Mutual information based real-time property extracting method

Abstract

A mutual information based real-time property extracting method comprises the following steps: 1) determining whether a current window is a first window; if so, entering the step 2); if not so, entering the step 3); 2) calculating the mutual information between two dimensions and forming a mutual information matrix; calculating the unit matrix of the mutual information matrix; performing propertydecomposition on the mutual information matrix; sequencing the property values and the property vectors; calculating the rate of contribution of each property vector; determining the component decision matrix of the front k property vectors with the accumulative rate of contribution of 85-95%; and mapping the data in the current window on the decision matrix; and 3) calculating the mutual information matrix in the current window; projecting to the unit matrix of the mutual information matrix in the previous window; performing property decomposition on the mutual information matrix; obtaining the component decision matrix of the front k property vectors with the accumulative rate of contribution of 85-95%; and projecting the data to the decision matrix to realize property extraction.

Description

technical field [0001] The invention relates to a real-time feature extraction method based on mutual information, which belongs to the technical fields of machine learning and data mining. Background technique [0002] In the past two years, the United States, Germany and China have successively proposed the "Advanced Manufacturing Partnership" report 2.0, "Industry 4.0 R&D White Paper", and "Made in China 2025", which opened the prelude to the transformation of the manufacturing industry to intelligent manufacturing. At the same time, the new concept of industrial big data has also appeared in front of people. Industrial big data is a new concept in the era of intelligent manufacturing. It generally refers to big data in the industrial field, including various data inside and outside the enterprise. With the successive implementation of intelligent transformation strategies, industrial big data has increasingly become the main driving force for the global manufacturing in...

AI Technical Summary

Problems solved by technology

It uses various variable parameters collected in the industrial production process to strictly monitor the production process, and through the analysis of production data, it improves the production process, optimizes the production process, and reduces energy consumption. However, due to its large volume and multi-source characteristics such as strong nonlinearity, and sparse value density make its analysis very difficult, and the requirements for analysis accuracy are relatively high. At the same time, the high-speed operation of the production line also requires higher real-time data processing

Therefore, in order to mine the information hidden in industrial big data better and faster, and accurately predict and analyze it, it is necessary to remove redundant attributes in the data, reduce the workload in the subsequent mining process, and improve the efficiency of data mining. In order to improve efficiency and performance, feature extraction must be performed on industrial big data. Traditional feature extraction algorithms have very strict requirements on data, and have certain requirements on the distribution and internal structure of data samples, and there are certain requirements on processing speed. Therefore, this paper proposes a real-time feature extraction algorithm based on mutual information (Feature Extraction Algorithm Based OnMutual Information, MIFE); this algorithm uses mutual information as the correlation coefficient for principal component analysis, and uses sliding windows to dynamically update data; then the current The combination of window data and new window data completes the feature extraction of the overall data

Images