Multi-scale prediction of water temperature using empirical mode decomposition with back-propagation neural networks

• The novel model which combines EMD and BPNN algorithm is presented to predict water temperature in intensive aquaculture..
• Using EMD technology adaptively decomposed the original water temperature data into a finite set of IMFs and a residue.
• EMD-BPNN has higher prediction accuracy and better generalization performance than standard BPNN and standard SVR.
• EMD-BPNN can be used as a suitable and effective modeling tool for predicting water temperature in intensive aquaculture.

In order to reduce aquaculture risks and optimize the operation of water quality management in prawn engineering culture ponds, this paper proposes a novel water temperature forecasting model based on empirical mode decomposition (EMD) and back-propagation neural network (BPNN). First, the original water temperature datasets are decomposed into a collection of intrinsic mode functions (IMFs) and a residue by EMD yields relatively stationary sub-series that can be readily modeled by BPNN. Second, both IMF components and residue is applied to establish the corresponding BPNN models. Then, each sub-series is predicted using the corresponding BPNN. Finally, the prediction values of the original water temperature datasets are calculated by the sum of the forecasting values of every sub-series. The proposed hybrid model was applied to predict water temperature in prawn culture ponds. Compared with traditional models, the simulation results of the hybrid EMD–BPNN model demonstrate that de-noising and capturing non-stationary characteristics of water temperature signals after EMD comprise a very powerful and reliable method for predicting water temperature in intensive aquaculture accurately and quickly.