What the IEA PVPS update covers
A September 2025 joint fact sheet from IEA PVPS Tasks 13 and 16 examines understanding, measuring and mitigating soiling losses. It includes on-site sensors, modelling and machine learning approaches, alongside cleaning optimisation and preventive measures during project design.
The fact sheet frames soiling as a source of energy loss, operating cost and forecast uncertainty. Its global context does not establish the recoverable loss at an individual plant. This 2026 review therefore focuses on the evidence connecting a local estimate to an actual maintenance decision.
Define what the model is estimating
Before comparing algorithms, specify the target: a soiling ratio, expected energy loss or a cleaning decision. Record how the reference measurement is maintained and how missing observations are treated. A model trained against an uncertain label can produce a precise-looking number without resolving the field uncertainty.
For the site review, keep irradiance, temperature, outages and operating restrictions available alongside the soiling estimate. A drop in power should trigger diagnosis rather than an automatic assumption about dust. Ask the analyst to explain how competing causes are excluded in the validation dataset.
Separate prediction quality from operational gain
A useful algorithm comparison uses a held-out period and states the error measure, seasons covered and data exclusions. Compare against a simple, documented baseline as well as other models. A better prediction score establishes a forecasting result; it does not alone measure additional energy delivered by cleaning.
To evaluate the intervention, record cleaning dates, affected rows, costs and the post-cleaning comparison method. Account for changing weather and any simultaneous maintenance. Keep prediction accuracy and recovered energy as separate results so that an operator can assess where the value actually arose.
Turn the estimate into a site decision
For an operating trial, agree how expected recoverable energy will be compared with cleaning cost and practical constraints. Include water availability, access, approved equipment and the forecast time until the next review. Define how uncertainty changes the decision when the estimated benefit is marginal.
The fact sheet's emphasis on site-specific mitigation supports evaluating local conditions. It does not establish one cleaning interval for all plants. Ask for operating records that show why a chosen schedule was appropriate for that site and season before carrying it into another project.
What the next update should disclose
A strong field report would identify the model version, sensor arrangement, monitoring window and intervention baseline. It would report both forecast performance and the net outcome of maintenance, including cases where cleaning was deferred or produced little measurable benefit.
For a buyer, request access to those records and an acceptance method tied to the intended plant. Avoid contractual guarantees based on a generic AI yield or maintenance-saving percentage. The useful improvement is a documented decision process whose performance can be checked against local measurements.
Frequently asked questions
Does AI guarantee a fixed PV yield increase?
No. A site must establish the initial loss, the intervention and the measured net outcome.
Is forecast accuracy the same as recovered energy?
No. Forecast accuracy measures prediction error; recovered energy requires a separate comparison of the maintenance intervention.
What date applies to the cited fact sheet?
IEA PVPS dates this joint Tasks 13 and 16 fact sheet September 2025; it is reviewed here in October 2026.
Use the news as a project input, then verify the scope.
This News item is editorial context. Confirm the final datasheet, applicable standard, project conditions and acceptance evidence before changing a design or purchase decision.
Discuss a project implication