Evaluation of 1-D tracer concentration profile in a small river by means of Multi-Layer Perceptron Neural Networks

摘要

The prediction of temporal concentration profiles of a transportedpollutant in a river is still a subject of ongoing research effortsworldwide. The present paper is aimed at studying the possibility ofusing Multi-Layer Perceptron Neural Networks to evaluate the wholeconcentration versus time profile at several cross-sections of ariver under various flow conditions, using as little informationabout the river system as possible. In contrast with the earlierneural networks based work on longitudinal dispersion coefficients,this new approach relies more heavily on measurements ofconcentration collected during tracer tests over a range of flowconditions, but fewer hydraulic and morphological data are needed.The study is based upon 26 tracer experiments performed in a smallriver in Edinburgh, UK (Murray Burn) at various flow rates in a 540 mlong reach. The only data used in this study were concentrationmeasurements collected at 4 cross-sections, distances between thecross-sections and the injection site, time, as well as flow rateand water velocity, obtained according to the data measured at the1st and 2nd cross-sections.The four main features of concentration versus time profiles at aparticular cross-section, namely the peak concentration, the arrivaltime of the peak at the cross-section, and the shapes of the risingand falling limbs of the profile are modeled, and for each of them aseparately designed neural network was used. There was also avariant investigated in which the conservation of the injected masswas assured by adjusting the predicted peak concentration. Theneural network methods were compared with the unit peak attenuationcurve concept.In general the neural networks predicted the main features of theconcentration profiles satisfactorily. The predicted peakconcentrations were generally better than those obtained using theunit peak attenuation method, and the method with mass-conservationassured generally performed better than the method that did notaccount for mass-conservation. Predictions of peak travel time werealso better using the neural networks than the unit peak attenuationmethod. Including more data into the neural network training setclearly improved the prediction of the shapes of the concentrationprofiles. Similar improvements in peak concentration were lesssignificant and the travel time prediction appeared to be largelyunaffected.