Multi‐objective auto‐encoder deep learning‐based stack switching scheme for improved battery life using error prediction of wind‐battery storage microgrid

Summary For any wind power generation system, battery energy storage is a suitable backup power unit for ensuring greater functionality by compensating the prediction error due to the variable nature of generation. For designing local energy management system, the hierarchical operation of distributed generation controllers, BES life, etc. is directly influenced by the prediction error profile. To design an effective local energy management system, it is imperative to obtain wind power generator control reference influencing the overall system stability and reducing the deterioration of life and power loss of the integrated battery energy storage system due to increased temperature. Thus, an efficient local energy management system for the doubly fed induction generator battery energy storage system is proposed here for the robust minimization of the prediction error and to increase controller responses and the life of the battery. A novel deep learning robust multilayer multi‐kernel extreme learning machine autoencoder algorithm is proposed here in order to obtain an improvement in the prediction error profile. This prediction algorithm produces optimal model generalization along with the minimization of reconstruction errors and uses simple matrix inversion for prediction in comparison with more complex deep learning neural networks. A new secondary controller is proposed to address the degradation of the life of the battery energy storage system due to increased temperature/power loss profile under prediction error. A model reference‐based battery temperature model, associated with multi‐objective optimization‐based temperature tolerance dynamic stack reconfiguration of battery stacks is incorporated in the proposed secondary controller. This secondary controller‐based local energy management system operation is targeted towards satisfying local load demand and grid dispatch, while ensuring an optimized battery stack performance under wind power generation discrepancies. In addition, a new adaptive slope‐based primary controller is designed for effective power sharing between the battery energy storage system and wind power generator based on % state of the charge of the energy storage system. Also, an independent distributed generator controller is studied in an elaborate manner for the DC link stability of the DC‐DC converter and voltage source converter for grid synchronization control.