Knowledge discovery and graph construction method for power production data business

Abstract

The invention discloses an electric power production data business-oriented knowledge discovery and graph construction method, which comprises the following steps of: 1) constructing an electric power production data business database of an integrated system, and storing real-time data into the database; 2) constructing a knowledge graph of multi-modal knowledge discovery; and 3) complementing knowledge graph relations based on reinforcement learning, and constructing knowledge discovery and graphs oriented to power production data services. According to the knowledge discovery and graph construction method oriented to the power production data business, efficient processing and utilization of data are achieved, related business information, technical knowledge, industry standards and deep relations of the power production data business objects are comprehensively and physically displayed, internal relations among entity objects are mined, and the knowledge discovery and graph construction efficiency is improved. And the user is helped to better understand and analyze the power production data service.

Description

Technical Field

[0001] The present invention relates to knowledge graph technology, and in particular to a method for knowledge discovery and graph construction for power production data services. Background Art

[0002] With the continuous development of the construction of power enterprise data centers, a large amount of power business production data has been connected to the data center. However, these data sources are scattered, large in quantity, and lack association relationships. The same data may exist in multiple different systems, and the same data may also come from different systems, making the data situation very complex.

[0003] There are multiple deficiencies in the existing technologies when dealing with power business data. First, there are often data island problems, and it is difficult to integrate and share data between different departments and systems, resulting in information inconsistency and redundancy. Second, data quality problems are also widespread, and errors and inconsistencies in the collection, storage, and processing processes may affect the accuracy and credibility of the data. In addition, existing methods usually process data in a batch mode, resulting in low processing efficiency and being unable to meet the real-time requirements of power production data services. At the same time, there is a lack of sufficient intelligent analysis capabilities, making it difficult to effectively mine patterns, anomalies, or associations from the data, which limits the deep application value of the data. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for knowledge discovery and graph construction for power production data services in view of the defects in the existing technologies.

[0005] The technical solution adopted by the present invention to solve its technical problems is: A method for knowledge discovery and graph construction for power production data services, comprising the following steps:

[0006] 1) Construct a power production data service database for an integrated system, and store real-time data in the database;

[0007] Among them, constructing a power production data service database for an integrated system includes: designing and constructing the database, then building a graph neural network model, connecting the database with the graph neural network, realizing the integration of the integrated system, and constructing a power production data service database for the integrated system;

[0008] 2) Construct a knowledge graph for multimodal knowledge discovery;

[0009] First, determine the main data modalities. For each modality, select appropriate feature extraction methods for knowledge extraction, construct sub-graphs, and fuse the knowledge extracted from different modalities to construct a comprehensive system knowledge graph. Finally, evaluate and optimize the constructed system knowledge graph;

[0010] 3) Knowledge graph relation completion based on reinforcement learning, constructing knowledge discovery and graph for power production data services;

[0011] 3.1) Define and model the environment based on the knowledge graph, where nodes represent entities in power production data services and edges represent the relationships between entities; each node and relationship is embedded into a vector space;

[0012] 3.2) Conduct reinforcement learning training on the agent;

[0013] The agent perceives the state by observing the environmental results. Each state consists of the current node and its related paths, and these paths are represented as vectors through embedding. The action is the relationship extension that the agent needs to execute in the current state, and the action space is composed of possible relationships in the knowledge graph. Design a reward function suitable for power production data services to guide the agent to obtain feedback based on the current state and the actions taken, and encourage the agent to expand the path to complete the knowledge graph relationship according to the reward value;

[0014] 3.3) Use the agent for path reasoning and optimization.

[0015] According to the above solution, in step 1), the design and construction of the database are carried out as follows:

[0016] 1.1) Determine the power production data to be collected, and select a suitable database for power production data, such as a relational database or a database for time-series data;

[0017] 1.2) Design the structure of the database tables, including entity objects and the relationships between entity objects;

[0018] 1.3) Design a data acquisition system to store real-time data in the database.

[0019] According to the above solution, in step 1), build a graph neural network model as follows:

[0020] Preprocess the data in the database, including data cleaning and normalization; then convert the data into a graph representation form, where nodes represent entity objects and edges represent the relationships between entities; select a neural network model suitable for graph data, and finally train and validate the neural network model. Use historical power production data to train the graph neural network model and use the validation set to optimize the model to obtain the graph neural network model.

[0021] According to the above solution, in step 1), connect the database with the graph neural network as follows:

[0022] Connect the data in the database with the graph neural network model to achieve data transmission and sharing;

[0023] Construct a knowledge graph; use a graph neural network model trained with historical power production data to construct the knowledge graph, and map the relationships between entity objects onto the graph;

[0024] Embed the knowledge graph into a database to obtain an integrated real-time power production data business database.

[0025] According to the above solution, in step 2), evaluate the constructed knowledge graph to determine whether its quality meets the integrated system indicators, ensure the performance of the model and the accuracy of knowledge extraction, then optimize according to the evaluation results, and regularly update the sub-graph to reflect the latest data and business situations.

[0026] According to the above solution, in step 3.2), perform reinforcement learning training on the agent. During the reinforcement learning training, the policy network receives the current state as input and outputs the probability distribution of the next action; by performing actions in the environment, observing rewards, and updating the policy network, use experience replay to stabilize the training and thus obtain a more reliable agent.

[0027] According to the above solution, in step 3.3), use the agent to perform path reasoning and optimization; specifically as follows:

[0028] After the agent is trained, use it for path reasoning. Given a starting node, the agent expands the path by selecting the most likely relationship, thereby inferring possible relationship paths.

[0029] The beneficial effects produced by the present invention are:

[0030] 1. The present invention proposes a knowledge discovery and graph construction method for power production data services. By constructing an integrated system, integrating the power production data business database with a graph neural network model, it realizes the efficient processing and utilization of data, comprehensively and systematically displays the relevant business information, technical knowledge, industry standards and their deep connections of power production data service objects, and excavates the internal connections between entity objects;

[0031] 2. The multi-level relationship path completion method based on reinforcement learning of the present invention enables the agent to expand the path by selecting the most credible relationship, realizes reasoning and relationship path completion in a large-scale knowledge graph, and helps users better understand and analyze power production data services. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The present invention will be further described below in conjunction with the drawings and embodiments. In the drawings:

[0033] Figure 1 is the method flow chart of the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0034] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0035] As Figure 1 shown, a knowledge discovery and graph construction method for power production data services includes the following steps:

[0036] 1) Construct a power production data service database for the integrated system and store real-time data in the database;

[0037] Among them, constructing a power production data service database for the integrated system includes: designing and constructing the database, then building a graph neural network model, connecting the database with the graph neural network, realizing the integration of the integrated system, and constructing a power production data service database for the integrated system;

[0038] The design and construction of the database are as follows:

[0039] 1.1) Determine the power production data to be collected and select a suitable database for power production data, such as a relational database or a database for time-series data;

[0040] 1.2) Design the structure of the database tables, including entity objects and the relationships between entity objects;

[0041] 1.3) Design a data acquisition system to store real-time data in the database.

[0042] For example, it is determined that the types of power production data to be collected include real-time monitoring data (such as voltage, current, temperature, etc.), historical operation data (such as equipment operation records in the past year), and equipment status data (such as the health status of generators and transformers). The data sources can obtain data from various sensors, monitoring systems, SCADA systems, etc.; secondly, select a suitable database for power production data, such as a relational database (MySQL, PostgreSQL) or a database for time-series data (InfluxDB, Prometheus); then design the database table structure, such as entity tables (generator tables, transformer tables), relationship tables (equipment connection tables), and data tables (real-time data tables); finally, design a data acquisition system. The data acquisition tool uses open-source tools or custom acquisition programs, and the data transmission protocol uses protocols such as MQTT and HTTP to transmit the collected data to the database.

[0043] The construction of the graph neural network model is as follows:

[0044] Preprocess the data in the database, including data cleaning and normalization; then convert the data into a graph representation, where nodes represent entity objects and edges represent the relationships between entities; select a neural network model suitable for graph data, and finally train and validate the neural network model. Use the historical power production data to train the graph neural network model and use the validation set to optimize the model to obtain the graph neural network model.

[0045] Preprocessing the data includes handling missing values and outliers. For example, use interpolation to fill in missing time series data, use statistical methods to detect and handle outliers, and standardize the data to a unified range, such as normalizing numerical values such as voltage and current between 0 and 1; the graph representation is in the form where each device (such as a generator, transformer) is a node, and the connection relationship between devices is an edge, and the weight of the edge can represent the strength or importance of the connection; select the GraphConvolutional Network (GCN) model, use historical data for training, and the training objective is to predict device status or detect abnormal connections, and use the cross-validation method to evaluate the model performance.

[0046] Connect the database to the graph neural network as follows:

[0047] Connect the data in the database to the graph neural network model to achieve data transmission and sharing;

[0048] Construct a knowledge graph; use the graph neural network model trained with historical power production data to construct the knowledge graph and map the relationships between entity objects onto the graph;

[0049] Embed the knowledge graph into the database to obtain an integrated real-time power production data business database.

[0050] In this embodiment, use SQLAlchemy to connect to the database, extract data and convert it into a graph representation, use Pandas to process the data, and transfer the data to the graph neural network model; extract device information from the database to construct nodes, extract device connection information from the database to construct edges, and use the NetworkX library to construct the knowledge graph;

[0051] Use visualization technology to display the results:

[0052] Use D3.js or the Plotly library to develop a web interface to display the knowledge graph;

[0053] 2) Construct a knowledge graph for multi-modal knowledge discovery;

[0054] First, determine the main data modalities. For each modality, select an appropriate feature extraction method for knowledge extraction, construct sub-graphs, and fuse the knowledge extracted from different modalities to construct a comprehensive system knowledge graph. Finally, evaluate and optimize the constructed system knowledge graph;

[0055] The specific operations are as follows: Determine that the main data modalities include real-time monitoring data (such as voltage, current, temperature, etc.), equipment status data (such as equipment operating status, health status), market information data (such as electricity prices, supply and demand conditions), and environmental factors (such as weather data, ambient temperature);

[0056] Use models such as ARIMA and LSTM to extract features of real-time monitoring data, use methods such as TF-IDF and Word2Vec to extract features of market information, and use methods such as PCA and LDA for feature dimensionality reduction and analysis; Select knowledge extraction models such as machine learning models (such as random forest, support vector machine), deep learning models (such as convolutional neural network, long short-term memory network), and statistical models (such as Bayesian network, Markov chain); Fuse the features of different modalities to construct a comprehensive system knowledge graph.

[0057] Evaluate the constructed knowledge graph to determine whether its quality meets the integrated system indicators, ensure the performance of the model and the accuracy of knowledge extraction, then optimize according to the evaluation results, and regularly update the sub-graphs to reflect the latest data and business situations.

[0058] Use metrics such as accuracy, recall rate, and F1-score to evaluate the quantity, and use cross-validation methods to evaluate the model performance; Adjust the feature extraction method according to the evaluation results, and adjust the model parameters such as learning rate and regularization parameter according to the evaluation results; Update the sub-graphs once a month or a quarter to reflect the latest data and business situations, and use incremental learning methods to avoid the overhead of re-training with all data.

[0059] 3) Based on the results of step 2), perform knowledge graph relationship completion based on reinforcement learning to construct a knowledge discovery and graph oriented to power production data services;

[0060] 3.1) Define and model the environment based on the knowledge graph. The nodes represent entities in the power production data service (such as generators, transformers), and the edges represent the relationships between entities (such as connection relationships, dependency relationships). Use methods such as Node2Vec or DeepWalk to embed the nodes and edges into the vector space.

[0061] 3.2) Conduct reinforcement learning training on the agent;

[0062] The agent perceives the state by observing the environmental results. Each state consists of the current node and its related paths. These paths are represented as vectors through embedding. The action is the relationship expansion that the agent needs to execute in the current state. The action space is composed of the possible relationships in the knowledge graph. A reward function suitable for the power production data service is designed to guide the agent to obtain feedback based on the current state and the actions taken, and encourage the agent to expand the path according to the reward value to complete the knowledge graph relationship.

[0063] The agent is trained with reinforcement learning. During the reinforcement learning training, the policy network receives the current state as input and outputs the probability distribution of the next action. By executing actions in the environment, observing the rewards, and updating the policy network, experience replay is adopted to stabilize the training and thus obtain a more reliable agent.

[0064] 3.3) Use the agent for path reasoning and optimization.

[0065] After the agent is trained, use it for path reasoning. Given a starting node, the agent expands the path by selecting the most likely relationship, thereby inferring the possible relationship path.

[0066] To process large-scale knowledge graphs, optimization methods such as sampling and distributed training are used to ensure the scalability of the model.

[0067] After obtaining the completed knowledge graph, visualization technology can be used to display the information in the knowledge graph, including real-time data, historical trends, relationships, etc., providing an intuitive interface to help users better understand and analyze the power production data service. The constructed integrated system can provide decision support for the power production business, integrate with other applications through the interfaces provided by the system, and help users make more scientific and reasonable decisions. It provides strong technical support for the power production business.

[0068] It should be understood that for those of ordinary skill in the art, improvements or transformations can be made according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention.

Claims

1. A knowledge discovery and graph construction method for power production data business, characterized in that: The following steps are involved: 1) Build an integrated system's power production data business database and store real-time data in the database; Among them, building the power production data business database of the integrated system includes: designing and building the database, then building a graph neural network model, connecting the database with the graph neural network, realizing the integration of the integrated system, and building the power production data business database of the integrated system; 2) Construct a knowledge graph for multimodal knowledge discovery; First, determine the main data modalities, select appropriate feature extraction methods for each modality to extract knowledge, build sub-graphs, and integrate the knowledge extracted from different modalities to build a comprehensive system knowledge graph. Finally, evaluate and optimize the constructed system knowledge graph. 3) Knowledge graph relationship completion based on reinforcement learning to build knowledge discovery and graph for power production data business; 3.1) Define and model a knowledge graph-based environment, where nodes represent entities in the power production data business and edges represent relationships between entities; each node and relationship is embedded in a vector space; 3.2) Perform reinforcement learning training on the agent; The agent perceives the state by observing the environmental results. Each state consists of the current node and its related paths. These paths are represented as vectors through embedding. The action is the relationship extension that the agent needs to perform in the current state. The action space consists of possible relationships in the knowledge graph. A reward function suitable for the power production data business is designed to guide the agent to obtain feedback based on the current state and the actions taken. The agent is encouraged to expand the path to complete the knowledge graph relationship based on the reward value. 3.3) Use intelligent agents to perform path reasoning and optimization.

2. The knowledge discovery and graph construction method for power production data business according to claim 1 is characterized in that: In step 1), the database is designed and constructed as follows: 1.1) Determine the power production data that needs to be collected and select an appropriate database for power production data; 1.2) Design the structure of the database table, including entity objects and the relationships between entity objects; 1.3) Design a data acquisition system to store real-time data in a database.

3. The knowledge discovery and graph construction method for power production data business according to claim 1 is characterized in that: In step 1), a graph neural network model is constructed as follows: Preprocess the data in the database including data cleaning and normalization; then convert the data into a graph representation, where nodes represent entity objects and edges represent the relationship between entities; select a neural network model suitable for graph data, and finally train and verify the neural network model. Use historical power production data to train the graph neural network model, and use the verification set to tune the model to obtain the graph neural network model.

4. The knowledge discovery and graph construction method for power production data business according to claim 1 is characterized in that: In step 1), the database is connected to the graph neural network as follows: Connect the data in the database with the graph neural network model to realize data transmission and sharing; Construct a knowledge graph; use the graph neural network model trained with historical power production data to construct a knowledge graph and map the relationships between entity objects onto the graph; The knowledge graph is embedded into the database to obtain an integrated real-time power production data business database.

5. The knowledge discovery and graph construction method for power production data business according to claim 1 is characterized in that: In step 2), the constructed knowledge graph is evaluated to determine whether its quality meets the integrated system indicators to ensure the performance of the model and the accuracy of knowledge extraction. It is then optimized based on the evaluation results and the sub-graph is regularly updated to reflect the latest data and business conditions.

6. The knowledge discovery and graph construction method for power production data business according to claim 1 is characterized in that: In the step 3.2), the agent is trained by reinforcement learning. During the reinforcement learning training, the policy network receives the current state as input and outputs the probability distribution of the next action. By executing actions in the environment, observing rewards, and updating the policy network, experience replay is used to stabilize the training and thus obtain a more reliable agent.

7. The knowledge discovery and graph construction method for power production data business according to claim 1 is characterized in that: In step 3.3), an intelligent agent is used to perform path reasoning and optimization; specifically, as follows: After the agent is trained, it is used to perform path reasoning. Given a starting node, the agent expands the path by selecting the most likely relationship, thereby inferring possible relationship paths.

8. An electronic device, characterized in that: include: one or more processors; as well as a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are caused to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images