In Situ Indirect Detection of Phosphate Concentration from Aquaculture Water Using Physico-limnological Sensor-Based Feed-Forward Neural Network

Authors:
Publisher website
Google Scholar
Abstract
High phosphate levels in aquatic ecosystems induce eutrophication, which causes algae to overgrow and disrupt biodiversity. Determining the phosphate concentration is significant for maintaining optimal ecological function and water quality. In this study, five machine learning regression techniques were investigated for detecting the phosphate level of an aquaculture environment. The computational models employed physico-limnological data such as pH level, electrical conductivity, and water temperature as predictors, acquired from the tilapia fishpond in Rizal, Philippines. These models were evaluated based on the defined criteria such as the predictive performance and mean absolute error. All machine learning models produced acceptable results with R 2 value greater than 0.8. Among these, the feed-forward neural network model is concluded to be the most effective phosphate prediction model with R 2 =0.93 and MAE=0.39. Furthermore, a hybrid approach of multigene genetic programming and genetic algorithm (MGGP-GA) was implemented for optimization of phosphate level. GPTIPSv2, a MGGP tool, was used to create symbolic models. The model with the highest predictive accuracy and lower complexity was configured as the key component of GA architecture. This optimization technique produced an optimum parameter value of 21.79°C for water temperature, 6.9 for pH, and 0.74 mS/cm for electrical conductivity with a fitness value evaluation of 7.521. With that, this approach serves as an effective technique for in situ managing the nutrient content, particularly the phosphate level of aquatic systems.

This publication has 21 references indexed in Scilit: