arXiv subjects

François Pacaud

Publications and source records attributed to François Pacaud.

29 records · Page 2Linked to original sources

arXiv2022

Optimization of a domestic microgrid equipped with solar panel and battery: Model Predictive Control and Stochastic Dual Dynamic Programming approaches

In this study, a microgrid with storage (battery, hot water tank) and solar panel is considered. We benchmark two algorithms, MPC and SDDP, that yield online policies to manage the microgrid, and compare them with a rule based policy. Model Predictive Control (MPC) is a well-known algorithm which models the future uncertainties with a deterministic forecast. By contrast, Stochastic Dual Dynamic Programming (SDDP) models the future uncertainties as stagewise independent random variables with known probability distributions. We present a scheme, based on out-of-sample validation, to fairly compare the two online policies yielded by MPC and SDDP. Our numerical studies put to light that MPC and SDDP achieve significant gains compared to the rule based policy, and that SDDP overperforms MPC not only on average but on most of the out-of-sample assessment scenarios.

math.OC↗

arXiv2022

Condensed interior-point methods: porting reduced-space approaches on GPU hardware

The interior-point method (IPM) has become the workhorse method for nonlinear programming. The performance of IPM is directly related to the linear solver employed to factorize the Karush--Kuhn--Tucker (KKT) system at each iteration of the algorithm. When solving large-scale nonlinear problems, state-of-the art IPM solvers rely on efficient sparse linear solvers to solve the KKT system. Instead, we propose a novel reduced-space IPM algorithm that condenses the KKT system into a dense matrix whose size is proportional to the number of degrees of freedom in the problem. Depending on where the reduction occurs we derive two variants of the reduced-space method: linearize-then-reduce and reduce-then-linearize. We adapt their workflow so that the vast majority of computations are accelerated on GPUs. We provide extensive numerical results on the optimal power flow problem, comparing our GPU-accelerated reduced space IPM with Knitro and a hybrid full space IPM algorithm. By evaluating the derivatives on the GPU and solving the KKT system on the CPU, the hybrid solution is already significantly faster than the CPU-only solutions. The two reduced-space algorithms go one step further by solving the KKT system entirely on the GPU. As expected, the performance of the two reduction algorithms depends intrinsically on the number of available degrees of freedom: their performance is poor when the problem has many degrees of freedom, but the two algorithms are up to 3 times faster than Knitro as soon as the relative number of degrees of freedom becomes smaller.

math.OC↗

arXiv2022

Batched Second-Order Adjoint Sensitivity for Reduced Space Methods

This paper presents an efficient method for extracting the second-order sensitivities from a system of implicit nonlinear equations on upcoming graphical processing units (GPU) dominated computer systems. We design a custom automatic differentiation (AutoDiff) backend that targets highly parallel architectures by extracting the second-order information in batch. When the nonlinear equations are associated to a reduced space optimization problem, we leverage the parallel reverse-mode accumulation in a batched adjoint-adjoint algorithm to compute efficiently the reduced Hessian of the problem. We apply the method to extract the reduced Hessian associated to the balance equations of a power network, and show on the largest instances that a parallel GPU implementation is 30 times faster than a sequential CPU reference based on UMFPACK.

cs.MS↗

arXiv2021

A Globally Convergent Distributed Jacobi Scheme for Block-Structured Nonconvex Constrained Optimization Problems

Motivated by the increasing availability of high-performance parallel computing, we design a distributed parallel algorithm for linearly-coupled block-structured nonconvex constrained optimization problems. Our algorithm performs Jacobi-type proximal updates of the augmented Lagrangian function, requiring only local solutions of separable block nonlinear programming (NLP) problems. We provide a cheap and explicitly computable Lyapunov function that allows us to establish global and local sublinear convergence of our algorithm, its iteration complexity, as well as simple, practical and theoretically convergent rules for automatically tuning its parameters. This in contrast to existing algorithms for nonconvex constrained optimization based on the alternating direction method of multipliers that rely on at least one of the following: Gauss-Seidel or sequential updates, global solutions of NLP problems, non-computable Lyapunov functions, and hand-tuning of parameters. Numerical experiments showcase its advantages for large-scale problems, including the multi-period optimization of a 9000-bus AC optimal power flow test case over 168 time periods, solved on the Summit supercomputer using an open-source Julia code.

math.OC↗

arXiv2021

A Feasible Reduced Space Method for Real-Time Optimal Power Flow

We propose a novel feasible-path algorithm to solve the optimal power flow (OPF) problem for real-time use cases. The method augments the seminal work of Dommel and Tinney with second-order derivatives to work directly in the reduced space induced by the power flow equations. In the reduced space, the optimization problem includes only inequality constraints corresponding to the operational constraints. While the reduced formulation directly enforces the physical constraints, the operational constraints are softly enforced through Augmented Lagrangian penalty terms. In contrast to interior-point algorithms (state-of-the art for solving OPF), our algorithm maintains feasibility at each iteration, which makes it suitable for real-time application. By exploiting accelerator hardware (Graphic Processing Units) to compute the reduced Hessian, we show that the second-order method is numerically tractable and is effective to solve both static and real-time OPF problems.

math.OC↗

arXiv2021

Leveraging GPU batching for scalable nonlinear programming through massive Lagrangian decomposition

We present the implementation of a trust-region Newton algorithm ExaTron for bound-constrained nonlinear programming problems, fully running on multiple GPUs. Without data transfers between CPU and GPU, our implementation has achieved the elimination of a major performance bottleneck under a memory-bound situation, particularly when solving many small problems in batch. We discuss the design principles and implementation details for our kernel function and core operations. Different design choices are justified by numerical experiments. By using the application of distributed control of alternating current optimal power flow, where a large problem is decomposed into many smaller nonlinear programs using a Lagrangian approach, we demonstrate computational performance of ExaTron on the Summit supercomputer at Oak RidgeNational Laboratory. Our numerical results show the linear scaling with respect to the batch size and the number of GPUs and more than 35 times speedup on 6 GPUs than on 40 CPUs available on a single node.

math.OC↗

arXiv2021

Mixed Spatial and Temporal Decompositions for Large Scale Multistage Stochastic Optimization Problems

We consider multistage stochastic optimization problems involving multiple units. Each unit is a (small) control system. Static constraints couple units at each stage. We present a mix of spatial and temporal decompositions to tackle such large scale problems. More precisely, we obtain theoretical bounds and policies by means of two methods, depending whether the coupling constraints are handled by prices or by resources. We study both centralized and decentralized information structures. We report the results of numerical experiments on the management of urban microgrids. It appears that decomposition methods are much faster and give better results than the standard Stochastic Dual Dynamic Programming method, both in terms of bounds and of policy performance.

math.OC↗

arXiv2021

Decentralized Multistage Optimization of Large-Scale Microgrids under Stochasticity

Microgrids are recognized as a relevant tool to absorb decentralized renewable energies in the energy mix. However, the sequential handling of multiple stochastic productions and demands, and of storage, make their management a delicate issue. We add another layer of complexity by considering microgrids where different buildings stand at the nodes of a network and are connected by the arcs; some buildings host local production and storage capabilities, and can exchange with others their energy surplus. We formulate the problem as a multistage stochastic optimization problem, corresponding to the minimization of the expected temporal sum of operational costs, while satisfying the energy demand at each node, for all time. The resulting mathematical problem has a large-scale nature, exhibiting both spatial and temporal couplings. However, the problem displays a network structure that makes it amenable to a mix of spatial decomposition-coordination with temporal decomposition methods. We conduct numerical simulations on microgrids of different sizes and topologies, with up to 48 nodes and 64 state variables. Decomposition methods are faster and provide more efficient policies than a state-of-the-art Stochastic Dual Dynamic Programming algorithm. Moreover, they scale almost linearly with the state dimension, making them a promising tool to address more complex microgrid optimal management problems.

math.OC↗

arXiv2019

Upper and Lower Bounds for Large Scale Multistage Stochastic Optimization Problems: Application to Microgrid Management

We consider a microgrid where different prosumers exchange energy altogether by the edges of a given network. Each prosumer is located to a node of the network and encompasses energy consumption, energy production and storage capacities (battery, electrical hot water tank). The problem is coupled both in time and in space, so that a direct resolution of the problem for large microgrids is out of reach (curse of dimensionality). By affecting price or resources to each node in the network and resolving each nodal sub-problem independently by Dynamic Programming, we provide decomposition algorithms that allow to compute a set of decomposed local value functions in a parallel manner. By summing the local value functions together, we are able, on the one hand, to obtain upper and lower bounds for the optimal value of the problem, and, on the other hand, to design global admissible policies for the original system. Numerical experiments are conducted on microgrids of different size, derived from data given by the research and development centre Efficacity, dedicated to urban energy transition. These experiments show that the decomposition algorithms give better results than the standard SDDP method, both in terms of bounds and policy values. Moreover, the decomposition methods are much faster than the SDDP method in terms of computation time, thus allowing to tackle problem instances incorporating more than 60 state variables in a Dynamic Programming framework.

math.OC↗

arXiv2018

Stochastic optimal control of a domestic microgrid equipped with solar panel and battery

Microgrids are integrated systems that gather and operate energy production units to satisfy consumers demands. This paper details different mathematical methods to design the Energy Management System (EMS) of domestic microgrids. We consider different stocks coupled together - a battery, a domestic hot water tank - and decentralized energy production with solar panel. The main challenge of the EMS is to ensure, at least cost, that supply matches demand for all time, while considering the inherent uncertainties of such systems. We benchmark two optimization algorithms to manage the EMS, and compare them with a heuristic. The Model Predictive Control (MPC) is a well known algorithm which models the future uncertainties with a deterministic forecast. By contrast, Stochastic Dual Dynamic Programming (SDDP) models the future uncertainties as probability distributions to compute optimal policies. We present a fair comparison of these two algorithms to control microgrid. A comprehensive numerical study shows that i) optimization algorithms achieve significant gains compared to the heuristic, ii) SDDP outperforms MPC by a few percents, with a reasonable computational overhead.

math.OC↗

arXiv2017

Stochastic decomposition applied to large-scale hydro valleys management

We are interested in optimally controlling a discrete time dynamical system that can be influenced by exogenous uncertainties. This is generally called a Stochas-tic Optimal Control (SOC) problem and the Dynamic Programming (DP) principle is one of the standard way of solving it. Unfortunately, DP faces the so-called curse of dimensionality: the complexity of solving DP equations grows exponentially with the dimension of the variable that is sufficient to take optimal decisions (the so-called state variable). For a large class of SOC problems, which includes important practical applications in energy management, we propose an original way of obtaining near optimal controls. The algorithm we introduce is based on Lagrangian relaxation, of which the application to decomposition is well-known in the deterministic framework. However, its application to such closed-loop problems is not straightforward and an additional statistical approximation concerning the dual process is needed. The resulting methodology is called Dual Approximate Dynamic Programming (DADP). We briefly present DADP, give interpretations and enlighten the error induced by the approximation. The paper is mainly devoted to applying DADP to the management of large hydro valleys. The modeling of such systems is presented, as well as the practical implementation of the methodology. Numerical results are provided on several valleys, and we compare our approach with the state of the art SDDP method.

math.OC↗