Energy Forecasting Model Fine-Tuning for Data-Scarce Buildings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Smart Energy Management Systems (SEMS) face challenges in accurately forecasting energy consumption in newly instrumented buildings due to reliance on historical data, leading to inaccurate energy estimations and performance issues.
Innovation Solution
A method utilizing Few-Shot Learning for training prediction models, which involves obtaining time series data from similar past buildings, pretraining a model with generic data, and fine-tuning it with limited data from the new building, using techniques like Gaussian normalization and R2-score evaluation to enhance adaptability and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional data-dependent models are used for energy forecasting in newly instrumented buildings, then the models rely on historical data, but the forecasting accuracy deteriorates due to limited or low-quality data availability
Solution Approach 1:
The patent applies preliminary action by pretraining the model backbone on generic energy consumption data from multiple buildings before deployment to new buildings. This pretraining establishes a foundational understanding of energy patterns that can be adapted later, allowing the model to perform reasonably well even when target building data is limited or of low quality.
Solution Approach 2:
The patent uses copying by replicating the pretrained backbone model structure and transferring knowledge from source buildings to target buildings through fine-tuning. The model architecture and learned features are copied and adapted to new contexts, enabling the system to leverage patterns from multiple buildings rather than relying solely on limited data from a single new building.
2Adaptability or versatility
If models are trained from scratch for each new building, then the models can be customized to specific building characteristics, but the training time and computational resources increase significantly
Solution Approach 1:
The patent applies preliminary action by pretraining the model backbone on generic energy consumption data from multiple buildings before deployment to new buildings. This pretraining establishes a foundational understanding of energy patterns that can be adapted later, allowing the model to perform reasonably well even when target building data is limited or of low quality.
Solution Approach 2:
The patent applies local quality by fine-tuning only the necessary components of the pretrained model using target building data. Instead of retraining the entire model, the system selectively adapts specific layers or parameters to capture building-specific characteristics, thereby achieving customization while minimizing additional training time and computational resources.
3Productivity
If generic pretrained models are used without fine-tuning, then the training speed is fast, but the forecasting accuracy deteriorates for specific building scenarios
Solution Approach 1:
The patent applies local quality by fine-tuning only the necessary components of the pretrained model using target building data. Instead of retraining the entire model, the system selectively adapts specific layers or parameters to capture building-specific characteristics, thereby achieving customization while minimizing additional training time and computational resources.
Solution Approach 2:
The patent applies partial action by performing selective fine-tuning rather than complete retraining. The system applies just enough additional training on target building data to adapt the pretrained model to specific building characteristics, avoiding the excessive computational cost of full retraining while still improving forecasting precision for the specific building scenario.
Data Source
AI summary
A method for training at least one model able to predict a power consumption or production of at least one electric equipment, also called target. The method includes: (a) obtaining time series data representing the evolution of the power consumption or production of the target over a first period of time, (b) comparing the target time series data to known time series data representing the evolution, over a second period of time, of the power consumption or production of known electric equipments, the second period of time being greater than the first period of time, to determine the k known time series data that are the most similar to the target time series data, (c) training of a first prediction model or backbone for each of the k known electric equipments, the backbone being able to predict the evolution over time of the consumption or the production of the corresponding known electric equipment, the backbone being trained on the corresponding time series data over the second period of time, and (d) training at least one second prediction model, called target model, by fine tuning at least one of the first trained prediction model on the time series data of the target over the first period of time.