Florence d'Alché-Buc

Neural Network Adaptive Modeling of Battery Discharge Behavior

Dynamic processes are often influenced by external conditions. We expand the neural network approximation capability to behavior modeling within an original hierarchical master-slave relation. Unlike the control theory paradigm, neural weights will replace “state variables” that may be impossible to measure. An application aiming at predicting the end of discharge for rechargeable batteries is fully described. This new battery management tool leads to accurate predictions (mean error is about 3 %) and its implementation into a portable equipment demonstrates that neural networks could be useful even for small size products. The system is further improved by on-line adaptation to actual conditions and individual behavior. This improvement reduces the error prediction to a low 1.5 %.

Asymptotic performances of a constructive algorithm

We present a numerical study of a neural tree learning algorithm, the “triolearning≓ strategy. We study the behaviour of the algorithm as a function of the size of the training set. The results show that a limited number of examples can be used to estimate both the network performance and the network complexity that would result from running the algorithm on a large data set.