基 于“AI+半 导 体 光 催 化”的 特 种 鱼 养 殖 水 体 绿 色 原 位 治 理 与智 能 调 控 方 案 研 究

Facing issues such as water quality fluctuations，empirical medication practices，extensive on-site control，and the simultaneous pressure to meet effluent standards in localized northern breeding of high-value specialty fish from the south，this paper aims at stable control of aquaculture water bodies and green management. Using multi-source data integration，spatiotemporal prediction，constrained decision control，and in-situ semiconductor photocatalytic purification methods，the study investigates water body state identification，risk warning，control decision-making，and pollutant reduction in specialty fish farming. A multi-source database covering water quality，meteorology，and fish growth stages was constructed to perform rolling predictions of dissolved oxygen，ammonia nitrogen，pH，temperature，and residual drug risks. Based on these predictions，control recommendations were generated considering fish physiological thresholds，equipment capacity，medication limits，and operational costs. The results show that：1）Multi-source sensing improves the continuity and integrity of water quality state identification，compensating for the shortcomings of manual pond inspection and single-point sampling；2）The CNN-Transformer hybrid model captures short-term water quality fluctuations and medium-to-long-term trends，providing early warnings for low dissolved oxygen，ammonia accumulation，and medication residue risks；3）Constrained decision-making can translate predictions into recommendations for feeding，oxygenation，water exchange，medication，and operation of purification units，while the broad-spectrum nano-composite photocatalytic network offers a low-energy pathway for reducing antibiotic residues and certain organic pollutants. The study demonstrates that the collaborative application of intelligent prediction，on-site control，and photocatalytic purification can provide technical support for water management and effluent reduction in high-value specialty fish farming.