Programme
The workshop will offer a mix of invited papers and contributed papers with ample time for discussion and reflection.
On Monday 15th April, there will be a session run jointly with the Warwick Global Research Priority on Energy followed by a reception.
A workshop dinner will be held on the evening of Tuesday 15th April.
Programme Outline
| Monday 14th April (MS.02 Zeeman Building) | Tuesday 15th April (MS.02) | Wednesday 16th April (MS.02) |
| 09:30 - 10:30 Registration 10:30 - 11:00 Tea/Coffee 11:00 - 11:10 Welcome 11:10 - 12:00 IT1: Stan Zachary 12:00 - 12:40 CT1: Monica Giulietti 12:40 - 14:00 Lunch 14:00 - 14:50 IT2: Andrey Bernstein 14:50 - 15:00 Tea/Coffee 15:00 - 15:30 Energy GRP: Introduction 15:30 - 17:30 Energy GRP: Facilitated Discussion 17:30 - 19:30 Energy GRP Reception 19:30 - Free Evening |
09:30 - 10:20 IT3: Michael Goldstein 10:20 - 11:00 CT2: Anthony Lawson 11:00 - 11:30 Tea/Coffee 11:30 - 12:10 CT3: Ksenia Chernysh 12:10 - 12:50 CT4: Marek Brabec 12:50 - 14:00 Lunch 14:00 - 14:50 IT4: Tim Bedford 14:50 - 15:30 CT5: Majid Al-Gwaiz 15:30 - 16:00 Tea/Coffee 16:00 - 16:50 IT5: Idris Eckley 16:50 - 17:30 CT6: Gruffudd Edwards 17:30 - 19:30 Break 19:30 - Workshop Dinner |
09:30 - 10:20 IT6: John Moriarty 10:20 - 11:00 CT7: Jhonny Gonzalez 11:00 - 11:20 Tea/Coffee 11:20 - 12:10 IT7: Valentin Bertsch 12:10 - 13:30: Lunch 13:30 - 15:00 Discussion and wrap-up Close Suggestion: Go round on the Energy Trail before you leave! |
Abstracts
IT1: Stan Zachary (Department of Actuarial Mathematics and Statistics, Heriot-Watt University, UK)
Title:Roles of energy storage in balancing power networks
Abstract: Electric power networks require that supply and demand be kept balanced on a minute-by-minute basis. Storage, which may be regarded as shifting energy through time, is able to assist in this process by (a) smoothing predicted imbalances, such as between known times of high and low demand, (b) buffering against sudden and unpredicted variations in either supply or demand. Typically the economic use of storage requires that it is simultaneously able to provide both these (and often also other) services. We discuss an integrated mathematical framework for optimal control of such storage and for the determination of its economic value.
We discuss also possible extensions to the time-shifting of demand.
IT2: Andrey Bernstein (Schoolf of Computer and Communication Sciences, EPFL, Switzerland)
Title: A Composable Method for Real-Time Control of Active Distribution Networks with Explicit Power Setpoints
Abstract: The classic approach to the control of medium and low voltage distribution networks involves a combination of both frequency and voltage controls at different time scales. It was designed with a central operation in mind. With the increased penetration of stochastic resources, distributed generation and demand response, it shows severe limitations both in the optimal/safe operation of these networks as well as in aggregating the network resources for upper-layer power systems. To overcome this difficulty, a radically different control philosophy was recently proposed by the authors, which enables power networks as well as their resources to directly communicate with each other in order to define explicit real-time setpoints for active/reactive power absorptions or injections. The framework is composable, i.e. subsystems can be aggregated into abstract models that hide most of their internal complexity. In order to ensure short-term safety and long-term optimality of the network, efficient stochastic optimization methods should used in order to compute the setpoints. In particular, in the proposed framework, it is assumed that the setpoints are computed approximately each 100 msec in order to cope with the fastest dynamics of the power system. Hence, the employed optimization methods should be adapted to this fast time scale. In this talk, we describe one such method that is based on projected online gradient descent algorithm. We also present the application of the method to a benchmark microgrid and evaluate its performance.
IT3: Michael Goldstein (Department of Mathematical Sciences, Durham University, UK)
Title: Bayesian uncertainty analysis for complex energy related systems modelled by computer simulators
Abstract: Most large and complex physical systems are studied by mathematical models, implemented as high dimensional computer simulators. There is a growing field of study which aims to quantify and synthesise all of the uncertainties involved in relating models to physical systems, within the framework of Bayesian statistics, and to use the resultant uncertainty specification to address problems of forecasting and decision making based on the application of these methods. This talk will give an overview of aspects of this emerging methodology, in the context of uncertainty and planning for future power systems.
IT4: Tim Bedford (Department of Management Science, Strathclyde Busines School, UK)
Title: Epistemic Uncertainties for Availability Risk in Future Offshore Wind Farms
Abstract: To increase capacity and reduce costs, new offshore wind farms use novel large-scale and complex technology which introduces potential systemic weaknesses in design, manufacturing and operational processes. As offshore wind farms accumulate experience such weaknesses should be resolved, but most likely at the cost of additional investment. The inability of availability models to represent the epistemic uncertainty about systemic weaknesses can lead to underestimation of future performance. From the perspective of assessing financial performance in order to make the investment case, understanding the sources of these uncertainties is critical, as they will typically impact on early life, which is the time at which loss of generation capacity has the largest negative effect on the project projected NPV.
As part of an EPSRC funded project, we have developed a generic simulation model for new offshore wind farms. This model aims to support decisions by understanding the impact of risks arising from design, manufacturing, installation and operation. The model outputs various parameters that indicate the impact of aleatory and epistemic uncertainties on the productivity of the farm.
We describe the model and the work we have done with engineering and industry experts to validate its structure, discuss the types of features that the model can deal with, and consider the potential applications.
(Joint work with Athena Zitrou, Lesley Walls, Keith Bell, Kevin Wilson)
IT5: Idris Eckley (Department of Mathematics and Statistics, Lancaster University, UK)
Title: Alias Detection and Spectral Correction for Locally Stationary Energy Time Series
Abstract: When analysing energy time series, for example with a view to forecasting, we implicitly make the assumption that the series is sampled at a suitable rate. Specifically we assume that it is free from the effect of aliasing. However is this a realistic assumption to make?
In practice it is all too easy to overlook aliasing when conducting an analysis of a time series. Indeed it is rarely tested for, even though a bispectrum-based test of aliasing for (stationary) time series was proposed by Hinich and Wolinsky in 1988. For locally stationary series the situation is a bit different in that aliasing can be intermittent, depending on whether the spectrum locally contains frequencies higher than the Nyquist rate or not.
This talk will introduce a wavelet-based method to separate the spectral components of a locally stationary time series into two classes: (i) aliased or white noise components and (ii) lower frequency uncontaminated components.
(Joint work with Guy Nason (Bristol)).
IT6: John Moriarty (School of Mathematics, University of Manchester, UK)
Title: If electricity were priced like wine: Stochastic control, Stopping and Storage
Abstract: We address the pricing of battery charging for a random future time of demand, for example in electric vehicles, when the writer of the contract is assumed to be a small buyer in a real-time spot market with limited access to electricity storage. We cast this problem of pricing a call option with physical delivery as one of monotone singular stochastic control and consider further the optimal time to write such a contract, adding a problem of optimal stopping. Our motivation is to classify the solutions and identify any unexpected features. For a simple model allowing mean reverting and potentially negative spot prices (the Ornstein-Uhlenbeck process) and a non convex terminal cost for making good any shortfall at the time of demand, we find that the optimal control strategies divide into three classes depending on the problem parameters. We solve two of these cases and also obtain the optimal stopping rules which, despite their unusual disconnected form, are exactly solvable.
IT7: Valentin Bertsch (Institute of Industrial Production, Karlsruhe Institute of Technology, Germany)
Title: An integrated approach for uncertainty handling in the layout optimisation of decentralised energy systems
Abstract: In the layout planning of distributed energy systems, including photovoltaic (PV) systems in combination with heat pumps and thermal storages, many different design options can be chosen. The dimensioning of the individual components, such as the size of the heat storages, has an immediate impact on the economic profitability of the system as a whole. Moreover, the economic profitability is subject to manifold uncertainties, e.g. the uncertainty related to the future development of electricity prices or the development of the electrical and thermal load as well as the solar PV generation profiles needs to be addressed adequately in the layout planning process. Therefore, an approach is presented integrating modules for (i) simulating consistent ensembles of input data, such as solar irradiation or temperature profiles, by a Markov process, (ii) transforming these initial profiles to consistent sets of PV generation and heat demand profiles and (iii) using the generated profiles in a (mixed-integer) linear optimisation. The approach is demonstrated for a residential quarter including approx. 70 households.
CT1: Monica Giulietti (Global Energy Group, Warwick Business School, UK)
Title: Revenues from Storage in a Competitive Market: Empirical Evidence from Great Britain
Abstract: In dealing with increasing amounts of intermittent supply into the electrical power market, technologies for electricity storage not reliant on the geographical features required for hydroelectric plant offer an alternative to building and operating additional peaking generation plants. The economic question is whether sufficient revenues can be raised from operating the storage facilities. There are many papers that consider the issue of energy storage within a broadly economic framework. These papers address a range of issues, but the key aim is to evaluate the profitability of storage. Despite the several papers in the area, there remain some significant gaps. This paper aims to generate a first contribution in this direction The novelty of our approach is twofold: 1) previous works are focused on optimization techniques to maximize profits considering electricity production costs, while our approach relies on the comparison between spot and future prices 2) previous methods are based on the analysis of arbitrage opportunities on long timescales and use actual prices. The focus of the present paper will be on short timescales (one-two days) which seem to be most promising and will use predicted prices instead of actual. Monetary benefits which can be achieved when investing in energy storage are based on the difference between the price that energy will command when it is released and the price paid when injected into the storage. We investigate this relationship using time series statistical techniques for various maturities of forward prices, using data on assessments of power prices for future delivery.
Our empirical analysis of a hypothetical storage facility is based on data on power prices and power futures assessment for Great Britain which have been obtained from Platts UK’s Power markets dataset. The initial results on the potential profitability of storage facilities with different levels of round-trip efficiency indicate that, when buying and selling base load power, positive revenues can be obtained with round-trip efficiency at the 60% and 70% level, although positive results tend to be concentrated in the years between 2005 and 2008.
Our results indicate that arbitrage opportunities for storage facilities particularly arise when energy is stored over a short-term period as little as a day and certainly no more than a week.
A further step in the analysis has involved examining the distributional properties of the extreme values of the price distribution, as these are the observations that generate the most profitable arbitrage opportunities. The identification of an appropriate distribution for extreme values requires the definition of a threshold beyond which the observed values can be classed as extreme. For the purpose of our analysis the thresholds for extreme values are exogenously determined by the chosen levels of round trip efficiency (60% and 70%, as discussed earlier). However, we find that in the case of round trip efficiency of 60% we have an insufficient number of observations for a reliable statistical analysis, while at the 70% threshold we can use 11.4% of the observations (335) from the original series. Using this subset of observations from the original series we estimated a Generalized Pareto distribution, obtaining highly significant form and scale parameters. This estimated distribution satisfactorily fits the theoretical one.
CT2: Anthony Lawson (Department of Mathematical Sciences, Durham University, UK)
Title: Emulation, Uncertainty and Planning Future Power Systems
Abstract: This presentation details the application of statistical emulators to the constraint cost problem that arises in power systems. Constraint costs are the costs that arise when finite transfer capabilities (i.e. power lines) restrict the transfer of energy capacity; and a more expensive capacity source must be used instead. The transfer capabilities can be increased at a cost, which would allow more capacity to be traded and reduce constraint costs. Cost benefit analysis is used to minimise the constraint costs that arise over a given period plus the cost of any reinforcement made. At the time investment decisions are made, there is great uncertainty over the future system background against which the system must be planned. Such uncertainties exist in installed generating capacities; the availability probability of these capabilities, the demand levels in the system and many more. Statistical simulators are used to estimate expected constraint costs in a system for a given background scenario. However, the full simulator is too computationally expensive to run under every set of input values desired. Therefore, statistical emulators must be used to model expected constraint costs under uncertainty. The statistical emulators can then be combined with accurately specified prior beliefs to estimate expected constraint costs under uncertainty.
This talk explores the use of statistical emulators to make reinforcement decisions. The particular problem considered is that of Britain's power system. Britain is divided into seven zones, roughly splitting Britain in a linear pattern from North to South. Constraint costs are modelled as arising at the boundaries between the zones.
The example presented will consider making reinforcement decisions for boundaries between Scotland and England and between North and Mid-England for the 2016 scenario when using data from 2010 predictions. Uncertainty will be considered in 3 variables: demand level, nuclear availability probability and CCGT availability probability. The methodology presented can easily be extended to deal with far more complicated scenarios.
(Joint work with Chris Dent and Michael Goldstein).
CT3: Ksenia Chernysh (School of Mathematical and Computer Sciences, Heriot-Watt University, UK)
Title: Control Strategies for conventional generation to reduce shock shortfalls
Abstract: We are interested in a mathematical toy-model, which explores control strategies for slow response thermal plants in dealing with shortfalls in supply in power systems and minimising associated costs. The shortfalls that we will be interested in occur due to sudden changes in wind power output, for example the movement of a frontal system over the U.K. or a gust forcing multiple turbines to feather their blades. The ramp constraints of conventional generation mean that they are not able to respond in normal circumstances to such events.
Instead we consider the idea of switching on additional conventional generation ahead of time. To select the amount of extra generation we use a series of rules that are aimed to minimise the average spending on both the excess conventional plant and the fast reacting capacity needed to deal any shortfall.
Throughout the talk we assume Poisson structure of underlying shortfall process. It simplifies the problem and helps to understand interesting features of possible optimal controls.
We will present both numerical simulation and theoretical results about the existence and nature of the optimal control.
CT4: Marek Brabec (Institute of Computer Science, Academy of Computer Sciences of the Czech Republic)
Title: How good are NWP-based solar radiation forecasts for photovoltaic farms and how to improve them?
Abstract: In order to assess various risks related to the erroneous prediction of the electric power produced by solar farms, we first conduct a detailed spatio-temporal analysis of the radiation prediction errors for several commonly used NWP (numerical weather prediction) models and their variants (MM5, WRF in various spatial resolution and versions). The errors are obtained by comparing the raw NWP and homogeneously calibrated NWP forecasts to in situ measurements spread over the Czech Republic. We consider short to medium forecast horizons, namely hourly forecasts for D0, D1 and D2. The analysis shows quite clearly that the NWP-based forecasts are under-smooth. They generate a lot of high frequency variability which is not followed by the data. Hence the main route to the improvement leads via model-based smoothing, both in time and space. Further, demonstrating severe systematic non-homogeneities of forecast errors in time and space, as well as with respect to variety explanatory variables, we will motivate a semiparametric model to be used for prediction quality improvements. The model builds on smoothness across both time and space and hence it is formulated as a penalized regression.
The resulting radiation predictions are then used as inputs for improved forecasts. Conversion between the radiation and electric power forecast can be viewed as an errors-in-variables problem. Additionally, it features several other complications. In fact, not only the variance of the power output response depends heavily on the (predicted) covariate, but in fact many other important distributional properties do change as well. For instance skewness changes dramatically due to the presence of both lower and upper bounds on the power output. We formulate a flexible GAMLSS approach to this problem and apply it on a selection of solar farms in the Czech Republic. When assessing the performance of the resulting power forecasts, we start with traditional measures like RMSE, MAE, but eventually we accentuate more focused characteristics like those related to the “ramp” prediction as well as prediction of other “stylized facts” relevant from practical network management point of view.
CT5: Majid Al-Gwaiz (Process & Control Systems Department, Saudi Aramco, Saudi Arabia)
Title: Supply Function Competition in Electricity Markets with Flexible, Inflexible, and Variable Generation
Abstract: In this paper we study the supply function competition between power-generation firms with different levels of flexibility. Inflexible firms produce power at a constant rate over an operating horizon, while flexible firms can adjust their output to meet the fluctuations in electricity demand. Both types of firms compete in an electricity market by submitting supply functions to a system operator, who solves an optimal dispatch problem to determine the production level for each firm and the corresponding market price. We study how firms’ (in)flexibility affects their equilibrium behavior and the market price. We also analyze the impact of variable generation (such as wind and solar power) on the equilibrium, with the focus on the effects of the amount of variable generation, its priority in dispatch, and the production-based subsidies. We find that the classic supply function equilibrium model overestimates the intensity of the market competition, and even more so as more variable generation is introduced into the system. The policy of economically curtailing variable generation intensifies the market competition, reduces price volatility, and improves the system’s overall efficiency. Moreover, we show that these benefits are most significant in the absence of the production-based subsidies.
CT6: Gruffudd Edwards (School of Mathemetical and Computer Sciences, Heriot-Watt University, UK)
Title: Ensemble-Based Probabilistic Forecasts for Wind Speeds at Multiple Look-Ahead Times
Abstract: The construction of probabilistic forecasts for the outputs of wind power generators has received considerable research attention in recent years, reflecting the substantial economic benefits of such forecasting given the rapid growth in wind energy capacity installed globally. However, while much of this work has focussed on wind power outputs/ speeds at single locations and single look-ahead times, recent work at Heriot-Watt has been concerned with joint predictive distributions for wind speeds.
Our forecasts are based upon ECMWF ensembles, and we are interested in the extent to which the time varying properties of ensemble members mimic observed behaviour, and how they may be corrected to do so better. Armed with temporally-joint distributions, we explore how the NWP-based forecasts may be updated based on observations at times intermediate between forecast issue times. Data has been obtained for a number of locations in southern Scotland, and we intend to extend our analysis to the spatio-temporal case in the future. Our work is in conjunction with MeteoGroup and National Grid.
CT7: Jhonny Gonzalez (School of Mathematics, University of Manchester, UK)
Title: Risk-sensitive optimal switching and applications to district energy systems
Abstract: We propose a new practical methodology for risk-sensitive stochastic optimal control, showing that material decreases in risk indices are possible with relatively little loss of average case performance. Cost optimisation of energy system assets has typically been carried out under the assumption of risk-neutrality, minimising average operational costs. The risk pro le of control strategies is thereby ignored, despite the fact that liberalised electricity markets can be highly volatile and financial risk is a material consideration. In a flexible energy system with cogeneration and heat storage, however, it is possible to exploit variation in wholesale price level or volatility (or both) by shifting heat demand through time and varying electricity demand, achieving a balance between low average cost and low volatility which takes account of risk preferences. Based on least squares Monte Carlo regression, our proposed method optimises an exponential objective function containing a risk sensitivity parameter which may then be tuned to achieve the desired tradeoff. We provide a realistic case study of a flexible district energy system, where local heat and electricity demand must be satisfied at minimum cost subject to stochastic price dynamics and the physical constraints of the system. In this example we compare risk-neutral and risk-sensitive optimal strategies and show consistent changes in economic risk under two different risk measures.
The Energy Trail
Workshop participants may be interested in embarking on an energy-themed walk through the campus: the Energy Trail covers 16 individual points of interest on campus showcasing the University’s world-class research and technology to solve the global energy challenge. More information about the Energy Trail and a map can be found here.