1. Voltage Control of Distribution Network Using an Artificial Intelligence Planning Method Jianing Cao 1 Keith Bell 1 Amanda Coles 2 Andrew Coles 2 1. Department of Electronic and Electrical Engineering 2. Department of Computer and Information Sciences University of Strathclyde, UK Jianing Cao – UK – Session 5 – 1112

2. Frankfurt (Germany), 6-9 June 2011 Background  Distribution Network active control Source: R.A.F. Currie, G.W. Ault, C.E.T. Foote, G.M. Burt, J.R. McDonald Figure 1. Example of a substation with active management facilities

3. Frankfurt (Germany), 6-9 June 2011 Objectives  Improve settings for controllers in a Distribution Network E.g. mechanically switched capacitors (MSC) & tap changing transformers  Minimise control actions & wear-and-tear on equipment  Plan control targets to minimise human intervention  Respect the voltage limits E.g. ± 6% for 33kV/11kV [1] (Case study: ± 5%) [1]: D.A. Roberts, SP Power Systems LTD, 2004, “Network management systems for active distribution networks – a feasibility study”

4. Frankfurt (Germany), 6-9 June 2011 Methodology  Multi-objective Artificial Intelligence planning method [2] – forecast demand and generation for a given period, e.g. a day (re-planning might be needed)  Load flow simulation – Linear sensitivity factors reflecting voltage changes with respect to control actions [2]:K. Bell, A.I. Coles, M. Fox, D. Long, A.J. Smith, 2009, "The Role of AI Planning as a Decision Support Tool in Power Substation Management", AI Communications, IOS Press, vol.22, 37-57.

5. Frankfurt (Germany), 6-9 June 2011 Details Planner objective function [2] –PM: plan metric T: transformer steps –M: MSC switchesLV/HV: low voltage/high voltage –α/β: cost of control/switch action from transformer/MSC –γ/δ: relative “cost” of voltage below 0.95 p.u / above 1.05 p.u [2]:K. Bell, A.I. Coles, M. Fox, D. Long, A.J. Smith, 2009, "The Role of AI Planning as a Decision Support Tool in Power Substation Management", AI Communications, IOS Press, vol.22, 37-57.

6. Frankfurt (Germany), 6-9 June 2011 Existing Planner Figure 2. Overview of the VOLTS system Source: Keith Bell, Andrew Coles, Maria Fox, Derek Long and Amanda Smith Using PDDL (planning domain definition language) 1. A domain file for predicates and actions 2. A problem file for objects, initial states & goal

7. Frankfurt (Germany), 6-9 June 2011 Software integration Network Parameters Planner parameters Derive sensitivity factors Metric-FF (planner) Run sequence of Load Flow Update sensitivity factors Voltages outside limits? Yes No Final plan Control actions/ Control targets Control actions/ Control targets

8. Frankfurt (Germany), 6-9 June 2011 Distribution Network Model Source: AuRA-NMS project

9. Frankfurt (Germany), 6-9 June 2011 Demand data in case study  Assumption of load Constant power factor for each load throughout the day Profiles follow National Grid’s half-hourly metered data  E.g. 30-Oct-2010

10. Frankfurt (Germany), 6-9 June 2011 Generation data in case study  Combined Heat and Power (CHP) Capacity of 4MW  Output maximum power when space & water heating needed  Power factor of unity or 0.8

11. Frankfurt (Germany), 6-9 June 2011 Simulation Process 1: setting base cases  Base case 1:  voltage target of transformer set to 1.0 per unit Run load flows to get tap settings & voltages  Base case 2:  tap position of transformers set to nominal (0) Run load flows to get voltages to compare

12. Frankfurt (Germany), 6-9 June 2011 Simulation Process 2: optimisation  Feed the planner with sensitivity factors & initial conditions from load flow results  Generate new transformer tap settings  In set of load flows, set tap positions according to the planner’s control output  Compare against the base case.

13. Frankfurt (Germany), 6-9 June 2011 Simulation results  Tap settings  Minimum voltage on the network PF1 mode: Base Case 1: transformer tap varied from -3% to -2% Base Case 2: minimum voltage 0.947 per unit at 18:00 Planner’s result: 0.96 per unit PFC/VC mode: Planner suggested no change from base case 2 since voltage is not beyond limits

14. Frankfurt (Germany), 6-9 June 2011 Summary  Conclusion Successful integration between the planner and a load-flow simulator A sequence of control settings were found Achieved required voltage profile with fewer tap changes in planned mode Hence, less wear-and-tear on the equipment

15. Frankfurt (Germany), 6-9 June 2011 Summary  Future development Larger distribution network with more controllers / loads/distributed generators Test another ‘worst case’ scenario – low demand & high generation The planner’s robustness to forecast errors to be tested Acknowledgement: the work described has been funded by EPSRC under research grant EP/D062721 Supergen ‘HiDEF’ programme Thank you for your attention.