Gasoline Blend Soft-Sensor Using Feature-Specific Blending Rules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing blending rules in oil refineries are not optimized for each feature, leading to inefficiencies in predicting and achieving desired blend properties, such as octane number, in gasoline blending.
Innovation Solution
A system and method that optimize blending rules by extracting relevant data from multiple sources, preprocessing it, selecting key features, creating soft-sensors, and determining the best blending rules using optimization techniques to predict desired blend properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple blending rules are used to predict blend properties, then prediction coverage is improved, but prediction accuracy deteriorates due to lack of optimization for each feature
Solution Approach 1:
The patent segments the blending rule selection process by creating separate optimized blending rules for each feature (property) of the blend components. Instead of using a single general blending rule for all properties, the system develops specific blending rules tailored to each feature such as octane number, research octane number, motor octane number, vapor pressure, and others. This segmentation allows each rule to be optimized independently for its specific feature, thereby maintaining high prediction accuracy across multiple different properties.
Solution Approach 2:
The patent applies local quality by making each blending rule specialized for its specific feature rather than using a uniform approach. Each blending rule incorporates feature-specific parameters and optimization criteria tailored to the particular property being predicted. For example, the octane number blending rule uses different correlation parameters and weighting factors compared to the vapor pressure blending rule, allowing each rule to capture the unique characteristics and blending behavior of its target feature.
2Reliability
If data from multiple sources is collected and processed, then prediction reliability is improved, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by performing extensive data preprocessing, cleaning, and feature selection before the actual blending rule optimization and prediction processes. The system pre-processes data from multiple sources including component properties, blend compositions, and experimental measurements, organizing and validating the data in advance. This preliminary data preparation ensures high-quality input data for the blending rules, improving prediction reliability while managing complexity through structured data handling performed beforehand.
Solution Approach 2:
The patent uses soft-sensors as intermediary elements that bridge the gap between raw data from multiple sources and the final blending predictions. These soft-sensors act as computational mediators that process and interpret data from various sources, transforming raw component and blend data into meaningful predictions of blend properties. The soft-sensors incorporate the optimized blending rules and serve as intermediaries that reconcile data from different sources, improving reliability while managing system complexity through a structured computational layer.
Data Source
AI summary
The gasoline blending is a critical aspect in oil refinery operations. There are multiple blending rules in prior art to predict the chemical or physical property of the blend. But none of the prior method focuses on obtaining the best blending rule for each feature used in creating the soft-sensor. A method and system for generating a soft-sensor for getting desired property of a blend by optimizing a set of blending rules for each feature used for creating soft-sensor have been provided using data from individual components of the blend. The method comprises automated soft-sensor creation using multiple data sources. Further, the method involves finding best blending rule for each feature used for building the soft-sensor or the blending model to predict property of the blend. The soft-sensor developed using data of components used for blending is adapted to predict mixture or blend property with limited tuning.