Renewable Energy Project Engineers have a lot of responsibility. They have to accurately predict the weather over the next 30+ years, design a power plant to minimize both construction and operation costs, and manage studies covering air space 500′ above ground to soil characteristics 30′ below it. With this diversity in workload comes a pressure for accuracy; small mistakes with these studies can be very costly. A 1% error in energy prediction for a 100MW facility can cost over $2 million in lifetime revenue.
Due to the range of tasks required, the ideal engineer must be well-rounded, with a strong engineering background. Typically, they have a degree in mechanical, civil, or environmental engineering. While perfect for building the skills needed to manage a project’s engineering studies, these curriculums don’t typically teach proficient Data Science skills, such as how to program, advanced statistics, and machine learning. The result is a workforce of renewable energy engineers processing massive amounts of data without the proper training (or time) to do so effectively. This poses a problem for the clean energy revolution.
But is this really a problem? Energy assessment software exists, and Excel templates can be used to tailor their outputs. Machine learning sounds unnecessarily complicated, just a buzz word that will require us to hire software engineers to slow down project development with no added value. Somebody shouldn’t need a computer science degree for an engineering job.
Sure, software does exist, Excel templates do work, and project engineers don’t need to be software engineers, but the software that exists bites, Excel templates bite worse, and the skill level required to write machine learning code is marginally more than what’s required to make a pun with the word byte. The cost of upskilling a project engineer’s data analysis skills is nothing (free open-source tools, free training material, little time required to learn), and it results in significant profit for the development firm overall (less mistakes, improved productivity, and better project insights). Here’s an example.
A very common task for a project engineer is estimating the energy that will be produced by a project over the course of its lifetime given different layout constraints. This analysis is vital to the economics of a project and thus highly scrutinized, but a project can expect to run a hundred different assessments by the time it completes development. These assessments are often rushed and performed in batches, where ten iterations are run at a time, with months between each batch. Here is the traditional process:
- Onsite weather data must be compiled, cleaned, and analyzed.
- Satellite weather datasets must be compiled and analyzed.
- Models must be drawn across the onsite data and the satellite data, with a new weather dataset predicted as a result.
- This weather data must be fed into an energy assessment software with specific inputs set to represent a single iteration of the project layout.
- The output of the energy assessment software must be fed into an Excel template, which further processes the results.
- A report is created to summarize the results of the Excel model.
- Steps 4–6 must be repeated for each layout iteration of the project.
Typically, Steps 1–3 of this process involve downloading CSVs from a variety of web sites and plugging them all into a series of Excel templates, copying and pasting the data from one template to the next. Sometimes application-specific software is used in addition to Excel. The "cleaning" process happens by manually combing through thousands of data points to remove the ones you don’t like (or doing nothing at all because you don’t have the time) and the results of the analysis are highly subjective. If you had two engineers repeat this process (or the same engineer try to do the process twice), you will get two different answers. The model used to forecast weather data is often very rudimentary, but that’s not all bad because there is way too much uncertainty with this process to warrant anything more complex. All told, this process can take a day to complete and will require detailed review by another engineer.
Steps 4–7, if set up well, are very monotonous and don’t require any critical thinking. They are the equivalent of asking somebody to click the same spots on a screen repeatedly without slipping up once. Manual mistakes can and often do happen, especially when this is done under a time crunch (which is every time).
Again, to be clear, this process works. Good engineers are diligent enough to catch the big mistakes and can write off all the small ones to uncertainty. Yes, the time sink is hefty, but the engineers are just expected to stay in the office late to get it done. Most importantly, anybody can be trained to do this.
Meanwhile, the margin on renewable energy projects continues to slim. The market is getting more and more competitive, and uncertainty in energy prediction must necessarily tighten if a project wants to stand a chance. On top of that, workload for engineers is growing exponentially, and any time they spend furiously clicking on a screen could better be spent, well, Engineering.
Fortunately, the entire process outlined above could be completed in the time it takes an engineer to get a cup of coffee. They would simply need to specify the layout scenarios they want to run and hit play. When they get back, they can put their caffeinated energy to use critically analyzing the results, something they couldn’t afford to do before. They can trust that the analysis has no manual errors, less uncertainty, and is completely reproducible by anybody on the team.
What is this black-magic button? It is something that can be built with no formal training and at no expense. Any engineer could write a simple program to automate this entire process, from grabbing the data, to analyzing it with advanced statistical models, then creating interactive visuals and attractive reports. That engineer can even tailor the program perfectly to match their company’s workflow, something 3rd party software could never do. I know because I’ve done this myself.
The benefits to building engineers’ data science skills go much further than this example. Fully automated, reproducible workflows save countless hours and mistakes, and sophisticated models (like machine learning) can unlock business insights with data that is currently underutilized. There are better ways for Renewable Energy project engineers to analyze their data, and better ways for development firms to use their engineers’ time.
These skills are remarkably easy to learn, and any project engineer could pick them up without needing a computer science degree. Putting an emphasis on this training will pay dividends to any engineering team. Knowing this, a harder question to answer is "why should engineers NOT study data science?"