Hybrid Cloud Storage Profile via Dynamic Confidence Method

The patent implements dynamic weight adjustment for different storage features based on environment factors. The system continuously updates the importance weights of storage features (such as security, cost, performance) according to changing business requirements and environmental conditions, enabling the storage management system to adapt dynamically without complete retraining. This resolves the contradiction by making the system flexible and adaptive while maintaining a relatively simple underlying architecture.

The patent changes the parameters of the deep learning model by adjusting feature weights rather than retraining the entire model. When environment factors change, the system modifies the weight parameters of existing features to reflect new priorities, allowing the model to adapt to new conditions while preserving the learned patterns from historical data. This approach maintains adaptability while avoiding the complexity of complete model retraining.

The patent performs preliminary training of the deep learning model on historical storage data to learn general patterns and relationships. Once trained, the model maintains these learned patterns while only adjusting feature weights in response to environment changes. This preliminary action establishes a solid foundation that reduces the need for frequent complete retraining, thereby maintaining prediction accuracy while minimizing training time and computational overhead.

Instead of performing full model retraining when environment factors change, the patent applies partial action by only adjusting the weight parameters of relevant features. This selective adjustment maintains prediction accuracy for the affected features while avoiding the computational expense of retraining the entire model, thus resolving the contradiction between accuracy and training time.

The patent implements feedback mechanisms where the deep learning model continuously predicts storage category and genre classifications, and these predictions are used to update the system's understanding of storage patterns. The model receives feedback from actual storage operations and environment factor changes, adjusting its predictions and feature weights accordingly. This feedback loop improves detection accuracy over time while maintaining system reliability and data security.

The patent uses a composite approach by combining multiple storage features (security requirements, performance characteristics, cost factors, capacity requirements) into a unified deep learning model that predicts storage category and genre. This composite model integrates diverse information sources to improve detection accuracy and reliability, making the detection process more robust despite the increased complexity of analyzing multiple features simultaneously.

The patent performs preliminary prediction of storage category and genre using the trained deep learning model before actual file allocation. This preliminary action provides advance guidance on where files should be allocated based on their characteristics and current environment factors, enabling fast allocation decisions that already account for security requirements and resource optimization. The model's pre-trained knowledge allows rapid classification without compromising security.

The patent dynamically adjusts the weight parameters of storage features based on environment factors before file allocation. When security concerns are elevated in the current environment, the model increases the weight of security-related features in its predictions, automatically prioritizing secure storage locations. This parameter adjustment enables the system to maintain high allocation efficiency while adapting to changing security requirements and resource conditions.