初始状态不确定的非线性过程系统状态估计的鲁棒粒子滤波方法

在过程系统的控制与故障检测等方面，状态估计发挥着重要作用。针对非线性过程系统状态估计过程中初始状态不确定性问题，提出一种鲁棒粒子滤波方法。该方法首先引入初始状态准确性间接判定准则，根据判定的结果来选择是否进行基于观测偏差反馈机制的初始状态迭代改进。初始值准确性较差时，可以通过初始状态迭代改进策略使最终的初始粒子更接近真实的初始状态，从而增加产生初始粒子的正确性概率，通过粒子滤波迭代得到更准确的状态估计结果。将提出的鲁棒粒子滤波方法与传统粒子滤波方法应用于两个非线性动态系统实例中，结果验证了所提出方法的有效性与鲁棒性。

A robust particle filter for estimating states in nonlinear process systems with uncertain initial states

State estimation is critical for both process control and fault detection. A robust particle filter was proposed to estimate states in nonlinear process systems with uncertainty of initial states, which an indirect acceptance criterion was introduced to determine accuracy of the initial states and then to decide the needs for iterative improvement on the initial states by the feedback mechanism of observation bias. In case that the initial states were inaccurate, the iterative improvement strategy would be triggered to adjust particles closer to the true initial states. Therefore, the probability of setting the correct initial states to particles was increased and the accuracy of state estimation was improved through particle filter iteration. When applied to two nonlinear dynamic systems, the proposed particle filter demonstrated much more effectiveness and robustness than the traditional particle filter.