Prediction of dissolved oxygen in aquaculture based on EEMD and LSSVM optimized by the Bayesian evidence framework

Abstract

In order to improve the accuracy and effectiveness of dissolved oxygen (DO) prediction, a combined forecasting model based on ensemble empirical mode decomposition (EEMD) and least squares support vector machine (LSSVM) is proposed. Firstly, the DO time series are decomposed into a group of relatively stable subsequences by ensemble empirical mode decomposition to reduce mutual influences among diverse trend information. Secondly, the decomposed subsequence is reconstructed by phase space reconstruction (PSR), and then, an LSSVM optimized by the Bayesian evidence framework prediction model of each sub-sequence is established. Lastly, we use Bp neural network to reconstruct the predicted values of each component to obtain the predicted value of the original DO sequence. This paper used the single point iterative method to achieve multi-step prediction in order to obtain forecasting results for 24 h into the future. EEMD-LSSVM is tested and compared with other algorithms in the Jiangsu Liyang huangjiadang special aquaculture farms. The experimental results show that the proposed combination prediction model of EEMD-LSSVM has a better prediction effect than WD-LSSVM, EEMD-ELM and standard LSSVM methods. The relative mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE) and the largest error (emax) for the EEMD-LSSVM model are 0.0261, 0.2161, 0.1721 and 0.0767, respectively. Consequently, it is clear that the EEMD-LSSVM model has high forecast accuracy and generalization ability.