使用粒子濾波器方法設計適用於不定延遲時間與封包遺失之網路化控制系統之估測器

Abstract

本研究之目的是為網路化控制系統設計估測器方法，首先針對網路化控制系統當中之網路與受控系統建立一以機率為基礎之模型。其中使用馬可夫鏈隨機變數描述資訊封包在網路中傳遞之情況，用以表現不定之延遲時間與遺失機率。 由於使用機率進行模型化，因而使用粒子濾波器方法為基礎設計對應之估測器，並應用粒子濾波器中常見之改良方法如重新取樣法以及正規化方法以及選擇適當之重要性取樣分布來改善估測效果。同時提出在高階系統與考慮到估測有效視窗長度情況下，對粒子濾波器當中計算權重之公式之近似，用以減少計算需求以及節省記憶空間，同時能取得合理之權重更新趨勢。 模擬與實驗時，選取一加熱系統為處理對象，其具有多觀測端點以及應用情境可遠端之特性，並且能在較少之控制命令下得到多個系統狀態之變化。將此系統所收集到之觀測數據透過網路傳輸並紀錄，而後送至估測器對估測方法作驗證。並且將估測結果與傳統之卡曼濾波器作比較，從而驗證此設計方式之可行性與效果。

The purpose of this research is to design a state observer specifically for networked controlled systems. We propose a probabilistic model for the network controlled system which consists of a Markov-chained random variable to represent the transferring status of each data packet. Because of the inherent probabilistic nature of the proposed model, we design the observer by using particle filtering techniques. To improve the performance, both resampling and regularization methods are incorporated into the particle filter, and the importance sampling function is carefully selected. In addition, we simplify the way to update the weightings of the samples in order to reduce the demanding computational power and the need for memory space, while to maintain reasonable state estimation accuracy. In the simulation and experiment parts, a heating system is chosen. The system has multiple sensing nodes for a remote application, and it can change various state variables with a few control commands. After collecting the temperature data of the system, the data are sent through a real network to the observer for state estimation. The result of the estimation is then compared with the outcome of a traditional Kalman filter in order to verify the feasibility and effectiveness of the proposed method.