Over the last few months, we have had some breakthroughs on improving the accuracy of our power plant monitoring. This is particularly true for for plants with many transmission lines connected in dense rights of way.
In these examples, you can see three market moving plants with a before and after picture of our generation estimates. Until now market participants had struggled with the available information on these plants and often were forced to make assumptions on their output levels or whether they were online at all.
Through a combination of our technology, persistence and enough historical observations, we have been able to dramatically improve our generation estimate on these plants. To all of our customers along this journey, we thank you for your constant reminders of how important it is to get this right.
The process of where to deploy sensors and how to combine the signals into an estimate of near real-time generation is a patented proprietary process. However, we like to share some basic steps and principles used to help our customers understand the basis of the data.
Our process includes:
- Deploying sensors along as many high voltage lines connected to the plant for which we can find a landowner.
- Screening of initial sensor data. Do we have noisy data? Are there too many lines interfering with each other?
- Examining data for historical operating patterns based on historical data from CEMS, NRC, ERCOT, EIA, EQR, and wind speed. Do we clearly see startups/shutdowns and capacity blocks.
- Choose training and validation periods and create calibration models for each facility.
- Peer review process - every released or re-calibrated plant goes through a peer review that often results in reworking the analysis or additional field work. We may want to add more sensors or move one to a different location. If the generation estimate looks good we will release a new plant as REVIEW status.
- Wait and observe how the values change over time and during line outage events. We assess if the calibration is stable or unstable during line outages and adjust to these different grid states.
- Review generation estimate against new reference data and recalibrate if necessary. Again we choose multiple training and validation periods and test the model again.
- Revise monitoring plan. What can we still work on to make this better? Conduct a field survey and begin to dis-aggregate multiple circuits?
We are always willing to share our strengths and weaknesses with our customers. In some cases, we may have a lot of noise at zero output. Or in other cases we may be very confident of the capacity blocks or individual generating unit output.
Customer Notes are provided in our interface to communicate all the relevant caveats around the data. Additionally, we post if something is going on like a line outage or even if our sensor is not working properly.
You can choose how to use our data with as complete an understanding as you need to assess the benefits or risks.
Restate Historical Data
As our final step we will restate history and publish that revised history to our customers. This is to provide the best and most accurate data series for analytical uses. In most cases, this is just the past few months of recent history.
In the event it is needed, we keep all the historical versions of each data record and can produce them upon request.
So, check it out and tell us if we are setting the bar higher for quality and accuracy.