Neural Network Architecture for Determining the Aging of Stationary Storage Systems in Smart Grids

Abstract

The estimation of the State-of-Health (SOH) of energy storage systems is a key task to ensure their reliable operation and maintenance. This paper investigates a new SOH determination method for stationary storage in Microgrids. Aging tests are conducted on NMC cells, with test profiles corresponding to Microgrids’ conditions. The focus of this work is on optimizing the learning process and the application of a Multilayer Perceptron (MLP) model to address this issue. This study introduces a novel approach of considering lag sequences, or time series data, to expedite the learning procedure and enhance prediction accuracy. A key advancement in this research is the usage of shorter time intervals to calculate the SOH, which not only reduces the learning time but also decreases the application time. This approach led to an overall reduction in computational effort when estimating the SOH. Energy is introduced as a new input parameter, resulting in improved modeling and more accurate SOH estimations. Furthermore, the MLP model achieved a Mean Squared Error (MSE) of 2.95 and a Mean Absolute Error (MAE) of 1.10, which are indicative of its strong predictive accuracy. Emphasis was also placed on the careful tuning and optimization of the neural network’s hyperparameters. The goal was to design a computationally efficient network that still yields optimal results. The findings demonstrate the effectiveness and potential of the MLP model in SOH estimation, underscoring the importance of the methodical model design and hyperparameter optimization.