Mean field approximation of an optimal control problem for the continuity equation arising in smart charging (Q6058509)
From MaRDI portal
scientific article; zbMATH DE number 7758726
Language | Label | Description | Also known as |
---|---|---|---|
English | Mean field approximation of an optimal control problem for the continuity equation arising in smart charging |
scientific article; zbMATH DE number 7758726 |
Statements
Mean field approximation of an optimal control problem for the continuity equation arising in smart charging (English)
0 references
1 November 2023
0 references
The author considers an optimal control problem for a finite population of hybrid processes, which intends to model the optimal charging of a large population of identical plug-in electric vehicles. He considers a population of \(n\) vehicles (\(n\in \mathbb{N}^{\ast }\)), a state space \([0,1]\times I\), where \(I\) is a finite set, a time interval \([0,T]\), time and space discretizations depending on \(n\), with time step \(\Delta t^{n}\) and space step \(\Delta s^{n}\), such that \(N_{T}^{n}=T/\Delta t^{n}\) and \( N_{s}^{n}=1/\Delta s^{n}\) are integers, and the set \(\mathcal{P}([0,1]\times I)\) of empirical probability measures on the space \([0,1]\times I\). The time mesh is \(\{0,t_{1}^{n},\ldots ,t_{k}^{n},\ldots ,T\}\), with \( t_{k}^{n}=k\Delta t^{n}\), for any \(k\in \{0,\ldots ,N_{T}^{n}\}\). The space mesh is \(\{0,y_{1}^{n},\ldots ,y_{p}^{n},\ldots ,1\}\), with \( y_{p}^{n}=p\Delta s^{n}\), for any \(p\in \{0,\ldots ,N_{s}^{n}\}\). For any \( \ell \in \{1,\ldots ,n\}\) and \(t\in \lbrack 0,T]\), the vehicle \(\ell \) is described by its state variable \(x_{t}^{\ell }=(i_{t}^{\ell },s_{t}^{\ell })\in I\times \lbrack 0,1]\), where the discrete variable \(i^{\ell }\) denotes the charging mode, which can switch deterministically and only at fixed times in \(\{t_{1},\ldots ,t_{N_{T-1}^{n}}\}\), and the continuous variable \( s_{t}^{\ell }\) represents the charge state, which is governed by an ODE depending on the charging mode. Between two jumps of the variable \(i^{\ell }\), the dynamics of \(x\) is deterministic and is given by: \(i_{t}^{\ell }=i_{t_{k}}^{\ell }\), \(\frac{ds_{t}^{\ell }}{dt}=b(i_{t}^{\ell },s_{t}^{\ell })\), for any \(t\in \lbrack t_{k},t_{k+1})\), with \(s_{t_{k}}^{\ell }=lim_{\tau \uparrow t_{k}}s_{\tau }^{\ell }\), the initial state \( (i_{0}^{\ell },s_{0}^{\ell })\) being given. The function \(b\) belongs to \( C^{1}(I\times \mathbb{R})\) and vanishes at the boundary of \([0,1]\). For the initial distribution \(\overline{m}\in \mathcal{P}^{n}([0,1]\times I)\), the author defines the admissible path space as: \(\mathcal{X}^{n}(\overline{m} )=\{\{(i^{\ell },s^{\ell })\}_{\ell \in \{1,\ldots ,n\}}\in D([0,T],I)\times C^{0}([0,T],[0,1])^{n}\mid \) the empirical distribution of \(\{(i_{0}^{\ell },s_{0}^{\ell })\}_{\ell }\) is equal to \(\overline{m}\) and \((i^{\ell },s^{\ell })\) satisfies the preceding equation\(\}\). The population of \(n\) processes is subject to the congestion constraints: \(\frac{1}{n}\sum_{\ell =1}^{n}1_{i}(i_{t}^{\ell })\leq D_{i}(t)\), for every \((i,t)\in I\times \lbrack 0,T]\), where \(D:[0,T]\rightarrow \mathbb{R}_{+}^{\ast }\) is given. The paper especially describes the mean field limit of the optimal control problem written as: \(\inf_{(i,s)\in \mathcal{T}^{n}(\overline{m},D)}\mathcal{J }^{n}(i,s)\), where \(\mathcal{T}^{n}(\overline{m},D)\) is the admissible set defined by: \(\mathcal{T}^{n}(\overline{m},D)=\{(i,s)\in \mathcal{X}^{n}( \overline{m})\mid (i,s)\) satisfies the congestion constraints\(\}\). The objective function is defined as \(\mathcal{J}^{n}(i,s)=\frac{1}{n} \sum_{l=1}^{n}\int_{0}^{T}c(t,i_{t}^{\ell },s_{t}^{\ell })dt+g(i_{T}^{\ell },s_{T}^{\ell })+\sum_{i,j\in I,j\neq i}\sum_{k=1}^{N_{T}^{n}-1}\sum_{p=0}^{N_{s}^{n}-1}L\left( \frac{ Q_{i,j}^{k,p}(i,s)}{\Delta t}\right) \Delta tQ_{i}^{k,p}(i,s)\), \( Q_{i,j}^{k,p}(i,s)\) being the proportion of vehicles among the processes with a state in \([y_{p},y_{p+1})\times \{i\}\) that switch their discrete state from \(i\) to \(j\) at time \(t_{k}\), \(Q_{i}^{k,p}(i,s)\) is the proportion of vehicles that have a state in \([y_{p},y_{p+1})\times \{i\}\) at time \( t_{k}^{-}\), \(L:\mathbb{R}\rightarrow \overline{\mathbb{R}}\) is a convex function, defined by: \(L(x)=l(x)\) if \(x>0\), \(L(x)=0\) if \(x=0\), \(L(x)=+\infty \) otherwise, where \(l\in C^{1}(\mathbb{R}_{+},\mathbb{R}_{+})\) is an increasing, strongly convex function, bounded from above by a quadratic function. For the description of the mean field limit of this finite population optimal control problem, the author introduces the continuity equation on the pair \((m,E)\in D([0,T],\mathcal{P}([0,1]\times I))\times \mathcal{M}^{+}([0,T]\times \lbrack 0,1]\times I^{2})\) written as: \(\partial _{t}m_{i}(t,s)+\partial _{s}(m_{i}(t,s)b_{i}(s))=-\sum_{i,j\in I,j\neq i}(E_{i,j}(t,s)-E_{j,i}(t,s))\), \((i,t,s)\in I\times (0,T)\times (0,1)\), where \(m\) is the distribution of the hybrid state of the infinite population of vehicles, and \(E\) is the measure of the switches of the discrete state variable among the population over the time. The initial condition \(m_{i}(0,s)=\overline{m}_{i}(s)\) is imposed for \((i,s)\in I\times (0,1)\). The density constraint \(m_{i}(t,[0,1])\leq D_{i}(t)\) is imposed for every \( (i,t)\in I\times \lbrack 0,T]\).\ The admissible set is \(\mathcal{S}( \overline{m},D)=\{(m,E)\mid m\in C^{0}([0,T],\mathcal{P}([0,1]\times I)\), \( E\in \mathcal{M}^{+}([0,T]\times \lbrack 0,1]\times I^{2})\), \(E_{i,j}\ll m_{i}\) and \(\frac{dE_{i,j}}{dm_{i}}\in L_{m_{i}}^{2}(0,T)\) \(\forall i,j\in I\) , \((m,E)\) is a weak solution to the preceding equation and satisfies the above constraint\(\}\). The objective function associated with this limit problem \(\mathcal{J}\) is defined for any \((m,E)\in \mathcal{S}(\overline{m} ,D)\) by: \(\mathcal{J}(m,E)=\sum_{i\in I}\int_{0}^{T}\int_{0}^{1}(c_{i}(t,s)+\sum_{j\in I,j\neq i}L\left( \frac{ dE_{i,j}}{dm_{i}}(t,s)\right) m_{i}(t,ds))dt+\sum_{i\in I}\int_{0}^{1}g_{i}(s)m_{i}(T,ds)\). The main result proves under hypotheses on the data and on the time and space steps that if \(\{\overline{m}^{n}\}_{n} \) weakly converges to \(\overline{m}\in \mathcal{P}([0,1]\times I)\), then \[ \inf_{(m,E)\in \mathcal{S}(\overline{m},D)} \mathcal{J}(m,E)\leq \liminf_{n\rightarrow \infty }\inf_{(i,s)\in \mathcal{T}^{n}(\overline{m}^{n},D)} \mathcal{J}^{n}(i,s). \] Further, there exists \(C>0\) which depends on the data such that for any \(n\in N\): \[ \inf_{(i,s)\in \mathcal{T}^{n}( \overline{m}^{n},D)}\mathcal{J}^{n}(i,s)-\frac{C}{n^{1/3}}\leq \inf_{(m,E)\in \mathcal{S}(\overline{m},D)} \mathcal{J}(m,E). \] For the proof, the author introduces an Eulerian formulation of the mean field limit of the following finite population optimal control problem and he proves an equivalence between the two problems, using a superposition principle. The first part of the main result is proved applying compactness arguments on sequences of solutions to the Eulerian problem. The second part is proved using regularity results of the solution to the mean field control problem and he builds an admissible control for the Eulerian formulation.
0 references
optimal control problem
0 references
hybrid state space
0 references
mean-field limit
0 references
regularity of solutions
0 references
finite population of hybrid processes
0 references
0 references
0 references
0 references
0 references
0 references