Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (7 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2017)

This dissertation presents techniques for the numerical modeling and control of building systems, with an emphasis on thermostatically controlled loads. The primary objective of this work is to address technical challenges related to the management of energy use in commercial and residential buildings. This work is motivated by the need to enhance the performance of building systems and by the potential for aggregated loads to perform load following and regulation ancillary services, thereby enabling the further adoption of intermittent renewable energy generation technologies. To increase the generalizability of the techniques, an emphasis is placed on recursive and adaptive methods which minimize the need for customization to specific buildings and applications.

The techniques presented in this dissertation can be divided into two general categories: modeling and control. Modeling techniques encompass the processing of data streams from sensors and the training of numerical models. These models enable us to predict the energy use of a building and of sub-systems, such as a heating, ventilation, and air conditioning (HVAC) unit. Specifically, we first present an ensemble learning method for the short-term forecasting of total electricity demand in buildings. As the deployment of intermittent renewable energy resources continues to rise, the generation of accurate building-level electricity demand forecasts will be valuable to both grid operators and building energy management systems. Second, we present a recursive parameter estimation technique for identifying a thermostatically controlled load (TCL) model that is non-linear in the parameters. For TCLs to perform demand response services in real-time markets, online methods for parameter estimation are needed. Third, we develop a piecewise linear thermal model of a residential building and train the model using data collected from a custom-built thermostat. This model is capable of approximating unmodeled dynamics within a building by learning from sensor data.

Control techniques encompass the application of optimal control theory, model predictive control, and convex distributed optimization to TCLs. First, we present the alternative control trajectory (ACT) representation, a novel method for the approximate optimization of non-convex discrete systems. This approach enables the optimal control of a population of non-convex agents using distributed convex optimization techniques. Second, we present a distributed convex optimization algorithm for the control of a TCL population. Experimental results demonstrate the application of this algorithm to the problem of renewable energy generation following.

This dissertation contributes to the development of intelligent energy management systems for buildings by presenting a suite of novel and adaptable modeling and control techniques. Applications focus on optimizing the performance of building operations and on facilitating the integration of renewable energy resources.

    Cover page: Building Energy Modeling and Control Methods for Optimization and Renewables Integration
    • Article
    Working Papers (2015)
    The short-term forecasting of building electricity demand is certain to play a vital role in the future power grid. Given the deployment of intermittent renewable energy sources and the ever increasing consumption of electricity, the generation of accurate demand-side electricity forecasts will be valuable to both grid operators and building energy management systems. The literature is rich with forecasting models for individual buildings. However, an ongoing challenge is the development of a broadly applicable method for electricity forecasting across geographic locations, seasons, and use-types. This paper addresses the need for a generalizable approach to electricity demand forecasting through the formulation of a stacking ensemble learning method. Rather than using a single model to predict electricity demand, our method uses a weighted linear combination of forecasts from multiple sub-models. By learning the model weights in real-time using electricity demand data streams and a moving horizon training technique, the method is more robust than a single model approach. By applying our method to electricity demand data sets for 8 different buildings, we show that this data-driven approach is capable of producing accurate multivariate forecasts for building level applications.
      Cover page: Building Electricity Load Forecasting via Stacking Ensemble Learning Method with Moving Horizon Optimization
      • Article
      Working Papers (2015)
      For thermostatically controlled loads (TCLs) to perform demand response services in real-time markets, recursive methods for parameter estimation are needed.  As the physical characteristics of a TCL change (e.g. the contents of a refrigerator or the occupancy of a conditioned room), it is necessary to update the parameters of the TCL model. Otherwise, the TCL will be incapable of accurately predicting its potential energy demand, thereby decreasing the reliability of a TCL aggregation to perform demand response. In this paper, we investigate the potential of an unscented Kalman filter (UKF) algorithm to identify a TCL model that is non-linear in the parameters. Experimental results demonstrate the parameter estimation of two residential refrigerators.
        Cover page: Recursive Parameter Estimation of Thermostatically Controlled Loads via Unscented Kalman Filter
        • Article
        Working Papers (2015)
        A fundamental requirement of the electric power system is to maintain a continuous and instantaneous balance between generation and load. The intermittency and uncertainty introduced by renewable energy generation requires the expansion of ancillary power system services to maintain such a balance. In this paper, we examine the potential of thermostatically controlled loads (TCLs), such as refrigerators and electric water heaters, to provide generation following services in real-time energy markets (1 to 5 minutes). To simulate the non-linear dynamics of hysteretic dead-band systems in a manner suitable for convex optimization, we introduce an alternative control trajectory representation of the TCLs and their discrete input signals. To perform distributed optimization across large populations of TCLs, we propose a variation of the alternating direction method of multipliers (ADMM) algorithm. Based on our simulation results, we have demonstrated the potential for controlling a population of TCLs within an error tolerance of 10 kW.
          Cover page: Generation Following with Thermostatically Controlled Loads via Alternating Direction Method of Multipliers Sharing Algorithm
          • Article
          Working Papers (2016)
          Energy systems (e.g. ventilation fans, refrigerators, and electrical vehicle chargers) often have binary or discrete states due to hardware limitations and efficiency characteristics. Typically, such systems have additional programmatic constraints, such as minimum dwell times to prevent short cycling. As a result, non-convex techniques, like dynamic programming, are generally required for optimization. Recognizing developments in the field of distributed convex optimization and the potential for energy systems to participate in ancillary power system services, it is advantageous to develop convex techniques for the approximate optimization of energy systems. In this manuscript, we develop the alternative control trajectory representation -- a novel approach for representing the control of a non-convex discrete system as a convex program. The resulting convex program provides a solution that can be interpreted stochastically for implementation.
            Cover page: Alternative Control Trajectory Representation for the Approximate Convex Optimization of Non-Convex Discrete Energy Systems
            • Article
            Working Papers (2016)
            Model predictive control (MPC) strategies show great potential for improving the performance and energy efficiency of building heating, ventilation, and air-conditioning (HVAC) systems. A challenge in the deployment of such predictive thermostatic control systems is the need to learn accurate models for the thermal characteristics of individual buildings. This necessitates the development of online and data-driven methods for system identification. In this paper, we propose a piecewise linear thermal model of a building. To learn the model, we present a Kalman filter based approach for estimating the parameters. Finally, we fit the piecewise linear model to data collected from a residential building with a forced-air heating and ventilation system and validate the accuracy of the trained model.
              Cover page: Piecewise Linear Thermal Model and Recursive Parameter Estimation of a Residential Heating System
              • Article
              Working Papers (2014)
              Typical residential HVAC systems employ mechanical or hard-coded deadband control behaviors that are unresponsive to changing energy costs and weather conditions. In this paper, we investigate the potential of electric baseboard heaters to maintain a comfortable temperature while optimizing electricity consumption given weather forecasts and price data. We first propose a distributed system architecture that utilizes mobile application platforms. We then develop, assemble, and deploy a sensor network and Internet server to collect real-time temperature data from an apartment. With these sensor streams, we identify a thermal model of the apartment. Finally, we propose a model predictive control algorithm and perform a software-in-the-loop simulation of the cloud-based system to demonstrate the economic advantage.
                Cover page: Model Predictive Control of Residential Baseboard Heaters with Distributed System Architecture
                Top