基於分數階PID類神經網路之模型預測控制應用於寬輸入電壓範圍直流-直流轉換器

Abstract

提出一種模型預測控制(Model Predictive Control, MPC)系統，以模型預測控制系統結合分數階比例-積分-微分類神經網路(Fractional-Order Proportional-Integral-Derivative Neural Network, FOPIDNN)電壓控制器，應用於直流-直流轉換器，不同於多模式操作控制(Multimode Operation Control, MOC)系統，轉換器在轉換模式切換過渡時的輸出電壓能夠更平滑且穩定。首先，本研究藉由推導出轉換器電路各開關導通狀態的電流斜率，以預測下一個時刻的電感電流、電感充放電電流斜率以及功率開關之導通工作週期(Duty Cycle)，並利用整數階比例-積分-微分(Proportional-Integral-Derivative, PID)電壓控制器對模型預測控制系統提供電感電流參考值做驗證，以確保模型預測控制系統能使轉換器電路之輸出電壓能快速穩定在理想電壓值。為了提高輸出功率之響應速度與穩定度，必須改良影響電感電流參考值的電壓控制器，而分數階比例-積分-微分(Fractional-Order Proportional-Integral-Derivative, FOPID)電壓控制器的控制參數更多，比起整數階PID有更高的控制靈活性，不過控制參數多同時也意味著必須花費額外的時間來試錯。因此，本研究進一步提出一FOPIDNN電壓控制器，以類神經網路(Neural Network, NN)動態調整參數的特性來提升FOPID產生電感電流參考值的精確度，實現快速響應且不受環境干擾地的輸出穩定功率。本研究首先以PSIM軟體模擬來評估所提出之方法之有效性和可行性，接著再應用於實際轉換器電路並以DSP TMS320F28335數位信號處理器實現控制系統。最後，比較四種實現控制系統，包括多模式操作控制系統、基於整數階PID之MPC控制系統、基於FOPID之MPC控制系統與本研究提出的基於FOPIDNN之MPC控制系統。結果顯示與前者方法相比，本研究所提方法應用於寬輸入電壓範圍直流-直流轉換器上，能夠更及時地補償電感電流參考值，除了輸出電壓響應速度更快，輸出電壓在轉換模式切換的過渡，平滑效果也明顯提高許多，有效輸出穩定的輸出功率。This paper proposes a Model Predictive Control (MPC) system for DC-DC converters, integrating a Fractional-Order Proportional-Integral-Derivative Neural Network (FOPIDNN) voltage controller. Unlike conventional Multimode Operation Control (MOC) systems, this approach ensures smoother and more stable output voltage transitions during mode switching. The proposed method involves deriving the current slopes for various switch states within the converter circuit to predict the inductor current, the inductor's charging and discharging slopes, and the duty cycle of the power switches for the next time step. A Proportional-Integral-Derivative (PID) voltage controller provides a reference inductor current to validate the MPC’s performance, ensuring the output voltage rapidly stabilizes at the desired level. To improve response speed and stability, this study enhances the voltage controller that generates the inductor current reference. The Fractional-Order PID (FOPID) voltage controller offers more control flexibility due to its additional tuning parameters compared to traditional PID controller, though these extra parameters require additional time for tuning. To address this, the study introduces a FOPIDNN voltage controller, which leverages neural network capabilities to dynamically adjust parameters, enhancing the accuracy of the inductor current reference and enabling faster response and robust power stability under varying environmental conditions. The proposed approach is first validated through simulations in PSIM, and then implemented on a real converter circuit using a DSP TMS320F28335 controller. Four control systems are compared: a MOC system with PID, an MPC system with PID, an MPC system with FOPID, and the proposed MPC system with FOPIDNN. The results demonstrate that the proposed method significantly improves inductor current reference compensation, yielding faster output voltage response and much smoother transitions during mode switching, ensuring more stable power output across a wide input voltage range.