Continuous-Time Sensor Prediction for Process Plant Time Series
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The discrepancy between the continuous nature of process engineering and the discrete operation of digital applications poses a challenge in accurately predicting time series data for process optimization, leading to inaccurate models due to missing data and complex, computationally intensive manual modeling.
Innovation Solution
A data-driven, continuous-time model combined with a differential equation solver is used to bridge the gap between continuous and discrete modeling, allowing for accurate prediction of time series data by initializing parameter structures in a continuous space and converting models to discrete states using a differential equation solver.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a discrete-time model is used to match digital application data, then the model can be trained using discrete data points, but important continuous process information is lost and parameters require very high numerical accuracy
Solution Approach 1:
The patent transitions from discrete-time modeling to continuous-time modeling, changing the temporal dimension from discrete samples to continuous functions. This allows the model to capture continuous process information while still being trained on discrete sensor data by using differential equations that describe the underlying continuous dynamics of the process.
Solution Approach 2:
The patent changes the parameter space from discrete time steps to continuous time parameters in differential equations. By using continuous-time state-space models with differential equations, the parameters represent continuous physical quantities and rates of change, eliminating the need for very high numerical accuracy in discrete parameters while preserving continuous process information.
2Reliability
If manual physical modeling is performed to create continuous-time models, then the model can accurately represent continuous processes, but the process becomes complex, computationally intensive, and prone to errors
Solution Approach 1:
The patent enables the model to automatically learn continuous-time dynamics directly from discrete sensor data without requiring manual physical modeling. The continuous-time state-space model with differential equations is trained autonomously on available data, eliminating the need for expert manual derivation of physical equations while maintaining model accuracy and reducing complexity.
3Ease of operation
If discrete data points are used to represent continuous processes, then the data can be processed by digital applications, but the continuous nature of the process is lost leading to inaccurate predictions
Solution Approach 1:
The patent introduces continuous-time differential equations as an intermediary layer between discrete sensor data and process predictions. The differential equations act as a mediator that transforms discrete data points into continuous process representations, allowing digital applications to process data while preserving the continuous nature of the underlying process for accurate predictions.
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a computer-implemented method and corresponding device for predicting time series data from at least one sensor (S) of a process plant, comprising the steps: - receiving discrete input data (DD) from the at least one sensor (S) and/or a monitoring device (Ü) over a time period, - receiving a pre-processed, dynamic model (M) which is at least partially data-based and characterizes a continuous process (P), - calculating an initial state of the model (M), - defining at least one prediction period with associated discrete time steps, - evaluating the continuous model (M) for the defined prediction periods using a differential equation solver (DES), - outputting discrete expected values of time series data for the defined prediction periods.The method according to the invention allows for an improvement in the operation and/or planning of a process engineering plant using the predicted time series data.