Operational Energy Yields

Focusing analyst eyes where they are most needed

For much of my career, prior to cofounding Clir at the start of 2017, I worked as a consultant in the renewable energy industry. As an energy analyst in the wind consulting world, energy yields (also known as EYs or energy assessments) were our bread and butter. Our department head referred to the EY process as a meat grinder. This has stuck in my head because it really rings true.

Consider:

• The input data for an EY comprises millions of data points representing the wind farm’s operational history and meteorological conditions at or near the site.
• This data is run through a series of calculations, i.e. the meat grinder.
• The most important output from an EY is a single number indicating the estimated long-term annual energy production of the farm (net P50). In other words, how many GWh of energy is the wind farm expected to produce per year going forward? Understanding this is key to understanding the performance and value of a wind farm.

Below I’ll outline Clir’s process for operational EYs. First though, I’ll start with a high-level review of Clir’s earlier EY process and the associated challenges.

Previous Energy Yield Process and Challenges

Clir’s previous EY process was/is fairly manual. Analyst expertise and judgement are required throughout the calculations. As a software company, Clir is keen to increase automation and reduce the amount of human involvement required, with the goal of scaling what we can deliver and our impact on the wind industry. The EY calculations are straightforward to automate, and properly labeled input data is easily available for any farm on the Clir App (thanks to the Clir data model). However, our earliest attempts at automating the EY process were not completely successful.

The challenging part of EY process automation is that the analyst is required to make certain decisions during the calculations, within the meat grinder, as seen in the figure below. There are often site-specific data integrity issues and other important considerations which require the analyst’s expertise on how to apply EY best practices. Automating the analyst’s judgement and discretion is challenging (as is getting analysts to move away from Excel towards a more scalable solution).

Clir’s previous energy yield process

New Energy Yield Process

The key breakthrough towards increased automation of the EY process was separating the analyst from the meat grinder. A new approach was developed whereby EY results are automatically calculated for all EY scenarios under consideration, of which there are often thousands. An EY scenario is defined by a unique set of EY calculation inputs and parameters. One example scenario is described in the table below.

One example energy yield scenario.

The number of possible scenarios in an EY analysis depends on the inputs available, which varies by farm. The figure below summarizes the different sets of inputs and parameters available for one example farm. In this case there would be 1,728 EY scenarios under consideration (4 x 4 x 4 x 3 x 3 x 3).

Inputs and parameters for an example energy yield analysis.

In this example, EY results are automatically calculated for each of the 1,728 scenarios. The analyst then reviews the precalculated EY results, in conjunction with Clir EY best practices and farm-specific considerations, to determine which scenario best represents the future production of the wind farm. The new process is illustrated below. Note that although the process is referred to here as new, it has been in place at Clir since earlier this year and it represents our current EY process.

Clir’s current energy yield process.

With results available for all scenarios, the analyst is able to quickly and easily carry out a sensitivity analysis to determine which inputs and parameters are most important. This ensures the quality of input data used and mitigates undesired influence of inconsistent or non-representative input data on the EY results.

Considering thousands of scenarios may be seen as adding complexity. However, these scenarios exist and should be considered regardless of which EY process is used for a given analysis. With Clir’s previous method, which is similar to the industry-standard method, the analyst would be expected to somehow consider all these scenarios and determine the preferred one. This often resulted in a repetitive and time-consuming process where the analyst:

• Calculates EY results for one scenario.
• Reviews EY results for that one scenario.
• Hums and haws about whether the inputs and parameters are appropriate.
• Repeats until the analyst and reviewers are happy.

This repetition is avoided with Clir’s new process because EY results for all potential scenarios are readily available at the analyst’s fingertips and can be easily compared and contrasted. Several visualizations of EY results from the new process are shown below.

Distribution of long-term annual energy production (net P50) considering all scenarios, coloured by the production source used for the scenario. In this case substation data is identified as a clear outlier, likely representing a data integrity issue. The median scenario net P50 is roughly 314 GWh/annum.

Sample heat map of long-term annual energy production (net P50). Each rectangle represents one scenario. The colour of the rectangle indicates the net P50 result in GWh/annum for that scenario. This heat map shows the influence of i) period subset, ii) data coverage threshold, iii) total loss factor threshold, and iv) reference source on the farm’s estimated long-term annual production (net P50). From this visualization it is evident that the farm performed better during the first half of the analysis period relative to each source of reference wind data. It is also observed that the estimated net P50 increases as the data coverage and loss factor thresholds are increased.

Distribution of correlation quality (r-squared) considering all scenarios, coloured by the reference wind speed source used for the scenario. In this case the ERA nodes correlate better with farm gross production than the MERRA-2 nodes across a wide range of scenarios.

The Result

The new process has become Clir’s standard approach and has already been used successfully for multiple farms. Benefits from the new process include:

• Faster delivery of EY results. Analyst time is focused on exercising expertise and judgement rather than running calculations.
• More robust and defensible results, with a wider and more holistic view of possible outcomes and a clearer understanding of the sensitivity of the results to different inputs and assumptions.
• Final results can be compared against certain benchmarks such as the median scenario to gain further confidence in the estimated long-term annual energy production.

I would like to thank Andrew Cameron and the Clir Renewable Analytics team for their collaboration on this initiative. Next steps include applying the new process to many more farms and expanding the process to encompass EY uncertainty analysis and the reconciliation of expected and observed lost production by cause.

Interested in working at Clir? Have a look at our open positions!