Predicting Island Grid Demand through Advanced Deep Learning Models

The Solution

Big Blue AI developed a high-precision energy forecasting architecture and interactive operational dashboard for HEDNO (Hellenic Electricity Distribution Network Operator), processing three years of hourly consumption data across Skiathos Island to optimize power distribution for both the flexibility and day-ahead energy markets.

• Short-Term Demand Forecasting: Developed a continuous short-term prediction capability to anticipate hourly energy usage, giving operators real-time visibility to keep the electrical grid balanced and stable.
• Day-Ahead Consumption Planning: Built a long-term forecasting tool that projects full-day energy needs in advance, helping the team plan resources efficiently for next-day power markets.
• Cleaned & Enriched Consumer Insights: Combined weather data and seasonal trends with historical usage patterns, while filtering out non-essential meters to isolate true consumer demand across 32 key locations.
• Interactive Grid Operations Dashboard: Created an easy-to-use visual dashboard that allows operators to easily track weather conditions, compare predicted versus actual energy demand, and make fast, informed grid management decisions.

The Challenge

HEDNO required an accurate, scalable forecasting solution to manage power demand on non-mainland grids like Skiathos Island. Grid operators faced significant complexity due to severe seasonal tourism swings, changing weather patterns, and irregular meter deployments across a three-year historical dataset. Without precise hourly and daily predictions, allocating grid resources for the flexibility market and preparing bids for the day-ahead market posed significant operational risks for grid stability and energy pricing.

The Impact

The deep learning deployment delivered exceptionally precise predictive performance for HEDNO’s grid management teams:

• 90% Accuracy for Day-Ahead Forecasts: The stacked LSTM model achieved a low Mean Absolute Percentage Error (MAPE) of 10 percent when forecasting full day-ahead energy consumption.
• 86.7% Accuracy for Hourly Load Forecasts: The hourly flexibility model achieved a MAPE of 13.3%, closely tracking real-time load spikes and dips.
• 3 Years of Raw Grid Data Streamlined: Filtered, cleaned, and standardized multi-year telemetry across 32 consumer meters, successfully accounting for gradual meter onboarding and extreme outliers.
• Replicable Model for Island Grids: Established a validated deep learning framework ready for rollout across other interconnected Greek islands with similar tourism and geographical profiles.

How Big Blue AI helped 

Big Blue AI designed the end-to-end machine learning pipeline, leading the feature engineering, time-series decomposition, and hyperparameter tuning of deep LSTM networks using Python, Keras, and TensorFlow. By converting complex, multi-year island grid telemetry into actionable predictions and integrating them into an intuitive Power BI dashboard, Big Blue AI provided HEDNO with a powerful, scalable decision-support tool for modern energy markets.