A combined model of dissolved oxygen prediction in the pond based on multiple-factor analysis and multi-scale feature extraction

Abstract

As dissolved oxygen (DO) is an important indicator of water quality in aquaculture, an accurate prediction for DO can effectively improve quantity and quality of product. Accordingly, a novel hybrid dissolved oxygen prediction model, which combines the multiple-factor analysis and the multi-scale feature extraction, is proposed. Firstly, considering that dissolved oxygen is affected by complex factors, water temperature and pH are chosen as the most relevant environmental factors for dissolved oxygen, using grey relational degree method. Secondly, the ensemble empirical mode decomposition (EEMD) is adopted to decompose the dissolved oxygen, water temperature and pH data into several sub-sequences, respectively. Then, the sample entropy (SE) algorithm reconstructs the sub-sequences to obtain the trend component, random component and detail component. Lastly, regularized extreme learning machine (RELM), a currently effective and stable artificial intelligent (AI) tool, is applied to predict three components independently. The prediction models of random component, detail component and trend component are RELM1, RELM2 and RELM3 respectively. The dissolved oxygen, water temperature and pH of the random component forms the input layer of RELM1, and predicted value of dissolved oxygen in the random component is the output layer of RELM1. The input and output of RELM2 and RELM3 are similar to that of RELM1. Final prediction results are obtained by superimposing three components predicted values. One of the main features of the proposed approach is that it integrates the multiple-factor analysis and the multi-scale feature extraction using grey correlation analysis and EEMD. Its performance is compared with several outstanding algorithms. Results for experiment show that the proposed model has satisfactory performance and high precision.