مشخصات پژوهش

The need for more accurate grid-instability forecasting is driven by the rapid growth of renewable energy. Wind and solar energy are inherently dependent on weather conditions, which fluctuate constantly. When these changes are not anticipated early enough, the grid may experience sudden mismatches between supply and demand. These imbalances can then lead to voltage deviations, frequency swings, and other forms of instability. Because renewable penetration keeps increasing each year, operators require predictive systems that can interpret environmental conditions and grid behavior together in real time. Traditional engineering approaches to stability assessment still play an important role, but they also have clear limitations. Physically based models can be computationally heavy, and they demand ongoing manual updates to reflect new infrastructure, distributed energy resources, and market regulations. They are also difficult to scale to large-area systems with thousands of nodes and millions of users. In contrast, data-driven methods can continuously learn from evolving grid conditions yet, many prior studies still use only electrical data and lack awareness of external factors that trigger instability. Research in multimodal forecasting has shown that combining different types of information such as SCADA measurements, meteorological data, and contextual features consistently enhances model robustness. Electricity market data, in particular, encodes economic pressures that strongly influence generation scheduling and consumer behavior. By incorporating these complementary signals into a single deep learning architecture, forecasts can become more accurate under complex, real-world operating scenarios. This capability is especially important as grids shift away from centralized fossil-fuel generators toward more distributed and unpredictable resources . Finally, the operational value of such forecasts is significant. Earlier detection of instability can give operat