Abstract:
Optimising power-generation efficiency has arisen as a critical challenge in the fastchanging environment of smart grids and urban ecosystems. Properly using energy
resources has become essential in today’s technologically advanced world. Smart Grids are
at the forefront of this shift, giving a complete energy consumption and distribution
approach. Given the delicate connection between climatic conditions and energy use
patterns, including weather data in these networks is an improvement and a necessity now.
Significant advances in the domain have used large amounts of weather data and
sophisticated models. Future studies can improve on these findings by combining
sophisticated time series models with meteorological data and optimising them for demand
response techniques in smart grid power generation. This study presents a novel strategy
combining smart grid electricity generation demand response mechanisms with hyperlocal
weather forecasts. The project attempts to improve the accuracy and reliability of power
generation estimates by using the capability of machine learning. Five distinct time series
and machine learning models - SARIMAX, Prophet, Holt-Winters, XGBoost, and LSTM
– have been integrated with hyperlocal meteorological data, encompassing precipitation,
relative humidity, temperature, and cloud cover. SARIMAX has shone out among the
individual models, with a MAPE of 4.92% and an MAE of 3.54, demonstrating its ability
to capture subtle seasonal patterns and autocorrelations. The hybrid model, an ensemble of
SARIMAX, Prophet, and Holt-Winters, outperformed the individual SARIMAX in
predicted accuracy, boasting an impressive 0.06% MAPE and an MAE of 4.43. When
paired with real-time data analytics, this demonstrates the transformational potential of
machine learning algorithms. A new aspect of this research is introducing a user-centric
dashboard, which provides a real-time display of anticipated data and model performance
indicators. Its versatility is enhanced further by user-specific customisation options, which
provide specialised forecasting insights over user-defined timeframes, increasing real-time
decision-making processes. The combination of demand response tactics with powerful
machine learning models, demonstrated by the hybrid model’s excellent performance,
offers a promising path toward increased flexibility and efficiency in smart grids and cities.
