Control Of Building Energy Systems Research Articles

Short-term thermal parameter prediction of building energy system based on MIE-JANET method

The precise prediction of thermal parameters in building energy systems is foundational for the efficient and stable operation of the system, aiding in the formulation of optimal building operation strategies. Currently, in the field of architecture, there is a lack of parameter prediction for thermal parameters that involve multiple devices and thermal processes. Moreover, the widely used long short-term memory (LSTM) algorithm in this field tends to have a large number of network parameters, leading to longer training times. Aiming at these problems, a short-term predictive method for thermal parameters named MIE-JANET is proposed. Mutual information entropy was employed to select relevant features from system operation data. Multiple short-term prediction models were constructed by JANET mdoel. The results indicate the superior predictive performance compared to similar algorithms. When considering insufficient data, the mean absolute error for the JANET decreases by 4.9% relative to LSTM. With the same structure, the training time of the JANET model was reduced by more than 40% compared to LSTM. Moreover, The accuracy of the room temperature model, return water temperature model, and supply water temperature model has significantly improved compared to the physical model. This approach provides a foundational model for the intelligent control of building energy systems.

Safe operation of online learning data driven model predictive control of building energy systems

Model predictive control is a promising approach to reduce the CO2 emissions in the building sector. However, the vast modeling effort hampers the widescale practical application. Here, data-driven process models, like artificial neural networks, are well-suited to automatize the modeling. However, the underlying data set strongly determines the quality and reliability of artificial neural networks. In general, the validity domain of a machine learning model is limited to the data that was used to train it. Predictions based on system states outside that domain, so-called extrapolations, are unreliable and can negatively influence the control quality.We present a safe operation approach combined with online learning to deal with extrapolation in data-driven model predictive control. Here, the k-nearest neighbor algorithm is used to detect extrapolation to switch to a robust fallback controller. By continuously retraining the artificial neural networks during operation, we successively increase the validity domain of the artificial neural networks and the control quality.We apply the approach to control a building energy system provided by the BOPTEST framework. We compare controllers based on two data sets, one with extensive system excitation and one with baseline operation. The system is controlled to a fixed temperature set point in baseline operation. Therefore, the artificial neural networks trained on this data set tend to extrapolate in other operating points. We show that safe operation in combination with online learning significantly improves performance.

A physical model with meteorological forecasting for hourly rooftop photovoltaic power prediction

Accurate photovoltaic power forecasting provides essential information for the flexible control of building energy systems. This paper proposes a physical model with environmental parameter prediction and an improved maximum power point tracking algorithm for hourly photovoltaic power forecasting. This study incorporates the forecast of meteorological parameters into the photovoltaic model so that the physical model proposed in this paper can achieve photovoltaic power prediction in the face of different weather conditions compared to previous photovoltaic physical models in the literature. A hierarchical clustering algorithm was used to obtain future hourly irradiance based on different weather conditions. The coordinate analysis method was used to calculate the hourly irradiance received on the surface of the photovoltaic panel. An equivalent circuit model was used to calculate the current-voltage characteristics of photovoltaic panels. Finally, the output voltage of the photovoltaic panel was adjusted by the improved maximum power point tracking algorithm to obtain the photovoltaic power. This algorithm can accelerate the calculation process and avoid long convergence times or oscillations near the optimal value. A photovoltaic project located in central China was selected as a case study to verify the accuracy of the prediction model. The long-term monitoring results show that the relative error of the predicted irradiance of the photovoltaic panel surface is 18.5%. The mean absolute error of the forecasted photovoltaic power was 15.9% in 120 consecutive days under various weather conditions, indicating that the model had high accuracy compared with traditional machine learning methods.

A new framework integrating reinforcement learning, a rule-based expert system, and decision tree analysis to improve building energy flexibility

This study presents a new framework that integrates machine learning and a domain knowledge-based expert system to improve building energy flexibility. In this framework, a rule-based expert system was used to maximize the self-consumption of solar photovoltaics (PV) power, while a reinforcement learning (RL) agent was constructed to efficiently optimize the grid power import for battery charging and facilitate decision-making for battery discharging in response to the time of use electricity prices. Meanwhile, a Classification and Regression Tree (CART) model was developed to quantitatively analyze the relationships between building energy flexibility and external variables of interest to enhance the explainability of the framework. This work integrates the advantages of safety and simpleness of the domain knowledge-based expert system and the exploration and optimization capability of RL into building energy management. The performance of the proposed framework was evaluated using the four-year data collected from a real net zero energy office building. The system cost reduction ratio, self-consumption ratio, and self-sufficiency ratio were used as the energy flexibility indicators. The results showed that the integration of the RL and the rule-based expert system was able to reduce the electricity cost and grid power consumption by 7.0% and 10.6% respectively, and increase the self-consumption of PV power by 9.2% as compared with the use of the rule-based expert system only. The CART analysis also showed that external conditions can significantly influence the level of building energy flexibility. For instance, the average daily cost reduction ratio was 0.89 out of 1.0 when the daily maximum solar irradiance was above 717.5 W/m2, while it decreased to 0.28 when the daily mean solar irradiance was below 62.4 W/m2. This strategy can be used to facilitate building demand-side management and improve the design and control of building energy systems for enhanced demand flexibility.

A Review of Reinforcement Learning for Controlling Building Energy Systems From a Computer Science Perspective

Energy efficient control of energy systems in buildings is a widely recognized challenge due to the use of low temperature heating, renewable electricity sources, and the incorporation of thermal storage. Reinforcement Learning (RL) has been shown to be effective at minimizing the energy usage in buildings with maintained thermal comfort despite the high system complexity. However, RL has certain disadvantages that make it challenging to apply in engineering practices. In this review, we take a computer science approach to identifying three main categories of challenges of using RL for control of Building Energy Systems (BES). The three categories are the following: RL in single buildings, RL in building clusters, and multi-agent aspects. For each topic, we analyse the main challenges, and the state-of-the-art approaches to alleviate them. We also identify several future research directions on subjects such as sample efficiency, transfer learning, and the theoretical properties of RL in building energy systems. In conclusion, our review shows that the work on RL for BES control is still in its initial stages. Although significant progress has been made, more research is needed to realize the goal of RL-based control of BES at scale.

Model predictive control of a building renewable energy system based on a long short-term hybrid model

Considering solar photovoltaic (PV) usage in building energy systems (BES), energy systems that combine batteries and solar PV (PVB) have been widely used in buildings. Ensuring the safety of the battery and the operation requirements of other types of equipment to the extent possible while achieving the optimization goal has become an important challenge in the implementation of this system. In this case, model predictive control (MPC) has gained increasing attention in the operation control of BES owing to its strong fitting ability. However, the prediction models at the core of MPC are not adequately designed for the BES optimization problem, and numerous studies have not yet been modeled and tested in real buildings. In this study, we used a long- and short-term hybrid prediction model in the MPC framework. The proposed prediction models and MPC framework were validated using a measured dataset from a real office building in Japan. The results show that, compared with the currently used baseline control logic, the proposed hybrid prediction MPC framework can improve the battery safety index by 81.6%, while the combined heat and power continuous operation index can also be improved by 36.4%. Compared with the conventional MPC framework, the proposed hybrid prediction MPC framework improves the optimization of off-grid operation by approximately 69% while maintaining the remaining optimization objectives.

Multi-agent reinforcement learning dealing with hybrid action spaces: A case study for off-grid oriented renewable building energy system

With the application of renewable energy in building energy systems (BES), an increasing number of power grids require building energy systems coupled to realize off-grid operation which is one type of energy flexible and grid responsive operations. In this case, deep reinforcement learning (DRL) algorithms have gained more and more attention in the operation control of BES due to their strong fitting ability and model-free utilization characteristics. However, mainstream DRL algorithms cannot solve the reinforcement learning problem of hybrid action spaces, which also restricts its further application in BES including variety of energy sources. In this paper, we firstly use a multi-agent deep reinforcement learning algorithm (MADRL) to solve the RL problem of hybrid action spaces in the building controls domain. The proposed algorithm is validated on a measured dataset of a real office building in Japan. The results show that compared to the currently used baseline control logic, MADRL can achieve a 60% improvement in off-grid operation tasks. For battery safety, MADRL can reduce unsafe battery runtime by at least 80%. Furthermore, through experiments we find that the single-agent DRL algorithm cannot solve the reinforcement learning problem with hybrid action spaces. The MADRL framework achieves stable training and optimization by layering problems and agents.

Two-time-scale coordinated optimal control of building energy systems for demand response considering forecast uncertainties

Predictive scheduling is essentially needed to optimize energy dispatch of building energy systems for demand response, as it can maximize the benefits considering future conditions. However, the optimized demand response probably cannot be achieved and energy demands probably cannot be fully satisfied in operation due to the large discrepancy between the predicted and actual conditions. Generic and comprehensive optimal control methods of building energy systems for demand response, which can achieve maximized benefits while ensuring satisfaction of actual energy demands, are still absent. In this study, a two-time-scale coordinated optimal control strategy is proposed to optimize energy dispatch between building energy systems for demand response considering forecast uncertainties. The control strategy includes two coordinated schemes: a stochastic scheduling scheme and a real-time optimal control scheme. Forecast uncertainties are quantified based on real meteorological data. The optimal start time of scheduling optimization horizon is investigated. The control strategy was tested on a platform with PV and battery, established based on the Zero Carbon Building in Hong Kong. Results show that the optimal scheduling optimization horizon start time is the initial of the off-peak period. The proposed strategy reduces up to 14.7% cost compared with existing strategies while satisfying actual energy demands.

Development of a Soft Actor Critic deep reinforcement learning approach for harnessing energy flexibility in a Large Office building

This research is concerned with the novel application and investigation of ‘Soft Actor Critic’ based deep reinforcement learning to control the cooling setpoint (and hence cooling loads) of a large commercial building to harness energy flexibility. The research is motivated by the challenge associated with the development and application of conventional model-based control approaches at scale to the wider building stock. Soft Actor Critic is a model-free deep reinforcement learning technique that is able to handle continuous action spaces and which has seen limited application to real-life or high-fidelity simulation implementations in the context of automated and intelligent control of building energy systems. Such control techniques are seen as one possible solution to supporting the operation of a smart, sustainable and future electrical grid. This research tests the suitability of the technique through training and deployment of the agent on an EnergyPlus based environment of the office building. The agent was found to learn an optimal control policy that was able to minimise energy costs by 9.7% compared to the default rule-based control scheme and was able to improve or maintain thermal comfort limits over a test period of one week. The algorithm was shown to be robust to the different hyperparameters and this optimal control policy was learnt through the use of a minimal state space consisting of readily available variables. The robustness of the algorithm was tested through investigation of the speed of learning and ability to deploy to different seasons and climates. It was found that the agent requires minimal training sample points and outperforms the baseline after three months of operation and also without disruption to thermal comfort during this period. The agent is transferable to other climates and seasons although further retraining or hyperparameter tuning is recommended.

Coordinated Multi-Time Scale Optimal Control of Building Energy Systems for Demand Response Considering Forecast Uncertainties

As an economical way to address the great challenge from increasing renewable energy generation of intermittent nature, demand response is attracting increasing attention recently. Predictive scheduling is widely used to optimize energy dispatch of building energy systems for demand response, since it can minimize cost considering time-varying electricity price. However, it probably cannot satisfy energy demands and achieve optimized demand response in actual operation due to forecast uncertainties. In this study, a coordinated multi-time scale optimal control strategy is therefore proposed to optimize energy dispatch between building energy systems for demand response considering forecast uncertainties. The control strategy coordinates a stochastic scheduling scheme and a real-time optimal control scheme using coordinating control variables. The selection of coordinating control variables is comprehensively analyzed. The optimal start time of scheduling optimization horizon is identified. Forecast uncertainties are quantified based on real meteorological data. The proposed control strategy was tested and evaluated on the energy system of the Zero Carbon Building in Hong Kong. Results show that it is essential and beneficial to consider forecast uncertainties and coordinate stochastic scheduling with real-time optimal control to ensure satisfaction of energy demands and minimization of energy cost in actual operation when providing demand response.

A Pattern-Recognition-Based Ensemble Data Imputation Framework for Sensors from Building Energy Systems

Building operation data are important for monitoring, analysis, modeling, and control of building energy systems. However, missing data is one of the major data quality issues, making data imputation techniques become increasingly important. There are two key research gaps for missing sensor data imputation in buildings: the lack of customized and automated imputation methodology, and the difficulty of the validation of data imputation methods. In this paper, a framework is developed to address these two gaps. First, a validation data generation module is developed based on pattern recognition to create a validation dataset to quantify the performance of data imputation methods. Second, a pool of data imputation methods is tested under the validation dataset to find an optimal single imputation method for each sensor, which is termed as an ensemble method. The method can reflect the specific mechanism and randomness of missing data from each sensor. The effectiveness of the framework is demonstrated by 18 sensors from a real campus building. The overall accuracy of data imputation for those sensors improves by 18.2% on average compared with the best single data imputation method.

Model predictive control of building energy systems with thermal energy storage in response to occupancy variations and time-variant electricity prices

Developing effective operational strategies for commercial buildings is a high priority as the global community seeks to reduce energy usage and greenhouse gas emissions. This study aims to validate the feasibility of a model predictive control (MPC) strategy for commercial building in response to occupancy variations and time-variant electricity prices in comparison to a conventional rule-based control (RBC) strategy. The building energy system included an air-cooled chiller, stratified chilled water thermal energy storage, two fan coil units, three heat exchangers, and five pumps. The optimal operations of the chiller and storage system were determined with the goal of minimizing the operating cost while maintaining the zone temperature at a cooling set point temperature during cooling operation hours. Artificial neural network was utilized as prediction models and metaheuristics algorithm was employed as optimization solver to construct a reliable and computationally manageable MPC controller. The simulation was performed for four days during the cooling season with a confirmed optimal prediction time horizon of 24 h and a control timestep of 1 h intervals. In conclusion, MPC reduced the total operating cost by 3.4% compared to the RBC, which prioritized the storage system operation to manage the thermal load.

Effects of various partitions on the accuracy of virtual in-situ calibration in building energy systems

Sensor errors significantly affect the operation and control of building energy systems. Thus, using correct and reliable sensors can effectively reduce the energy consumption of building energy systems. Virtual in-situ calibration (VIC), which is based on Bayesian inference and the Markov Chain Monte Carlo method, can help avoid the installation of new sensors, effectively reduce the systematic and random errors of sensors, and improve the reliability of collected data. In the current study, six working conditions were designed to check the robustness and accuracy of the proposed VIC technology. It's shown that for the systematic and random error of various sensors, the improved component calibration method is the most accurate. Contrarily, the whole calibration method is the least accurate, whereas the performance of the three local calibrations is intermediate. The maximum errors of the whole, local, and component calibration methods were 973%, 112.7%, and 30%, respectively. This verifies that component calibration effectively reduces the possibility of abnormal data and also enhances the reliability of sensor measurements. For the whole and local calibration methods, the constraints between various parameters are significantly reduced because of the random and uncertain coefficients; consequently, a majority of the correction results deviate from their true values. It is not necessary to determine the sensitivity and normalizing coefficients through a complicated optimization algorithm or historical experience by using the proposed component calibration. Therefore, the component calibration method features advantages in terms of computational time and correction accuracy.

Model-based multi-objective predictive scheduling and real-time optimal control of energy systems in zero/low energy buildings using a game theory approach

Online optimal control of energy systems plays a significant role in achieving the zero/low energy goal and high energy efficiency for zero/low energy buildings. Increasing amount of studies are conducted on the multi-objective optimal control of building energy systems in recent years since multiple objectives are often concerned in practice. The weighted sum method is widely used to solve the multi-objective optimization problems for online optimal controls. However, it is often difficult and even impractical to determine proper weights for objectives of different natures. Existing studies on online optimal control of energy systems in zero/low energy buildings focused on the predictive scheduling, and very few studies addressed the real-time optimal control aspect which is also essential in operation. In this study, a coordinated online multi-objective control strategy, consisting of two control optimization schemes, is therefore proposed for predictive scheduling and real-time optimal control of energy systems in zero/low energy buildings. A cooperative game theory-based method is adopted for the online multi-objective optimizations. The control strategy was tested and evaluated via simulation of the energy systems in a zero energy building with battery storage on two typical days. Control variables and weights of objectives were optimized to minimize the combined optimization objective involving energy cost and grid impact. The test results show that it is essential and beneficial to coordinate the predictive scheduling and real-time optimal control in actual operation. The cooperative game theory-based method is effective for the online multi-objective optimization without the need of setting weights of different objectives.

Distributed model predictive control of building energy systems coupled to geothermal fields

In building energy systems (BES), operation strategies are a key factor to exploit the full potential of renewable energy sources and thus to reduce the CO2 emission and energy costs. However, advanced operation strategies like model predictive control require high engineering effort, which limits the applicability. To ease the implementation of model predictive control in BES, we present a flexible and robust distributed control framework. The framework supports the use of different model types and timescales for the various subsystems, allowing the user to tailor models from white to black modelling according to the available system knowledge. In a case study, we apply this framework to control a real-life industrial hall. We further simulate a case in which the hall is supplied by a geothermal field to investigate the sustainable long-term control of the field. We vary the temperature setpoint of the hall and adjust the thermostats in the office rooms. Since the set temperature of the hall and offices differs, the model predictive control framework finds an economic trade-off between the requirements of those two zones. Judging from these results, the approach and our implementation could be an important contribution to energy efficient operation of buildings with renewable energy sources.

A systematic feature selection procedure for short-term data-driven building energy forecasting model development

An accurate building energy forecasting model is the key for real-time model based control of building energy systems and building-grid integration. Data-driven models, though have lower engineering cost during their development process, often suffer from poor model generalization caused by high data dimensionality. Feature selection, a process of selecting a subset of relevant features, can defy high dimensionality, increase model interpretability, and enhance model generalization. In building energy modeling research, features are often selected based on domain knowledge. There lacks a comprehensive methodology to guide a systematic feature selection procedure when developing building energy forecasting models.In this research, a systematic feature selection procedure for developing a building energy forecasting model is proposed which attempts to integrate statistical analysis, building physics and engineering experiences. The proposed procedure includes three steps, i.e., (Step 1) feature pre-processing based on domain knowledge, (Step 2) feature removal through filter methods to remove irrelevant and redundant variables, and (Step 3) feature grouping through wrapper method to search for the best feature set.Two case studies are presented here using both simulated and real building data. The simulated building data are generated from a medium-size office building (a DOE reference building) simulation model. The real building data are obtained from a medium-size campus building in Philadelphia, PA. In both cases, the energy forecasting models that are developed using proposed systematic feature selection procedure is compared with models using other feature selection techniques. Results show that the models developed using proposed procedure have better accuracy and generalization.

A novel hybrid agent-based model predictive control for advanced building energy systems

The development of new energy efficient components and complex energy concepts in recent years has heightened the need to design advanced control strategies. The main objective of this research is to develop a control strategy for building energy systems to save primary energy by applying the concept of multi-agent systems. Most of the advanced control strategies have been developed to be energy efficient, while their objectives are obtained from energy analysis. However, an energy analysis is unable to provide information on the quality of energy streams flowing through a system. In this study, exergy is selected as the objective of the optimization. To reach the goal of this research, an agent-based control for building energy systems using the exergy cost functions is developed. Agent-based control, which takes into account the interactions among the components of the system, offers a promising solution to the need for more advanced control strategies for complex building energy systems. The classical agent-based control developed in this study is combined with model predictive control, which leads to a novel hybrid agent-based model predictive control for the optimization of advanced building energy systems from an exergy point of view. For evaluation purposes, a case study is defined and modeled, which is controlled by a reference control and the agent-based under the same circumstances through software-in-the-loop simulations. The results show that the agent-based control is able to reduce the primary energy consumption by 2 percent while maintaining the room air temperature at the same level of the reference case.

Prioritised objectives for model predictive control of building heating systems

Advantages of Model Predictive Control (MPC) strategies for control of building energy systems have been widely reported. A key requirement for successful realisation of such approaches is that strategies are formulated in such a way as to be easily adapted to fit a wide range of buildings with little commissioning effort. This paper introduces an MPC-based building heating strategy, whereby the (typically competing) objectives of energy and thermal comfort are optimised in a prioritised manner. The need for balancing weights in an objective function is eliminated, simplifying the design of the strategy. The problem is further divided into supply and demand problems, separating a high order linear optimisation from a low order nonlinear optimisation. The performance of the formulation is demonstrated in a simulation platform, which is trained to replicate the thermal dynamics of a real building using data taken from the building.

A Modelica library for the agent-based control of building energy systems

Building energy systems account for one third of the primary energy demand in the OECD member countries. Due to global warming and growing energy scarcity, a higher energy efficiency of these systems is required. As a result, building energy systems become more complex and often contain multiple heat and cold suppliers. The selection of the ideal supplier based on the current system state requires new control strategies. Multi agent systems depict a promising technology for the problem. There are agent-based frameworks available for many programming languages, but not for the modelling language Modelica, which is commonly used for building simulation. In the course of this work, a Modelica library for the agent-based control of building energy systems based on market mechanisms is introduced. The structure of the library, different types of agents, the concepts of agent communication and the trading of capacity adjustments are discussed. The library offers a variety of cost functions in order to realize different optimization goals. The functionality of the concept is proven with a simulation example of a building energy system controlled with agents from the introduced library. The system depicts a plug&play solution to optimize the performance of complex building energy systems with interchangeable optimization goals. Due to communication via Ethernet, the control system can be coupled to real energy systems.

Evaluation of Reinforcement Learning Control for Thermal Energy Storage Systems

This paper describes a simulation-based investigation of machine-learning control for the supervisory control of building energy systems. Model-free reinforcement learning control is investigated for the operation of electrically driven cool thermal energy storage systems in commercial buildings. The reinforcement learning controller learns to charge and discharge a thermal storage tank based on the feedback it receives from past control actions. The learning agent interacts with its environment by commanding the thermal energy storage system and extracts cues about the environment solely based on the reinforcement feedback it receives, which in this study is the monetary cost of each control action. No prediction or system model is required. Over time and by exploring the environment, the reinforcement learning controller establishes a statistical summary of plant operation, which is continuously updated as operation continues. The controller learns to account for the time-dependent cost of electricity (both time-of-use and real-time pricing), the availability of thermal storage, part-load performance of the central chilled water plant, and weather conditions. Though reinforcement learning control proved sensitive to the selection of state variables, level of discretization, and learning rate, it effectively learns a difficult task of controlling thermal energy storage and displays good performance. The cost savings compare favorably with conventional cool storage control strategies but do not reach the level of predictive optimal control.