June 11 - 14, 2013
Keywords of the presentation: Buildings, Decentralized Control, Clustering
In this talk we consider the modeling and control of building systems represented as large and complex systems. First, we present a motivation for why we believe a centralized approach to optimal operation may not be the best approach. Then, we present an approach for clustering building zones so as to create a decentralized architecture that balances achievable performance with tolerance to sensor/actuator faults. Here, performance is measured by temperature regulation. The clustering procedure is an agglomerative one in which we define 2 metrics that result in a Pareto-optimal tradeoff and then search for a Nash-like equilibrium between them. Subsequently, we design a model predictive control (MPC) approach for performing on-line optimization of controlled HVAC within a building. An Energy-Plus simulation of a commercial building under MPC control is given to demonstrate the approach.
Time permitting, we will discuss the HVAC mechanical systems within buildings. The discussion will start from a complex systems-level viewpoint and drill down to physical modeling of multi-phase flows in heat-pump/AC systems.
Forecasts will play an increasingly important role in the next generation of control systems. In nominal conditions, predictions of system dynamics, human behavior and environmental conditions can be used by the control algorithm to improve the performance of the resulting system. However, in practice, constraint satisfaction, performance guarantees and real-time computation are challenged by the complexity of the engineered system and uncertainty in the environment where the system operates. Read More...
In this talk I will present the theory and tools that we have developed over the past ten years for the systematic design of predictive controllers for uncertain linear and nonlinear systems. I will first provide an overview of our theoretical efforts. Then, I will focus on our recent results in stochastic predictive control design for smart buildings.
More info on: www.mpc.berkeley.edu
Keywords of the presentation: optimal control, diagnostics, commercial buildings, intelligent buildings
Compared to other energy sectors (e.g., transportation, industrial), there are significant challenges in realizing energy efficiency improvements in buildings due to the structure of the marketplace and the fact that buildings are not mass produced. In particular, intelligent control and diagnostic strategies that can optimize operations have not been widely implemented in buildings because each system tends to be unique and the costs associated with developing and updating specialized software can be prohibitive. Even so, the opportunities for energy and cost savings through optimized operations are very significant and there is renewed interest in capturing these savings. This talk will provide an overview of the barriers and opportunities for intelligent operations in buildings. The general problem of optimal supervisory control with time-of-use and demand charges will be posed and example savings results will be presented for small commercial buildings. A number of general supervisory control and diagnostic strategies and algorithms will be presented that can be embedded within individual controllers to substantially reduce site-specific software engineering costs. The use of modeling, optimization, and testing in the development and evaluation of embedded solutions is critical and will be addressed.
Keywords of the presentation: reduced-order model, optimal control
In most cases mathematical models for building thermal control assume that the state of the zone-air can be represented as a single temperature - the well-mixed hypothesis. In cases such as non-uniform solar loads, or buoyancy-driven cooling such models are problematic. In recent work we have developed data-driven approaches wherein results from CFD simulations are used to develop linear-time-invariant reduced-order models (ROM). In this talk we discuss two such approaches: one based on a subspace iteration to optimally reduce the dynamic order, and a second, ad-hoc approach, that captures important long-term behavior. The ROM is being used by colleagues at Purdue to study implementations of Model Predictive Control (MPC), which has emerged as a preferred approach to optimal closed-loop control of building energy systems. MPC control is commonly realized in update form by solving a sequence of finite-dimensional nonlinear programming problems (NLP) that approximate the discrete-time behavior of the building thermal dynamics on the remaining time horizon. It is of interest to compare NLP results with controls arising from applying optimality conditions to a corresponding continuous-time optimal control formulation. Our objectives are to benchmark the NLP solutions and to develop heuristics that lead to more efficient NLP parameterizations. We study a rather simple, three-state model of a single zone with comfort limits imposed as a state-inequality constraint.
Keywords of the presentation: Energy Modeling, Uncertianty, Control
In an attempt to enhance the performance of the built environment, models are being used everywhere to better design or operate buildings. Whole-building energy models are being used for design trades and for compliance (e.g. LEED) in design firms and these or reduced order versions are used for online predictive optimization or fault diagnostics after the building is commissioned. The models used vary greatly in their complexity, size (the number of physical attributes of the building they intend to capture), computational cost and flexibility (access to dynamic states etc.). In this talk we discuss different modeling approaches, uncertainty in these models and how to overcome some of the computational challenges that different models introduce. We will also discuss the notion of closing some gaps between different modeling methods to provide seamless methods to use the same models across the design and operation stages of a buildings life cycle.
Keywords of the presentation: Buildings, Energy Modeling, Software
EnergyPlus is the US Department of Energy’s flagship software tool for forward modeling of energy-related performance of buildings. This presentation will provide an overview of the project with an emphasis on numerical challenges. The talk will summarize a Request For Information and Funding Opportunity Announcement that DOE is planning related to an effort to reengineer the software with help from the private sector.
Keywords of the presentation: stochastic model predictive control, rule extraction, occupant-driven uncertainty, distributed model predictive control, building/grid feedback
This talk describes currently ongoing work related to optimal building control in which a range of computational challenges have to be addressed
a) Stochastic Optimal Control of Mixed Mode Buildings Considering Occupant-Driven Uncertainty
In the first part of this presentation, we present a methodology for determining viable control rules for commercial building systems that account for occupant driven uncertainty. The methodology consists of three steps; first, offline stochastic model predictive control (SMPC) utilizing detailed building simulation models is used to determine optimal setpoint schedules. Second, the optimal SMPC results are used to inform a machine learning algorithm for rule extraction, specifically a classification and regression tree (CART). Third, the CART model results are refined and transformed into readable building automation system (BAS) code. A representative set of results are provided to illustrate the approach.
b) Distributed Residential HVAC Model Predictive Control for Load Prediction and Peak Demand Management
In the second part, we turn to residential building control. Within the past few years, in-home internet gateways and smart thermostats have become the technology standard for residential demand-side management of electric loads though utility sponsored demand response (DR) programs. At the same time, the computational power of these devices has increased to a point that these devices are able to learn the physical characteristics of a home and perform complex energy simulations. In the near future, these devices may obviate the need for traditional demand response for managing residential HVAC electric load, replacing it instead with highly distributed network of load predicting and optimizing systems using model predictive control techniques. We discuss a simulation environment coupling electric grid modeling software with distributed, residential MPC 'agents' based on inverse gray box models that respond to grid signals and optimize operation in order to achieve a particular utility objective. Initial results on the impact to grid operation and stability will be discussed.
This talk is about design of feedback gains that achieve a desirable
tradeoff between quadratic performance of distributed systems and
controller sparsity. Our approach consists of two steps. First, we
identify sparsity patterns of the feedback gains by incorporating
sparsity-promoting penalty functions into the optimal control problem,
where the added terms penalize the number of communication links in the
distributed controller. Second, we optimize feedback gains subject to
structural constraints determined by the identified sparsity patterns.
In the first step, the sparsity structure of feedback gains is
identified using the alternating direction method of multipliers, a
powerful algorithm well-suited to large optimization problems. This
method alternates between promoting the sparsity of the controller and
optimizing the closed-loop performance, which allows us to exploit the
structure of the corresponding objective functions. In particular, we
take advantage of the separability of the sparsity-promoting penalty
functions to decompose the minimization problem into sub-problems that
can be solved analytically. Several examples are provided to demonstrate
the effectiveness of the developed approach and the accompanying
software LQRSP (available at www.umn.edu/~mihailo/software/lqrsp/).
Keywords of the presentation: Data center energy management, Real-time sensor measurements, Physics models, Numerical simulations
We consider the problem of modeling air flow, heat transfer, and humidity in buildings equipped with a network of sensors and a data management system which gathers the sensor data in real time. Such sensor measurements serve as input data for the boundary value problems for systems of partial differential equations comprising the physical models. First, we describe a simplified physics model for simulating air and heat transfer in data centers and show results from a case study indicating that the model produces approximations to temperature distributions which compare favorably with 3D experimental thermal measurements, even when the amount of input information provided by the real-time sensor data is limited. We then consider simulating humidity along with air flow and heat transfer, specifically in environments where natural convection is dominant. We will discuss modeling techniques that facilitate performing simulations of the physical phenomena at hand and which are suitable for coupling with measurements gathered via sensor networks.
Keywords of the presentation: fault detection and diagnostics; EnergyPlus; real-time; building energy performance
Building energy systems often consume approximately 16% more energy than is necessary due to system deviation from the design intent. Identifying the root causes of energy waste in buildings can be challenging largely because energy flows are generally invisible. To help address this challenge, we present a model-based, real-time whole building energy diagnostics and performance monitoring system. The proposed system continuously acquires performance measurements of heating, ventilation and air-conditioning, lighting and plug equipment usage and compare these measurements in real-time to a reference EnergyPlus model that either represents the design intent for the building or has been calibrated to represent acceptable performance. A proof-of-concept demonstration in two real buildings will be presented. The challenges and key learning will be discussed.
Keywords of the presentation: Distributed control, networks, heat flow, power flow
Many problems in optimization of transportation networks for heat and power can be stated in terms of matrices with non-negative coefficients. Moreover, dynamical models for such systems often have monotone step responses. This has great advantages in design and verification of controllers for large-scale networks. In particular optimal controllers can be computed using linear programming, with a complexity that scales linearly with the number of states and interconnections. Hence two fundamental advantages are achieved compared to classical methods for multivariable control: Distributed implementations and scalable computations. We will present several examples and look forward to discussing the relevance for energy-efficient buildings.
For about ten years several research groups at ETH have been investigating the design and implementation of advanced techniques for the control of indoor climate and their impact on energy use and comfort. In my talk I will summarize these efforts on building modeling and identification, and the use of weather, occupancy and utility price forecasts for control, including information on the uncertainty (stochastic and robust control). I will also describe an on-going project on implementing and testing a predictive control scheme on a large office building in Basel.
Keywords of the presentation: gaussian process, large-scale, scalability, buildings
We review applications and algorithmic challenges of Gaussian Process (GP) modeling. GP is a powerful and flexible uncertainty quantification and data analysis technique that enables the construction of complex models without the need to specify algebraic relationships between variables. This is done by working directly in the space of the kernel or covariance matrix. In addition, it derives from a Bayesian framework and, as such, it naturally provides predictive distributions. We describe how these features can be exploited in Measurement and Verification (M&V) tasks and in the construction of building surrogate models from detailed physical counterparts. Training a GP modeling, however, presents a highly complex computational pattern. This requires the maximization of a general likelihood function with respect to the kernel hyperparameters. In addition, the kernel matrix appears in inverse form so even a single function evaluation has cubic complexity with the number of training points. We discuss the technical difficulties arising from this complex pattern and present some strategies available to date to deal with them.Read More...