Battery energy storage system scheduling based on variable-step prediction and online dynamic nonlinear programming

The increased penetration of renewable energy sources has exacerbated the issue of peak shaving in power systems. To address this challenge, Battery Energy Storage Systems (BESS) emerge as a crucial solution. However, due to the limitations of BESS capacity, unrestricted charging/discharging is not...

Full description

Saved in:

Bibliographic Details
Published in:	Journal of energy storage Vol. 127; p. 116874
Main Authors:	Cai, Jun, Cai, Yuxin, Chen, Liang, Yan, Ying, Cheok, Adrian David, Chen, Zhong, Zhang, Xin
Format:	Journal Article
Language:	English
Published:	Elsevier Ltd 15.08.2025
Subjects:	Battery Energy Storage System (BESS) Convolutional Neural Network (CNN) Long Short-Term Memory (LSTM) Online Updating Dynamic Nonlinear Programming Algorithm Based on Variable-Step Prediction (OUDNLP-VSP) Peak shaving Peak shaving Battery Energy Storage System (BESS) Long Short-Term Memory (LSTM) Online Updating Dynamic Nonlinear Programming Algorithm Based on Variable-Step Prediction (OUDNLP-VSP) Convolutional Neural Network (CNN)
ISSN:	2352-152X
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	The increased penetration of renewable energy sources has exacerbated the issue of peak shaving in power systems. To address this challenge, Battery Energy Storage Systems (BESS) emerge as a crucial solution. However, due to the limitations of BESS capacity, unrestricted charging/discharging is not feasible. Present short-term programming strategies overlook this factor, leading to BESS being either fully charged or insufficiently charged, thereby impacting subsequent programming. In response, an Online Updating Dynamic Nonlinear Programming Algorithm Based on Variable-Step Prediction (OUDNLP-VSP) is proposed, aiming to minimize power fluctuations in the grid. OUDNLP-VSP is divided into two phases. In the day-ahead phase, a combined Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) Network based forecasting model is utilized to obtain an approximate load profile for the following day, which is then used to formulate a charging/discharging strategy. In the intraday phase, the forecasting model is employed to predict the load value for the next time step, which can enable real-time optimization of the charging/discharging strategy established in the day-ahead phase. •OUDNLP-VSP algorithm resolves BESS scheduling via dynamic nonlinear programming with online updates.•Day-ahead uses CNN-LSTM 96-step predictions for BESS strategy. Intraday refines via single-step predictions.•PI controller iteratively adjusts real-time predictions to enhance accuracy at each timestep.•Nonlinear programming updates sequences/constraints in real-time using latest forecast data.
ISSN:	2352-152X
DOI:	10.1016/j.est.2025.116874