Achin Jain

Data-driven modeling, control and tools for cyber-physical energy systems

Demand response (DR) is becoming important as the volatility on the grid continues to increase. Current DR approaches are either completely manual or involve deriving first principles based models which are extremely cost and time prohibitive to build. We consider the problem of data-driven DR for large buildings which involves predicting the demand response baseline, evaluating fixed DR strategies and synthesizing DR control actions. We provide a model based control with regression trees algorithm (mbCRT), which allows us to perform closed-loop control for DR strategy synthesis for large buildings. Our data-driven control synthesis algorithm outperforms rule- based DR by 17% for a large DoE commercial reference building and leads to a curtailment of 380 kW and over

Data Predictive Control for building energy management

45,000 in savings. Our methods have been integrated into an open source tool called DR-Advisor, which acts as a recommender system for the buildings facilities manager and provides suitable control actions to meet the desired load curtailment while maintaining operations and maximizing the economic reward. DR-Advisor achieves 92.8% to 98.9% prediction accuracy for 8 buildings on Penns campus. We compare DR-Advisor with other data driven methods and rank 2nd on ASHRAEs benchmarking data-set for energy prediction.

Data Predictive Control for Peak Power Reduction

Decisions on how to best optimize energy systems operations are becoming ever so complex and conflicting, that model-based predictive control (MPC) algorithms must play an important role. However, a key factor prohibiting the widespread adoption of MPC in buildings, is the cost, time, and effort associated with learning first-principles based dynamical models of the underlying physical system. This paper introduces an alternative approach for implementing finite-time receding horizon control using control-oriented data-driven models. We call this approach Data Predictive Control (DPC). Specifically, by utilizing separation of variables, two novel algorithms for implementing DPC using a single regression tree and with regression trees ensembles (random forest) are presented. The data predictive controller enables the building operator to trade off energy consumption against thermal comfort without having to learn white/grey box models of the systems dynamics. We present a comprehensive numerical study which compares the performance of DPC with an MPC based energy management strategy, using a single zone building model. Our results demonstrate that performance of DPC is comparable to an MPC controller, with only 3.8% additional cost in terms of optimal objective function and within 95% in terms of R2 score, thereby making it an alluring alternative to MPC, whenever the associated cost of learning the model is high.

Learning and Control Using Gaussian Processes

Decisions on how best to optimize todays energy systems operations are becoming ever so complex and conflicting such that model-based predictive control algorithms must play a key role. However, learning dynamical models of energy consuming systems such as buildings, using grey/white box approaches is very cost and time prohibitive due to its complexity. This paper presents data-driven methods for making control-oriented model for peak power reduction in buildings. Specifically, a data predictive control with regression trees (DPCRT) algorithm, is presented. DPCRT is a finite receding horizon method, using which the building operator can optimally trade off peak power reduction against thermal comfort without having to learn white/grey box models of the systems dynamics. We evaluate the performance of our method using a DoE commercial reference virtual test-bed and show how it can be used for learning predictive models with 90% accuracy, and for achieving 8.6% reduction in peak power and costs.

Data-Driven Model Predictive Control with Regression Trees—An Application to Building Energy Management

Building physics-based models of complex physical systems like buildings and chemical plants is extremely cost and time prohibitive for applications such as real-time optimal control, production planning and supply chain logistics. Machine learning algorithms can reduce this cost and time complexity, and are, consequently, more scalable for large-scale physical systems. However, there are many practical challenges that must be addressed before employing machine learning for closed-loop control. This paper proposes the use of Gaussian Processes (GP) for learning control-oriented models: (1) We develop methods for the optimal experiment design (OED) of functional tests to learn models of a physical system, subject to stringent operational constraints and limited availability of the system. Using a Bayesian approach with GP, our methods seek to select the most informative data for optimally updating an existing model. (2) We also show that black-box GP models can be used for receding horizon optimal control with probabilistic guarantees on constraint satisfaction through chance constraints. (3) We further propose an online method for continuously improving the GP model in closed-loop with a real-time controller. Our methods are demonstrated and validated in a case study of building energy control and Demand Response.

Data-driven Switched Affine Modeling for Model Predictive Control

Model Predictive Control (MPC) plays an important role in optimizing operations of complex cyber-physical systems because of its ability to forecast system’s behavior and act under system level constraints. However, MPC requires reasonably accurate underlying models of the system. In many applications, such as building control for energy management, Demand Response, or peak power reduction, obtaining a high-fidelity physics-based model is cost and time prohibitive, thus limiting the widespread adoption of MPC. To this end, we propose a data-driven control algorithm for MPC that relies only on the historical data. We use multi-output regression trees to represent the system’s dynamics over multiple future time steps and formulate a finite receding horizon control problem that can be solved in real-time in closed-loop with the physical plant. We apply this algorithm to peak power reduction in buildings to optimally trade-off peak power reduction against thermal comfort without having to learn white/grey box models of the systems dynamics.

Data Predictive Control for Building Energy Management: Poster Abstract

Abstract Model Predictive Control (MPC) is a well consolidated technique to design optimal control strategies, leveraging the capability of a mathematical model to predict the system’s behavior over a predictive horizon. However, building physics-based models for large scale systems, such as buildings and process control, can be cost and time prohibitive. To overcome this problem we propose in this paper a methodology to exploit machine learning techniques (i.e. regression trees and random forests) in order to build a state-space switched affine dynamical model of a large scale system only using historical data. Finite Receding Horizon Control (RHC) setup using control-oriented data-driven models based on regression trees and random forests is presented as well. A comparison with an optimal MPC benchmark and a related methodology is provided on an energy management system to show the performance of the proposed modeling framework. Simulation results show that the proposed approach is very close to the optimum and provides better performance with respect to the related methodology in terms of cost function optimization.

On the design and tuning of linear model predictive control for wind turbines

Decisions on how best to optimize todays energy systems operations are becoming ever so complex and conflicting such that model-based predictive control algorithms must play a key role. However, learning dynamical models of energy consuming systems such as buildings, using grey/white box approaches is very cost and time prohibitive due to its complexity. This paper presents data-driven methods for making control-oriented model for peak power reduction in buildings. Specifically, a data predictive control with regression trees (DPCRT) algorithm, is presented. DPCRT is a finite receding horizon method, using which the building operator can optimally trade off peak power reduction against thermal comfort without having to learn white/grey box models of the systems dynamics. We evaluate the performance of our method using a DoE commercial reference virtual test-bed and show how it can be used for learning predictive models with 90% accuracy, and for achieving 8.6% reduction in peak power and costs.