Research On The Construction Of Multi-source Energy Knowledge Graph For Smart Home

At a time when automation and intelligence are respected,the work around the clock in all walks of life generates hundreds of millions of data volumes.In order to improve the effective use of industry data,research on the combination of artificial intelligence and specific fields has become a new way forward.In the era of high semantic interconnection,knowledge graph,with its excellent semantic expression ability,knowledge storage and reasoning ability,effectively solves the knowledge organization and intelligent application of data.The technology related to the construction and application of knowledge graph has also become the research content that scholars are keen on.The world’s limited reserves,environmental pollution and climate-affecting fossil energy issues make the road to sustainable human development very difficult.Developing and utilizing renewable energy and reducing the proportion of fossil fuels is an effective way to achieve green development and social well-being.Under the background of the development of smart grid,the behavior of power users to obtain profits by dispatching power to assist in solving the power demand under the premise of ensuring their own power demand has arisen.Urban and rural residents are a part of electricity users that cannot be ignored.Distributed renewable energy generation on the residential side can enable the residential side to achieve efficient interaction with the power grid,and the utilization rate of renewable energy will also be greatly improved.Home energy management systems connect homes to smart grids and increase the overall use of renewable energy by directing energy demand to off-peak times and improving energy savings.However,the effective information contained in the massive system-related data generated by household electricity consumption has not been fully exploited and utilized,resulting in a waste of data resources.The powerful knowledge expression ability and storage reasoning ability of knowledge graph provide a solution to this problem.This topic builds a knowledge graph of energy management for smart homes.The research content covers the construction of knowledge graphs and expertise in the field of electricity,including the following aspects:（1）For the knowledge extraction part of the knowledge graph construction research,firstly design and use the combined BERT and BiGRU-Attention-CRF model to complete the identification task of power entities.The BERT model is introduced to generate high-quality word vector representations considering contextual information,and bi-directional gating unit-attention mechanism-conditional random field（Bi GRU-Attention-CRF）is used to sequentially label and decode the semantic encoding output from the previous layer.The experimental comparison results of this model and several models with high frequency of use on the data set designed in this project and the MSRA data set prove that the model has a good effect on the task of entity extraction.The calibration of the power knowledge extraction results refers to the nomenclature specification of the power grid industry,the professional representation of the power system,and the Chinese entity classification standard.In addition,a relation extraction model based on a multi-class relational attention mechanism is proposed,taking the idea of remote supervised multi-instance learning as the starting point,and reducing the model’s dependence on manual annotation.The effect of the model is verified by comparison experiments with several other models.（2）In the part of knowledge storage,this topic selected Neo4 j for knowledge graph storage based on various reasons.This thesis is based on the triple data consisting of entities and entity relationships obtained by the power knowledge extraction task,and based on the extracted content,the knowledge fusion of triples is carried out,entity synonyms are mined,and ambiguous entities are distinguished and synonymous entities.coreference resolution and other work.The fused knowledge triples are stored in the selected Neo4 j to complete the final knowledge graph drawing and storage,covering power entities,entity attributes and associations between entities.In addition,the language Cypher can query and verify the content of the knowledge map,and the visualization platform of Neo4 j can complete the display of the knowledge map.（3）In order to efficiently utilize the constructed knowledge graph,this thesis implements a knowledge platform system connected to the Neo4j database with the help of B/S model architecture,and returns the corresponding results in the database to the knowledge platform page.Finally,a knowledge platform based on knowledge graph is constructed through system testing.