Discrete Sensor Data Prediction Using Temporal Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing prediction techniques for discrete sensor data, particularly in unbalanced binary or discrete sensor data scenarios, struggle to accurately forecast due to data sparsity and imbalance, leading to inaccuracies and system failures when sensors malfunction.
Innovation Solution
A method and system that introduce temporal uncertainty into sparse discrete sensor data by converting it into pseudo-continuous data, using LSTM techniques to predict corrected continuous data based on historical patterns, and then reconverting it back to discrete format, ensuring continuous data flow even during sensor failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing forecasting techniques are used for discrete sensor data, then the prediction process can be performed, but the prediction accuracy deteriorates due to data sparsity and imbalance
Solution Approach 1:
The patent transforms discrete sensor data parameters into continuous parameters by introducing temporal uncertainty. Instead of working with binary discrete values, the system converts them to continuous values within ranges, enabling the use of LSTM models that require continuous input data for accurate predictions.
Solution Approach 2:
The patent introduces an intermediary transformation process that converts discrete sensor data into continuous data representations. This intermediary step involves creating pseudo-continuous data from discrete values, which then serves as input for the LSTM prediction model, bridging the gap between discrete data and continuous prediction requirements.
2Measurement precision
If neural network techniques such as RNN or LSTM are used for unbalanced discrete sensor data, then prediction can be performed, but the prediction accuracy deteriorates because most data is 'No' and few samples of 'Yes' are available
Solution Approach 1:
The patent changes the parameter representation from discrete binary values to continuous values with temporal uncertainty. By representing discrete 'Yes' and 'No' as continuous ranges rather than fixed values, the model can better handle the imbalance and learn meaningful patterns from the minority class samples.
Solution Approach 2:
The patent adds a temporal dimension to the data representation by introducing uncertainty ranges over time. This transforms the problem from a static discrete classification task to a dynamic continuous prediction task, allowing the LSTM model to leverage temporal patterns to improve prediction accuracy for imbalanced data.
3Reliability
If sensor networks operate without forecasting techniques, then the system structure remains simple, but the system reliability deteriorates when sensors fail leading to missing or corrupt data
Solution Approach 1:
The patent applies preliminary action by using LSTM models to predict future sensor values before actual sensor failures occur. The system continuously learns from historical data and maintains prediction capabilities that can fill gaps when sensors fail, ensuring continuous data flow without requiring complex real-time intervention systems.
Data Source
AI summary
This disclosure relates generally to a method and system for prediction of correct discrete sensor data, thus enabling continuous flow of data even when a discrete sensor fails. The activities of humans/subjects, housed in a smart environment is continuously monitored by plurality of non-intrusive discrete sensors embedded in living infrastructure. The collected discrete sensor data is usually sparse and largely unbalanced, wherein most of the discrete sensor data is ‘No’ and comparatively only a few samples of ‘Yes’, hence making prediction very challenging. The proposed prediction techniques based on introduction of temporal uncertainty is performed in several stages which includes pre-processing of received discrete sensor data, introduction of temporal uncertainty techniques followed by prediction based on neural network techniques of learning pattern using historical data.