A Hybrid LSTM–FNN Framework for Safety-Constrained Energy Management in Mining Microgrids

This paper presents a novel framework for the development of a real-time energy management system for mining microgrids, which integrates the benefits of a long short-term memory (LSTM) network and a feedforward neural network (FNN) for the prediction of the load and solar power, and the optimization of the dispatch, respectively, while ensuring the safety of the microgrid through the application of a convex safety filter. In the proposed framework, the LSTM provides probabilistic multi-step forecasts of load and photovoltaic generation, capturing the high volatility characteristic of mining operations with ramp rates up to 5 MW/min. The FNN approximates the optimal power dispatch policy, enabling sub-millisecond inference times essential for real-time control. The convex safety filter projects the FNN’s proposed actions onto the feasible set defined by operational constraints, ensuring voltage regulation within ±0.1% and preventing safety violations. The framework was validated using operational data from Jwaneng Mine, Botswana, within a MATLAB/Simulink co-simulation environment that couples discrete-time EMS decisions (15-min intervals) with continuous-time electrical dynamics (1-s resolution). Simulation results demonstrate an 18.7% reduction in operational costs, renewable energy utilization of 79.1%, voltage deviation of only 0.08%, and constraint violations reduced to 0.3% of intervals. The complete system achieves end-to-end latency of 50.8 ms, with the core optimization requiring just 0.8 ms, satisfying the stringent real-time requirements of mining microgrid control.

The authors would like to thank Botswana International University of Science and Technology (BIUST). The authors would like to thank the management and engineering teams at Jwaneng Mine for their support.