Author Contributions
Conceptualization, A.S.H. and A.M.; methodology, A.S.H., K.B. and A.M.; software, A.M. and K.B.; validation, A.S.H., A.M. and K.B.; formal analysis, A.S.H. and K.B.; investigation, A.S.H., A.M. and K.B.; resources, A.M. and K.B.; data curation, A.M.; writing—original draft preparation, A.S.H. and K.B.; writing—review and editing, A.S.H., A.M. and K.B.; visualization, A.S.H., K.B. and A.M.; supervision, K.B. and A.M.; project administration, K.B. and A.M.; funding acquisition, K.B. All authors have read and agreed to the published version of the manuscript.
Figure 1.
35th-scale Centipod wave energy converter by Dehlsen Associates, LLC.
Figure 1.
35th-scale Centipod wave energy converter by Dehlsen Associates, LLC.
Figure 2.
Degrees of freedom for dynamic modeling of Centipod WEC: (a) baseline configuration and (b) model with mooring lines.
Figure 2.
Degrees of freedom for dynamic modeling of Centipod WEC: (a) baseline configuration and (b) model with mooring lines.
Figure 3.
Normalized radiation damping of the Centipod WEC: (a) surge axis, (b) heave axis, and (c) pitch axis.
Figure 3.
Normalized radiation damping of the Centipod WEC: (a) surge axis, (b) heave axis, and (c) pitch axis.
Figure 4.
Normalized added mass of the Centipod WEC: (a) surge axis, (b) heave axis, and (c) pitch axis.
Figure 4.
Normalized added mass of the Centipod WEC: (a) surge axis, (b) heave axis, and (c) pitch axis.
Figure 5.
Normalized excitation amplitude of the Centipod WEC: (a) surge axis, (b) heave axis, and (c) pitch axis.
Figure 5.
Normalized excitation amplitude of the Centipod WEC: (a) surge axis, (b) heave axis, and (c) pitch axis.
Figure 6.
Deep network design for RL actor and critic in MATLAB for the heave and pitch PTOs: (a) deep network for the critic and (b) deep network for the actor.
Figure 6.
Deep network design for RL actor and critic in MATLAB for the heave and pitch PTOs: (a) deep network for the critic and (b) deep network for the actor.
Figure 7.
WECSim digital twin of 3-pod Centipod WEC with PTOs in heave and pitch axes.
Figure 7.
WECSim digital twin of 3-pod Centipod WEC with PTOs in heave and pitch axes.
Figure 8.
Wave spectrum of the mean of a cluster of sea states for WEC-Sim simulation.
Figure 8.
Wave spectrum of the mean of a cluster of sea states for WEC-Sim simulation.
Figure 9.
RL agent training through custom Simulink environment interfaced to WEC Emulator via Ethernet UDP.
Figure 9.
RL agent training through custom Simulink environment interfaced to WEC Emulator via Ethernet UDP.
Figure 10.
RL-DDPG controller for 2-DoF 3-pod CENTIPOD WEC.
Figure 10.
RL-DDPG controller for 2-DoF 3-pod CENTIPOD WEC.
Figure 11.
RL DDPG agent training for heave PTO: (a) training with unconstrained PTO force and (b) training with constrained PTO force, .
Figure 11.
RL DDPG agent training for heave PTO: (a) training with unconstrained PTO force and (b) training with constrained PTO force, .
Figure 12.
RL DDPG agent training for pitch PTO: (a) training with unconstrained PTO force and (b) training with constrained PTO force, .
Figure 12.
RL DDPG agent training for pitch PTO: (a) training with unconstrained PTO force and (b) training with constrained PTO force, .
Figure 13.
Nonlinear MPC controller for digital twin of three-pod Centipod device.
Figure 13.
Nonlinear MPC controller for digital twin of three-pod Centipod device.
Figure 14.
Heave average electrical power output per pod with linear waves enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 14.
Heave average electrical power output per pod with linear waves enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 15.
Pitch average electrical power output per pod with linear waves enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 15.
Pitch average electrical power output per pod with linear waves enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 16.
Heave average electrical power output per pod with nonlinear buoyancy and Froude–Krylov excitations enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 16.
Heave average electrical power output per pod with nonlinear buoyancy and Froude–Krylov excitations enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 17.
Pitch average electrical power output per pod with nonlinear buoyancy and Froude–Krylov excitations enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 17.
Pitch average electrical power output per pod with nonlinear buoyancy and Froude–Krylov excitations enabled in WECSim: (a) with unconstrained PTO force and (b) with constrained PTO force, .
Figure 18.
Heave instantaneous electrical power output per pod with unconstrained PTO force: (a) with linear wave conditions in WECSim and (b) with nonlinear buoyancy and Froude−Krylov excitations enabled in WECSim.
Figure 18.
Heave instantaneous electrical power output per pod with unconstrained PTO force: (a) with linear wave conditions in WECSim and (b) with nonlinear buoyancy and Froude−Krylov excitations enabled in WECSim.
Figure 19.
Heave PTO force output of NMPC and RL controllers with nonlinear buoyancy and Froude−Krylov excitations enabled in WECSim.
Figure 19.
Heave PTO force output of NMPC and RL controllers with nonlinear buoyancy and Froude−Krylov excitations enabled in WECSim.
Table 1.
Nomenclature for WEC dynamics.
Table 1.
Nomenclature for WEC dynamics.
Symbol | Unit | Description |
---|
| , rad/s | Generalized velocity |
| | Generalized displacement |
| | Auxiliary state variable for radiation force dynamics |
| | Force from wave radiation in axis due to velocity in axis |
| | Buoyancy restoring force |
| | Fluid damping force |
| | Force due to wave excitation |
| | Power take-off actuation force |
| | Pod mass |
| | Infinite frequency generalized radiation added mass in axis due to acceleration in axis. |
| | Buoyancy restoring constant |
|
| Fluid quadratic damping constant |
| | Added mass due to wave radiation in axis due to acceleration in axis |
|
| Radiation damping due to wave radiation in axis due to velocity in axis |
| | Impulse response function for wave radiation |
| ,
| Mechanical impedance |
| | Gravity constant |
| | Water density |
Table 2.
Nomenclature for nonlinear MPC formulation.
Table 2.
Nomenclature for nonlinear MPC formulation.
Symbol | Description |
---|
| Set of system variables |
| Prediction horizon |
| State vector of WEC dynamics |
| Manipulated variable vector, PTO force/torque () |
| Finite-horizon terminal cost penalty |
| Polynomial of system variables |
| Constant weighting matrices |
| Constant column vectors |
| Column vectors of nonlinear functions of state variables |
| Column vectors of nonlinear functions of state variables |
| Excitation force disturbance vector, () |
| Some real number |
Table 3.
WEC system parameters.
Table 3.
WEC system parameters.
Parameter | Value |
---|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
Water depth | |
Pod volume | |
Pod immersed volume | |
Mooring line length | |
Mooring line type | Chain |
Mooring no of lines | |
Mooring line diameter | |
Mooring mass density in air | |
Mooring damping ratio | |
Mooring stiffness | |
Mooring transverse drag coefficient | |
Mooring transverse added mass coefficient | |
Mooring tangential drag coefficient | |
Mooring tangential added mass coefficient | |
Table 4.
RL DDPG agent properties for heave and pitch control.
Table 4.
RL DDPG agent properties for heave and pitch control.
RL Agent Options | Value |
---|
Sample time | 0.1 |
Target smooth factor | 1 × 10−6 |
Discount factor | 0.95 |
Mini batch size | 512 |
Length of experience buffer | 1 × 106 |
Noise variance | 0.3 |
Variance decay rate | 1 × 10−5 |
Target update frequency | 1 |
Table 5.
RL critic properties for heave and pitch control.
Table 5.
RL critic properties for heave and pitch control.
RL Critic Options | Value |
---|
Representation | RL Q-value |
Learn rate | 0.1 |
Gradient threshold | inf |
Action feature input layer size | 1 |
State feature input layer size | 1 |
Action and critic fully connected layer-1 size (for unconstrained ) | 64 |
Action and critic fully connected layer-2 size (for constrained ) | 50 |
Critic common fully connected layer-2 (FC1) size | 1 |
Table 6.
RL actor properties for heave and pitch control.
Table 6.
RL actor properties for heave and pitch control.
RL Critic Options | Value |
---|
Representation | RL deterministic |
Learn rate | 0.1 |
Gradient threshold | inf |
Optimizer momentum | 0.95 |
State feature input layer size | 1 |
Actor fully connected layer-1 size (unconstrained ) | 32 (heave), 16 (pitch) |
Actor fully connected layer-1 size (constrained ) | 25 (heave), 16 (pitch) |
Actor fully connected layer-2 size | 1 |
Optimizer for pitch (constrained or unconstrained ) | Root mean square propagation (RMS-Prop) |
Optimizer for heave (for constrained ) | Root mean square propagation (RMS-Prop) |
Optimizer for heave (for unconstrained ) | Stochastic gradient descent with momentum (SGDM) |
Table 7.
Mean of the cluster of sea states for Centipod digital twin simulation in WECSim.
Table 7.
Mean of the cluster of sea states for Centipod digital twin simulation in WECSim.
Sea-State Parameter in WECSim | Value |
---|
Significant wave height [m] | 2.5 |
Peak wave period [s] | 7.35 |
Spectrum type of ocean waves | Pierson Moskowitz (PM) |
Class of waves | Irregular |
Table 8.
Moving mean of electrical output power [kW] per PTO in WEC-Sim with irregular waves.
Table 8.
Moving mean of electrical output power [kW] per PTO in WEC-Sim with irregular waves.
Controller Algorithm | Linear Wave Conditions | Nonlinear Buoyancy and Froude–Krylov Excitation |
---|
Unconstrained | | Unconstrained | |
---|
NMPC | 17 | 16 | Unstable | 17 |
L-DDPG | 32 | 17 | 85 | 37 |
| Average electrical power [kW] for pitch |
NMPC | 4.50 | 2.80 | Unstable | 3.25 |
RL-DDPG | 5.25 | 3.10 | 6.50 | 3.30 |
Table 9.
Timings stats for NMPC and RL DDPG control.
Table 9.
Timings stats for NMPC and RL DDPG control.
| Training Time [h] | Task Execution Time (TET) [s] |
---|
NMPC heave and pitch combined | - | |
NMPC per DoF | - | |
RL-DDPG heave | | |
RL-DDPG pitch | | |