Optimizing long short-term memory recurrent neural networks using ant colony optimization to predict turbine engine vibration

摘要

This article expands on research that has been done to develop a recurrentneural network (RNN) capable of predicting aircraft engine vibrations usinglong short-term memory (LSTM) neurons. LSTM RNNs can provide a moregeneralizable and robust method for prediction over analytical calculations ofengine vibration, as analytical calculations must be solved iteratively basedon specific empirical engine parameters, making this approach ungeneralizableacross multiple engines. In initial work, multiple LSTM RNN architectures wereproposed, evaluated and compared. This research improves the performance of themost effective LSTM network design proposed in the previous work by using apromising neuroevolution method based on ant colony optimization (ACO) todevelop and enhance the LSTM cell structure of the network. A parallelizedversion of the ACO neuroevolution algorithm has been developed and the evolvedLSTM RNNs were compared to the previously used fixed topology. The evolvednetworks were trained on a large database of flight data records obtained froman airline containing flights that suffered from excessive vibration. Resultswere obtained using MPI (Message Passing Interface) on a high performancecomputing (HPC) cluster, evolving 1000 different LSTM cell structures using 168cores over 4 days. The new evolved LSTM cells showed an improvement of 1.35%,reducing prediction error from 5.51% to 4.17% when predicting excessive enginevibrations 10 seconds in the future, while at the same time dramaticallyreducing the number of weights from 21,170 to 11,810.