本篇論文提出了一個使用神經網路來補償跨網控制系統所產生的遺失封包問題。神經網路輸入的是經由記憶體儲存的前一刻與前兩刻的訊號資訊,用於補償出此刻所遺失的訊號,藉此增加系統追蹤的準確性與穩定性。在訓練神經網路的階段時收集各種情形下的八階系統或是無人機飛行狀態的前一刻跟前兩刻以及此刻的狀態與控制訊號,藉由上述的資訊來進行多個神經網路補償器的訓練,並藉由之前是否有遺失過訊號的資訊在不同的補償器之間進行判斷切換。完成多重補償器與判斷器之後即可部屬於跨網路控制系統上,然後使用Simulink進行無人飛行器與八階系統模擬以證明此方法是否在有遺失訊號的情況下減少因遺失訊號所造成的影響。
This dissertation proposes a neural network to compensate for packet loss caused by cross-network control systems. The input of neural network is stored the signal that is two moments ago by memory, and used to calculate for lost signals at this moment, thereby increasing the accuracy and stability of system tracking. In the training data, collect the information of the state and control that is different drone's flight paths and eight-order system, and the training of the plurality of neural network compensators is performed by the above information. The switch between different compensators by whether or not the information of the lost signals in previous moment, after completing the multi-compensator and switch, it can be deployed on the network control system. Next, we simulate the UAV and eight-order system in the Simulink and prove this method can improve the performance of system tracking control.