Ammonia nitrogen prediction in surface water based on bidirectional gated recurrent unit

For more accurate prediction of NH4+-N, this paper proposes a novel hybrid forecast model (CCB) that uses complementary complete ensemble empirical mode decomposition with adaptive noise (CCEEMDAN) and bidirectional gated recurrent unit (BiGRU) neural network. Firstly, the original NH4+-N data is decomposed into several relatively simple components by CCEEMDAN. Subsequently, BiGRU neural network is employed to predict each component. The final forecast result is obtained by the summation of all the prediction results for the decomposed components. NH4+-N data of Poyang Lake that was monitored from June, 2017 to February, 2020 is used to evaluate the proposed forecast model. Mean absolute percentage error (MAPE) of the forecast result by our model is 3.38% for 1day ahead forecast, 6.82% for 7days ahead forecast and 9.41% for 15days ahead forecast. Moreover, CCB model shows better forecast performance than the competitor models. Results demonstrate that CCB model has a powerful forecast capacity, and it can be effectively used for the analysis and decision-making in water resource management. © 2022, Editorial Board of China Environmental Science. All right reserved.