InstaCast™ provides ultra high resolution intra-hour solar forecasts. In order to anticipate local events in the forthcoming minutes, on-site images of the sky are the most valuable source of information. InstaCast™ processes those images to precisely and reliably forecast solar production, and deliver it in a convenient way.


1-minute updates of the forecasts

1-minute updates of the forecasts

With rapid forecast updates down to 1 minute, InstaCast™ offers unmatched short-term prediction of cloud events and solar production drops.

Easy interfacing with a plant's control system

Easy interfacing with a plant's control system

InstaCast™ has a proven compatibility with major energy management & power control solutions from the market.

100% offline mode for remote sites

100% offline mode for remote sites

An embedded version of InstaCast™ integrated within Reuniwatt’s sky imagers enables it to be used for microgrid projects in off-grid sites.

Forecasting methodology

The physical phenomenon of the steep fluctuations in photovoltaic production is the cloud. Depending on the clouds’ altitude class and type, the consequences on production will be different: number of events per unit of time, amplitude of attenuation, reflection, abrupt change in cloud cover (ramp rate). These impacts are characterised by the implementation of signal processing, modal and frequency analysis and semantics, on the local images obtain using our sky imagers.

Our forecasting methogology involves a multi-steps procedure.

Sky image acquisition

Local images of the sky are acquired at a high frequency (1 min or less), as well as local weather measurements. Since the quality of this input data influences the accuracy of the intra-hour solar forecasts, Reuniwatt decided to design and build its own imagers, including the Sky InSight™ and the Sky Cam Vision™, after testing various sky-imaging systems available on the market since 2011.

Irradiance & production forecasts

First, a geometric calibration using the position of the sun – known with precision by astronomical calculations of solar geometry – is performed in order to associate each pixel with a position in the sky expressed with the zenith angle and azimuth. The figure on the right shows the lines representing certain values of zenith angles.

This “inverse projection” step produces an image with a uniform scale, where each pixel corresponds to the same area of sky. Then, the sky image is segmented into two classes {cloud, clear sky}. At the same time, the field of movement of the clouds is calculated on the projected images. Knowing the cloudy state of the sky at time t and assuming that this displacement is constant over the next few moments, it is possible to estimate future cloud states and derive a radiation measurement.

Formatting & transmission

Once computed, forecasts are formatted in a convenient way to be integrated in the plant’s management system, for an improved control of the plant. Reuniwatt’s InstaCast™ solution has a proven compatibility with major energy management & power control system providers. The intra-hour solar forecasts can be delivered locally in an embedded mode (Modbus TCP), or through the Internet (RESTful API or SFTP).


  • Several variables available: solar irradiance (GHI/DNI/GTI…), output power, drops

  • 1-minute updates

  • 30-minute horizon at 1-minute granularity

  • Probabilistic forecasts

  • 360°x 180° Field of View

  • Tailored distant (http, sftp…) or local (MODBUS) data transfer

Use cases and performance evaluation

The question of the performance of the forecasting method is, of course, of utmost importance in judging the interest of the tool. The answer, however, remains complex, particularly concerning the metrics used to evaluate forecasts.

We begin here by presenting the performance of InstaCast™ compared to a more basic statistical forecasting system.

InstaCast™ is the most performing forecasting method for 1-min to 30-min forecasts; it improves the statistical methods by more than 30% at a 5-min horizon.

However we will see in the examples below that minimizing the nMAE is not the main performance criteria depending on the use case; using sky cam make it possible to use a deterministic approach that is a must-have for some forecast applications like ramp detection or spinning reserve control.

Ramp detection

Purely deterministic forecasts are used in this case (orange curve vs actual production in blue), and permit to determine with the best accuracy the global shape and amplitude of solar drops, which is particularly useful for some cases like ramp control (curtailment). However, such forecasts are not very efficient in terms of nMAE.

Battery management

In this case, the forecast (orange curve) is optimised in order to reduce the nRMSE (root mean square error). Very large deviations are avoided at each time step. The counterpart is that the dynamics linked to the passage of the clouds are strongly smoothed. Such forecasts are notably useful in cases of battery storage management.

Spinning reserve control

The orange curve here represents the 1% percentile forecast; production will exceed the forecast 99% of the time. This type of conservative forecast makes it possible in particular to control the spinning reserve of gensets, in the case of a PV-diesel hybrid system.

You might also be interested in

satellite forecast

Intraday solar forecasts with HourCast™

Sky InSight™, the infrared all-sky imager

Solar forecasting for microgrids and off-grid projects