Smart Technology-Based Platform for Advanced Energy Management (TIGENER)

Project Description and Objectives

The main objective of the project is to develop an innovative remote management product based on smart technologies for managing, analyzing, and processing data, in order to ensure an accurate assessment of energy consumption.

With the aim of contributing to greater energy efficiency and optimizing the use of resources, the overall objective of the project is to develop a new product based on smart technologies capable of managing, analyzing, and processing data, in order to ensure an accurate assessment of recorded consumption. These technologies will be designed to provide a logical and consistent response in terms of operation, organizing the processed data effectively. Their implementation will result in a significant improvement in energy efficiency and optimal use of energy in the systems and facilities where they are applied, generating substantial economic and environmental benefits.

The project involves the development of advanced technologies related to the solution architecture, the use of conversational artificial intelligence, image and audio processing, and IoT. In addition, all of this will follow a multiplatform approach, automating both backend development and API documentation generation. For this reason, an initial software version (prototype) will be developed and deployed on the existing OTEA solution, and the developments will be validated through a pilot test in an industrial environment, connected to a self-consumption installation.

  • Place of implementation: A Coruña
  • Implementation period: 02/2024–10/2026
  • Project budget: €477,921.00
  • Amount financed by the EU: €139,171.30

Results

Year 2024

During this 2024 phase, significant volumes of technical information were managed and processed in order to configure the core of the smart platform.

  • Artificial Intelligence Parameters:
    • Multiple models were analyzed and tested: LLaMA (version 1), GPT-3.5, GPT-3.5-Turbo, GPT-4o, GPT-4o-mini, Gemini, Mixtral, and Mistral
    • Variable temperature settings (0.9, 0.7, 0.5) were applied to balance creativity and response accuracy
    • Advanced techniques such as RAG (Retrieval-Augmented Generation), fine-tuning, and prompt engineering were implemented using libraries such as LangChain
  • Energy Data Processing:
    • Data structures in JSON format were defined for communication between the AI and the OTEA platform.
    • Linear regression methods were applied to calculate baselines, along with degree-hour and degree-day methodologies to analyze climate impact
    • For information quality, interpolation and clustering models were used to reconstruct data gaps

Summary of Milestones Achieved (Milestone 1 in 2024)

The project successfully completed the analysis, design, and initial testing phases of the main modules planned for the first year:

  1. Benchmarking and Energy Management Module:
    • Completion of the design and development of a ranking and comparison system between facilities.
    • Implementation of advanced filters (by number of offices, indicators, and date ranges) and visualization of average and hourly consumption.
    • Successful validation of comparative reports based on real energy meter data.
  2. Virtual Assistant (Nerea – Phase 1):
    • Definition of a robust architecture based on Docker, SQL, FastAPI/Django, and an AI Agent microservice.
    • Selection of ChatGPT as the most viable model due to its accuracy in Spanish and response times, after OpenSource models and Gemini were discarded during pilot testing.
    • Demonstrated ability to generate dashboards, download data, and create charts using natural language.
  3. Artificial Intelligence and Diagnostics:
    • Successful execution of local models (LLaMA, Mixtral) to predict machinery failures, identifying that Wizard-based models provide higher response quality.
    • Identification of critical limitations, such as the difficulty current models have in performing complex mathematical calculations and their tendency to generate “hallucinations,” which made it possible to adjust the strategy toward a hybrid logic.
  4. Flexible Demand:
    • Technical integration with the Bamboo platform was achieved, defining the flows for sending consumption data and receiving activation orders for demand control.
  5. Strategic Prioritization:
    • It was decided to postpone the Billing module until Milestone 2 (2025) in order to concentrate resources on advancing Artificial Intelligence, which required 41% more development hours than initially planned due to the rapid evolution of the sector.

Year 2025

This 2025 phase reflects an intensification of work in the areas of artificial intelligence and conversational assistants.

Summary of Milestones Achieved (Milestone 2 in 2025)

  1. Evolution of the Virtual Assistant (Nerea):
    • Phase 2: The transition from a simple-response model to an advanced conversational system based on state graphs (LangGraph) was completed, enabling the management of complex processes and automatic error correction. Persistence in SQL Server was implemented for long-term memory, together with structural security against prompt injection.
    • Phase 3 (Brought Forward): The capability to read technical manuals was integrated using a RAG architecture and the FAISS library, and a hybrid diagnostic engine combining AI with physical rules programmed in Python was designed.
  2. Bill Simulator:
    • A tool capable of replicating the Spanish tariff system (updated to 2026) was developed and validated, enabling the comparison of real vs. simulated bills and the detection of billing errors through the analysis of active energy, reactive energy, and power.
  3. Carbon Footprint:
    • The technical analysis for filling data gaps was completed using methods such as mean values, proportionality, and the KNN algorithm.