Optimal dispatching of renewable energy-based urban microgrids using a deep learning approach for electrical load and wind power forecasting

Optimal load dispatching plays a vital role in improving the reliability and efficiency of renewable energy systems. This research presents a Mixed-Integer Linear Programming (MILP) approach for optimizing a power system's daily operational cost while increasing its resilience, including a wind turbine, battery, and conventional grid. Deep learning and statistical models along with a novel hybrid model, were developed and used to forecast the 3 days ahead load demand and wind power output. Testing these models shows that the proposed hybrid model could predict load with more accuracy than other models and it could reduce the root mean squared error by 22% to 44% for load forecasting and by 10.5% to 16.6% for wind speed prediction. The MILP model is applied for optimizing the load dispatch of an urban microgrid. The results of the dispatching model show that adding battery storage not only can bring down the grid-connected daily operational cost (from $8.4/day cost to $109.8/day income) and increase the resilience of the system by providing an off-grid mode, but also can extend its lifetime through minimization of degradation cost. The results also indicate that the degradation cost of batteries will contribute to a bigger portion of the operational costs in an off-grid mode in comparison to that of wind power curtailment cost. This research can inform effective and logical decisions for urban micro-grids and direct better integration and use of renewable energy systems in urban areas.