AI-Based Energy Management and Optimization for Urban Infrastructure: A Case Study in Trikala, Greece ^†

Abstract

This study explores an ongoing work for AI-driven framework for energy management and optimization within urban infrastructures, demonstrated through a case study in Trikala, Greece. The approach integrates smart monitoring systems, real-time analytics, and predictive algorithms to enhance energy efficiency across municipal infrastructures. By leveraging AI, such as machine learning, for demand forecasting and automated decision-making, the system reduces energy waste while supporting sustainability goals. The findings will highlight significant improvements in energy utilization and propose a scalable model applicable to other smart cities pursuing digital energy transitions.

1. Introduction

Urban energy systems are undergoing a paradigm shift toward decentralization, digitalization, and decarbonization. In this transition efforts, artificial intelligence (AI)-enabled energy management systems (EMSs) have emerged as key enablers of operational efficiency, renewable integration, and smart infrastructure control [1]. These systems allow cities to optimize energy flows in near real-time, forecast demand, and dynamically respond to pricing signals (when available), aligning energy consumption with sustainability goals [1].

AI-based EMS platforms have proven particularly effective in complex municipal environments involving multiple distributed assets, including photovoltaic systems (PV), battery energy storage systems (BESS), and public infrastructure such as drinking-water and wastewater facilities. Recent studies emphasize the potential of predictive optimization and machine learning in enhancing both operational reliability and energy efficiency at the urban scale [2].

Towards these, the Municipality of Trikala, Greece—a labeled City in the EU’s Mission Cities initiative—is deploying ABB’s OPTIMAX^® (ABB AG, Mannheim, Germany) to manage an approximate 10 MW energy ecosystem. The platform, which is under development, will integrate near real-time data from over 130 assets (public buildings, water infrastructure, schools and future PV plants), and performs intra-day and day-ahead optimization via cloud-based analytics. With the support of AI algorithms and on-line SCADA connectivity, the system will support the city’s ambition to achieve climate neutrality by 2030 while introducing a replicable model for intelligent municipal energy governance [3].

This paper presents the system architecture, implementation strategy, and expected outcomes of Trikala’s OPTIMAX ongoing deployment. It contributes to the academic discourse on smart cities by showcasing a real-world AI-driven EMS in action.

2. Material and Methods

The AI-based energy management framework that will be developed in Trikala will utilize ABB’s OPTIMAX^®, a modular, cloud-hosted system configured for predictive analytics, real-time control, and multi-source monitoring of municipal energy infrastructure. The solution will be hosted on the municipality’s cloud environment and integrated with existing SCADA systems via OPC UA protocol, ensuring secure, scalable interoperability across assets (Figure 1).

2.1. System Architecture

The OPTIMAX^® system will interface with three main infrastructure domains:

• Water and Wastewater Infrastructure: Including 50 pumping stations and 25 local stations with dataloggers, plus one wastewater treatment plant (WWTP), accounting for a combined load of ~2 MW.
• Public Buildings: A set of 60 buildings, of which 10 will be fully monitored (energy consumption and analytics) and 50 will be reported through periodic CSV file uploads.
• Renewable Assets: A 1 MWp photovoltaic plant and two Battery Energy Storage Systems (BESS) will be included in the future, forming the distributed energy resources (DER) layer.

The architecture follows a hybrid data acquisition strategy—real-time OPC UA integration for online systems and periodic file transfers (SFTP) for static or aggregated building data. This flexible structure supports comprehensive system coverage without overhauling legacy infrastructure.

2.2. Data Integration and Signal Design

The system will ingest approximately 1400 unique signals from ~138 assets. These will include real-time sensor data (e.g., electricity consumption and analytics), equipment statuses, energy production logs, and environmental inputs. Signal configuration adheres to standard numeric or string-based identifiers to ensure compatibility with OPTIMAX’s AI modules (Figure 2).

Each signal will be mapped to one or more analytics functions, such as:

• Load forecasting (using historical consumption and weather data)
• Intra-day dispatch optimization
• Day-ahead planning based on price signals (when available) and PV forecasts
• Anomaly detection and efficiency diagnostics

2.3. AI-Driven Forecasting and Optimization

OPTIMAX^® can employ supervised learning and time-series modeling techniques for short- and medium-term load forecasting, leveraging historical and real-time data streams. Forecast outputs can be fed into optimization engines that generate control signals or dispatch strategies, depending on constraints such as tariff schedules, PV production curves, and grid export limitations [4].

These capabilities can enable predictive energy balancing across the municipality’s distributed infrastructure and allow for rule-based or autonomous adjustments to operations (e.g., shifting pumping loads to off-peak hours). As demonstrated in previous smart grid deployments, such dynamic EMS configurations can yield substantial efficiency gains and emissions reductions [5,6].

3. Expected Results

The deployment of ABB OPTIMAX^® in the city of Trikala will demonstrate early-stage effectiveness in preparing municipal infrastructure for intelligent, AI-driven energy optimization. Although the project is in its implementation phase (scheduled for completion in 2026), initial system integration and modeling outputs allow for the following observations:

3.1. Load Coverage and System Responsiveness

Preliminary signal mapping and load modeling indicate the need for integration of approximately 1400 signals across 138 municipal assets, covering an estimated peak capacity of 10 MW. The asset load distribution (Figure 3) will demonstrate a balanced demand profile across water, wastewater, public buildings, and future renewable systems. The integration process will validate system responsiveness, with SCADA-connected infrastructures supplying near real-time input to control modules. Furthermore, the integration with the Energy Service Office and the Climate Neutrality Hub will ensure that technical outputs can be directly translated into actionable policy insights, facilitating transparent tracking of energy and emissions KPIs [3].

3.2. Forecasting Accuracy and Optimization Potential

Simulated intra-day forecasts using historical consumption data and weather inputs yielded promising accuracy levels. Initial tests showed that load deviation from predicted baseline remained within ±6% for water systems and ±8% for building clusters during dry-weather scenarios. These results align with established accuracy ranges for short-term AI-based load forecasting in urban grids [1]. Such precision, even in test phase, enables reliable optimization routines, including tariff-based load shifting and peak shaving.

3.3. Energy Flow Optimization Scenarios

Using synthetic data from historical trends, OPTIMAX can be configured to run control simulations for energy flow dispatch across PV, BESS (if available), and consumption nodes. Primitive results through first tests with raw and hypothetical data showed that:

• PV self-consumption ratios could be increased by 22% by rescheduling pumping operations to daylight hours.
• Energy cost savings for public buildings (when simulated against time-of-use pricing) ranged from 12% to 18%.
• Load balancing between BESS and grid could import reduced peak loads by approximately 15% in modeled weeks.

These findings suggest that AI-based predictive optimization can significantly enhance energy efficiency and economic performance in municipal systems [2].

4. Discussion—Ongoing Work

The ongoing deployment of AI-based energy management tool in Trikala (Figure 4) represents a critical step in operationalizing climate neutrality at the municipal level. While early integration tests and simulations show strong potential for load forecasting, control responsiveness, and cost reduction, several components of the project remain under development. These ongoing activities are focused on fine-tuning data integration, expanding stakeholder engagement, and validating optimization routines under real-time operational constraints.

4.1. Final Integration and Live Testing

The final stages of the system integration involve harmonizing signal streams from multiple infrastructure types—especially those reliant on non-standard data formats (e.g., CSV uploads from municipal schools and buildings). Ensuring consistency in metadata, time synchronization, and fault tolerance is essential to support continuous operation and reliable AI inference. In parallel, ABB’s OPTIMAX^® modules are being calibrated to local energy market conditions, weather dependencies, and regulatory constraints. [7]

4.2. Energy Policy and Governance Interfaces

The Climate Neutrality Hub’s digital tools—including the Energy Service Office, GHG Emissions Platform, and Participation Dashboard—are being developed in parallel with the technical backend. These governance layers are designed to democratize energy data, provide transparent policy feedback, and increase citizen trust in AI-based optimization decisions. Ongoing workshops are focused on co-designing performance indicators that reflect both environmental and social priorities [8].

4.3. Scalability and Replication

Another area of focus is validating the replicability of the Trikala model. A modular architecture, decentralized data ingestion, and flexible SCADA interfacing support the potential for deployment in other EU cities with comparable infrastructure. Future phases will analyze how the system can adapt to cities with variable renewable capacity, mixed energy tariffs, or different levels of digital maturity.

4.4. AI Governance and Ethical Considerations

As the project will soon be entering real-time optimization phases, the transparency and explainability of AI decisions becomes more crucial. The scientific team and partners are currently assessing methods for visualizing control decisions, setting override thresholds, and integrating human-in-the-loop controls to comply with the EU AI Act and maintain institutional accountability [1]. Ensuring these safeguards are in place is vital for long-term trust and operational safety.

4.5. Future Research Directions

Several research avenues are being explored:

• Incorporating dynamic pricing algorithms for BESS operation optimization
• Enhancing digital twin fidelity to better simulate fault scenarios
• Exploring federated learning for cross-municipality AI model sharing
• Evaluating citizen behavior modeling to inform DSM strategies