A guide to Artificial Intelligence

Our CENTAUR® Gates technology uses AI to reduce urban flood risk. But what does AI really mean? We explain in layperson’s terms.

What is Artificial Intelligence?

Artificial intelligence, or AI, is often thought of as the emulation of human intelligence. This creation of a machine that has common sense (the ability to perceive, understand, and judge things) is known as artificial general intelligence, or AGI.

As a science, artificial intelligence – a term coined by John McCarthy in 1955 – is generally recognised as the ability of a computer programme to think and learn and to imitate the human condition.

Artificial intelligence is so-called because the intelligence does not belong to biological being, which would be known as natural intelligence (NI). Machine intelligence is a less common term but captures this distinction.

Types of AI such as fuzzy logic (FL) and artificial neural networks (ANN) allow us to make computers, via carefully designed and enacted mathematical algorithms, capable of specific tasks.  This is often referred to as narrow AI. The algorithm outputs can be translated into useful operational information to support decisions, or can directly enact decisions.

AI is useful where processes have recurrent patterns that are understood to be complex and variable; ANNs can be trained to learn patterns in these circumstances. Where human knowledge is available but processes are shrouded in variability, FL can be used to rationalise decisions.

There are three types of AI application in the wastewater industry:

Artificial Neural Networks

• Generally used to learn from “big data” and then to predict or detect events
• Generally used where phenomena are predictable but poorly understood or complex

Fuzzy Logic

• Generally used where expert knowledge is available, and
• Where phenomena are understood in principle but are variable

Genetic Algorithms

• Used for optimisation of decisions or systems
• Often used in combination with ANNs and FLs, often for their refinement

Throughout these blogs, we aim to consider where each of these types of AI are useful – alone or in combination – and where they have drawbacks and advantages in particular applications. We will consider the emergence of the different types of AI from academia and their practical application for business benefits.

What are the different types of artificial intelligence?

The main distinction here is between “weak AI”, more usefully termed “narrow AI”, and “strong AI”. This is a distinction between non-sentient machine intelligence used for a narrow task, and sentient machine intelligence with consciousness and mind. The latter remains hypothetical.

People have extended the “strong AI” definition by talking about “artificial general intelligence” (AGI), where a machine has the ability to apply intelligence to any problem. AGI is generally taken as meaning to be at least as smart as a human. Superintelligence is beyond this, referring to AI which surpasses the brightest human intelligence.

However, strong AI isn’t on the horizon and isn’t currently relevant to the water industry. Narrow AI is coming into use in the water industry, and is likely to come into further use in many more applications in the near future.

Approaches

Narrow AI can usefully be classified by approach, and this tends to be what is done in the water industry. The approaches commonly used in the water industry are all branches of soft computing, which can be broken down at a high level as follows:

Machine Learning

• The field of computer science that gives computers the ability to learn without being explicitly programmed (Koza et al, 1996)
• Includes artificial neural networks

Fuzzy Logic

• The analysis of analogue input values in terms of logical variables that take on continuous values between 0 and 1, in contrast to classical or digital logic, which operates on discrete values of either 1 or 0 (true or false, respectively) (Pedrycz, 1993, Hájek, 1998)
• Or, logic to deal with concepts that cannot be expressed as the “true” or “false” but rather as “partially true”

Evolutionary Computation

• Optimisation algorithms inspired by biological evolution; they use candidate solutions in population-based trial-and-error computations.
• These algorithms commonly include genetic algorithms, ant colony, and particle swarm optimisations

Probabilistic Methods

• Probabilistic models such as Bayesian networks and Markov chain models have been applied to water industry phenomena such as pipe failure
• However, this is less recognised as AI and is more traditionally recognised as applied statistics

“Chaos Theory” and “Creative Computing” are the other branches of soft computing, where we are focussing on AI in the water industry.

Before considering the forms of narrow AI above-identified in the water industry, it is worth considering how we encounter them in our everyday lives:

Artificial Neural Networks

• Speech recognition commonly uses artificial neural networks (and Markov chain) algorithms in their operation
• Image recognition for medical diagnostics and smartphone apps that can recognise everyday objects are based on artificial neural networks

Fuzzy Logic

• Automated gear transmission systems use several variables (speed, acceleration, throttle opening, rate of change of throttle opening, engine load) and weight each of these in a fuzzy aggregate to decide on gear changes
• Antilock brakes, with wheel circumferential speed and vehicle speed as input variables
• Interpretation of handwriting
• Washing machines, which sense the load size, detergent amount and then track water turbidity and make decisions on this
• Television: ambient lighting, time of day and so on to adjust parameters such as screen brightness, colour, contrast and sound
• Criminal search systems: combining photo and descriptive analysis (tall, young)
• CGI – huge scale armies created to have random yet orderly movements

Genetic Algorithms

• Mostly used “off-line” in design applications – Brainz published a useful list of 15 applications: https://www.brainz.org/15-real-world-applications-genetic-algorithms/
• g. aircraft wing design
• g. optimising trading strategies
• g. product design to fit the market

Understanding how and where AI can be applied is important. There is no doubt that efficiency and benefits can be achieved.

Application of Artificial Neural Networks (ANNs) in Wastewater

ANNs are generally used to learn from ‘big data’ and then to predict or detect events. They are used where phenomena are poorly understood or very complex but obey patterns.

Data needs to be available of a sufficient extent and quality that an algorithm can be designed to learn from it and to sufficiently generalise within the overall landscape of possibilities. A downfall of ANNs, therefore, can be insufficient data, or a bias towards data which focuses on specific areas of the overall landscape. A further downfall can be a change in an aspect of a system which is not picked up in the original input data.

ANNs have been used to replicate the behaviour of complex models which are too slow to run real-time application, for example in urban flood prediction.

ANNs are being used to tackle emerging climate related problems. Urban flooding is an increasing prevalent side-effect of climate change and urbanisation. Hence, the focus on modelling has increased. ANNs have been used to model and predict water levels, flow rates and flood volumes in both fluvial environments and in wastewater networks.

Application of Fuzzy Logic in Wastewater

In wastewaster, FL has been used where expert knowledge is available, and where phenomena are understood in principle, but the outputs are variable.

A study conducted by Pedrycz (1993) observed that although alternative approaches such as GAs and ANNs can perform just as well as FL in many cases, FL has the advantage that the solution to the problem can be cast in terms that human operators can understand, so that their experience can be used in the design of the controller. This makes it easier to mechanise tasks that are already successfully performed by humans.

For these reasons, FL is finding many applications in wastewater, especially those involving an element of control.

In wastewater, FL has been widely used in control applications, e.g. pump station control and optimisation of energy, control of additives in treatment, control of activated sludge plants, energy saving in the aeration process, in-line control of non-linear pH neutralisation, optimisation of nitrogen removal and aeration energy consumption in wastewater treatment plants. FL has also been used in blockage detection, state estimation in anaerobic wastewater treatment, CSO performance optimisation and management in near-real-time.

Application of Genetic Algorithms in Wastewater

GAs are generally used off-line in the optimisation of decisions or systems. This might be in model calibration, or in asset investment optimisation looking over extended time periods.

For network design and rehabilitation planning, much of this began in water distribution networks and progressed then onto wastewater networks and then from infrastructure to non-infrastructure. GAs are now in common use in asset investment planning in the water industry.

A study by Mackle et al. (1995) used a simple GA for pump scheduling in a water supply system; their objective was to minimise the costs of pumping by taking advantage of low-cost electricity tariffs and additional storage in the system. The GA was necessarily used offline but the advantages coming from them were built into operational policies.

GAs have been used in combination with ANNs for applications in wastewater systems control by Hadja et al. (1998). Since GAs are likely to perform too slowly for online application, ANNs were trained to approximate GA results.

GAs are now commonly used in modelling packages for optimisation problems, for example in looking at management strategies for large networks.

To realise the benefits of AI, we need to internalise the power in better products and more robust platforms which have been designed to guard against the downsides of AI. These downsides include the inability to communicate effectively.

How does CENTAUR® fit into this?

CENTAUR® is an intelligent autonomous system for local urban flood risk reduction. It utilises untapped network capacity. It does this through the operation of a gate to control flow based on an intelligent algorithm which leverages local water level data. CENTAUR® is a self-managing, easily deployed system, which can be less costly than capital and space intensive solutions.

In beta format, it is operational in Coimbra in Portugal, contributing to flood protection of a World Heritage Site. Further implementations are planned in France in parallel to full market launch.

The control intelligence for CENTAUR® is a fuzzy logic algorithm. Genetic Algorithms are being trialled off-line in the optimisation of the FL algorithm.

It could be said that the increased network handling capacity and financial benefits (capital avoided) come from the algorithm. However, without the robustness of monitoring and communication technologies as the enablers, the effective application of AI would not be possible.

Much of the design of the system has been focussed on reliability and fail-safes. The system uses specially engineered communications to guarantee signal without latency. An online dashboard connects to the system hub and gives operational visibility. Although the dashboard isn’t necessary for the operation of the system, it introduces convenience features. The gate technology is purpose-designed for the application and for easy deployment. It has physical fail-safes in the form of overflows which keep upstream risk to an absolute minimum. Sensors are designed to give reliable data at low power, with special installation techniques to avoid in-manhole problems. However, if communication or sensor failures should occur the system remains safe.

Although not necessary for the autonomous operation of the system, an online dashboard connects to the system Hub and provides operational visibility.

Technology can guard against any unintended consequences of AI. The design and application diligence are as important as the AI itself. With this, it is important to note that expertise in modelling, hardware, communication and software design from across the EU in a Horizon 2020 project.

References

Hájek, P. (1998). Metamathematics of fuzzy logic (4 ed.). Springer Science & Business Media.

Koza, J.R., Bennet, F.H., Andre, D., Keane, M.A. (1996). Automated Design of Both the Topology and Sizing of Analog Electrical Circuits Using Genetic Programming. Artificial Intelligence in Design ’96. Springer, Dordrecht. pp. 151-170

Pedrycz, W. (1993). Fuzzy Control and Fuzzy Systems (2 ed.). Research Studies Press Ltd

Hadja, P., Novotny, V., Feng, X. & Yang, R. (1998). Simple feedback logic, genetic algorithms, and artificial neural networks for real-time control of a collection system. Water and Science Technology, 38(3), 187-195

Mackle, G., Savic, D.A. & Walters, G.A. (1995) Application of genetic algorithms to pump scheduling for water supply. In Proc. of Genetic Algorithms in Engineering Systems: Innovations and Applications, GALESIA’95. IEE Conference Publication 414, 400-405

Pedrycz, W. (1993). Fuzzy Control and Fuzzy Systems (2 ed.). Research Studies Press ltd.

Acknowledgement

The CENTAUR® project in the case study has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 641931.