Residual-Based Risk Prediction for Industrial Process Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the risk of uncertain quantities in industrial processes and electrical networks are inadequate, especially when nonzero knowledge about the behavior of these quantities is available.
Innovation Solution
A computer-implemented method that refines the risk assessment by using a prediction model to estimate future values of uncertain quantities, determining residuals based on past values, and calculating a new risk level that accounts for the uncertainty in the probability distribution of these residuals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If stochastic methods are used to estimate risk with zero prior knowledge, then the method is universally applicable, but the risk estimation is overly conservative and less accurate
Solution Approach 1:
The patent applies preliminary action by incorporating prior knowledge about the probability distribution of uncertain quantities before performing risk estimation. Instead of starting from zero knowledge, the method pre-establishes distribution characteristics (mean, variance, or other parameters) based on historical data or expert knowledge, which then guides the subsequent risk analysis and reduces conservatism in the estimates
Solution Approach 2:
The patent changes parameters by transitioning from assuming zero knowledge (uniform distribution) to utilizing specific distribution parameters (mean, variance, skewness, kurtosis, or full probability density functions) that characterize the uncertain quantities. This parameter enrichment allows for more accurate risk estimation while maintaining computational tractability through analytical solutions
2Measurement precision
If prior knowledge about uncertain quantities is utilized, then risk estimation accuracy improves, but the method becomes more complex
Solution Approach 1:
The patent employs parameter changes by systematically incorporating distribution parameters (mean, variance, and higher-order moments) to characterize uncertain quantities. This allows the method to adapt to different levels of available knowledge while maintaining a unified analytical framework, improving measurement precision without proportionally increasing complexity
Solution Approach 2:
The patent applies segmentation by dividing the prior knowledge into distinct components: distribution type identification, parameter estimation (mean, variance, higher-order moments), and uncertainty characterization. This modular approach allows each component to be handled separately and combined systematically, managing complexity while improving accuracy
3Reliability
If conservative risk estimates are used, then safety margins are increased, but operational efficiency decreases
Solution Approach 1:
The patent changes parameters by replacing conservative uniform distribution assumptions with informed probability distributions that reflect actual system behavior. This allows for tighter, more realistic risk bounds that maintain safety assurance while reducing unnecessary safety margins, thereby improving operational efficiency
Solution Approach 2:
The patent applies feedback by using historical data and observed system behavior to continuously refine the probability distribution parameters of uncertain quantities. This feedback loop enables the risk estimates to become progressively less conservative as more information becomes available, balancing safety and efficiency dynamically
Data Source
Figure 1a
Figure 1b
Figure 2a~2b
AI summary
A computer-implemented method (100) for refining the risk α that an uncertain quantity x in an industrial process (1a) and/or electrical network (1b) will exceed a given threshold value x̃, the method comprising the steps of: • providing (110) an initial desired value α∗ of the risk α; • providing (120) a prediction model (2) that is configured to predict, from values xt, xt-1, ..., xt-k of the quantity x, future values xt+1, ..., xt+l of this quantity x; • providing (130) a history x# of measured past values xt,xt-1, ..., xt-k of the quantity x; • determining (140), for this history x#, differences between values x̂t,x̂t-1, ..., x̂t-k predicted by the prediction model and the measured past values xt, xt-1...,xt-k as residuals (3); • determining (150) an estimated probability density function (4) of these residuals (3); • determining (160), based at least in part on the history x#, and/or on the residuals (3), a confidence set to which the estimated probability density function (4) belongs, said confidence set comprising probability density functions that would be equally plausible given the history x# and/or the residuals (3); and • determining (170), based at least in part on the dimension of the confidence set , a new risk α' such that, if the risk α that x exceeds x̃ is less than α', it is guaranteed that this risk α will not exceed the initial desired value α∗ even if the residuals should behave according to any one of the probability density functions in the confidence set D instead of the estimated probability density function (4).