The complexity of the network, paired with factors like lag time and multiple environmental variables made this system a perfect candidate for model predictive control (MPC).
Protecting infrastructure — As precious as gold or oil - the control of water being released for human, livestock or environmental use is critical for such a scarce commodity.
Accounting for multiple variables — Factors like farmers purchasing water for irrigation downstream, rainfall runoff and evaporation all make it more difficult to predict flow and water levels.
Calculating for lag time — There is a time lag of up to 10 Hrs from changes made to release flows at the dam, to when those changes are seen at the farthest compliance point.
Supporting compliance and water quality — As a public utility and a purveyor of a critical natural resource, Melbourne Water puts water conservation at the core of its business.
While the water industry is relatively mature when it comes to automation and system optimization, quality, process variability, outages and downtime are still persistent challenges.
The complexity of calculations, the duration of focus and the speed and precision required are poorly suited to human capabilities. MPC applies a data science layer on top of existing regulatory controls to continuously monitor and predictively optimize process behavior.
MPC analyzes a set of variables (e.g. flows from tributaries, rainfall, evaporation, irrigation consumption, etc.) in real time and continuously, predicting and driving the set point adjustments to maintain optimal system function, removing the strain of extended observation from conventional operation eliminating the root cause of variability, outages and downtime.
A primer on desalination
- Some of the demand for water in Australia is met through a range of desalination plants converting sea water into freshwater, a costly process and massive energy draw.
- The process uses reverse osmosis, resulting in very hard water. Remineralization is required to soften it and improve taste, which increases costs.
- The energy required to pump water to the end consumer is yet another additional cost.
Model predictive control reduces process variability to achieve plant obedience and manages the uplift within process constraints
The outputs from the Pavilion8 system are adjustments to set points in the existing regulatory control system
The normal variability present in any operating process results in a standard deviation from desired control targets
With model predictive control we reduce the variability and standard deviation
Once the process is under tighter control we're able to operate closer to process specification limits
Melbourne Water saved billions of liters of water every year with model predictive control
Our work made it possible to save 2 billion liters of water annually, conserve energy and reduce strain on the workforce while improving compliance. We helped Melbourne Water set the bar for autonomous operations and responsible stewardship over a critical natural resource.