Abstract
The aquaponic system can satisfy the different needs of aquatic products and plants by utilizing nutrient flow to improve the economic benefit. Nitrogen is a key nutrient element in such process. High nitrogen concentration can worsen water quality, which may further result in mass death of typical fish and/or shellfish. Therefore, it is particularly important to monitor the nitrogen concentration to manage aquatic products and plants growth in aquaponic systems. Sensors for measuring nitrogen concentration are commercially available, but they are often expensive and unreliable in service. At the same time, the research on the soft sensor of nitrogen concentration is very limited. Therefore, this paper proposes a new adaptive filtering-based soft sensor method for real-time estimation of total nitrogen concentration. This method provides accurate prediction by integrating mechanistic model, online measurements (e.g. fish biomass, temperature) and infrequent offline measurement of total nitrogen. A moving-horizon estimation (MHE) algorithm is used for joint state and parameter estimation, thus allowing correction of mismatch between model parameters and the real process. Furthermore, the appropriate frequency of offline nitrogen measurement is determined to balance between the soft sensor accuracy and cost. Through computer simulation study of an aquaponic system, the proposed approach is effective to provide promising real-time prediction performance. By applying the correction method, the RMSE (root-mean-square error) of nitrogen concentration estimation is reduced by 31.86% on average compared to the simulated practical situation. In conclusion, the proposed soft sensor method could provide useful monitoring of the total nitrogen in aquaponics, and such information could be used for optimizing the operations.