Showing 1–5 of 5 results for author: Moura, S J

Safe Learning MPC with Limited Model Knowledge and Data

Authors: Aaron Kandel, Scott J. Moura

Abstract: This paper presents an end-to-end framework for safe learning-based control (LbC) using nonlinear stochastic MPC and distributionally robust optimization (DRO). This work is motivated by several open challenges in LbC literature. In particular, many control-theoretic LbC methods require subject matter expertise in order to translate their own safety guarantees, often manifested as preexisting data… ▽ More This paper presents an end-to-end framework for safe learning-based control (LbC) using nonlinear stochastic MPC and distributionally robust optimization (DRO). This work is motivated by several open challenges in LbC literature. In particular, many control-theoretic LbC methods require subject matter expertise in order to translate their own safety guarantees, often manifested as preexisting data of safe trajectories or structural model knowledge. In this paper, we focus on LbC where the controller is applied directly to a system of which it has no or extremely limited direct experience, towards safety during \textit{tabula-rasa} or ``\textit{blank slate''} model-based learning and control as a challenging case for validation. This explores the boundary of the status-quo in control theory relating to requirements for subject matter expertise. We show under basic and limited assumptions on the underlying problem, we can translate probabilistic guarantees on feasibility to nonlinear systems using results in stochastic MPC and DRO literature whose relevance we formally extend in a mathematical analysis. We also present a coupled and intuitive formulation for persistence of excitation (PoE), and illustrate the connection between PoE and applicability of the proposed method. Our case studies of vehicle obstacle avoidance and safe extreme fast charging of lithium-ion batteries reveal powerful empirical results supporting the underlying DRO theory. Our method is widely applicable within the LbC domain to, for example, airborne wind energy systems, vehicle obstacle avoidance, and energy storage systems management. It is also applicable to quantifying uncertainty beyond the LbC case. △ Less

Submitted 21 August, 2023; v1 submitted 1 April, 2020; originally announced April 2020.

Hopfield Neural Network Flow: A Geometric Viewpoint

Authors: Abhishek Halder, Kenneth F. Caluya, Bertrand Travacca, Scott J. Moura

Abstract: We provide gradient flow interpretations for the continuous-time continuous-state Hopfield neural network (HNN). The ordinary and stochastic differential equations associated with the HNN were introduced in the literature as analog optimizers, and were reported to exhibit good performance in numerical experiments. In this work, we point out that the deterministic HNN can be transcribed into Amari'… ▽ More We provide gradient flow interpretations for the continuous-time continuous-state Hopfield neural network (HNN). The ordinary and stochastic differential equations associated with the HNN were introduced in the literature as analog optimizers, and were reported to exhibit good performance in numerical experiments. In this work, we point out that the deterministic HNN can be transcribed into Amari's natural gradient descent, and thereby uncover the explicit relation between the underlying Riemannian metric and the activation functions. By exploiting an equivalence between the natural gradient descent and the mirror descent, we show how the choice of activation function governs the geometry of the HNN dynamics. For the stochastic HNN, we show that the so-called "diffusion machine", while not a gradient flow itself, induces a gradient flow when lifted in the space of probability measures. We characterize this infinite dimensional flow as the gradient descent of certain free energy with respect to a Wasserstein metric that depends on the geodesic distance on the ground manifold. Furthermore, we demonstrate how this gradient flow interpretation can be used for fast computation via recently developed proximal algorithms. △ Less

Submitted 13 November, 2019; v1 submitted 4 August, 2019; originally announced August 2019.

Joint Fleet Sizing and Charging System Planning for Autonomous Electric Vehicles

Authors: Hongcai Zhang, Colin J. R. Sheppard, Timothy E. Lipman, Scott J. Moura

Abstract: This paper studies the joint fleet sizing and charging system planning problem for a company operating a fleet of autonomous electric vehicles (AEVs) for passenger and goods transportation. Most of the relevant published papers focus on intracity scenarios and adopt heuristic approaches, e.g., agent based simulation, which do not guarantee optimality. In contrast, we propose a mixed integer linear… ▽ More This paper studies the joint fleet sizing and charging system planning problem for a company operating a fleet of autonomous electric vehicles (AEVs) for passenger and goods transportation. Most of the relevant published papers focus on intracity scenarios and adopt heuristic approaches, e.g., agent based simulation, which do not guarantee optimality. In contrast, we propose a mixed integer linear programming model for intercity scenarios. This model incorporates comprehensive considerations of 1) limited AEV driving range; 2) optimal AEV routing and relocating operations; 3) time-varying origin-destination transport demands; and 4) differentiated operation cost structure of passenger and goods transportation. The proposed model can be computational expensive when the scale of the transportation network is large. We then exploit the structure of this program to expedite its solution. Numerical experiments are conducted to validate the proposed method. Our experimental results show that AEVs in passenger and goods transportation have remarkable planning and operation differences. We also demonstrate that intelligent routing and relocating operations, charging system and vehicle parameters, e.g., charging power, battery capacity, driving speed etc., can significantly affect the economic efficiency and the planning results of an AEV fleet. △ Less

Submitted 1 November, 2018; originally announced November 2018.

Data-driven Chance-constrained Regulation Capacity Offering for Distributed Energy Resources

Authors: Hongcai Zhang, Zechun Hu, Eric Munsing, Scott J. Moura, Yonghua Song

Abstract: This paper studies the behavior of a strategic aggregator offering regulation capacity on behalf of a group of distributed energy resources (DERs, e.g. plug-in electric vehicles) in a power market. Our objective is to maximize the aggregator's revenue while controlling the risk of penalties due to poor service delivery. To achieve this goal, we propose data-driven risk-averse strategies to effecti… ▽ More This paper studies the behavior of a strategic aggregator offering regulation capacity on behalf of a group of distributed energy resources (DERs, e.g. plug-in electric vehicles) in a power market. Our objective is to maximize the aggregator's revenue while controlling the risk of penalties due to poor service delivery. To achieve this goal, we propose data-driven risk-averse strategies to effectively handle uncertainties in: 1) The DER parameters (e.g., load demands and flexibilities) and 2) sub-hourly regulation signals (to the accuracy of every few seconds). We design both the day-ahead and the hour-ahead strategies. In the day-ahead model, we develop a two-stage stochastic program to roughly model the above uncertainties, which achieves computational efficiency by leveraging novel aggregate models of both DER parameters and sub-hourly regulation signals. In the hour-ahead model, we formulate a data-driven distributionally robust chance-constrained program to explicitly model the aforementioned uncertainties. This program can effectively control the quality of regulation service based on the aggregator's risk aversion. Furthermore, it learns the distributions of the uncertain parameters from empirical data so that it outperforms existing techniques, (e.g. robust optimization or traditional chance-constrained programming) in both modelling accuracy and cost of robustness. Finally, we derive a conic safe approximation for it which can be efficiently solved by commercial solvers. Numerical experiments are conducted to validate the proposed method. △ Less

Submitted 16 August, 2017; originally announced August 2017.

Joint Planning of PEV Fast-Charging Network and Distributed PV Generation Using the Accelerated Generalized Benders Decomposition

Authors: Hongcai Zhang, Scott J. Moura, Zechun Hu, Wei Qi, Yonghua Song

Abstract: Integration of plug-in electric vehicles (PEVs) with distributed renewable resources will decrease PEVs' well-to-wheels greenhouse gas emissions, promote renewable power adoption and defer power system investments. This paper proposes a multidisciplinary approach to jointly planning PEV fast-charging stations and distributed photovoltaic (PV) power plants on coupled transportation and power networ… ▽ More Integration of plug-in electric vehicles (PEVs) with distributed renewable resources will decrease PEVs' well-to-wheels greenhouse gas emissions, promote renewable power adoption and defer power system investments. This paper proposes a multidisciplinary approach to jointly planning PEV fast-charging stations and distributed photovoltaic (PV) power plants on coupled transportation and power networks. First, we develop models of 1) PEV fast-charging stations; 2) highway transportation networks under PEV driving range constraints; 3) PV power plants with reactive power control. Then, we formulate a two-stage stochastic mixed integer second order cone program (MISOCP) to determine the sites and sizes of 1) PEV fast-charging stations; 2) PV power plants. To address the uncertainty of future scenarios, a significant number of future typical load, traffic flow and PV generation curves are adopted. This makes the problem large scale. We design a Generalized Benders Decomposition Algorithm to efficiently solve it. To the authors' knowledge, this work is the first that jointly plans both PEV fast-charging stations and PV plants with consideration for PEV driving range limits and reactive PV power control. We conduct numerical experiments to illustrate the effectiveness of the proposed method, and validate the benefits of the joint planning and adopting advanced PV reactive power control. △ Less

Submitted 14 March, 2017; v1 submitted 23 February, 2017; originally announced February 2017.