Design And Implementation Of Intelligent Customer Service Q&A System For Power Grid

With the increasing demand for automation and efficiency of the power grid,the power industry needs to accelerate the pace of building a highly intelligent and intelligent platform.The power system is the most complex system and network established by human beings,and the application of AI has been running through the development of the entire industry.Based on its own positioning,the State Grid Customer Service Center actively embraces the new era of intelligence,uses information technology innovation to accelerate service transformation,builds an intelligent operation and maintenance platform,and builds a "smart family" to continuously improve the service level of the power industry and meet the beautiful needs of the people.The combination of intelligent question answering technology and the power service industry is both a demand and an inevitability.However,the power grid operation and maintenance customer service Q&A system for the professional model field is different from the open and emotional Q&A systems on the market,and some conventional application technologies cannot be used directly.The problem is still not well resolved.The power operation and maintenance service platform needs a fast,convenient,accurate,intelligent and comprehensive intelligent question and answer platform.This paper uses natural language processing and other technologies,combined with neural network model,to build a set of accurate and efficient electric power operation and maintenance customer service question and answer system.In order to adapt to serviceability and matching efficiency,the question answering system designed in this paper has two matching techniques.For the question and answer database,a BERT-Bi LSTM-GCN matching model is proposed.The data is input into the vector obtained by the BERT model,and the predicted label sequence can be obtained through the Bi LSTM network.The GCN network can capture the grammatical features of sentences and reduce invalidation.The number of tags,the encoding layer makes the semantic representation of the text and the syntactic features of the sentence present a complementary relationship,and finally the candidate answer sequence is obtained through similarity calculation.Through experimental comparison,this model can capture the semantic features of sentences better than the traditional BERT model because of its attached extraction layer.Although the number of parameters has increased,the overall recognition rate is better.For the power knowledge graph,a question and answer based on the power knowledge graph is designed.matching model.As long as the entities and entity-entity relationships in the user’s question are determined,the correct answer can be found on the knowledge graph.In the entity extraction part,the Bi GRU-CRF model structure is used to extract the question entity.The Bi GRU network first extracts the features of the preprocessed knowledge corpus,obtains the forward hidden state through forward propagation,and obtains the backward hidden state through back propagation.The hidden state output is spliced to obtain the complete hidden state,thereby obtaining automatically extracted sentence features.Then use the label information before and after to add constraints to the last predicted label through the CRF layer,construct a linear chain with each label as a node,and use a two-dimensional matrix to store the transition score from one label to another label to determine the output sequence.In relation extraction,this section uses an AttBi LSTM model,which is a structure of attention mechanism plus bidirectional long short-term memory neural network.An attention-weighted summation of the outputs of a bidirectional long short-term memory neural network is performed using an attention mechanism.Although the model is relatively conventional,the extraction effect is very impressive.This paper analyzes the requirements of the power operation and maintenance customer service Q&A system,designs the system architecture,determines the technical route,and builds the power operation and maintenance customer service Q&A system with four layers: business display layer,information analysis layer,answer extraction layer and data storage layer.Through the verification test of system evaluation index and user evaluation index,the function of the system is intact.