Font Size: a A A

Research On Soft-sensing Methods Of Nutrients Concentration In Coastal Waters

Posted on:2023-02-15Degree:MasterType:Thesis
Country:ChinaCandidate:X Y WangFull Text:PDF
GTID:2530306617469974Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Nutrients in seawater are the basic factors of marine primary productivity,of which concentrations are closely related to plankton activities.Insufficient nutrients limit the growth of plankton,whereas excessive nutrients cause water eutrophication,thereby leading to ecological hazards.Therefore,monitoring of nutrient concentration is of great significance to maintaining the ecological environment and preventing disasters of marine.However,the current regular collection of nutrient concentration data in coastal waters generally has high cost,and it comes with problems like chemical reagent contamination and sampling dangers.To this end,this thesis proposes a soft-sensing method for determining nutrient concentration in coastal waters,in which machine learning methods are used to model the concentration rule between the target nutrient factor and other related water quality factors,the concentration of target factor is calculated using measured data of other water quality factors.In this thesis,dissolved inorganic nitrogen(DIN)and dissolved silicate(DSi)are taken as the research objects.We use the water quality monitoring dataset in coastal waters of Weihai and comprehensively consider the associated features,such as spatiotemporal,chemical,physical,meteorological and hydrological features,as well as land-based information,and then establish soft-sensing models of nutrient concentration based on the XGBoost model.Due to the high expense of monitoring data in coastal waters,we also propose a data consumption analysis approach.The soft-sensing models are multi-dimensionally observed and compared in terms of the data volume required for model training in order to find out more efficient softsensing models and more cost-effective training data volume.Data consumption analysis is particularly useful in high-cost scenarios.The main works of this thesis are summarized as follows.First,we preprocess the water quality monitoring data through a series of steps,which include the removal of direct correlation factors,feature simplification and feature combination,data normalization and associated feature selection.Second,we propose DIN and DSi softsensing methods based on XGBoost.We first group the monitoring stations by clustering technology thus the study region is divided into many sub-regions.And then we build softsensing models on each sub-region and improve the performance of the soft-sensing method through two key steps,which are distance features supplement and direct correlation factors supplement.Third,we analyze data consumption of soft-sensing models of DIN and DSi in coastal waters.A data consumption cube model is proposed to analyze the amount of training data required by soft-sensing model from three dimensions of time,spatial and characteristic.The time dimension refers to the time span of collecting data set,the spatial dimension is the spatial extent of monitoring area and the feature dimension is the number of characteristic factors collected by a monitoring station.The experiments verified the effectiveness of the proposed soft-sensing methods of nutrients concentration in coastal waters.Additionally,by the data consumption analysis,the most cost-effective time span,spatial extent and numbers of characteristic factors of softsensing model are also presented.
Keywords/Search Tags:Soft-sensing method, Nutrients, Data consumption analysis, XGBoost, Machine learning
PDF Full Text Request
Related items