Simulation,  Modeling,  &  Decision  Science  AMES LABORATORY

Merged Environment for Simulation and

Analysis (MESA)

Project Investigator: Dr. Mark Bryden

Post-doc Research Associate Dr. Paolo Pezzini

Date: 04/12/2018

DOE Award Number DE-AC02-07CH11358

Period of Performance FY2017

Objectives

1. Enabling high efficiency high turn down operation of hybrid power systems

2. Enabling rapid design and analysis of complex power systems

3. Creating on the fly updates for control systems for existing power plants

Traditional Control Strategies

• Model-based design

• Tuned for a specific operating point

Limitations

Characterization of each operating point

• Process instability

• Adaptability cannot be reached without system identification

High efficiency high turn down operation

• Model-free design

• Adaptable on-the-fly

• Reconfigurable on different power plants

• Intelligent

Agent-based Control

• Multi-agents emulates intelligent control

• Agents can coordinate their behavior

• Agents are not limited to a set of models

Multi-agents in building automation

Multi-agents are mainly used for supervisory applications..

Local  Temperature  

Agent

Local  Illumination  

Agent

Local  Air  Quality  Agent

Central  Agent Optimizer• Model  Predictive  Control• Ant  colony  optimization• …

Supervisory  control  (minutes or  hours timescale)

Dynamic  control  (milliseconds timescale)

Multi-agents for dynamic control of energy systems

Goal of Multi-agent Control Strategies

Grouping sensors and actuators in computational agents

Agent  NAgent  1

Sensors

Local  Decision

Actuator……

Sensors

Local  Decision

Actuator

Agent  Selection  Criteria

Random  Probability  of  ActionAgent Selection Criteria- Set of rules to avoid simultaneously actions

Random Probability of action- Emergent behavior in social insects

- Determines frequency of action taken

Resource sharing algorithm

Shared Resource (Blocks) establishes cooperation and sharing

Blocks are a discrete unit of change to an actuator

Exchange of blocks when the current state is not at nominal condition

Block  Repository

Agent  NAgent  1

……

Sensors

Current  State

Actuator

Sensors

Current  State

Actuator

Multi-agents control tested on Hyper

1. Multi-input multi-output (MIMO)

- 2 sensors and 2 actuators

- 2 agents

2. Multi-input single-output (MISO)

- 1 sensor and 2 actuators

- 2 co-worker (homogenous) agents

Turbine Speed

Electric Load

Current State

Block Repository

Agent 1 Blocks

Cathode Mass Flow

Cold Air Bypass

Current State

Agent 2 Blocks

CurrentState

Bleed-air

Turbine Speed

CurrentState

Auxiliary Fuel Valve

Turbine Speed

Block  Repository

Agent 1 Agent 2

Electric  Load Cold-­air  Bypass

Turbine  Speed FC  Cathode  Airflow

40800

40600

40400

40200

40000

39800

39600

39400

39200

39000

38800

38600

38400

38200

38000

Turb

ine

Spe

ed, R

PM

220200180160140120100806040200Time, Sec

100

95

90

85

80

75

70

65

60

55

50

45

40

35

30

Electric Load, kW

0.90

0.85

0.80

0.75

0.70

0.65

0.60

0.55

0.50

0.45

0.40

0.35

0.30

Cat

hode

Mas

s Fl

ow, k

g/s

220200180160140120100806040200Time, Sec

90

85

80

75

70

65

60

55

50

45

40

35

30

Cold A

ir Bypass, %

Turbine Speed

Electric Load

Cathode Mass Flow

Cold Air Bypass

System Response - MIMO Control Command - MIMO System Response - Resource Sharing (BS=0.75, PR=90%) Control Command - Resource Sharing (BS=0.75, PR=90%)

0.90

0.85

0.80

0.75

0.70

0.65

0.60

0.55

0.50

0.45

0.40

0.35

0.30

Cat

hode

Mas

s Fl

ow, k

g/s

28024020016012080400Time, Sec

130

120

110

100

90

80

70

60

50

40

30

Cold A

ir Bypass, %

40800

40600

40400

40200

40000

39800

39600

39400

39200

Turb

ine

Spe

ed, R

PM

28024020016012080400Time, Sec

90

85

80

75

70

65

60

55

50

45

40

35

30

25

Electric Load, kW

Cathode Mass Flow

Cold Air Bypass

Turbine Speed

Electric Load

System Response - MIMO Control Command - MIMO System Response - Resource Sharing (BS=2, PR=90%) Control Command - Resource Sharing (BS=2, PR=90%)

Multi-agents vs Centralized MIMO control

Multi-input single-output (MISO)

Nominal  design

[%]Auxiliary  fuel  valve  range  of  operation  

Bleed-­air  valve  range  of  operation  

Nominal condition:

• Fuel cell thermal heat drives the gas turbine

• Closed valves position to maximize system efficiency

• Valve opening during transient operation

Cold  Air  Bypass

Air  

Exhaust

MixingVolume

Generator

Hot  AirBypass

Bleed  AirBypass

Anode

Cathode

Electrolyte

FuelPre-­Combustor

Fuel

Combustor

AuxiliaryFuel

Co-worker agents algorithm

Asynchronous agent selection criteria

• Saturation limit as agent selection criteria

• 10% Probability of action Current

State

Bleed-air

Turbine Speed

CurrentState

Auxiliary Fuel Valve

Turbine Speed

Block  Repository

Agent 1 Agent 2

Turbine load following – turn down

Turbine Speed Deviation

• In open-loop: 1000 rpm

• Multi-agent: 500 rpm

Engineered systems that are built from, and depend

upon, the seamless integration of computational

algorithms and physical components

National  Science  Foundation  (NSF),  2016,  “Cyber-­‐Physical  Systems,”  program  solicitation  16-­‐549

Rapid design and analysis of complex power systems

NETL’s Cyber-physical System

Hybrid Performance Project (HYPER)

Design and Validation of Advanced Sensors• Advanced Sensing• High Temperature & Harsh Environment

Applications• Smart Sensors• Sensor Placement

Design and Validation of Novel Control Strategies

• Distributed Intelligent Control• Smart Actuation• Efficient Decision Making• Digital twin

Controls update for existing power plants

Technical Issues• Part-load operation

• Increase degradation of components

• Controls are designed for based–load operation

Health Parameter Degradation Profile

J. S. Litt, K. I. Parker, S. Chatterjee, “Adaptive Gas Turbine Engine Control for Deterioration Compensation Due to Aging,”Proceedings of the 16th International Symposium on Airbreathing Engines (ISABE–2003–1056), NASA/TM—2003-­‐212607

Degradation rate is function of• Effective cycles

• Physical age

Degradation  Rate

Effective  Cycles

Online System Identification

• Deviation in the model parameter due to aging of equipment

• Unsupervised real-time learning algorithm to support dynamic control

ProcessInputs,  uk Outputs,  yk

Model Model  Outputs,𝑦"#

Parameters  Update(Recursive  Least  Square)

error

𝒇(𝒖𝒌, 𝝑∗)

∆𝜗∗

Summary

1. Enabling high efficiency high turn down operation of hybrid power systems

2. Enabling rapid design and analysis of complex power systems

3. Creating on the fly updates for control systems for existing power plants

Future work

Co-worker/Homogenous agents

• Fuel cell model in the loop – test reproducibility

• Supercritical CO2 power cycle

Virtual plant and cyber-physical track• ASME Power and Energy Conference

Existing power plants

• Online system identification

Acknowledgements

MESA Team in AmesDr. Peter Finzell

Dan Bell (PhD grad, Iowa State U)

Zach Reinhard (PhD grad, Iowa State U)

Tina Akinyi (PhD grad, Iowa State U)

Wadood Daoud (PhD grad, Iowa State U)

Hyper Team in NETL

Dr. David Tucker

Dr. Larry Shadle

Dr. Farida Nor Harun

Dr. Nana Zhou

Mr. Harry Bonilla