For Gilles Estivalet data is first and foremost a decision-making instrument, and it is of immense value for the optimization of a power plant’s performance.
Conclusions drawn from proper data analysis can help increase not only a plant’s technical performance, but also the performance of the team that manages it. These two aspects collectively make up the field of asset performance, and the mastery of this field is pivotal to reaching high levels of performance for any large-scale solar PV endeavour as each percent of additional performance contributes to additional levels of profitability.
Each additional percent has a tremendous impact on financial returns. Data cannot suffer any slight inaccuracy.
According to Gilles, data is the cement which unites operation, maintenance and asset management of solar portfolios and a key to understanding the true scope of asset performance. First is operation management, it includes of all tasks related to plant supervision, performance management and engineering. Data related to operation management take the shape of AC or DC data fetched directly from the plant. Operation data also determines the plant’s long term direction through quality auditing and viability analysis of upgrades, retrofitting, or even plant decommissioning. Linked to operation data is maintenance data, which includes the technical maintenance of the power plant’s hardware and software systems, such as curative and preventative measures. At the end is asset management, the layer of oversight that binds the two previous categories together. At its core, asset management covers the financial side of a solar power plant’s performance, from tax filings to insurance. Although there is significant overlap between these three categories, Gilles explains that they require different core competences and what unites these categories is the data they share.
In an ideal world, operations managers should understand the full financial implications of their operational decisions, and conversely asset managers should be thoroughly aware of the technical specifications that ultimately determine a plant’s financial performance. However, the three categories tend to be offered by different service providers. Research by Greentech Media shows that barely 50% of service providers offer comprehensive suites for both asset management and operations management.
Despite their differences, however, the three aspects of asset performance share one key component: they are driven by information. The cornerstone of any proper managerial decision is the data it is based on, and asset performance management is even more reliant on data due to the sheer volume of information involved in the operation of a solar PV plant. To illustrate just how large these data sets can be, Gilles recalled using over 4,500 indicators in the evaluation of the performance of a 300 MW solar PV plant. The clear majority of this raw data is initially scattered, and in this disorganized state its usefulness is limited.
“The challenge is to organize heterogeneous data into a single homogeneous system to ensure its accuracy, quality, security and universal availability. Said differently, the challenge is to transform raw data into valuable insights.”
Raw data is essentially operational data. In addition to the AC / DC data provided by panels and transformers, there is the abundant information provided by the plant’s sensors, like irradiance and temperature, as well as the plant’s engineering information. There are also third party data sources, for example weather forecasts or market projections. It is interesting to note that the level of accuracy of forecasts can vary considerably across the industry. As an example, Gilles explained that the forecasts used by RTE, the operator of Europe’s largest high-voltage transmission system, average around a 4-5% accuracy level. Simultaneously linking all these sources of data is a complicated task, but it is vital for the proper analysis of asset performance, especially for teams operating across countries or continents. Taking the ‘Pyramid process’ as a starting point does significantly help to get higher quality data.
Acquiring raw data can often become a surprisingly daunting challenge. The primary goal of data acquisition is to achieve low latency access to raw data using big data methods and solutions. The immense amount of input and output bandwidth required for such a system necessitates significant processing power and in some cases a powerful bi-directional API. The infrastructure requirements for correct data acquisition should not be taken lightly.
Once raw data has been acquired it must be verified for quality and accuracy, a process Guillaume calls data quality management. Since the location of most large scale solar PV plants is away from human activity, often these plants suffer from a lack of reliable data connections. Missing data can severely impact the quality of a dataset, so establishing a stable network connection is a vital part of data quality assurance. A poor network connection can also result in outdated data, which can be especially detrimental to time-sensitive indicators like temperature. Erroneous data is the hardest type to detect according to Guillaume, especially if the margin of error is low. He notes that modern machine learning techniques like data cleaning have significantly improved the discovery and resolution rate of such problems over the past few years.
When the data has been verified it can be analyzed to draw important conclusions. Analytics can range from simple charts and visualization tools to complex functions and algorithmic processing. What is truly important, however, is the step after data analytics. Data intelligence, as Guillaume aptly calls it, goes a step beyond analytics, which still relies on the interpretation of human analysts. Considering the immense volume of data involved in the management of solar PV assets, machine learning is the perfect tool to obtain further insight from analyzed data. The immense growth in hardware capabilities, especially processing power, has unlocked the ability of modern Artificial Intelligence (AI) to support big data. AI not only helps with analyzing data, but also with predicting future data based on the current patterns exhibited by, for example, the plant’s level of production or the price of the spot energy market.