Integration of renewable energy sources based on LSTM with Harris Hawks Optimization for power system reactive power and power flow

A hybrid framework combining Long Short-Term Memory and Harris Hawks Optimization (LSTM + HHO) is introduced to facilitate Reactive Power Optimization (RPO) and Power Flow Control (PFC) in power systems integrated with Renewable Energy (RE). In academic discourse, two-stage approaches address the prediction and optimization phases independently and sequentially. The suggested model distinguishes itself from these techniques by operating a closed feedback loop between the two phases. The LSTM temporal forecasts modify the HHO search domain at each optimization period based on the anticipated operational condition. System Loading (SL), Renewable Penetration (RP), and Voltage Stability Margin (VSM) are optimized through a novel adaptive weight-coupling approach that incorporates a forecast-uncertainty metric derived from Monte Carlo dropout applied to the LSTM to establish the HHO objective weights. Consequently, forecast dependability modifies the metaheuristic search priorities at each interval, with diminished forecast confidence prioritizing voltage regulation and reactive power assistance. The model is tested on the IEEE 30-bus system, modified to include 90 MW of wind and solar generation. Installed capacity penetration in this system is 31.8%. The research uses three operational scenarios: baseline case, High Renewable Penetration (HRP), and Extreme Weather Conditions (EWC). Statistical robustness was tested using 30 independent trials per technique. The results show that LSTM + HHO reduced total active power loss by 6.92% to 12.46% across the three scenarios. The voltage profile improved by 39.7%–52.7%. The reactive power reserve margins are 38.7% to 67.3% greater than the Newton-Raphson with Interior Point Optimization baseline. Load-balancing improvements range from 9.7% to 12.8% over the same baseline. Cycle execution times vary from 3.25 to 4.12 seconds . This aligns with the real-time response of Supervisory Control and Data Acquisition (SCADA) systems (4 to 10 seconds in simulated environments), but it has not been validated in actual deployment.