The contribution of this work is twofold: (1) We introduce a collection of ensemble methods for time series forecasting to combine predictions from base models. We demonstrate insights on the power of ensemble learning for forecasting, showing experiment results on about 16000 openly available datasets, from M4, M5, M3 competitions, as well as FRED (Federal Reserve Economic Data) datasets. Whereas experiments show that ensembles provide a benefit on forecasting results, there is no clear winning ensemble strategy (plus hyperparameter configuration). Thus, in addition, (2), we propose a meta-learning step to choose, for each dataset, the most appropriate ensemble method and their hyperparameter configuration to run based on dataset meta-features.

Automated algorithm selection and hyperparameter tuning facilitates the application of machine learning. Traditional multi-armed bandit strategies look to the history of observed rewards to identify the most promising arms for optimizing expected total reward in the long run. When considering limited time budgets and computational resources, this backward view of rewards is inappropriate as the bandit should look into the future for anticipating the highest final reward at the end of a specified time budget. This work addresses that insight by introducing HAMLET, which extends the bandit approach with learning curve extrapolation and computation time-awareness for selecting among a set of machine learning algorithms. Results show that the HAMLET Variants 1-3 exhibit equal or better performance than other bandit-based algorithm selection strategies in experiments with recorded hyperparameter tuning traces for the majority of considered time budgets. The best performing HAMLET Variant 3 combines learning curve extrapolation with the well-known upper confidence bound exploration bonus. That variant performs better than all non-HAMLET policies with statistical significance at the 95% level for 1,485 runs.

We study a Combinatorial Multi-Bandit Problem motivated by applications in energy systems management. Given multiple probabilistic multi-arm bandits with unknown outcome distributions, the task is to optimize the value of a combinatorial objective function mapping the vector of individual bandit outcomes to a single scalar reward. Unlike in single-bandit problems with multi-dimensional action space, the outcomes of the individual bandits are observable in our setting and the objective function is known. Guided by the hypothesis that individual observability enables better trade-offs between exploration and exploitation, we generalize the lower regret bound for single bandits, showing that indeed for multiple bandits it admits parallelized exploration. For our energy management application we propose a range of algorithms that combine exploration principles for multi-arm bandits with mathematical programming. In an experimental study we demonstrate the effectiveness of our approach to learn action assignments for 150 bandits, each having 24 actions, within a horizon of 365 episodes.

In neural architecture search, the structure of the neural network to best model a given dataset is determined by an automated search process. Efficient Neural Architecture Search (ENAS), proposed by Pham et al. (2018), has recently received considerable attention due to its ability to find excellent architectures within a comparably short search time. In this work, which is motivated by the quest to further improve the learning speed of architecture search, we evaluate the learning progress of the controller which generates the architectures in ENAS. We measure the progress by comparing the architectures generated by it at different controller training epochs, where architectures are evaluated after having re-trained them from scratch. As a surprising result, we find that the learning curves are completely flat, i.e., there is no observable progress of the controller in terms of the performance of its generated architectures. This observation is consistent across the CIFAR-10 and CIFAR-100 datasets and two different search spaces. We conclude that the high quality of the models generated by ENAS is a result of the search space design rather than the controller training, and our results indicate that one-shot architecture design is an efficient alternative to architecture search by ENAS.

Automated hyperparameter tuning aspires to facilitate the application of machine learning for non-experts. In the literature, different optimization approaches are applied for that purpose. This paper investigates the performance of Differential Evolution for tuning hyperparameters of supervised learning algorithms for classification tasks. This empirical study involves a range of different machine learning algorithms and datasets with various characteristics to compare the performance of Differential Evolution with Sequential Model-based Algorithm Configuration (SMAC), a reference Bayesian Optimization approach. The results indicate that Differential Evolution outperforms SMAC for most datasets when tuning a given machine learning algorithm - particularly when breaking ties in a first-to-report fashion. Only for the tightest of computational budgets SMAC performs better. On small datasets, Differential Evolution outperforms SMAC by 19% (37% after tie-breaking). In a second experiment across a range of representative datasets taken from the literature, Differential Evolution scores 15% (23% after tie-breaking) more wins than SMAC.

The environmental impacts of medium to large-scale buildings receive substantial attention in research, industry, and media. This article studies the energy savings potential of a commercial soccer stadium during day-to-day operation. Buildings of this kind are characterized by special purpose system installations like grass heating systems and by event-driven usage patterns. This work presents a methodology to holistically analyze the stadium’s characteristics and integrate its existing instrumentation into a Cyber-Physical System, enabling to deploy different control strategies flexibly. In total, seven different strategies for controlling the studied stadium’s grass heating system are developed and tested in operation. Experiments in winter season 2014/2015 validated the strategies’ impacts within the real operational setup of the Commerzbank Arena, Frankfurt, Germany. With 95% confidence, these experiments saved up to 66% of median daily weather-normalized energy consumption. Extrapolated to an average heating season, this corresponds to savings of 775MWh and 148t of CO2 emissions. In winter 2015/2016 an additional predictive nighttime heating experiment targeted lower temperatures, which increased the savings to up to 85%, equivalent to 1GWh (197t CO2) in an average winter. Beyond achieving significant energy savings, the different control strategies also met the target temperature levels to the satisfaction of the stadium’s operational staff. While the case study constitutes a significant part, the discussions dedicated to the transferability of this work to other stadiums and other building types show that the concepts and the approach are of general nature. Furthermore, this work demonstrates the first successful application of Deep Belief Networks to regress and predict the thermal evolution of building systems.

The environmental impacts of medium to large-scale buildings receive substantial attention in research, industry, and media. This article studies the energy savings potential of a commercial soccer stadium during day-to-day operation. Buildings of this kind are characterized by special purpose system installations like grass heating systems and by event-driven usage patterns. This work presents a methodology to holistically analyze the stadium’s characteristics and integrate its existing instrumentation into a Cyber-Physical System, enabling to deploy different control strategies flexibly. In total, seven different strategies for controlling the studied stadium’s grass heating system are developed and tested in operation. Experiments in winter season 2014/2015 validated the strategies’ impacts within the real operational setup of the Commerzbank Arena , Frankfurt, Germany. With 95% confidence, these experiments saved up to 66% of median daily weather-normalized energy consumption. Extrapolated to an average heating season, this corresponds to savings of 775MWh and 148t of CO 2 emissions. In winter 2015/2016 an additional predictive nighttime heating experiment targeted lower temperatures, which increased the savings to up to 85%, equivalent to 1GWh (197t CO 2 ) in an average winter. Beyond achieving significant energy savings, the different control strategies also met the target temperature levels to the satisfaction of the stadium’s operational staff. While the case study constitutes a significant part, the discussions dedicated to the transferability of this work to other stadiums and other building types show that the concepts and the approach are of general nature. Furthermore, this work demonstrates the first successful application of Deep Belief Networks to regress and predict the thermal evolution of building systems.

Due to its significant contribution to global energy usage and the associated greenhouse gas emissions, existing building stock's energy efficiency must improve. Predictive building control promises to contribute to that by increasing the efficiency of building operations. Predictive control complements other means to increase performance such as refurbishments as well as modernizations of systems. This survey reviews recent works and contextualizes these with the current state of the art of interrelated topics in data handling, building automation, distributed control, and semantics. The comprehensive overview leads to seven research questions guiding future research directions.

One of the key information pieces in improving energy efficiency of buildings is the appliance level breakdown of energy consumption. Energy disaggregation is the process of obtaining this breakdow ...

Fossil fuels serve a substantial fraction of global energy demand, and one major energy consumer is the global building stock. In this work, we propose a framework to guide practitioners intending to develop advanced predictive building control strategies. The framework provides the means to enhance legacy and modernized buildings regarding energy efficiency by integrating their available instrumentation into a data-driven predictive cyber-physical system. For this, the framework fuses two highly relevant approaches and embeds these into the building context: the generic model-based design methodology for cyber-physical systems and the cross-industry standard process for data mining. A Spanish school's heating system serves to validate the approach. Two different data-driven approaches to prediction and optimization are used to demonstrate the methodological flexibility: (i) a combination of Bayesian regularized neural networks with genetic algorithm based optimization, and (ii) a reinforcement learning based control logic using fitted Q-iteration are both successfully applied. Experiments lasting a total of 43 school days in winter 2015/2016 achieved positive effects on weather-normalized energy consumption and thermal comfort in day-to-day operation. A first experiment targeting comfort levels comparable to the reference period lowered consumption by one-third. Two additional experiments raised average indoor temperatures by 2K. The better of these two experiments only consumed 5% more energy than the reference period. The prolonged experimentation period demonstrates the cyber-physical system-based approach's suitability for improving building stock energy efficiency by developing and deploying predictive control strategies within routine operation of typical legacy buildings.

In recent reflections on environmental impacts of buildings, medium to large scale sports stadiums have gained substantial attention. These stadiums of e.g.~professional soccer teams are characterized by special system installations like grass heating systems serving the crucial commercial asset(s) and by event-driven usage patterns. Public buildings of this size imply situation-specific operational modes combined with high levels of safety and comfort requirements. In this paper we provide experimental verification of the energy savings potential of a professional soccer stadium's grass heating system during day-to-day operation. Our supervisory holistic control based on state of the art information and communication technology (ICT) is verified by seven experiments which we executed within the real operational setup of the Commerzbank Arena in Frankfurt, Germany. Our experiments operated different control strategies of increasing complexity. In winter 2014/2015 we achieved weather normalized energy savings of more than 56% compared to the last heating season. In an average heating season this would amount to savings of approximately 780 MWh and 150 t CO_2. At the same time we violated minimum temperature targets less than 6% of the time. These results stress the feasibility and benefits of applying holistic context-aware control strategies to large scale legacy consumption systems using supervisory ICT platforms. We demonstrate significant efficiency improvements and establish a new energy baseline that future control strategy evolutions will have to benchmark against.

Outsourcing of IoT-data is a common concept when using third-party services, e.g. an external building management service is given access to the building's sensor measurements. However, the measurements captured from building systems often need to be protected from unauthorized access, because they are linked to persons, processes or business secrets, and therefore data protection requirements apply. Nevertheless, due to the large data volumes, external storage is desirable from a business perspective.We propose the concept of a Security Broker based on symmetric encryption schemes. It offers a key management scheme to flexibly create short decryption keys for time intervals or specific ranges of measurement values. We further show how the scheme can be generalized or a combination of the schemes can be applied. We implemented a prototype in Java and analyzed its performance. The evaluation proved the mechanism's applicability for mainstream applications running on off-the-shelf computing equipment.

In this paper we analyze operational energy data of the cooling, ventilation and heating systems of the professional soccer stadium Commerzbank Arena in Frankfurt, Germany. We analyze data collected over a six month period in 2014 statistically and show that depending on the stadium's operational context consumption patterns vary largely among the different systems resulting in very different behaviors. The results provide insights into what drives the energy consumption for different systems of a large commercial sports facility: the static heating system is purely dependent on outside air temperature, ventilation exhibits a pronounced daily consumption pattern irrespective of the temperature and cooling is driven by a combination of event operation and air temperature. These insights will allow us to predict, plan and balance the energy demands of different subsystems more accurately, resulting in energetic improvements of the stadium operation in the form of load shedding while maintaining the systems' service levels.

This paper analyzes the potential of knowledge discovery from sensed data, which enables real-time systems monitoring, management, prediction and optimization in smart buildings. State of the art data driven techniques generate predictive short-term indoor temperature models based on real building data collected during daily operation. The most accurate results are achieved by the Bayesian Regularized Neural Network technique. Our results show that we are able to achieve a low relative predictive error for each room temperature in the range of 1.35% - 2.31% with low standard deviation of the residuals.

Recently, reflections on modern sports stadiums' environmental impacts have gained substantial attention. Large-scale stadiums of e.g. professional soccer teams are characterized by having installations of grass heating systems serving the crucial commercial asset and at the same being the sub-system with the highest yearly thermal energy consumption. Public buildings of this size imply situation-specific operational modes combined with high levels of safety and comfort requirements. In this paper we provide a first study on the energy savings potential of a professional soccer stadium's grass heating system during day-to-day operation. In practice, limited heating capacities of the arena have to be adhered to, which causes the current operation to often result in under-performance of other, less critical facility units. Our analysis of dynamic operational and contextual data serves as foundation for long-term energy efficiency measures. We study relevant parameters related to the current control schemes and the stadium's context. Concretely, the grass root temperature as critical observable is studied with respect to weather conditions and the resulting thermal behavior. We provide an improved control strategy and quantify the anticipated savings of this strategy to be as high as 34% compared to the last heating season. For the future, the documented thermal characteristics will enable the formulation of more advanced control strategies to positively influence the grass heating operation. This will lead to further improvements in balancing the heating demand across all thermal facility sub-systems by integrating operational context with forecasts of the thermal behavior in the future.

Constantly increasing energy efficiency requirements for buildings call for top-quality systems and services for optimizing the building energy life cycle. The deep integration of ICT systems in heterogeneous building environment demands new approaches, advanced technical algorithms and sophisticated tools. In this paper we introduce our Intelligent Energy Management Platform (INTELLEM), an integrative solution for monitoring, analyzing and visualizing energy performance across multiple buildings. We verified our architectural approach by implementing a prototype capturing real data streams of considerable volume over a prolonged period of time, proving that our event-driven approach ensures the responsiveness of the whole system. We implemented a rich Web-interface for visualization, comprising a set of components for comprehensive analysis of building energy performance.