pictogram Future-Load-Model

Research Project: Future-Load-Model for Predicting Representative Load Collectives

Institute of Machine Components (IMA)

Function development to form representative load collectives based on past load for predictive diagnosis of lead-acid batteries
[Photo: Tamer Tevetoglu]

As part of a predictive diagnostic model, the Future Load Model is being used to predict future loads (time series data) that significantly influence the remaining lifetime of the lead-acid battery. For this purpose, past sensor data of the battery are first stored and grouped into representative load cases using unsupervised machine learning methods. Then, a feedback neural network (RNN) based on a so-called long short-term memory (LSTM) is trained and being used to predict future signal data. The now predicted data is assigned to the representative load cases by a classifier, e.g. Random Forest. By using the scattering of the data points within a load case, the confidence intervals of the predicted signal data are derived. Now the remaining lifetime can be calculated based on the future loads.

detailed project description (.pdf)


    To the top of the page