The Smart Predicting of Algal Concentration for Safer Drinking Water Production with Sensor Data
(2023) In Sensors 23(11).- Abstract
To better predict the timely variation of algal blooms and other vital factors for safer drinking water production, a new AI scanning–focusing process was investigated for improving the simulation and prediction of algae counts. With a feedforward neural network (FNN) as a base, nerve cell numbers in the hidden layer and the permutation and combination of factors, etc., were fully scanned to select the best models and highly correlated factors. All the factors involved in the modeling and selection included the date (year/month/day), sensor data (temperature, pH, conductivity, turbidity, UV254-dissolved organic matter, etc.), lab measurements (algae concentration) and calculated CO2 concentration. The new AI scanning–focusing... (More)
To better predict the timely variation of algal blooms and other vital factors for safer drinking water production, a new AI scanning–focusing process was investigated for improving the simulation and prediction of algae counts. With a feedforward neural network (FNN) as a base, nerve cell numbers in the hidden layer and the permutation and combination of factors, etc., were fully scanned to select the best models and highly correlated factors. All the factors involved in the modeling and selection included the date (year/month/day), sensor data (temperature, pH, conductivity, turbidity, UV254-dissolved organic matter, etc.), lab measurements (algae concentration) and calculated CO2 concentration. The new AI scanning–focusing process resulted in the best models with the most suitable key factors, which are named closed systems. In this case study, models with highest prediction performance are the (1) date–algae–temperature–pH (DATH) and (2) date–algae–temperature–CO2 (DATC) systems. After the model selection process, the best models from both DATH and DATC were used to compare the other two methods in the modeling simulation process: the simple traditional neural network method (SP), where only date and target factor as inputs, and a blind AI training process (BP), which considers all available factors as inputs. Validation results show that all methods except BP had comparable results for algae prediction and other water quality factors, such as temperature, pH and CO2, among which DATC displayed an obviously poorer performance through curve fitting with original CO2 data compared to that of SP. Therefore, DATH and SP were selected for the application test, where DATH outperformed SP due to the uncompromised performance after a long training period. Our AI scanning–focusing process and model selection showed the potential for improving water quality prediction by identifying the most suitable factors. This provides a new method to be considered in the enhancing of numerical prediction for the factors in water quality prediction and broader environment-related areas.
(Less)
- author
- Yu, Han LU ; Li, Jing LU ; Holmer, Linda and Köhler, Stephan J.
- organization
- publishing date
- 2023
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- AI, algal bloom, closed system, drinking water supply, lake, neural network
- in
- Sensors
- volume
- 23
- issue
- 11
- article number
- 5151
- publisher
- MDPI AG
- external identifiers
-
- pmid:37299878
- scopus:85161576222
- ISSN
- 1424-8220
- DOI
- 10.3390/s23115151
- language
- English
- LU publication?
- yes
- id
- 06b7f0f3-6dcd-4e97-9ea6-bb1e4df6bf80
- date added to LUP
- 2023-08-16 15:44:46
- date last changed
- 2024-04-20 00:49:51
@article{06b7f0f3-6dcd-4e97-9ea6-bb1e4df6bf80,
abstract = {{<p>To better predict the timely variation of algal blooms and other vital factors for safer drinking water production, a new AI scanning–focusing process was investigated for improving the simulation and prediction of algae counts. With a feedforward neural network (FNN) as a base, nerve cell numbers in the hidden layer and the permutation and combination of factors, etc., were fully scanned to select the best models and highly correlated factors. All the factors involved in the modeling and selection included the date (year/month/day), sensor data (temperature, pH, conductivity, turbidity, UV254-dissolved organic matter, etc.), lab measurements (algae concentration) and calculated CO<sub>2</sub> concentration. The new AI scanning–focusing process resulted in the best models with the most suitable key factors, which are named closed systems. In this case study, models with highest prediction performance are the (1) date–algae–temperature–pH (DATH) and (2) date–algae–temperature–CO<sub>2</sub> (DATC) systems. After the model selection process, the best models from both DATH and DATC were used to compare the other two methods in the modeling simulation process: the simple traditional neural network method (SP), where only date and target factor as inputs, and a blind AI training process (BP), which considers all available factors as inputs. Validation results show that all methods except BP had comparable results for algae prediction and other water quality factors, such as temperature, pH and CO<sub>2</sub>, among which DATC displayed an obviously poorer performance through curve fitting with original CO<sub>2</sub> data compared to that of SP. Therefore, DATH and SP were selected for the application test, where DATH outperformed SP due to the uncompromised performance after a long training period. Our AI scanning–focusing process and model selection showed the potential for improving water quality prediction by identifying the most suitable factors. This provides a new method to be considered in the enhancing of numerical prediction for the factors in water quality prediction and broader environment-related areas.</p>}},
author = {{Yu, Han and Li, Jing and Holmer, Linda and Köhler, Stephan J.}},
issn = {{1424-8220}},
keywords = {{AI; algal bloom; closed system; drinking water supply; lake; neural network}},
language = {{eng}},
number = {{11}},
publisher = {{MDPI AG}},
series = {{Sensors}},
title = {{The Smart Predicting of Algal Concentration for Safer Drinking Water Production with Sensor Data}},
url = {{http://dx.doi.org/10.3390/s23115151}},
doi = {{10.3390/s23115151}},
volume = {{23}},
year = {{2023}},
}