Document Type

Thesis - Open Access

Award Date


Degree Name

Master of Science (MS)


Electrical Engineering and Computer Science

First Advisor

Zhen Ni


Adaptive dynamic programming, Experience replay, Prioritized experience replay, Reinforcement learning, Shunt active filters, Virtual synchronous machines


Adaptive dynamic programming (ADP) controller is a powerful neural network based control technique that has been investigated, designed, and tested in a wide range of applications for solving optimal control problems in complex systems. The performance of ADP controller is usually obtained by long training periods because the data usage efficiency is low as it discards the samples once used. Experience replay is a powerful technique showing potential to accelerate the training process of learning and control. However, its existing design can not be directly used for model-free ADP design, because it focuses on the forward temporal difference (TD) information (e.g., state-action pair) between the current time step and the future time step, and will need a model network for future information prediction. Uniform random sampling again used for experience replay, is not an efficient technique to learn. Prioritized experience replay (PER) presents important transitions more frequently and has proven to be efficient in the learning process. In order to solve long training periods of ADP controller, the first goal of this thesis is to avoid the usage of model network or identifier of the system. Specifically, the experience tuple is designed with one step backward state-action information and the TD can be achieved by a previous time step and a current time step. The proposed approach is tested for two case studies: cart-pole and triple-link pendulum balancing tasks. The proposed approach improved the required average trial to succeed by 26.5% for cart-pole and 43% for triple-link. The second goal of this thesis is to integrate the efficient learning capability of PER into ADP. The detailed theoretical analysis is presented in order to verify the stability of the proposed control technique. The proposed approach improved the required average trial to succeed compared to traditional ADP controller by 60.56% for cart-pole and 56.89% for triple-link balancing tasks. The final goal of this thesis is to validate ADP controller in smart grid to improve current control performance of virtual synchronous machine (VSM) at sudden load changes and a single line to ground fault and reduce harmonics in shunt active filters (SAF) during different loading conditions. The ADP controller produced the fastest response time, low overshoot and in general, the best performance in comparison to the traditional current controller. In SAF, ADP controller reduced total harmonic distortion (THD) of the source current by an average of 18.41% compared to a traditional current controller alone.

Library of Congress Subject Headings

Intelligent control systems.
Dynamic programming.
Adaptive control systems.
System design.


Includes bibliographical references (126-146)



Number of Pages



South Dakota State University


In Copyright - Educational Use Permitted