Catastrophe filter algorithm

Abstract

The invention relates to a catastrophe filter algorithm. A large quantity of state catastrophes exist in the natural world, and therefore signal processing becomes very important for the kind of systems. In combination with system characteristics and state information, system noise and observation noise are regarded as two stochastic control variables, a catastrophe potential function is established, a singular point set is solved, catastrophe characteristics are evaluated through a catastrophe series method, then the normalized membership degree of the state catastrophes is calculated to express the degree of the state catastrophes, if catastrophes happen on a state completely, a last state predication value is not considered in the filter calculation in this time, and state evaluation is performed in combination with a sampling point data predication value and a current measurement value. If catastrophes happen on the state partially, a using state predication value is calculated according to the membership degree of the catastrophes, and then state evaluation is performed in combination with the sampling point data predication value and the current measurement value. The catastrophe filter algorithm has the advantages that a mathematical model and statistics characteristics of the system noise and measurement noise do not need to be known accurately, and the catastrophe filter algorithm is particularly suitable for performing signal processing on strong-nonlinearity and strong-stochastic systems.

Description

technical field [0001] The invention relates to a filter estimation method of abrupt signal, which is especially suitable for signal processing of strong nonlinear and strong random system. technical background [0002] In nature, in addition to gradual and continuous smooth changes, there are also a large number of sudden changes and jumps, such as gyro drift signals, the output signal of the radio altimeter when the plane is flying horizontally over the mountain, and the output of inertial navigation. Error signals, error signals in transfer alignment, GPS SA error signals, etc. These signals have mutation characteristics. Since the signal has no deterministic frequency spectrum, it is impossible to extract useful signals by conventional filtering methods. [0003] The existing KF, EKF and UKF and their improved filtering methods are widely used, especially the UKF filtering has high precision and fast convergence speed. The prediction of covariance, and then recursion a...

AI Technical Summary

Problems solved by technology

Since the signal has no deterministic spectrum, it is impossible to extract useful signals by conventional filtering methods

[0003] The existing KF, EKF and UKF and their improved filtering methods are widely used, especially the UKF filtering has high precision and fast convergence speed. The prediction of covariance, and then recursion and update, but the UKF algorithm belongs to the expansion of the classic KF filter, and is also based on the accurate mathematical model and the known statistical characteristics of system noise and measurement noise. When the system or environment changes drastically , the statistical characteristics of the noise will change greatly, and the filter accuracy and stability will decrease, or even diverge

In order to improve the UKF’s lack of adaptive ability to mutations, Sage–Husa filtering, robustness filtering, strong tracking filtering, fading factor algorithm, etc. have appeared. However, these algorithms are based on strict mathematical reasoning and harsh assumptions. However, these conditions are often difficult to satisfy in practical systems.

Images