LSTM-Based Prediction of Water Quality Parameters System in Backwaters

Aquaculture provides food security, self-employment, and a sustainable source of income. At the same time, risks involved is very high as the fishermen practice traditional approaches to maintain the culture systems. The unforeseen changes in the water quality parameters may cause mass mortality of fishes leading to economic losses. To address this issue, an experimental investigation is carried out by monitoring water quality parameters in an exsisting aquaculture eco-system. LSTM is utilized to predict the water quality parameters 90 minutes in advance, which provides sufficient time window for fishermen to take appropriate precautions. Performance analysis of three such different LSTMs architecture has been conducted. It has been observed that, the Bi-directional LSTM can better model the dynamic nature of the data. Also, the outliers in the predicted values have been identified by employing Gaussian distribution model. From the experiment, it can be seen the performance of developed outlier detection system is acceptable. The decision support system thus developed, supports the culturists for maintaining the aquaculture eco system in a favorable condition.