
AI-Driven Decision Support for Energy Trading and Portfolio Management
Introduction
In Brazil’s centrally dispatched power system, the National Grid Operator (ONS) often dispatches thermal plants outside the economic merit order due to transmission and energy security constraints. At the same time, the fast growth of wind and solar has sharply increased renewable curtailment. Both effects create significant uncertainty in the cash flows of generators and traders, affecting contract pricing, exposure management and operational planning.
Existing market tools could not forecast these variables with the accuracy and speed required by middle-office decisions, and the analyses were spread across spreadsheets and disconnected tools. The client needed a single platform connecting price, dispatch and curtailment forecasts to its commercial, operational and regulatory decisions.
Methodology
Within an R&D project regulated by ANEEL (the Brazilian electricity regulator), Mip Wise was responsible for the architecture and full development of the system, integrating the forecasting and financial models built by the project team into a single platform. Mip Wise also contributed to the research fronts on optimization and thermal plant flexibility. The project followed these phases:
- Decision process mapping: Translation of the client’s middle-office processes (portfolio management, trading strategy, maintenance planning, budget control and contract penalty monitoring) into system modules, data flows and business rules.
- System architecture: A modular platform built in Python, with independent computation engines, a PostgreSQL database, a reactive web interface and strict data contracts validated at every layer. The system runs in Docker containers on AWS, with automated testing and continuous deployment.
- Data and model integration: Automated ingestion of the official ONS and CCEE studies and of market data through APIs, feeding engines for marginal cost and spot price (PLD) projection, forward curves for mark-to-market, merit-order and out-of-merit thermal dispatch, and renewable curtailment, the last two based on machine learning. A settlement engine replicates the CCEE market rules at hourly granularity to compute each plant’s financial results.
- Optimization research: Mip Wise proposed a new Mixed-Integer Linear Programming (MILP) formulation that represents the multi-period transitions between operating modes of combined-cycle plants in Brazil’s day-ahead dispatch model, which currently treats these transitions as instantaneous. This work resulted in a scientific paper submitted to an international journal.
- Diverter damper analysis: Mip Wise contributed to the analysis and mathematical modeling of combined-cycle plants with diverter dampers, equipment that allows gas turbines to run in simple cycle. The models supported a technical, economic, and system-level valuation of installing this equipment.
- Deployment and evolution: Unit, integration and stress tests, training workshops, and a technical and user manual supported the handover. Mip Wise continues to maintain the system and deliver improvements.
Results
The system is in production at Âmbar Energia, where it runs weekly on the official ONS and CCEE data to support portfolio and trading decisions. Five case studies with the client’s real decision processes validated the platform.
- Single decision platform: Price, dispatch and curtailment forecasts and the financial settlement of each plant are now available in one environment, replacing spreadsheets and disconnected tools.
- Better trading decisions: In backtests, a portfolio strategy guided by the system’s forecasts delivered R$ 3.0 million (6.1%) more than a static price-trigger strategy. A short-term trading strategy based on the system’s signals accumulated R$ 3.2 million, against R$ 0.2 million for a traditional technical-analysis benchmark.
- More realistic dispatch modeling: The proposed formulation for mode transitions avoids physically unrealistic schedules and distortions in hourly prices that arise when transitions are treated as instantaneous.
- Value of flexibility: Simulations showed that diverter dampers can reduce renewable curtailment by 26% and increase system ramp capacity by 36%, with an option value that exceeds the equipment cost in three of the five plants analyzed.
Facing a similar challenge?
Every operation has its own constraints. Tell us what you have in mind and we will talk about what could be done in your case, with no obligation.