Adaptive model predictive climate control of multi-unit buildings using weather forecast data

https://doi.org/10.1016/j.jobe.2020.101449Get rights and content

Highlights

  • Saving 13% in building energy consumption.

  • Estimating Building model on-the-fly.

  • Energy optimization based on model predictive control method

Abstract

Energy use in buildings contributes a large part in global energy demand. To reduce energy use in this group of consumers, specially in cold seasons, an automatic control technique is proposed. In this paper, a model predictive controller (MPC) is employed to minimize the boiler activation time. The method uses the building model and incorporates the weather forecast data to act on the actuator in an optimal fashion while treating the user comfort constraints. This technique, as a part, can be embedded into the building energy management system. The building model parameters are obtained via an online identification process using unscented kalman filter (UKF). This identification is performed on-the-fly so the model of a building is continuously updated. The results of the system identification as well as the control performance are shown via Monte Carlo simulations, and compared with the results of a conventional control law. The comparison shows that the proposed method saves %13 energy consumption in the boiler activation.

Introduction

Energy management in buildings plays an important role in minimizing the global energy consumption. Providing an optimal temperature in buildings and workplaces is a significant step towards the energy management, regardless of the local climate. Buildings account for 20–40% of the world's total energy consumption, and this consumption is increasing in developed countries at 0.5–5% per annum [1]. Given that weather forecast data is available in the coming hours, designing and testing a predictive controller is an attractive idea for building control systems. The model predictive controller has two main parts: Selection of the mathematical model and the definition of the optimization problem.

In various studies, the energy consumption reduction potentials has been studied using model predictive control (MPC) [[2], [3], [4], [5]]. A hybrid multi-layered MPC is proposed in Refs. [3] in which the resident comforts are evaluated by the use of zone operative temperature criteria and the energy consumption is considerably reduced. Ref. [4] has compared a predictive controller with an on/off control scheme in which a linear optimization problem is defined using a temperature penalty that determines the regulation weight of the indoor temperature. The result showed an 18% reduction in the energy consumption. Overall, MPC is able to provide comfort to residents while reducing energy consumption compared to conventional BMS buildings [5]. In addition, the authors of [6] developed the MPC genetic algorithm, which significantly reduced the duration of residents’ thermal discomfort. However, in these studies the parameters are not estimated online and the probable changes in the system are not taken into account. Also, the climate scenario does not change in these studies and only a single unit as been considered.

Design of a model-based controller requires the system to be identified and a challenging issue in the use of MPC [7] for building automation is the lack of a proper model. There are three main structures for the modeling of a building system: Black-box, white-box, and gray-box. In Ref. [8] for building control purposes, an overview of three model structures is presented. The white-box model needs a detailed knowledge of the technical properties of the building and the use of analytic equations [9]. The black-box model however is described by output measurements corresponding to known inputs and fitting a nonlinear function to the obtained data (see Ref. [10]). The white-box and black-box models have disadvantages: The former has a vast number of parameters which makes the system complicated to be analyzed and the latter contains less parameters that are physically meaningless [11]. A gray-box model is a combination of the two methods. In Ref. [12,13], some new methods for improving the building gray-box model are presented using the information obtained from the closed-loop system. Yudong Ma et al. studied the thermal model of a two-mass room at which the controller is designed using the weather forecast and the comfort of the inhabitants [14]. A low-order modeling technique for describing the thermal behavior of a construction system is proposed by authors of [15,16].

Estimation of the state variables and the system parameters are discussed in several publications. In Ref. [17,18] an off-line parameter estimation is performed for two different systems of refrigeration and cold-water chiller, respectively. An artificial neural network has been introduced in Ref. [17], while different linear models has been validated in Ref. [18]. On-line parameter estimation using recursive least squares has been used for a heating ventilation and air conditioning (HVAC) system in Ref. [19]. Another example of on-line estimation using RLS is seen by authors of [20] that estimated a zone variable air volume system with seven parameters. In another studies, an unscented Kalman-filter (UKF) [21] and an extended Kalman-filter (EKF) [22] were used to estimate the thermal parameters of the building models. An EKF algorithm has recently been introduced. This algorithm works alternately manner and is based on parameter estimation and state prediction [23]. However, The state-of-the-art online estimations presented in the previous studies can be divided into two general categories: 1) single-zone buildings 2) multi-zone buildings. In these studies, the heating system for each zone is individually controlled and only the effect of the building zones is considered.

Sturzenegger et al. studied the modeling and optimization of energy consumption in different sectors [24,25]. In 2014, the first valid modeling framework for the building MPC design using a MATLAB toolbox was obtained [24], and in 2016, it is shown that using this toolbox together with solving an optimization problem, energy consumption is reduced about 17% in comparison to RBC [25]. In Ref. [26] an MPC controller with a building ventilation system is designed for a single classroom using a three-resistor-two-capacitor model which is linearized by means of a Taylor series expansion and demonstrates the role of MPC in comfort management of a thermal system.

In [27] the building model is trained and an optimal control framework is studied for the concurrent cooperation of HVAC and the battery energy saving while satisfying industrial standards and minimizing the peak load demand. In another example in Refs. [28], a learning-based framework is proposed to maintain resident comfort and reduce energy consumption.

In this paper, a gray-box model is used for the building system in which the dynamics of the model is developed using the heat transfer equation while the solar radiation heat is considered as an external disturbance. In the proposed model, all parameters have physical meanings and are estimated using the UKF algorithm. This nonlinear estimation strategy provides a flexible model that captures any probable changes in the system, i.e., adapts to a dynamic environment. After a pure estimation interval (star-up phase) the system parameters are converged to the actual values and the control loop can be closed as the model becomes reliable (steady-state phase). In this phase, a centralized MPC problem is solved at every sampling time. Since the heating boiler has an on/off behavior, a binary optimization is carried out within the MPC framework. The time delay in the system is also considered and approximated as a system parameter in order to be obtained in the estimation process. Finally, a system of two-unit building is solved as a numerical example. Therefore, the contributions of this paper in comparison to the state-of-the-art literature can be summarized as follows:

  • (1)

    An adaptive MPC scheme is utilized for a temperature control at which the online estimation helps the solution to be still optimal for a dynamic environment with variable parameters.

  • (2)

    A comparably simple gray-box model is selected in order to achieve a low feasible calculation time, while beside its parameters, the actuation time delay is considered and taken to the estimation process in order to make the algorithm applicable for a wide range of buildings with different pipeline systems.

  • (3)

    A two-unit model is considered in this paper which provides the buildings interaction to be taken into account.

  • (4)

    Compared to previous studies, a central heating system on a two-unit building has been investigated and therefore parameter estimation including time delay for the central controller is very important.

The remainder of this paper is structured as follows: In Section 2, the structures of the system and its analytic equations are presented. In Section 3, the estimation, optimization, and control methodologies are described in details. In Section 4, simulation results are presented and some discussions are made. Section 5 is dedicated to the concluding remarks.

Section snippets

System description

In this section, first a multi-unit building system is qualitatively described, then the gray box and equivalent circuit models are formulated as well. The system is modeled in order to design a model predictive controller based on the time delay, physical characteristics of the units, and the transmittance coefficient of the valves.

Control and estimation methodology

In this section, first, the parameters of the defined model is estimated using an online estimator. Then, the boiler activation time is determined using the predicted outside temperature and the system model behavior. This is done through an optimization, in which the energy consumption is minimized and the comfort zones of the units are treated.

Simulation results

This section presents the simulation results for a numerical example to show the effectiveness of the proposed controller design methodology. First, the results of the online estimation are presented, then the results of the controller design are presented. Finally, the proposed method is compared with the conventional climate control.

In Fig. 4, Fig. 5, the estimated parameters for units 1 and 2 are shown, respectively. The results of the Monte-Carlo simulation shows converging behavior for

Conclusions

In this paper, an automatic predictive control method is studied that minimizes the energy consumption in buildings while the estimation process makes the proposed methodology valid for a wide rage of buildings with different pipeline systems encountered by various dynamic environments. The control method (based on MPC) uses the weather forecast data as well as the building model, and tries to minimize the boiler activation time during cold seasons while treating the comfort zone constraints,

CRediT authorship contribution statement

Mohammad M. Mazar: Formal analysis, Software. Amin Rezaeizadeh: Conceptualization, Supervision, Writing - review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (30)

Cited by (9)

  • Occupant preferences on the interaction with human-centered control systems in school buildings

    2023, Journal of Building Engineering
    Citation Excerpt :

    Human-centered control systems have become viable with the development of predictive control systems such as Model Predictive Control (MPC). These control strategies have drawn attention due to their efficiency and robustness, as they incorporate weather conditions, occupancy, building physics, etc., to predict the desired indoor setpoints [3,4]. Another feature of MPC presents the integration of the occupant preference as a direct input into the control algorithm, where occupants can report their perceived comfort e.g., whether it’s too hot or too bright.

  • Model predictive control for energy-efficient optimization of radiant ceiling cooling systems

    2021, Building and Environment
    Citation Excerpt :

    The MPC controller reduced the energy consumption of the RCCUV system by 13.2%, compared to the PID (proportion integral differential) controller during the 2-h experimental period, confirming that the optimal performance of the MPC is superior to the PID control [31]. Mohammad M. Mazar [32] applied the MPC to minimize the energy consumption during the boiler's start-up time, using a building model combined with weather forecast data to operate the actuator in the best possible way while staying within the thermal comfort limits. A comparative analysis with conventional control methods showed that the proposed MPC saved 13% energy during the boiler's start-up time.

  • Model predictive control strategy using encoder-decoder recurrent neural networks for smart control of thermal environment

    2021, Journal of Building Engineering
    Citation Excerpt :

    The building sector is a major energy consumer in modern society, accounting for more than 30% of global final energy consumption [1]. Building automation and control systems are critical to improving building energy efficiency while ensuring a healthy and comfortable built environment [2,3–5] Numerous control strategies, such as fuzzy control [6,7], proportion-integration-differentiation (PID) control [8,9], and model predictive control (MPC) [10–13], etc., have been investigated to be applied for efficient operation of building systems in the past decade. In particular, increasing attention is paid to the MPC controller, due to its great advantage to realize optimal control and to deal with the trade-off among multiple control objectives [14].

  • A thermal comfort-driven model predictive controller for residential split air conditioner

    2021, Journal of Building Engineering
    Citation Excerpt :

    It has been shown that model predictive control (MPC) could be an efficient control strategy in controls of the HVAC system accounting for the future disturbances while maintaining the given thermal comfort constraints [29]. For example, Mazar et al. [30], Wu et al. [31], Schwingshackl et al. [32], and Ruano et al. [33] used MPC in thermal comfort-in-loop controllers. In these studies, to maintain thermal comfort, constrained optimization problems (e.g., minimization of HVAC energy consumption and tracking error between observed and reference set-point temperature) were solved over a prediction horizon.

  • All you need to know about model predictive control for buildings

    2020, Annual Reviews in Control
    Citation Excerpt :

    However, in some cases, linear dynamics might not be sufficiently accurate and can result in hampering the performance of the predictive controller (Kim, Witmer, Brownson, & Braun, 2014). In case of inaccurate disturbance models, stochastic (Drgoňa, Kvasnica, Klaučo, & Fikar, 2013; Parisio, Fabietti, Molinari, Varagnolo, & Johansson, 2014) or adaptive (Mazar & Rezaeizadeh, 2020) data-driven methods have been applied for mitigating the uncertainties associated with the weather forecast errors. For instance, (Liu, Paritosh, Awalgaonkar, Bilionis, & Karava, 2018) use a probabilistic time-series autoregressive model to quantify solar irradiance uncertainty.

  • Data-Driven Control for a Building Heating System

    2023, 11th RSI International Conference on Robotics and Mechatronics, ICRoM 2023
View all citing articles on Scopus
View full text