Synthetic Data Training for Emission Classification Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The reliance on human effort to extract, understand, and accurately categorize growing amounts of unstructured enterprise data poses a challenge for organizations to comply with the Greenhouse Gas Protocol and develop effective energy efficiency measures to reduce emissions.

Innovation Solution

A method and system utilizing a large language model to generate synthetic data, which is used to train a field machine learning model to classify real-world enterprise data into pre-defined emission classes, minimizing the need for manual labeling and enabling efficient classification and explanation of emission scopes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If human effort is used to extract and categorize emissions data, then accuracy of categorization is improved, but productivity deteriorates due to exponentially growing data volume

Engineering Contradiction:
Improveaccuracy of categorizationVSAvoidproductivity
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system performs preliminary actions by generating synthetic training data that pre-encodes GHG Protocol classification rules and patterns. This synthetic data is used to train the machine learning model in advance, enabling it to automatically categorize emissions data without requiring human experts to manually process each data point, thus resolving the contradiction between maintaining high categorization accuracy and processing exponentially growing data volumes efficiently

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary machine learning model that acts as a bridge between raw emissions data and GHG Protocol classifications. The model is trained on synthetic data that captures the complex classification rules, enabling it to serve as an automated intermediary that maintains categorization accuracy while dramatically increasing processing productivity compared to direct human analysis

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of information

If more unstructured enterprise data is collected, then completeness of emissions accounting is improved, but device complexity deteriorates due to data processing requirements

Engineering Contradiction:
Improvecompleteness of emissions accountingVSAvoiddevice complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The system extracts and isolates the essential classification patterns from GHG Protocol guidelines into synthetic training data, separating the complex rule-based knowledge from the raw data processing task. This extraction allows the machine learning model to focus on pattern recognition rather than implementing complex business logic, thereby maintaining complete emissions accounting while reducing processing complexity

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transforms the complex categorical classification problem into a parameter optimization problem by training the machine learning model to predict emission scope classifications. The model learns optimal parameter weights and decision boundaries from synthetic data, converting the complex rule-based GHG Protocol guidelines into learned parameters that simplify the classification process while maintaining accounting completeness

Inventive Principle:
Principle #35Parameter changes

3Reliability

If manual labeling of emissions data is performed, then reliability of classification is improved, but loss of time deteriorates

Engineering Contradiction:
Improvereliability of classificationVSAvoidloss of time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system creates synthetic copies of labeled emissions data that capture the statistical patterns and classification patterns of manually labeled data. These synthetic copies are used to train the machine learning model, enabling it to achieve reliability comparable to manual labeling while processing data at automated speeds, thus resolving the contradiction between maintaining classification reliability and reducing time loss

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent performs preliminary action by pre-training the machine learning model on synthetic data that encodes reliable classification patterns from GHG Protocol guidelines. This preliminary training enables the model to automatically classify new emissions data with high reliability without requiring time-consuming manual verification for each data point, thereby maintaining reliability while eliminating time loss

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4485297A1Generation and use of classification model from synthetically generated data
Publication Date: 2025.01.01 SERVICES PETROLIERS SCHLUMBERGER SA
  • EP4485297A1 patent drawingFigure 1
  • EP4485297A1 patent drawingFigure 2
  • EP4485297A1 patent drawingFigure 3

AI summary

A method for training and using a field machine learning (ML) model to classify emission data is presented. The method includes generating synthetic data by a large language model (LLM) by prompting the LLM with emission classes and few shot examples. The synthetic data includes multiple synthetic data instances and corresponding instance labels. A training dataset is obtained from the synthetic data. The method further includes training the field ML model with training instances which are synthetic data instances from the training dataset and corresponding training labels. The field ML model generates a predicted probability distribution of a training output class corresponding to a training instance. The method further includes adjusting a model parameter weight of the field ML model to minimize a categorical cross-entropy loss function calculated based on the generated predicted probability distribution. The trained field ML model is used to classify emission data.