Self-adaptive online filtering method

Abstract

The invention provides a self-adaptive online filtering method which is an improved kalman filtering method which based on improvement of self-adaptive current statistics models. The method includes the following steps: firstly using kalman filtering recursion so as to perform online denoising; secondly, using the self-adaptive current statistics models to capture the wave characteristics of data,and at the same time removing color noise in the process of estimation; and finally, introducing a sliding window, and using an error compensation method which is self-adaptive to smoothing filtering, converging measurement noise to a truth value, and removing the measurement noise, thus making the result more precise. The self-adaptive online filtering method herein uses a plurality of self-adaptive models to real-time modify parameters, addresses the problem of high-precision online denoising measurement of data of sensors, and is universally practical.

Description

technical field [0001] The invention relates to the technical field of signal processing, in particular to an adaptive online filtering method. Background technique [0002] With the advent of the era of big data, all kinds of data will be transmitted to the data center through various sensor devices such as mobile phones, smart bracelets, and shared bicycles. The requirements for accuracy and reliability are getting higher and higher; however, in actual systems, wrong measurement methods, aging of sensor elements, defective data collection methods or external interference will cause noise in the measurement data, And any noise may cause serious errors and wrong results to the analysis and use of the data. Therefore, how to process these data in real time, realize online denoising, and improve data accuracy is of great significance. [0003] In practical systems, it is often necessary to process sensor measurement data online in real time to achieve real-time data transmis...

AI Technical Summary

Problems solved by technology

[0003] In practical systems, it is often necessary to process sensor measurement data online in real time to achieve real-time data transmission, such as pedestrian and vehicle trajectory measurement, civil structure deformation monitoring, etc., so methods that require a large amount of offline data such as wavelet transform filtering, frequency domain filtering does not apply

Secondly, a lot of measurement data measured by the sensor is highly mobile, so traditional methods such as SG filtering method, BISE method, smoothing filtering method, exponential smoothing filtering method, and classic Kalman filtering cannot retain the rapidly changing data. characteristics, the denoising effect is not accurate enough

Finally, due to the existence of measurement noise in the measurement process, it will greatly affect the accuracy of the data. People often use empirical knowledge to give the variance of measurement noise. This method is not accurate enough

[0004] In the existing research results and technical methods, there is no high-precision online denoising method that can solve the above three types of problems at the same time, resulting in weak versatility and low accuracy of the existing methods

Images