Establishment Of A Water "Three Nitrogens" Concentrations Prediction Model Basing On Stacked Autoencoder-BP Neural Network

Ammonia nitrogen,nitrite nitrogen and nitrate nitrogen in water are commonly named“three nitrogens”.Among the three nitrogens,the ammonia nitrogen and nitrite nitrogen are highly toxic to the aquatic animals.Therefore,monitoring the concentrations of the ammonia nitrogen and nitrite nitrogen is very critical for the culture of aquatic animals.There are several drawbacks of the traditional methods which are used to measure the concentrations of the“three nitrogens”,such as high cost,labourous,time cost etc.,which hinder their widely application in aquaculture.Therefore,it is an urgent need to develop another novel method to predict the concentrations of the“three nitrogens”in water.In this report,we focused our studies on the development of novel prediction method.The obtained results are as follows:1.The establishment of online water monitoring system.We established an online water monitoring system in our laboratory,so that the temperature,pH value,dissolved oxygen and oxidation-reduction potential were recorded from the water in tanks.Meanwhile,the actual concentrations of the“three nitrogens”in water were measured using traditional methods.2.The collection of the concentrations of the“three nitrogens”.The actual concentration of the ammonia nitrogen,nitrite nitrogen and nitrate nitrogen in water was determined using Nessler's reagent dectrophotometry,alpha naphthalene colorimetric method,and dual-wavelength ultraviolet spectrophotometry,respectively.Basing on the analysis of the data of the actual concentrations of“three nitrogens”,pH value,temperature,oxidation-reduction potential,dissolved oxygen,the results showed that when the pH value was increased,the oxidation-reduction potential was decreased.While when the temperature was decreased,the dissolved oxygen was enhanced,indicating that there were obvious effects of pH value and temperature on the concentrations of the“three nitrogens”.3.The establishment of the prediction models for the“three nitrogens”.To establish the prediction models for the“three nitrogens”,the key point was to figure out the relationship among the concentrations of the“three nitrogens”and the water parameters.The data after pretreatment were used as the original data which were used for SAE neural network training.Thereafter,unsupervised greed training method was applied.The learnt characteristics were used for the supervision and training of BP neural network.The model was optimized using the back propagation(BP)algorithm.The results showed that the prediction model R~2 of the nitrite nitrogen after training was 0.95,and root mean square error of the prediction(RMSEP)was 0.099709,indicating that the model could accurately predict the concentration of nitrate nitrogen in water.While the prediction model R~2 of the ammonia nitrogen after training was 0.86757,and RMSEP was0.1405,suggesting that the model could betterly predict the concentration of ammonia nitrogen in water.By contrast,the prediction model R~2 of the nitrate nitrogen after training was 0.71824,and RMSEP was 0.29729,indicating that the model could not accurately predict the concentration of nitrate nitrogen in water.The established model will pave a new way for developing online system for monitoring the water“three nitrogens”concentrations in the future.