Research On Water Quality Prediction Model Of Urban Sewer Networks Based On Multi-source Data Fusion

Due to the impacts of accelerated urbanization and population growth,the water quality problem of urban sewer networks is becoming increasingly prominent.However,it is difficult to be discovered in time and solved efficiently because of the wide distribution of urban sewer networks and the poor timeliness of traditional water quality detection.In recent years,with the rapid development of technology,a large number of studies have shown that the construction of water quality prediction model of urban sewer networks is one of the most effective methods to achieve reasonable control of the water quality in urban sewer networks.It can avoid the time limitation of traditional water quality detection methods,to realize the timely prediction of the water quality in the sewer networks,and help to discover and solve the water quality problems of the urban sewer networks in time.However,the existing prediction models still have the shortcomings of low prediction accuracy,high modelling cost and poor practical application ability because water quality of urban sewer networks is very complicated.To solve the above problems,this study proposed an integrated water quality prediction model of urban sewer networks based on low-cost and easily accessible multi-source data（that is,indicator data from various sources such as environmental indicator data,social indicator data,water quantity indicator data,and easily detectable water quality indicator data）,which took deep learning algorithm as ontology and combined with various scientific methods,to achieve high-precision prediction of drainage water quality.In this paper,the water quality of urban sewer networks was continuously monitored,and the spatial-temporal variation law of water quality of sewer networks was analyzed,so as to determine the optimal multi-source data acquisition frequency.Then,it was determined that GRU had the best performance in multi-source data mining by comparing the performance of various algorithms.After that,the GRU,MGS and MSA were organically integrated to establish an accurate and efficient water quality prediction model for urban sewer networks.The MGS method was coupled to make the prediction model have the function of network structure and performance self-optimization.And combined with MSA method,the contribution degree of each input indicator to the prediction results was quantified,so as to realize the model input variable optimization mode based on the analysis of MSA.This study analyzed and discussed the three functions of the model（construction of mapping relationship between multi-source data and water quality data,self-optimization of model network structure and performance,and optimization of input variables based on the contribution degree of input variables）by means of case application analysis.The results showed that:（1）The model had a strong ability to capture and learn the characteristics of multi-source data of urban system and water quality data of sewer networks,and can construct strong and stable mapping relationship between them.Based on the characteristics of multi-source data with high frequency acquisition,this model was expected to achieve high frequency prediction of water quality of sewer networks.（2）The model can achieve high prediction accuracy for typical water quality indicators（BOD₅,COD,NH₄⁺-N,TN and TP）and key water quality indicators such as fat,oil and grease（FOG）and heavy metals in sewer networks,and the average R² value can reach more than 0.90.（3）The network structure of the GRU in the prediction model had a great influence on the prediction performance of the model,and the self-optimization function in the model can realize the optimal network structure to achieve the best prediction performance.（4）In the process of predicting the water quality of sewer networks,the contribution degree of multi-source data to the model prediction results was different,and a certain number of key input variables had strong interaction,so as to achieve high accuracy of the prediction results.In general,the model constructed in this study can predict the water quality of the urban sewer networks with high accuracy,and can effectively reflect the temporal and spatial changes of water quality at each node of the sewer networks,and timely grasp the temporal-spatial distribution of pollutants in the sewer networks.This not only helps to quickly check and trace the source of water quality problems in the sewer networks,but also provides a basis for the daily inspection of the sewer networks and the formulation of maintenance plans.In addition,the model can realize real-time prediction of the water quality of the sewer networks,and can provide timely and reliable basis for the early warning of abnormal or excessive water quality of the sewer networks,which will provide an important guarantee for the stable operation of the downstream wastewater treatment plants,and is of great significance to the development of urban smart water platform.