A new continuous learning method is applied to the problem of optimizing the selection of services in response to user requests in an active computer network simulation environment. The learning is an enhanced version of the 'snap-drift' algorithm, a hybrid form of learning that employs the complementary modes: fast, minimalist (snap) learning; and slower drift (towards the input patterns) learning, in a nonstationary environment where new patterns are continually introduced. Snap is based on Adaptive Resonance Theory, and drift on Learning Vector Quantization. The new algorithm swaps its learning style between the two modes of self-organisation when declining performance levels are received, but maintains the same learning style during episodes of improved performance. Performance updates occur at the end of each epoch. Reinforcement also occurs by maintaining successful adaptations, since learning is enabled with a probability that increases with declining performance. The method is capable of rapidly re-learning and is used in the design of a modular neural network system, Performanceguided Adaptive Resonance Theory. Simulations demonstrate the learning is stable, effective and able to discover alternative solutions in response to new performance requirements and significant changes in the stream of input patterns.