Research on Prediction Algorithm of Dissolved Oxygen in Aquatic Products Based on Improved LSTM Algorithm

This paper discusses a prediction method for dissolved oxygen in aquatic products based on the maximum information coefficient (MIC) and the long short-term memory (LSTM) algorithm. Firstly, the paper emphasizes the importance of dissolved oxygen in aquaculture, which is a key factor affecting the survival and growth of aquatic organisms. Despite the complexity of factors such as temperature, light, water mobility, and biological activity that influence the dynamics of dissolved oxygen, predicting its precise levels remains a formidable challenge. To address this issue, we present the maximum information coefficient (MIC) as an innovative tool for correlation analysis. Unlike traditional correlation coefficients, MIC is able to detect linear, nonlinear, and many-to-many relationships, which is useful for revealing potentially complex relationships between dissolved oxygen and other environmental variables. By calculating the MIC values between environmental variables and dissolved oxygen, we can screen out the most influential factors for the prediction of dissolved oxygen. Subsequently, long short-term memory (LSTM) neural networks were employed to predict dissolved oxygen using selected important environmental variables as inputs. LSTM is a unique recurrent neural network known for its distinctive gating mechanism, which allows the network to effectively capture long-term dependencies in time series data. We input historical dissolved oxygen data and relevant environmental variables into a trained LSTM model to predict future dissolved oxygen levels. The experimental results indicate that the prediction method utilizing MIC and LSTM exhibits strong performance in predicting the accuracy of dissolved oxygen, surpassing traditional statistical models and several other machine learning methods.