Brain meets Chill

This paper provides insights from two novel AI-based energy optimisation projects developed under the Sustainability and Innovation function for leading global technology companies across North America. The pioneering initiatives leveraged advanced IoT mapping, granular metering and sophisticated Machine Learning models to significantly improve energy efficiency and provide greater operational visibility for large-scale data centres. The project in the heart of Silicon Valley, California, achieved a remarkable 40% reduction in energy consumption. In contrast, the pilot developed in Allin, Texas, is still being refined, with results currently withheld from public disclosure.

Building on this global exposure and expertise, this paper also incorporates a valuable perspective from one of the novel testbed projects initiated by Saudi Arabia’s leading District Cooling provider. The pilot use case, conducted on a large District Cooling plant serving one of the mega-developments in Riyadh, further underscores the progress made in leveraging AI solutions for energy optimisation. Led by a well-established international consulting firm specialising in AI and digital solutions, the ongoing study produced a highly accurate predictive model demonstrating an impressive 96% accuracy within two hours. This milestone is considered significant progress in intelligent load demand forecasting and energy management for large-scale infrastructure in Saudi Arabia and the Middle East.

New Murabba aims to redefine the concept of smart built environments and sustainable urban living on a global scale. More than just the world’s modern downtown, it is founded on cutting-edge natural resource optimisation strategies, integrated from the very inception of its development. New Murabba is leveraging AI-powered predictive modelling, digital twin technologies and integrated energy-efficient solutions to reduce energy demand and carbon emissions, while enhancing real-time resource management based on informative data. These advancements are being delivered through strategic partnerships with leading global technology firms and innovation hubs. At the heart of this transformative vision stands the Mukaab – an iconic, immersive megastructure that masterfully fuses cultural heritage with cutting-edge technological innovation. It serves as a powerful emblem of New Murabba’s commitment to low-carbon design, AI-integrated operations and future-ready infrastructure at every level, in full alignment with Saudi Arabia’s Vision 2030.

These groundbreaking initiatives position Saudi Arabia’s giga and mega projects as regional pioneers and influential global contributors to the evolving narrative of sustainability, adaptive design and intelligent building innovation.

TOWARDS LIVABLE AND SUSTAINABLE URBAN ENVIRONEMENTS

Amid the climatic challenges and rapid urban expansion shaping major cities in the Middle East, the pursuit of livable, sustainable and energy-efficient urban environments has become a top priority from various perspectives and scales. District Cooling systems represent a transformative approach for modern climate control in this evolving landscape. District Cooling offers scalable, modular configurations that enhance efficacy, sustainability and operational excellence.

The District Cooling concept centralises the cooling process by chilling water and delivering it through underground piping, producing a cooling effect on a district-wide scale. The approach offers many benefits related to energy efficiency, carbon emission reduction, integration of clean technologies and appealing urban space planning.

In a region where cooling demand can represent 65% of the power demand in summer, these benefits become more than enticing; they become essential to meeting short- and long-term economic growth and prosperity goals. To achieve a stable and efficient operational scheme, cooling demand must be predicted accurately and granularly for a meaningful period.

AI analytics represent a powerful enablement tool that provides enormous opportunities for energy optimisation, operational excellence and asset performance management across the lifespan. Seamless integration with Computerised Maintenance Management System (CMMS) is critical for optimising operational efficiency and maintaining data integrity. Predictive analytics capabilities can proactively manage fluctuations in cooling demand by varying the outputs and limiting systems’ interdependency inefficiencies during different operational modes.

ROLE OF AI IN DEMAND PREDICTIONS AND FAILURE DIAGNOSTICS

AI-powered load prediction makes District Cooling plants much more innovative and resilient. By anticipating demand profiles, AI helps operators predict anomalies through alerts and equipment failure diagnosis before they increase in severity. These proactive features are essential in this region, where cooling demands are high and volatile and represent a severe burden on the power grid during the hot season.

Key elements monitored by Machine Learning models include the following factors:

Weather: Air temperature, humidity, wind speed and solar radiation are all great contributors to cooling usage. AI models can layer on localised microclimate data to provide sufficient coverage of the multiple weather types present in the Kingdom.

Occupancy and use: Buildings generate a lot more heat at peak occupancy. AI learns the unique and sometimes rapid changes in usage transition of the diverse building typologies, such as residential tall towers, malls, multi-dimensional commercial buildings and large-scale mixed-use developments.

Variations in part loads and peak loads: AI can detect and monitor hourly, daily, weekly and seasonal changes in cooling demand and mitigate the risk of service requests during part-load and peak-load phases.

Various techniques are used to integrate the above factors in a single or multiple prediction model to meet the intended goals.

LEVERAGING AI TECHNIQUES FOR REAL-TIME DECISIONS

Utilisation of advanced AI methods for load predictions and consequent operational improvements is paramount to achieving maximum efficiency in District Cooling while reducing energy wastage. By using accurate predictive models, District Cooling operators and end-users can foresee spikes and drops in cooling demand, allocate resources more efficiently and minimise the overall operation costs. This requires an effective pair of statistical models and Machine Learning implementations, as follows:

Statistical Models: These provide high-level understanding of consumption patterns and cyclical variations, and can often be considered baselines or components to be used within hybrid models, such as: Linear Regression, which is a traditional method that studies past cooling demand with regard to major predictors, such as ambient air temperature, humidity or occupancy proxies. It is only capable of identifying fundamental consumption trends and would typically use hyperbolic factors for similarly complex, non-linear relationships.

Time Series Models (SARIMA): The Seasonal Autoregressive Integrated Moving Average model is beneficial for forecasting cooling loads with deep cyclical characteristics, which is standard with climatic energy demands. This model can capture short-term fluctuations to long-term cycles, making it a handy tool for DC companies when planning for an optimised cooling supply strategy.

Machine Learning Models: AI models significantly elevate predictive accuracy by acquiring complex, non-obvious patterns residing in massive, multi-variate, time-based cooling load granular data. Predictive analytics often identifies and resolves these patterns far better than traditional statistical methods, particularly in quickly changing environments. The models considered are as follows:

Ensemble Learning: This includes Random Forest & Gradient Boosting models, like XGBoost and LightGBM). These powerful techniques yield added predictive robustness and reliability through the strategic means of weighted averaging decision trees. They are superior at addressing rich, high-dimensional data while accounting for the many simultaneous influencing factors (small-scale weather forecast data, dynamic occupancy change and complex historical demands). The advantages make these methods highly applicable in the diverse, fluid and volatile demand pattern.

Support Vector Machines (SVMs): Highly useful if the data exhibit complex, non-linear relationships, like many energy consumption patterns. They are also helpful for small, high-quality datasets or when residuals pattern demonstrates difficult attributes, particularly where low-accuracy regression models have dissimilarities due to complex boundaries (the case with cooling load prediction).

Artificial Neural Networks (ANNs) and Long Short-Term Memory (LSTM): ANN-type models and profound learning frameworks, like LSTMs, are well-suited for analysing time-based sequential data. The approach allows a vital learning experience from previous cooling demands data over time to improve estimates of future demand by identifying long-range dependencies. They are particularly applicable in fast-paced District Cooling environments, where energy demand evolves and changes throughout the day and seasons.

Reinforcement Learning (RL): Besides predicting demand for cooling, RL can optimise District Cooling systems’ real-time operational response to variations. RL entails a competent AI agent that learns and adapts behaviour over time, using experiences to improve its operation, while simultaneously seeking to minimise energy consumption and maximise performance under varying demand profiles. The operation of District Cooling systems provides an adequate cooling supply and balances it against demand while limiting operational costs. RL enables AI agents to make adaptive, data-driven choices by learning from experiential feedback instead of simply using a static set of rules.

DATA FOUNDATIONS FOR AI SOLUTIONS

The development of reliable, high-performance, AI-powered cooling load prediction and optimisation models requires specifications for fundamental data inputs. Accurate, consistent, high-quality, granular and comprehensive datasets are critical for developing reliable, robust, adaptive and scalable models. The five pillars of data requirements are as follows:

● Historical data: High-resolution and accurate consumption data, ideally over one year, can effectively capture complex changepoints, trends, patterns and anomalies.

● Real-time and forecasted weather data: Granular weather data sets include outdoor temperature, humidity, wind speed/direction and solar radiation. Local weather station data for accurate real-time and forecast data will be critical.

● Calendar data: These pertain to information on weekends, public holidays and key known scheduled events that may significantly influence building occupancy and use patterns.

● Building characteristics and metadata: These include detailed data types for building data, such as building type (e.g., residential, commercial, mixed-use governmental), building area total, intended occupancy, building age and typical operating hours.

● Operational data: These include real-time telemetry from cooling systems’ chillers, pumps, cooling towers, thermal energy storage tanks and other related network components (e.g., setpoints, flow rate, pressure differentials, power) for monitoring the performance and health of the systems’ emergent response.

MACHINE LEARNING: QUATIFIABLE ADVANTAGES

As construction and expansion continue to quickly modernise cities in the Middle East, combining AI with District Cooling systems offers a wide range of distinct and quantifiable benefits:

● Decreased energy use: AI can reduce energy use by 25-35% over conventional standalone cooling systems, depending on the type of heat rejection, thermal storage capacity and type of control system. This means considerable savings on operational costs.

● Reduced emissions: Helps achieve environmental and sustainability certifications, such as LEED and Mostadam.

● Decreased operational costs: Predictive maintenance and intelligent load management minimise wear and tear on complicated and costly equipment. More importantly, they reduce downtime, disruption and O&M costs.

CONCLUSION: BREAKTHROUGHS IN NEXT-GEN EFFICACY

AI is changing the District Cooling paradigm from a conventional cooling approach to a responsive, innovative, self-optimising energy network. By harnessing predictive analytics, advanced Machine Learning techniques and real-time data management, the District Cooling industry can achieve unprecedented energy efficiency – as low as 0.6 kW/TR, or better – alongside significant water resources conservation (2 Gal/TR). These tremendous improvements provide efficacy and long-term sustainability by extending infrastructure lifespan and significantly lowering total maintenance costs.

At their core, AI-driven load forecasting and operational efficiency measures are not merely technological advancement options or educational showcases; they mark a fundamental shift towards climate resilience and efficacy in a rapidly evolving economic landscape. By anticipating demand fluctuations and dynamically adapting operational strategies in real time, AI drives transformative advancements in energy management. It delivers exact responses to cooling demand variability through data-driven decision-making, ensuring exceptional efficiency and operational agility across large-scale infrastructure.

The writer is Senior Director – Design Management, New Murabba. He may be reached at <snezar@newmurabba.com>