Short-term load forecasting using multiple-source boosting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Accurate short-term load forecasting is challenging due to limited training data, especially for newly built houses, and changing consumption patterns, which require machine learning models to adapt quickly with small datasets, while emerging factors like renewable energy and electric vehicles introduce additional complexity.
Innovation Solution
The proposed method employs a Multiple-source Boosting based Deep Transfer Regression (MBDTR) framework, using deep regression models learned from multiple source domains via gradient boosting to improve forecasting performance in data-scarce target domains, minimizing negative transfer through selective and customized transfer processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used for short-term load forecasting, then the models can be trained with available data, but the forecasting accuracy deteriorates when training data is limited or when consumption patterns change
Solution Approach 1:
The patent combines multiple source domain datasets with different consumption patterns into a unified target domain model. By merging heterogeneous data sources (residential, commercial, industrial loads) through domain adaptation techniques, the system creates a comprehensive forecasting model that achieves high accuracy even when individual source domains have limited data availability.
Solution Approach 2:
The patent develops a universal forecasting framework that can handle multiple domain types (residential, commercial, industrial) and various consumption patterns simultaneously. The domain adaptation mechanism enables a single model to function effectively across diverse scenarios, making the system universally applicable regardless of data scarcity in any specific domain.
2Adaptability or versatility
If machine learning models are trained to adapt quickly to changing consumption patterns, then the models can respond to emerging factors like renewable energy and electric vehicles, but the complexity of the modeling process increases
Solution Approach 1:
The patent performs preliminary domain adaptation and feature alignment before actual forecasting. By pre-processing source domain data to match target domain characteristics and pre-training models on aggregated data, the system reduces the complexity of adapting to new patterns when they emerge, as the foundational adaptation work is already completed.
Solution Approach 2:
The patent dynamically adjusts model parameters and domain adaptation weights based on changing consumption patterns. When emerging factors like renewable energy or electric vehicles affect load patterns, the system modifies parameter configurations and re-weighting schemes to accommodate new patterns without requiring complete model redesign, thus managing complexity while maintaining adaptability.
3Measurement precision
If domain adaptation techniques are applied to transfer knowledge from source domains to target domain, then forecasting performance improves in data-scarce scenarios, but the risk of negative transfer increases
Solution Approach 1:
The patent implements feedback mechanisms that monitor the effectiveness of domain adaptation in real-time. By evaluating whether knowledge transfer from source domains is improving or degrading target domain performance, the system can dynamically adjust adaptation weights or switch between different source domains, thereby preventing negative transfer while maintaining high forecasting performance.
Data Source
AI summary
A method, computer program, and computer system are provided for load forecasting. Datasets corresponding to source machine learning models and a target domain base model are identified. A set of forecasting models corresponding to the identified datasets are learned. An ensemble model is determined from the learned set of forecasting models based on gradient boosting. An available resource is allocated based on the ensemble model.