Multi-Step Time Series Prediction Using Seasonal Lag Encoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional time series data prediction systems are limited to one-step/single-step predictions and fail to capture non-linear properties and seasonal information, restricting their applications.
Innovation Solution
A processor-implemented method and system for time series data prediction that determines the order of seasonality, selects encoders based on Partial Auto Correlation Function (PACF) significant lags, generates a combined context vector, and trains encoder-decoder pairs using back-propagation through time to generate multi-step predictions based on seasonal lag information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If single-step prediction using non-seasonal linear model is used, then system complexity is low, but prediction capability and application extent are limited
Solution Approach 1:
The prediction system is segmented into multiple encoders (first encoder for non-seasonal lags, second encoder for seasonal lags) that process different aspects of historical data separately. This segmentation allows the system to capture both non-seasonal and seasonal patterns independently, improving prediction capability while maintaining manageable complexity through modular architecture
Solution Approach 2:
The system transitions from single-step prediction to multi-step prediction by adding the time dimension. The encoder-decoder architecture processes historical data sequences and generates predictions for multiple future time steps, expanding the prediction horizon from one time step to multiple time steps simultaneously
2Use of energy by moving object
If single-step prediction is used, then computational resources are saved, but non-linear properties and seasonal information are not captured
Solution Approach 1:
Context vectors serve as intermediaries that bridge historical data and predictions. The first context vector captures non-seasonal patterns from historical data, while the second context vector captures seasonal patterns. These context vectors are combined to form a comprehensive representation that preserves both non-linear properties and seasonal information for multi-step predictions
Solution Approach 2:
The historical data processing is segmented into two separate encoder pathways: one for non-seasonal lags and another for seasonal lags. This segmentation allows efficient processing of different data characteristics independently, capturing non-linear properties through the non-seasonal encoder and seasonal information through the seasonal encoder, thereby reducing information loss
Data Source
AI summary
State of the art systems that are used for time series data prediction have the disadvantage that perform only one step prediction, which has only limited application. Disadvantage of such systems is that extent of applications of such single step predictions are limited. The disclosure herein generally relates to time series data prediction, and, more particularly, to time series data prediction based on seasonal lags. The system processes collected input data and determines order of seasonality of the input data. The system further selects encoders based on the determined order of seasonality and generates input data for a decoder that forms encoder-decoder pair with each of the encoders. The system then generates time series data predictions based on seasonal lag information distributed without redundance between encoder and decoder inputs.