Experimental and Computational Approaches to AI-Driven Load Forecasting and Dynamic Pricing in Smart Grids

Increasingly complex modern power systems, necessitated by integrating renewable energy sources, the electrification of transport, and the dynamism of consumer behaviour, have created an urgent need to develop forecasting and adaptive pricing mechanisms. Long-standard statistical and econometric techniques are fast but cannot describe nonlinear trends and data of high dimension that occur in smart grid settings. This research proposes a hybrid solution that includes heterogeneous artificial intelligence (AI) systems to forecast loads and a reinforcement learning-based algorithm to set prices dynamically. Weather and socio-economic variables augmented historical utility data were utilised to train forecasting models on artificial neural networks (ANN), long short-term memory networks (LSTM), gated recurrent units (GRU), and random Forest ensembles. The combination method proved the most accurate, with a vast improvement in the error measures compared to when individual models were used. Reinforcement learning was used to develop adaptive tariff schemes that react to changes in real-time demand and reduced peak load by 15% and consumer costs by 8% compared to their baseline pricing schemes. To make transparent, explainable machine learning approaches like SHAP and LIME were incorporated, which allow interpretable insights into the demand forecast and the pricing choice. These results show that AI-based systems have the potential to increase grid stability, improve cost-efficiency, and build consumer confidence, providing a scalable and sustainable way forward for smart energy system evolution.