Soft-sensing Of Ammonia-nitrogen Concentration In Sea Cucumber Culture Water Quality Based On Data Expansion And Migration Learning

In the process of intensive sea cucumber aquaculture,water quality is an important factor affecting the profitability of aquaculture,and ammonia nitrogen concentration,as one of the important parameters to measure water quality,is very important in the monitoring of aquaculture water quality.In the breeding process,the high concentration of ammonia nitrogen will have a bad impact on the living environment of sea cucumbers,and cause damage to the surrounding water quality and ecology,the impact on the water quality deterioration of intensive and high-density aquaculture at this stage is more severe.Therefore,in the process of intensive sea cucumber aquaculture,real-time monitoring of ammonia nitrogen concentration in water quality is particularly necessary.However,in the actual process of breeding sea cucumbers,the determination of ammonia nitrogen concentration in aquaculture mostly relies on the methods such as artificial sampling and chemical analysis,and the current water quality status cannot be obtained quickly.Soft sensing technology can realize real-time monitoring of ammonia nitrogen concentration through the analysis and modeling of other related parameters of aquaculture water quality.However,in the process of model training,insufficient number of samples can easily lead to overfitting.High-quality data sets can be used to improve the model,especially the training process of the learning model.Therefore,expanding the existing data set is an effective method to prevent the occurrence of overfitting.However,the particularity of ammonia nitrogen measurement makes manual collection of water quality batch detection and labeling very labor-intensive and material resources.In contrast,automatic generation of data samples is more feasible.For this reason,in order to realize the real-time monitoring of ammonia nitrogen concentration during sea cucumber culture,the following methods are proposed:1.In order to solve the problem of insufficient samples of water quality factors in ammonia nitrogen soft sensing modeling of aquaculture water quality and can not meet the requirements of high-quality modeling,a data augmentation algorithm based on improved tablegan is proposed.The data generation algorithm based on generator,classifier and discriminator is designed and implemented,and the public power consumption data set is augmented and compared with soft sensing modeling by using this augmented algorithm.The results show that the augmented data set is based on back propagation neural network(BP),radial basis function(RBF),stochastic configuration networks(SCN)In the bagging SCN modeling method,the test accuracy is about176.9 lower than the root mean square error of the original data on the test set,which proves the effectiveness of the improved tablegan data augmentation algorithm.2.In order to ensure the diversity of modeling samples,considering that the influencing factors of ammonia nitrogen concentration in sea cucumber culture are similar to those in turbot culture,the ammonia nitrogen concentration model in turbot culture is used for sea cucumber culture modeling by using the migration learning method based on the previous results of our experimental group in ammonia nitrogen soft measurement of turbot culture water quality.With the advantage of deep learning in feature extraction,a DeepSCN based modeling method of ammonia nitrogen concentration in aquaculture water quality is proposed.The turbot data set expanded by TableGAN is used for DeepSCN modeling experiment,which is transferred to the sea cucumber breeding process,and the weight of similar data is increased by eliminating dissolved oxygen to improve the migration effect.In addition,in order to solve the possible over fitting problem caused by the data generation model,early stopping is added to the sea cucumber data generation model to maintain the stability of the generated data,and the soft sensing model is pre-trained and fine-tuned.The results show that compared with the SCN and Bagging-SCN ammonia nitrogen soft sensing models previously proposed by our group,the model test accuracy based on data expansion and DeepSCN is higher,which can effectively improve the quality of soft sensing modeling,By comparing the root mean square error and model training time between the migration model and the direct augmented model,the migration models have better effects.The root mean square error of the test set is reduced by 0.00339 and the training time is shortened by about 15.7s.It has guiding significance for the regulation of water quality in the process of sea cucumber breeding.