Simulation And Prediction Of Water Quality In Industrial Aquaculture System By Using Machine Learning Method

Aquaculture is important for coastal region.It is critical for aquaculture to ganrantee the aquatic product quality while water quality is essential for aquaculture production and aquatic product quality safety.Water quality parameters accurately represent the water quality of aquaculture system.However,water quality real-time monitoring possesses high operation and maintenance cost.Therefore,it is necessary to develop the stable,accurate,and useful water quality simulation and prediction models for precise management of aquaculture,sustainbale development of aquaculture,and protection of coastal aquatic environment.This study used machine learning method to simulate and predict five water quality parameters including p H,dissolved oxygen（DO）,ammonia nitrogen（NH₃-N）,nitrite nitrogen（NO₂-N）,and nitrate nitrogen（NO₃-N）of industrial aquaculture system.The ensemble empirical mode decomposition and wavelet transform signal technology were used to reduce noise.The accuracy and reliability of machine-learning-based models were also investigated.The main results of this study included:（1）Four machine learning models including BP neural network,RBF neural network,support vector machine（SVM）and least squares support vector machine（LSSVM）were used to simulate water quality,and SVM model was determined as the best model for simulating water quality of industrial aquaculture system.Compared with other 3 models,all simulation results of SVM model by using different datasets including data of published article,measured data,and large-sample-quantity data were stable and accurate with simulation accuracy>95%for each water quality parameter.（2）The ensemble empirical mode decomposition（EEMD）signal denoising technology was used to denoise the water quality data.The water quality simulation prediction model based on EEMD-machine-learning model was constructed.EEMD noise reduction technology enhanced the simulation accuracy of BP neural network,RBF neural network and LSSVM model,showing that EEMD might be useful for improving some machine learning models such as neural network.However,and the simulation accuracy of EEMD-SVM model was lower than that of SVM model,illustrating that EEMD might be not suitable for SVM model.（3）Wavelet transform（W）signal denoising technology was used to denoise the water quality data.A water quality simulation and prediction model based on wavelet transform and machine learning model was constructed.Data before and after W noise reduction did not change much to cause negligible improvement in the simulation result of the four models.It should be noted that the simulation accuracy of particular water quality parameter of individual models was slightly improved.（4）The genetic algorithm was introduced to optimize the SVM model,and the water quality prediction model of support vector machine based on genetic algorithm（GA-SVM）was constructed.The simulation accuracy of GA-SVM model was not satisfactory.Compared with the SVM model,GA-SVM optimization was not suitable for the simulation and prediction of water quality parameters in industrial aquaculture farms.