Company Overview 

INESC P&D Brasil is a non-governmental Scientific and Technological Institution (STI) organized as a private, nonprofit association in Brazil.  It coordinates cooperation in research, development, and technology transfer together with Brazilian public universities and INESC TEC (Portugal).  The institute supports and executes R&D&I projects and provides advanced consulting, often in partnership with industry and under international funding programs.  Its work includes electric power systems and broader engineering-driven innovation, with a focus on turning research into practical, high-impact outcomes. 

The challenge is to forecast short-term (up to six hours ahead) distributed PV generation with enough accuracy for real-world grid operations, particularly in contexts where utilities cannot rely on expensive satellite or sky-imaging infrastructure. This challenge is compounded under net metering schemes, where smart meter readings blend PV production with local consumption, making the underlying signal harder to model and predict. In addition, utilities often face practical constraints such as limited or fragile data access and significant data gaps, which requires robust data quality control and gap-filling approaches so smart meter data can be reliably used for grid forecasting and analysis. 

This issue matters in Cape Verde because, under net metering and growing distributed PV, the grid operator needs accurate short-term forecasts from smart meter data to set reserves and optimize dispatch in a fossil-dependent island system, improving reliability and reducing operating costs.  

The Solution 

Our solution turns existing smart meter data into actionable short-term PV net metering generation forecasts for utility operations. It automatically validates data quality and restores missing measurements with an innovative gap-filling approach, so utilities can rely on forecasting even when datasets are incomplete. By fusing information from neighboring meters, effectively using the distributed PV fleet as a sensor network, the platform strengthens the signal and delivers more reliable predictions for day-to-day decision-making. 

It leverages the distributed PV fleet as a virtual sensor network, mining spatio-temporal patterns directly from existing billing and smart meter data, so utilities can deliver reliable forecasts without investing in satellite feeds or sky-imaging hardware. 

How It Works  

The platform turns existing smart meter/billing data into utility-grade, short-term (up to 6-hour ahead) distributed PV forecasts by validating data quality, repairing missing measurements with information from neighboring meters, and then running a spatio-temporal forecasting engine that outputs actionable scenarios. 

How it works: 

Imports smart meter data, runs quality checks (QC flags), fills gaps via horizontal inpainting across multiple meters, then generates probabilistic forecasts (min/most likely/max). 

In practice: 

Operators use a web app with maps and charts to monitor the network and export forecast outputs for daily operations. 

Tools / platforms / data used: 

Smart meter/billing data (including net metering signals), Django backend, web UI (SB Admin 2), Leaflet maps, ApexCharts, Google Drive export. 

Technologies leveraged: 

QC routines (CQ1/CQ2), adapted LDMM for gap-filling (“horizontal inpainting”), and a kernel-based Semantic Model for spatio-temporal forecasting. 

Who uses it: 

Utility grid operations and planning teams (e.g., ELECTRA/ONSEC environment). 

Field applications (examples): 

Reserve sizing (rotating reserve), economic dispatch support, visibility into “invisible” distributed generation, and detection of metering/data issues or non-compliance patterns (e.g., generation at night). 

Impact on the Ground 

The solution delivers utility-grade, short-term (up to 6-hour ahead) distributed PV forecasts from existing smart meter/billing data, helping operators reduce uncertainty in daily grid operation, especially in environments with net metering and incomplete datasets. It directly supports reserve sizing (rotating reserve) and more efficient dispatch decisions by improving visibility into distributed generation at feeder/island scale. 

It contributes to digitalization and resilience by converting fragmented metering data into operational intelligence through automated data quality control and innovative gap-filling, enabling forecasting without adding expensive satellite/sky-imaging infrastructure. It also strengthens data accessibility and usability for utilities by improving how smart meter data is acquired, validated, and operationalized. 

Key figures 

• Up to 6-hour ahead PV generation forecasting horizon. 

• ~50% average reduction in prediction error for 6-hour forecasts. 

• ~94% confidence level for the forecast interval over the full 6-hour horizon. 

• Data quality challenge observed: up to 82.5% missing/flagged data in some subsets. 

Results so far 

The system was developed and validated in a real utility context in Cabo Verde (Santiago Island), reaching a mature demonstration level (TRL 8). Key milestones include an integrated web application (maps, charts, export), deployment-oriented workflows for operators, and an initial financial viability assessment supporting commercialization planning.

What’s next 

Next steps focus on scaling beyond the pilot by replicating deployments across additional islands and regions, and by broadening the product offering based on netmetering insights (including forecasts for injected energy, consumption, and gross PV generation). The project also plans to launch a startup to commercialize DEEP as a revenuegenerating solution for utilities across Africa and Small Island Developing States (SIDS), using a tiered licensing model linked to the number of connected Solar Home Systems (SHS) and data monetization opportunities for regulators. A seed round is being pursued to formalize the legal entity, explore synergies with other Digital Energy Challenge projects, hire a growth lead to drive market expansion, and secure the initial working capital needed for operations.