Electricity Consumption Forecasting Engine

Energy-intensive companies need reliable consumption forecasts for procurement, financial planning and sustainability management. Forecasts based mainly on historical consumption often ignore weather, site assets, operational changes, and expected production.

The challenge is to use 15 minute load data profile data supplied by the TEELS API or via MCP and to optimize a baseline forecasting function by identifying and integrating the most relevant external and company-internal drivers.

The solution should systematically evaluate which external and internal drivers provide genuine predictive value and quantify their contribution relative to a historical baseline. Participants are encouraged to benchmark different modelling approaches, ranging from statistical time-series models to machine-learning methods and to explore feature engineering, model selection, and hyperparameter optimization. Attention should be paid to robust validation strategies, overfitting, uncertainty estimation and model interpretability. The resulting forecasting approach should generalize across heterogeneous locations, consumption profiles and companies rather than being optimized for a single dataset.

Expected Outcome

The goal is a reusable approach for different companies, locations, and consumption structures. The working prototype should:

• consume quarter-hourly load profile data from TEELS load profile engine via API or MCP
• process the quarter-hourly consumption data 
• combine external drivers with company-specific data
• select forecast drivers and optimize location-level models
• measure accuracy with transparent metrics
• provide visualized outputs which are useable for decision makers in procurement, finance and sustainability management
• build the solution independently from TEELS, but keep in mind that an integration of forecasting module you develop might happen later
• optional extension “measurement hierarchy”: map decentralized main meters, sub-meters, and other electricity measurements without double counting into a hierarchical structure (meter wizard)

Modern tools like TEELS provide automated insights into load profiles for different stakeholders such as DSOs (distribution system operator), TSOs (transmission system operator), energy service providers or energy managers.

Reliable forecasts are a key foundation for data-driven energy management for all of these groups.

Better forecasts can reduce procurement volume risks, improve electricity cost planning and support sustainability reporting requirements. Additional site and operational information also make forecasts more responsive to business changes and therefore more precise.

The challenge combines data engineering, forecasting, model evaluation, and business analytics in a practical prototype with financial and sustainability impact while interfacing TEELS as a potential new standard in the DACH region for load data analysis.

Detecon anticipates a future increase in demand for smart load data analysis and forecasting tools. Once integrated and deployed, the developed solution shall be used for Detecon’s core business in consulting and creating financial and sustainable added-value for energy-intense customers (e.g. through optimized energy procurement, increased implementation speed and accuracy for sustainability reportings, etc.)

Data redeemed from TEELS may include quarter-hourly consumption, baseline forecasts, site informations and technical features (e.g. new machinery, added renewable energy, increasing implementation and use of EV charging, etc.), weather and calendar data and production or volume data. Meter hierarchy data may be added for the optional extension from actual production sites.

This challenge is ideal for participants with skills in:

• Data Engineering: to integrate time-series, external, master, and optional meter hierarchy data
• Data Science / Forecasting: to engineer features, optimize models, evaluate accuracy, and explain results
• Business Analytics / Data Visualization: to translate forecasts into usable visual outputs