Enhanced Power System Responsiveness through Load Control

Transcription

1 Enhanced Power System Responsiveness through Load Control Ian A. Hiskens Vennema Professor of Engineering Professor, Electrical Engineering and Computer Science Acknowledge: Duncan Callaway, Univ of California, Berkeley Zhongjing Ma, Beijing Institute of Technology Scott Backhaus, Misha Chertkov, Nic Sinitsyn, Los Alamos National Lab Johanna Mathieu, Soumya Kundu, Mads Almassalkhi, Ian Beil, Univ of Michigan IEEE SmartGridComm Venice, Italy November 4, 2014

2 Load control overview Non-disruptive load control is standard in communications so why not in electricity delivery. Load control offers enhanced system responsiveness at the local, regional and system-wide levels. Frequency regulation: primary (droop) and secondary (AGC) control. Load shifting. Tracking renewable generation to reduce variability. Local congestion management (distribution). Overload, voltage collapse prevention (transmission). The design of controls that interact appropriately across the three levels of the grid, and with other controllable devices, is still very open. Incorporating load control (and distributed generation) into aggregate distribution network models is extremely challenging. But absolutely necessary for meaningful assessment of system dynamic performance. 2/35

3 Load control Competing objectives: Local control objective, e.g., Maintain temperature close to setpoint. Deliver required charge to PEV by specified time. System service, e.g., Balance renewable generation output. Load control strategies must be consistent with the legacy system operating philosophy. Centralized control of large numbers of loads is impractical. 3/35

4 Advanced metering infrastructure (AMI) Latency of up to 45 minutes. Having the infrastructure is one thing knowing what to do with it is another story. 4/35

5 Presentation structure 1. Motivation 2. Decentralized control/coordination 3. Load aggregation 4. Overload management in distribution networks 5. Building control 6. Final remarks 5/35

6 Naïve strategies PEV charging control strategies will be vitally important for ensuring largescale adoption of plug-in EVs does not cause generation scheduling problems. MISO summer load demand Time-based charging strategy 6/35

7 What can go wrong? Price-based strategy for charging electric vehicles: charge when price falls below a lower threshold, cease charging when price rises above an upper threshold. 7/35

8 Presentation structure 1. Motivation 2. Decentralized control/coordination 3. Load aggregation 4. Overload management in distribution networks 5. Building control 6. Final remarks 8/35

9 Decentralized control of PEVs The following charging negotiation procedure takes place prior to the actual charging interval: 1) The utility broadcasts base demand to PEVs. 2) Each PEV proposes its optimal strategy with respect to a common aggregate PEV demand broadcast by the utility: 3) The utility collects all the individual strategies proposed in (2), and updates the aggregate PEV demand accordingly. This updated aggregate demand is rebroadcast to all PEVs. 4) Repeat (2) and (3) until the optimal strategies proposed by the PEVs no longer change. 9/35

10 Decentralized control of PEVs Theorem: A collection of charging strategies for an infinite population of PEVs is a Nash equilibrium, if (i) minimizes the cost function, with respect to a fixed, and (ii), for all, i.e., can be reproduced by averaging the individual optimal control trajectories of all PEV agents. Mean field result. Under certain mild assumptions, the Nash equilibrium: Exists, by Brouwer s fixed point theorem. Unique, by the contraction mapping theorem. Can be obtained by a convergent iterative process. Satisfies a valley filling property which gives globally optimal cost. 10/35

11 Decentralized charging control Damping is zero: 11/35

12 Decentralized charging control Damping is a small positive value: 12/35

13 Auction-based games for PEV charging Vehicles submit bid profiles: price: maximum energy request: Incentive compatibility holds under the progressive second price auction mechanism. Cross-elasticity between charging times complicates analysis. The marginal valuation at each time-step is dependent upon the energy request at that time and the total energy request over the entire charging horizon. Main result: The efficient bid profile over the charging horizon is a Nash equilibrium of the underlying game. No PEV can improve their payoff (utility minus payment) by unilaterally changing their bid profile. 13/35

14 Presentation structure 1. Motivation 2. Decentralized control/coordination 3. Load aggregation 4. Overload management in distribution networks 5. Building control 6. Final remarks 14/35

15 Hysteresis-based load control Consider thermostatically controlled loads (TCLs). For cooling loads: Regions: Blue loads are in the off state. Red loads are in the on state. Control strategy: Increase load by lowering set-point. Decrease load by raising set-point. Temperature 15/35

16 Load control: tracking wind variations Controlling 60,000 AC loads to follow wind variations. From Callaway: Tapping the energy storage potential in electric loads. 16/35

17 PEV charging control: tracking Hysteresis-based load control can be extended to loads that require a certain amount of energy, but have some flexibility in when they receive that energy. PEV charging, refrigeration, dehumidifiers, pool pumps,... Tracking wind variability 17/35

18 State-space modeling of hysteretic control State-space modelling results in a nonlinear hybrid dynamical system. Nonlinear because states and inputs multiple together. Hybrid due to the influence of rapidly changing inputs. Period-4 orbit, Input period = 12.4 min Period-3 orbit, Input period = 15.6 min 18/35

19 Bifurcation diagram Analysis of period-adding bifurcations was achieved using the Poincare map: where is the input period. Varying the input period gave the bifurcation diagram: 19/35

20 Chaos Periodic behaviour is separated by regions of aperiodic response. The accuracy of the state-space model reduces dramatically within the aperiodic regions. This suggests high sensitivity and is indicative of chaos. Input period = 28.8 min 20/35

21 Presentation structure 1. Motivation 2. Decentralized control/coordination 3. Load aggregation 4. Overload management in distribution networks 5. Building control 6. Final remarks 21/35

22 Distribution network overloads Plug-in electric vehicles: Charging: Ampères per PEV Typical household connection 10-20A 240V) Scenario: en masse over-night charging Uncoordinated PEV charging = distribution network overload! 22/35

23 Dynamical model of PEV charging Distributed model predictive control (MPC) used to regulate power through the transformer. Communication between the loads and the transformer. 23/35

24 Iterative, distributed solution Open-loop coordinated charging, driven by pseudo-price PRICE MANAGER (subgradient step) PEV SCHEDULER (local optimization) Centralized solution is recovered for. 24/35

25 Case study: temp and load profiles Temperature within limits, despite 5% inaccuracy in background load 25/35

26 AIMD control algorithm Additive-increase Multiplicative-Decrease (AIMD) algorithm. Well-studied method for allocating global resources to distributed agents. Each load steps up charging at a rate α until the global threshold is reached, at which time each charger reduces its charging rate by a factor β. In steady-state, this results in equal charging rate for each participant. 26/35

27 AIMD control on a distribution feeder Transformer temperature Individual PEV load profiles 27/35

28 Presentation structure 1. Motivation 2. Decentralized control/coordination 3. Load aggregation 4. Overload management in distribution networks 5. Building control 6. Final remarks 28/35

29 Fast demand response: building HVAC Typical HVAC system for a large commercial building. 29/35

30 Demand response test results 30/35

31 Efficiency Average efficiency over a range of tests was 46%. 31/35

32 Presentation structure 1. Motivation 2. Decentralized control/coordination 3. Load aggregation 4. Overload management in distribution networks 5. Building control 6. Final remarks 32/35

33 Load control (resource) availability The amount of load that is available for control must be measured or estimated. Load disaggregation. For example, what is the current population of controllable TCLs or PEVs or pool pumps? Typical data: Frequent (seconds) noisy measurements of feeder/transformer aggregate power demand. Infrequent (minutes) information from individual loads. Approaches: State estimation based on Kalman filtering. Matrix completion concepts. Uncertainty is inevitable. Communications requirements and limitations introduce cost/capability tradeoffs. 33/35

34 Analysis and control Multiple (conflicting) control objectives. Hierarchical control structures with higher- and lower-level objectives that may conflict at times, e.g., frequency regulation may require a load increase but that would cause an overload of distribution transformers. Analysis. Dynamic behaviour: load synchronization, effects on grid dynamics. Games, interactions between decentralized decisionmakers. Effects on load efficiency. Modelling Analysis requirements. Control simplification. 34/35

35 Conclusions Significant actuation can be achieved through coordinated non-disruptive control of highly distributed loads. Technical issues: control structure, nonlinearity (bifurcations), latency, interoperability, data security, What is the (financial) value of load control? To utilities, aggregators, consumers? Social issues: incentives/rewards for consumers to participate in (non-disruptive) fast-acting, demand response schemes. Should consumers be financially rewarded, or required to participate as a condition of service? (The latter applies in the case of cell phone service.) Modelling of load behaviour has always been challenging. Distributed generation and controllable loads provide further nontrivial complications. 35/35

36 Primary references 1. D.S. Callaway, Tapping the Energy Storage Potential in Electric Loads to Deliver Load Following and Regulation, with Application to Wind Energy, Energy Conversion and Management, Vol. 50, 2009, pp D.S. Callaway and I.A. Hiskens, Achieving Controllability of Electric Loads, Proceedings of the IEEE, Vol. 99, No. 1, January 2011, pp R. Hermans, M. Almassalkhi and I.A. Hiskens, Incentive-based Coordinated Charging Control of Plug-in Electric Vehicles at the Distribution-Transformer Level, Proceedings of the American Control Conference, Montreal, Canada, June S. Kundu and I.A. Hiskens, State-space Modelling of Hysteresis-based Control Schemes, Proceedings or the European Control Conference, Zurich, Switzerland, July 2013, pp S. Kundu, N. Sinitsyn, S. Backhaus and I.A. Hiskens, Modeling and Control of Thermostatically Controlled Loads, Proceedings of the 17th Power Systems Computation Conference, Stockholm, Sweden, August Z. Ma, D.S. Callaway and I.A. Hiskens, Decentralized Charging Control of Large Populations of Plug-in Electric Vehicles, IEEE Transactions on Control Systems Technology, Vol. 21, No. 1, January 2013, pp Z. Ma, S. Zou, X. Liu and I.A. Hiskens, An Efficient Game for Coordinating Electric Vehicle Charging, submitted to IEEE Transactions on Automatic Control. 8. S. Kundu and I.A. Hiskens, Nonlinear Dynamics of Hysteresis-Based Load Controls, Proceedings of the 19 th IFAC World Congress, Cape Town, South Africa, August I. Beil, I.A. Hiskens and S. Backhaus, Round-Trip Efficiency of Fast Demand Response in a Large Commercial Air Conditioner, submitted to Energy and Buildings. 10. G.S. Ledva, E. Vrettos, S. Mastellone, G. Andersson and J.L. Mathieu, Applying Networked Estimation and Control Algorithms to Address Communication Bandwidth Limitations and Latencies in Demand Response, to appear Proceedings of the 48 th Hawaii International Conference on System Sciences, January /35