rapid data recovery in the event of accidental damage or loss. A strict data access control strategy is formulated, adopting a role-based access control (RBAC) mechanism to allocate data access permissions according to user roles and permissions, ensuring the legitimate use of data. A data quality monitoring system is established to regularly check and evaluate the integrity, accuracy, and consistency of data, promptly identifying and resolving data quality issues. Develop a data management platform, build a unified data management interface, and realize centralized management, unified scheduling, and efficient utilization of data; provide rich data query, retrieval, and analysis functions, and support users to analyze and mine data through SQL (Structured Query Language) queries, visualization reports, and other methods, providing strong data support for research and application.

[0078] Step S3: Design of low-rank adaptation instruction method.

[0079] To address the specific characteristics of the target power grid's scheduling tasks, an instruction method is designed that efficiently integrates domain knowledge with the scheduling model. By deeply studying the principles of low-rank adaptation technology and analyzing the language structure and semantic characteristics of the target power grid's scheduling instructions, suitable instruction templates are developed. This significantly reduces the number of training parameters (i.e., the first target parameters) while maintaining the performance of the scheduling model, improving the training efficiency and adaptability of the scheduling model. This enables the scheduling model to more accurately understand the target power grid's scheduling needs, providing technical support for subsequent scheduling decisions.

[0080] Step S31: Analyze the characteristics of power grid dispatch instructions;

[0081] The collected instruction samples (i.e., historical dispatch instructions) are decomposed to identify key components, including the operation object (e.g., generators, transformers), operation type (e.g., start-up, shutdown, adjustment), operation conditions (e.g., time, status), and expected result (e.g., power adjustment, voltage stabilization). This decomposition clarifies the syntactic structure and logical relationships of the instructions. Natural language processing techniques are then used to extract the semantic features of the instruction components, understanding the specific meaning and role of each component in the target power grid dispatch. For example, identifying the operation object (generator), operation type (increasing active power), and expected result (improving grid power supply capacity) in the instruction "increase generator active power".

[0082] Step S32, design the instruction template;

[0083] Based on the characteristics of historical power grid dispatch instructions obtained from the analysis, the structure of the initial instruction template is defined, including fixed parts (such as operation verbs and object types) and variable parts (such as specific values ​​and time). The parameters in the initial instruction template are determined; these parameters can be adjusted according to specific circumstances to generate instructions for specific tasks. For example, parameters may include time, equipment name, parameter type (such as power and voltage), and target value. Professional knowledge and experience in the target power grid dispatching field are integrated into the design of the initial instruction template (i.e., the initial instruction template is modified based on the historical dispatching experience of the target power grid), resulting in a modified instruction template. This ensures that the generated target dispatch instructions conform to the physical laws and operational procedures of the target power grid. For example, it ensures that the operation sequence in the target dispatch instructions conforms to the target power grid's safe operating procedures, and that the parameter range is within the allowable range of the equipment.

[0084] Step S33: Verify the validity of the instruction template.

[0085] The optimized instruction template was applied to a simulated target power grid dispatching scenario to test its applicability and accuracy under different conditions, ensuring that the optimized instruction template can effectively guide the target dispatching model to generate correct target dispatching instructions.

[0086] Step S4: Construct a large-scale power grid scheduling model (i.e., a target scheduling model).

[0087] Target power grid dispatch involves massive amounts of data and complex decision-making, which traditional methods struggle to handle efficiently. However, large-scale power grid dispatch models can integrate historical and real-time data from the target power grid, deeply learn its operational patterns, and accurately predict future states. Combined with low-rank adaptation instructions, these models can quickly generate effective target dispatch instructions for different dispatch scenarios, improving dispatch efficiency and accuracy and contributing to the stable operation of the target power grid.

[0088] Step S41: Select a suitable base model (i.e., the initial scheduling model).

[0089] This study investigates the architecture and performance of mainstream large language models, analyzing their performance and application cases in natural language processing tasks. Based on the characteristics of the target power grid scheduling task, it evaluates the model's ability to understand technical terms, perform logical reasoning, and process historical data, ensuring the model can adapt to the complex needs of the target power grid domain. Simultaneously, it considers the model's scalability and flexibility to facilitate subsequent parameter updates and structural expansions based on changes in the target power grid's scheduling requirements. Finally, a model with high compatibility and adaptability to the target power grid domain is selected as the base model (i.e., the initial scheduling model), laying a solid foundation for subsequent model fine-tuning and application.

[0090] Step S42, prepare the training dataset;

[0091] The initial historical dataset for the target power grid is constructed by selecting data directly relevant to the dispatching task from the collected target power grid data, such as historical operational data and historical dispatching instructions, and combining this with historical meteorological data for the corresponding historical time periods. The initial historical dataset undergoes preprocessing, including organizing power grid dispatching experts to annotate the data, clarifying its meaning and purpose to improve data quality, and converting the annotated data into a format suitable for training the initial dispatching model, ensuring that the data can be effectively read and processed by the initial dispatching model. Finally, the target historical dataset is divided into training and testing datasets, typically in a 7:3 or 8:2 ratio, to ensure the comprehensiveness and accuracy of the initial dispatching model training and testing, and to provide high-quality data support for the initial dispatching model training.

[0092] Step S43: Perform fine-tuning training of the model;

[0093] Based on the training instruction templates corresponding to the historical training scheduling instruction sets in the training dataset, key parameters for fine-tuning training are determined, such as the learning rate, batch size, and number of training epochs. The learning rate is generally set between 1e-5 and 1e-4, and the batch size is selected from 32 to 128 depending on the GPU memory and the initial scheduling model size. Tools such as TensorBoard are used to monitor the loss function value, accuracy, and other metrics in real time during the training process, promptly identifying and addressing overfitting or underfitting issues. Then, low-rank adaptation techniques are introduced to optimize traditional fine-tuning methods. Based on the training instruction templates corresponding to the historical training scheduling instruction sets in the training dataset, the first target parameters of the initial scheduling model are determined, such as weights, biases, and thresholds. By training only a small number of first target parameters, the number of parameters adjusted during the training process is reduced, lowering computational and storage costs. Detailed records are kept of parameter settings, training time, and changes in loss values ​​during the training process to ensure the efficiency and traceability of the initial scheduling model training, providing a detailed record for the optimization and evaluation of the initial scheduling model.

[0094] Step S44: Evaluate model performance.

[0095] Based on the scheduling requirements of the target power grid, evaluation metrics such as accuracy, recall, F1 score, and inference speed are set to comprehensively measure the performance of the trained scheduling model. The trained scheduling model is evaluated on a test dataset, and the values ​​of the above metrics are calculated to understand the generalization ability and predictive performance of the trained scheduling model. After the predictive performance and generalization ability of the trained scheduling model meet the requirements, a test scheduling model for the target power grid is obtained. The test scheduling model is compared and analyzed with the initial scheduling model and other power grid scheduling models to identify its strengths and weaknesses. Finally, an evaluation report is written, detailing the evaluation process, indicator results, problem analysis, and optimization suggestions, providing a basis for continuous improvement of the scheduling model and ensuring that the scheduling model can meet the actual application needs of the target power grid.

[0096] Step S5: Integrate knowledge from the power grid field.

[0097] This approach combines specialized knowledge from the power grid dispatching domain with a test dispatching model to enhance the model's understanding and decision-making capabilities regarding the target power grid. By constructing an initial knowledge graph and integrating professional knowledge such as power grid equipment parameters, operational constraints, and physical principles, the test dispatching model can more accurately understand the background and requirements of the target power grid dispatching tasks. A knowledge injection mechanism is designed to effectively integrate this domain knowledge into the test dispatching model, ensuring that the model fully absorbs and applies this knowledge. Knowledge alignment is achieved by combining the initial knowledge graph containing domain knowledge with the internal representation of the test dispatching model, enhancing the accuracy and interpretability of the test dispatching model in target power grid dispatching tasks. The knowledge fusion effect is verified to ensure that the resulting target dispatching model can provide dispatching suggestions that better conform to the operating rules of the target power grid in practical application scenarios, thereby improving the professionalism and reliability of the target dispatching model.

[0098] Step S51: Construct the initial knowledge graph;

[0099] Constructing an initial knowledge graph is crucial for structuring and integrating professional knowledge in the power grid domain into the test scheduling model. First, a comprehensive collection of professional knowledge regarding the target power grid, including equipment parameters, operational constraints, physical principles, and scheduling procedures, as well as historical scheduling cases and expert experience, is gathered. This information serves as the nodes and relationships of the initial knowledge graph. Next, the entities (such as generators, transformers, and lines) and relationships (such as "connection," "impact," and "belong to") of the initial knowledge graph are defined, constructing its framework. Then, the collected knowledge is populated into the initial knowledge graph, establishing connections between entities and forming a structured knowledge network of the target power grid. Finally, the initial knowledge graph is regularly updated and maintained to ensure it reflects the latest developments and changes in the target power grid, providing up-to-date domain knowledge support for the target scheduling model.

[0100] Step S52: Design a knowledge injection mechanism;

[0101] The knowledge injection mechanism is designed to establish an effective connection between the initial knowledge graph and the test scheduling model. First, a dedicated interface is designed in the test scheduling model to receive and process knowledge from the initial knowledge graph. Then, a mapping algorithm is developed to map entities and relationships in the initial knowledge graph to the parameters or activation functions of the test scheduling model, enabling the model to understand and utilize this knowledge. Next, a specific knowledge injection process is defined, including knowledge extraction, transformation, and loading steps, ensuring that the knowledge from the initial knowledge graph can be smoothly integrated into the test scheduling model. Finally, parameters in the injection process, such as the learning rate and knowledge weights, are adjusted to optimize the knowledge injection effect, allowing the test scheduling model to better absorb and apply domain knowledge.

[0102] Step S53: Achieve knowledge alignment;

[0103] Knowledge alignment is a crucial step in ensuring that the knowledge in the initial knowledge graph matches the internal representation of the test scheduling model. First, the entities and relations in the initial knowledge graph are aligned with the internal representation of the test scheduling model, ensuring that the test scheduling model can correctly identify and process this knowledge, resulting in the target knowledge graph. Next, based on the structure of the target knowledge graph, the parameters (i.e., the second target parameters) or architecture of the test scheduling model are adjusted to obtain the adjusted scheduling model, making its representation match the target knowledge graph. Then, the adjusted scheduling model is trained using the aligned data (i.e., the target knowledge graph), enhancing its ability to understand and apply the target knowledge graph, resulting in the target scheduling model. Finally, experiments are conducted to verify the effectiveness of knowledge alignment, ensuring that the target scheduling model can accurately combine the knowledge in the target knowledge graph with its internal representation, thereby improving the scheduling accuracy and interpretability of the target scheduling model.

[0104] Step S54: Verify the knowledge fusion effect.

[0105] Verifying the effectiveness of knowledge fusion is a crucial step in evaluating the effects of knowledge injection and alignment. First, a verification experiment is designed, outlining the experimental plan, including experimental data, evaluation metrics, and procedures, to verify the performance of the target scheduling model after knowledge fusion. Next, the target scheduling model with fused knowledge is compared with a test scheduling model without fused knowledge to assess the performance improvement brought about by knowledge fusion, focusing on the model's accuracy and interpretability in the target power grid scheduling task. Then, power grid scheduling experts are invited to evaluate the output of the target scheduling model, and their feedback is collected to understand its performance in practical applications. Finally, based on the verification results and expert feedback, problems in the knowledge fusion process are analyzed, and the target scheduling model and knowledge injection mechanism are optimized and adjusted to ensure that knowledge fusion effectively improves the performance and practicality of the test scheduling model.

[0106] Step S6: Construction and optimization of the scheduling system.

[0107] By developing functional modules such as a real-time status assessment module, a risk warning module, and a decision suggestion generation module into the target scheduling model, and constructing a scheduling system for the target power grid based on this model, the system can achieve real-time perception and analysis of the target power grid's operating status. This helps dispatchers quickly identify potential risks and provide scientific scheduling suggestions. Simultaneously, a user-friendly interface is designed for the scheduling system, enabling dispatchers to easily input commands, query information, and obtain scheduling decision support, improving work efficiency and the accuracy of determined target scheduling commands. The construction of the scheduling system combines the intelligent analysis capabilities of the target scheduling model with the professional knowledge of dispatchers, forming a human-machine collaborative working mode, enhancing the intelligence level and emergency response capabilities of the target power grid scheduling. Furthermore, this scheduling system will provide strong support for the safe, stable, and efficient operation of the target power grid, realizing the intelligent upgrade of target power grid scheduling.

[0108] Step S61: Develop the core functional modules;

[0109] Developing core functional modules is crucial for achieving target power grid dispatch decisions. First, a real-time status assessment module is developed. This module reads real-time data from the target power grid (i.e., current operating data), performs data preprocessing and feature extraction, and utilizes the real-time status assessment module of the target dispatch model for real-time analysis. It outputs the current operating status of the target power grid, including key indicators such as equipment status, load conditions, and power balance, helping dispatchers understand the grid's operational status in a timely manner. Next, a risk warning module is developed. Based on the target power grid's current operating data and status, and combined with risk assessment algorithms, it outputs risk identification results for the target power grid. This allows for rapid identification and warning of potential risks in the target power grid's operation, such as equipment overload, voltage exceeding limits, and frequency anomalies, promptly reminding dispatchers to take appropriate measures. Then, a decision suggestion generation module is developed. Based on the target power grid's current operating status and risk identification results, it selects an appropriate target instruction template, calls the large-scale power grid dispatch model, and performs decision reasoning based on current operating data and meteorological data to generate reasonable dispatch decision suggestions (i.e., target dispatch instructions), such as adjusting generation plans, optimizing load allocation, and switching equipment status, providing dispatchers with scientific decision support. Finally, a historical data analysis module was developed to deeply mine and analyze the historical operation data of the target power grid, extract key features and patterns, provide rich historical experience support for the target scheduling model, and help the target scheduling model better understand the operation trend of the target power grid.

[0110] Step S62, design the user interface;

[0111] The user interface design for the dispatching system aims to enhance user experience and ensure dispatchers can efficiently formulate target dispatch instructions for the target power grid. Requirements analysis and user surveys were conducted to understand dispatchers' operating habits and information needs, determining the interface's functions and layout. The interface layout was designed, rationally arranging the positions of each functional module, such as placing the real-time data display area in a prominent position for easy access to key information by dispatchers. Simultaneously, a simple and intuitive operation flow was designed to ensure users can complete complex operations with minimal clicks. An interface prototype was created and tested, with dispatchers invited to participate in trials, feedback collected, and the interface optimized to ensure usability and practicality. For example, based on feedback, shortcut buttons were added, and the data display method was optimized to improve work efficiency.

[0112] Step S63: Perform functional integration testing;

[0113] Functional integration testing is crucial for ensuring the stable operation and functional accuracy of the dispatching system. A detailed test plan should be developed, covering functional and performance testing to ensure comprehensiveness and effectiveness. A test environment similar to the actual target power grid dispatching environment should be constructed, including hardware equipment, software systems, and network configurations, to ensure the representativeness of the test results. Integration testing should be performed on each functional module of the dispatching system, verifying functionality according to test cases, and recording test results and identified issues. For example, if the response time of the dispatching decision suggestion generation module is found to be too long during testing, the problem should be resolved by optimizing the algorithm and adjusting dispatching system parameters. Simultaneously, performance testing of the dispatching system should be conducted to evaluate indicators such as response time and throughput, ensuring stable operation under high loads. Based on the test results, the performance of the dispatching system should be continuously optimized to improve its overall efficiency.

[0114] Step S64: Collect user feedback.

[0115] Collecting user feedback is crucial for optimizing the dispatching system. System training should be provided to power grid dispatchers to familiarize them with the operation and usage of the dispatching system. The system should then be tested in practical work, providing a feedback channel for users. Diverse feedback channels should be established, such as online feedback forms, email, and instant messaging tools, to facilitate users submitting questions and suggestions at any time. User feedback should be collected and organized regularly, and issues should be categorized, statistically analyzed, and identified to pinpoint common problems in the dispatching system and key user concerns. Based on user feedback, the dispatching system should be optimized and improved, such as adjusting the interface layout and optimizing functional modules, to enhance its usability and user satisfaction. Simultaneously, the optimized dispatching system should be tested again by users, forming a closed loop of continuous iterative optimization to ensure that the dispatching system meets the actual needs of users.

[0116] The low-rank adaptation instruction-based power grid dispatching operation auxiliary decision-making method possesses efficient model adaptation and optimization capabilities. Traditional large-model fine-tuning methods, such as full-parameter fine-tuning, typically require significant computational resources and time, especially in the field of power grid dispatching, where they are inefficient and costly given the massive scale of model parameters and complex power grid data. The low-rank adaptation instruction method, by introducing low-rank constraints, decomposes the model parameter update matrix into a low-rank component. This requires training only a small number of low-rank decomposed matrix parameters (i.e., the first target parameters), rather than all parameters, significantly reducing computational load and memory consumption during training and improving the efficiency of target dispatching model adaptation. Simultaneously, this method, while maintaining the original model performance, enables the target dispatching model to adapt more quickly to the specific tasks and scenarios of the target power grid's dispatching decisions, better meeting the real-time and rapid response requirements of power grid dispatching and providing more timely and effective decision support for target power grid dispatching.

[0117] Meanwhile, this method enables the effective integration of knowledge from the power grid dispatching domain with the target dispatching model. Traditional artificial intelligence methods applied to power grid dispatching often suffer from insufficient understanding of the specialized knowledge in this domain, potentially leading to inaccuracies and unreliability in the target dispatching model's output. The low-rank adaptation instruction method, through specific design, integrates various professional knowledge, operational rules, and experiences from the power grid dispatching domain into the fine-tuning training process of the target dispatching model. For example, using low-rank adaptation technology, key information such as the physical characteristics, equipment parameters, and operational constraints of the target power grid is transformed into instructions or prompts that the model can learn and understand, guiding the model's learning process and enabling it to more accurately understand and grasp the essence and characteristics of power grid dispatching problems. In this way, when making dispatching decisions, the target dispatching model can fully consider the specialized characteristics and actual operational requirements of the target power grid, generating more realistic, professional, and operable dispatching instructions, effectively improving the quality and effectiveness of the target power grid's dispatching decisions.

[0118] The above optional implementation methods achieve at least the following effects: guided by the training instruction template, not only can the number of model parameters to be adjusted be reduced, but the model can also learn the decision logic behind historical scheduling instructions more clearly, avoiding blind training, improving the model's learning efficiency and decision quality, and improving the accuracy of the target scheduling instruction determination results; the injection of the initial knowledge graph enables the target scheduling model to have a certain "power grid common sense", and can make more reasonable scheduling decisions based on operating rules and equipment characteristics, avoiding the generation of target scheduling instructions that violate basic physical or operational principles.

[0119] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0120] This embodiment also provides a power grid dispatch instruction determination device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0121] According to an embodiment of this application, an apparatus embodiment for implementing a method for determining dispatch instructions for a power grid is also provided. Figure 3 This is a schematic diagram of a power grid dispatch instruction determination device according to an embodiment of this application, as shown below. Figure 3As shown, the above-mentioned power grid dispatch instruction determination device includes a data acquisition module 302, a first determination module 304, and a second determination module 306. The device will be described below.

[0122] The data acquisition module 302 is used to collect the current operating data and current meteorological data of the target power grid;

[0123] The first determining module 304, connected to the data acquisition module 302, is used to determine the target instruction template of the target power grid based on the current operating data.

[0124] The second determining module 306, connected to the first determining module 304, is used to determine the target scheduling instruction of the target power grid based on the current operating data, current meteorological data, and target instruction template, using the target scheduling model.

[0125] In the power grid dispatch instruction determination device provided in this application embodiment, by setting up a data acquisition module 302, a first determination module 304, and a second determination module 306, the device aims to obtain the target dispatch instruction of the target power grid by collecting the current operating data and current meteorological data of the target power grid and adopting the target dispatch model. This achieves the technical effect of improving the accuracy of the target dispatch instruction determination result of the target power grid, thereby solving the technical problem of inaccurate power grid dispatch instruction determination results in related technologies.

[0126] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0127] It should be noted that the data acquisition module 302, the first determining module 304, and the second determining module 306 mentioned above correspond to steps S102 to S106 in the embodiments. The instances and application scenarios implemented by the above modules and their corresponding steps are the same, but they are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of the device, can run on a computer terminal.

[0128] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0129] The aforementioned power grid dispatch instruction determination device may further include a processor and a memory. The data acquisition module 302, the first determination module 304, the second determination module 306, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0130] The processor contains a core that retrieves the corresponding program unit from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0131] This application provides a non-volatile storage medium storing a program that, when executed by a processor, implements a method for determining power grid dispatch instructions.

[0132] This application provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: collecting current operating data and current meteorological data of a target power grid; determining a target instruction template for the target power grid based on the current operating data; and determining a target scheduling instruction for the target power grid using a target scheduling model based on the current operating data, current meteorological data, and the target instruction template. The device in this document may be a server, PC, etc.

[0133] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program with the following method steps: collecting current operating data and current meteorological data of the target power grid; determining the target instruction template of the target power grid based on the current operating data; and determining the target scheduling instruction of the target power grid by adopting a target scheduling model based on the current operating data, current meteorological data, and target instruction template.

[0134] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0139] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0140] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using 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 technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0141] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0142] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for determining dispatch instructions for a power grid, characterized in that, include: Collect current operating data and current meteorological data of the target power grid; Based on the current operating data, the target instruction template for the target power grid is determined; Based on the current operating data, the current meteorological data, and the target instruction template, the target scheduling model is used to determine the target scheduling instruction for the target power grid.

2. The method according to claim 1, characterized in that, Before determining the target dispatch instruction for the target power grid using the target dispatch model based on the current operating data, the current meteorological data, and the target instruction template, the method further includes: Obtain the initial historical dataset of the target power grid, wherein the initial historical dataset includes the historical operating data, historical dispatch instructions and historical meteorological data of the target power grid during a predetermined historical period; The initial historical dataset is preprocessed to obtain the target historical dataset of the target power grid; The target historical dataset is divided into a training dataset and a test dataset; Using the training dataset, the parameter values ​​of the first objective parameter of the initial scheduling model are adjusted to obtain the trained scheduling model; The trained scheduling model is tested using the test dataset to obtain a test scheduling model. Knowledge injection is performed on the test scheduling model to obtain the target scheduling model.

3. The method according to claim 2, characterized in that, The step of adjusting the parameter values ​​of the first objective parameter of the initial scheduling model using the training dataset to obtain the trained scheduling model includes: Determine the training instruction template corresponding to the historical training scheduling instruction set in the training dataset, wherein the training instruction template is used to guide the initial scheduling model to generate scheduling instructions that conform to the target power grid operation logic by learning the historical training scheduling instruction set; Based on the training instruction template, the first target parameter is determined; The training scheduling model is obtained by adjusting the parameter values ​​of the first target parameter using the historical training run dataset and historical training meteorological dataset in the training dataset.

4. The method according to claim 3, characterized in that, Determining the training instruction template corresponding to the historical training scheduling instruction set in the training dataset includes: For any historical training scheduling instruction in the historical training scheduling instruction set, the historical training scheduling instruction is decomposed to obtain multiple components of the historical training scheduling instruction. Based on the aforementioned components, natural language processing techniques are used to determine the semantic features corresponding to each of the aforementioned components. Based on the semantic features corresponding to the multiple components, determine any initial instruction template for any historical training scheduling instruction; Based on the historical dispatching experience of the target power grid, any initial instruction template is modified to obtain any modified instruction template of any historical training dispatching instruction. The correction instruction templates corresponding to multiple historical training scheduling instructions in the historical training scheduling instruction set are determined by determining any one of the correction instruction templates. The correction instruction templates corresponding to the multiple historical training scheduling instructions are merged to obtain the training instruction template.

5. The method according to claim 2, characterized in that, The step of injecting knowledge into the test scheduling model to obtain the target scheduling model includes: Construct an initial knowledge graph of the target power grid; The initial knowledge graph is injected into the test scheduling model to obtain the target scheduling model.

6. The method according to claim 5, characterized in that, The step of injecting the initial knowledge graph into the test scheduling model to obtain the target scheduling model includes: The initial knowledge graph is aligned with the test scheduling model to obtain the target knowledge graph. Based on the target knowledge graph, the structure of the test scheduling model and the parameter values ​​of the second target parameter are adjusted to obtain the adjusted scheduling model; The target knowledge graph is used to train the adjustment scheduling model to obtain the target scheduling model.

7. The method according to any one of claims 1 to 6, characterized in that, The step of determining the target instruction template for the target power grid based on the current operating data includes: Based on the current operating data, the current operating status of the target power grid is determined; Based on the current operating status, the risk identification result of the target power grid is determined; Based on the risk identification results, the target instruction template is determined.

8. A device for determining dispatch instructions for a power grid, characterized in that, include: The data acquisition module is used to collect the current operating data and current meteorological data of the target power grid; The first determining module is used to determine the target instruction template of the target power grid based on the current operating data; The second determining module is used to determine the target scheduling instruction of the target power grid based on the current operating data, the current meteorological data, and the target instruction template, using a target scheduling model.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions, which are adapted to be loaded by a processor and executed by the power grid scheduling instruction determination method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the power grid dispatch instruction determination method according to any one of claims 1 to 7.

Images