Neural Networks
For more than three decades, SAIC has been researching technologies that mold the future. Neural network technology, which is an enabler of artificial intelligence, is an example of the technological innovations that we apply across a wide range of areas to help clients reduce costs and add value. Neural network technology has been applied to a wide variety of problems: image analysis, equipment maintenance, performance monitoring of a fossil fuel electric generating plant, and control of non-linear systems, to name a few examples. Another example, Intelligent Control of Building Energy Systems (ADEPT), is described below.
Our ADEPT artificial intelligence system uses a neural network and genetic algorithm to make buildings "smarter." ADEPT helps optimize the performance and minimize the operating costs associated with electric and gas-powered cooling systems. Taking into consideration weather, building usage, and electric rate information, ADEPT automatically develops an operating plan to meet cooling requirements at the lowest possible cost. In addition to saving a building owner money, ADEPT can help utilities operate more efficiently (which can reduce emissions and conserve natural resources) and pass savings on to their customers.
The ADEPT system uses two technologies that emulate the human brain's thought processes: an artificial neural network and a genetic algorithm. The neural network "learns" to recognize patterns with minimum information (weather and building operation information). The algorithm develops and evaluates operating schedules based on a "survival of the fittest" selection methodology.
Once the neural network has learned the building's operational characteristics and the associated cooling equipment, it considers the weather forecast and energy costs for the next day. The system re-evaluates its decisions every 15 minutes to meet the building's cooling requirements. If those requirements have not been met, the neural network and genetic algorithm make the appropriate adjustments.
The neural network adjusts by "re-learning." At the end of each day, the neural network compares its predictions to actual operation. If the difference between the prediction and the actual operation do not meet performance criteria, the neural network updates its knowledge base by incorporating the latest and best information available.
A finalist in the 1998 Computerworld Smithsonian Awards for innovations in information technologies in the field of energy, ADEPT was first demonstrated at a San Diego high school, where it saved 6% of the school's total energy costs during the summer. In a demonstration for a utility, ADEPT increased the use of thermal energy storage from 3% to 46%, resulting in a 13% reduction in average daytime demand. For commercial customers, ADEPT's ability to take into consideration real-time pricing is estimated to result in cost savings of approximately 10%.
ADEPT is just one example of how SAIC applies neural networks to meet the needs and objectives of our customers.
