On March 18, 2024,
IOM - Solutions
held a virtual Workshop on AI Applications in Fault Detection and Maintenance Strategies for PV and CSP Technologies. In this Workshop event, we covered the latest developments in O&M of solar power plants and how AI is integrated into this field.
PVRADAR presented by its CEO -
Thore Müller
- spoke about: “Physics-based AI is revolutionizing soiling estimation and cleaning optimization in large scale PV”.
“PVRADAR is a software platform for transparent assessment of soiling and snow related performance losses, optimization of cleaning strategies and modeling of economic scenarios.
PVRADAR gives project developers the tools to quantify lifetime risks and find cost-optimal solutions starting on the first day of development: assess soiling losses based on satellite data, select a cost-optimal cleaning strategy from dozens of alternatives in our database, simulate bifacial gain improvements from the use of albedo enhancing materials.”
Muller spoke during its presentation on how we can use methods like AI, and machine learning to predict conditions during the design phase of the power plant and prepare in the most optimal way to mitigate them, maybe as early as the design, and then be prepared to when they occur during operations.
Muller mentioned a recent study by KWh analytics claiming that “Solar assets are underperforming by an average of 8 %” based on PV magazine article published on May 3, 2023. For Muller, the average PV project is designed based on standard values for climate-related losses that bear no relation to reality. In this regard, PVRADAR empowers project developers to achieve world-class performance and reliability across all climatic zones, transparently and traceable.
Muller mentioned that at PVRADAR they set the course for world-class performance through three main topics: Soiling risk assessment, cleaning cost optimization, and Albedo improvement.
During the presentation, Muller focuses more on the first topic which is soiling risk mitigation. PVRADAR develops and provides soiling mitigation reports for a specific location. They offer analysis of 12 months of soiling data for a specific location, based on advanced models that predict soiling with high accuracy and provide insight into how you can mitigate it and make savings.
PVRADAR software can perform high-precision soiling mitigation for a specific site instead of relying on 2%, a standard value that cannot reflect the reality of some regions. The software can optimize the cleaning of this area and thus reduce soiling losses to a much lower level and make decisions during the operation of solar power plants.
Captured From Muller's talk
In the second part of the presentation, Muller explained why it is important to assess soiling and optimize cleaning during the design phase. Taking a solar project in the design phase when there is no data and measurements of soiling on this site, the majority relies on general assumptions, for example of soiling losses being 2% every month to then estimate the cleaning costs. Muller said that by using this standard expectation, soiling is wrongly estimated, and a lower cleaning budget is associated, leading the operations team to subsequently face several problems in maintaining high performance of the solar power plant. This lack of estimation of cleaning budgets leads EPC to opt for cheaper cleaning solutions and strategies which also impacts subsequent performance. The solution is to be therefore able to make data-driven decisions. Muller claimed that it's important to start as early as possible to avoid very costly reiterations, as the later the design needs to be changed, the more costly it is.
Muller mentioned that at PVRADAR they use advanced modeling techniques to provide holistic techno-economic optimization of the cleaning strategy for power plants and to determine which cleaning technology and frequency are the best.
Muller explained how data, algorithms, and AI can help us to solve the problem. The main relevant models’ inputs are soiling conditions, Meteorological conditions, PV plant technical design, PV plant market environment, cleaning technologies technical parameters, and finally cleaning costs.
Captured from Muller's talk
The solution developed by PVRADAR is based on 5 essential steps:
Step 01: Define the problem based on the project location and design.
Step 02: Estimate historic soiling losses.
Step 03: Define cleaning alternatives (Dry or wet cleaning; Robotic or Manual cleaning…).
Step 04: Simulate the future based on model's long-term yield and soiling losses.
Step 05: Cleaning optimization and costs (define the cleaning systems, machines numbers, and frequency…).
Muller then presented how the developed models are validated using both satellite and ground measurements. Muller also presented the methodology followed for a real project that PVRADAR worked on.
