@Article{ee.2026.075509,
AUTHOR = {Lei Shen, Qiang Gao, Shanyun Gu, Wei Li, Jun Li, Jianquan Li, Ruyi Xia, Jie Ji},
TITLE = {Lightweight Prediction-Driven Rolling Scheduling for Off-Grid Construction Microgrids under Variable Electric Demand},
JOURNAL = {Energy Engineering},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/energy/online/detail/25975},
ISSN = {1546-0118},
ABSTRACT = {Aiming at the contradiction between green energy consumption and diesel dependence in temporary construction camps under the condition of “weak data-weak communication”, this paper puts forward a collaborative framework of off-grid light storage and firewood storage with tight coupling of “prediction-scheduling”. The prediction layer constructs a lightweight IOOA-CNN-BiLSTM-Markov model: CNN extracts the spatial characteristics of tower crane shadow and cloud cluster, BiLSTM captures the bidirectional time series dependence, Markov residual compensates the non-stationary disturbance, and uses the improved Osprey algorithm to complete the small sample superparameter self-tuning at the edge of ARM, thus realizing the error-oriented compression in the scene of lack of weather channels. The scheduling layer is designed to improve the affine adjustable rolling optimization driven by the snake optimization ISO. With the prediction interval-scene dual mode as the input, the start-stop frequency of diesel and the energy storage SOC are modeled as integer-continuous variables synchronously, and the three-objective MILP of “minimum diesel + minimum load loss + minimum light rejection” is solved in seconds, thus achieving the multi-objective coordination of green power priority, energy storage arbitrage and diesel compensation. By building a photovoltaic, energy storage and diesel hardware-in-the-loop platform and loading a typical temporary site with an impact load of 80–400 kW, the system verifies the adaptability and deployment of the proposed framework under the extreme conditions of “zero power grid, zero history and weak sensing”.},
DOI = {10.32604/ee.2026.075509}
}