A Modular Connectionist Architecture For Learning Piecewise Control Strategies

Methodologies for designing piecewise control laws, such as gain scheduling, are useful because they circumvent the problem of determining a fixed global model of the plant dynamics. Instead, the dynamics are approximated using local models that vary with the plant's operating point. We describe a multi-network, or modular, connectionist architecture that learns to perform control tasks using a piecewise control strategy. The architecture's networks compete to learn the training patterns. As a result, a plant's parameter space is partitioned into a number of regions, and a different network learns a control law in each region.