Hazardous Waste Generation And Discharge Anomaly Detection Technology Study Based On Machine Learning And Knowledge Graph

The variety of hazardous wastes,the complexity of their composition,the wide range of industries involved,which makes it difficult for management departments to identify and monitor it.Therefor the use of intelligent methods to correlate information on hazardous waste generation and emissions with enterprise production is essential to achieve the detection of hazardous waste anomalies.Compared to traditional methods,machine learning can significantly reduce labour and time costs with high accuracy rates,providing technical support for hazardous waste anomaly detection.At the same time,knowledge graph technology can visualize the relationship between complex hazardous waste generation information,enterprise production information and other information in a graphical structure,which not only assists managers and enterprises to quickly query and learn about hazardous waste,but also enables rapid auditing of abnormal values of hazardous waste in enterprises.This study established the relationship between hazardous waste and enterprise production information based on machine learning and knowledge graph technology to achieve anomaly detection of hazardous waste in enterprises.Firstly,the key hazardous waste-related enterprises in Hebei Province were screened and experimental data samples were identified,and the material names of the enterprises were normalised using unique thermal coding to transform the textual data of the enterprises’ industrial production into numerical data for machine learning.Secondly,based on the Graph Convolutional Neural(GCN)network model in machine learning,enterprises were labelled with industry categories,and enterprises’ raw and auxiliary materials or products were used as input parameters to learn the characteristics of raw and auxiliary materials or product information used by enterprises under each industry,and then realised the function of using enterprises’ raw and auxiliary materials or product information to predict enterprises’ industry categories.Accuracy,precision,recall and F1 value(the summed average of precision and recall)were used as evaluation criteria for the model to predict industry results,and the best data for the model results were selected to summarise the hazardous waste generation characteristics of each industry using the raw and auxiliary materials and products used by enterprises and the hazardous waste generated.Then,based on the National Hazardous Waste Inventory(2021Edition)and the industrial production information of enterprises,the graph database Neo4 j was used to build a knowledge graph with hazardous waste as the core,and to portray the network relationships between entities such as hazardous waste information,enterprises,raw and auxiliary materials,products and processes.Finally,identified the abnormalities of hazardous waste reported by enterprises based on the industry hazardous waste characteristics,and visualise the identified abnormal values based on the knowledge graph to audit the hazardous waste abnormalities.The results showed that the GCN model predicts enterprise industry categories using enterprise products as input parameters better than the results using raw and auxiliary materials as input parameters,and the accuracy,precision,recall and F1 values can reach:77.05%,84.52%,74.69% and 0.7768 respectively.Based on the established industry hazardous waste generation characteristics and hazardous waste list knowledge graph of hazardous waste anomaly detection process,110 enterprises in Shijiazhuang’s five key industries were detected for hazardous waste,and a total of 23 enterprises were identified to have abnormal hazardous waste data,ultimately achieving rapid identification and audit of under-reporting and mis-reporting of hazardous waste by enterprises.