Data-driven modelling seeks to provide a mapping between the inputs and
outputs of a given system, with little prior process knowledge, and is emerging
as an attractive option for prediction and classification in water systems. Over
the next decade, advancements in the general progresses of ICT (hard and soft)
such as in data analytics, accelerated computing power and mega-networking
(already becoming available) will help solve the frustrating fragmentation of
scientific information in the field of water resources. The increase of CPU power
(massive parallel computing, cloud computing, etc.) extends the possibilities
of numerical and data-driven models and of 3-D/augmented reality displays, and
Web 2.0/3.0 opens up access to information sources to millions of new users.
New developments and products in the fields of micro-sensors, alternative power
supply and wireless telecoms all revolutionise the whole domain of real-time
monitoring and consequently real-time management.
Deep learning [20] is when big data intersects with machine learning. Techniques
allow the tackling of problems that exceed human understanding. Deep learning’s
important innovation is to have neural nets learn categories incrementally,
attempting to model lower-level categories (like letters) before attempting to acquire
higher-level categories (like words). Deep learning excels at this sort of problem and
has transformed the application of AI in the last decade. Their usage has been
directly facilitated by big data on top of algorithmic breakthroughs. Next-generation
formulations of deep artificial neural networks allow for the direct transition
from data to action. Such state-of-the-art algorithms include unsupervised feature
extraction as part of the data-driven learning. Deep learning technology has
the potential to leverage the power of big data and help develop an insight-driven
culture. Deep learning has been applied to visualising high-dimensional smart water
meter data [21]. T-SNE [19] is a technique that can be used for human-intuitive
(two-dimensional) visualisation of high-dimensional data. The parametric version
of t-SNE uses deep neural networks.
2 Data Analytics: Prediction
Predictive analytics involves using the patterns of past behaviour to predict
behaviour in the future. Predictive analytics encompasses a variety of statistical
techniques from predictive modelling, machine learning and data mining that
analyse current and historical facts to make predictions about future (or unknown)
events. Predicting future values of time series is one such application useful in
many water resource domains. Since there is a set of temporal ordered observations
for which (and working on the assumption that) there exist serial correlations
along the series, previous observations can be used to predict future values. The
task is essentially one of function approximation, i.e. to approximate the underlying
continuous valued function producing the time series.
Data Science Trends and Opportunities for Smart Water Utilities
9
outputs of a given system, with little prior process knowledge, and is emerging
as an attractive option for prediction and classification in water systems. Over
the next decade, advancements in the general progresses of ICT (hard and soft)
such as in data analytics, accelerated computing power and mega-networking
(already becoming available) will help solve the frustrating fragmentation of
scientific information in the field of water resources. The increase of CPU power
(massive parallel computing, cloud computing, etc.) extends the possibilities
of numerical and data-driven models and of 3-D/augmented reality displays, and
Web 2.0/3.0 opens up access to information sources to millions of new users.
New developments and products in the fields of micro-sensors, alternative power
supply and wireless telecoms all revolutionise the whole domain of real-time
monitoring and consequently real-time management.
Deep learning [20] is when big data intersects with machine learning. Techniques
allow the tackling of problems that exceed human understanding. Deep learning’s
important innovation is to have neural nets learn categories incrementally,
attempting to model lower-level categories (like letters) before attempting to acquire
higher-level categories (like words). Deep learning excels at this sort of problem and
has transformed the application of AI in the last decade. Their usage has been
directly facilitated by big data on top of algorithmic breakthroughs. Next-generation
formulations of deep artificial neural networks allow for the direct transition
from data to action. Such state-of-the-art algorithms include unsupervised feature
extraction as part of the data-driven learning. Deep learning technology has
the potential to leverage the power of big data and help develop an insight-driven
culture. Deep learning has been applied to visualising high-dimensional smart water
meter data [21]. T-SNE [19] is a technique that can be used for human-intuitive
(two-dimensional) visualisation of high-dimensional data. The parametric version
of t-SNE uses deep neural networks.
2 Data Analytics: Prediction
Predictive analytics involves using the patterns of past behaviour to predict
behaviour in the future. Predictive analytics encompasses a variety of statistical
techniques from predictive modelling, machine learning and data mining that
analyse current and historical facts to make predictions about future (or unknown)
events. Predicting future values of time series is one such application useful in
many water resource domains. Since there is a set of temporal ordered observations
for which (and working on the assumption that) there exist serial correlations
along the series, previous observations can be used to predict future values. The
task is essentially one of function approximation, i.e. to approximate the underlying
continuous valued function producing the time series.
Data Science Trends and Opportunities for Smart Water Utilities
9