Decarbonization of Campus Energy Systems

Transcription

1 Decarbonization of Campus Energy Systems G RID- F RIENDLY E LECTRIFIED D ISTRICT H EATING AND C OOLING WITH T HERMAL S TORAGE JA de Chalendar, PW Glynn, J Stagner and SM Benson jdechalendar@stanford.edu

2 Meet the team (1/2) Prof. Sally M. Benson Energy Resources Engineering: Geologic storage of CO 2 / Subsurface characterization and monitoring / Technologies and energy systems for a low-carbon future. Prof. Peter W. Glynn Management Science & Engineering: Simulation / Computational probability / Queueing theory / Statistical inference / Stochastic modeling. Joe Stagner: Led the development of the initial Stanford Energy Systems Innovations project. Two senior undergraduate students. Energy Resources Engineering department. 2

3 Meet the team (2/2) My background: École Polytechnique (France), Stanford MSc research: pore-scale imaging and modeling for CO 2 sequestration, Internships: IMPA / Geli / PNNL, Stanford PhD research: topic of today. 3

4 Proposed outline Motivation Stanford Energy Systems Innovations an ideal case study Current work Optimal scheduling for SESI (in preparation) More advanced demand response (WIP) Some ideas for future work & discussion 4

5 Threads for this talk 1. District heating and cooling options should be considered more. 2. Problems with long-term objectives but short-term decisions. 3. Derive generally applicable insights from real-life systems. 4. What do you think?

6 Shifts in the Power Grid (1/2) Need to reduce carbon emissions drives: Electrification, Large-scale renewables. Ø Shifts in the management of the power grid are necessary. Residential (6%) Industry (23%) Transport (23%) Other (10%) Electricity & Heat (42%) World CO2 emissions from fuel emissions by sector in Adapted from IEA (2016). The CAISO duck curve and non-dispatchable generation in

7 Shifts in the Power Grid (2/2) New challenges associated with both large-scale renewables and increased electrification (uncertain fluctuations, inertia, availability, ) Ø Turn to the demand side for new control options. Increasing interest in the (historically passive) distribution system level. Electric loads that offer energy services and flexibility have value. Ø Heat pumps or large manufacturing machines, but also electric vehicles, household appliances, etc. Increasing number of temporal and physical scales are involved. 7

8 Why study district heat? (1/2) An energy transfer that is dependent on temperature. Heat is big, dirty, and associated with slower timescales than electricity. Why district energy? Ø District energy networks have historically played a small but important role at a local level, Ø Motivating example of wind power in China. Rapid urbanization prompts further questions around the City of the Future : Ø What is the economic case and carbon impact of electrified district heating and cooling? 8

9 Why study district heat? (2/2) Three generations of district heating Ø Steam-carrying concrete ducts (Manhattan, 1877; Paris, 1920s), Ø Pressurized hot water in concrete ducts (Soviet era), Ø Lower temperature hot water in prefabricated, pre-insulated pipes (most Scandinavian systems today). Towards 4 th Generation District Energy? Ø From a recycling system for waste heat to multi-energy inputs from a variety of thermal sources. In the context of calls for Energy Systems Integration. 9

10 10 Stanford s Campus as an Ideal Case Study

11 Stanford s Campus as an Ideal Case Study Stanford s campus is master-metered. The SESI project is big (~36 MWe peak load) and centrally managed by an Energy Management System. Source of real data and insights as well as an ideal testbed. The entire distribution feeder can be viewed as a Virtual Power Plant. 11

12 Stanford Campus Electric Billing Structure Direct Access to California s electricity markets. Two-part energy tariff: monthly demand charge + hourly energy usage price. Day of year Electricity price ($/kwh) 12

13 13 Stanford s Campus as an Ideal Case Study

14 Meeting Stanford s energy requirements Three main streams for energy consumption. Cooling (GJ) Heating (GJ) Power (GJe) Day of year Each row corresponds to one day, each column to one hour.

15 The case for heat recovery Daily thermal loads and maximum daily load that can be served through Heat Recovery Chillers. 4 Thermal load (TJ) Cold Hot Cold (HRC) Hot (HRC) From data collected for the Stanford campus. Assumes load is served at beginning of day and HRCs produce 1.3x more hot than cold.

16 Characterizing loads K-means clustering applied to energy loads: Power is occupancy-driven, thermal loads are seasonal. Thermal loads are more difficult to predict than electric power consumption. Cooling (GJ) 214 (1) 0 (0) 72 (.13) 471 (.87) 27 Power (GJe) Confusion matrix for weekday/weekend using computed clusters for 757 days. Heating (GJ)

17 Characterizing loads Clustering of normalized consumption profiles into k=4 representative days. Identification of a shift in power consumption profiles in the Spring of 2017 this is most visible on weekends. 4.5MWe of rooftop solar come online Normalized power Power (GJe) 28

18 18 Stanford Campus as an Ideal Case Study

19 Stanford Campus as an Ideal Case Study 1. What does it mean to be flexible and to provide energy services? 2. What is the value of this flexibility? 19

20 Facility manager s generic problem (state, action, exogenous information): Policy: System model: Objective: Consider a rational agent, not irrational end-users. 20

21 Current work #1 CARBON- AWARE SCHEDULING FOR SESI

22 Scheduling the CEP - Problem formulation (1/2) Planning problem based on historical data, Formulated as a linear program. 22

23 Scheduling the CEP - Problem formulation (2/2) Decision variables: power to machines at every time step. State variable: amounts of energy in thermal storage (water in tanks). Linearize demand charge objective by introducing an auxiliary variable to represent the monthly peak demand. Formulate the problem of generating optimal schedules for the Central Energy plant as a linear program with ~130k variables and 170k constraints (hourly operations for a year). Energy prices & Power, heating, cooling loads min < %, ' > Subject to )' + ' 0 Hourly operations schedule for CEP machines 23

24 Scheduling the CEP - Implementation Implemented using JuMP (Julia) and solved using Gurobi. Linear program with ~130k variables and 170k constraints (hourly operations for a year). Data processing using Pandas (Python). Solves in a couple seconds on a laptop. 24

25 Optimal operations schedule base case Chillers and Heat Recovery Chillers are scheduled to to avoid the expensive early evening hours and minimize the cost of the aggregate imports from the power grid. Grid imports (GJe) Chiller output (GJ) HRC output (GJ) Day of year Heatmaps for key operating variables. Each row corresponds to one day and each column to one hour. 25

26 Optimal operations schedule base case Optimized loads billed under a demand charge system have a typical flat profile. Jan 16 Dec 16 In the Winter, HRCs are used in combination with gas-fired heaters; in the Summer, with electric Chillers. 26 Summer (Aug 3 to 10) Winter (Jan 3 to 10)

27 27 Accounting for carbon

28 Accounting for carbon Aggregate campus emissions were reduced by 68% through the SESI project. Can we do better by leveraging the current infrastructure? 28

29 California Grid Carbon Intensity Scenarios in mid-2020s Emissions impact the atmosphere on a timescale of years, but scheduling decisions must be made on a timescale of hours. Ø Average Emissions factors can be used to determine the carbon intensity depending on the generation mix. 16 Actuals High solar High solar & wind Day of year kgco 2 /MWe 30 Heat maps for the hourly AEFs (kgco 2 /MWhe) of the California power grid energy mix in the mid-2020s. Each row corresponds to one day, each column to one hour, and all images use the same colorbar.

30 Demonstrating Short-Term Flexibility of SESI (1/3) Three different scenarios for CAISO AEFs in mid-2020s. Four different operating conditions are tested for the optimization program: 1. Current operations (minimize electric bill), 2. Current operations without thermal storage, 3. $100 per tonne carbon tax, 4. Carbon optimal strategy. We determine the changes in cost and carbon emissions. 30

31 31 Demonstrating Short-Term Flexibility of SESI (2/3)

32 32 Demonstrating Short-Term Flexibility of SESI (2/3)

33 33 Demonstrating Short-Term Flexibility of SESI (3/3)

34 Carbon Case Study Conclusions The Central Energy Plant is a flexible load. The CEP represents only a fraction of aggregate campus loads (~20% of current carbon and financial footprints). Ø Impact of optimization on aggregate campus emissions and financials can only be limited. There is high value for thermal storage if load shaping has high value. Ø Demand Charge Management. Ø Carbon Management with high daily variability in carbon intensity of grid generation mix. 34

35 Quick Recap A facility like SESI used to be a point on a map. Ø Now a black box that gives hourly behavior as a function of price (price loosely defined). Who is the target? 1. Manager for a facility like SESI 2. ISO/TSO 3. Policy maker 35

36 Current work #2 TOWARDS MORE ADVANCED DEMAND RESPONSE

37 Capacity Bidding Programs Load-following through dynamic pricing is limited by Demand Charge. Ø Capacity Bidding Programs? Monthly capacity bid for DR program is made. J-1 Start of new billing cycle for demand charge; loads are observed. J+X J-5 Energy price for next 24 hours is H,J,M A DR event is called. H+720, known, temperature forecast ready. J+30, M+1 Schematic for CBP timeline.

38 Capacity Bidding Programs Participating in a CBP requires adequate treatment of risk. Designed to avoid the need for more generation in periods of peak demand. Ø The CBP transfers the burden of uncertainty associated with net load forecasting from the utility to the customer. Almost functions as the reverse of a Demand Charge ($ value as well). Hope to start pilot this May.

39 Capacity Bidding Problem formulation Scheduling problem (for one month): where and or

40 Conclusions so far The CBP shifts some responsibility for supply/demand balancing to the customer. Capacity mechanisms are likely to play an increasing role in renewable energy systems. Looking for metrics to define flexibility. Ø Towards a DR supply curve?

41 Future work

42 Future work plans and ideas Temperatures, Pipe system, Seasonal storage, Exploiting building-level controls, Large-scale EV charging, Operational supply curve for various types of DR, Other campuses (capacity expansion).

43 Threads for this talk 1. District heating and cooling options should be considered more. 2. Problems with long-term objectives but short-term decisions. 3. Derive generally applicable insights from real-life systems. 4. What do you think?

44 Acknowledgements Global Climate and Energy Project, State Grid Corporation of China Bits & Watts Graduate Fellowship.

45 Q&A