In many cases, the production of hydroelectricity is managed at the watershed scale, not just from individual plants. Producers must monitor and control their operations over a whole system of interconnected water reservoirs, rivers, and artificial ducts.
An example of such a system is given in this schematic:
Plant operators constantly take decisions on the modulation of water flows through turbines for immediate electricity production, but can also pump water through artificial ducts to transfer production potential from one plant to another, or to defer it to later times.
These decisions are taken in a context of fluctuating electricity prices (hence of the revenue from selling it to the grid), changing water inputs (brought by rain and melting glaciers), and evolving operational conditions (e.g. the conversion efficiency of turbines is not constant). Moreover, instantaneous decisions have an impact over the future because of safety and technical constraints : for example a turbine cannot be instantly shut off, the level of a reservoir must stay in given limits, etc.
Hence, the optimization of the production revenue is a complex problem, and the operators try to follow a production plan usually computed by the R&D department to optimize the revenue over a long period of time (typically a year). It is frequently updated (up to several times a week) as the operating conditions and anticipated future evolve.
The objective of this challenge is to develop a solution helping:
- Planners (R&D and financial) to build and update production plans and guidelines, and estimate the profitability and sustainability of their operations over the long term,
- Production operators to make immediate decisions taking into account present conditions and future scenarios. This is a new development compared with the common practice, where planning is centralized and there is no assistant tool in the production centers.
The decision optimization must lead to maximal profits, and be computed as quickly as possible.
How will it be measured?
The solution will be tested on a large amount of operational data available collected for different scenarios.
It will be benchmarked against known available algorithms in terms of the quality of the solution (maximizing profits) and computation time.
A mathematical formalization of this problem can be found in the embedded slides (in French).
For each of the scenarios, the data is available over a full year, and consists of:
- Electricity prices, hourly values
- Water inputs for each entry point of the system, in m3/s, day-averaged values.
A sample of such data is given in the plot below.
Each scenario is also characterized by a set of constraints, for example minimal thruflows, bounded reservoir volumes, production curves (function of thruflow and upstream reservoir volume), etc.
This challenge is sponsored by GE Digital.
The core optimization module can be deployed on Predix as a service accessible to third parties for a price following the platform’s pricing principles.
A more advanced app, providing direct business-value insights, could integrate one of GE business units offerings to its clients, in which case a licensing agreement would be concluded between the app developer and GE.