Time Series Prediction Using Encoded Historical Window Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time series prediction technologies are not effective for fine-granularity and long-period time series data, leading to low prediction efficiency and precision, particularly in scenarios where data collection granularity is lower than the periodic trend, such as cloud platforms with user requests and internet of things devices, resulting in increased cold starts and resource wastage.
Innovation Solution
A time series data prediction method using a decoder with multiple layers and encoders that process historical time series data to extract relative location and change trend information, employing a Fourier transform convolution unit and multi-head self-attention mechanism to efficiently predict future traffic patterns, reducing the need for complete historical data and minimizing accumulated errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fine-granularity and long-period time series data is collected for accurate prediction, then prediction precision is improved, but data amount becomes excessively large leading to low prediction efficiency
Solution Approach 1:
The patent segments the time series data processing by dividing historical data into multiple time windows and using multiple encoders to process different segments separately. This segmentation allows the system to handle fine-granularity data without processing the entire long-period dataset as one unit, thereby maintaining prediction precision while improving processing efficiency through parallel computation of segmented data.
Solution Approach 2:
The patent transforms the one-dimensional time series data into multi-dimensional feature representations through multiple encoders that extract different characteristics (relative location information, change trend information). This dimensionality change allows the system to capture essential patterns from fine-granularity data without requiring processing of all raw data points, thus improving prediction efficiency while maintaining precision.
2Measurement precision
If complete historical time series data is used for prediction, then prediction accuracy is improved, but accumulated errors increase and processing complexity increases
Solution Approach 1:
The patent extracts only the essential features from complete historical time series data using multiple encoders, rather than processing all raw data. The encoders extract relative location information and change trend information, which are the critical elements needed for accurate prediction. This extraction process reduces processing complexity while maintaining prediction accuracy by focusing on essential patterns.
Solution Approach 2:
The patent performs preliminary processing of historical data by pre-computing encoded features (relative location information and change trend information) using multiple encoders before the actual prediction task. This preliminary action prepares the data in an optimized format that reduces accumulated errors and simplifies subsequent prediction operations, thereby reducing overall processing complexity.
3Measurement precision
If more computing resources are allocated to handle large data amounts, then prediction precision is improved, but resource waste increases when traffic is low
Solution Approach 1:
The patent implements dynamic resource allocation by using predicted traffic patterns to adjust the number of container instances actively serving requests. The system continuously predicts future traffic based on historical patterns and dynamically scales resources accordingly, ensuring sufficient computing power is available during high-traffic periods for accurate predictions while reducing resource allocation during low-traffic periods to minimize waste.
Solution Approach 2:
The patent performs preliminary prediction of future traffic patterns using historical data analysis. This preliminary action allows the system to anticipate traffic demands and proactively adjust resource allocation before actual traffic peaks occur, ensuring that computing resources are optimally configured to handle predicted loads without over-provisioning during low-traffic periods, thus reducing resource waste while maintaining prediction precision.
Data Source
AI summary
A time series data prediction method and apparatus, and a storage medium are provided. The method includes: obtaining current time series data collected in a current time window that is adjacent to and precedes a prediction time window in a current time period, and obtaining a plurality of groups of historical time series data separately collected in a same target time window of a plurality of historical time periods; encoding the plurality of groups of historical time series data by using a plurality of encoders respectively, to obtain a plurality of historical time series features, where each historical time series feature represents relative location information and change trend information of each group of historical time series data in the target time window; and determining, predicted time series data corresponding to a target object in the prediction time window.