Refuelling Nuclear Reactors – a multistage optimisation problem? Richard Overton, British Energy, Gloucester Mathematics 2010 – Saddlers’ Hall, London, April 22nd 2010 BE/FORM/COMM/050 Revision 000 Contents 1. A brief background to British Energy’s reactors and electricity production 2. A simplified description of the refuelling an Advanced Gas-cooled Reactor 3. A simple model for a stable equilibrium state 4. Formulating as an detailed time-dependent optimisation problem 5. The objective, constraints, and different approaches to optimising 6. Observations on obtaining an optimum (a personal view) BE/FORM/COMM/050 Revision 000 Prologue A youngster’s view of mathematics : 1. Logic 2. “A Right Answer” – no uncertainty 3. Precision 4. Perfection BE/FORM/COMM/050 Revision 000 British Energy’s Reactors > British Energy operates… > a Pressurised Water Reactor (PWR) at Sizewell (Suffolk) BE/FORM/COMM/050 Revision 000 British Energy’s Reactors (2) > …. and 14 Advanced Gas-cooled Reactors (AGRs) BE/FORM/COMM/050 Revision 000 British Energy’s Reactors (3) > … that’s 15 nuclear reactors at 8 sites around the UK BE/FORM/COMM/050 Revision 000 British Energy’s Reactors – scale of operations > British Energy produces 15-20% of UK electricity from nuclear power > Each AGR produces 500-600 MW of electricity (½ million “1 bar fires”). > Together all our reactors save emission of around 35 million tonnes CO2 annually (equivalent to half the UK’s car emissions) > Each reactor’s daily electricity production sells for around £½ m > Fuel costs are around 20% of selling price > Optimisation can potentially help (a) maximise reactor output (b) minimise fuel costs BE/FORM/COMM/050 Revision 000 Generation of energy from nuclear fission > Heat energy produced by splitting (fission) of uranium atoms by neutrons Heat (Uranium) BE/FORM/COMM/050 Revision 000 Reactor schematic > Enrichment : Proportion of U235 increased from natural (0.7%) to about 3.5% to assist fission > More neutrons produced by fission – chain reaction Fuel assemblies Reaction rate controlled by neutron-absorbing rods partially inserted into reactor BE/FORM/COMM/050 Revision 000 Gas Coolant Steam Reactor Primary Circuit Turbine Alternator Boiler Water Water Secondary Circuit Condenser Continuous Supply of Cooling Water Water Pump Pump Pump BE/FORM/COMM/050 Revision 000 Refuelling an Advanced Gas-cooled Reactor BE/FORM/COMM/050 Revision 000 The AGR Fuel Element Eight elements as a ‘fuel assembly’ loaded into each channel BE/FORM/COMM/050 Revision 000 About 300 fuel channels per reactor…. BE/FORM/COMM/050 Revision 000 Typical Distribution of Channel Powers 332 fuel channels New fuel Old Fuel BE/FORM/COMM/050 Revision 000 Typical distribution of Fuel ‘Age’ BE/FORM/COMM/050 Revision 000 The need to refuel… Operation & Generation → fuel becomes less reactive control rods withdrawn slowly Each channel needs refuelling every 6-7 years (generally the older fuel) Every few weeks a batch of 8 or so channels are refuelled Not a trivial operation ….. BE/FORM/COMM/050 Revision 000 AGR Fuelling Machine BE/FORM/COMM/050 Revision 000 A simple refuelling model BE/FORM/COMM/050 Revision 000 ‘deadtime’ d Shutdown period Scaled Reactor Power 1 ‘deadtime’ d Refuelling batch 3 ‘deadtime’ d Refuelling batch 2 Refuelling batch 1 Simplified Refuelling Process p n channels refuelled r per day n channels refuelled r per day 0 n channels refuelled r per day 0 n channels refuelled r per day 1095 N refuelling batches N 1n d 1p n/r d70 dn/r )11ppddnn 70 N (p i1 1165 1165 Nn N r rr Fuel demanded by reactor 1165 Time (days) = Fuel supplied to reactor BE/FORM/COMM/050 Revision 000 Number of batches Equivalent days at full power Fractional ‘load factor’ Daily income (£) 1095i n N n (1 p)i(d n / r ) D 1095 N (1 p)(d n / r ) (1095i n)(1 p)(d n / r ) 1095 n ( 1 p ) i ( d n / r ) L 1165 G.L Can now determine sensitivity to n,p,d,r and help define long term refuelling strategy BE/FORM/COMM/050 Revision 000 Results from a simple model Annual Reactor Output Demand exceeds supply Supply exceeds demand Load low enrichment fuel Fuel supply capability (elements/year) BE/FORM/COMM/050 Revision 000 Results from a simple model Annual Reactor Output Demand exceeds supply Supply fulfills demand Load higher enrichment fuel Fuel supply capability (fuel assemblies/year) BE/FORM/COMM/050 Revision 000 A more detailed refuelling model [ BE/FORM/COMM/050 Revision 000 Control rod position Reactor Power Timeline of repeated refuelling & operation Time BE/FORM/COMM/050 Revision 000 Channel Power Effect of age on channel power 5 Age of fuel (yrs) BE/FORM/COMM/050 Revision 000 Effect Effectofofhigher age on enrichment channel power on channel power Channel Power Maximum channel power allowed Higher enrichment again Higher enrichment 5 6 Age of fuel (yrs) 7 BE/FORM/COMM/050 Revision 000 Effect of gadolinium poison on channel power Channel Power Maximum channel power allowed Effect of gadolinium ‘poison’ 1.5 5 6 Age of fuel (yrs) 7 BE/FORM/COMM/050 Revision 000 Optimisation of the detailed model [ BE/FORM/COMM/050 Revision 000 Objective Function & Constraints > Objective : Maximise fuel life, or generation, or ‘profit’ over a long period (years), in which there are k refuelling batches i.e. find max ΣFi > Constraints: Keep power profile across reactor as flat as possible i.e. limit peak & spread ‘Shutdown requirement’ ensures reactor can be shutdown safely Max {Pi} ≤ P’ maximum power limit Max {Pi} - Min {Pi} ≤ S’ limit spread in channel powers Use 2 enrichments (usually fixed zones in reactor) > … and many other constraints not mentioned here! BE/FORM/COMM/050 Revision 000 Mathematical Model > Choices (“controls”) at refuelling batch k : No. of channels to refuel nk Size of refuelling batch k (k choices) Channels for refuelling (Ii)k = 0 ……………..…….. do not refuel channel i = 1 ……………………. refuel channel i 300C choices (~1015) …. for i = 1,2,… nk nk Fuel type to use (ei)K ε { E1, E2 } ………… Enrichment choice (of 2) (qi)K ε { Q1, Q2,…QL } ……... Poison choice (typically 2-3) …. for i = 1,2,… nk BE/FORM/COMM/050 Revision 000 Mathematical Model (2) > Constraints Let Pik = power in channel i after refuelling batch k Pik ≤ Pmax i = 1, 300 Power limit for each channel (300 constraints) Maxi {Pik} – Mini {Pik} ≤ S Range of powers limited (1 constraint) ∫ Pik(t) dt ≤ A Energy output (Age) limit on fuel assembly (300 constraints) SDk (nk, (Ik,ek,qk)) ≥ SDmin Shutdown capability (1 constraint) Typically 20000 constraints to cover a 4 year period (30 batches) Constraint functions are nonlinear, evaluated by ‘black box’ physics model (PANTHER code) Importantly (1) Refuelling a channel affects neighbouring channels (2) Refuelling choice affects next choice for neighbouring channels BE/FORM/COMM/050 Revision 000 Optimisation options (1) > Option 1 : Exhaustive evaluation > Choosing 8 channels for one refuelling batch > In practice, will choose from oldest 10% of fuel 300 15 C8 ~ 10 choices 30 7 C8 ~ 10 choices 8 > Must also choose refuelling batch size 4-12 channels → 10 choices > To be useful, need to evaluate over k = 20 refuelling batches (3 years) : 6000 C160 choices > ….and also need to choose fuel types for each refuelled channel Each choice requires one evaluation of the objective …and evaluation of all constraints (PANTHER physics model) BE/FORM/COMM/050 Revision 000 Optimisation options (2) > Option 2 : Heuristic approach > Choose oldest 8 channels for refuelling > Quickly leads to violation of range of power constraint > Have programmed in a set of heuristic rules, based on knowledge/experience, but still performs poorly BE/FORM/COMM/050 Revision 000 Optimisation options (3) > Option 3 : Formal ‘textbook’ methods > Iterative hillclimbing approach? > Newton-Raphson? …with penalty functions to handle constraints x n 1 x n f '(xn ) f "(xn ) > Far too high dimensionality, discrete choices, ‘black box’ model (no derivatives) BE/FORM/COMM/050 Revision 000 Optimisation options (4) > Option 4 : Linear/integer programming > Piecewise linear approximation to physics model > Integer 0/1 variables > Refuelling choice (Ii)k (0/1) variables > Again, far too high dimensionality, beyond computational feasibility BE/FORM/COMM/050 Revision 000 Optimisation options (5) > Option 5 : Dynamic programming > Exploit multistage structure to subdivide problem into k subproblems > Use Bellman’s recursive Principle of Optimality > Fi(Si-1) = Maxui { fi(ui, Si-1) + Fi+1(Si (ui, Si-1)) } for i = 1, 2, 3 …k > Relies on state Si at stage i depending only on state at stage i-1 (Si-1) and choices made at stage i (ui) 8 > Still ~ 10 choices of channels to refuel at each stage > But unfortunately state Si depends on choices ui, ui-1, ui-2,… made at several previous refuellings (the gadolinium poison effect persists over several refuelling batches) BE/FORM/COMM/050 Revision 000 Optimisation options (6) > Option 6 : Genetic Algorithms > Collaborative work with Imperial College > “Breed” from initial population of possible solutions > Random selection of characteristics (refuelling channels, fuel type) from 2 individual solutions to create new solution. > Evaluate constraints > Evaluate objective > If improvement, keep in population with pre-determined probability > Does need many tens of thousands of evaluations to get good solutions BE/FORM/COMM/050 Revision 000 Objective Function (1) > First attempts with Genetic Algorithm > Objective : maximise Age of fuel > Tends to > (a) minimise batch size (take each fuel assembly to its age limit) > (b) create more refuelling batches => loss of generation > Instead, need a financially based objective function…. BE/FORM/COMM/050 Revision 000 Objective Function (2) Let G = value of 1 full power day generation F1 = cost of 1 assembly of fuel type 1 F2 = cost of 1 assembly of fuel type 2 fpd (i) = number of full powers days generation after refuelling batch i Then in N refuelling ‘outages’ and generation runs, the value of generation is : N G fpd (i ) i 1 The cost of the fuel for N refuelling batches is : N (F n ( i ) F n ( i )) i 1 1 1 2 2 BE/FORM/COMM/050 Revision 000 Objective Function (3) And a suitable objective function to maximise is then: Z N 1 N 1 i 1 N 1 i 1 G. fpd (i ) [ F1n1 (i ) F2 n2 (i )] G. fpd ' ( N ) F1n'1 ( N ) F2 n' 2 ( N )) N fpd(i ) fpd' ( N ) (a (n (i ) n (i ))b(1 p) m(i )c(1 p)) i 1 i 1 1 2 BE/FORM/COMM/050 Revision 000 Objective Function (4) > Success with ~50000 to 100000 evaluations of objective function, giving choice of different refuelling batches to refuelling planner BE/FORM/COMM/050 Revision 000 > However, in practice : > Other constraints appear – Early refuelling of a channel required Occasional breakdown of fuelling machine – smaller refuelling batch > Objective function changes (selling price, fuel cost) Step change, contractual change, government intervention > Improvements in safety – Additional restrictions on fuel age or power : new or changed constraints > Solution needs to be robust to such changes i.e. avoid significant losses if rules change BE/FORM/COMM/050 Revision 000 UK Baseload Power Prices to July 2008 (annuals) 90 85 Electricity Pool BETTA Market NETA Market Tight supply/demand balance in the oil and coal markets push prices to record levels 80 75 70 65 Station Gate Market Prices 60 Market Price - £/MWh Further Oil & Gas price rises NBP Market Prices Cold winter & high gas prices Rising oil, coal & gas prices Oct '02 TXU collapses E.on 55 buys generation & supply business 50 45 40 Sept '02 BE announces Apr '02 state-aid 1.5GW plant is mothballed 35 30 25 Oil & Gas price rises Apr '03 2 GW of plant closes 2.3GW is mothballed Increasing security over winter gas supplies Fear of Winter 05 gas supply Effect of LCPD & high carbon shortage 20 15 Oct '03 3GW of plant returns 10 prices Weak gas & mild winter weather Dec '03 EU ETS enters front year 5 0 Aug-99 Feb-00 Aug-00 Feb-01 Aug-01 Feb-02 Aug-02 Feb-03 Aug-03 Feb-04 Aug-04 Trade Date Feb-05 Aug-05 Feb-06 Aug-06 Feb-07 Aug-07 Feb-08 BE/FORM/COMM/050 Revision 000 Uranium Ore prices Historic Uranium Spot Prices 150 140 130 120 110 90 80 70 60 50 40 30 20 2006 $ 2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 1988 1986 1984 1982 1980 1978 1976 1974 0 1972 10 1970 US$ / lb U3O8 100 Outturn $ BE/FORM/COMM/050 Revision 000 Conclusion > For complex systems such as AGR refuelling: > Not worth the effort in optimising too finely > Optimisation can be undermined by step change in objective or constraints > System managers do not necessarily seek optimum > Develop training on amelioration techniques and know when to stop! BE/FORM/COMM/050 Revision 000 References D. Barrable, J. H. Kershaw, R. S. Overton, K. Brearley and D. R. Gray “Increasing fuel irradiations in advanced gas-cooled reactors in the UK” Nuclear Energy, 37 A. K. Ziver , C. C Pain, C. R. E. de Oliveira, A. J. H. Goddard and R. S. Overton “Genetic algorithms and artificial neural networks for loading pattern optimisation of advanced gas-cooled reactors.” Annals of Nuclear Energy 31, Issue 4 BE/FORM/COMM/050 Revision 000 Acknowledgements > Grateful thanks to: Vanessa Strong, John Foulkes, Andy Williams, Nigel Rhodes, Chris Hall, Rik Bahia (colleagues at British Energy) … and Kay & Liz Overton BE/FORM/COMM/050 Revision 000 Epilogue Trainee mathematician Precision Perfection A right answer BE/FORM/COMM/050 Revision 000 Metamorphosis in the attic Mature industrial Trainee mathematician mathematician/physicist Precision Perfection Puts up with uncertainty A right answer Tolerates imperfection Faced with intractable problems With apologies to Oscar Wilde BE/FORM/COMM/050 Revision 000 Questions Any questions? BE/FORM/COMM/050 Revision 000