Ensemble Machine Learning for Long-Term Time Series Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models face limitations in making accurate long-term predictions, requiring large amounts of data and significant computational resources, which is inefficient and often results in incorrect forecasts, especially in applications where data and computation power are constrained.
Innovation Solution
An ensemble-based machine learning model architecture that processes time series data using multiple machine learning models, such as ARIMA and LSTM, to generate future predictions, either non-overlapping or averaged predictions based on defined steps, conserving resources and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large capacity neural networks are used for long-term forecasting, then prediction accuracy may improve, but computing resources and data requirements increase significantly
Solution Approach 1:
The patent divides the forecasting task into multiple segments by using multiple smaller machine learning models instead of one large capacity neural network. Each model processes a portion of the time series data and generates predictions for specific future steps, which are then combined to produce the final forecast. This segmentation approach reduces the computational burden on individual models while maintaining overall prediction accuracy.
Solution Approach 2:
The patent merges the outputs of multiple machine learning models to create a comprehensive forecasting system. By combining predictions from different models that process different portions of the data or generate predictions for different time steps, the system achieves accurate long-term forecasting without requiring any single model to be excessively large or complex.
2Adaptability or versatility
If multiple machine learning models are used to generate overlapping predictions, then prediction coverage improves, but computational complexity increases
Solution Approach 1:
The patent dynamically adjusts the number of models, the number of predictions each model generates, and the skip steps between predictions based on the specific forecasting requirements and available computational resources. This dynamic configuration allows the system to optimize between prediction coverage and computational complexity for different scenarios.
Solution Approach 2:
The patent changes key parameters such as the number of machine learning models, the number of future steps each model predicts, and the skip steps between predictions to balance prediction coverage with computational complexity. By tuning these parameters, the system can adapt to different forecasting needs while managing resource consumption.
Data Source
AI summary
A device may receive time series data, and may define a first quantity of steps into past data utilized to make future predictions, a second quantity of steps into the future predictions, and a third quantity of steps to skip in the future predictions. The device may determine whether the second quantity is equal to the third quantity. When the second quantity is equal to the third quantity, the device may process the time series data, with a plurality of machine learning models, to generate a plurality of future predictions that do not overlap, may merge the plurality of future predictions into a list of future predictions, and may provide the list for display. When the second quantity is not equal to the third quantity, the device may process the time series data, with the plurality of machine learning models, to generate another plurality of future predictions that do overlap.