A method and system for adaptive sequence window selection based on normalized information entropy

CN122241423APending Publication Date: 2026-06-19BEIJING C&W ELECTRONICS GRP

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: BEIJING C&W ELECTRONICS GRP
Filing Date: 2026-03-06
Publication Date: 2026-06-19

AI Technical Summary

Technical Problem

Existing technologies lack systematic and theoretical support in selecting time series window lengths, resulting in poor model performance and an inability to adapt to diverse time series data characteristics. In particular, they struggle to ensure the effective utilization of computing resources in complex system monitoring and intelligent early warning.

Method used

An adaptive sequence window selection method based on normalized information entropy is adopted. By receiving the time series data stream and the candidate window range, the sliding window is traversed, a unique identifier is generated, the probability of subsequences is statistically calculated, the normalized information entropy value is calculated, the optimal modeling window is dynamically selected, and the optimal window length is output to configure the sequence model.

Benefits of technology

It achieves adaptive characterization of the essential characteristics of time series, dynamically adapts to multiple input data sources, improves the robustness and adaptability of the model, and is suitable for various application scenarios such as anomaly detection, classification and recognition, and future trend prediction, while significantly reducing the consumption of computing resources.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122241423A_ABST

Patent Text Reader

Abstract

This application relates to an adaptive sequence window selection method and system based on normalized information entropy, belonging to the field of computer data processing technology. The window selection method includes: receiving a time-series data stream and a preset candidate window range; traversing each window length within the candidate window range and extracting all continuous subsequences from the time-series data stream using a sliding window method with a step size of 1; generating a unique identifier for each subsequence and calculating the occurrence probability of each subsequence; calculating the original information entropy corresponding to each window length based on the occurrence probability of each subsequence; calculating the normalized information entropy value based on the original information entropy and the maximum achievable entropy corresponding to each window length; selecting the optimal modeling window from the candidate window range according to a preset selection strategy; and outputting the optimal modeling window length to configure the input sequence length of the sequence model. This application can automatically determine the optimal window length by combining the statistical characteristics of the time series itself.

Need to check novelty before this filing date? Find Prior Art