SOL-MIND

UPVD BQR 2026 project dedicated to explainable diagnostics, cognitive mapping and topological analysis of photovoltaic systems.

Overview

SOL-MIND (Solar Multivariate Interpretable Navigation and Diagnosis-Mapping) is a research project funded through the UPVD BQR 2026 program and developed within the PROMES-CNRS laboratory.

The project aims to establish a new generation of explainable diagnostic methodologies for photovoltaic systems by combining artificial intelligence, topological data analysis, interpretable metrics, and cognitive representations of system behavior.

The project addresses one of the major limitations of current photovoltaic diagnostic systems: the lack of interpretability of machine learning models. While many artificial intelligence approaches can detect anomalies with high accuracy, they often fail to provide understandable explanations of system behavior. SOL-MIND seeks to bridge this gap by developing physically meaningful representations of photovoltaic operating conditions.

Scientific Context

Photovoltaic systems generate large amounts of electrical, thermal, environmental, and operational data. Current monitoring platforms typically rely on threshold-based alarms or black-box machine learning models, which provide limited insight into the physical causes of detected anomalies.

Recent advances in explainable artificial intelligence (XAI), topological data analysis (TDA), and manifold learning offer promising opportunities to represent complex system behaviors within low-dimensional and interpretable spaces. However, these approaches remain largely unexplored for photovoltaic diagnostics.

SOL-MIND aims to create a unified framework capable of representing normal and faulty photovoltaic operating conditions through cognitive maps that can be directly interpreted by engineers and maintenance operators.

Main Objectives

The project pursues the following scientific objectives:

  • Develop explainable metrics for photovoltaic system monitoring.
  • Create interpretable low-dimensional representations of photovoltaic behaviors.
  • Investigate topological data analysis techniques for anomaly characterization.
  • Develop cognitive maps linking operating conditions to fault mechanisms.
  • Integrate physical knowledge and artificial intelligence into a common diagnostic framework.
  • Improve the transparency and trustworthiness of AI-based photovoltaic monitoring systems.

Methodology

The project is organized around three main scientific axes:

Methodological framework of the SOL-MIND project showing the integration of explainable metrics, photovoltaic monitoring infrastructure and cognitive mapping approaches for interpretable photovoltaic diagnostics.

Axis 1 – Explainable Hybrid Metrics

Development of physically meaningful indicators derived from electrical, environmental, thermal, and operational measurements.

Examples include:

  • Performance indicators
  • Variability indicators
  • Stability metrics
  • Topological descriptors
  • Multi-source diagnostic signatures

The objective is to establish physically interpretable metrics capable of describing photovoltaic operating conditions beyond conventional performance ratios and threshold-based indicators.

Axis 2 – Diagnostic-Oriented Data Infrastructure

Design and deployment of an acquisition and monitoring platform capable of collecting and synchronizing heterogeneous data sources.

The platform includes:

  • Photovoltaic electrical measurements
  • Environmental sensors
  • Meteorological information
  • Operational logs
  • Fault databases
  • Thermal measurements
  • Data quality monitoring tools

Special attention is devoted to data traceability, interoperability, reproducibility and FAIR data management principles.

Axis 3 – Cognitive Mapping and Explainable AI

Construction of low-dimensional cognitive representations using:

  • Topological Data Analysis (TDA)
  • Manifold Learning
  • Explainable Artificial Intelligence (XAI)
  • Hybrid physics-informed approaches

The final objective is to generate interpretable maps of photovoltaic system behavior capable of supporting fault diagnosis and maintenance decisions while remaining understandable to human experts.

Expected Outcomes

The project is expected to deliver:

  • Novel explainable diagnostic metrics
  • Cognitive maps of photovoltaic system behavior
  • Open-source software tools
  • Scientific publications in high-impact journals
  • International conference contributions
  • New collaborations in the field of explainable AI for energy systems
  • FAIR photovoltaic datasets
  • Foundations for future national and European research projects

Scientific Impact

SOL-MIND contributes to the development of trustworthy artificial intelligence for renewable energy systems.

By improving interpretability and transparency, the project seeks to facilitate the deployment of AI-based monitoring tools in real industrial environments while increasing user confidence in automated diagnostic systems.

The methodologies developed within SOL-MIND may also be transferable to batteries, smart grids, industrial processes, cyber-physical systems and predictive maintenance applications.

Funding

UPVD BQR 2026

Host Institution

PROMES-CNRS (UPR 8521)

Université de Perpignan Via Domitia (UPVD)

Keywords

Explainable Artificial Intelligence (XAI) · Photovoltaic Systems · Fault Diagnosis · Predictive Maintenance · Topological Data Analysis · Cognitive Mapping · Renewable Energy · Smart Monitoring · Data Analytics · Digital Twins

Project Status

Ongoing

References