OK, we all get it. Those aging assets that were obscured in the glare of traders’ profit visions a mere two years ago are now the key to our very survival. Asset
optimization is the current buzz phrase, and it seems like everyone is offering
it. But what is asset optimization, and what do these offerings actually do?

Webster defines optimization as: “… an act, process, or methodology
of making something (as a design, system, or decision) as fully perfect, functional,
or effective as possible; specifically: the mathematical procedures … involved
in this.”

To understand what someone means by asset optimization, you can start by asking
yourself whether they are offering: 1) an act: the doing of a thing; 2) a process:
a series of actions or operations conducing to an end; or 3) a methodology:
a body of methods, rules, and postulates employed by a discipline.

There is no doubt that every power generation enterprise is inherently performing
some degree of asset optimization. Every day, employees from the bottom to
the top of the organization are making decisions and performing acts intended
to make their assets fully perfect, functional, and effective. There
are also plenty of engineering and consulting firms that can offer specialized
acts, or train your employees with methodologies designed to improve the effectiveness
of their acts.

I’m sure that we can readily agree that not all of these acts
have the desired effect, and even the ones that lead to a positive impact still
leave significant room for improvement. It’s not easy in
large organizations for employees to understand the impact of the acts they
perform, much less know the optimal combination of acts they should be performing.

What if it were possible for every individual in a power generation enterprise
to better understand the business impact of key decisions they make, prior
to making them? What if it were possible to implement a real-time process for
continuously improving the performance
of your assets? What if you could coordinate key local decisions or actions
toward the global impact they have on the company’s
bottom line?

This sounds like a pretty tall order, and it is. Power generation enterprises
are by definition massive, complex, distributed, and highly dynamic. They are
driven by multiple and often conflicting objectives and must operate within
tight constraints. Hundreds of thousands of decisions across equipment, units,
plants, and portfolios are executed within an enterprise each day, so is it
realistic to expect that we can capture all of this data, integrate it, and
then convert it into coordinated action toward a single objective? (see Figure
1).

One thing is for sure: putting monitoring tools in the hands of operators and
digital dashboards in the hands of executives will come nowhere close to solving
the problem. Monitors and dashboards only tell you what is happening; much
more important is knowing why it is happening and how to change it. It is critical
to get information about what is happening in the hands of executives so they
can make better strategic decisions. It is even more important to get knowledge
of why
it is happening and how to change it in the hands of those operating
the plants, the people who have a real-time impact.

But how is it possible for control room operators to understand the repercussions
of their local actions – pulling a lever here or there– on the company’s bottom line? How can maintenance engineers plan maintenance
activities to maximize not just plant up-time but also overall profitability?
How can units be dispatched in a way that accounts for current variable costs
and future availability in the face of highly uncertain power, fuel, and allowance
markets? How can all of these decision makers, acting over vastly different
time scales, adapt as plant and market conditions change?

The answer to the problem is a revolutionary approach based on our understanding
of the mechanics of the human brain, known as learning theory. This artificial
intelligence-based approach implies a radical shift in mindset from deductive
to inductive reasoning and opens up an entirely new world of asset optimization
possibilities and opportunities.

Incredible Machine

The brain is truly the most incredible machine that we know. Just
like an enterprise, it is massive, complex, distributed, and dynamic. Unlike
today’s typical enterprise, however, it has billions of parts that work
together with an amazing degree of coordination. The workers in the brain continuously
cooperate to solve global problems inductively and with relative ease.

To explain why a learning theory approach offers such promise for optimizing
power generation enterprises, it is first necessary to describe how the brain
functions (see Figure 2).

The brain is a collection of approximately 10 billion decision makers called
neurons. While these neurons are massively interconnected, each neuron is only
synaptically connected to
a few thousand other neurons. This means that very few neurons in the brain
have a direct connection to the external world. They can only communicate with
the local neurons that surround them.

What we know about neurons is that they behave in a binary fashion. They either
fire or they don’t fire in response to other neurons that fire in their
vicinity. When they do fire it is because they have received a precise pattern
of input activity from neighboring neurons that happened to fire at the same
time. This leads to a cascade of firing neurons. Those at the beginning might
have been responding to visual stimulation from seeing
a ball approach and those at the end could be driving motor movement in the
arm to allow us to catch it.

Credit Assignment

We know that the input patterns a neuron responds to have been learned from
correlated past events, and that the neuron continues to adapt its response
based on the global success of the task at hand. The concept that is at the
heart of learning theory is known as credit assignment. How does a neuron deep
within the recesses of the brain, which has no direct ability to see whether
or not the ball was actually caught, get credited or penalized for its role
in performing the task?

There are lots of theories and algorithms for the credit assignment problem,
some more biologically plausible than others, but what is significant here
is that they all provide a process, specifically the mathematical procedures
involved in this, for coordinating a massive number of local decisions toward
global objectives.

Learning theory assigns credit to the neurons that fire
in response to a particular situation, and this is a critical underpinning
of this proposed new approach to asset optimization. What if the same kind
of credit could be applied to each of the thousands of local decisions taken
each day within a power generation enterprise? With that kind of “enterprise
learning” capability, we would have the ability to adapt and adjust local
parameters to affect a desired global outcome. A power generation enterprise
could be likened to a living, breathing, learning being.

Leveraging Investments

This may seem like a futuristic dream, but the technologies exist today
for making it a reality. Furthermore, power generators have already made most
of
the required investments in instrumentation, automation, and monitoring. These
investments are just waiting to
be tapped. We have already started leveraging these investments on
a much smaller scale using artificial neural networks to solve local optimization
problems. Ten to 15 percent of US generators have real-time combustion optimization
systems and together are saving an estimated $250 million per year from the
NOx and heat-rate reductions they provide. Similar optimization systems have
recently been implemented on other subsystems as well
(e.g., soot blowing, steam temperature, selective catalytic reduction, selective
non-catalytic reduction, and flue gas desulferization systems).

The current systems are only scratching the surface
of the potential of learning theory, however. People are approaching the optimization
problems independently, building small brains to solve each problem without
any synaptic connections between them. This will not work. The systems we are
trying to optimize are highly interdependent, and it will not be possible,
and could even be dangerous, to optimize them without integration. Learning
theory provides
the ability to perform this integration, but it will require a radically new
approach to the design of the software systems from which these optimizers
are built.

Fortunately, there is such an approach to software design now available.
Microsoft and others have invested billions in shifting their focus from PC
to enterprise
applications. Bill Gates has bet the company on a radically new framework for
developing all software applications, called .NET. The result provides exciting
opportunities for innovation across the board, but in particular to the application
of learning theory to the challenge of asset optimization. I refer to this
marriage as enterprise learning.

This concept is beginning to gain momentum. In fact,
the US Department of Energy recently partnered with
NeuCo to invest $18 million in a four-year demonstration project through its
Clean Coal Power Initiative to prove out
the potential of integrated optimization systems within the power industry.
The initiative will leverage and accelerate
the investment NeuCo has already made in the commercialization of
a suite of optimization products in a real-time integrated software environment.
The result will be one large integrated brain that provides real-time optimization
of the equipment, units, and plants
in a power generator’s portfolio toward business objectives.

Summary

Asset optimization requires more than monitors in the control
room and digital dashboards in the boardroom. It requires arming
the people running the plants with the ability to translate business goals
into operational actions. A new form of asset optimization achieved through
the application of learning theory makes this possible and it will have a tremendous
impact on the future of power generation enterprises. The technology to build
truly adaptive enterprises will be available to progressive power generators
within five years, and it will be deployed throughout the industry within 10
years. The power generation industry, with its heavy investments in data, control,
and automation systems, is able to take the benefits
of this disruptive technology to a whole new level – to build truly adaptive,
learning enterprises.