An Energy Control Center for a Network of Distributed Generators

1
An Energy Control Center for a Network of
Distributed Generators

• By Etienne Dupuis
• Supervisor Dr. J.H Taylor

2
Topics

• Power systems and Distributed Generators.
• A control center for distributed generators.
• Renewable energy and forecasting.
• Control center algorithms.
• A new Unit Commitment Algorithm.

3
Power Systems

• Large units and high voltage transmission lines.
• Fifteen units in New Brunswick.
• Installed capacity of 3948 MW.
• 6665 km of transmission lines.

Source www.nbpower.com
4
Distributed Generators

• Ratings from tens of kW to a few MW.
• Both renewable and non-renewable technologies are
  available.
• Some units can be situated close to the customer.
• The NIMBY factor and deregulation favor
  distributed generation.

5
ASPRIs Distributed Generators

• Renewable
• Wind Turbine, 30kW
• Dorchester, NB
• Small Hydro, 25kW
• York Mills, NB

• Non-Renewable
• Fuel Cell,
• Saint-Johns, NL
• Micro-gas turbine,
• Fredericton, NB

6
This Project

• Aggregate the controls of distributed generators.
• Provides a basis to include distributed
  generators in the economic dispatch and ancillary
  service dispatch.

Integrating distributed generators
7
Optimization of DG generation

• Forecasting / Bidding
• Generation Scheduling
• Economic Dispatch
• Unit Commitment
• Ancillary services
• Hydro-Thermal Scheduling

8
Forecasting

• Forecasting wind speed improves the scheduling of
  our power system.
• Forecasting the hydraulic head of hydro units
  improves the hydro-thermal schedule.
• Classical time series forecasting, neural
  networks and meteorological methods will be
  investigated in this project.

9
Time Series Analysis

• The correlation between measurements is used to
  estimate an ARMA model to fit the data.
• Extensions are available for handling seasonal
  series.

Lag 1 and lag 2 correlations for MA models
10
Neural Networks

• Feed-forward neural networks are composed of an
  input, hidden and output layer.
• The inputs are weighted, summed and passed trough
  a non-linear function before being used as input
  to the next layer.
• The weights of the network are adjusted so that
  the output of the network approximates that of
  the system.

11
Meteorology

• Forecasts from environment Canada are a start.
• Numerical weather prediction provide increased
  lead times.
• Ensemble forecasts are an interesting way to
  estimate confidence in the forecast.

Source weatheroffice.ec.gc.ca
12
Economic Dispatch

• Cost of thermal generators are expressed as
  quadratic functions of their power output.
• The optimization is constrained by the physical
  limits of the generators and the need to meet the
  power demand.

Cost Contours for 2 generators
13
Unit Commitment

• Another optimization problem.
• Which thermal units to assign to meet demand and
  minimize cost.
• Unlike the Economic Dispatch problem, Unit
  Commitment is hard!
• To find the optimal solution for N generators, we
  could have to perform an economic dispatch 2N
  times.

14
Lagrangian Relaxation

• A commonly used solution to the Unit Commitment
  problem.
• The solution is iterative and determines which
  unit to commit based on the profitability for a
  given marginal cost of power.
• This method does not always yield an optimal
  solution.

15
Another problem with Lagrangian Relaxation

• Identical units are an irritant because they get
  committed by the algorithm at the same time.
• If distributed generation becomes widely used,
  identical units are bound to turn up.

16
Solving the UC by sintering

• Use algebra to obtain the quadratic parameters of
  the optimal path.
• We obtain the optimal commitments for the two
  units over their full feasible range.

17
Algorithm output

• We end up with 5 quadratic curves which represent
  the lowest cost as a function of power for these
  units.
• The good news is we can keep going!

18
Cost Curve Sintering for 20 units

• The sintering method returned the same commitment
  vector as CE 100 of the time.
• Sintering ran in 1.437 seconds, it took 220
  minutes to run CE at P5 resolution.
• Sintering yields more info.

19
Error Analysis

• The difference between the costs returned by the
  two methods is introduced by the tolerance of the
  CE method.
• The circles in the stem plot are x10 accuracy.

20
Computation time for sintering

• Sintering took 69 minutes
• Predicted time for CE, 6.41050
• YEARS!

21
Solving the UC for multiple hours

• Start-up costs make solving the UC more
  difficult.
• Sintering is a good match to Dynamic Programming,
  because it provides a list of good unit
  combinations.

22
Including transition costs

• The movie on the right shows the effect of
  progressively adding faster states to the
  sintered commitments.
• DP is run over 100 time instances, with a
  sinusoidal forcing function.

23
Sintered curves for bidding

• ProfitP( / W)-Cost.
• The degree of uncertainty of the wind forecast
  could be used alongside the profit curve to make
  a bid into the power pool.

24
Hydro-thermal scheduling

• Power from hydro plants is a function of flow,
  hydraulic head and turbine efficiency.
• Constraints on the drawdown and storage can be
  significant factors.
• Plants can be coupled hydraulically in series or
  parallel.
• Solutions are specific to the hydro system under
  study.

25
Hydrological Forecasting

• Real time hydrometric data is available from
  Water Survey of Canada.
• Water levels are a function of precipitation,
  soil saturation, vegetation and other factors.

Source Environment Canada
26
References

• Power Generation Operation Control, Allen J.
  Wood, Bruce F. Wollenberg.
• Time Series Analysis Forecasting and Control
  third edition, George E.P. Box, Gwilym M.
  Jenkins, Gregory C. Reinsel.
• Artificial Neural Networks, Forecasting Time
  Series, V. Rao Vemuri, Robert D. Rogers.
• Wind Power Prediction using Ensembles, Riso
  institute.
• Unit Commitment A bibliographical Survey,
  Narayana Prasad Padhy.

27
Significance

• Power system scheduling at the distributed
  generator level enables this technology.
• Generator aggregation could lead to a new
  solution to the unit commitment problem.
• Renewable generators make the specific goals of
  ASPRI coherent with those of utilities
  incorporating wind power to their system.

28
Questions ?