Operations Optimization

Our Approach

HydroForecast™ contains an end-to-end watershed operations optimization system, comprised of three components:

  1. A reservoir and river network simulation environment which can be configured to represent the reservoirs, hydropower assets, river network, environmental considerations and regulatory requirements of any river system.
  2. A price forecasting machine learning model which takes market signals, flow forecasts and asset operations as inputs and predicts future power prices.
  3. Reinforcement learning optimization agents which predict an optimal asset operation plan coordinated among all assets on a river system, taking into account environmental, regulatory and other stakeholder considerations.

How It Works

Reinforcement learning (RL) is a machine learning technique popularized when AlphaGo, an RL agent, defeated a world champion Go player.

RL is concerned with how to take actions in an environment so as to maximize some notion of cumulative reward. RL’s ability to balance multiple objectives and sacrifice short term gains for larger, long-term rewards has propelled major recent advancements in strategic, transformative decision-making for integrated watershed management.

By creating multiple RL agents, each of which prioritize different objectives such as overall revenue, risk reduction, and near term revenue, the system suggests a few near-optimal operations plans to human operators and allow them to weigh the relative merits of each plan and select the one which works best for their organization at that time.

Together this system produces hourly asset operational recommendations optimized to maximize revenue, improve safety, reduce risk, support environmental objectives and meet regulatory requirements.