Delft University of Technology ENergy-Based analysis and control of the grid: dealing with uncertainty and mARKets Tjerk Stegink Mahya Adibi Claudio De Persis Dimitri Jeltsema Nima Monshizadeh Arjan van der Schaft Jacquelien Scherpen Real-time dynamic pricing Introduction I nterfacing the growing supply of renewable energy sources to the power grid presents several challenges demanding for appropriate control strategies that guarantee stable operation. Additionally, any control strategy for future smart grids cannot disregard economical considerations that allow producers and consumers to fairly share utilities and costs associated with the generation and consumption of energy. In this research we develop a novel approach for the modeling, analysis and control of smart grids based on energy functions which also lend themselves to the integration of dynamic pricing algorithms that allow to consider economical factors in the control of smart grids. Model of the power network Based on the gradient method applied to (4), we obtain the distributed market dynamics power generation: power demand: auxiliary variable: electricity price: generators: inverters: loads: T MG θ̈G + AG θ̇G = −DG Γsin(D θ) + uG AI θ̇I = −DI Γsin(D T θ) + uI ALθ̇L = −DLΓsin(D T θ) + PL (1) where θ = col(θG , θI , θL), D = blockcol(DG , DI , DL). G I L L L Figure 1: A power network consisting of generators, inverters and loads. Control areas Another model of the power system is obtained when dividing the grid into large control areas, each having a power supply and demand: (2) M θ̈ + Aθ̇ = −DΓsin(D T θ) + u − PL Optimal power dispatch W e consider two optimal power dispatch problems. The first one focuses only on the power producers and the second one on both producers as well as consumers. Minimizing generation cost Each generator and inverter is associated to a quadratic cost function. X X Ci (ui ) = qi ui2 min. generation cost: min supply-demand matching: s.t. i X (3) i ui = i X PL,j Extensions: • Congestion (of e.g. the transmission lines or nodal power production) • Power transmission costs T he classical complex phasor representation of sinusoidal voltages and currents generalizes to arbitrary waveforms leading to the notion of the time-varying power triangle. Using the Hilbert Transform H{·}, the analytical signal representation of the voltage u(t) and current i(t) is formulated as √ jα(t) √ jβ(t) u(t) = u(t) + jH{u(t)} = U(t) 2e , i(t) = i(t) + jH{i(t)} = I (t) 2e , √ with j := −1, U(t) := |u(t)|, α(t) := arg{u(t)}, etc.. 1 S(t) := u(t)i ∗(t) = P(t) + jQ(t), 2 1 1 with P(t) := 2 u(t)i(t) + H{u(t)}H{i(t)} and Q(t) := 2 H{u(t)}i(t) − u(t)H{i(t)} . The instantaneous power then takes the form: p(t) = u(t)i(t) = P(t) 1 + cos 2α(t) + Q(t)sin 2α(t) , where P(t) = U(t)I (t)cos ϕ(t) , Q(t) = U(t)I (t)sin ϕ(t) , and ϕ(t) := α(t) − β(t). Budeanu's reactive and distortion power revised Budeanu's reactive power can be related to energy oscillations, but only in an average sense. Indeed, consider distorted waveforms: X √ X √ u(t) = Uk 2cos(kωt + αk ) , i(t) = Ik 2cos(kωt + βk ). k k Then, the active power PA and Budeanu's reactive power QB are obtained from averaging P(t) and Q(t) over a period: Z T X 1 PA := P(t)dt = Uk Ik cos(ϕk ), T 0 k Z T X 1 QB := Q(t)dt = Uk Ik sin(ϕk ), T 0 k with ϕk := αk −βk . Let DP (t) := P(t)−PA, DQ (t) := Q(t)−QB , IP (t) := I (t)cos ϕ(t) , and IQ (t) := I (t)sin ϕ(t) , then Budeanu's distortion power DB is decomposed into two parts: ( 2 2 2 2 D := kUk kI k − P P PU A, 2 2 2 2 2 2 DB := S − PA − QB = DPU + DQU , with DQ2 U := kUk2kIQ k2 − QB2 , We associate U(PL) to the total utility of the power consumers. X X max. social welfare: max U(PL) − C (u) = Ui (PL,i ) − Ci (ui ) u s.t. i X ui = X i PL,i i (4) i Distributed controllers T Definition of power under nonsinusoidal conditions j Social welfare problem supply-demand matching: (6) The nonsinusoidal equivalent of the classical complex power is defined by the time-varying complex power I u −∇C (u) + λ − ω ∇U(PL) − λ + ω −D T λ Dv − u + PL Time-varying complex power G control areas: = = = = where τu , τL, τv , τλ > 0 are time-scales. Interconnecting (2) with (6) gives an asymptotically stable closed-system w.r.t. the optimal points of (4) in addition to zero frequency deviation. C onsider a simplified model of a power network containing three types of elements, which are generators, inverters and frequency dependent loads: τu u̇ τLṖL τv v̇ τλλ̇ wo distributed controller designs are proposed to tackle each of the optimal power dispatch problems. where DPU and DQU are measures of fluctuation around the active power and Budeanu's reactive power, respectively, relative to the voltage. i(t) P(t) 1Ω G Internal-model-based controller Based on model (1) and optimization problem (3), consider the controller X δ̇i = − aij (δi − δj ) − qi−1ωi (5) j ui = qi−1δi . Zero frequency deviation is achieved with an optimal power generation identified by (3), despite the uncertainty of a constant unknown PL. Extensions: A variant of (5) can also deal with a time-varying sinusoidal power demand which have known amplitude and frequency but unknown offset and phase. u(t) 2H ωφ(t) u(t)=10√2cos(t)+5√2cos(5t) PA D P (t) time Q(t) S(t) QB DQ (t) Q(t) φ(t) time P(t) Figure 2: A series RL circuit supplied by a nonsinusoidal voltage source. Here ωϕ(t) = ϕ̇(t) denotes the instantaneous frequency.