Water Treatment Model Selection for Adaptive Chemical Dosing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing water treatment methods for seawater desalination plants face challenges in real-time chemical dosing optimization due to reliance on sampling experiments and operator knowledge, making it difficult to adapt to changes in feed water conditions.
Innovation Solution
A device and method for selecting an optimal water treatment model through a model storage part, model generation part, and model evaluation part, which generates and evaluates variable models based on training data to determine the champion model with the smallest error, allowing for adaptive chemical dosing optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If sampling experiments and operator knowledge are used for chemical dosing, then chemical dosing can be performed, but real-time adaptation to feed water changes is difficult
Solution Approach 1:
The patent replaces manual sampling experiments and operator knowledge-based control with an automated machine learning system. The system uses multiple water treatment models (seed models, variable models, and optimal models) that are automatically trained and evaluated using historical water quality data and chemical dosing results, enabling real-time adaptation without manual intervention.
Solution Approach 2:
The system performs self-learning and self-optimization by automatically training models with historical data, evaluating model performance, and selecting the best-performing model for chemical dosing control. The model training and selection process occurs autonomously without requiring continuous operator input or manual experimentation.
2Reliability
If multiple water treatment models are maintained for evaluation, then optimal model selection is possible, but device complexity increases
Solution Approach 1:
The patent segments the model management system into distinct functional modules: a seed model storage unit for storing base models, a variable model generation unit for creating adapted models, and an optimal model storage unit for storing the best-performing models. This segmentation allows each module to be independently managed and evaluated, reducing overall system complexity while maintaining multiple models for reliable selection.
Solution Approach 2:
The system performs preliminary model training and evaluation by maintaining a library of pre-trained seed models and pre-generated variable models before actual chemical dosing operations. This preliminary preparation allows the system to quickly select from pre-evaluated models rather than training new models in real-time, reducing operational complexity while ensuring reliable model selection.
3Reliability
If model training and evaluation is performed continuously, then optimal model is maintained, but processing time and computational resources increase
Solution Approach 1:
The patent implements periodic model training and evaluation instead of continuous processing. The system trains variable models from seed models at predetermined intervals using historical water quality data, and evaluates models periodically to identify performance changes. This periodic approach maintains model reliability while significantly reducing computational time and resource consumption compared to continuous training and evaluation.
Data Source
AI summary
A device for selecting an optimal model includes: a model storage part including a seed model storage place in which a seed model is stored, and an optimal model storage place in which an existing optimal model is stored; a model generation part configured to use training data to generate a variable model; and a model evaluation part configured to prepare evaluation data, and use the evaluation data to select a champion model from among a plurality of evaluation target models including the seed model, the existing optimal model, and the variable model by evaluating the plurality of evaluation target models.