A semantic large model optimization method and system for power scenarios

Abstract

This application provides a method and system for optimizing a semantic large-scale model for power scenarios, relating to the field of data processing technology. The method includes: collecting characteristic data of power scenarios and classifying and labeling them by attributes; then, standardizing and preprocessing this data; designing a semantic large-scale model functional architecture; performing semantic mapping matching; identifying semantic association rules in the power scenario; and performing association clustering to obtain a functional semantic label dataset. A semantically optimized large-scale model is trained and optimized to obtain a semantically optimized large-scale model; and these models are then semantically topologically connected to finally generate a semantically optimized large-scale model for power scenarios. This application solves the technical problem that existing technologies struggle to identify terminology in power scenarios, leading to performance degradation when processing complex data, further affecting low recognition accuracy and processing efficiency. By optimizing the semantic large-scale model for power scenarios, accurate semantic recognition and efficient processing of power scenarios are achieved, improving the intelligence level of power systems.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a semantic large model optimization method and system for power scenarios. Background Art

[0002] With the advancement of energy transformation and smart grid construction, the power industry has an increasingly urgent need for intelligent technology. As an important intelligent tool, semantic big models can achieve in-depth understanding and intelligent processing of text, voice and other data in power scenarios, providing strong support for the intelligent upgrade of the power industry. In recent years, with the rapid development of deep learning technology, semantic big models have made significant progress in the field of natural language processing. These models have powerful text understanding, generation and reasoning capabilities, providing a technical basis for the optimization of semantic big models for power scenarios. Power scenarios involve a large number of professional terms, equipment names, operating instructions, etc., which put forward higher requirements on the accuracy and reliability of semantic big models. At the same time, the data in power scenarios often have characteristics such as temporal and spatial characteristics, and semantic big models are required to effectively process these complex data.

[0003] At present, although the existing semantic large models have achieved certain results in the field of natural language processing, they still have some limitations when facing power scenarios.

[0004] In summary, existing technologies have difficulty in identifying terms in power scenarios, resulting in performance degradation when processing complex data, further affecting technical issues such as low recognition accuracy and processing efficiency. Summary of the invention

[0005] The purpose of this application is to provide a semantic large model optimization method and system for power scenarios, so as to solve the technical problem that the existing technology is difficult to identify the terms of power scenarios, resulting in performance degradation when processing complex data, further affecting the low recognition accuracy and processing efficiency.

[0006] In view of the above problems, the present application provides a semantic large model optimization method and system for power scenarios.

[0007] In the first aspect, the present application provides a semantic big model optimization method for power scenarios, and the method is implemented by a semantic big model optimization system for power scenarios, wherein the method includes: S1: collecting and acquiring a power scenario characteristic data set, constructing a power scenario attribute classifier, and classifying and labeling the power scenario characteristic data set based on the power scenario attribute classifier to obtain a power scenario multidimensional attribute label data set; S2: performing standardized preprocessing on the power scenario multidimensional attribute label data set to obtain a standard power scenario multidimensional attribute label data set, and designing a power semantic big model functional architecture according to the power scenario application requirements; S3: performing semantic mapping and matching based on the power semantic big model functional architecture and the standard power scenario multidimensional attribute label data set to obtain power scenario semantic association rules, and performing association clustering on the standard power scenario multidimensional attribute label data set based on the power scenario semantic association rules to obtain a power scenario functional semantic label data set; S4: using the power semantic big model functional architecture to train and optimize the power scenario functional semantic label data set to obtain a power scenario functional semantic model set, and performing semantic topological connection on the power scenario functional semantic model set to generate a power scenario semantic optimization big model.

[0008] In the second aspect, the present application also provides a semantic big model optimization system for power scenarios, which is used to execute a semantic big model optimization method for power scenarios as described in the first aspect, wherein the system includes: a data set acquisition module, the data set acquisition module is used to collect and acquire power scenario characteristic data sets, construct a power scenario attribute classifier, and classify and label the power scenario characteristic data sets based on the power scenario attribute classifier to obtain a power scenario multidimensional attribute label data set; a functional architecture acquisition module, the functional architecture acquisition module is used to perform standardized preprocessing on the power scenario multidimensional attribute label data set to obtain a standard power scenario multidimensional attribute label data set, and design a power scenario according to the power scenario application requirements. A power semantic big model functional architecture; an association clustering module, which is used to perform semantic mapping and matching based on the power semantic big model functional architecture and the standard power scenario multidimensional attribute label data set, obtain power scenario semantic association rules, and perform association clustering on the standard power scenario multidimensional attribute label data set based on the power scenario semantic association rules to obtain a power scenario functional semantic label data set; a model acquisition module, which is used to use the power semantic big model functional architecture to train and optimize the power scenario functional semantic label data set to obtain a power scenario functional semantic model set, perform semantic topological connection on the power scenario functional semantic model set, and generate a power scenario semantic optimization big model.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] Through S1: collecting and acquiring power scenario characteristic data sets, constructing power scenario attribute classifiers, and classifying and labeling the power scenario characteristic data sets based on the power scenario attribute classifiers to obtain power scenario multidimensional attribute label data sets; S2: performing standardized preprocessing on the power scenario multidimensional attribute label data sets to obtain standard power scenario multidimensional attribute label data sets, and designing the power semantic big model functional architecture according to the power scenario application requirements; S3: performing semantic mapping and matching based on the power semantic big model functional architecture and the standard power scenario multidimensional attribute label data sets to obtain power scenario semantic association rules, and based on the power scenario semantic association rules, the standard power scene The multi-dimensional attribute label data set is associated and clustered to obtain the power scene function semantic label data set; S4: the power scene function semantic label data set is trained and optimized by using the power semantic big model functional architecture to obtain the power scene function semantic model set, and the power scene function semantic model set is semantically topologically connected to generate a power scene semantic optimization big model, which effectively solves the problem that the existing technology is difficult to identify the terms of the power scene, resulting in performance degradation when processing complex data, and further affecting the technical problems of low recognition accuracy and processing efficiency. Through the optimization of the power scene semantic big model, accurate semantic recognition and efficient processing of the power scene are realized, which improves the intelligence level of the power system.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0013] Figure 1 A flowchart of a semantic large model optimization method for power scenarios for this application;

[0014] Figure 2 This application provides a structural diagram of a semantic large model optimization system for power scenarios.

[0015] Description of reference numerals:

[0016] Data set acquisition module 11, functional architecture acquisition module 12, association clustering module 13, model acquisition module 14. DETAILED DESCRIPTION

[0017] This application provides a semantic big model optimization method and system for power scenarios, which solves the technical problem that the existing technology has difficulty in identifying terminology of power scenarios, resulting in performance degradation when processing complex data, and further affecting low recognition accuracy and processing efficiency. Through the optimization of the semantic big model of power scenarios, accurate semantic recognition and efficient processing of power scenarios are achieved, thereby improving the intelligence level of the power system.

[0018] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0019] Embodiment 1

[0020] Please see attached Figure 1 The present application provides a semantic large model optimization method for power scenarios, wherein the method is applied to a semantic large model optimization system for power scenarios, and the method specifically includes the following steps:

[0021] S1: Collect and obtain a power scenario characteristic data set, construct a power scenario attribute classifier, and classify and label the power scenario characteristic data set based on the power scenario attribute classifier to obtain a power scenario multidimensional attribute label data set;

[0022] Specifically, data is collected from all aspects of the power system, such as power generation, transmission, distribution, and power consumption, including equipment operating status, operation records, fault reports, maintenance logs, etc. These data can be structured data, such as tables, database records, or unstructured data, such as text descriptions, images, and videos. The collected data is preprocessed by cleaning, deduplication, and standardization to ensure data quality and consistency. According to the characteristics of the power scenario, key features are extracted, such as equipment type, operation type, fault type, etc. According to the nature of the problem and the characteristics of the data, a suitable machine learning algorithm or deep learning model is selected to build an attribute classifier, such as support vector machine, random forest, convolutional neural network, etc. Use a labeled data set to train the attribute classifier. The preprocessed power scenario characteristic data set is input into the trained attribute classifier. The attribute classifier classifies and labels each sample in the data set according to the learned rules and patterns, such as labeling a fault report as a transformer fault or a line fault. In addition to the basic classification labels, more dimensional labels can be added to the data set according to specific needs, such as timestamp, geographic location, severity, etc.

[0023] S2: Performing standardized preprocessing on the electric power scenario multidimensional attribute label data set to obtain a standard electric power scenario multidimensional attribute label data set, and designing a functional architecture of an electric power semantic large model according to the electric power scenario application requirements;

[0024] Specifically, remove duplicate, invalid or erroneous data records to ensure data accuracy and completeness. Convert data from different sources or formats into a unified format, such as CSV, JSON, etc., to facilitate subsequent processing and analysis. For attribute labels in text form, such as equipment type, fault type, etc., they can be converted into numerical data by using methods such as unique hot encoding and label encoding to facilitate model processing. Ensure that all numerical data use unified units, such as physical quantities such as voltage and current should be unified into standard units in the International System of Units. By normalizing the data to scale to the [0,1] interval or standardizing the data to a distribution with a mean of 0 and a standard deviation of 1, the effects of dimensions and magnitudes between different attributes can be eliminated to improve the training efficiency and performance of the model. After standardized preprocessing, a multidimensional attribute label dataset of standard power scenarios is obtained. The dataset has high quality, unified format and units, and provides a good foundation for subsequent model training and optimization. In-depth understanding of the specific application requirements of power scenarios, such as equipment fault prediction, operation instruction understanding, power system status assessment, etc., to clarify the functions and performance requirements that the model needs to support. According to the results of demand analysis, the power semantic model is divided into multiple functional modules, such as text input processing module, semantic understanding module, task execution module, etc. Each module is responsible for processing specific tasks or functions. For each functional module, select appropriate technologies and algorithms to implement its functions. For example, the Transformer model can be used as the basic model of the semantic understanding module, and fine-tuned and optimized in combination with professional knowledge in the power field. Define the input and output interfaces and data interaction methods between each functional module to ensure the collaborative work and data consistency between modules.

[0025] S3: Perform semantic mapping and matching based on the functional architecture of the power semantic big model and the standard power scenario multidimensional attribute label data set to obtain power scenario semantic association rules, and perform association clustering on the standard power scenario multidimensional attribute label data set based on the power scenario semantic association rules to obtain a power scenario functional semantic label data set;

[0026] Specifically, a pre-trained power semantics model, such as a Transformer-based model, is used to convert each label in the standard power scenario multidimensional attribute label dataset into a high-dimensional semantic vector. The similarity between different label semantic vectors, such as cosine similarity and Euclidean distance, is calculated to quantify the degree of semantic association between labels. A similarity threshold is set. When the similarity between two or more labels exceeds the threshold, it is considered that there is a semantic association between them, thereby extracting the semantic association rules of the power scenario. These rules can be expressed in the form that if label A appears, label B is also likely to appear. The power scenario semantic association rule set contains the semantic association information between different attribute labels in the power scenario. According to the characteristics and requirements of the power scenario, a suitable clustering algorithm is selected, such as density-based DBSCAN and hierarchical-based AGNES. These algorithms can use the similarity information in the semantic association rules for clustering. The standard power scenario multidimensional attribute label dataset is used as input, and the selected clustering algorithm is executed in combination with the power scenario semantic association rules. During the clustering process, the algorithm divides the labels into different clusters according to the degree of semantic association between them. Generate a functional semantic label for each cluster, which can summarize the common characteristics or functions of all labels in the cluster. For example, a cluster may contain all labels related to transformer fault prediction, then the functional semantic label of the cluster can be related to transformer fault prediction. The power scenario functional semantic label dataset, in which each functional semantic label represents a specific power scenario function or task, provides more clear and targeted guidance for the subsequent optimization and application of semantic large models.

[0027] S4: Use the power semantic big model functional architecture to train and optimize the power scenario functional semantic label data set to obtain a set of power scenario functional semantic models, perform semantic topological connection on the set of power scenario functional semantic models, and generate a power scenario semantic optimization big model.

[0028] Specifically, the power scene functional semantic label dataset is divided into a training set, a validation set, and a test set for model training, validation, and testing. Based on the functional architecture of the power semantic big model, the model parameters and structure are initialized. These parameters and structures will serve as the basis for model training. The model is trained using the training set, and the model parameters are adjusted through the back propagation algorithm and optimizers such as gradient descent to minimize the prediction error or loss function. During the training process, the validation set can be used for model selection and hyperparameter adjustment. The trained model is evaluated using the test set to calculate the performance of the model on various performance indicators, such as accuracy, recall, F1 score, etc. Ensure that the model has good generalization ability on unseen data. According to the model evaluation results, the model is iteratively optimized. This can include adjusting the model structure, adding data enhancement technology, using more advanced optimization algorithms, etc. Through multiple iterations, the performance of the model is continuously improved. A set of trained and optimized power scene functional semantic models, which have good performance on different functions or tasks of the power scene. According to the correlation and dependency between the power scene functional semantic models, a suitable semantic topological connection strategy is determined. This can include connections based on functional hierarchy, connections based on task relevance, etc. According to the determined connection strategy, the various models in the set of power scenario functional semantic models are integrated. This can be achieved through model fusion techniques such as weighted average, voting mechanism, etc. or multi-model integration methods in deep learning such as knowledge distillation, model compression, etc. In the model integration process, the semantic topology structure is adjusted and optimized according to actual needs to ensure the collaborative work and smooth information transmission between different functional models. The entire power scenario semantic optimization large model is globally optimized and verified to ensure its overall performance and stability in various power scenario functions. The power scenario semantic optimization large model integrates multiple semantic models for different power scenario functions, has powerful semantic processing capabilities and good task performance, and provides strong support for the intelligent upgrade of the power industry.

[0029] Furthermore, a power scenario attribute classifier is constructed in S1, including:

[0030] S11: Acquire a power scenario attribute knowledge base, extract knowledge attributes from the power scenario attribute knowledge base, and obtain a power scenario entity knowledge attribute set;

[0031] S12: performing in-depth hierarchical division on each knowledge attribute in the electric power scenario entity knowledge attribute set, determining entity attribute hierarchical information, and arranging and integrating the electric power scenario entity knowledge attribute set according to the entity attribute hierarchical information to obtain an electric power scenario system knowledge attribute set;

[0032] S13: filling the content of each knowledge attribute in the power scenario system knowledge attribute set in turn based on the power scenario attribute knowledge base to obtain a power entity attribute content set;

[0033] S14: The power scenario system knowledge attribute set and the power entity attribute content set are used as classifier nodes to construct the power scenario attribute classifier.

[0034] Specifically, a knowledge base containing rich power scenario attributes is selected as the data source. Natural language processing techniques such as named entity recognition and relationship extraction are used to extract entities and their attributes related to power scenarios from the knowledge base. The extracted attributes are cleaned to remove duplicate, invalid or low-quality data and integrated into a set of power scenario entity knowledge attributes. The characteristics of each attribute and the relationship between them are deeply analyzed. According to the importance, universality, and specificity of the attributes, they are divided into different levels, such as basic attributes, core attributes, and extended attributes. According to the defined hierarchical information, the power scenario entity knowledge attribute set is orderly arranged and integrated to form a power scenario system knowledge attribute set. The power scenario attribute knowledge base is used again to find and match the corresponding content or value for each attribute. Ensure that the filled content matches the corresponding attribute and is accurate. Integrate the filled attributes and content into a power entity attribute content set. According to the characteristics and requirements of the power scenario, design a suitable attribute classifier architecture, such as a decision tree, a neural network, etc. The power scenario system knowledge attribute set and the power entity attribute content set are used as input data for the classifier. Use appropriate algorithms to train the classifier, and improve the performance of the classifier by adjusting parameters and optimizing the model structure. Use independent validation datasets to evaluate the performance of the classifier to ensure that it has good accuracy and reliability in practical applications.

[0035] Furthermore, the power scenario semantic association rules obtained in S3 include:

[0036] S31: abstracting each attribute label in the standard power scenario multidimensional attribute label data set into a semantic element to obtain a power scenario label semantic element set;

[0037] S32: Acquire architecture classification factors, where the architecture classification factors include function type and function level, and semantically describe the functional architecture of the power semantic big model based on the architecture classification factors to obtain a semantic description set of functional architecture nodes;

[0038] S33: dividing the functional architecture of the electric power semantic model into nodes based on the functional architecture node semantic description set, and obtaining a functional architecture partition node set;

[0039] S34: Perform semantic association matching on each architecture node in the functional architecture partition node set and the power scenario label semantic element set respectively to obtain the power scenario semantic association rule.

[0040] Specifically, each label in the standard power scenario multidimensional attribute label dataset is deeply analyzed to understand its meaning and context. The core meaning of each label is abstracted into one or more semantic elements. These semantic elements are the basic units that constitute the label concept. Each attribute label, such as equipment type and fault type, is abstracted into a semantic element for functional association matching. All extracted semantic elements are integrated into a set, namely, the power scenario label semantic element set. Function type and function level are clarified as the main factors for architecture classification. Model function evaluation scope, such as the level of fault evaluation for distribution equipment is smaller than that for all power equipment. Each architecture classification factor is described in detail semantically to ensure that it can accurately reflect its role and significance in the functional architecture of the power semantic big model. The semantic descriptions of all functional architecture nodes are integrated into a set, namely, the functional architecture node semantic description set. According to the functional architecture node semantic description set, the definition and scope of each node are clarified. According to the definition, the functional architecture of the power semantic big model is divided into multiple independent functional architecture partition nodes. All divided functional architecture partition nodes are integrated into a set, namely, the functional architecture partition node set. Select a suitable semantic association matching algorithm, such as similarity calculation based on vector space, rule-based matching, etc. Match each node in the functional architecture partition node set with the elements in the power scene label semantic element set. According to the matching results, extract the semantic association rules between the power scene label semantic elements and the functional architecture partition nodes. These rules reflect the deep connection between different attribute labels and functional architecture nodes in the power scene.

[0041] Furthermore, the power scenario functional semantic model set is obtained in S4, including:

[0042] S41: performing model framework analysis on each architecture node in the functional architecture of the electric power semantic big model to determine a semantic model training framework set;

[0043] S42: training the power scenario function semantic label data sets respectively based on the semantic model training framework set to obtain a power scenario function semantic model set;

[0044] S43: Performing performance evaluation on the power scenario functional semantic model set to obtain a semantic model performance evaluation parameter set;

[0045] S44: Iteratively optimize the power scenario function semantic model set based on the semantic model performance evaluation parameter set to obtain the power scenario function semantic model set.

[0046] Specifically, analyze the characteristics of each architecture node, including its data processing capability, computational complexity, scalability, etc. Establish standards for selecting model training frameworks, such as training efficiency, model performance, ease of use, etc. Based on the analysis results and selection criteria, determine one or more suitable semantic model training frameworks for each architecture node, and determine the model framework based on the function of each node, such as convolutional neural network, recurrent neural network or Transformer, etc., to form a set of semantic model training frameworks. Prepare the corresponding power scenario function semantic label dataset according to the needs of each architecture node. Initialize the model parameters and structure in the selected training framework. Train each model using the prepared dataset, and adjust the model parameters through the optimization algorithm to minimize the training error. After the training is completed, save the obtained power scenario function semantic model to form a power scenario function semantic model set. Determine the standards for evaluating model performance, such as accuracy, recall rate, F1 score, training time, etc. Prepare an independent test dataset for evaluating performance. Use the test dataset to evaluate each model in the power scenario function semantic model set and record the evaluation results. Integrate the evaluation parameters of all models into a set, namely the semantic model performance evaluation parameter set. Conduct in-depth analysis of the semantic model performance evaluation parameter set to identify the model's performance bottlenecks and room for improvement. Formulate optimization strategies based on the analysis results, such as adjusting the model structure, improving the training algorithm, and adding data enhancement. Iterate and optimize the power scenario functional semantic model set according to the formulated optimization strategy, and perform performance evaluation after each optimization to verify the optimization effect. Update the model with improved performance after optimization verification to the power scenario functional semantic model set.

[0047] Further, the training of the power scene functional semantic label dataset in S42 includes:

[0048] Performing sensitive data identification on the power scenario functional semantic label data set to obtain a power scenario sensitive data set;

[0049] Performing sensitivity rating on the power scenario sensitive data set according to a preset classification level to obtain a power scenario sensitive data level;

[0050] Determine a data desensitization coefficient according to the power scenario sensitive data level, perform data desensitization on the power scenario sensitive data set based on the data desensitization coefficient, and obtain a power scenario functional semantic label desensitized data set;

[0051] The power semantic big model functional architecture is used to train and optimize the power scenario functional semantic label desensitized data set.

[0052] Specifically, clarify which types of data are sensitive data, such as personal privacy information and key business data. Use sensitive data identification tools or algorithms to scan the power scenario functional semantic label data set to identify the sensitive data. Integrate the identified sensitive data into a separate data set, namely the power scenario sensitive data set. Preset different sensitivity levels, such as high, medium, and low, and clarify the standards for each level. Rate each piece of data in the power scenario sensitive data set according to the rating standards to determine its sensitivity level. Label each piece of sensitive data with its corresponding sensitivity level to form a data set with power scenario sensitive data level labels. Based on the sensitivity level of the power scenario sensitive data, determine the corresponding data desensitization coefficient. The desensitization coefficient indicates the degree of data desensitization. The higher the sensitivity level, the greater the data desensitization coefficient, ensuring that the security and privacy of power data are effectively protected during the model optimization process. The desensitization coefficient determines the degree and method of data desensitization. Use the desensitization algorithm and desensitization coefficient to desensitize the power scenario sensitive data set. Desensitization methods may include data replacement, data perturbation, data anonymization, etc. Verify the desensitized data to ensure that the desensitization process protects the sensitivity of the data while not affecting the usability and analytical value of the data. Integrate the desensitized data into a desensitized dataset of power scenario functional semantic labels. Use the desensitized dataset of power scenario functional semantic labels as training data to ensure that the format and quality of the data meet the requirements of model training. Initialize the model parameters and structure in the functional architecture of the power semantic large model to prepare for training. Use the desensitized dataset to train the model, and continuously adjust the model parameters through the optimization algorithm to minimize the training error and improve the generalization ability of the model. Regularly evaluate the performance of the model, such as accuracy and recall, during the training process, and make necessary optimizations and adjustments based on the evaluation results.

[0053] Further, the electric power scenario functional semantic model set is obtained in S44, including:

[0054] Constructing a semantic model optimization strategy space, wherein the semantic model optimization strategy space includes performance evaluation parameters of the historical semantic model, optimization strategy parameters, and corresponding model optimization effect data;

[0055] Performing strategy matching screening based on the semantic model performance evaluation parameter set and the semantic model optimization strategy space to obtain a semantic model optimization strategy memory library set;

[0056] A global comparison and optimization is performed within the semantic model optimization strategy memory set to determine a model optimization strategy parameter set, and based on the model optimization strategy parameter set, the power scenario function semantic model set is iteratively optimized to obtain the power scenario function semantic model set.

[0057] Specifically, collect performance evaluation data of historical semantic models, including key indicators such as accuracy, recall, and F1 score. Identify various possible optimization strategies and their parameters, such as learning rate adjustment, model structure adjustment, and data enhancement strategy. Record the application effect of each optimization strategy on the historical model, including performance improvement, training time change, etc. Integrate the above information to build a multi-dimensional semantic model optimization strategy space. Compare the performance evaluation parameters of the current model with the performance parameters of the historical model in the strategy space. Based on the comparison results, find the optimization strategy that is effective for similar performance models under historical conditions. Integrate the matched optimization strategies and their effect data into an optimization strategy memory set. Traverse all strategy combinations in the optimization strategy memory set and evaluate their potential improvement effects on the current model set. Based on the historical effect data and the current model status, predict the application effect of each strategy combination. Select the optimization strategy combination with the best prediction effect to form a model optimization strategy parameter set. According to the specific parameter settings in the model optimization strategy parameter set, adjust the training configuration, model structure, or data processing method of the power scenario functional semantic model. Retrain the model with the adjusted configuration to ensure that the model can fully learn and apply the improvements brought by the optimization strategy. Evaluate the performance of the optimized model on an independent validation dataset to ensure the effectiveness of the optimization strategy. Update the verified performance-improved model to the power scenario functional semantic model set to complete a round of optimization iteration.

[0058] Furthermore, the present application also includes:

[0059] Performing control effect index fitting analysis on the semantic model optimization strategy space and constructing a semantic model optimization effect evaluation function;

[0060] Embed the semantic model optimization effect evaluation function into the semantic model optimization strategy memory library set, and set the memory library value probability and parameter optimization step length;

[0061] Activate the semantic model optimization strategy memory library set, use the memory library value probability and parameter search step to perform global optimization in the semantic model optimization strategy memory library set until a preset number of iterations are reached, and output the model optimization strategy parameter set.

[0062] Specifically, historical optimization strategies and their corresponding effect indicator data are extracted from the semantic model optimization strategy space. Mathematical methods such as regression analysis and neural networks are used to fit the extracted data to analyze the relationship between optimization strategy parameters and effect indicators. Based on the fitting results, a semantic model optimization effect evaluation function is constructed, which can predict the model performance improvement under given optimization strategy parameters. The optimization effect evaluation function is used as part of the memory library collection to ensure that the function can be easily called during the optimization process. For each optimization strategy in the memory library, a value probability is set, which indicates the possibility of selecting the strategy during the optimization process, and a parameter optimization step size is set, which indicates the step size when adjusting the strategy parameters. A preset number of iterations is set as the termination condition of the optimization process. The optimization process is started and the optimization strategy is selected from the memory library according to the value probability. In each iteration step, the selected optimization strategy parameters are adjusted according to the parameter search step size. The model performance under the current parameter combination is predicted and evaluated using the embedded optimization effect evaluation function. The value probability and / or parameter settings in the memory library are updated according to the evaluation results, so that a better strategy parameter combination is more likely to be selected in subsequent iterations. After reaching the preset number of iterations, the optimization process is stopped. Output the model optimization strategy parameter set obtained through global optimization.

[0063] In summary, the semantic large model optimization method for power scenarios provided by this application has the following technical effects:

[0064] Through S1: collecting and acquiring power scenario characteristic data sets, constructing power scenario attribute classifiers, and classifying and labeling the power scenario characteristic data sets based on the power scenario attribute classifiers to obtain power scenario multidimensional attribute label data sets; S2: performing standardized preprocessing on the power scenario multidimensional attribute label data sets to obtain standard power scenario multidimensional attribute label data sets, and designing the power semantic big model functional architecture according to the power scenario application requirements; S3: performing semantic mapping and matching based on the power semantic big model functional architecture and the standard power scenario multidimensional attribute label data sets to obtain power scenario semantic association rules, and based on the power scenario semantic association rules, the standard power scene The multi-dimensional attribute label data set is associated and clustered to obtain the power scene function semantic label data set; S4: the power scene function semantic label data set is trained and optimized by using the power semantic big model functional architecture to obtain the power scene function semantic model set, and the power scene function semantic model set is semantically topologically connected to generate a power scene semantic optimization big model, which effectively solves the problem that the existing technology is difficult to identify the terms of the power scene, resulting in performance degradation when processing complex data, and further affecting the technical problems of low recognition accuracy and processing efficiency. Through the optimization of the power scene semantic big model, accurate semantic recognition and efficient processing of the power scene are realized, which improves the intelligence level of the power system.

[0065] Embodiment 2

[0066] Based on the semantic large model optimization method for power scenarios in the aforementioned embodiment, the present application also provides a semantic large model optimization system for power scenarios, please refer to the attached Figure 2 , the system comprising:

[0067] The data set acquisition module 11 is used to collect and acquire a power scenario characteristic data set, construct a power scenario attribute classifier, and classify and label the power scenario characteristic data set based on the power scenario attribute classifier to obtain a power scenario multidimensional attribute label data set;

[0068] Functional architecture acquisition module 12, the functional architecture acquisition module 12 is used to perform standardized preprocessing on the electric power scenario multidimensional attribute label data set to obtain a standard electric power scenario multidimensional attribute label data set, and design a functional architecture of the electric power semantic large model according to the electric power scenario application requirements;

[0069] An association clustering module 13, wherein the association clustering module 13 is used to perform semantic mapping and matching based on the functional architecture of the electric power semantic big model and the standard electric power scenario multidimensional attribute label data set to obtain electric power scenario semantic association rules, and to perform association clustering on the standard electric power scenario multidimensional attribute label data set based on the electric power scenario semantic association rules to obtain an electric power scenario functional semantic label data set;

[0070] The model acquisition module 14 is used to use the power semantic big model functional architecture to train and optimize the power scene functional semantic label data set, obtain a set of power scene functional semantic models, perform semantic topological connection on the set of power scene functional semantic models, and generate a power scene semantic optimization big model.

[0071] Furthermore, the data set acquisition module 11 in the system is also used for:

[0072] S11: Acquire a power scenario attribute knowledge base, extract knowledge attributes from the power scenario attribute knowledge base, and obtain a power scenario entity knowledge attribute set;

[0073] S12: performing in-depth hierarchical division on each knowledge attribute in the electric power scenario entity knowledge attribute set, determining entity attribute hierarchical information, and arranging and integrating the electric power scenario entity knowledge attribute set according to the entity attribute hierarchical information to obtain an electric power scenario system knowledge attribute set;

[0074] S13: filling each knowledge attribute in the power scenario system knowledge attribute set with content in turn based on the power scenario attribute knowledge base to obtain a power entity attribute content set;

[0075] S14: The power scenario system knowledge attribute set and the power entity attribute content set are used as classifier nodes to construct the power scenario attribute classifier.

[0076] Furthermore, the association clustering module 13 in the system is also used for:

[0077] S31: abstracting each attribute label in the standard power scenario multidimensional attribute label data set into a semantic element to obtain a power scenario label semantic element set;

[0078] S32: Acquire architecture classification factors, where the architecture classification factors include function type and function level, and semantically describe the functional architecture of the power semantic big model based on the architecture classification factors to obtain a semantic description set of functional architecture nodes;

[0079] S33: dividing the functional architecture of the electric power semantic model into nodes based on the functional architecture node semantic description set, and obtaining a functional architecture partition node set;

[0080] S34: Perform semantic association matching on each architecture node in the functional architecture partition node set and the power scenario label semantic element set respectively to obtain the power scenario semantic association rule.

[0081] Furthermore, the model acquisition module 14 in the system is also used for:

[0082] S41: performing model framework analysis on each architecture node in the functional architecture of the electric power semantic big model to determine a semantic model training framework set;

[0083] S42: training the power scenario function semantic label data sets respectively based on the semantic model training framework set to obtain a power scenario function semantic model set;

[0084] S43: Performing performance evaluation on the power scenario functional semantic model set to obtain a semantic model performance evaluation parameter set;

[0085] S44: Iteratively optimize the power scenario function semantic model set based on the semantic model performance evaluation parameter set to obtain the power scenario function semantic model set.

[0086] Furthermore, the system further comprises a training optimization module, wherein the training optimization module is used to:

[0087] Performing sensitive data identification on the power scenario functional semantic label data set to obtain a power scenario sensitive data set;

[0088] Performing sensitivity rating on the power scenario sensitive data set according to a preset classification level to obtain a power scenario sensitive data level;

[0089] Determine a data desensitization coefficient according to the power scenario sensitive data level, perform data desensitization on the power scenario sensitive data set based on the data desensitization coefficient, and obtain a power scenario functional semantic label desensitized data set;

[0090] The power semantic big model functional architecture is used to train and optimize the power scenario functional semantic label desensitized data set.

[0091] Furthermore, the system further comprises a semantic model set acquisition module, wherein the semantic model set acquisition module is used to:

[0092] Constructing a semantic model optimization strategy space, wherein the semantic model optimization strategy space includes performance evaluation parameters of the historical semantic model, optimization strategy parameters, and corresponding model optimization effect data;

[0093] Performing strategy matching screening based on the semantic model performance evaluation parameter set and the semantic model optimization strategy space to obtain a semantic model optimization strategy memory library set;

[0094] A global comparison and optimization is performed within the semantic model optimization strategy memory set to determine a model optimization strategy parameter set, and based on the model optimization strategy parameter set, the power scenario function semantic model set is iteratively optimized to obtain the power scenario function semantic model set.

[0095] Furthermore, the system further includes an optimization strategy parameter set acquisition module, and the optimization strategy parameter set acquisition module is used to:

[0096] Performing control effect index fitting analysis on the semantic model optimization strategy space and constructing a semantic model optimization effect evaluation function;

[0097] Embed the semantic model optimization effect evaluation function into the semantic model optimization strategy memory library set, and set the memory library value probability and parameter optimization step length;

[0098] Activate the semantic model optimization strategy memory library set, use the memory library value probability and parameter search step to perform global optimization in the semantic model optimization strategy memory library set until a preset number of iterations are reached, and output the model optimization strategy parameter set.

[0099] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1The semantic large model optimization method for power scenarios and the specific examples in the first embodiment are also applicable to the semantic large model optimization system for power scenarios in this embodiment. Through the above detailed description of the semantic large model optimization method for power scenarios, those skilled in the art can clearly know the semantic large model optimization system for power scenarios in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method description.

[0100] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0101] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A semantic large model optimization method for power scenarios, characterized in that: The method comprises: S1: Collect and obtain a power scenario characteristic data set, construct a power scenario attribute classifier, and classify and label the power scenario characteristic data set based on the power scenario attribute classifier to obtain a power scenario multidimensional attribute label data set; S2: Performing standardized preprocessing on the electric power scenario multidimensional attribute label data set to obtain a standard electric power scenario multidimensional attribute label data set, and designing a functional architecture of an electric power semantic large model according to the electric power scenario application requirements; S3: Perform semantic mapping and matching based on the functional architecture of the power semantic big model and the standard power scenario multidimensional attribute label data set to obtain power scenario semantic association rules, and perform association clustering on the standard power scenario multidimensional attribute label data set based on the power scenario semantic association rules to obtain a power scenario functional semantic label data set; S4: Use the power semantic big model functional architecture to train and optimize the power scenario functional semantic label data set to obtain a set of power scenario functional semantic models, perform semantic topological connection on the set of power scenario functional semantic models, and generate a power scenario semantic optimization big model.

2. The method according to claim 1, characterized in that The power scene attribute classifier is constructed in S1, including: S11: Acquire a power scenario attribute knowledge base, extract knowledge attributes from the power scenario attribute knowledge base, and obtain a power scenario entity knowledge attribute set; S12: performing in-depth hierarchical division on each knowledge attribute in the electric power scenario entity knowledge attribute set, determining entity attribute hierarchical information, and arranging and integrating the electric power scenario entity knowledge attribute set according to the entity attribute hierarchical information to obtain an electric power scenario system knowledge attribute set; S13: filling the content of each knowledge attribute in the power scenario system knowledge attribute set in turn based on the power scenario attribute knowledge base to obtain a power entity attribute content set; S14: The power scenario system knowledge attribute set and the power entity attribute content set are used as classifier nodes to construct the power scenario attribute classifier.

3. The method according to claim 1, characterized in that The power scenario semantic association rules obtained in S3 include: S31: abstracting each attribute label in the standard power scenario multidimensional attribute label data set into a semantic element to obtain a power scenario label semantic element set; S32: Acquire architecture classification factors, where the architecture classification factors include function type and function level, and semantically describe the functional architecture of the power semantic big model based on the architecture classification factors to obtain a semantic description set of functional architecture nodes; S33: dividing the functional architecture of the electric power semantic model into nodes based on the functional architecture node semantic description set, and obtaining a functional architecture partition node set; S34: Perform semantic association matching on each architecture node in the functional architecture partition node set and the power scenario label semantic element set respectively to obtain the power scenario semantic association rule.

4. The method according to claim 1, characterized in that The power scenario functional semantic model set obtained in S4 includes: S41: performing model framework analysis on each architecture node in the functional architecture of the electric power semantic big model to determine a semantic model training framework set; S42: training the power scenario function semantic label data sets respectively based on the semantic model training framework set to obtain a power scenario function semantic model set; S43: Performing performance evaluation on the power scenario functional semantic model set to obtain a semantic model performance evaluation parameter set; S44: Iteratively optimize the power scenario function semantic model set based on the semantic model performance evaluation parameter set to obtain the power scenario function semantic model set.

5. The method according to claim 4, characterized in that The step S42 of training the power scene functional semantic label dataset includes: Performing sensitive data identification on the power scenario functional semantic label data set to obtain a power scenario sensitive data set; Performing sensitivity rating on the power scenario sensitive data set according to a preset classification level to obtain a power scenario sensitive data level; Determine a data desensitization coefficient according to the power scenario sensitive data level, perform data desensitization on the power scenario sensitive data set based on the data desensitization coefficient, and obtain a power scenario functional semantic label desensitized data set; The power semantic big model functional architecture is used to train and optimize the power scenario functional semantic label desensitized data set.

6. The method according to claim 4, characterized in that The electric power scenario functional semantic model set is obtained in S44, including: Constructing a semantic model optimization strategy space, wherein the semantic model optimization strategy space includes performance evaluation parameters of the historical semantic model, optimization strategy parameters, and corresponding model optimization effect data; Performing strategy matching screening based on the semantic model performance evaluation parameter set and the semantic model optimization strategy space to obtain a semantic model optimization strategy memory library set; A global comparison and optimization is performed within the semantic model optimization strategy memory set to determine a model optimization strategy parameter set, and based on the model optimization strategy parameter set, the power scenario function semantic model set is iteratively optimized to obtain the power scenario function semantic model set.

7. The method according to claim 6, characterized in that The step of determining a set of model optimization strategy parameters further includes: Performing control effect index fitting analysis on the semantic model optimization strategy space and constructing a semantic model optimization effect evaluation function; Embed the semantic model optimization effect evaluation function into the semantic model optimization strategy memory library set, and set the memory library value probability and parameter optimization step length; Activate the semantic model optimization strategy memory library set, use the memory library value probability and parameter search step to perform global optimization in the semantic model optimization strategy memory library set until a preset number of iterations are reached, and output the model optimization strategy parameter set.

8. A semantic large model optimization system for power scenarios, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the system comprises: A data set acquisition module, the data set acquisition module is used to collect and acquire a power scenario characteristic data set, construct a power scenario attribute classifier, and classify and label the power scenario characteristic data set based on the power scenario attribute classifier to obtain a power scenario multidimensional attribute label data set; A functional architecture acquisition module, wherein the functional architecture acquisition module is used to perform standardized preprocessing on the electric power scenario multidimensional attribute label data set to obtain a standard electric power scenario multidimensional attribute label data set, and to design a functional architecture of an electric power semantic large model according to the electric power scenario application requirements; An association clustering module, the association clustering module is used to perform semantic mapping and matching based on the functional architecture of the electric power semantic big model and the standard electric power scenario multidimensional attribute label data set to obtain electric power scenario semantic association rules, and to perform association clustering on the standard electric power scenario multidimensional attribute label data set based on the electric power scenario semantic association rules to obtain an electric power scenario functional semantic label data set; The model acquisition module is used to train and optimize the power scenario function semantic label data set using the power semantic big model functional architecture, obtain a set of power scenario function semantic models, perform semantic topological connection on the set of power scenario function semantic models, and generate a power scenario semantic optimization big model.

Images