Improvements to the Modeling of Theoretical Power

Theoretical power is a fundamental pillar of performance analysis on the Novasense Portal. It represents the calculated output of a solar plant based on weather conditions and the technical characteristics of the plant. Two approaches coexist for calculating it: a physics-based model, comparable to a detailed simulation, and a machine-learning model that relies on the actual historical production of each inverter. The improvements presented in this article mainly concern the physics-based model, which is the most widely used modeling method on the Novasense Portal.

Overview of Improvements

Reduction of Seasonal Bias

The previous model had a strong tendency to significantly overestimate winter production. The new model is optimized to represent much more accurately the real-world performance of typical rooftop plants with moderate panel tilts up to 20°. This configuration corresponds to the vast majority of plants monitored on the Novasense Portal. Of course, the model continues to deliver excellent results for steeper tilts as well.

Consideration of Degradation

Until now, the modeling relied on a degradation assumption corresponding to that of a practically new plant. As a result, production from older plants was systematically overestimated. The new modeling now incorporates the plant's commissioning date and applies a fixed degradation rate of 0.5% per year of operation. This value is based on several meta-studies and aims to represent not only module degradation, but also the increase in mismatch losses over time. It does not, however, account for losses that can be mitigated through maintenance, such as evolving soiling or decreasing component availability.

Consideration of Maximum Inverter Power

Going forward, the nominal — or maximum — power of inverters is also integrated into the calculation. This enables more realistic modeling for plants with heavily undersized inverters, an increasingly common situation due to injectable power limitations on the grid.

Consideration of Power Limitation

The artificial limitation of plant power to comply with grid connection constraints is a rapidly expanding practice. It is now possible to integrate these constraints into the anomaly categorization as well (power control). If you would like to enable this feature for your plants, please feel free to contact your reseller's support.

Machine Learning

Improvements are also being made regularly to plants modeled using machine learning. Much like conversational agents based on large language models, machine-learning models can occasionally hallucinate or be imprecise when the dataset used for training is not of sufficient quality. Optimizations at the model training level have been implemented, particularly for plants whose power is dynamically controlled.

Implications for Users

As of the deployment of these changes in mid-July 2026, customers can expect more accurate theoretical power. A jump in the calculated data is therefore possible, particularly with regard to the Energy Performance Index (EPI), from this date. Please note that the Performance Ratio (PR) calculation is not impacted in any way, as it relies on potential power rather than theoretical power. To learn more about these indicators, please visit our dedicated KPI calculation page.