Abnormal-Tag Report Generation for LLM Sensor Summaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Processing systems in facilities like chemical plants face high computational loads when using large language models to generate reports on operating status due to the large number of measurement data items, which can overwhelm the model.
Innovation Solution
A report generation system that assigns unique tags to measurement data items, identifies abnormal behavior, and generates prompts for a large language model to describe only abnormal data, reducing computational load by excluding normal data and using reference data for comparison.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If all measurement data items are input into the large language model to generate a report on operating status, then the report comprehensiveness is improved, but the computational load on the large language model increases
Solution Approach 1:
The patent extracts only the abnormal measurement data items from the complete set of measurement data. The abnormality determination unit identifies data items that deviate from normal ranges or patterns, and only these extracted abnormal items are input to the large language model for report generation. This extraction principle resolves the contradiction by eliminating normal data that would increase computational load without adding valuable information to the report.
Solution Approach 2:
The patent segments the measurement data processing into distinct stages: data collection, abnormality determination, and report generation. By dividing the processing pipeline, the system can filter data before it reaches the large language model, ensuring that only relevant abnormal data segments are processed by the computationally intensive model, thus reducing overall computational load while maintaining report quality.
2Productivity
If measurement data items are filtered to reduce computational load, then the processing efficiency is improved, but the report information completeness may deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the abnormality determination unit continuously monitors measurement data against established normal ranges and patterns. This feedback loop ensures that only data items exhibiting actual abnormalities are selected for report generation, preventing the loss of important information while maintaining high processing efficiency. The feedback mechanism validates that filtering does not compromise information completeness.
Solution Approach 2:
The patent performs preliminary abnormality determination before the large language model processing stage. By pre-identifying and flagging abnormal data items in advance, the system prepares a curated dataset that ensures information completeness is maintained while enabling efficient subsequent processing by the large language model, thus resolving the contradiction between efficiency and completeness.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A processing system includes sensors that monitor its operating status. Each of multiple measurement data items corresponding to measurement values collected by the sensors is assigned a tag unique to the sensor that collects the measurement data item. The report generation system includes a processor configured to identify, as an abnormality tag, a tag for a measurement data item that indicated abnormal behavior during a report target period among the measurement data items, generate a prompt including a report target data item and an instruction to generate a description of a behavior reflected in the report target data item, the report target data item corresponding to a measurement data item acquired during the report target period and assigned a tag matching the abnormality tag, acquire the description from the large language model by inputting the prompt into the large language model, and generate the report using the description.