Research On Prediction Method Of Fish Feeding Based On Machine Learning

In aquaculture,the level of feeding is directly related to aquaculture efficiency and production benefit.At present,there are many breeding modes and complex aquaculture environment in China.Therefore,how to achieve accurate prediction of fish feeding under different farming modes is a key scientific problem in production.With the continuous development and wide application of machine learning technology,it has become advance and tendency to apply it for the prediction of fish feeding.This paper intends to use fish of swimming as the research object,and to use the machine learning algorithm such as self-adaptive fuzzy-neural network,BP neural network and thought evolution algorithm for the characteristics of traditional breeding mode and intensive farming mode.The prediction of feeding was studied,including:1)Firstly,analysis of the main factors affecting fish feeding,including environmental factors,fish physiological factors,nutritional factors and management factors.This paper focuses on the environmental factors and the physiological factors of fish.By analyzing the relationship between these influencing factors and fish feeding,it lays the foundation for the proposed feeding prediction model.2)Secondly,aiming at the problem that the amount of culture information in traditional breeding mode is small and difficult to obtain,a two-parameter adaptive fuzzy network(ANFIS)feeding prediction model is proposed.Using water temperature and fish weight as input language variables,achieving the accurate prediction of feeding amount under the traditional farming mode basing on building two-parameter feeding prediction model by ANFIS through fuzzy reasoning.The experimental results show that the correlation between the predicted results of the proposed model and the actual feeding value is 0.98,which not only can refine the feeding rate table,but also save labor cost,and has a guiding role on fish feeding in the traditional farming mode.3)Finally,aiming at the characteristics of large number of types and quantities of sensor data in intensive farming mode,a multi-parameter MEA-BP(Mind Evolutionary Algorithm,MEA)dynamic feeding prediction model is proposed.The water temperature,dissolved oxygen concentration,body' weight and mantissa were used as input variables,and the evolutionary algorithm with strong global optimization ability was used to optimize the parameters of BP neural network.Finally,a multi-parameter-based feeding prediction model was constructed.The experimental results show that the correlation coefficient of the test sample fitting curve is 0.962.Compared with the feeding rate table method,the predicted value of the proposed model is closer to the measured value,and the model performance is better than the feeding rate table.The accuracy of the feeding amount prediction is improved,and the accurate and dynamic prediction of the feeding amount of fish under intensive farming is realized.