Real-Time Variable-Weight Aquaculture Water Quality Index Evaluation Method Based on Adaptive Sliding Window

Establishing an aquaculture water quality index (WQI) enables comprehensive water quality assessment and ensures aquaculture safety; however, existing WQI techniques are constrained by delays in conventional sampling and analysis methods, and their fixed weighting coefficients lack responsiveness to real-time changes. This article proposes a real-time variable-weight assessment method, termed "dynamic improvement entropy method (D-IEM)," based on an adaptive sliding window (AVSW). An Internet of Things (IoT) water quality monitoring system is developed to acquire real-time data, and a variable-weight WQI model is designed using pH, nonionic ammonia, chemical oxygen demand (COD), and phosphate. The D-IEM method dynamically calculates WQI weighting coefficients through information entropy and AVSW. In order to address the high cost and potential data loss in phosphate detection, interpretable evolutionary algorithms (EAs) optimize an extremely randomized trees (ERTs) model, which predicts phosphate concentrations and ensures WQI stability. The model demonstrates excellent performance, achieving a mean error of 1.136% for phosphate prediction under an equally weighted WQI assessment. Experimental results confirm that D-IEM effectively captures WQI weight change trends, enabling dynamic weight calculation and facilitating online, real-time aquaculture water quality assessment.