in interpreting the risk of iron failures [11]; in relating water quality and age in
drinking water [12]; in exploring the rate of discolouration material accumulation
in drinking water [13]; and for geospatial burst behaviour [14]. The SOM is an
unsupervised ANN model which resembles the way biological brain maps
spatially order their responses by modelling those self-organising and adaptive
learning features of the brain [15]. Trained vectors are positioned on a regular
low-dimensional grid in a spatially ordered fashion hence facilitating improved
visualisation, readily enabling presentation and interpretation. SOMs are noise
tolerant; this property is highly desirable when sparse data are used, and thus
there are many potential applications in the water industry. The SOM (see example
in Fig. 3) contains colour-coded hexagons that summarise all of the component
planes that represent individual variables. Each hexagonal cell represents individual
neurons, which are the mathematical linkages between the input and output
layers. In the component planes for individual variables, the colouring or shading
corresponds to actual numerical values for the input variables that are referenced
in the scale bars adjacent to each plot. Blue shades show low values, and red
corresponds to high values. Visual inspection and comparison of the component
planes allow examination of how variables vary against each other. Figure 3
shows SOM analysis of data from a water quality sensor monitoring in a test loop
facility [16] for a 28-day period measuring water quality parameters every minute.
While absolute accuracy was uncertain, the way in which these parameters are
related in a time invariant fashion can be displayed by the component planes of
Fig. 2 Vision of an integrated future
Data Science Trends and Opportunities for Smart Water Utilities
7
drinking water [12]; in exploring the rate of discolouration material accumulation
in drinking water [13]; and for geospatial burst behaviour [14]. The SOM is an
unsupervised ANN model which resembles the way biological brain maps
spatially order their responses by modelling those self-organising and adaptive
learning features of the brain [15]. Trained vectors are positioned on a regular
low-dimensional grid in a spatially ordered fashion hence facilitating improved
visualisation, readily enabling presentation and interpretation. SOMs are noise
tolerant; this property is highly desirable when sparse data are used, and thus
there are many potential applications in the water industry. The SOM (see example
in Fig. 3) contains colour-coded hexagons that summarise all of the component
planes that represent individual variables. Each hexagonal cell represents individual
neurons, which are the mathematical linkages between the input and output
layers. In the component planes for individual variables, the colouring or shading
corresponds to actual numerical values for the input variables that are referenced
in the scale bars adjacent to each plot. Blue shades show low values, and red
corresponds to high values. Visual inspection and comparison of the component
planes allow examination of how variables vary against each other. Figure 3
shows SOM analysis of data from a water quality sensor monitoring in a test loop
facility [16] for a 28-day period measuring water quality parameters every minute.
While absolute accuracy was uncertain, the way in which these parameters are
related in a time invariant fashion can be displayed by the component planes of
Fig. 2 Vision of an integrated future
Data Science Trends and Opportunities for Smart Water Utilities
7