Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867573
Amin Vahidi-Moghaddam, Kaian Chen, Zhaojian Li, Yan Wang, Kai Wu
Model predictive control (MPC) is a state-of-the-art control method that can explicitly tackle system constraints. However, its high computational cost still remains an open challenge for embedded systems. To achieve satisfactory performance with manageable computational complexity, a spatial temporal filter (STF)-based data-driven predictive control framework is developed to systematically identify system dynamics and subsequently learn the MPC policy using STF-based function approximations. Specifically, an online nonlinear system identification method that satisfies persistence of excitation (PE) is developed by using a discrete-time concurrent learning technique. An STF-based function approximation is then employed to learn the nonlinear MPC (NMPC) policy based on the identified model. Furthermore, a discrete-time robust control barrier function (RCBF) is introduced to guarantee system safety in the presence of additive disturbances and system identification errors. Finally, simulations on the cart inverted pendulum are performed to demonstrate the efficacy of the proposed control synthesis.
{"title":"Data-Driven Safe Predictive Control Using Spatial Temporal Filter-based Function Approximators","authors":"Amin Vahidi-Moghaddam, Kaian Chen, Zhaojian Li, Yan Wang, Kai Wu","doi":"10.23919/ACC53348.2022.9867573","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867573","url":null,"abstract":"Model predictive control (MPC) is a state-of-the-art control method that can explicitly tackle system constraints. However, its high computational cost still remains an open challenge for embedded systems. To achieve satisfactory performance with manageable computational complexity, a spatial temporal filter (STF)-based data-driven predictive control framework is developed to systematically identify system dynamics and subsequently learn the MPC policy using STF-based function approximations. Specifically, an online nonlinear system identification method that satisfies persistence of excitation (PE) is developed by using a discrete-time concurrent learning technique. An STF-based function approximation is then employed to learn the nonlinear MPC (NMPC) policy based on the identified model. Furthermore, a discrete-time robust control barrier function (RCBF) is introduced to guarantee system safety in the presence of additive disturbances and system identification errors. Finally, simulations on the cart inverted pendulum are performed to demonstrate the efficacy of the proposed control synthesis.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125374367","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867601
Kuan-Yu Tseng, J. Shamma, G. Dullerud
We propose a novel gradient-based reprogrammable iterative learning control (GRILC) framework for autonomous systems. Performance of trajectory following in autonomous systems is often limited by mismatch between a complex actual model and a simplifed nominal model used in controller design. To overcome this issue, we develop the GRILC framework with offline optimization using the information of the nominal model and the actual trajectory, and online system implementation. In addition, a reprogrammable learning strategy is introduced which incorporates directly the learned primitives stored into a library, for future motion planning. The proposed method is applied to the autonomous time-trialing example. The simulation and experimental results illustrate the effectiveness and robustness of the proposed approach.
{"title":"Low-Fidelity Gradient Updates for High-Fidelity Reprogrammable Iterative Learning Control","authors":"Kuan-Yu Tseng, J. Shamma, G. Dullerud","doi":"10.23919/ACC53348.2022.9867601","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867601","url":null,"abstract":"We propose a novel gradient-based reprogrammable iterative learning control (GRILC) framework for autonomous systems. Performance of trajectory following in autonomous systems is often limited by mismatch between a complex actual model and a simplifed nominal model used in controller design. To overcome this issue, we develop the GRILC framework with offline optimization using the information of the nominal model and the actual trajectory, and online system implementation. In addition, a reprogrammable learning strategy is introduced which incorporates directly the learned primitives stored into a library, for future motion planning. The proposed method is applied to the autonomous time-trialing example. The simulation and experimental results illustrate the effectiveness and robustness of the proposed approach.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"157 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115179075","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867763
A. Acernese, A. Yerudkar, C. D. Vecchio
With the advent of industry 4.0, machine learning (ML) methods have mainly been applied to design condition-based maintenance strategies to improve the detection of failure precursors and forecast degradation. However, in real-world scenarios, relevant features unraveling the actual machine conditions are often unknown, posing new challenges in addressing fault diagnosis problems. Moreover, ML approaches generally need ad-hoc feature extractions, involving the development of customized models for each case study. Finally, the early substitution of key mechanical components to avoid costly breakdowns challenge the collection of sizable significant data sets to train fault detection (FD) systems. To address these issues, this paper proposes a new unsupervised FD method based on double deep-Q network (DDQN) with prioritized experience replay (PER). We validate the effectiveness of the proposed algorithm on real steel plant data. Lastly, we compare the performance of our method with other FD methods showing its viability.
{"title":"A Novel Reinforcement Learning-based Unsupervised Fault Detection for Industrial Manufacturing Systems","authors":"A. Acernese, A. Yerudkar, C. D. Vecchio","doi":"10.23919/ACC53348.2022.9867763","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867763","url":null,"abstract":"With the advent of industry 4.0, machine learning (ML) methods have mainly been applied to design condition-based maintenance strategies to improve the detection of failure precursors and forecast degradation. However, in real-world scenarios, relevant features unraveling the actual machine conditions are often unknown, posing new challenges in addressing fault diagnosis problems. Moreover, ML approaches generally need ad-hoc feature extractions, involving the development of customized models for each case study. Finally, the early substitution of key mechanical components to avoid costly breakdowns challenge the collection of sizable significant data sets to train fault detection (FD) systems. To address these issues, this paper proposes a new unsupervised FD method based on double deep-Q network (DDQN) with prioritized experience replay (PER). We validate the effectiveness of the proposed algorithm on real steel plant data. Lastly, we compare the performance of our method with other FD methods showing its viability.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116006883","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867720
Neelay Junnarkar, Emmanuel Sin, P. Seiler, D. Philbrick, M. Arcak
This letter considers assignment problems consisting of n pursuers attempting to intercept n targets. We consider stationary targets as well as targets maneuvering toward an asset. The assignment algorithm relies on an n × n cost matrix where entry (i, j) is the minimum time for pursuer i to intercept target j. Each entry of this matrix requires the solution of a nonlinear optimal control problem. This subproblem is computationally intensive and hence the computational cost of the assignment is dominated by the construction of the cost matrix. We propose to use neural networks for function approximation of the minimum time until intercept. The neural networks are trained offline, thus allowing for real-time online construction of cost matrices. Moreover, the function approximators have sufficient accuracy to obtain reasonable solutions to the assignment problem. In most cases, the approximators achieve assignments with optimal worst case intercept time. The proposed approach is demonstrated on several examples with increasing numbers of pursuers and targets.
{"title":"Fast Assignment in Asset-Guarding Engagements using Function Approximation","authors":"Neelay Junnarkar, Emmanuel Sin, P. Seiler, D. Philbrick, M. Arcak","doi":"10.23919/ACC53348.2022.9867720","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867720","url":null,"abstract":"This letter considers assignment problems consisting of n pursuers attempting to intercept n targets. We consider stationary targets as well as targets maneuvering toward an asset. The assignment algorithm relies on an n × n cost matrix where entry (i, j) is the minimum time for pursuer i to intercept target j. Each entry of this matrix requires the solution of a nonlinear optimal control problem. This subproblem is computationally intensive and hence the computational cost of the assignment is dominated by the construction of the cost matrix. We propose to use neural networks for function approximation of the minimum time until intercept. The neural networks are trained offline, thus allowing for real-time online construction of cost matrices. Moreover, the function approximators have sufficient accuracy to obtain reasonable solutions to the assignment problem. In most cases, the approximators achieve assignments with optimal worst case intercept time. The proposed approach is demonstrated on several examples with increasing numbers of pursuers and targets.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122990677","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867190
Raunak Srivastava, Rolif Lima, K. Das
Autonomous capture of an unknown non-cooperative aerial target is a complex task requiring real-time target localization, trajectory prediction and control. The current work presents an estimation and control system for a quadrotor in order to track and grab a dynamic aerial target. A Kalman filter is used to estimate and predict the target’s pose in real time, which is further used as reference by a Model Predictive Controller for tracking and grabbing the target while chasing it. The efficacy of the proposed controller is demonstrated through repeated trials with varying initial conditions and target maneuvers. The controller is compared with a conventional PD controller. The accuracy of estimation algorithms and the ability of the proposed controller to neutralize the target is demonstrated by means of simulations in gazebo.
{"title":"Aerial Interception of Non-Cooperative Intruder using Model Predictive Control","authors":"Raunak Srivastava, Rolif Lima, K. Das","doi":"10.23919/ACC53348.2022.9867190","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867190","url":null,"abstract":"Autonomous capture of an unknown non-cooperative aerial target is a complex task requiring real-time target localization, trajectory prediction and control. The current work presents an estimation and control system for a quadrotor in order to track and grab a dynamic aerial target. A Kalman filter is used to estimate and predict the target’s pose in real time, which is further used as reference by a Model Predictive Controller for tracking and grabbing the target while chasing it. The efficacy of the proposed controller is demonstrated through repeated trials with varying initial conditions and target maneuvers. The controller is compared with a conventional PD controller. The accuracy of estimation algorithms and the ability of the proposed controller to neutralize the target is demonstrated by means of simulations in gazebo.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"92 2","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114040695","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867288
Erfaun Noorani, J. Baras
Robustness is a key enabler of real-world applications of Reinforcement Learning (RL). The robustness properties of risk-sensitive controllers have long been established. We investigate risk-sensitive Reinforcement Learning (as a generalization of risk-sensitive stochastic control), by theoretically analyzing the risk-sensitive exponential (exponential of the total reward) criteria, and the benefits and improvements the introduction of risk-sensitivity brings to conventional RL. We provide a probabilistic interpretation of (I) the risk-sensitive exponential, (II) the risk-neutral expected cumulative reward, and (III) the maximum entropy Reinforcement Learning objectives, and explore their connections from a probabilistic perspective. Using Probabilistic Graphical Models (PGM), we establish that in the RL setting, maximization of the risk-sensitive exponential criteria is equivalent to maximizing the probability of taking an optimal action at all time-steps during an episode. We show that the maximization of the standard risk-neutral expected cumulative return is equivalent to maximizing a lower bound, particularly the Evidence lower Bound, on the probability of taking an optimal action at all time-steps during an episode. Furthermore, we show that the maximization of the maximum-entropy Reinforcement Learning objective is equivalent to maximizing a lower bound on the probability of taking an optimal action at all time-steps during an episode, where the lower bound corresponding to the maximum entropy objective is tighter and smoother than the lower bound corresponding to the expected cumulative return objective. These equivalences establish the benefits of risk-sensitive exponential objective and shed lights on previously postulated regularized objectives, such as maximum entropy. The utilization of a PGM model, coupled with exponential criteria, offers a number of advantages (e.g. facilitate theoretical analysis and derivation of bounds).
{"title":"A Probabilistic Perspective on Risk-sensitive Reinforcement Learning","authors":"Erfaun Noorani, J. Baras","doi":"10.23919/ACC53348.2022.9867288","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867288","url":null,"abstract":"Robustness is a key enabler of real-world applications of Reinforcement Learning (RL). The robustness properties of risk-sensitive controllers have long been established. We investigate risk-sensitive Reinforcement Learning (as a generalization of risk-sensitive stochastic control), by theoretically analyzing the risk-sensitive exponential (exponential of the total reward) criteria, and the benefits and improvements the introduction of risk-sensitivity brings to conventional RL. We provide a probabilistic interpretation of (I) the risk-sensitive exponential, (II) the risk-neutral expected cumulative reward, and (III) the maximum entropy Reinforcement Learning objectives, and explore their connections from a probabilistic perspective. Using Probabilistic Graphical Models (PGM), we establish that in the RL setting, maximization of the risk-sensitive exponential criteria is equivalent to maximizing the probability of taking an optimal action at all time-steps during an episode. We show that the maximization of the standard risk-neutral expected cumulative return is equivalent to maximizing a lower bound, particularly the Evidence lower Bound, on the probability of taking an optimal action at all time-steps during an episode. Furthermore, we show that the maximization of the maximum-entropy Reinforcement Learning objective is equivalent to maximizing a lower bound on the probability of taking an optimal action at all time-steps during an episode, where the lower bound corresponding to the maximum entropy objective is tighter and smoother than the lower bound corresponding to the expected cumulative return objective. These equivalences establish the benefits of risk-sensitive exponential objective and shed lights on previously postulated regularized objectives, such as maximum entropy. The utilization of a PGM model, coupled with exponential criteria, offers a number of advantages (e.g. facilitate theoretical analysis and derivation of bounds).","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114450696","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867658
Vlad Mihaly, Mircea Şuşcă, E. Dulf, P. Dobra
Fractional order calculus in modelling and control applications has been increasingly popular in every scientific field due to its versatility and superiority in various instances. However, the main issue of fractional order elements is related to their implementation. Usually, integer order elements are used through different approximation methods to implement such elements. None of the existing methods discuss the introduction of a fractional element in a fixed-structure robust synthesis method. MATLAB realp objects do not allow the use of other than classical arithmetic operations, while existing approximations use exponential terms. To overcome this shortcoming, this work proposes and offers an algorithm for a method of approximation useful in robust controller synthesis.
{"title":"Approximating the Fractional-Order Element for the Robust Control Framework","authors":"Vlad Mihaly, Mircea Şuşcă, E. Dulf, P. Dobra","doi":"10.23919/ACC53348.2022.9867658","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867658","url":null,"abstract":"Fractional order calculus in modelling and control applications has been increasingly popular in every scientific field due to its versatility and superiority in various instances. However, the main issue of fractional order elements is related to their implementation. Usually, integer order elements are used through different approximation methods to implement such elements. None of the existing methods discuss the introduction of a fractional element in a fixed-structure robust synthesis method. MATLAB realp objects do not allow the use of other than classical arithmetic operations, while existing approximations use exponential terms. To overcome this shortcoming, this work proposes and offers an algorithm for a method of approximation useful in robust controller synthesis.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"112 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122116427","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867408
Azwirman Gusrialdi, Z. Qu
This paper considers a cooperative control problem in presence of unknown attacks. The attacker aims at destabilizing the consensus dynamics by intercepting the system’s communication network and corrupting its local state feedback. We first revisit the virtual network based resilient control proposed in our previous work and provide a new interpretation and insights into its implementation. Based on these insights, a novel distributed algorithm is presented to detect and identify the compromised communication links. It is shown that it is not possible for the adversary to launch a harmful and stealthy attack by only manipulating the physical states being exchanged via the network. In addition, a new virtual network is proposed which makes it more difficult for the adversary to launch a stealthy attack even though it is also able to manipulate information being exchanged via the virtual network. A numerical example demonstrates that the proposed control framework achieves simultaneously resilient operation and real-time attack identification.
{"title":"Cooperative Systems in Presence of Cyber-Attacks: A Unified Framework for Resilient Control and Attack Identification","authors":"Azwirman Gusrialdi, Z. Qu","doi":"10.23919/ACC53348.2022.9867408","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867408","url":null,"abstract":"This paper considers a cooperative control problem in presence of unknown attacks. The attacker aims at destabilizing the consensus dynamics by intercepting the system’s communication network and corrupting its local state feedback. We first revisit the virtual network based resilient control proposed in our previous work and provide a new interpretation and insights into its implementation. Based on these insights, a novel distributed algorithm is presented to detect and identify the compromised communication links. It is shown that it is not possible for the adversary to launch a harmful and stealthy attack by only manipulating the physical states being exchanged via the network. In addition, a new virtual network is proposed which makes it more difficult for the adversary to launch a stealthy attack even though it is also able to manipulate information being exchanged via the virtual network. A numerical example demonstrates that the proposed control framework achieves simultaneously resilient operation and real-time attack identification.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122126087","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867562
Weidi Yin, A. Tivay, J. Hahn
❖ Hemorrhage resuscitation and intravenous(IV) sedation can interfere with each other in a conflicting manner, possibly driving a patient to a dangerous physiological state: (i) hemorrhage resuscitation can weaken sedation effect by lowering sedative concentration in the blood; (ii) sedation can weaken hemorrhage resuscitation effect by inducing vaso-/veno-dilation and low blood pressure.
{"title":"Hemodynamic Monitoring via Model-Based Extended Kalman Filtering: Hemorrhage Resuscitation and Sedation Case Study","authors":"Weidi Yin, A. Tivay, J. Hahn","doi":"10.23919/ACC53348.2022.9867562","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867562","url":null,"abstract":"❖ Hemorrhage resuscitation and intravenous(IV) sedation can interfere with each other in a conflicting manner, possibly driving a patient to a dangerous physiological state: (i) hemorrhage resuscitation can weaken sedation effect by lowering sedative concentration in the blood; (ii) sedation can weaken hemorrhage resuscitation effect by inducing vaso-/veno-dilation and low blood pressure.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129540921","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-08DOI: 10.23919/ACC53348.2022.9867232
Matthew F. Singh, Michael Wang, Michael W. Cole, ShiNung Ching
System identification poses a significant bottleneck to characterizing and controlling complex systems. This challenge is greatest when both the system states and parameters are not directly accessible, leading to a dual-estimation problem. Current approaches to such problems are limited in their ability to scale with many-parameter systems, as often occurs in networks. In the current work, we present a new, computationally efficient approach to treat large dual-estimation problems. In this work, we derive analytic back-propagated gradients for the Prediction Error Method which enables efficient and accurate identification of large systems. The PEM approach consists of directly integrating state estimation into a dual-optimization objective, leaving a differentiable cost/error function only in terms of the unknown system parameters, which we solve using numerical gradient/Hessian methods. Intuitively, this approach consists of solving for the parameters that generate the most accurate state estimator (Extended/Cubature Kalman Filter). We demonstrate that this approach is at least as accurate in state and parameter estimation as joint Kalman Filters (Extended/Unscented/Cubature) and Expectation-Maximization, despite lower complexity. We demonstrate the utility of our approach by inverting anatomically-detailed individualized brain models from human magnetoencephalography (MEG) data.
{"title":"Efficient identification for modeling high-dimensional brain dynamics","authors":"Matthew F. Singh, Michael Wang, Michael W. Cole, ShiNung Ching","doi":"10.23919/ACC53348.2022.9867232","DOIUrl":"https://doi.org/10.23919/ACC53348.2022.9867232","url":null,"abstract":"System identification poses a significant bottleneck to characterizing and controlling complex systems. This challenge is greatest when both the system states and parameters are not directly accessible, leading to a dual-estimation problem. Current approaches to such problems are limited in their ability to scale with many-parameter systems, as often occurs in networks. In the current work, we present a new, computationally efficient approach to treat large dual-estimation problems. In this work, we derive analytic back-propagated gradients for the Prediction Error Method which enables efficient and accurate identification of large systems. The PEM approach consists of directly integrating state estimation into a dual-optimization objective, leaving a differentiable cost/error function only in terms of the unknown system parameters, which we solve using numerical gradient/Hessian methods. Intuitively, this approach consists of solving for the parameters that generate the most accurate state estimator (Extended/Cubature Kalman Filter). We demonstrate that this approach is at least as accurate in state and parameter estimation as joint Kalman Filters (Extended/Unscented/Cubature) and Expectation-Maximization, despite lower complexity. We demonstrate the utility of our approach by inverting anatomically-detailed individualized brain models from human magnetoencephalography (MEG) data.","PeriodicalId":366299,"journal":{"name":"2022 American Control Conference (ACC)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-06-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129700098","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：