How AI, ML, RL, and DL Are Transforming Building Management Systems (BMS)

Major innovations are reshaping how buildings are managed, monitored, and optimised. Leading to significant improvements in:

• energy efficiency
• operational cost savings
• resource utilisation
• occupant comfort
• operational resilience and service facility excellence

By leveraging data-driven insights, these systems are becoming smarter, greener, and more efficient. Setting new benchmarks for sustainability and operational excellence.​ Renewable-directed Building Energy Management Systems (BEMS), often complementing BMS in fully integrated building solutions that also increasingly utilise IoT systems for monitoring and control. ​ 

Building Automation and Management Systems (BAMS) are now also integrating advanced automation, AI, and IoT technologies to enable predictive capabilities and strategic management of these building systems. ​ 

These systems, together, monitor and control broader building-wide operations, such as HVAC, lighting, alarms, and security. Helping buildings save energy, improve comfort for users, and help make substantial cost savings. ​ 

This article therefore explores the transformative impact of key building management technologies, supported by real-world examples and key data points that demonstrate their impact on energy use, cost-savings, occupant-comfort and human well-being. 

AI, BMS, IoT, and connectivity are interconnected technologies that collectively transform building operations, energy management, and sustainability into secure, connected entities and ecosystems. ​BMS type systems serving as centralised platforms for collecting data, dashboarding and automating the control and monitoring of building systems such as Heating, Ventilation and Air Conditioning (HVAC), lighting, and security, for ensuring operational efficiency. IoT technologies providing the infrastructure for real-time data collection through interconnected sensors and devices. Monitoring variables such as temperature, humidity, air quality, energy usage, and occupancy levels. ​ 

Secure and resilient connectivity acts as the backbone of these systems. Enabling seamless but secure and private communication and data exchanges between many disparate IoT devices, BMS, and AI systems (and their system protocols). Also, Facilitating secure interactions with external networks such as electrical power grids, off-site data sets, and district heating systems. ​AI processing the collected data using advanced analytics and ML models to enable predictive control, maintenance, fault and anomaly detection, and system optimisation. ​ 

Together, these technologies create a synchronised ecosystem that supports intelligent decision-making, dynamic energy-management, and trusted automated adjustments. ​Transforming traditional buildings into proactive, adaptive, and flexible environments that support sustainability and NetZero efforts. ​​ 

IoT is playing a transformative role in building management. ​IoT devices, such as smart sensors and devices, collect real-time data on parameters such as energy usage, temperature, humidity, air quality, occupancy, equipment performance, and structural health: ​​ 

• IoT helps automates BAMS systems such as lighting, HVAC, and cooling based on occupancy and environmental conditions, allowing disparate devices to communicate, send data, and receive control commands securely. ​Integrating both legacy and new building systems to help coordinated operations, especially in older building stock. Averting failures, leaks, downtime, and lowering repair costs. ​ 
• IoT also empowers Digital Twins – virtual replicas of physical buildings. Allowing real-time visualisation, simulation, optimisation, and system learning. ​By automating responses to dynamic building, user, or environmental conditions learned through AI and ML, IoT, BMS, and BAMS can adjust ventilation, heating, and lighting based on occupancy, building condition, or environmental factors. Improving management efficiency across large or multiple facilities. ​ 
• IoT sensors also track important parameters that affect human comfort such as temperature, humidity, particulate matter, and CO2 concentrations to maintain healthy indoor conditions and enhance comfort.​ IoT platforms providing device and connectivity management through dashboards for real-time decision-making.  
• IoT technologies also support smart-energy systems through energy monitoring, adaptive control systems, and demand-side management. Promoting energy-saving practices and enabling dynamic resource allocation for efficient use of energy, water, and other important resources. 

As IoT solutions are modular and scalable, they help create interconnected systems that are retrofittable and adaptable to new and future technologies. ​When integrated with advanced AI, IoT further helps optimise building systems. Improving operational efficiency and occupant experiences through the use of large broad, local, and temporal data sets, simulation, and continuous learning. Transforming traditional BMS into intelligent, adaptive, and sustainability supporting systems.

AI, ML, RL, and DL are reshaping building management systems in distinct ways: 

• AI (Artificial Intelligence): AI acts as the ‘brain’ or ‘mind’ for buildings, enabling systems to learn from data, make decisions, and improve over time. ​It powers modern BMS by performing tasks such as self-learning, decision-making, and dynamic optimisation. ​AI tools leverage historical, real-time, localised, and wider data from IoT sensors to adjust building operations dynamically (including lighting, heating, energy use, ventilation, and HVAC). ​Predictive analytics anticipate energy demand, reducing waste and enhancing efficiency. ​ 

• ML (Machine Learning): ML, a subset of AI, teaches systems to find patterns in data and make predictions. ​ML algorithms like artificial neural networks (ANNs), support vector machines (SVMs), and random forests (RF) are widely applied for energy forecasting, fault detection, and anomaly spotting in collected and incoming datasets. ​ 

• DL (Deep Learning): DL, a subset of ML, uses layers of “virtual neurons” to analyse complex data and make accurate predictions. ​DL models like convolutional neural networks (CNNs) and long short-term memory (LSTM) networks excel at handling large-scale, high-dimensional, and time-dependent building management data. ​ 

• RL (Reinforcement Learning): RL focuses on training systems to make decisions by interacting with their environment and learning from outcomes. ​ For example, RL algorithms dynamically adjust HVAC settings based on real-time data and complex decisions, such as occupancy levels and weather conditions, to minimise energy consumption while maintaining occupant comfort. ​ 

• Hybrid Models combine ML and DL to improve their accuracy and robustness. For example, ML-based frameworks optimised by nature-inspired algorithms like particle swarm optimisation (PSO) and firefly algorithms (FA) enhance prediction accuracy by automatically tuning predefined hyperparameters (a) as the data emerges. ​Also playing a pivotal role in air quality and safety monitoring by calibrating sensors and detecting inconsistencies in data. ​​​​Hybrid deep learning architectures, such as CNN-BiLSTM (b), help predict pollutant concentrations and optimise HVAC operations. Analysing usage patterns of electric appliances and CO2 levels to detect occupancy with high accuracy and for estimating power consumption in distributed systems. ​​ 

(a) Hyperparameters are parameters in machine learning models that are set before the training process begins and are not learned from the data during training. They control the behaviour, structure, and performance of the model, influencing how the model learns and generalises from the data. Requiring careful selection and tuning to ensure effective and efficient model performance.

(b) CNN-BiLSTM improves energy management predictions by leveraging spatial and temporal analysis capabilities, reducing errors, and achieving high accuracy in forecasting energy consumption and environmental parameters. It is particularly effective for multi-step-ahead forecasting and dynamic energy scenarios.

AI and IoT technologies are revolutionising BMS in several key areas: 

Energy Management

Thermal energy (heating and cooling) accounts for nearly 50% of final energy usage in EU buildings and is therefore one of the biggest costs. Globally, HVAC systems being the largest energy consumers in buildings, accounting for up to 49% of electricity use. Lighting and electrical systems, however, are also significant components and BMS will increasingly need to adapt to the demands of scaled EV charging. ​AI and ML help optimise energy consumption by predicting energy needs in these dynamic and multifaceted environments. Adjusting systems through BMS and IoT control accordingly.  

AI-assisted HVAC systems have been found to achieve significant energy savings in commercial buildings through a variety of mechanisms. ​​​​RL frameworks being particularly useful as they learn from their environment. Balancing energy use and comfort to help create autonomous, localised systems that benefit users. ​​​Some research and case study examples of the efficiencies available through the use of AI include: 

• AI-based smart thermostats in offices have demonstrated energy savings of up to 17.20% during winters, while achieving energy savings of up to 37% using AI models for HVAC control.  

• ​Residential and educational buildings have reported savings of up to 23% and 21%, respectively, achieving up to 24.29% in cost reductions using AI. 

• Occupancy detection systems have helped save up to 8.1% in energy use and improved fed-back ‘thermal-comfort6’ by between 43% and 73%. ​(Thermal comfort is measured using metrics like Predicted Mean Vote (PMV), Predicted Percentage Dissatisfied (PPD), indoor CO2 levels, and percentage improvements in occupant satisfaction).

• MagicBox, Madrid: An LSTM-based predictive model for HVAC energy consumption, achieved a high-performing test error rate (NRMSE) of 0.13, enabling real-time energy estimations and as a result reduced operational inefficiencies. 

• The EDGE Building in Amsterdam utilised 30,000 IoT sensors to optimise lighting, ventilation, and temperature control, resulting in a 17% reduction in energy consumption. 

• At Aalborg University in Denmark, ML models identified over 100,000 fault instances such as “Light_On_No_Occupancy,” thereby reducing unnecessary energy consumption and improving operational efficiencies. 

• The Sydney Opera House in Australia achieved energy savings of 25% and annual cost savings of $1.5 million through AI-driven BEMS (https://www.academia.edu/18669761/Sydney_opera_house_case_study_report).  

• Similarly, the University of California, Berkeley, significantly reduced energy, and water consumption using AI-enabled BEMS for a range of key projects. ​​ 

• Digital twins combined with AI algorithms have also been found to reduce energy consumption by as much as 50% by improving real-time decision-making and active responses. ​ 

AI and ML, when combined with BMS, make buildings more comfortable for people by adjusting the temperature and lighting based on real-time data. ​​Predicting thermal comfort levels and adjusting heating, cooling or ventilation systems accordingly. ​​​​​​​ 

AI and RL are not only important for saving energy and thermal comfort, however, but also for maintaining air and building quality. When supported with sensors, AI helps analyse occupant behaviours and preferences to optimise indoor environmental quality (IEQ). Typically including temperature, lighting, air quality, humidity, CO2 levels, and volatile organic compound pollutants (VOCs). Linking with HVAC systems to automatically adjust ventilation rates and air purification systems based upon internal and external conditions and pollutants. ​​ 

Indoor air quality monitoring is particularly crucial for short-term and long-term health. ​​​​Ensuring safety, particularly in residential or workplace settings and more industrial environments. ​There being many factors that can affect air quality: 

• Volatile Organic Compounds (VOCs): Emissions from building materials, paints, adhesives, furnishings, carpets, and cleaning products can degrade IAQ and cause health issues such as respiratory irritation, headaches, and long-term organ damage. Other concerns being ozone (O3) and nitrogen dioxide (NO2)​. Maintaining Total Volatile Organic Compounds (TVOCs) within the specified ranges is recommended for optimal IAQ. For example, The EU directives and standards that define and set limits for VOCs include the following: 

• Directive 2004/42/CE: Regulates VOC emissions from paints, varnishes, and vehicle refinishing products. 

• Directive 2010/75/EU: Focuses on industrial emissions, including VOCs. ​ 

• Industrial Emissions Directive (IED): Governs VOC emissions from industrial processes. 

• Directive 2023/1791: Includes environmental impacts and energy efficiency, potentially addressing VOC emissions. 

• Directive 2024/1275: Addresses VOC emissions as part of indoor air quality and energy efficiency standards in building performance. 

• DIN 1946-6: Defines acceptable indoor air quality levels, including VOC thresholds. 

• UNI EN ISO 16000: Provides recommendations for sensor placement to ensure accurate VOC measurements. 

• It is important to note, however, that Natural VOCs (Volatile Phyto-Organic Compounds – VPOCs) are also released by plants. These can enhance air quality but can also contribute to VOC levels on many types of sensors giving misleading reports. Context, therefore, is important. 

• Particulate Matter (PM): Fine particles from construction activities, HVAC systems, internal or external pollution sources can infiltrate indoor spaces and cause health or breathing problems. Sensors usually discriminate between particle sizes (i.e. PM2.5 and PM10), as they can affect respiratory systems in different ways. 

• Carbon Dioxide (CO₂): Elevated CO₂ levels due to poor ventilation can lead to discomfort, reduced cognitive function, and health issues. For example, maintaining indoor CO2 levels below 1000 ppm is highlighted as a measure of occupant comfort in many research studies. Quality of CO2 sensors, however, is important as measurement accuracy is critical to their effectiveness. 

• Formaldehyde is a particularly dangerous VOC found in many building materials and furnishings, being a known irritant and carcinogen. There are sensors that can detect for this compound. 

• Humidity and Temperature: Improper humidity levels (the optimal range being: 30%-60%) can promote dangerous Mould growth which is harmful to health (particularly in residential stings). Proper ventilation and temperature control being essential to controlling its formation and spread. 

• Radon is a naturally occurring radioactive gas that poses significant health risks, particularly when accumulated in indoor environments. Long-term exposure is the primary concern, as health effects may only become apparent years later. The dangers associated with radon include: 
  • Lung Cancer Risk: Radon is the second leading cause of lung cancer after smoking. Radon decays into radioactive particles that can be inhaled, damaging lung tissue over time. 
  • Indoor Accumulation: Radon can seep into buildings through cracks in floors, walls, and foundations, especially in areas with high radon concentrations based on geographic location and soil composition. Poor ventilation can exacerbate the build-up indoors, increasing exposure risks. 
  • No Immediate Symptoms: Radon exposure does not cause immediate health symptoms, making it difficult to detect without proper testing. 

• Noise is a pollutant that can affect the well-being of individuals. Particularly in their home environments. Measuring noise levels can help expose issues and areas of potential conflict and discomfort. 

Buildings are increasingly using less predictable renewable energy sources (RES) for power generation. Promoting sustainability goals such as zero-energy buildings, reduced carbon footprints, and green certifications including LEED and BREEAM. ​AI and ML can help balance energy supply and demand dynamically and in real-time. Ensuring the efficient and optimum use of RES for energy generation, storage, and consumption, by integrating and controlling solar panels, wind turbines, geothermal systems, energy storage solutions (ESS) and distribution systems. Enabling real-time monitoring, predictive analytics, and adaptive control to dynamically adjust energy usage based on weather conditions, occupancy patterns, energy demand, load balancing requirements and addressing challenges like power generation intermittency, energy trading, and electric vehicle charging.  

​Cloud-based platforms, AI and integrated Digital Twins can help further improve scalability, system maintenance, and energy scheduling: 

• Energy Production and Consumption: Real-time data from smart meters measure and transmit dynamic energy consumption for analysis. Smart grids powered by AI help integrate this highly dynamic data with intermittent RES. Important for helping the viability of these systems and reducing greenhouse gas (GHG) emissions. ​ ​​​ 

• Solar Radiation and Renewable Energy Data: Solar irradiance, ambient temperature, wind speed, and diffuse solar radiation are measured and used for assessing wind power, solar heat gain, indoor temperature effects, and building data sets for forecasting energy production capacity, availability, and demand needs. 

• Data from lithium-ion batteries, EVs, and other energy storage solutions help manage intermittency and optimise the increasingly critical ESS capacity and availability for storing RES generated power. 

• ML algorithms can also detect or predict early anomalies in any of the equipment’s performance. Extending the life of complex systems and expensive RES and ESS equipment. ​​ AI-driven predictive maintenance systems being shown to reduce costs, delays and downtimes in many such building systems. 

While AI and smart building technologies offer many benefits, there are also challenges to overcome in implementing these systems. For example: 

• Data Quality: AI models need large amounts of accurate data, but many buildings lack proper sensors and data collection systems. ​​​For instance, only 0.60% of non-residential buildings in Catalonia have implemented predictive control systems, being indicative of low adoption rates in many regions. A significant portion of buildings (86.40%) [and surprisingly, many offices (99.10%)] also lack data digitisation. Limiting the application of advanced analytics and predictive controls in many important settings.  

• Security: Security concerns around AI systems, IoT, and smart technologies include vulnerabilities to cyberattacks, unauthorised access, data breaches, and risks to sensitive data integrity and privacy. ​​Challenges arise from inadequate security measures in legacy systems, fragmented communication protocols, reliance on third-party cloud services, and interoperability issues across diverse platforms. ​​ 

• Privacy Concerns: Collecting data from building occupants raises ongoing ethical questions about privacy and appropriate use. ​​​​Ethical concerns such as algorithmic biases and compliance with international data protection laws further complicate AI systems. Solutions involving advanced encryption, secure communication protocols, blockchain technology, privacy-preserving methods, and security-by-design principles, alongside regular audits, updates, and differentiated access controls to safeguard data and infrastructure have been variously discussed in research as potential heightened solutions in the AI building management paradigm. ​ 

• Integration: Existing building systems may not always be compatible with AI and ML technologies, requiring upgrades. ​​​Integration challenges include fragmented communication protocols, interoperability issues across diverse platforms, lack of standardisation in data formats and IoT devices, scalability limitations, high acquisition costs, and difficulties in retrofitting legacy systems with modern technologies. ​​Additional problems involve semantic heterogeneity, data quality issues, technical complexities, and the need for advanced skills. Solutions proposed include adopting standardised communication protocols (e.g., BACnet, MQTT, Zigbee), leveraging middleware platforms and APIs, employing modular and scalable architectures, using frameworks like BIM, blockchain, and ontology-based systems, enhancing cybersecurity, utilising techniques like data fusion and distributed computing, and fostering collaboration among stakeholders. Training programs, real-world trials, and community-building initiatives being recommended for further support seamless integration and operational efficiency. 

• Scalability: AI-based algorithms require significant computational resources, making deployment in large-scale systems challenging and requires the securing of these dynamic resources for real-time systems.  

• Model Generalisation:  Adapting AI models to diverse building environments and high-frequency data remains a key research focus as economies of scales require that they be adaptable to many environments. 

• Explainability and Transparency: A lack of explainable AI solutions and challenges in understanding AI model predictions can hinder trust, usability, and their adoption. Research is therefore ongoing into how to make AI more explainable and configurable to enable more everyday use cases to become available.  

Despite these challenges, the future looks highly promising for the building management sector as AI opens significant opportunities to integrate intelligent and resource efficient systems with distributed IoT sensors. ​​Many of these systems require secure inter-connectivity, where we can expect: 

• Better Energy Efficiency: AI-powered systems will continue to reduce energy use and costs. ​​​​ 

• Lower Maintenance Costs: Predictive maintenance frameworks will become more advanced, saving time and money. ​​​​ 

• Enhanced Reliability: Buildings will become more reliable and flexible with self-learning and self-updating systems. ​​ 

• Generative AI: This technology will simplify data analysis and improve energy modeling accuracy for users. ​ 

• More Collaborative Global Research: Increased research in Europe and worldwide will help the critical building management sector meet stringent sustainability goals. ​​​ 

​​​Fast Machine Learning (FastML) is also being developed to make real-time applications even more efficient. ​FastML works by employing techniques such as model quantization, pruning, and hardware optimisation using specialised hardware to enhance computational efficiency and inference speed. ​It accelerates various stages of the machine learning pipeline, including data preprocessing, model training, and inference, while maintaining model accuracy and reducing computational overhead. ​FastML should improve AI systems by enabling real-time or near-real-time decision-making, optimising energy management, enhancing occupant comfort, ensuring safety, and supporting tasks like fault detection, energy efficiency optimisation, and predictive maintenance. ​Potentially making it ideal for resource-constrained edge environments and dynamic applications such as BMS, BEMS and BAMS. 

AI, ML, RL, and DL are helping transform building management systems, making them smarter, greener, and more efficient. ​​From saving energy to improving comfort, these technologies are paving the way for a sustainable future. ​​​​While challenges remain, the opportunities are immense, and the benefits are already being felt in many buildings around the world. ​​As these technologies continue to evolve, we can look forward to smarter, more sustainable buildings that help improve our lives and protect the broader environment. ​​​​