Enhancing Dynamic Stability in Weak Power Grids: A Hybrid Deep Learning and Swarm Intelligence Framework, a case study of Western Kenya Distribution Network

The relentless push to integrate inverter-based Renewable Energy Sources (RES) into weak transmission grids creates a difficult balancing act: how to minimize costs without risking the grid's dynamic stability. This paper presents a new solution for the Western Kenya 132/220kV grid: a Tri-Hybrid Artificial Intelligence Framework. This system is designed to bridge the gap between accurate forecasting and secure power dispatch. It combines a Long Short-Term Memory (LSTM) network for precise weather prediction, physics-based models for realistic solar and wind simulation, and a Hybrid Particle Swarm Optimization - Grey Wolf Optimizer (PSO-GWO) to make dispatch decisions. Unlike standard approaches that treat stability as an afterthought, the proposed framework embeds non-linear checks—specifically for Frequency Nadir (>49.5Hz), Voltage Recovery, and Critical Clearing Time—directly into the decision-making loop. When tested on the Western Kenya network, this approach showed clear advantages over standard Genetic Algorithms (GA) and PSO. It cut operational costs by 7.0% ($5,810/hr versus $6,250/hr), reduced carbon emissions by 17.7%, and converged to a solution 60% faster. More importantly, the sensitivity analysis pinpointed a hard limit for Solar Hosting Capacity at 65 MW; going beyond this point compromises system inertia. N-1 contingencies were effectively handled, keeping the frequency nadir safely at 49.65 Hz and fixing long-standing voltage issues at Kisumu. These results prove that it is possible to maintain a resilient grid even with high renewable penetration, offering a practical roadmap for other emerging economies facing similar grid weaknesses.