Financial time sequence prediction method and system based on frequency domain enhancement and morphological embedding

By employing frequency domain enhancement and morphological embedding methods, the problems of missing local semantic features and numerical drift in financial stock data prediction are solved, achieving high-precision stock price prediction and stable market trend following, thus improving the robustness of the model.

CN122453522APending Publication Date: 2026-07-24CHONGQING UNIV OF POSTS & TELECOMM
0 Cites 0 Cited by

Patent Information

Application Number
CN202610616776.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-05-07
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing models suffer from problems such as missing local semantic features, data augmentation disrupting long-term trends, and numerical drift in non-stationary data when processing financial stock candlestick data. These issues lead to large prediction errors, especially when the market is experiencing sharp fluctuations and the models are unable to accurately capture market turning points.

Method used

This method employs frequency domain enhancement and morphological embedding. It extracts deep features of K-lines through CNN morphological embedding, uses frequency domain enhancement to preserve low-frequency trends and applies perturbation to high-frequency components, and combines an adaptive inverse normalization strategy to achieve high-precision prediction of stock prices.

Benefits of technology

It achieves high-precision stock price prediction, with the average absolute error reduced to 23.62 and the relative error rate below 0.67%. It maintains the ability to follow trends without lag even in volatile markets, significantly improving the robustness of the model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the technical field of artificial intelligence, deep learning, financial science and technology and time sequence prediction, in particular to a financial time sequence prediction method and system based on frequency domain enhancement and morphological embedding, and the method comprises the steps: firstly obtaining the daily line-level opening price, the highest price, the lowest price, the closing price and the trading volume multi-dimensional K-line data of a stock market, and carrying out the moving average smoothing; reordering column dimensions, extracting K-line morphological features through multilayer one-dimensional convolution, and projecting the K-line morphological features into high-dimensional vectors; in the training stage, frequency domain enhancement is performed on data, and only high-frequency components are disturbed; the features are input into an iTransform encoder, and time sequence features are extracted; and finally, restoring a real predicted value through self-adaptive reverse normalization. According to the method, morphological embedding, frequency domain refinement enhancement and adaptive anti-normalization are matched with iTransform modeling, so that stock price high-precision prediction is realized, the numerical value cliff effect is eliminated, the trend following and inflection point capturing capability is improved, and the overall robustness of the algorithm is enhanced.
Need to check novelty before this filing date? Find Prior Art