DLinear
Description
This API calls the DLinear model, which can be used for time series forecasting.
DLinear Model Architecture

This model is a combination of the time series decomposition method used in Autoformer and FEDformer with a linear layer.
Time Series Decomposition Process:
- Generate Moving Average.
- The moving average of the time series data is calculated.
- Remove Moving Average.
- The moving average is removed to decompose the data into trend and seasonality components.
Learning Process:
- Decompose into Trend and Seasonality.
- After removing the moving average, the data is split into trend and seasonality components for training.
- Apply Linear Layers to Each Component.
- A single linear layer is applied to each of the trend and seasonality components for learning.
- Sum the Two Outputs.
- The predictions for the trend and seasonality are summed to calculate the final prediction.
API module path
from api.v2.model.DLinear import DLinear
Parameters
Window_size
- Specifies the length of the input sequence.
- Example
- window_size = 10
Forecast_size
- Specifies the number of time points to predict.
- Example
- forecast_size = 1
Kernel size
- Specifies the kernel size for decomposition.
- default : 25
- Example
- kernel_size = 25
Individual
- Specifies the whether to use individual linear layers for each feature.
- default : False
- Example
- individual=False
Feature_size
- Specifies the input feature size.
- Example
- If data's shape (x,x,1).
- feature_size = 1
Use_revin
- Specifies whether to use RevIN normalization.
- default : False
- Example
- Use_revin = False
Multi_feature
- Specifies whether to apply a fully connected (FC) layer for multiple features.
- default : False
- Example
- multi_feature = false
Example Sample Code (Python)
Results

Check the entire module code.
datahub/api/v2/model/DLinear.py at main · machbase/datahub
All Industrial IoT DataHub with data visualization and AI source - machbase/datahub