Unlocking new standards in solar forecasting to maximize asset efficiency

At Amperon, we’re committed to driving innovation and delivering cutting-edge solutions to solve the energy industry’s most critical challenges. Since launching our asset-level solar forecast earlier this year, we’ve iterated and scaled a product that delivers industry-leading accuracy results, seamless data integration, and an intuitive user experience for forecasting generation from solar photovoltaic (PV) assets.

With enhanced machine-learning (ML) capabilities, new weather ensemble features, and improved physics-based modeling, our customers are experiencing 10% accuracy improvements on average. Our updated asset solar capabilities are setting a new benchmark for forecasting accuracy, empowering asset owners to optimize their operations and financial returns.

Why accurate solar generation forecasting matters

The rapid growth in global solar capacity has been a cornerstone of grid decarbonization. Solar asset owners, including independent power producers (IPPs), investor-owned utilities (IOUs), and public power entities are at the forefront of this transition, and are managing renewables as critical components of their generation portfolio.

While advancements in PV technology, global supply chains, and policy initiatives have all accelerated the deployment of assets, managing and optimizing those assets is an increasingly complex challenge. From balancing net demand as a load serving entity (LSE) to maximizing financial returns in wholesale markets, operational and financial success hinges on reliable and accurate generation forecasts.

With recent feature releases and improvements in our platform and our ML-based forecasting engine, Amperon is solving the following problems:

Managing Net Demand to Minimize Real-time Exposure: LSEs must track and manage their net demand (total demand minus renewable generation) to avoid financial exposure in the real-time markets and ensure regulatory compliance through resource adequacy and grid stability rules. Under-forecasting solar generation can create significant financial instability as LSEs are forced to procure electricity in volatile real-time price markets.

Optimizing PV Asset to Maximize Financial Returns: Asset owners, including IPPs, need accurate forecasts to confidently schedule energy generation in wholesale markets, optimizing their energy scheduling for day-ahead markets and maximizing financial returns.

What makes Amperon’s asset solar forecasts unique?

Forecasting solar generation is complex, influenced by challenges in weather forecasting, limitations of physics-based modeling, and shortcomings of ML-based techniques. At Amperon, we’ve made significant strides to address these challenges through recent product and engineering investments. On average, our forecasts now show ~10% relative improvement in capacity normalized mean absolute error, cnMAE, for existing customers. (Read more about cnMAE and other renewable energy forecast accuracy metrics here). Here’s what’s driving those accuracy improvements:

Amperon’s ensemble weather
Accurately forecasting solar relies heavily on precise weather data, including variables like irradiance, cloud cover, and temperature – among dozens of others – that impact generation output. Traditionally, errors introduced by third-party weather vendors negatively impact downstream generation forecasts. To address this weather error risk from any individual weather vendor, Amperon has developed a dynamically weighted ensemble weather model. This approach combines the strengths of multiple third-party weather vendors, leveraging their unique accuracy benefits in our final prediction while minimizing the risk of errors introduced from any single provider. This feature is available in our short-term forecast product for both asset-level solar and wind.

Amperon's weather ensemble and model improvement for asset-level solar forecasts

Integrating historical asset availability and curtailment data
Accurately forecasting generation requires a clear understanding of drivers of past performance, especially periods of underperformance (i.e., lower-than-expected output). For solar asset operators, it’s crucial to distinguish whether reduced generation was caused by weather-related factors or external variables like utility-mandated curtailment or planned maintenance downtime. To address this challenge, Amperon now allows customers to upload historic curtailment, potential power, and plant availability data. This additional data improves our ML models by ensuring they correctly attribute performance issues and ignore exogenous factors like curtailment and maintenance. This feature is available in our solar short-term forecast, solar sub-hourly forecast, and wind short-term forecast.

Model improvements to incorporate an asset’s physical attributes
Traditionally, the solar industry has relied on physics-based modeling to estimate generation output, using physical attributes and weather forecasts as key inputs. While valuable, this approach has inherent accuracy limitations. At Amperon, we’ve leveraged our expertise in ML-driven forecasts to enhance asset-level renewables forecasting. By incorporating physical characteristics of the solar panels themselves into our ML models, we’ve significantly improved training and prediction accuracy. This hybrid approach is particularly effective for assets with limited historical generation data. This feature is available in our solar short-term and sub-hourly forecast products.

‍