Feed intake prediction model for group fish using the MEA-BP neural network in intensive aquaculture

Abstract

In aquaculture, the accurate prediction of feed intake for group fish is considered to be crucial to any feeding system. Previous studies mainly used mathematical statistics to establish the mapping relationship between feed intake and influencing factors. The result was easily influenced by subjective experience. To solve the above issues, this paper proposed a feed intake prediction model for group fish using the back-propagation neural network (BPNN) and mind evolutionary algorithm (MEA). Firstly, four factors, including water temperature, dissolved oxygen, the average fish weight and the number of fish were selected as the input of the BPNN model. Secondly, the initial weight and threshold of the BPNN were optimized by the MEA to improve the matching precision. Finally, the prediction model was achieved after training. Experimental results showed that the correlation coefficient between the predicted and measured values reached 0.96. And the root mean squared error, mean square error, mean absolute error, mean absolute percent error of the model was 6.89, 47.53, 6.17 and 0.04, respectively. In addition, the proposed method also had the better nonlinear fitting ability than BPNN and GA-BP. By using an intelligent optimization algorithm, the mapping relationship between fish intake and environmental factors was automatically established, thus avoiding the subjectivity of traditional methods. Therefore, it can lay a theoretical foundation for the development of intelligent feeding equipment and meet the needs of the smart fishery.