# Workshop Programme

## for period 22 - 26 April 2013

### Energy Systems Week: Management of Variability and Uncertainty in Energy Systems

22 - 26 April 2013

Timetable

Monday 22 April | ||||

09:00-10:00 | Registration and Welcome | |||

09:00-10:00 | Registration and Welcome | |||

10:00-10:40 | Pinson, P (Technical University of Denmark) |
|||

Stochastic Power Generation From Renewables: Forecasting and Optimization Challenges for its Optimal Integration | Sem 1 | |||

Renewable energy sources and demand are directly influenced by the weather. In contrast to the electric load, whose variations at an aggregated level are quite smooth and fairly predictable based on key meteorological variables, the variability and (lack of) predictability of renewable power generation induce new challenges in power systems and market operations. This variability and lack of predictability obviously depend upon the type of renewable energy source, location, time of the year and prevailing weather conditions. They are of crucial importance at a wide range of temporal and spatial scales, e.g., local and short-term for control problems or region-wide and longer-term for planning and investment problems. In practice, the operational management of renewable energy in power systems and electricity markets requires dedicated forecasts of power generation, which may be of deterministic or probabilistic nature. They are used as input to decision-making problems which are more or less advanced depending upon the decisions to be made and the expertise of the decision-maker. In view of the uncertainties involved the forecasts should optimally be probabilistic, in the form of marginal predictive densities or even space-time scenarios, while decision-making problems should be solved in a stochastic optimization framework. In that respect maybe one of the key to successful decision-making lies at the interface between forecasting and optimization. Indeed, the more advanced stochastic optimization problems, supposed to yield better decisions in view of uncertainties, necessitate forecasts of ever-increasing complexities. These get quite difficult to generate and verify, potentially leading to less valuable decisions overall. Also, practical aspects might step in, preventing the application and operational implementation of the advanced approaches proposed by academics. The interface between forecasting and optimization will be discussed based on real-world test cases, but also toy models in power systems operations and electricity markets. |
||||

10:40-11:00 | Coffee and Tea | |||

11:00-11:45 | Pollitt, M (University of Cambridge) |
|||

Electricity Market Reform in the UK | Sem 1 | |||

This talk will discuss the recent proposed changes to electricity markets in the UK known as Electricity Market Reform (EMR). It will discuss how the four elements of the proposed reforms fit together, their economic rationale, practical implementation difficulties and the extent to which they will impact on the operation of the electricity market. |
||||

11:45-12:30 | Smeers, Y (Universiteit Louvain) |
|||

Trade-offs, Uncertainty, Coordination and Market Design in Restructuring for Decarbonisation | Sem 1 | |||

Economic activities are coordinated through prices or quantities. Coordination by quantities is conducted through optimization; equilibrium models are used to analyze coordination by prices. The market generally chooses the best instrument but this selection seems to have been particularly difficult so far in the restructuring of the electricity sector. This is the question of the market design that must separate activities driven by prices from those that require more centralization through quantities. While this process remains largely unfinished in the traditional system, it is posed in new terms by decarbonisation. The presentation addresses the question as follows. Starting from residual questions in today market design of the electricity sector so far, it examines the new problems on the basis of three recent documents with the view of uncovering economic coordination problems where methodological advances of mathematical nature (of the computable economic equilibrium type) are required. |
||||

12:30-14:00 | Lunch, Churchill College | |||

14:00-14:30 | Brossat, X (EDF, France) |
|||

A Range of Methods for Electricity Consumption Forecasting | Sem 1 | |||

For Electricité de France the forecast of electricity consumption is a fundamental problem which has been studied for the last twenty years. It is necessary to be able to provide customers and at the same time, optimize the production at different horizons of time. Results of operating models that use non linear regression or ARMAX methods are satisfying with a current accuracy of 1.5% for the forecast of the following day. But, they have to be continually fitted to be adapted to some very difficult periods of time and to the change of consumption. For a few years, due to the new competitive environment, the electrical load curve has become less regular. Its shape and level which depended essentially on climatic exogenous variables has become more affected by economical and ecological variables. The data is not always available and the time series used are often short. So, we have tried to apply the following alternative methods to answer to problems like adaptivity, nonstationarity, parsimony, lack of data, necessity of forecast interval. In this presentation we will display the operating models and those different classes of models which we applied to electrical consumption forecast. For each model we will present the method used, we will show some practical results and we will discuss the benefits and drawbacks of it.(adaptive Kalmann, GAM, combining algoritms, KWF, Bayesian Methods, ..) |
||||

14:45-15:15 | Goude, Y (EDF, France) |
|||

Semi-Parametric Models for Electricity Consumption Forecasting | Sem 1 | |||

Electricity load forecasting faces rising challenges due to the advent of innovating technologies such as smart grids, electric cars and renewable energy production. For utilities, a good knowledge of the future electricity consumption stands as a central point for the reliability of the network, investment strategies, energy trading, optimizing the production etc. Generalized Additive Models have been investigated recently to forecasts electricity consumptions at EDF R&D. These models achieve an interesting trade-off between accuracy of forecasts and adaptation to different data sets thanks to their semi-parametric structures. We apply GAM models on different data sets corresponding to different practical applications at EDF and show how these models can be used for real forecasts at different horizon and geographical scale. |
||||

15:30-15:45 | Tea and Coffee | |||

15:45-16:30 | Leboudec, J Y (EPFL, Lausanne) |
|||

Real-Time Storage and Demand Management | Sem 1 | |||

We review recent results on storage and demand management obtained using stochastic modelling. First, we study a model of storage used as a secondary reserve in an electrical grid. We analyze storage scheduling policies, in particular the fixed offset policy of Bejan et al.; we consider the impact of energy conversion efficiency and of the quality of renewable predictions. We find that there is a bound on the performance of any scheduling policy, which is tight for large enough storage capacity. We also find strategies that outperform the proposed fixed level policies. Second, we study the effect of energy storage on the efficiency of the real time electricity market. We show the existence of a competitive equilibrium; further, storage is beneficial in the sense that the resulting price process in presence of storage is smoother and concentrates on the marginal production cost. However, we find that consumers and stand-alone storage operators have an incentive to under-dimension their storage system. Third, we propose a system-wide model for real-time demand response. The model shows that evaporation, namely the fraction of delayed demand that eventually disappears, plays a central role. The model also suggests that large backlogs of delayed demand may accumulate and make predictions difficult. |
||||

16:45-18:45 | Wine Reception, Isaac Newton Institute of Mathematical Sciences |

Tuesday 23 April | ||||

09:30-10:00 | Edwards, G (Heriot-Watt University) |
|||

Forecasting Aggregated Wind Power Availability Over Extended Regions | Sem 1 | |||

The effective operation of a large power system with significant penetration of wind generation requires as much information as possible about the uncertainty in aggregated wind power output levels close to real-time. In a Bayesian framework, short-term joint posterior predictive distributions for wind speeds are desirable, allowing calculation of the aggregated power output distribution. Given the complex and dynamic spatiotemporal structures associated with wind resource availability, it is clear that the most accurate predictive distributions are obtained from adaptive multivariate statistical models involving the most timely information from across the system's extended area, and utilising a number of meteorological variables. As such, dynamic linear models (DLMs) are presented as a suitable and novel forecasting framework for wind power. This presentation will report on initial results of research into the best DLM model structures for producing such predictive distributions. This work is a collaboration between Heriot-Watt and Lancaster Universities, involving Dr Stan Zachary (Heriot-Watt), and Dr Idris Eckley and Dr Rebecca Killick (Lancaster). |
||||

10:15-10:45 | Plumptre, P (National Grid) |
|||

Wind Problems for the Electricity Planner and Operator | Sem 1 | |||

Paul Plumptre covers six problems, posed by the increasing proportion of Wind powered generation to the Operator and Planner of the GB power system. Paul has thirty years experience of such issues for National Grid |
||||

11:00-11:30 | Coffee and Tea | |||

11:30-12:00 | Nason, G (University of Bristol) |
|||

Modelling and Forecasting of Network Time Series | Sem 1 | |||

This talk concerns the modelling and forecasting of data acquired on the nodes of a network observed through time. Many data sets exist that are acquired directly on the nodes of a network, or there is often an implicit network that be identified or constructed. An important challenge for modelling network data is taking proper account of the high-dimensional inter-node distributional associations, as well as modelling these through time. We present a novel technique for dimension reduction using the network version of the recently devloped "lifting one coefficient at a time" transform which exhibits excellent decorrelation properties both in space and in time. We explain the method of dimension reduction and demonstrate its utility on data arising from epidemiology and wind energy. |
||||

12:00-12:30 | Eckley, I (University of Lancaster) |
|||

Coherence Analysis of Multivariate Time Series | Sem 1 | |||

Data collection systems are widely used within the energy sector to record process activity across energy generations sites. These loggers are capable of sampling data at high rates, at a number of locations and recording multiple process aspects at each location. Such series are typically non-stationary in nature, with potentially time-varying dependence between the various series components. In this talk we consider the problem of modelling and estimating the coherence structure within such time series. In particular we focus on the challenge of identifying whether the dependence between a pair of components is direct or indirectly driven by other components of the series, illustrating our approach using an example from wind energy. |
||||

12:30-14:00 | Lunch, Churchill College | |||

14:00-14:30 | Grindrod, P (University of Reading) |
|||

Domestic Behaviour and Demand: What Can Be Learned from Analysis of Smart Meter Data and How This May Be Useful? | Sem 1 | |||

We take a careful look at the variation in demand patterns form smart meter data and show that while these may be grouped in an unsupervised manor, such data driven classes are very poorly correlated to socio-economic and asset data. We are not like our neighbours!. We also consider the problem of forecasting smart meter behavior for individual households or small groups of households at substation level. These profiles are very spiked and the law of large numbers cannot rescue the situation for the substations. Thus we consider a methodology for producing rolling forecasts for such data. With smart storage such forecasts could be valuable to the consumer on their side of the meter and the distribution network operator on the other side of the meter. Realistic peaks are required which in turn effects the definition of forecast errors to be minimized: and indeed errors from forecasting peaks earlier are certainly better than errors from forecasting peaks later. We consider some implications for customers and their take up of new technologies, time of day tariffs, and forward planning. |
||||

14:30-15:00 | Brayshaw, D (University of Reading) |
|||

Physically-Based Modelling Approaches to Power-System Forecasting: Weather and Climate | Sem 1 | |||

Physically-based atmospheric models are powerful tools and are widely used in Numerical Weather Prediction (NWP) at timescales of hours to days. This talk will discuss the construction of these models, their application to power-forecasting, and the use of probabilistic strategies to reliably interpret their output. The advantages and disadvantages of NWP compared to pure statistical models will be discussed. The scope of the forecasting problem will then be extended to consider the potential for longer-range predictability and the affects of a changing climate on the energy system. Ongoing research exploring (a) the use of long-range weather forecasting for energy applications (weeks to seasons ahead), and (b) the impact of future climate on energy systems, will be discussed. |
||||

15:00-15:30 | Tea and Coffee | |||

Chair: Andrew Richards | ||||

15:30-17:00 | Richards, A | |||

Forecasting in Energy Networks Discussion | Sem 1 |

Wednesday 24 April | ||||

09:30-10:15 | Meyn, S (University of Florida) |
|||

Control of the Grid in 2020, and How Economics Can Help Us | Sem 1 | |||

Control of the grid takes place on many time-scales, and is analogous to many other control problems, such as confronted in aviation. There is decision making on times scales of days, weeks, or months; much like the planning that takes place for ticket sales for a commercial airline. Hourly decision making of energy supply is analogous to the chatter between pilot and air traffic controller to re-adjust a route in response to an approaching thunderstorm. Then, there is regulation of the grid on time-scales of seconds to minutes; consider the second-by-second movement of the ailerons on the wings of an airplane, in response to disturbances from wind and rain hitting the moving plane. There are also transient control problems: The recovery of the grid following one generator outage is much like the take-off or landing of an airplane. In this talk we survey control issues in the grid, and how the introduction of renewables brings new and interesting control problems. We also explain the need for economic theory to guide the formulation of contracts for resources needed for reliable real-time control. |
||||

10:30-11:00 | Coffee and Tea | |||

11:00-12:30 | Bialek, J | |||

Optimization in Energy Markets Discussion | Sem 1 | |||

12:30-14:00 | Lunch, Isaac Newton Institute for Mathematical Sciences | |||

14:00-14:30 | Lang, P (UK Power Networks) |
|||

Challenges - Change is the Only Constant | Sem 1 | |||

Peter Lang from UK Power Networks will introduce the role of Distribution Network Operators and describe existing systems used to manage the networks. He will also describe UK Power Networks' initiative projects funded through the Low Carbon Network Fund and the challenges still facing DNOs. |
||||

14:00-17:00 | Open for Business | |||

14:30-15:00 | Dixon, R (Department of Energy & Climate Change) |
|||

Reflecting and Communicating Uncertainty in Analysis | Sem 1 | |||

What lessons can analysts at DECC learn from other organisations about how best to reflect and communicate uncertainty in the analysis they undertake |
||||

15:00-15:30 | Bialek, J (University of Durham) |
|||

Technical Limits of Penetration of Non-Synchronous Wind and PV Generation in an AC Interconnection | Sem 1 | |||

Wind and PV generation is usually connected to the grid using an asynchronous interface: induction generator in the case of wind and DC/AC converter in the case of PV. As the operation of the power system is based around synchronous operation of synchronous generators used in traditional coal/gas/nuclear/hydro power stations , there is a technical limit of how much asynchronous generation can be connected to an AC interconnection. Recent studies in Ireland indicate that the limit is about 50-75% of the total generation operating at any time. The talk will discuss the physical reasons for the limit and possible means of overcoming it. |
||||

15:30-16:00 | Tea and Coffee | |||

16:00-17:00 | Meyn, S | |||

Panel Discussion | Sem 1 |

Thursday 25 April | ||||

09:30-10:00 | Kubik, M (University of Reading) |
|||

Responding to the Challenges of Wind Variability | Sem 1 | |||

The variability of energy supplied from renewable resources is commonly identified as one of the major challenges of integrating renewable energy with existing power systems. In Northern Ireland, with a 40% renewable energy target by 2020, this is a particularly challenging proposition. No existing large-scale electricity grid is able to operate without some minimum level of conventional generation, which is required for both system security and to maintain power quality. This minimum stable generation level restricts the amount of wind energy that can be used to satisfy system demand, any excess of which must be wasted ('curtailed'), at cost, if it cannot be stored. A further system operator concern is the occurrence of low probability, high swings in wind generation that exceed the operational characteristics of conventional units, requiring, for example, fast acting peaking plant to respond to a sudden drop in wind power. As wind power increasingly displaces conventional generation from operating, the capacity of this generation to respond to wind variability diminishes. The purpose of this talk is to overview a research project investigating both these complex issues using a scenario based approach and a 32 year long reanalysis wind data set, validated alongside historic Northern Ireland and Great Britain wind generation data. Both challenges are quantified and important factors in order to reduce the system impact of curtailment and improve capability are proposed. Power plant modifications, interconnection and battery storage are identified as potential solutions to the challenges faced in Northern Ireland. The complex interrelated nature of interconnected highlights a further mathematical need to understand to what extent interconnector capacity can be relied upon. |
||||

10:00-10:30 | Schulze, T (University of Edinburgh) |
|||

A Scenario Decomposition Method for Stochastic Unit Commitment Problems | Sem 1 | |||

In recent years the expansion of energy supplies from volatile renewable sources has triggered an increased interest in stochastic optimization models for generation unit commitment. Solving this problem directly is computationally intractable for large instances. In this talk we outline how a Dantzig-Wolfe reformulation can be used to decompose multistage stochastic unit commitment problems by scenarios. We develop a dually stabilized column generation framework which can handle convex quadratic and piecewise linear generation costs and is capable of solving stochastic unit commitment problems to optimality. We use a dual initialization procedure to hot start our method. Numerical results are given to illustrate that convergence can be achieved within a few iterations of our method. |
||||

10:30-11:00 | Cruise, J (Heriot-Watt University) |
|||

A Series of Interesting Power System Models | Sem 1 | |||

In this talk we will introduce a series of three simplistic models designed to capture future and on going issues with wind integration. Each model is designed to qualitatively improve our understanding of the associated issue. In each case we will motivate the model and highlight the simplifying assumptions we have made. This talk serves as an introduction to the later talks by Ksenia Chernysh, Glyn Eggar, and Stan Zachary who will discuss the analysis of these models. |
||||

11:00-11:30 | Coffee and Tea | |||

11:30-12:00 | Chernysh, K (Heriot-Watt University) |
|||

Controlling Conventional Generation to Minimize Forecast Error Cost | Sem 1 | |||

Nowadays renewable power sources are extremely important. However, their unpredictability makes effective managing of power system difficult. System operators predict net demand, where net demand is power produced by renewable sources subtracted from the total demand. The power system acts autonomously to cover predicted net demand but is unable to deal with errors in prediction. If the local supply fails to meet demand the shortfall must be met by imported power via interconnectors. Although, imported power is expensive. Our main objective is to minimize cost by reducing power purchased from abroad. This can be done by scheduling of additional conventional power plants. Ramp constraints lead to the need to carry out pre-emptive actions. In this talk we present an initial model for considerably large prediction errors and explore some properties of its optimal management. |
||||

12:00-12:30 | Eggar, G (Heriot-Watt University) |
|||

Using Storage to Control Uncertainty in Power Systems | Sem 1 | |||

This talk will discuss a simple probabilistic model exploring how storage can be effectively used to aid the operation of an electricity network with uncertain energy inputs. What is the best way of scheduling conventional units in advance to make optimum use of the variable and hard to predict renewable resource? The solution to this problem will depend, amongst other things, on oThe quality of the renewable prediction process oThe capacity of the store oThe efficiency and maximum release rate of the store oRamp constraints and delayed reaction of conventional generation oThe risk appetite of the network operator |
||||

12:30-14:00 | Lunch, Churchill College | |||

14:00-14:30 | Zachary, S (Heriot-Watt University) |
|||

Valuing Storage and Demand-Side Management | Sem 1 | |||

Abstract: Storage of electrical energy can be viewed as shifting energy through time. To assess the economic value of such storage, we may assume that energy is always available at a price which is time-dependent, so that the value becomes that of being able to shift energy from times when it is cheap to times when it is (otherwise) expensive. We consider the mathematics of optimal control under conditions in which there are both capacity and rate constraints on storage, and under conditions where this activity may be sufficiently large in scale as to itself impact on market prices. In particular we show that time horizons for optimal decision making are typically short - of the order of a day. We discuss also possible extensions to cases where future prices are uncertain, and to time-shifting of demand. |
||||

Chair: Frank Kelly | ||||

14:30-15:00 | Kelly, F | |||

Optimisation of Storage and Demand Management Discussion | Sem 1 | |||

15:00-15:30 | Tea and Coffee | |||

15:30-17:00 | Kelly, F | |||

Optimization of Storage and Demand Management Discussion (Continued) | Sem 1 |

Friday 26 April | ||||

Chair: Robert Leese | ||||

09:30-11:00 | Problem Scoping, Identification of Mathematical Grand Challenges in Energy Systems - Breakout Discussion Groups | |||

11:00-11:30 | Coffee and Tea | |||

11:30-12:30 | Leese, R (Mathematics KTN) |
|||

Problem Scoping, Identification of Mathematical Grand Challenges in Energy Systems - Summary | Sem 1 | |||

12:30-14:00 | Lunch, Churchill College |