The power of artificial intelligence to shape the future of solar industry

In the solar PV industry, AI can contribute to processing the considerable amount of data coming from PV projects more quickly and efficiently. The input drawn from this analysis improves asset management and constitutes a powerful strategic tool to guide the project owners’ decision-making processes to maximise their return on investment. Some of these AI-based processes are related to automated inspections, energy forecasting, demand response, and performance analytics, and include:

1. Drone infrared (IR) flyovers: instead of using representative samples, fleets of drones make it possible to perform IR scans of entire PV sites, thus optimising the inspection time. The raw data created by these sitewide scans are then analysed by AI models; and then, if needed, further ground IR testing may be performed on specific areas.
    
2. Smart” Data Acquisition System (DAS) integration: Alarms have always been a part of DAS, but AI contributes to improving the accuracy and relevance of the information provided. Some Smart DAS service providers now generate monthly reports with detailed information and recommendations, such as optimal washing dates of certain module strings, or inform of systemic inverter-specific issues, together with the number of days after which critical failures are to be expected.
    
3. Machine Learning (ML) performance analytics: Similar to Smart DAS, stand-alone performance analysis has started to rely more on ML techniques in order to discover the aforementioned systemic issues. With data available sometimes in a 1-minute (or even sub-minute) temporal resolution, that can be combined with data coming from some years of operation, ML models are becoming robust enough to be relied upon for noteworthy observations.
    
4. Module defect detection: Anomalies in solar panels can be identified using special cameras that capture IR and electroluminescence (EL) images. If defects are present in a module, specific patterns will be generated, making it possible to identify and classify them. With these metrics, it is possible to generate robust ML models to automatically analyse and point out the areas affected by the defects, also determining their impact on the performance of the device.
    
5. Energy forecasting: AI models can be used to forecast the expected production output of PV projects, sometimes days ahead of real-time. This calculation relies on the quality of the weather forecasts, and it is used by owners of power plants and utilities to plan daily operations.
    
6. Demand response: Using similar tactics, utilities employ end user consumption data to predict future demand on the grid. This enables utilities to determine how much more or how much less energy will be needed. Depending on the situation, operating PV plants can be curtailed due to oversaturation, or additional power stations (e.g., traditional power plants) can be fired up to meet the energy demand.
    
7. Performance analytics: AI can be used to monitor and optimise the performance of solar PV projects in real-time. AI algorithms can be used to analyse data from sensors and other monitoring devices to identify issues and inefficiencies and to recommend adjustments to improve performance.
    

At Enertis Applus+, specific AI-based tools like the Advanced Performance Analytics Application (A-PAA) and Smart PV Inspection Tool (SPI) have been developed to perform tasks related to items (1), (3), and (4).

Part of running a PV plant involves properly maintaining the equipment onsite to maximise power output. With several thousands of modules and hundreds of inverters on any given site, this task has proven to be very challenging. Until recent years, PV project owners had to send a licensed professional to the site to perform a walkthrough and take photos and infrared scans of a representative sample of modules. The expert would then make general recommendations on how to proceed with fixing and replacing modules, or even reset the expectations for power output. These processes may take a long time, especially when modules must be shipped to a third-party testing laboratory. This means that it may be weeks before systemic issues are discovered and solved, thus negatively impacting potential project financing opportunities, as well as Power Purchase Agreement (PPA) and construction timelines.

Recent advances in technology and software development have provided a solution to accelerate such long processes by using a combination of drone IR flyovers and AI models. The Smart PV Inspection Tool (SPI), for example, brings speed and accuracy to IR drone inspections of solar assets. It is responsible for processing, in a matter of minutes, the thousands of images that are collected from IR flyovers, searching for multiple types of defects that may impact the performance of an asset, automatically classifying their criticality, and geolocating the faulty modules within the plant. It guarantees the delivery of the final results in a couple of days, allowing for a quick response from the project owners. Further time optimisation can be achieved using mobile laboratories, that can be sent to the PV plant to perform onsite testing of the faulty modules, saving time and money.

Asset performance analysis is another central point where AI is helping analysts to obtain more information from the data recorded from the SCADA systems. Traditional monthly performance analysis provides a more high-level view of system health and cannot inform on how much potential production was lost due to downtime events or other losses. The combination of advanced data science tools and ML models has opened new frontiers, allowing data processing at any component hierarchical level, and at small intervals such as 1-minute timestep.

An example of this is Enertis Applus+ Advanced Performance Analytics Application (A-PAA), used to calculate key metrics such as the real PR evolution for each component, how much energy has been lost during downtime events at the meter level all the way up to the string level, and the optimal behavior of the different device types based on their historical data. Separate analysis can also be conducted to determine losses due to different causes, such as long-term system degradation, utility-enforced curtailment, soiled modules, and non-functioning single-axis trackers, and to calculate actual Ohmic losses between the inverter and point-of-interconnection.

As with other AI tools, the results of an A-PAA run can be used to help fine-tune forward-looking P50 expectations, as well as to inform project owners and asset managers of the most likely culprits of the sub-performance. Another attractive capability of such architecture is its ability to process any type of information provided, regardless of the data format. Therefore, if a portfolio has multiple PV plants with different SCADA systems, they can all be integrated into one single platform, thus centralising the information, and allowing for a direct comparison of the performance across the assets.

To conclude, it is clear that AI applications are already playing a key role in transforming the solar PV industry, as they contribute to improving the efficiency and security of PV plants, and in pushing the transition towards a more sustainable energy system. Looking ahead, we expect research on Large Language Models (LLM) to explore their impact on our industry. There are exciting times ahead for clean power.  

Written by Lucas Viani, Data Science Team Manager, and Brian Custodio, Director, Data Science and Consulting for North America. Published in May’s issue of Energetica Magazine.