Forecasting Irregular Time Series with Missing Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to effectively forecast future values in multivariate irregularly sampled time series data with missing values, as they do not account for the irregular sampling intervals and missing data points, leading to inaccurate predictions.
Innovation Solution
A processor uses a recurrent neural network, specifically gated recurrent units, to input non-missing values, mask values, and time interval values to predict missing values and forecast future data points, including the timing of next snapshots, by employing a custom loss function and imputation techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing forecasting methods are used on irregularly sampled time series data with missing values, then the method is simple to implement, but the forecasting accuracy deteriorates due to inability to account for irregular sampling intervals and missing data points
Solution Approach 1:
The patent applies preliminary action by pre-processing the time series data to extract and encode temporal characteristics (sampling intervals, missing value patterns, seasonality indicators) before feeding them to the forecasting model. This preprocessing step prepares the data in advance to account for irregular sampling and missing values, improving forecasting accuracy without requiring complex modifications to the core model architecture.
Solution Approach 2:
The patent introduces intermediary components including temporal encoding layers and attention mechanisms that act as mediators between the raw irregularly sampled data and the forecasting model. These intermediaries transform the irregular time series into a format suitable for standard forecasting architectures, bridging the gap between complex data characteristics and simpler model structures.
2Measurement precision
If a recurrent neural network is used to capture long-term dependencies in irregularly sampled time series, then the forecasting accuracy improves, but the computational time and complexity increase
Solution Approach 1:
The patent segments the time series processing into distinct components: temporal pattern extraction, seasonality detection, and trend analysis. By dividing the complex task of capturing long-term dependencies into separate modular components, the system achieves accurate forecasting while reducing overall computational burden through specialized efficient implementations of each segment.
Solution Approach 2:
The patent leverages periodic action by detecting and exploiting seasonal patterns in the time series data. By identifying periodic components and using them to guide the forecasting process, the model can make accurate predictions about future values at corresponding seasonal points without requiring excessively long training sequences, thus reducing computational time while maintaining accuracy.
3Loss of information
If imputation techniques are applied to fill missing values in time series data, then the completeness of data improves, but the risk of introducing biased predictions increases
Solution Approach 1:
The patent applies feedback by using the observed data patterns and temporal relationships to dynamically adjust imputation strategies. The model learns from the actual data distribution and uses this feedback to generate imputed values that are consistent with the underlying patterns, reducing bias introduction while maintaining data completeness. The feedback loop ensures imputed values align with observed behavior.
Solution Approach 2:
The patent changes parameters by adapting imputation methods based on local data characteristics rather than using fixed imputation rules. By adjusting imputation parameters dynamically according to the specific temporal context, seasonality, and missingness patterns, the system fills missing values in a way that preserves prediction reliability while achieving data completeness.
Data Source
AI summary
In an approach for forecasting in multivariate irregularly sampled time series, a processor receives time series data having one or more missing values. A processor determines, from the time series data, non-missing values present in the time series data. A processor determines, from the time series data, zero or more mask values for the time series data. A processor determines time interval values. A processor inputs the one or more missing values, the non-missing values, the zero or more mask values, and the time interval values into a recurrent neural network. A processor determines a predicted value for the one or more missing values.