在現今電力電子與再生能源技術快速發展的時代，多電平逆變器（Multilevel
Inverter, MLI）因其具備高輸出電壓品質、低諧波失真及低開關損耗等特性，已成為
高功率電能轉換應用中不可或缺的關鍵技術。然而，選擇性諧波消除脈波寬度調變
（Selective Harmonic Elimination Pulse Width Modulation, SHEPWM）控制策略涉及高度非
線性的多重方程求解，使得即時運算與硬體實現面臨挑戰。
本論文針對十一階級聯式H橋多電平逆變器（CascadedH-Bridge,CHB-MLI），提出
一套結合改良型粒子群演算法（ModifiedParticleSwarmOptimization, MPSO）與深度神
經網路（DeepNeuralNetwork,DNN）模型之SHEPWM控制方法。研究首先利用MPSO
求得不同調變指數(ModulationIndex) 下的最優開關角度，作為DNN模型的訓練資料，
以實現開關角度預測模型，並以VerilogHDL完成整體硬體架構設計。
本文的研究過程以Python進行演算法驗證，後續完成以Cell-based為基礎的完整晶
片設計流程，並透過FPGA進行系統驗證。模擬與實驗結果顯示，本方法能有效抑制
低次諧波、提升輸出波形品質，並顯著降低濾波器尺寸與運算延遲，展現其於多電平
逆變器即時控制應用中的可行性與優勢。

In recent years, multilevel inverters (MLIs) have played a crucial role in modern power
electronics and renewable energy systems due to their advantages of high output voltage qual
ity, low harmonic distortion, and reduced switching losses. However, the Selective Harmonic
Elimination Pulse Width Modulation (SHEPWM) technique involves solving highly nonlin
ear transcendental equations, which poses significant challenges for real-time computation and
hardware implementation.
This thesis focuses on an eleven-level Cascaded H-Bridge Multilevel Inverter (CHB-MLI)
and proposes a SHEPWM control method that combines the Modified Particle Swarm Opti
mization (MPSO) algorithm with a Deep Neural Network (DNN). The MPSO algorithm is first
employed to obtain optimal switching angles under different modulation indices, which are then
used to train the DNNmodelforfast, non-iterative prediction of switching angles. The proposed
DNN adopts the ReLU and piecewise linear approximation (PLAN) Sigmoid activation func
tions, and the overall hardware architecture is implemented using Verilog HDL.
Theproposedmethodisverified through algorithm validation in Python, followed by a com
plete cell-based chip design flow and FPGA implementation. Simulation and experimental re
sults demonstrate that the proposed approach effectively suppresses low-order harmonics, im
proves output waveform quality, and significantly reduces filter size and computation latency,
proving its feasibility and advantages in real-time multilevel inverter control applications.

[1] J. Rodriguez, J.-S. Lai, and F. Z. Peng, “Multilevel inverters: a survey of topologies, con
trols, and applications,” IEEE Transactions on industrial electronics, vol. 49, no. 4, pp.
724–738, 2002.
[2] Y.-S. Lai and F.-S. Shyu, “Topology for hybrid multilevel inverter,” IEE Proceedings
Electric Power Applications, vol. 149, no. 6, pp. 449–458, 2002.
[3] I.Colak, E.Kabalci, andR.Bayindir, “Reviewofmultilevelvoltagesourceinvertertopolo
gies and control schemes,” Energy conversion and management, vol. 52, no. 2, pp. 1114
1128, 2011.
[4] S. Debnath, J. Qin, B. Bahrani, M. Saeedifard, and P. Barbosa, “Operation, control, and
applications of the modular multilevel converter: A review,” IEEE transactions on power
electronics, vol. 30, no. 1, pp. 37–53, 2014.
[5] M. S. Dahidah, G. Konstantinou, and V. G. Agelidis, “A review of multilevel selective
harmonic elimination pwm: formulations, solving algorithms, implementation and appli
cations,” IEEE Transactions on Power Electronics, vol. 30, no. 8, pp. 4091–4106, 2014.
[6] W. Fei, X. Du, and B. Wu, “A generalized half-wave symmetry she-pwm formulation for
multilevel voltage inverters,” IEEE Transactions on Industrial Electronics, vol. 57, no. 9,
pp. 3030–3038, 2009.
[7] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceedings of ICNN’95
international conference on neural networks, vol. 4. ieee, 1995, pp. 1942–1948.
[8] D. Tian and Z. Shi, “Mpso: Modified particle swarm optimization and its applications,”
Swarm and evolutionary computation, vol. 41, pp. 49–68, 2018.
[9] R. K. Ursem, “Diversity-guided evolutionary algorithms,” in International conference on
parallel problem solving from nature. Springer, 2002, pp. 462–471.
[10] M. Leshno, V. Y. Lin, A. Pinkus, and S. Schocken, “Multilayer feedforward networks
withanonpolynomialactivationfunctioncanapproximateanyfunction,”Neuralnetworks,
vol. 6, no. 6, pp. 861–867, 1993.
[11] F. Agostinelli, M. Hoffman, P. Sadowski, and P. Baldi, “Learning activation functions to
improve deep neural networks,” arXiv preprint arXiv:1412.6830, 2014.
[12] S.-i. Amari, “Backpropagation and stochastic gradient descent method,” Neurocomputing,
vol. 5, no. 4-5, pp. 185–196, 1993.
[13] R. Hecht-Nielsen, “Theory of the backpropagation neural network,” in Neural networks
for perception. Elsevier, 1992, pp. 65–93.
[14] P. J. Werbos, “Backpropagation through time: what it does and how to do it,” Proceedings
of the IEEE, vol. 78, no. 10, pp. 1550–1560, 2002.
[15] D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning representations by back
propagating errors,” nature, vol. 323, no. 6088, pp. 533–536, 1986.
[16] T. Yu and H. Zhu, “Hyper-parameter optimization: A review of algorithms and applica
tions,” arXiv preprint arXiv:2003.05689, 2020.
[17] A. Tisan, S. Oniga, D. Mic, and A. Buchman, “Digital implementation of the sigmoid
function for fpga circuits,” Acta Technica Napocensis Electronics and Telecommunica
tions, vol. 50, no. 2, p. 6, 2009