The Study For Water-saving Potential Model According To Recurrent Neural Network

The water-saving potential is the water-saving effect achieved against the same specificwater unit because of using new water-saving appliances, improving water-saving technologyor taking some kind of water-saving measures. The water-saving potential relate to thebuilding of water-saving capacity and the building of wading management capacity, and hasan important guiding role in water conservation, and even has an mandatory role in thebuilding of water conservation capacity.There are many indicators when calculating water-saving potential, so, it is difficult tocalculate approximately using a linear function, however, the existence of stock water-savingand incremental water-saving cause some difference between water-saving potential and theactual amount of water-saving. Because the calculation of water saving potential is nonlinear,a new way of calculating water-saving potential is proposed in this paper, which includesanalysing water-saving indicators, analysing the relationship between water-saving indicatorsand water-savingusing potential by using Elman neural network, which has the capability ofinformation delay and delay information feedback.Picking up index data by using sample selection and index selection theory, andnormalizing the matrix which is consisted of index of different years. The training and testingsamples are generated, the Elman neural network is builded and the network trainingparameters are setted by picking up matrix data. The network training is processed byinitializing the network and using the training sample. Finally, the test results and test errorsare given by using test sample to test the network.The results outputed by selecting appropriate water-saving indexs and networkparameters, show that the training error of the network can achieve the requiredaccuracy.When the hidden layer nodes is14, the prediction error of the amount ofwater-saving is lesser and conformity is99.07%, which means that the Elman neural networkis self-learning and adaptive when hidden layer nodes is14, which can be used in theWater-saving potential.