Rate of Adaptation in Control Systems

Adaptive control systems are capable of giving near opt imum performance in the face of changing input command and noise characteristics, changing dynamics of the controlled processes, and changing goal or mission requirements. The principles of feedback control are used by the adaptation mechanism to control the structure of systems. The quality of adaptation is given by the system "misadjustment ," the ratio of the mean increase in mean square error (from adjustments based on finite statistical data) divided by the minimum mean square error. A control system can adapt to a major change in process statistics in about 10 times the impulse response time of the system itself, with a misadjustment of only 20%. Faster adaptation is possible with patternrecognizing adaptive control systems tha t use longer term experiences. Pattern-recognizing niters may be composed of adaptive Adaline "neurons ." A new electric circuit element called the "memistor" (a resistor with memory) has been devised to facilitate the realization of the Adaline neuron. It is a compact rugged electrochemical element whose resistance can be controlled reversibly by electroplating. The experiences of the neuron are stored in resistance values in a simple and directly usable form.