Selection And Comparison Of AI Methods Of LID Water Quality Prediction In Sponge City

The rapid development of modern society and the increasing development of urban construction.Water pollution has become an urgent problem in urban architecture design.Sponge City rationally control the amount of rain and flood in a city,so that the city can be like a sponge for prevention,storage and treatment.Among them,runoff pollution control is an important goal of the construction of sponge cities.By predicting the effluent quality of Low-Impact Development(LID)facilities in Sponge City,it can provide early warnings for the non-compliance of the effluent quality of LID facilities,thereby reducing urban runoff pollution,and providing the decision support of LID construction programs that meet the effluent quality standards under extreme weather conditions.At present,research on the use of artificial intelligence methods for water quality prediction in LID facilities is still blank.BP neural network and SVM are commonly used artificial intelligence algorithms in water quality prediction.The LSTM neural network has been widely studied in recent years.These three algorithms are selected to conduct a comparative study on the water quality prediction of LID facilities.The main research works were as follows:(1)Learn to understand the characteristics and application overview of the six commonly used water quality prediction algorithms: time series,regression analysis,grey model,mathematical statistics,artificial neural network and support vector machine regression;(2)The principle and characteristic of BP neural network,SVM and LSTM neural network were analyzed emphatically.Water quality data are characterized by many influencing factors and complex relationships.The BP neural network has nonhazardous dynamic properties,and its ability to block incoming and outgoing data is also very strong.Considering a series of very complex nonlinear problems in the urban design process,we can abstract the neuron simulation into a network to establish the relevant water quality prediction model.Support vector machine is a very effective method to deal with nonlinear problems.It is established on the basis of statistical theories without prior knowledge and is more suitable for some relatively small sample calculation processes.LSTM neural network has a unique neuron structure,which is more suitable for time series learning than CNN.Moreover,the LSTM gating system has the ability of self-circulatory weight adjustment,which solves the problem that RNN is easy to rely on for a long time and the gradient disappears to a certain extent.(3)Multiple rainfall data of a super large-scale urban sponge city pilot urban area were Collected in 2018,including rainfall,maximum rainfall intensity,runoff,runoff coefficient,etc.After principal component analysis,the BP neural network,SVM algorithm and the LSTM neural network were used to establish prediction models,and the effluent quality of LID facilities were predicted by experiments.The experimental results show that the root mean square error(RMSE)of the BP neural network algorithm is 29.221,the coefficient of efficiency(CE)is 0.835;the root mean square error(RMSE)of the SVM algorithm is 34.956,and the coefficient of efficiency(CE)is0.764;the root mean square error(RMSE)of the LSTM neural network algorithm is15.417,the coefficient of efficiency(CE)is 0.954.Therefore,the LSTM neural network is better than the BP neural network algorithm and SVM algorithm in the prediction accuracy of the effluent quality of the LID facility.