smart grids and the American Way

The system must become automated, because decision speeds increasingly are becoming too fast for humans to manage.

By Roger Anderson and Albert Boulanger

The Aug. 14, 2003, blackout in the northeastern United States was not an isolated event. Slow response times of mechanical switches, lack of automated analysis of problems, and an inability to see the whole grid in real time are contributing to a noticeable increase in failures of the present electric grid. These problems have caused a noticeable increase in blackouts and brownouts since 1998.

The situation is not getting any better. The problems that have caused the recent spate of blackouts will propagate cascading failures of the grid more and more frequently, unless we create a more intelligent grid control system. The system must become automated, because decision speeds increasingly are becoming too fast for humans to manage.

This is a vital national security interest. If present trends continue, a blackout enveloping half the continent is not out of the question.

The grid also must evolve into a beyond-state-of-the-art system that the United States requires for its next 20 years of economic growth, and must be prepared for a time when it will be required to transmit and distribute 50 percent more power than it does today.

Cutting-edge technologies will be required to create what we call the Smart Grid of the Future that can accommodate the additional power from massive, remote solar and wind farms. These sources are environmentally benign, but erratic and unpredictable in their generation capacity. In addition, large-scale storage of electricity to accommodate the erratic nature of such green power sources will be required in elevated reservoirs, in super-batteries, flywheels, and magnets, and in underground compressed air and natural gas caverns.

We believe a distributed storage-generation smart grid will be required as demand grows in the United States, and as the country interconnects more and more with Canada and Mexico. At the same time, the Smart Grid must also cope with the added burden of defending vital economic resources against terrorism. The insertion of such computer upgrades, along with new hardware like superconducting fault current limiters, transformers and storage devices, digital power controllers, and next-generation nanotechnology transmission lines will require testing and development into an operational system before they can be deployed widely.

Before 1997, the frequency of blackouts had a simple relationship with the number of customers affected. That relationship has broken down, with large outages becoming more common—a sign the power grid is becoming unstable.

The management of the Smart Grid will require digital control, automated analysis of problems, and automatic switching capabilities more familiar to the Internet. An analogy would be routers sold by Cisco that break messages into packets and send them over different routes to relieve congestion, only to reassemble them at the destination into your next e-mail.

In short, the present U.S. electric grid will not work on any scale—local, state, national, or international—at the higher loads and more diverse generation sources required in the future, let alone if the terrorist threat becomes more severe. Failing to upgrade the system will leave us unprepared and, ultimately, in the dark.

With any complex machine like the electric grid, there will always be the occasional failure. But since 1998, the frequency and magnitude of blackouts has increased at an alarming rate, deviating from the stable, predictable "fractal" pattern of the previous 15 years. Blackouts in Chicago, Delaware, Atlanta, New Orleans, and New York in 1999, San Francisco and Detroit in 2000, and the infamous California "problems" of 2001 deviated from the predictable behavior of the 1980s and early 1990s.

This deviation in the frequency of outages compared to the number of customers affected by each should have provided a clear warning that the system was going unstable long before Aug. 14, 2003. As demand, sources of energy, and distances between demand and supply increase, the grid will become increasingly vulnerable not only to blackouts set off by equipment failures and weather, but also to terrorist attacks.

The specter of terrorist attacks across multiple power grids exacerbates the uncertainty inherent in the complex array of decision-making variables faced by utility operating engineers. But as engineers and managers maximize the value of their investment in computer networks, distributed sensors, and secure communication systems to make better, safer, and more profitable operating decisions, the threat of terrorism raises a fundamental dilemma for improving energy infrastructure performance and safety. How can we strengthen the electric power system against terrorist attack while not losing the needed advantages that come with increased computer control?

One critical aspect of the overall solution to this problem lies in the creation of computer-based simulations that provide a learning environment for gaining experience in responding to threats. Electric power engineers and operators must be trained to respond to threats: to electric generation and distribution systems under terrorist attack as well as to the normal threats to the power transmission grids, such as from weather and mechanical failures.

What's more, they must also deal with the power grid's ever-increasing interconnectivity to the web of other infrastructure grids spanning water, gas, telecommunications, transportation, automation, and fuel systems—a latticework crisscrossing many company and jurisdictional boundaries that all depend ultimately on the power grid.

Three frames from a threat simulation: An outage in Maine ripples through the Northeast power grid, turning off the lights of millions of customers. The Aug. 14, 2003, blackout extended west to Ohio.

The grid is an integrated infrastructure with millions of possible failure points, and computer simulators much like the ones used in manufacturing and aerospace industries are needed. Such a threat simulator would focus on critical failure trees that are grid-specific, and provide feedback loops and reinforcement learning to get better and better at their tasks over time.

The linking of events is where the threat simulator becomes most powerful. Specific events spanning the regional, local, and client grid networks can be analyzed and transformed into planned responses to specific attacks that the computer has seen before in simulation. Much the way advanced computers learn to play chess or backgammon, a computer learning system searches for failure dependencies. Thus, operators can be trained and ready for unforeseen contingencies.

The threat simulator identifies and optimizes answers to such problems impacting the grid as identifying new threats, finding new failure points, and determining propagation patterns (or domino effects). A threat simulator must automatically and continually learn as it absorbs, stores, and makes relationships among discrete runs of the simulator. The threat simulator learning system starts up loaded with a basic framework derived from playing through generic "tabletop" exercises. War-gaming experts pose problems and work through the solutions with grid operations experts. The lessons learned are electronically recorded as the exercise proceeds.

This sort of threat simulator uses reinforcement learning, a kind of dynamic programming that has many attractive properties for this kind of control problem. Neuro-dynamic programming, for example, is used to tackle computationally expensive evaluations. And another technique, vector quantization, simplifies the number of variables to be computed and reduces the tendency of machine learning programs to overgeneralize from a limited data set.

One implication of applying machine learning techniques to grid control is that the sensory intelligence has to be pushed to the "last mile" of the grid—to the point where it meets the customer—in order to make it sufficiently smart.

Before the end of the decade, cheap silicon devices will be attached to most manufactured items—even inert things like steel plates. These devices will have more memory and processor power than present-day laptop computers and will enable parts to identify themselves and find their location using GPS and wireless triangulation. They will communicate wirelessly with central control using secure ultra-wideband communications, all the while incorporating sensors on chips that interact with other sensors using modern electronic standards. The sensors, computer, and communicator will be self-contained and operational for the life of the asset, from its creation to its destruction.


It is crucial to add this kind of real-time sensing and control if the future grid is to fully exploit synergies among various power sources, including wind and solar. We believe the key to exploiting these synergies is ubiquitous silicon associated with all critical assets in the electric grid, from generation to storage, transmission, distribution, and finally consumption.

Cheap silicon in the field will rewrite the way we manage the electric grid. Widespread wireless computing on every critical node of the Smart Grid will deliver business intelligence. It will incorporate self-healing, self-organizing, Web services, and peer-to-peer computing among networks of connected assets. Each field computer will have enough memory so that it can capture its own best practices and be its own data historian.

One key benefit to successful application of such microelectro- mechanical technologies is having the vital assets of the power grid be able to sense not only their own well-being, but also the price for optimal business performance of each node. A sensor network that "rewrites the SCADA equation" must have its price driven down to dollars per asset (from the thousands of dollars today).

In addition, we must build "dashboards" for executives and operational people to manage the grid in real time. This sort of performance optimization via two-way communications with millions of such field nodes is a significant command-and-control challenge.

Once a threat simulator capability is in place, a computerized control capability could then be established to model and better understand the electric grid. At present, no model exists that can visualize the entire North American grid.

Computer assistance is particularly needed because the grid is exhibiting more and more behavior characteristic of chaotic systems. Since electricity on the grid is free-flowing, it moves as power pulses at nearly the speed of light. But, in reality, electrons flow back and forth between power plants and consumers at speeds slower than the flow of these power pulses. Electrons can move along the copper and aluminum wires and through the transformers of the transmission grid at velocities 100 to 1,000 times slower than light. This disparity produces the nonlinear behavior of the system that affects the flow of electricity in unpredictable ways, causing cascading failures like that experienced on Aug. 14, 2003.

But a smart grid needs more than just new software. New hardware is needed to provide buffers to the cascades of failures that are caused by congestion and disruptions of the flow of electricity—failures that propagate across the grid.

The Smart Grid will use new technologies such as MEMS, superconducting flywheels (above) to store energy, and wires of carbon nanotubes (top) to transmit electricity with no line losses.

Grid operators need the option of redirecting electricity around obstacles and disturbances at the speeds needed to forestall failures. Right now, flow can be redirected only through the use of mechanical circuit breakers at speeds adequate for most present, but not future, conditions.

One tool for increasing speeds is a power switch that combines thyristors to redirect flow with capacitors that provide buffering storage. Other technology will be needed to smooth the intermittent and unpredictable flow from large-scale wind and solar farms.

Among the hardware that must be added to the grid are distributed storage and generation hardware, such as high-temperature superconductivity storage. Superconductors have no resistance to electricity flow at supercritical temperatures, but when heated up—for example, by an electricity spike—the wires become resistive and limit the propagation of the power surge. Surges and sags in power could be handled using this natural property of superconductors in fractions of a second without shutting down the whole system.

Superconductor technology is at present too expensive for widespread use. But the current regulatory system makes such investments unlikely in any event. For example, regulators do not allow utilities to recover the costs of purchasing such equipment through consumer electricity rates. Yet these large-scale storage systems must be in place before we can decentralize the grid to accommodate significant amounts of smaller distributed generation and distributed storage capabilities.

This new grid hardware, called power controllers, would also allow operators to charge a higher fee for high-quality power, while no additional fees would be charged to users that don't need completely stable power. The ability to switch the flow of electrons is required for this type of dual power system, wherein more reliable power can be delivered (at an added cost) only to those consumers who need it, while lower-quality, less reliable, less expensive power could be delivered to the rest of us.

Presently, we all get the same high-quality, expensive power: 99.999 percent of the time, it is within a strict range of voltage and frequency, the five-nines in the power business. Most residential customers would gladly put up with momentary brownouts in exchange for markedly lower rates. But such brownouts would wreck a semiconductor assembly line—and a manufacturer would pay dearly to avoid one. The added revenue generated by a dual system could attract the private capital that is needed to upgrade and maintain the long-distance transmission system.

In the long term, there is great promise that high-temperature superconductivity and nanotechnologies will deliver several breakthroughs that could revolutionize the grid. This future grid could transmit electricity through woven rope made from carbon nanotubes. Such "quantum wire" could have electrical conductivity that is higher than copper's at one-sixth the weight, and at twice the strength of steel. A grid made up of such transmission wires would have no line losses or weather dependencies, eliminating the need for massive emergency generation capacity, and the grid could be buried without any special handling or sheathing.

It would be an enormous undertaking. There are more than 700,000 miles of transmission lines in the United States alone as of 2004. The factories required to weave quantum wire in the quantities needed for the grid haven't been built, or even planned.

In order to provide efficient, clean, plentiful, safe, and secure energy to power continued economic development with lessened environmental impact, this country will need to build a smart grid. How we get there from here is a hard question. Currently, there are no incentives for fixing the grid beyond short-term patches like laying additional transmission wires around congestion.

That strategy is much like urban highway construction. The more lanes a city provides, the more traffic the road attracts, producing more congestion, requiring more lanes, and so on. Also, the grid cannot be experimented with live. We must be certain that the grid is capable of handling each new technology before it is deployed. It is not an option to connect new gadgets directly to the grid, and accidentally cause massive, cascading blackouts.

The problem with creating such Smart Grid improvements is that, according to Technology Review magazine last year, the electricity industry has among the lowest research and development expenditures of all industries in the whole world. We must all recognize the electric grid as vital to our life, liberty, and pursuit of happiness if we are to fix this impending crisis.


Roger Anderson and Albert Boulanger are researchers at the Lamont-Doherty Earth Observatory in Palisades, N.Y.

Return to Index