AI-Powered PV Analysis for Solar Parks
How energy-weighted AI analysis makes hidden yield losses visible
Many solar parks appear unremarkable in monitoring, yet still lose energy yield. Jaroona uses energy-weighted AI analysis to reveal economically relevant deviations and prioritize maintenance in a targeted way.
The Problem: Hidden Energy Losses During Operations
Photovoltaic systems do not always lose performance suddenly. Many relevant issues develop gradually—individual module strings produce less yield, modules age at different rates, soiling affects output only at certain times of day, or MPP tracking operates inefficiently under specific irradiance conditions.
Partial Shading
Trees, structural elements, soiling, or temporary obstructions can significantly affect individual strings. Shading is especially critical during high-yield midday hours.
Module and String Issues
Defective cells, faulty bypass diodes, contact issues, or damaged connectors often lead to local performance losses that are barely noticeable in the overall picture of the system.
Gradual Degradation
Modules do not always age evenly. Certain areas of a system can lose performance faster without the total yield dropping dramatically right away.
Inefficient MPP Tracking
Maximum power point tracking is intended to find the system's optimal operating point. If it works incorrectly or too slowly under certain conditions, permanent efficiency losses occur.
In large solar parks, a single underperforming string can disappear in the overall picture. If it is not detected, the loss accumulates over the years of operation into significant financial damage.
Why Conventional Monitoring Often Misses Relevant Deviations
Many monitoring and SCADA systems analyze PV data at fixed time intervals. One minute in the early morning is treated statistically the same as one minute at midday. From a technical perspective, that is simple and understandable. From an economic perspective, however, it is problematic.
A performance drop at 07:00 has a different meaning at low irradiance than the same performance drop at 12:00 during peak production. When both events are weighted equally, it creates distortion.
The key point
Not every minute is worth the same. An analysis that treats every minute equally can misjudge the economic impact of a deviation.
Problem 1
Low-yield periods distort the analysis. Morning and evening hours make up a large share of the time series, but contribute relatively little to daily energy output.
Problem 2
Critical midday losses are underestimated. Especially during high irradiance, the greatest economic losses occur.
Problem 3
Relevant deviations disappear in the noise. Without economic weighting, it is difficult to distinguish statistical noise from truly relevant performance losses.
The Jaroona approach: energy-weighted AI analysis
The energy-weighted analysis reverses the classic logic. It does not only ask: “When did an anomaly occur?” but above all: “How much energy yield was affected at that moment?” This brings actual economic relevance to the center.
1
Time-based analysis
Every measurement counts equally. An anomaly at 07:00 and one at 12:00 are treated statistically in a similar way.
2
Energy-weighted analysis
Every measurement is weighted according to the energy yield affected. High-yield hours automatically receive a higher weight.
3
Economically precise assessment
Anomalies are not only detected technically, but prioritized according to their actual yield loss.
In practice, this means: A string that runs normally for 90% of the day, but significantly underperforms during the midday peak, may not be prioritized by classical systems. The energy-weighted AI analysis, on the other hand, recognizes that this exact anomaly is economically relevant.
Illustrative Dashboard Views
The following diagrams show, by example, how an AI-powered PV performance analysis can be displayed in a dashboard. They visualize key evaluations: the difference between time-based and energy-weighted analysis, the detection of conspicuous strings, the prioritization of plants by economic damage, and a possible alert and notification system in the event of outages or underperformance.
The displays are based on illustrative sample data and show how operators can identify technical deviations faster, assess them economically, and plan maintenance measures more precisely.
From Deviation to Technical Cause
A simple alarm is often not enough for the operation of a solar park. What matters is not only that a deviation exists, but why it occurs and what action should follow.
Processed Data Sources
- String current and string voltages
- Inverter power and MPP behavior
- Irradiance and weather data
- Temperature data
- Historical performance data
- Changes over days, weeks, and months
- Maintenance and fault information
Detected Patterns
From this data, the system identifies patterns that are difficult to access with traditional analyses:
- Recurring performance deviations in individual strings
- Gradual degradation
- Time-of-day-dependent shading effects
- Anomalous control behavior at the inverter
The goal is not to create additional data overload. The goal is a prioritized, technically actionable recommendation.
Underperformance of a Single String
The Scenario
String 34 shows 12% lower performance compared to comparable reference strings. However, the deviation does not occur evenly throughout the day, but mainly during periods of high irradiance.
A classic monitoring system might still classify this string as largely unremarkable. The energy-weighted AI analysis, by contrast, recognizes that the deviation occurs precisely during the economically most important hours.
Possible Technical Causes
- Module defect or cell degradation
- Partial shading
- Foreign object on modules
- Faulty bypass diode
- Contact problem
- MPP control problem at the inverter
The operational advantage lies in prioritization: instead of carrying out a blanket inspection of all strings, maintenance can be targeted specifically at String 34. This reduces search effort, shortens response times, and increases the likelihood of quickly resolving the economically relevant fault.
Economic Impact: Small Deviations, Big Effects
Even a few percentage points of performance improvement can have significant economic impacts in commercial PV systems and large solar parks.
1–1.5 MW
Single System
Small improvement, but already a noticeable increase in yield with 3–10% avoidable losses.
10 MW
Solar Park
Clear economic leverage - performance losses quickly add up to significant amounts.
100 MW
Fleet
Performance losses can add up to several hundred thousand or even millions of euros per year.
The AI analysis acts like a virtual additional PV output. No extra modules are installed, no additional space is required, and no new hardware is needed. The added yield comes from better data analysis, faster fault detection, and more targeted maintenance.
Predictive Maintenance: From Reacting to Predicting
Classic maintenance often responds to visible faults, alarms, or periodic inspection schedules. This is necessary, but not optimal. Modern PV portfolios require a more predictive operational approach.
By linking SCADA data, real-time sensor values, weather data, cleaning logs, O&M tickets, maintenance records, and component history, a much better basis for decision-making is created.
Early Warning
Technical issues are detected before they accumulate into significant yield losses. AI identifies typical precursors to faults such as slowly declining string output or recurring deviations under certain weather conditions.
Prioritization
Maintenance is not scheduled according to rigid routines, but based on economic relevance. Resources are deployed where the greatest impact is expected.
Availability
Unplanned outages and long response times can be reduced. The operational principle is: maintenance happens before yield is lost — not after.
Combining with Drone Inspections
Operational data analysis and drone inspection complement each other ideally. Operational data analysis shows where a system is underperforming and how economically relevant the deviation is — but not always which physical defect is actually present. Drone inspections, on the other hand, reveal physical anomalies such as hotspots, module damage, soiling, or shading.
The Efficient Diagnostic Chain
01
Data Detection
SCADA and string data show a relevant deviation in system operation.
02
AI Prioritization
The energy-weighted AI analysis prioritizes the affected string according to economic relevance.
03
Targeted Drone Flight
The drone flies specifically to the conspicuous area — no blanket inspection of the entire system.
04
Physical Confirmation
Thermography or RGB images confirm the physical defect on the module or string.
05
Repair Recommendation
Maintenance receives a concrete, data-based repair recommendation with clear prioritization.
Instead of checking hundreds of modules indiscriminately, the drone is deployed exactly where the data analysis suspects the greatest economic damage. The inspection effort is significantly reduced.
Portfolio Analysis for Operators of Multiple Plants
The benefit is especially strong at portfolio level. Operators with many sites face a central question: Which plant is currently losing energy yield — and where is the economic damage greatest?
Without a standardized, normalized analysis, this question is difficult to answer. Plants differ by location, age, technology, irradiance profile, maintenance history, and operating conditions. The Jaroona analysis creates a comparable view across all plants.
Normalized Benchmarking
Plants are compared fairly, even though they have different site conditions.
Automatic Flagging
The platform automatically identifies plants that are significantly below expectations.
O&M Prioritization
Maintenance budget and deployment planning are concentrated where the greatest economic impact is expected.
Portfolio-Wide Transparency
Management, operations, and asset administration receive a shared basis for decision-making.
Especially with large portfolios, this prioritization can make the difference between reactive administration and proactive performance management.
Integration into Existing System Landscapes
The solution is not intended as a replacement for existing IT systems, but rather as an intelligent analytics layer on top of existing data sources. A system change is usually not necessary.
SCADA Systems
Historical and live operating data are automatically retrieved and analyzed. No manual data transfer is required.
Inverter APIs
String-level data can be integrated directly from inverters or manufacturer platforms.
O&M Software
Analysis results can be transferred as prioritized tickets or action recommendations into existing maintenance systems.
Asset Management Systems
Performance metrics, yield forecasts, and optimization potential can be integrated into asset reports and executive dashboards.
This preserves the existing operating structure. The AI complements existing systems with a financially focused analysis and prioritization layer.
Who is the solution especially relevant for?
The energy-weighted AI analysis is especially relevant for organizations that professionally operate, manage, or finance solar assets.
Solar park operators
Operators of ground-mounted systems benefit from earlier fault detection, fewer hidden yield losses, and more targeted maintenance.
Utilities and IPPs
For companies with PV systems in their own portfolio, the solution is a tool for securing returns, optimizing operating costs, and increasing performance transparency.
Infrastructure and energy investors
Investors need reliable performance data for due diligence, valuation, reporting, and sale preparation.
O&M service providers
Maintenance companies can expand their services with data-driven insights and offer customers not just reactive repairs, but prioritized performance optimization.
Scalability: From a Single System to a Large Portfolio
The methodology is platform-independent and scalable. It can start with a single system and later be expanded to large portfolios. In this way, the platform grows with the operator: from the first system to hundreds of sites in multiple countries.
Single System
Detailed string analysis, root-cause identification, and maintenance prioritization for a single PV system.
Solar Park
Analysis of thousands of strings, automatic detection of underperformance, and targeted inspection planning at the park level.
Portfolio
Site-wide benchmarking, ranking by optimization potential, and portfolio-level resource prioritization across all locations.
Long-Term Perspective: Solar Plant Management of the Future
The energy-weighted AI analysis is a starting point for a more comprehensive development. In the coming years, PV systems will increasingly be controlled in a data-driven, automated, and predictive way.
1
AI-Optimized Inspection
Hotspots and performance deviations are automatically correlated. Inspection intervals are dynamically adjusted to the actual condition of the system.
2
Proactive Maintenance Planning
The system predicts failure probabilities and schedules maintenance measures before yield losses occur.
3
Standardized Portfolio Benchmarking
Systems are made comparable by location, age, technology, and operating conditions.
4
Dynamic Digital Twin
Each system receives a digital model that can be used to simulate maintenance measures, component replacements, or expansions.
This is how traditional monitoring evolves into active solar plant management that connects technical data, financial metrics, and operational measures.
Conclusion: Better Data Analysis Instead of More Hardware
Many yield losses in solar parks do not occur because data is missing, but because existing data is not evaluated economically enough. Classic monitoring detects obvious faults, but often overlooks deviations that only appear under specific irradiance conditions and are precisely therefore especially relevant.
Jaroona’s energy-weighted AI analysis takes exactly this approach. It evaluates performance deviations based on their actual impact on energy yield, identifies technical patterns at the string and plant level, and prioritizes maintenance according to economic damage.
For operators, energy companies, investors, and O&M service providers, this creates a more precise, scalable, and economically focused approach to operating modern PV systems.
The Central Advantage
More yield does not come from additional modules or additional area, but from smarter analysis, faster fault detection, and targeted measures.
Economically focused
Deviations are evaluated based on actual yield loss — not by timestamp.
Technically precise
Patterns at the string and plant level are detected and assigned to technical causes.
Scalable
From a single plant to a large fleet — without system changes and without new hardware.
Identify hidden yield losses in your PV portfolio?
Start with a pilot project based on your existing SCADA, inverter, or O&M data. Jaroona analyzes your plant’s optimization potential and shows which deviations are truly relevant from a financial standpoint.
Use existing data
No system change, no new hardware — we work with your existing SCADA and inverter data.
Fast start
The pilot project quickly reveals where financially relevant deviations exist in your plant.
Clear results
You receive a prioritized overview of optimization opportunities — concrete, understandable, and action-oriented.
Your Partner for AI & Machine Learning
Tailored AI solutions for businesses — from strategic consulting to successful implementation.
© 2026 Jaroona. All rights reserved.