Solar Irradiance Nowcasting Model With Threshold-Based Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing solar irradiance nowcasting models are static and fail to adapt to dynamic environmental changes, leading to deteriorated predictive accuracy and inefficiencies in power grid operations.
Innovation Solution
A solar irradiance nowcasting system with a storage unit and processing unit that continuously accumulates recent sky images and GHI data, periodically checks for a threshold, and initiates a retraining process to adapt the model using deep learning architectures like CNN-LSTM, VGG16, and ResNet18, ensuring up-to-date predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a static pre-trained deep learning model is used for solar irradiance nowcasting, then the initial predictive accuracy is achieved, but the model fails to adapt to dynamic environmental changes over time, leading to deteriorated predictive accuracy
Solution Approach 1:
The patent implements a dynamic model update mechanism where the pre-trained deep learning model is periodically retrained using newly accumulated sky images and GHI data. This transforms the static model into a dynamic system that continuously adapts to changing environmental conditions, resolving the contradiction between initial accuracy and long-term adaptability.
Solution Approach 2:
The system changes the training data parameters over time by accumulating new environmental data (sky images and GHI measurements) and using this updated data to retrain the model. This parameter change approach allows the model to maintain predictive accuracy while adapting to new environmental conditions.
2Reliability
If the model is retrained frequently to maintain accuracy, then predictive accuracy is improved, but computational resources and time are consumed
Solution Approach 1:
The patent implements periodic retraining where the model is retrained at predetermined time intervals or when accumulated data reaches a threshold. This periodic action balances the need for maintaining accuracy with the cost of retraining, avoiding both excessive retraining and model obsolescence.
Solution Approach 2:
The system performs preliminary data accumulation and preprocessing before retraining is initiated. By continuously accumulating and preprocessing sky images and GHI data in advance, the system prepares training data ready for when retraining is triggered, reducing the actual retraining time and resource consumption.
3Adaptability or versatility
If more recent data is accumulated for retraining, then model adaptability is improved, but data storage requirements increase
Solution Approach 1:
The patent extracts only the essential and recent data needed for model retraining, rather than storing all historical data. By selectively accumulating recent sky images and corresponding GHI data, the system achieves sufficient model adaptability while minimizing storage requirements.
Solution Approach 2:
The system uses a partial dataset approach by accumulating only a predetermined amount of recent data (not the complete historical record) for retraining purposes. This partial action provides sufficient adaptability improvement while controlling storage consumption.
Data Source
AI summary
A solar irradiance nowcasting system is provided. The system includes a storage unit and a processing unit. The storage unit is configured to store a structured dataset, an unstructured dataset, and a solar irradiance nowcasting model. The processing unit generates predicted GHI values based on current sky images using the model and adaptively maintains the model by executing operations including accumulating recent sky images and corresponding GHI data into the datasets, periodically checking if their accumulated number reaches a specified threshold, and initiating retraining upon reaching the threshold. The retraining process involves using the accumulated recent sky images and GHI data to update the model, ensuring accurate and up-to-date predictions.