Flexible operation and advanced control for energy systems
Monday 7th January 2019 to Friday 11th January 2019
10:30 to 11:20  Registration and morning coffee  
11:20 to 11:30  Welcome from David Abrahams (Isaac Newton Institute)  
11:30  AM  Problems vs Methods  
11:30 to 12:30 
Warren Powell (Princeton University) A Unified Framework for Stochastic Optimization in Energy Session: Problems vs Methods
A Unified Framework for Stochastic Optimization in Energy<br>
Warren B. Powell<br>
Dept. of Operations Research and Financial Engineering<br>
Princeton University<br>
<br>
Energy systems offer a variety of forms of uncertainty that have to be accommodated to ensure a reliable source of power. The modeling of these sequential decision problems under uncertainty has lacked the kind of canonical framework long enjoyed by deterministic problems. I will introduce a modeling framework that is completely general, which involves three mathematical challenges: 1) machine learning (there are up to five classes of learning problems), 2) uncertainty modeling, and 3) designing policies, which are functions for making decisions. There are two fundamental strategies for creating policies, each of which further divides into two subclasses, creating four classes of policies. These four (meta)classes of policies are universal, in that any method used to solve a sequential decision problem will be drawn from this set. The four classes are illustrated in the context of several applications in energy systems. An energy storage application is then used to demonstrate that each of the four classes of policies might be best depending on the characteristics of the data. <br>
<br>
<br>

INI 1  
12:30 to 13:30  Lunch at Westminster College  
13:30  PM  Industrial Needs Expression  
13:30 to 14:30 
Andrew Haslett (Energy Technologies Institute) Creating user value through mathematics – the challenge of hidden states Session: Industrial Needs Expression
Energy systems are large complex, and expensive, often significantly regulated and delivered through policy interventions. They are paid for through user charges, supported by taxation. Moving away from fossil fuels creates new systems challenges to meet user needs. Understanding these needs is an important element in meeting them, as the supply paradigm changes.Energy use is incidental to daily life – visiting friends, going to work, eating, keeping clean and comfortable etc are the things that matter to users. Control has a critical part to play in managing large loads such as vehicle charging or space heating. How do we understand the user intent behind the energy use and represent it in our mathematics?There are two kinds of hidden states in energy use systems – physical and psychological states. Control systems have direct impact on physical states, and psychological states through the user experience. This talk considers the functional needs of the user, their need for agency and the transaction cost from their perspective of engaging with the control systems for their house or car. It does not consider consumer segmentations on functional, affective and symbolic dimensions. Although in principle, hidden physical states can be discovered by measurement, often this is not practical.Control of heating systems and vehicle charging should aim to deliver satisfied users at lowest system cost. Although the talk will not cover in detail how to estimate current supply side state and forecast its coevolution with aggregate demand over the control period, it will consider in general terms how this might interact with user needs and willingness to pay.The talk will not cover mathematical solutions to the challenges. It is intended to illuminate the realworld problems that will require advanced mathematics to deliver effective solutions.

INI 1  
14:30 to 15:30 
Patrick Panciatici (R)evolution of large electrical systems: Needs and Challenges Session: Industrial Needs Expression
Historical power systems are emblematic examples of system of systems. Now, we are living a major evolution and perhaps a revolution in electrical grids. The term “smart grid” is used everywhere without a very precise definition. The concept of cyberphysical System of Systems seems a good framework to capture the essence of this (r)evolution.<br>
The “cyber” layer is going to play a key role in the system reliability. Indeed, more and more controls are embedded in subsystems which become “intelligent” and partially autonomous. The system behavior will be imposed by the interactions between these “intelligent” agents driven by local pieces of software rather than by physical laws. <br>
Different examples showing this trend will be presented from synchronization of inverters to aggregated game for demand response management.<br>
We must pay attention; possible negative impacts could occur when a certain critical level of penetration of new devices and processes will be reached. It is not easy, we must review our historical approach in order to specify behaviors which were yesterday imposed by physical laws and didn’t need any specification and which will be tomorrow defined by software in local controls.<br>

INI 1  
15:30 to 16:00  Afternoon Tea  
16:00 to 17:00 
Hungpo Chao (PJM) Electricity market reform to enhance the energy market pricing mechanism: observations from PJM Session: Industrial Needs Expression Electricity market reform to enhance the energy and reserve pricing mechanism: Observations from PJM<br> <br> by Hungpo Chao<br> <br> Abstract<br> <br> For more than 20 years, the PJM wholesale markets have successfully worked to promote competition, produce stable energy prices and attract competitive resource investments to ensure efficient and reliable operations. However, in recent years, the PJM markets have been undergoing a significant transition. While such transitions have also occurred elsewhere, each has resulted in some unique challenges. <br> This paper examines issues regarding efficient price formation in the energy and reserve markets under nonconvex. In principle, with nonconvexity, no market clearing prices exist without side payments. In a poolbased wholesale electricity market, one of the greatest challenges unmatched in scale and complexity is that in the dayahead and realtime markets, after running a mixed integer programming model for solving a security constrained economic commitment and dispatch problem to determine the market allocations, a pricing model is employed to determine the market clearing prices and side payments in ways that must promote economic efficiency, consistent incentives and revenue sufficiency. <br> <br> One of the most severe limitations of the current pricing mechanism (locational marginal pricing or LMP) is that LMP is not incentive compatible. This limitation has caused adverse effects in operations and investments. Building on the classic Lagrangian dual formulation, this paper extends the existing pricing method in a way that is dominant strategy incentive compatible in a competitive market with a large number of independent suppliers, and like a VickeryClarkGrove mechanism, truthful revelation would become a dominant strategy. The convex hull pricing method or called the extended LMP, is a wellknown case which yields the minimum uplift. Moreover, integer relaxation is a computationally practical implementation that ensures incentive compatibility producing generally good, and often exact, approximations to ELMP solutions if the cost functions are homogeneous of degree one. <br> <br> As market continues to evolve with flattening demand growth, flattening supply curves with low marginal costs and penetration of renewable resources with zero marginal cost, nonconvex conditions will become growingly important. A key advantage of enhanced pricing mechanism is that it would form price signals in ways that would foster economic efficiency in operations and investments, demand participation and market innovation. <br> 
INI 1  
17:00 to 18:00  Welcome Wine Reception at INI 
09:00  AM  Methods vs Problems  
09:00 to 10:00 
Ben Godfrey (None / Other) Market design for flexible distribution networks Session: Methods vs Problems 
INI 1  
10:00 to 11:00 
Gilles Louppe (Université de Liège) Automated parameter inference and data modelling with deep learning Session: Methods vs Problems 
INI 1  
11:00 to 11:30  Morning Coffee  
11:30 to 12:30 
Javad Lavaei (University of California, Berkeley) Computational Methods for Nonlinear Power Operational Problems: Convex Reformulations and NearLinear Time Algorithms Session: Methods vs Problems
Coauthors: Somayeh Sojoudi (UC Berkeley), Richard Zhang (UC Berkeley), Salar Fattahi (UC Berkeley), Igor Molybog (UC Berkeley), Ming Jin (UC Berkeley), SangWoo Park (UC Berkeley)In this talk, we will study a set of nonlinear power optimization and decisionmaking problems, namely power flow, optimal power flow, state estimation and topology error detection. We will propose different conic relaxation and approximation techniques to solve these nonconvex problems. We will prove that such conic problems could be solved in near linear time due to intrinsic properties of realworld power networks. We will offer case studies on systems with as high as 14,000 nodes.

INI 1  
12:30 to 13:30  Lunch at Westminster College  
13:30  PM  Demand Dispatch & Microgrids  
13:30 to 14:30 
Juan Miguel Morales González (Universidad de Málaga) Electricity demand forecasting and bidding via datadriven inverse optimization Session: Demand Dispatch & Microgrids
A method to predict the aggregate demand of a cluster of priceresponsive consumers of electricity is discussed in this presentation. The priceresponse of the aggregation is modelled by an optimization problem whose defining parameters represent a series of marginal utility curves, and minimum and maximum consumption limits. These parameters are, in turn, estimated from observational data using an approach inspired from duality theory. The resulting estimation problem is nonconvex, which makes it very hard to solve. In order to obtain good parameter estimates in a reasonable amount of time, we divide the estimation problem into a feasibility problem and an optimality problem. Furthermore, the feasibility problem includes a penalty term that is statistically adjusted by cross validation. The proposed methodology is datadriven and leverages information from regressors, such as time and weather variables, to account for changes in the parameter estimates. The estimated priceresponse model is used to forecast the power load of a group of heating, ventilation and air conditioning systems, with positive results. We also show how this method can be easily extended to be used for demandside bidding in electricity markets.

INI 1  
14:30 to 15:30 
Anna Scaglione (Arizona State University) On modeling dispatchable loads in grid operation: the good, the bad and the ugly Session: Demand Dispatch & Microgrids
The past ten years of research had produced a variety of models for managing flexible loads, paving the way for addressing congestion in the grid by enabling a more efficient dynamic pricing of electricity. However, real change has been hard to come by in practice. The goal of this talk is to review such models, highlighting the difference between distributed algorithms, that seek to decompose the problem, and aggregate representations that map large populations of flexible loads onto spinning reserves. The objective is to highlight the challenges that exist in transforming and remaining compatible with established retail and wholesale market practices and legacy systems and how new abstractions may be necessary to rip the benefits of flexible load.

INI 1  
15:30 to 16:00  Afternoon Tea  
16:00 to 17:00 
JeanYves Le Boudec (EPFL  Ecole Polytechnique Fédérale de Lausanne) Realtime operation of microgrids Session: Demand Dispatch & Microgrids Coauthors: Paolone, Marione (EPFL), Reyes, Lorenzo (EPFL), Bernstein, Andrey (NREL), Wang, Cong (EPFL)<br> <br> Very large amounts of renewable electricity generation, combined with a large penetration of plugin electric vehicles, may cause considerable stress to electrical grids and to the system of spinning reserves. This problem can be solved by controlling the huge number of electrical resources that are located in distribution grids, such as thermal loads, stationary batteries, charging stations and curtailable power generators. However, this poses a number of new challenges in terms of online computation and scalability. In this talk we discuss how these challenges are solved by COMMELEC, a system of realtime software agents developed at EPFL and deployed in several grids. We also introduce elements of a theory about how uncertainty on power injections affects the controllability of the grid.<br> <br> Slides available at http://icawww1.epfl.ch/PS_files/MESW01LEB2019.pdf 
INI 1  
17:00 to 18:00  Poster session 
09:00  AM  Markets & System Operation  
09:00 to 10:00 
Rene Aid (Université ParisDauphine) Optimal Electricity Demand Response Contracting Session: Markets & System Operation
Coauthors: Dylan Possamaï (Columbia University), Nizar Touzi (Ecole Polytechnique)We address the moral hazard underlying demand response contracts in electricity markets, we formulate the interaction problem between producer and the consumer by means of a PrincipalAgent problem. The producer, acting as the Principal, is subject to the generation costs related to the level and the volatility of generation, thus accounting for the limited flexibility of production. Based on the continuoustime consumption of the Agent, representing a single consumer, she sends an incentive compensation in order to encourage him to reduce his average consumption and to improve his responsiveness defined as the volatility of his consumption. We provide closedform expression for the optimal contract that maximizes the utility of the principal in the case of linear energy valuation. We provide rationality for the form of the observed demandresponse contracts, that is a fixed premium for enrolment and a proportional price for the energy consumed. However, we show that the pre mium should be an increasing function of the duration of the demand response event. Further, we show that optimal contracting allows the system to bear more risk as the resulting consumption volatility may increase, but the corresponding risk is now optimally shared between the two actors. We calibrate of our model to publicly available data of the London demandresponse trial, and we infer that a significant increase of responsiveness can be expected by the implementation of the control of the consumption volatility. We find that the responsiveness control would lead a significant increase of the value of the producer. We examine the stability of our explicit optimal contract by performing appropriate sensitivity analysis, and show that the linear approximation of the energy value function of the consumer provides a robust approximation of the optimal contract.

INI 1  
10:00 to 11:00 
Lang Tong (Cornell University); (Chalmers University of Technology) Towards seemless operation: a new look at interface scheduling and market operation Session: Markets & System Operation 
INI 1  
11:00 to 11:30  Morning Coffee  
11:30 to 12:30 
Anthony Papavasiliou (Université Catholique de Louvain) Transmission capacity allocation in zonal electricity markets Session: Markets & System Operation We propose a novel framework for modelling zonal electricity markets, based on projecting the constraints of the nodal network onto the space of the zonal aggregation of the network. The framework avoids circular definitions and discretionary parameters, which are recurrent in the implementation and study of zonal markets. Using this framework, we model and analyze two zonal market designs currently present in Europe: flowbased market coupling (FBMC) and availabletransfercapacity market coupling (ATCMC). We develop cuttingplane algorithms for simulating FBMC and ATCMC while accounting for robustness of imports/exports to single element failures, and we conduct numerical simulations of FBMC and ATCMC for a realistic instance of the Central Western European system under 768,000 different operating conditions. We find that FBMC and ATCMC are unable to anticipate congestion of branches interconnecting zones and branches within zones, and that both zonal designs achieve similar overall cost efficiencies 0.5% difference in favour of FBMC), while a nodal market design largely outperforms both of them (5.9% better than FBMC). These findings raise the question of whether it is worth for more European countries to switch from ATCMC to FBMC, instead of advancing directly towards a nodal design. 
INI 1  
12:30 to 13:30  Lunch at Westminster College  
13:30 to 18:00  Free afternoon  
19:30 to 22:00  Formal Dinner at Christs College 
09:00  AM  Optimization & Control  
09:00 to 10:00 
Florian Doerfler (ETH Zürich) Realtime feedback optimization on the power flow manifold Session: AM  Optimization & Control I will focus on online optimization of AC power systems in closed loop. In contrast to the conventional approach where an optimal power flow solution is computed offline and online controllers enforce these setpoints, our objective is to design an adaptive feedback controller that steers the system robustly and in real time to the optimal operating point. Our methodological approach is based on online algorithms for manifold optimization that can be applied in feedback with realtime measurements and actuation. We treat the power flow equations as implicit constraints that are naturally enforced by the physics and hence give rise to the power flow manifold. Based on our theoretical results for this type of optimization problems, we propose a projected gradient descent scheme on the power flow manifold. In detailed simulation case studies we validate the performance of our algorithm and show that it reliably tracks the timevarying optimum of the underlying AC optimal power flow problem.<br> <br> Coauthors: Adrian Hauswirth (ETH Zurich), Saverio Bolognani (ETH Zurich), Gabriela Hug (ETH Zurich) 
INI 1  
10:00 to 11:00 
Andrea Simonetto (IBM Research) TimeVarying Optimization: Algorithms and Applications in Power Systems Session: AM  Optimization & Control
Continuously varying optimization programs have appeared as a natural extension of timeinvariant ones when the cost function, the constraints, or both, depend on a time parameter and change continuously in time. This setting captures relevant control, signal processing, and machine learning problems. Recently, running and predictioncorrection methods have been put forward to set up iterative algorithms that sample the continuouslyvarying optimization program at discrete time steps track the optimizer(s) trajectory while it evolves in time up to an asymptotical error bound.In this talk, we will review current stateoftheart algorithms in timevarying optimization, with a special emphasis on applications in power grids. We will touch upon timevarying AC optimal power flow problems, realtime optimization of aggregations of distributed energy resources, as well as dynamic distribution state estimation.

INI 1  
11:00 to 11:30  Morning Coffee  
11:30 to 12:30 
Emiliano Dall'anese (University of Colorado) Feedbackbased online algorithms for timevarying optimization: theory and applications in power systems Session: AM  Optimization & Control
The talk focuses on the synthesis and analysis of online algorithmic solutions to control systems or networked systems based on performance objectives and engineering constraints that may evolve over time. Particular emphasis is given to applications in power systems operations and control. The timevarying optimization formalism is leveraged to model optimal operational trajectories of the systems, as well as explicit local and networklevel constraints. The design of the algorithms then capitalizes on an online implementation of primaldual projectedgradient methods; the gradient steps are, however, suitably modified to accommodate actionable feedback in the form of measurements from the network  hence, the term feedbackbased online optimization. By virtue of this approach, the resultant running algorithms can cope with model mismatches in the algebraic representation of the system states and outputs, they avoid pervasive measurements of exogenous inputs, and they naturally lend themselves to a distributed implementation. Under suitable assumptions, Qlinear convergence to optimal solutions of a timevarying convex problem is shown. On the other hand, under a generalization of the MangasarianFromovitz constraint qualification, sufficient conditions are derived for the running algorithm to track a KarushKuhnTucker point of a timevarying nonconvex problem. Examples of applications in power systems will be provided. <br>
<br>
Joint work with: A. Simonetto, Y. Tang, A. Bernstein, and S. Low.

INI 1  
12:30 to 13:30  Lunch at Westminster College  
13:30 to 14:30 
Andy Sun (Georgia Institute of Technology) A new distributed algorithm for nonconvex network flow problems with convergence guarantees Session: PM  Optimization & Control 
INI 1  
13:30  PM  Optimization & Control  
14:30 to 15:30 
Pär Holmberg (None / Other) Central versus SelfDispatch in Electricity Markets Session: PM  Optimization & Control In centralized markets, producers submit detailed cost data to the dayahead market, and the market operator decides how much should be produced in each plant. This differs from decentralized markets that rely on selfcommitment and where producers send less detailed cost information to the operator of the dayahead market. Ideally centralized electricity markets would be more effective, as they consider more detailed information, such as startup costs and noload costs. On the other hand, the bidding format is rather simplified and does not allow producers to express all details in their costs. Moreover, due to uplift payments, producers have incentives to exaggerate their costs. As of today, US has centralized wholesale electricity markets, while most of Europe has decentralized wholesale electricity markets. The main problem with centralized markets in US is that they do not provide intraday prices which can be used to continuously update the dispatch when the forecast for renewable output changes. Intraday markets are more flexible and better adapted to deal with renewable power in decentralized markets. Iterative intraday trading in a decentralized market can also be used to sort out coordination problems related to nonconvexities in the production. The downside of this is that increased possibilities to coordinate increase the risk of getting collusive outcomes. Decentralized dayahead markets in Europe can mainly be improved by considering network constraints in more detail. 
INI 1  
15:30 to 16:00  Afternoon Tea  
16:00 to 17:00 
Sean Meyn (University of Florida) Irrational Agents and the Power Grid Session: PM  Optimization & Control
For decades power systems academics have proclaimed the need for real time prices to create a more efficient grid. The rationale is economics 101: proper price signals will lead to an efficient outcome. In this talk we will review a bit of economics 101; in particular, the definition of efficiency. We will see that the theory supports the realtime price paradigm, provided we impose a particular model of rationality. It is argued however that this standard model of consumer utility does not match reality: the products of interest to the various "agents" are complex functions of time. The product of interest to a typical consumer is only loosely related to electric power  the quantity associated with price signals. There is good news: an efficient outcome is easy to describe, and we have the control technology to achieve it. We need supporting market designs that respect dynamics and the impact of fixed costs that are inherent in power systems engineering, recognizing that we need incentives on many timescales. Most likely the needed economic theory will be based on an emerging theory of efficient and robust contract design.

INI 1  
17:00 to 18:00 
Orcun Karaca (ETH Zürich) CoreSelecting Mechanisms in Electricity Markets Session: PM  Optimization & Control
Previous work on electricity market auctions considers the payasbid and the locational marginal pricing (LMP) mechanisms. In both mechanisms, generators can bid strategically to influence their profits since these mechanisms do not incentivize truthful bidding. As an alternative, under the VickreyClarkeGroves mechanism, truthful bidding is the dominantstrategy Nash equilibrium. Despite having this theoretical virtue, coalitions of participants can influence the auction outcome to obtain higher collective profit. In this talk, we characterize the exact class of coalitionproof mechanisms as the coreselecting mechanisms. In addition to being coalitionproof, we show that these mechanisms generalize the economic rationale of the LMP mechanism. Namely, these mechanisms are the exact class of mechanisms that ensure the existence of a competitive equilibrium in linear/nonlinear prices. This implies that the LMP mechanism is also coreselecting, and hence coalitionproof. In contrast to the LMP mechanism, coreselecting mechanisms exist for a broad class of electricity markets, such as ones involving nonconvex costs and nonconvex constraint sets. In addition, they can approximate truthfulness without the pricetaking assumption of the LMP mechanism. Finally, we show that they are also budgetbalanced. Our results are verified with case studies based on optimal power flow test systems and the Swiss reserve market.

INI 1 
09:00  AM  Storage & Data analytics  
09:00 to 10:00 
Clemence Alasseur (EDF, France) A meanfield game model for management of distributed storages for the power system Session: Storage & Data analytics
We consider a stylized model for a power network with distributed local power generation and storage. This system is modeled as a network connection of a large number of nodes, where each node is characterized by a local electricity consumption, has a local electricity production (e.g. photovoltaic panels), and manages a local storage device. Depending on its instantaneous consumption and production rate as well as its storage management decision, each node may either buy or sell electricity, impacting the electricity spot price. The objective at each node is to minimize energy and storage costs by optimally controlling the storage device. In a noncooperative game setting, we are led to the analysis of a nonzero sum stochastic game with N players where the interaction takes place through the spot price mechanism. For an infinite number of agents, our model corresponds to an Extended MeanField Game (EMFG). We are able to compare this solution to the optimal strategy of a central planner and in a linear quadratic setting, we obtain and explicit solution to the EMFG and we show that it provides an approximate Nashequilibrium for Nplayer game.

INI 1  
10:00 to 11:00 
Simon Tindemans (Delft University of Technology) Optimal dispatch of heterogeneous batteries to maximise security of supply Session: Storage & Data analytics
Coauthors: Michael Evans (Imperial College London), David Angeli (Imperial College London). We consider the problem of dispatching a fleet of heterogeneous batteries (i.e. energyconstrained generators) to prevent or minimise power shortage scenarios. In the general case on which nothing is known about future power requirements, three significant results are derived. First, a greedy policy exists that uniquely maximises the time until the fleet is first unable to supply demand. This policy implicitly establishes a `feasible set’ of power requests that can be satisfied by the fleet. Second, an analytical transformation is presented that expresses this feasible set in a graphical form instead of a procedural form (i.e. by invoking the policy). The graphical representation also provides a measure of the flexibility penalty due to heterogeneity. Third, it is shown that the policy can be extended to handle scenarios with unavoidable power shortages, in which case it minimises the energy not supplied. The fact that the greedy policy results in bestcase securityof supply performance suggests it is suitable to be used as a reference policy for battery dispatch within system adequacy studies. We present a discrete time algorithm that is tailored for this use case and show results for a Great Britain case study.

INI 1  
11:00 to 11:30  Morning Coffee  
11:30 to 12:30 
Meng Wang (Rensselaer Polytechnic Institute) Highdimensional data analytics using lowdimensional models in power systems Session: Storage & Data analytics
Phasor Measurement Units and smart meters provide finegrained measurements to enhance the system visibility to the operators and reduce blackouts. The recent wealth of data is revolutionizing the conventional modelbased power system monitoring and control to a modern datadriven counterpart. One recent research interest is to develop computationally efficient datadriven methods to convert data into information.<br>
<br>
This first part of the talk discusses our proposed missing data recovery and error correction methods for synchrophasor data. The low data quality currently prevents the implementation of synchrophasordatabased realtime monitoring and control. This second half of the talk discusses our proposed privacypreserving data collection framework for smart meters. We developed load pattern extraction methods from highly noisy and quantized smart meter data such that the estimated load pattern is only accurate for the operator, and the information is obfuscated to a cyber intruder with partial measurements. The common theme of the two projects is to exploit the intrinsic lowdimensional structures in the data to develop fast algorithms for nonconvex problems with analytical performance guarantees.

INI 1  
12:30 to 13:30  Lunch at Westminster College  
13:30  PM  Broadening the Perspective  
13:30 to 14:30 
Pierre Gaillard (INRIA Paris  Rocquencourt); (ENS  Paris) Bandit algorithms for power consumption control Session: Broadening the Perspective
We are interesting in optimizing price signals sent by an electricity provider to its customers so has to modify their electricity consumption. We formulate this problem as a sequential problem in which the electricity provider send signals and sequentially observes corresponding feedback. The mathematical theory of bandits will be adapted to this exploration  exploitation problem.<br>
<br>
This is a joint work with Margaux Brégère, Gilles Stoltz and Yannig Goude

INI 1  
14:30 to 15:30 
John Moriarty (Queen Mary University of London); Ana Busic (INRIA); Steven Low (CALTECH (California Institute of Technology)); Louis Wehenkel (Université de Liège); Pierre Pinson (Danmarks Tekniske Universitet) Wrapup discussion session Session: Broadening the Perspective 
INI 1 