Research And Development Of Intelligent Response System For Elevator Emergency Handling Oriented To Task-based Dialogue

In recent years,with the rapid development of artificial intelligence in the field of natural language processing,more and more related applications continue to appear in our lives,among which the human-computer interaction conversation system is the hot spot of the current research in the industry and academia.In order to solve the conflict problem of alarm system of the traditional elevator emergency handling service platform,this paper establishes a response system based on the elevator emergency situation.When the elevator breaks down and people get stuck,the system can conduct voice guidance and comfort for the trapped people,then interact and share data with the rescue center through the cloud platform,so as to realize the efficiency and accuracy of intelligent response.According to the application type,the dialogue system is mainly divided into taskbased and non-task-based.Among them,the realization of task-based dialogue system is divided into two ways: based on pipeline and based on end to end.In order to meet the real-time processing requirements of elevator emergency handling,this paper chooses the task-based dialogue system based on PIPELINE to realize the development of the response system under the elevator emergency scenario.The main work of this paper is as follows:(1)This paper improved three combined model algorithms based on the Bert.To actualize slot filling and intention recognition tasks of natural language understanding module,this paper reproduces the function of the combined model of original BERT as the baseline.BERT+CRF,Bert + Label Attention and Bert fur-training,all of them have shown the better results than the Baseline model by comparison.(2)A Chinese chat model algorithm based on Word2 Vec + Seq2 Seq + Attention +Beam Search structure is proposed.Based on the Seq2 Seq structure,this paper implemented the Chinese chat model on small order of magnitude of the chat corpus.Based on the Seq2 Seq + Attention + Beam Search model structure,the experiment compared the influence of word vectors on the final experimental results,and finally concluded that the Word2 vec + Seq2 Seq + Attention + Beam Search model structure can better generate the reply of chat dialogue.(3)The response system combined with Raspberry Pi was successfully developed in the elevator emergency response scenario.First,through detailed demand analysis,the whole system process framework has been designed;Then the open source framework RASA,endpoint detection,Baidu AI and other technologies are applied to code and build each module of the response system,as well as the function of interacting with the cloud platform.Finally,combined with Raspberry Pi and other hardware devices(pickers,acoustics),a complete response system interaction process is realized in the elevator emergency scenario.