Lavorro’s AI/ML driven Analytics solution for Solar Energy Plants

Figuring out methods to maximize electricity output from solar panels is a challenge for utilities and solar developers alike. Small variations in daily performance can significantly affect the long-term output of an array. Three variables that can help power providers manage expectations and maximize performance of utility-scale plants, which are:

  • Global Horizontal Irradiance (GHI) is the amount of terrestrial irradiance falling on a surface horizontal to the surface of the earth. GHI can be measured with a variety of instruments. The most common instrument used to measure GHI is called a pyranometer. GHI accounts for 71.6% of performance variations.
  • The impact of temperature and module type on performance. All semiconductors perform better at lower temperatures and most lose capacity factor at higher temperatures.
  • The third major factor is the inverter load ratio (ILR), which describes how much direct current generation from a project’s modules is converted to alternating current by the inverters and sent to the grid. In summer months when the solar resource is strongest, and adding modules and higher-powered inverters can also add cost to arrays, but it can also lead to power “clipping” — an inverter preventing modules from generating at full output to avoid being overloaded.

Reliability and Failure Modes:

PV module's operating life is largely determined by the stability and resistance to corrosion of the materials from which it is constructed. Manufacturer's guarantees of up to 20 years indicate the quality of bulk silicon PV modules currently being produced. Nevertheless, there are several failure modes and degradation mechanisms which may reduce the power output or cause the module to fail. Nearly all of these mechanisms are related to water ingress or temperature stress leading to loss of hermiticity, short & open circuits, p-n junction failure degradation due to metal migration etc. How much energy will any single panel or solar plant contribute to the grid on any particular day? The answer to that question has historically been tricky to answer.

Sometimes it’s cloudy. Sometimes the wind doesn’t blow. In order to further rely on the sun and the wind and make the energy they create a bigger part of the grid, we must be able to understand how much power will be available, what the demand for it could be and then use this energy to meet consumer needs as efficiently as possible. Weather data, satellite feeds, predictive analytics and machine learning, decides how power can reach the grid on a reliable and much more consistent basis. Forecasting technology and big data can help solve another challenge, too. Maintaining solar farms -- sometimes with hundreds or even thousands of panels spread across large regions -- can be a difficult, expensive and sensitive process. Advances in monitoring have made checking on these plants much more efficient.

Lavorro AI/ML Solutions:

Big data analytics to detect underperformance and identify when inefficiencies occur without having someone on site. The company’s Virtual Irradiance (VI), a solar management program, works by collecting ground level sunlight-intensity data to signal when panels aren’t performing at expected rates, and it sends an alert that repairs or maintenance are needed.

Environmental monitoring is critical to optimizing the efficiency of solar power generation. Beginning with site assessment, on-site meteorological monitoring factors into overall project performance and ROI. Weather stations provide reliable data on parameters such as solar radiation, wind speed and direction, temperature, and precipitation that influence location and efficiency of solar panels. Weather-corrected solar generation data based on measured irradiance from the weather station versus modeled irradiance when calculating actual performance versus modeled for contractual obligations. Additionally, weather data will be the basis of certain alarm conditions as well as reporting. Weather information is highly useful for load and production forecasting.”

Lavorro’s advanced analytics includes formulating statistical matrices, implementing actionable performance alarms, and conducting artificial intelligent models to predict future performance. This can work alongside an effective SCADA system that includes platforms and tools to evaluate plant performance and manage assets. Along with machine learning and deep learning models to revolutionize the way owners and operators manage their renewable energy assets. A Real-Time Weather Adjusted Performance Index measures the power and characteristics of a site (such as irradiance, back panel temperature, power generation, and DC and AC capability) to provide a real-time snapshot of the site’s performance. Implemented at the device level and the site level, it is a valuable Key Performance Indicator (KPI). By reviewing a site’s KPIs, engineers can:

  • Proactively troubleshoot an underperforming device and/or site
  • Respond to a performance alarm based on a low KPI
  • Forecast energy production by integrating one or more weather services
  • Realistically predict the site’s performance using machine learning (ML)
  • techniques based on historical data and actual site conditions

Anomaly Detections and Monitoring does not only depend on predefined formulas or account for the conditions at solar sites, such as shaded and soiled PV panels. A goal for more accurate energy prediction is the ability to pinpoint a device’s performance anomalies in real time. To do this, an individual model can be built for each site, incorporating information about the given site’s conditions into the model. An outlier detection model, consisting of one class support vector machine (SVM) technique, was built to monitor a device’s performance by comparing the measured and predicted energy at a device/site level in real time. The abnormal energy generation is highlighted. When an anomaly is detected, an alarm can be issued to alert the operator of the issue.

As a starting point, Lavorro needs to ingest a vast amount of detailed fault history information at many levels, from entire sites to the string level. This fault data provides essential “learning material” for machine learning modeling. Detecting and predicting faults can positively impact operational performance. The predictive analysis can help to schedule maintenance activities, maintain the correct amount and types of spare parts in inventory, and reduce the potential for device malfunctions. Overall, this will reduce downtime, diminish lost energy, and increase the lifetime of a solar asset.