Research On Intelligent Prediction Method Of Dissolved Oxygen In Aquaculture

With the rapid development of aquaculture industry,strengthening the key technology research of aquaculture water quality prediction has become one of the key contents of fishery production.It can enhance the ability of disaster mitigation and prevention of aquaculture,and ensure the safety of aquaculture production.Dissolved oxygen(DO)is an important indicator of water quality in aquaculture ponds.Insufficient or excessive oxygen will adversely affect the living environment of fish.Establishing accurate and practical prediction model has important practical significance for aquaculture industry.At present,the application of Internet of Things and big data technology in predicting dissolved oxygen is still in the exploratory and trial stage.In order to make water quality prediction practical,the key is to improve the accuracy of prediction model,and the second is to realize online prediction.This paper takes dissolved oxygen(DO)in aquaculture as the research object,builds a multi-scale IoT system for aquaculture,establishes DO prediction model by machine learning method,and realizes fast response DO online prediction application.The main research contents of this paper are as follows:(1)In order to provide a comprehensive and accurate data source for DO prediction model,firstly,the process of DO change in aquaculture ponds was sorted out,and the relationship between DO and multi-scale environmental factors in space and time was analyzed.The water quality parameters needed to be monitored were obtained,and the data source of the prediction model was determined.Then multi-scale environmental information is obtained by NB-IoT technology,and three kinds of environmental information(water quality sensor data,water quality chemistry on-line detection data and meteorological data)are sent from NB module to IoT cloud service platform.Cloud platform equipment docking and northward design are realized,which provides multi-scale environmental sources for DO prediction.(2)Through autocorrelation analysis of dissolved oxygen time series,it is found that dissolved oxygen has strong non-linear,non-stationary and chaotic characteristics,which verifies the predictability of time series.The idea of "decomposition-prediction-reconstruction" is proposed,and a time series prediction model of dissolved oxygen(EEMD-LSSVM)is constructed.The model first decomposes the time series by ensemble empirical mode decomposition(EEMD).Then the decomposed components are analyzed by sample entropy method(SE),and the same qualitative components are merged.To reduce the number of models,three components,random,detail and trend,are reconstructed.The C-C method is used to reconstruct the phase space of the three components.The input and output vectors of the prediction model are constructed reasonably.Then,the parameters of least squares supportvector machine(LSSVM)are optimized by adaptive ant colony optimization(AACO),and the optimal prediction model structure is established to obtain the prediction results of three components.Finally,the final prediction results are obtained by BP neural network.Experiments show that the accuracy of AACO to optimize LSSVM is higher than that of particle swarm optimization to optimize LSSVM and artificial bee colony algorithm to optimize LSSVM.The prediction accuracy of EEMD-LSSVM is higher than that of EMD-LSSVM model,DWT-LSSVM model,BP model,ELM model,AMIMA model and standard LSSVM model.The RMSE,MAPE and MAE of the model are 0.1745,0.0074 and 0.1309 respectively,and the determination coefficient R2 is 0.9843.It has better prediction accuracy and generalization ability,and is an effective short-term prediction model for DO single factor time series.(3)High-end and refined aquaculture provides robust long-term forecasting.A novel multi-factor dissolved oxygen prediction model based on similar day clustering is proposed.Water temperature,pH,ammonia nitrogen and light were selected as the most relevant environmental factors of dissolved oxygen by grey correlation method.The K-means clustering method is used to select the similar day data with high similarity to the predicted day,and the optimal sample is obtained.EEMD decomposition,sample entropy merging,ELM prediction and BP recombination were used for combination prediction.The model is compared with the model without similar day clustering and all sample sets for prediction.The results show that the model has the highest fitting degree and the smallest error compared with other models(MF-SVM,MF-ELM and MF-LR).Further,the EEMD-LSSVM model of single-factor time series and this model are compared under the condition of constant weather and abrupt weather.It is found that the model is more suitable for long-term prediction,and has stronger generalization ability and better robustness for abrupt weather.(4)In order to realize the online practicality of dissolved oxygen prediction in aquaculture,an application design based on two-stage middleware response speed optimization was proposed.The first level middleware provides a unified,simple protocol format and highly reusable standardized interface for environmental data and commands of different devices.The second level middleware performs data analysis,extraction and packaging for DO prediction.Information is returned to the application system in a simple form.It greatly reduces the amount of database operation and the heavy data processing burden of the application system.The application system of Web and Android is realized.The mixed programming is carried out by using MATLAB and Java,and the dissolved oxygen prediction algorithm model is imported.The online practicality of dissolved oxygen prediction results is realized.