Moving-Horizon State Initializer for Artificial Pancreas Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current state estimators for artificial pancreas systems face challenges with sensor recalibrations, asynchronous data handling, and plant-model mismatches, leading to suboptimal insulin delivery and potential overdelivery or offset errors.
Innovation Solution
A moving-horizon optimization strategy that fits a continuous-time function to CGM data, accommodating sensor recalibrations by including discontinuities and handling asynchronous data sampling, while ignoring recalibration discontinuities during sampling, to construct a model state without relying on a specific model, thus mitigating plant-model mismatches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a recursive state estimator is used to initialize MPC predictions, then the implementation is straightforward, but sensor recalibrations cause lively dynamics leading to meaningless predictions and potential overdelivery
Solution Approach 1:
The patent applies a dynamic approach by using a moving-horizon state initializer that adapts to changing conditions. Instead of a fixed recursive estimator, the system continuously updates the initial state estimates by optimizing over a moving horizon of recent data, allowing it to adapt to recalibrations without causing lively dynamics. The initializer adjusts its window and parameters based on current system conditions, maintaining reliability after recalibration events.
2Ease of manufacture
If a fixed sample-period state estimator is used, then implementation is simple, but asynchronous CGM data causes over-estimation of rate of change and veering off CGM trajectory
Solution Approach 1:
The patent implements a dynamic moving-horizon approach that adapts to asynchronous data arrival. The optimizer dynamically adjusts the horizon window and re-optimizes state estimates based on the actual timing of CGM measurements. This allows the system to handle variable sample periods and asynchronous data without over-estimating rates of change, keeping predictions on the CGM trajectory.
Solution Approach 2:
The system uses feedback from the actual CGM measurement times and values to continuously refine state estimates. The moving-horizon optimizer incorporates feedback about data arrival times and uses this information to adjust its estimation, preventing veering off the trajectory that occurs with fixed-sample-period estimators.
3Ease of manufacture
If model-based recursive state estimators are used, then implementation is straightforward, but plant-model mismatch prevents offset-free estimates even in steady-state
Solution Approach 1:
The patent creates a moving-horizon copy of the recent system behavior and optimizes state estimates to match this copied behavior rather than relying on the potentially mismatched model. By copying and optimizing over recent actual system responses, the system achieves offset-free estimates without depending on perfect model accuracy.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A state-estimator for the estimation or initialization of the state of a discrete-time state-space dynamical model based on sensor measurements of the model output, comprising fitting a continuous-time function to acquired sensor measurement data-points of each model output, and subsequently sampling the continuous time function at exactly the sample-period of the state-space dynamic model for which the state is being estimated or initialized, in order to construct a model state via a synthesized output trajectory.