Feature Extraction and Prediction of Water Quality Based on Candlestick Theory and Deep Learning Methods

In environmental hydrodynamics, a research topic that has gained popularity is the transmission and diffusion of water pollutants. Various types change processes in hydrological quality are directly related to meteorological changes. If these changing characteristics classified effectively, this will be conducive application deep learning theory pollution simulation. When periodically monitoring quality, data were represented with candlestick chart, different classification features displayed. The from area 2012 2019 generated 24 results line physics laws. Therefore, prediction method was proposed classify process improve accuracy based on theory, visual geometry group, gate recurrent unit (CT-VGG-GRU). method, after periodic changes by graphically, extracted VGG network its advantages graphic feature extraction. Then, other scenario parameters fused as input time series model, pollutant concentration sequence at predicted station constituted output model. Finally, hybrid model combining graphical formed, used continuous multiple stations Lijiang River watershed train validate Experimental indicated that, compared comparison models, such back propagation neural (BPNN), support vector regression (SVR), GRU, VGG-GRU, had highest accuracy, especially for extreme values. Additionally, trend closer real situation, which information could fully CT-VGG-GRU theory. For indicators DO, CODMn, NH3-N, mean absolute errors (MAE) 0.284, 0.113, 0.014, root square (RMSE) 0.315, 0.122, 0.016, symmetric percentage (SMAPE) 0.022, 0.108, 0.127, respectively. established achieved superior computational performance. Using river obtained effectively also retained, made more explanatory. can provide new prediction.