Precise Energy and Load Forecasting

There is no universal forecasting solution. Our flexible approach combines advanced statistical and AI-based methods to develop the optimal model for specific operational requirements.

• Deterministic forecasting: Precise expected values
  Deterministic models provide a precise forecast value for each time step (e.g. "10.2 MWh at 2:15 p.m."). These point forecasts are easy to interpret and form the basis for many operational decisions. To achieve this, we apply advanced algorithms such as gradient boosting (XGBoost) and neural networks (TCN, Transformer, Few-Shot Learning), which capture complex relationships from historical data and external influencing factors (e.g. production data and weather conditions).
• Probabilistic forecasting: Managing uncertainty intelligently
  We go beyond single-value predictions by providing realistic forecast intervals (e.g. "with a probability of 90%, consumption will range between 9.8 and 10.6 MWh"). Similar to a weather forecast that indicates the probability of rain, this approach directly quantifies uncertainty and risk. This enables more informed, risk-based decision-making, for example when hedging energy volumes on the market. To achieve this, we apply advanced methods such as quantile regression and ensemble models.
• Hierarchical forecasting: Consistency across the entire organization
  Companies typically have hierarchical structures: machines form lines, lines form facilities. Using hierarchical forecasting methods, we ensure that forecasts remain mathematically consistent across all levels. The sum of subsystem forecasts therefore corresponds exactly to the forecast for the overall site. This prevents inconsistent planning assumptions and ensures that all departments work with a consistent data foundation.

Every project starts with a clear business objective. Together with you, we determine the forecasting goals and create the foundation for successful project implementation.

• Defining the forecasting objective: What should be optimized? Is the focus on minimizing energy costs, avoiding peak loads, optimizing storage operation, or supporting strategic energy procurement planning?
• Defining the forecast horizon: We define the required forecasting horizon, which strongly influences model selection:
  • Short-term forecasts (intraday to 48 Hours): For operational shift planning, spot market trading, and real-time load management.
  • Medium-term forecasts (days to weeks): For optimizing energy procurement, maintenance planning, and the operation of energy storage systems.
  • Long-term forecasts (months to years): As a reliable basis for budgeting, capacity planning, and strategic investment decisions.
• Defining success metrics: We define how forecasting performance will be evaluated – not only from a technical perspective (e.g. forecast error), but also from an economic perspective (e.g. cost savings).

In this phase, we establish the foundation for high forecasting accuracy. We identify, collect, and prepare all relevant data to provide the forecasting models with the best possible information base.

• Identification of relevant data sources: We use historical load profile data as the foundation and enrich it with valuable internal and external predictors, such as production schedules, weather forecasts, and calendar data.
• Data integration and cleansing: All data streams are synchronized on a common timeline. We implement robust processes for data validation, outlier correction, and the intelligent imputation of missing values.
• Feature engineering: We derive meaningful features from raw data that serve as important signals for forecasting models, such as weekdays, shift indicators, and temperature dependencies.

This phase represents the technical core of the project. Here, we develop, train, and validate forecasting models tailored specifically to your data and objectives.

• Systematic model selection: Based on the forecasting objective and data complexity, we evaluate a broad range of models to ensure optimal predictive performance.
• Training and rigorous validation: The model is trained using historical data, and its performance is objectively evaluated using a separate validation dataset (backtesting) before deployment.
• Application of our core methodology: We implement the probabilistic and hierarchical approaches described in our methodology to quantify uncertainty and ensure forecast consistency.
• Objective performance evaluation: Model performance is assessed using industry-standard metrics (e.g. MAE and RMSE for point forecasts; Pinball Loss for probabilistic forecasts).

A forecasting solution only delivers value when it is seamlessly integrated into operational processes and continuously provides reliable results.

• Forecasting as a service: We provide forecasting models via standardized interfaces (e.g. REST APIs), enabling easy integration into existing energy management systems, dashboards, and trading tools.
• Operational automation: The entire workflow – from data updates and forecast generation to result delivery – is fully automated.
• Continuous monitoring and adaptive retraining: We continuously monitor forecasting performance during live operation. When operating conditions change (e.g. new systems or modified processes), automated retraining is triggered to ensure consistently high forecasting accuracy.

Use cases & their specific benefits: Putting forecasts into practice 

Our load forecasts create direct economic value in key business areas:

• Optimized energy procurement and trading: Purchase the right amount of energy at the right time on the spot market, or secure long-term contracts based on a well-founded demand forecast. Avoid costly balancing energy purchases.
• Efficient scheduling and production planning: Synchronize your production schedules with forecasted energy prices. Strategically schedule energy-intensive processes during low-price periods.
• Grid fee optimization through peak load forecasting: We accurately forecast your 15-minute peak loads. This allows you to take timely countermeasures (e.g., shutting down consumers, using battery storage) and significantly reduce your grid fees.
• Management of on-site power generation and storage: Optimize the operation of your PV systems, CHP units, and battery storage. Our forecasts help you maximize self-consumption and optimally charge and discharge your storage systems.