Soft Sensing Of Ammonia Nitrogen Concentration In Aquaculture Based On Stochastic Configuration Networks And Its Improvement

To solve the problems of high cost,high consumption,and difficulty in real-time and effective detection of the concentration of ammonia nitrogen in the water body during the intensive seawater aquaculture process,based on the relevant water quality parameters collected during the cultivation of turbot in the laboratory intensive circulating seawater aquaculture control system,this paper analyzes the relevant variables that affect the concentration of ammonia nitrogen in the water body.According to the requirements for easy measurement of auxiliary variables,combined with actual conditions,the temperature,dissolved oxygen,conductivity and pH of the aquaculture water body obtained by the intensive seawater circulation control system in the turbot culture process were selected as auxiliary variables.Establish a corresponding soft measurement model of ammonia nitrogen concentration,and then realize real-time and effective monitoring of ammonia nitrogen concentration in aquaculture water.By analyzing and comparing the characteristics and deficiencies of the algorithms of BP network,random vector function link(RVFL)and stochastic configuration networks(SCNs),the use of stochastic configuration networks have fast learning ability and high approximation performance,the stochastic configuration networks are introduced into the soft measurement process of ammonia nitrogen concentration in intensive aquaculture water.To avoid the influence of the randomness of SCNs parameter selection on the model measurement performance,a genetic algorithm(GA)optimized SCNs soft sensor model was proposed.The model optimizes the preselected weights and threshold matrix of SCNs based on the genetic algorithm and uses the optimized preselected matrix for the construction of the SCNs model.Then based on the relevant water quality parameters obtained from the turbot aquaculture based on the intensive circulating seawater aquaculture system,the constructed GA-SCNs model is used to measure the concentration of ammonia nitrogen in the aquaculture water,and the measurement results are compared with the SCNs,BP,RVFL models for the measurement of the concentration of ammonia nitrogen in the water.The experimental results show that the GA-SCNs model has a longer running time than other models,but it has a better measurement effect.In addition,because bagging can effectively reduce the variance of the integrated model while maintaining the model deviation,and the advantages of effectively improving the performance of the unstable algorithm,a soft measurement modeling method for bagging integrated stochastic configuration networks are proposed.This method selects SCNs with fast learning speed and strong approximation as the base learner,uses bootstrap to generate multiple different training subsets,trains multiple different base learners SCNs,and then takes the output of each SCNs model.The average value is used as the output of the Bagging-SCNs model.Based on the relevant water quality parameters obtained during the turbot breeding in the lab intensive cyclic seawater control system,the soft-sense modeling of Bagging-SCNs,SCNs,RVFL,and Bagging-RVFL was performed for 20 consecutive times,and statistical analysis was performed on the output of each model.Comparing the average of the root mean square error(RMSE),the maximum absolute error(MAE),and the average absolute percentage error(MAPE)of the predicted output of different models,it is shown that Bagging-SCNs has better measurement accuracy and higher stability,and it can avoid the problem that the running time of the model is significantly increased when genetic algorithm optimizes the SCNs of network parameters.The validity of the proposed Bagging-SCNs model in measuring ammonia nitrogen concentration in intensive marine aquaculture water is further verified,which has certain guiding significance for monitoring aquaculture water bodies.