Analysis And Prediction Of Dissolved Oxygen Concentration Changes In Lakes Based On PSO-LSTM

Dissolved oxygen is a crucial environmental indicator in water bodies,which has an important impact on the survival and development of aquatic organisms.Improving the accuracy of forecasting dissolved oxygen can play an early warning role in managing water pollution,discovering potential pollution events in time and taking preventive measures to minimize the impact.This paper proposes a Long Short-Term Memory(LSTM)model optimized by Particle Swarm Optimization(PSO),referred to as PSO-LSTM,and uses this method to predict the Spatiotemporal variation of lake dissolved oxygen in Eagle Mountain Lake(EML).The method uses both vertical and horizontal data to predict future dissolved oxygen levels.Experimental results show that PSO-LSTM predicts better than other methods.The main work and conclusions of this paper are as follows:(1)By analyzing the temporal and spatial characteristics of 21 monitoring points in the vertical profile of Eagle Mountain Lake measured and recorded every two hours from April25 to October 29,2019,it was found that dissolved oxygen(Dissolved Oxygen,DO)The highest point is often on the lake surface,and the lowest point of DO is often at the bottom of the lake,which is related to the oxygen exchange in the surface water body.The phenomenon of dissolved oxygen stratification decreased from October,and the dissolved oxygen stratification was obvious in other months.The dissolved oxygen content is the lowest in August and September,and the average dissolved oxygen content in April is the highest,which is related to temperature and light.In early summer,most of the dissolved oxygen concentration ranges around 8mg/L,and the dissolved oxygen at the lake bottom and lake surface changes greatly within a day;in summer,the dissolved oxygen concentration is around 4mg/L,and the dissolved oxygen at the lake bottom and lake surface still changes the most;The concentration of dissolved oxygen is evenly distributed between6-8mg/L,and the solubility curve is close to a straight line,that is,the upper and lower concentrations are consistent,indicating that the dissolved oxygen in the lake is evenly mixed at this time.Under the influence of wind and waves,vertical mixing occurs,which is conducive to oxygen supplementation.In early summer and autumn,dissolved oxygen levels are generally lower.(2)Through the analysis of the temporal and spatial characteristics of the monthly data of six monitoring points in the horizontal profile of Eagle Mountain Lake from January 1987to February 2022,the annual average data range is 7-10mg/L,1998,2010,2019 The dissolved oxygen concentration at all monitoring points increased in 2010,and tended to be stable at other times.According to the monthly average analysis,the average dissolved oxygen in June,July,and August is the lowest.The monthly average dissolved oxygen has a wide range of changes and is more affected by seasons.(3)Prediction of dissolved oxygen concentration in lakes by PSO-LSTM model.In the comparative experiment,four models including BP neural network,support vector regression,and LSTM were constructed to predict the dissolved oxygen concentration.The prediction accuracy of the model shows that the prediction effect of PSO-LSTM is better on different spatial scales.In vertical water bodies,the PSO-LSTM model exhibited higher fitting accuracy than the LSTM,BP and SVR models,with a 14%improvement in~2.In addition,in the future prediction task,the PSO-LSTM model outperforms the LSTM,BP and SVR models,with significant improvements in RMSE and MAE metrics,and a slight improvement in~2.Compared to the BP model,the PSO-LSTM model exhibits more reliable error results for future predictions.At the same time,compared with the SVR model,the PSO-LSTM model has achieved a large improvement in RMSE and MAE indicators,and a small improvement in~2.In horizontal water bodies,the PSO-LSTM model performed the best among the four prediction models,with the smallest RMSE and MAE values and the highest~2 value.Compared with LSTM,BP and SVR models,the PSO-LSTM model shows significant improvement in all three metrics.Overall,compared with traditional models,the PSO-LSTM model exhibits superior performance and accuracy in the prediction of dissolved oxygen concentration,providing scientific support for water quality prediction and watershed management.In summary,compared with other methods,the new model proposed in this paper can more accurately predict the future changes in dissolved oxygen concentration,and it also has advantages in different space and time scales,providing a basis for effective protection and governance of water quality sources.Reliable data guidance.