Journal ArticleDOI

# A convex approach to a class of non-convex building HVAC control problems: Illustration by two case studies

15 Apr 2015-Energy and Buildings (Elsevier)-Vol. 93, pp 269-281
TL;DR: In this article, a convexification approach is presented for a class of non-convex optimal/model predictive control problems more specifically applied to building HVAC control problems.

### 1. Introduction

• In the context of energy-efficient buildings HVAC control has gained increasing attention in recent years.
• This leads to the risk of designing a non-working controller on the real system or a working controller with suboptimal results.
• De Ridder et al. and Verhelst [12, 5] used mathematical model-based control methods for HybGCHP systems, which allow global optimization.
• There xist analytical formulas for a bilinear function or a rational function of two varibles.

### 2. PART I: Theoretical Foundations

• First, the authors introduce a class of non-convex control p blems which include the optimal/model predictive control of HybGCHP systems.
• Next, the proposed convex relaxation method for the given non-convexcontrol problems is detailed.
• Since in Section 3.2 the proposed convex relaxation is applied to optimal control of HybGCHP systems and the results are compared with its dynamic programming-based control results, the authors also give a short oveview of the dynamic programming at the end of this part.

### 2.1.2. Non-convex Model Predictive Control Problem Class

• The rest of descriptions of variables and functions are the same as in Section 2.1.1.
• In MPC control, the control input over the prediction horizon is calculated at every time step and the first element of the control input vector is applied at the current time step.
• At the next time step, the same calculation procedure is peated.
• Due to this implementation scheme, MPC is sometimes called receding horizon control.

### 2.2.1. Overview of convex optimization

• This means that if the authors take any two points inS and draw a line segment between these two points, then every point on that line segmentalso belongs toS. Definition 2.2 (Convex Function [13]).
• This definition means that if the authors take any two points on the graph of a convex functionf and draw a straight line between them, then the portion of thefunction between these two points will lie below this straight line.
• Armed with the above definitions of convex sets and functions, the authors are now ready to define a convex optimization problem.
• Definition 2.3 (Convex Optimization Problem: the most general form [13]).
• Given a convex functionf , a convex setS and the decision variable vectorx, the associated convex optimization problem is defined as minimizef(x) subject tox ∈ S. Definition 2.4 (Convex Optimization Problem: less general form [13]).

### 2.2.2. Convex Envelope

• Next, the concept of convex envelope is defined.
• Given a continuous functionk(x), its convex envelope, denoted byconvk(x), over a convex setS is defined as the pointwise supremum of all convex functions which are majorized byk(x): convk(x) = sup{r(x)| r convex andr(y) < k(y) ∀y ∈ S}.
• The overall system of relaxed constraints consists of 18 scalar convex constraints, which can be compactly represented as Efrw ≥ gfr-c(x1, x2), (12) whereEfr is a constant vector andgfr-c(x1, x2) is a convex function with respect to the variablesx1, x2.
• The derivation, although simple, is lengthy and hence was skipped.

### 2.4. Recap of Dynamic Programming

• Dynamic programming is a closed-loop, global optimal contrl method (global optimal up to approximations due to state-input gridding and interpolations).
• It is based on the “principle of optimality” [16] which simply says that in a multi-stage process whatever the previous states are, the remaining decisions must be optimal with regard to the state following from the current state.
• This principle allows the optimal control problem of aK-stage process to be recursively formulated starting from the last stage.
• For dynamic programming-based control methods, the most important issue is to have an accurate model with minimum number of states and inputs due to the famous curse of dimensionality problem [16, 17].
• The reader is referred to references [16, 17] for details on dynamic programming.

### 3. PART II: Applications

• The purpose of this section is to test the proposed convexification-based control methods first on a benchmark HVAC building control system case study from the literature to which the authors apply the convexified MPC control method and compare these control results with the adaptive and fuzzy control results from the literature.
• Next, the optimal control of a HybGCHP system is considered,which is the main application example for the developed methods.
• For the HybGCHP system, the convexified optimal control results are compared with the dynamic programming control results of the same system to assess the performanceof th convexified optimal control method.

### 3.1. Application to a Benchmark Problem

• The benchmark case study considered in this paper is the building heating control system, as described by Calvino et al. [18] and Chaudry and Das [19].
• The model parameter values used in Eq.(14) are taken from Chaudhry and Das [19] and are given in Table 1.
• Figure 3 shows the corresponding results when convexified MPC is usedwith Np = 10.

### 3.2.1. HybGCHP System Description

• It is assumed that the heat demand (Q̇h) is provided by the heat pump and the boiler and the cold demand (Q̇c) is provided by the passive cooler (using a heat exchanger instead of an active chiller) and the chiller.
• The coefficients of performance given by the above expressions depend on the temperatures of the source and the emission system, as expressed by COPhp = fhp(Tf , Tsw,h), COPpc = fpc(Tf , Tsw,c), COPch = fch(Ta, Tsw,c), whereTsw,h, Tsw,c represent the supply water temperature for heating and supply water temperature for cooling, respectively.
• Next, the authors will discuss these constraints and present their expressions.

### 3.2.2. Heat & Cold Demand Satisfaction

• The building heating and cooling demands should be satisfiedwith some acceptable violation margins: Q̇h(t)−.
• Note that the margins are taken to be time-dependentto allow different degrees of flexibility over time.
• During critical demand load periods these margins can be set very strictly.
• It is assumed thatQ̇h andQ̇c are given and hence building modeling is not included in the optimization.

### 3.2.3. Circulating Fluid Temperature Bounds

• The cooling of a building requires heat injection into the ground during summer.
• This increases the ground temperature towards winter,which, in turn, increases COPhp.
• The ground temperature, which is represented indirectly byTf , should be kept below the supply water temperature,Tsw,c, for passive cooling of the building.
• Similarly, heating of a building requires heat extraction from the ground.

### 3.2.5. Borehole Dynamics

• In the equivalent diameter approach, the heat transfer from the Utube is approximated by the heat transfer from a single pipe with a hypothetical diameter through which the heat exchanging fluid circulates.
• T are soil nodal temperatures andp is the known parameter vector including thermal, physical and other parameters of the system (diffusivitiesαg, αs, conductivitieskg, ks, different radii rfg, rgs, discretization step sizes,etc.) andunet is the net heat power injected to the per borehole length.
• POD is a flexible model-order reduction method compared to other model order reduction methods, also known as Remark 2.
• Here, the authors consider optimal control which is an open-loop control method where control actions are c lculated off-line and the accuracy of these control actions is depending on the“simulation performance” of the model, not on its “prediction performance” because in classical optimal control measured values are not used and everythingis done off-line.
• A too simple model cannot have a good simulation performance over a long period (like one year or multiple years), as shown in Figure 6.

### 3.2.6. Non-convex Optimal Control Problem for Total EnergyCost Minimization

• (23j) With this formulation, the non-convexity of the optimal control problem comes from Eq.(23j): a bilinear term.
• Note that the rational term includingTa in the cost function does not create any non-convexity becauseTa is not a decision variable or a function of decision variables.
• In the next subsection, Eq.(23j) will be replaced by its convex approximation.

### 3.2.7. Convexified Optimal Control Problem and Comparison with Dynamic Programming

• Note that all other constraints are already/or can be easilyput into the form given in Eq.(7).
• To apply dynamic programming for assessing the performanceof onvexified optimal control, the optimization problem given by Eq.(23)is equivalently reformulated with two inputs instead of four by assuming zero violat n margins (see Eq.(17)) to alleviate the curse of dimensionality problem in dynamic programming.
• This is done for allfe sible gridded states.
• Figures 7(a)-7(b) show the time evolution of the mean temperature ofthe circulating fluid (Tf ), Figures 7(c)-7(d) show the heat pump power and heating demand (Q̇hp, Q̇h), Figures 8(a)-8(b) show the chiller power and cold demand (Q̇ch, Q̇c) and finally Figures 8(c)-8(d) show the evolution of accumulated cost (J).

### 3.3. HybGCHP System Control Using Convexified MPC

• The non-convex model predictive control (NMPC) problem forHybGCHP system is shortly described as follows.
• Here, the authors do not present the results for convexified MPC, they just wanted to show how the proposed idea can be used in the context of NMPC.
• In practice this will cause the following problems.
• Firstly, the large-scale model with 506 states is not observable since there are 506 states and only one input-output pair.

### 4. Conclusion

• In this paper a convexification approach using convex envelopes for hard-tosolve non-convex optimization problems involving rational and/or bilinear terms of decision variables was proposed.
• For the first case study of an HVAC building control system, the performance of convexified MPC was compared to the performance of fuzzy and adaptive control from the literature.
• For the second application, the total energy costminimization of buildings with a HybGCHP system, convexified optimal control was used and its results were compared to dynamic programming based control results, which is a closedloop, global optimal control method (global optimal up to approximations in the gridding of states/inputs and used interpolations).
• The overall message of this paper is that given a non-convex optimization/control problem of thermal systems, the first step should be to analyse the given system in terms of the non-convex terms and then investigate whether convex envelopes for the associated terms exist or not.
• The accuracy degree of the approximation of the original non-convex optimization problem by a convex one may be very case dependent.

Did you find this useful? Give us your feedback

##### Citations
More filters
Journal ArticleDOI
TL;DR: This paper provides a unified framework for model predictive building control technology with focus on the real-world applications and presents the essential components of a practical implementation of MPC such as different control architectures and nuances of communication infrastructures within supervisory control and data acquisition (SCADA) systems.

276 citations

Journal ArticleDOI
TL;DR: In this paper, a compact overview of the state-of-the-art in modeling of ground-coupled heat pump (GCHP) systems and an in-depth review of their optimal control along with the associated research challenges are given.
Abstract: In this paper, a compact overview of the state-of-the-art in modeling of ground-coupled heat pump (GCHP) systems and an in-depth review of their optimal control along with the associated research challenges are given. The main focus is on optimal control but since design of an optimal controller may require a model, a relatively short literature review of modeling approaches is also discussed. Adopting the adage “a picture is worth a thousand words”, we tried to include a minimal number of representative schematics and result figures for some of the reviewed studies for clarity and a better understanding of the presented material. In addition to the literature review, we included our comments, points of view, alternative solutions and some potential future directions. This review paper is useful both for engineers and researchers involved in modeling and optimal control of GCHP systems. The second part of the paper, “Ground-Coupled Heat Pumps: Part 2 – Literature Review and Research Challenges in Optimal Design”, focuses on the literature review on optimal design and the associated design challenges for GCHP systems.

97 citations

Journal ArticleDOI
, T. Leurs2
TL;DR: In this article, a data-driven control approach for building climate control is presented based on reinforcement learning, where the underlying sequential decision making problem is cast into a Markov decision problem, after which the control algorithm is detailed.

93 citations

Posted Content
, T. Leurs2
TL;DR: Model-assisted batch reinforcement learning is applied to the setting of building climate control subjected to dynamic pricing and it is found that within 10 to 20 days sensible policies are obtained that can be used for different outside temperature regimes.
Abstract: Driven by the opportunity to harvest the flexibility related to building climate control for demand response applications, this work presents a data-driven control approach building upon recent advancements in reinforcement learning. More specifically, model assisted batch reinforcement learning is applied to the setting of building climate control subjected to a dynamic pricing. The underlying sequential decision making problem is cast on a markov decision problem, after which the control algorithm is detailed. In this work, fitted Q-iteration is used to construct a policy from a batch of experimental tuples. In those regions of the state space where the experimental sample density is low, virtual support samples are added using an artificial neural network. Finally, the resulting policy is shaped using domain knowledge. The control approach has been evaluated quantitatively using a simulation and qualitatively in a living lab. From the quantitative analysis it has been found that the control approach converges in approximately 20 days to obtain a control policy with a performance within 90% of the mathematical optimum. The experimental analysis confirms that within 10 to 20 days sensible policies are obtained that can be used for different outside temperature regimes.

81 citations

### Cites background from "A convex approach to a class of non..."

• ...When considering building climate control, MPC has received considerable attention in the recent literature [6, 7, 8, 26]....

[...]

Journal ArticleDOI
TL;DR: The residential and commercial building sector is known to use around 40% of the total end-use energy and is considered to be the largest energy consumer sector in the world as mentioned in this paper.
Abstract: The residential and commercial building sector is known to use around 40% of the total end-use energy and, hence, is considered to be the largest energy consumer sector in the world [1]. Approximately half of this energy is used for heating/cooling, ventilation, and air-conditioning (HVAC), and this usage is increasing 0.5?5% per year in developed countries [2]. The distribution of energy use percentages within the building for the United States is shown in Figure 1. This trend is similar for the rest of the world.

77 citations

##### References
More filters
Book
01 Mar 2004
TL;DR: In this article, the focus is on recognizing convex optimization problems and then finding the most appropriate technique for solving them, and a comprehensive introduction to the subject is given. But the focus of this book is not on the optimization problem itself, but on the problem of finding the appropriate technique to solve it.
Abstract: Convex optimization problems arise frequently in many different fields. A comprehensive introduction to the subject, this book shows in detail how such problems can be solved numerically with great efficiency. The focus is on recognizing convex optimization problems and then finding the most appropriate technique for solving them. The text contains many worked examples and homework exercises and will appeal to students, researchers and practitioners in fields such as engineering, computer science, mathematics, statistics, finance, and economics.

33,341 citations

Book
01 May 1995
TL;DR: The leading and most up-to-date textbook on the far-ranging algorithmic methododogy of Dynamic Programming, which can be used for optimal control, Markovian decision problems, planning and sequential decision making under uncertainty, and discrete/combinatorial optimization.
Abstract: The leading and most up-to-date textbook on the far-ranging algorithmic methododogy of Dynamic Programming, which can be used for optimal control, Markovian decision problems, planning and sequential decision making under uncertainty, and discrete/combinatorial optimization. The treatment focuses on basic unifying themes, and conceptual foundations. It illustrates the versatility, power, and generality of the method with many examples and applications from engineering, operations research, and other fields. It also addresses extensively the practical application of the methodology, possibly through the use of approximations, and provides an extensive treatment of the far-reaching methodology of Neuro-Dynamic Programming/Reinforcement Learning.

10,834 citations

### "A convex approach to a class of non..." refers background or methods in this paper

• ...The reader is referred to references [16, 17] for details on dynamic programming....

[...]

• ...For dynamic programming-based control methods, the most important issue is to have an accurate model with minimum number of states and inputs due to the famous curse of dimensionality problem [16, 17]....

[...]

Proceedings ArticleDOI
02 Sep 2004
TL;DR: Free MATLAB toolbox YALMIP is introduced, developed initially to model SDPs and solve these by interfacing eternal solvers by making development of optimization problems in general, and control oriented SDP problems in particular, extremely simple.
Abstract: The MATLAB toolbox YALMIP is introduced. It is described how YALMIP can be used to model and solve optimization problems typically occurring in systems and control theory. In this paper, free MATLAB toolbox YALMIP, developed initially to model SDPs and solve these by interfacing eternal solvers. The toolbox makes development of optimization problems in general, and control oriented SDP problems in particular, extremely simple. In fact, learning 3 YALMIP commands is enough for most users to model and solve the optimization problems

7,676 citations

### "A convex approach to a class of non..." refers methods in this paper

• ...The convexified optimal control problem is coded using YALMIP [30] in MATLAB and calling CPLEX [31] as solver....

[...]

Book
01 Jan 1970

3,442 citations

Journal ArticleDOI
TL;DR: For nonlinear programming problems which are factorable, a computable procedure for obtaining tight underestimating convex programs is presented to exclude from consideration regions where the global minimizer cannot exist.
Abstract: For nonlinear programming problems which are factorable, a computable procedure for obtaining tight underestimating convex programs is presented. This is used to exclude from consideration regions where the global minimizer cannot exist.

2,053 citations