Machine-Learned Reactivity Prediction From Measured Reaction Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reactivity prediction methods rely on calculated activation free energy (ΔG‡) from quantum chemical simulations, which have significant deviations from measured values, leading to inaccurate synthesis route predictions.
Innovation Solution
A reactivity prediction system that utilizes a learned model to predict reaction parameters based on measured data, incorporating a storage unit to store a learned model that relates measured reaction parameters to reaction conditions, and a processing unit to predict reaction parameters using this model, including features like a soft sensor for direct measurement and machine learning to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If quantum chemical calculation is used to predict activation free energy, then synthesis route can be evaluated, but prediction accuracy deteriorates due to significant deviation from measured values
Solution Approach 1:
The patent replaces the quantum chemical calculation mechanism with a machine learning-based prediction mechanism. The system uses a learned model trained on experimental data to predict activation free energy, substituting the theoretical calculation approach with a data-driven approach that directly learns from measured values, thereby achieving higher prediction accuracy and reliability
Solution Approach 2:
The patent introduces an intermediary layer between reaction conditions and activation free energy prediction. The learned model acts as an intermediary that has been trained on the complex relationship between multiple reaction conditions (solvent, temperature, catalyst, etc.) and measured activation free energy values, enabling accurate prediction without direct quantum chemical calculation
2Measurement precision
If multiple reaction conditions are considered for accurate prediction, then prediction accuracy improves, but system complexity increases
Solution Approach 1:
The patent creates a universal learned model that handles multiple reaction conditions (solvent type, temperature, catalyst, additives, reagents) through a single predictive system. The model is trained to simultaneously process various input conditions and predict multiple reaction parameters, reducing overall system complexity while maintaining high prediction accuracy
Solution Approach 2:
The patent transforms multiple reaction condition parameters into a standardized input format for the machine learning model. By changing the representation of reaction conditions into suitable input features for the learned model, the system can accurately process diverse conditions without increasing operational complexity
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
A reactivity prediction system (40) includes: a storage unit (43) that stores a learned model (MD) that has learned a relationship between measured reaction parameters and reaction conditions under which the reaction parameters are measured; and a processing unit (44) that predicts reaction parameters from input reaction conditions using the learned model (MD).