Temporal Relation-Effect Layer Module for Time-Series Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI-based time-series analysis methods struggle to effectively decompose time-series data into similar patterns within AI model layers, and they fail to consider the relations between different time-series patterns, limiting their application in real-world systems.
Innovation Solution
A deep learning neural network with a temporal relation-effect layer module, comprising a temporal decomposition gate module, a temporal relation gate module, and a temporal effect gate module, is used to improve time-series data-based deep learning prediction performance by learning temporal relations and effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional AI-based time-series analysis methods (CNN, RNN, LSTM, GRU) are used, then the model structure is established, but the input time-series data cannot be decomposed into similar time-series patterns within the corresponding layer, and the relations between different time-series patterns cannot be considered
Solution Approach 1:
The patent applies segmentation by decomposing the time-series analysis task into distinct functional modules: a temporal decomposition module that separates time-series data into multiple patterns, and a temporal relation module that analyzes relationships between these patterns. This modular segmentation allows each component to specialize in specific aspects of temporal analysis, improving prediction accuracy while managing complexity through structured division of labor.
Solution Approach 2:
The patent introduces an intermediary temporal decomposition module that acts as a bridge between the input time-series data and the main AI model layers. This intermediary component decomposes the raw time-series data into multiple similar time-series patterns before feeding them into the CNN, RNN, LSTM, or GRU layers, enabling the model to effectively process and relate different temporal patterns without overwhelming complexity.
2Measurement precision
If all potential factors in time-series data are considered, then the prediction performance is improved, but it becomes difficult to pinpoint which factors affect each other in what relations
Solution Approach 1:
The patent segments the complex web of temporal factors into distinct decomposed time-series patterns, each representing specific temporal characteristics. By separating the analysis into pattern decomposition and pattern relationship analysis stages, the system can handle all potential factors systematically while making their interrelations detectable through the structured output of the temporal decomposition module.
Solution Approach 2:
The temporal decomposition module serves as an intermediary that transforms the complex mixture of all potential factors into organized, separable time-series patterns. This intermediary representation makes it possible to detect and measure factor relations by analyzing the relationships between the decomposed patterns, rather than attempting to directly analyze the tangled original data.
3Adaptability or versatility
If existing speech recognition models trained on adult voices are applied to infants' voices, then the model structure is available, but the recognition rate is low due to differences in voice characteristics
Solution Approach 1:
The patent applies local quality by training the temporal decomposition module specifically on infants' voice data to capture the unique temporal patterns and characteristics of infant speech. This specialized training enables the model to adapt to the local qualities of infant voices (such as different pitch ranges, rhythm patterns, and phonetic structures) rather than applying a generic adult-voice-trained model, thereby improving recognition rate while maintaining model versatility.
Data Source
AI summary
Provided are a method and system for providing time-series prediction deep learning neural network by a time-series prediction application run by at least one processor of a terminal include: training a temporal relation-effect layer module; inputting first time point time-series data into the temporal relation-effect layer module; acquiring second time point time-series data according to the first time point time-series data from the temporal relation-effect layer module; and providing the second time point time-series data, wherein the temporal relation-effect layer module includes a temporal decomposition gate module, a temporal relation gate module, and a temporal effect gate module.