Study On The Prediction Methods Of Water Quality And Water Levels In Riverside Well Field

The law of the groundwater closely related to surface water and can be convertedinto each other, through which effectively increases the exploitation of groundwater inriverside well field. In many areas especially the hydropenic cities of the northern China,the governments often alongside the river water arrange wells to intake water as the mainwater supply of the city, Shenyang is the typical representative. Hunhe flows throughShenyang, directly involves in groundwater water cycle, and it takes an important part ofrecharge of groundwater. At present, however, the water quality of Hunhe is concerned,making the water quality difficult to guarantee, drinking water unsafety. Therefore,further the research of forecast methods of groundwater level and water quality surroundHunhe, to well manage ground water source field and for the security to provide securityof safe quality of water.Firstly, we study on the several of groundwater dynamic prediction methods andanalysis that the theories such as the gray system theory exist the deficiencies of short offorecast time、weakening the law of randomness data insufficient and so on, but thegeneralization ability of the Artificial Neural Network (ANN) model and the SupportVector Machine (SVM) model are very strong, what’s more, the robustness and faulttolerance are also good, which are suitable to be used in the areas of lack of data material.So in this paper, we select the neural network model and the support vectormachine model as the groundwater quality and water level prediction models.Establish the neural network model based on the Matlab platform and realize thesupport vector machine (SVM) model in the Windows DOS environment, comparing theprediction results of the two kinds of model to conclud the main conclusions as follows:1）In the BP water quality prediction model,we use2000-2008water quality datas asthe training sample，the2008datas as the tested samples；Use the single well andmulti-wells methods to predicte the water quality. Input and output layerneuron number is8and1, the errors is least; Take the principle of the less of modeliteration times and the high accuracy prediction to determine the number of hidden nodes.And concludes that node number of implicit layer in the single well water qualityprediction is six,the mult wells is nine; the layer node number of water level forecastingmodel is nine. 2）Through analysising the prediction results of ammonia nitrogen and nitratenitrogen, they show that the accuracy of the multi-wells prediction is higher than thesingle well. However, because of the highth of repeatability of nitrite nitrogen dataresulting in low learning model generalization ability, so at this moment, the accuracy ofmulti-wells water quality is very low.3）Compare the predict results of the ammonia concentration (2010-2014) using thesingle well and multi-wells with the true value (2000-2009), the single well predictioncan not eliminate of big value. However, the model prediction of multi-well evenlydistributed and eliminate the influence of big values indicate that the multi-wellpredictive reliability is better.4）In the water quality and water level prediction model, the support vector machineforecasting results precision are higher than the neural network model, which indicatethat the support vector machine (SVM) model has better generalization ability.