Fine Dust Source Inference Using CNN-RNN Public Data Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting fine dust concentration and identifying sources lack accuracy and efficiency, particularly in real-time monitoring and emergency response, due to the complexity of fine dust generation factors and the reliance on average concentration predictions.
Innovation Solution
A method and device using local public data, combining convolutional neural networks (CNN) for image classification and recurrent neural networks (RNN) for time-series data, to convert time-series data into image datasets, infer fine dust generation grades, and apply class activation mapping (CAM) for source identification, while maintaining standard deviation without relying on root mean square error (RMSE) loss functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If observation stations are installed to measure fine dust concentrations in real time, then monitoring capability is improved, but the ability to predict and infer sources is insufficient
Solution Approach 1:
The patent combines CNN-based image classification for detecting fine dust generation situations with RNN-based time-series analysis for concentration prediction. This merging of different AI approaches enables both real-time monitoring and accurate prediction, resolving the contradiction between measurement capability and predictive reliability
Solution Approach 2:
The system introduces an intermediary AI prediction system that processes data from observation stations and adds predictive capabilities. This intermediary layer transforms raw monitoring data into actionable predictions and source inferences, enhancing the overall system reliability without replacing the original measurement infrastructure
2Measurement precision
If CNN-based image classification is used for fine dust generation detection, then detection accuracy is improved, but integration with time-series prediction is complex
Solution Approach 1:
The patent segments the prediction system into distinct functional modules: a CNN-based image classification module for detecting fine dust generation situations, and an RNN-based module for time-series concentration prediction. This segmentation allows each module to specialize in its strength while reducing integration complexity through defined interfaces
Solution Approach 2:
The system design creates a universal framework where the CNN image classification results serve as input features for the RNN prediction model. This multi-functional architecture enables the same system to perform both detection and prediction tasks, simplifying integration while maintaining high accuracy
3Reliability
If RNN-based time-series data is used for concentration prediction, then prediction capability is improved, but accuracy without proper loss functions decreases
Solution Approach 1:
The patent optimizes the RNN training by carefully selecting and tuning loss function parameters and training hyperparameters. This parameter optimization ensures that the RNN model achieves high prediction accuracy while maintaining its capability to handle time-series data effectively
Solution Approach 2:
The system implements feedback mechanisms where prediction results are continuously evaluated and used to refine the model. This feedback loop maintains high prediction accuracy by continuously optimizing the model based on actual performance metrics and emerging patterns in the data
4Measurement precision
If class activation mapping is applied for source inference, then source identification accuracy is improved, but computational requirements increase
Solution Approach 1:
The patent applies class activation mapping selectively to identify the most relevant features for source inference, rather than processing the entire dataset. This partial action approach maintains high source identification accuracy while significantly reducing computational energy requirements by focusing only on the most informative activations
Data Source
AI summary
Disclosed are a method of predicting a fine dust concentration and inferring a fine dust source by using local public data and a prediction and inference device. The method of predicting a fine dust concentration and inferring a fine dust source by using local public data includes generating time-series data related to fine dust by collecting public data in a specific region in a predetermined chronological order and determining whether fine dust is generated in the specific region by converting pieces of time-series data collected in consecutive times into an image dataset for training and by training the image dataset for training in a convolution neural network (CNN)-based image classification model.