Building expert systems
Machine learning an artificial intelligence approach volume II
Machine learning an artificial intelligence approach volume II
Adaptive learning of process control and profit optimization using a classifier system
Evolutionary Computation
| Hi-index | 0.00 |
Nickel cadmium batteries are an important source of power for aerospace applications. One such application is being developed at the Marshall Space Flight Center (MSFC) for use with the Hubble Space Telescope. A battery testbed has been built at MSFC to aid in that development. In addition, the Nickel Cadmium Battery Expert System (NICBES) was developed by Martin Marietta Corporation to assist NASA engineers in battery management.This paper describes an extension to NICBES which will make it more effective as a battery management tool. The extension involves the incorporation of classifier system machine learning techniques into a subsystem of NICBES. The principal reason for suggesting this extension is the nature of battery management itself. There is still much which is unknown about these batteries and the factors affecting their performance [2]. Hence, battery management might be said to be as much an art as a science and relies heavily on the expertise of the battery manager. NICBES is an attempt to incorporate that battery expertise into an expert system. One difficulty, however, is that battery behavior is likely to change over time in unforseen ways. This detracts from the usefulness of the expert system. Consequently, the battery manager who is using NICBES as a tool would be required to make changes to the expert system in order to accomodate the changed parameters of battery behavior. This should be the function of the knowledge engineer, however, not the battery expert. This is an example of the familiar problem of knowledge acquisition in knowledge engineering. The solution presented here is to use machine learning techniques to help overcome the knowledge acquisition problem. The expert system then interacts at a high level with the battery manager and undertakes adaptation on itself in order to determine new rules conforming to the changed parameters of the power system. The basic principles of learning classifier systems based on genetic algorithms will be presented first. Next, a brief description of NICBES will be given, particularly the advice subsystem to which the learning component will be added. A discussion of specific techniques by which machine learning can be incorporated into this particular rule-based expert system will follow. This discussion will come under the headings of the bit-string representation of rules, the initial rule population, an evaluation function for this system, and the genetic operators. Finally, some comments will be made concerning the implementation of a user interface for a system such as this.