Causal-Guided HyperMixer Prediction Model for Aquaponic Water-Environment Diagnosis
Authors
Files
Abstract
This paper presents a causal-guided HyperMixer prediction model for aquaponic water-environment diagnosis. To identify abnormal water conditions in the combined fish-plant circulation system early and prevent harm to the health of fish, nitrification stability, plant nutrient absorption, etc., disruption of regular farm operations should be avoided. The method builds multi-source observation windows by collecting dissolved oxygen, pH, ammonia nitrogen, nitrite, nitrate, temperature, electrical conductivity, turbidity, oxidation-reduction potential, feeding intensity, pump status, aeration state, water-return flow and zone metadata from fish tanks, hydroponic beds, biofilters and sump tanks. A causal screening branch reduces misleading weights caused by routine feeding, short sensor gaps, cleaning events, pump cycles, temporary tank disturbance and local root blockage. The HyperMixer branch then conducts a light-weight temporal and channel interaction of the water-quality variables to interpret oxygen decline, ammonia rise, temperature fluctuation, pH drift and circulation instability simultaneously. Conduct experiments on a simulated aquaponics monitoring dataset to perform multi-zone, cross-unit, ablation and disturbance tests. Compared with the previous machine-learning, recurrent, convolutional and mixer baselines, the proposed model has improved diagnosis accuracy, high-risk recall, warning precision and false-review control at a relatively low computational cost for edge deployment. The output is a staff review card that contains a diagnostic label, risk level, affected area, main causal factor, confidence level and suggested inspection action. Therefore, the model is to be used as an aquaponics decision-support tool for warning prioritisation and cause analysis, not as an autonomous water-control command.