Entretiens du Centre Jacques Cartier
Lausanne,
15-16 novembre 2012
L’électricité intelligente: vers des
systèmes à valeur ajoutée
Session 3: Gestion de la demande
«Stockage par Supercondensateurs
et Programmation Dynamique»
3
a)
b)
c)
Une charge très typique avec de fortes variations
de puissance (bidirectionnelles) : L’ascenseur
Optimal Energy Management of an Improved
Elevator with Energy Storage Capacity based on
Dynamic Programming
E. Bilbao1 P. Barrade1 I. Etxeberria-Otadui2 A. Rufer1
S. Luri2 I. Gil3
1Industrial Electronics Laboratory 2IK4-IKERLAN 3Elevator Innovation Centre
STI-IEL-LEI, EPFL Technology Research Centre Orona EIC 1015 Lausanne, Switzerland 20500 Arrasate-Mondrag´on, Spain 20120 Hernani, Spain
Innovación Tecnológica
Outline
1 Introduction
2 Objectives and Problem Statement
3 Dynamic Programming Based Energy Manager 4 Implementation and Validation
Outline
1 Introduction
2 Objectives and Problem Statement
3 Dynamic Programming Based Energy Manager
4 Implementation and Validation
Introduction
Energy EfficiencyElevator sector (VDI 4707 Part 1). Possible solution: To add an ESS.
Regenerative energy recovering. Additional functionalities.
Degrees of freedom increasing: Energy Management Strategy requirement.
Rules-based, cutoff frequencies, fuzzy logic, ANN, DP.
Grid Scaps Braking Resistor Elevator Objective
The proposal, development, implementation and validation in a real test tower with energy storing capacity of an optimal Dynamic Programming based Energy Management Strategy.
Outline
1 Introduction
2 Objectives and Problem Statement
3 Dynamic Programming Based Energy Manager
4 Implementation and Validation
Objectives and Problem Statement
Objectives for the Energy Storage System Operation
To reduce energy losses in the braking resistor.
To reduce short-term power peaks absorbed from the grid.
Problem Statement
SOC of the ESS is charged/discharged by: Grid energy: unidirectional and controllable by the EMS.
Elevator energy: bidirectional and non-controllable (stochastic). Braking resistor: unidirectional and
controllable by the EMS (security element).
Operating Range SOCmax SOCmin ESS State of Charge
Statistical Energy Modeling
Proposal of a
General Energy and Statistical Description (GESD)
GESD Representation of an ElevatorThe number of passengers and distance
define the energy requirement (wk):
Traction mode: wk >0
Regenerative mode: wk<0
Probability of occurrence (Pwk):
Several missions are repeated in a period of time, increasing the probability. Different missions have a similar energy requirement, increasing the probability. GESD can be updated on-line monitoring solely the elevator power profile.
-300 0 30 60 90 0.012 0.024 0.036 GESD Representation
Elevator Energy Requirement (w k) [kJ] El e va tor P rob ab ili ty ( Pwk )
Outline
1 Introduction
2 Objectives and Problem Statement
3 Dynamic Programming Based Energy Manager
4 Implementation and Validation
Dynamic Programming Based Energy Manager
Dynamic Programming Principle
Solving sequential decision problems.
Economics and computer science applications. Breaking the sequence of problems in smaller ones. Subproblems are solved evaluating a cost function:
Maximizing the benefits. Minimizing the costs. The solution is:
Policy of decisions for Deterministic systems (DDP). Table of decisions for Stochastic systems (SDP).
Dynamic Programming Based Energy Manager
Elevator Application with Energy Storing CapacitySequential decision problem: energy absorbed from the grid. Stochastic Dynamic Programming (SDP): elevator energy requirements are unknown.
The cost function based on the stock management theory:
T otal Cost = V ariableCost + StorageCost + ShortageCost
Stochastic Cost Function Maps
Circles are the states of the system: ESS state of charge.
Transitions: elevator mission. Planes are the decisions: energy from the grid.
Outline
1 Introduction
2 Objectives and Problem Statement
3 Dynamic Programming Based Energy Manager
4 Implementation and Validation
Implementation
DP based EMSOff-line implementation. Sliding window of 7 missions. EMS Cost Function Parameters:
Variable cost c 1 Storage cost h 55 Shortage cost p 5 Table of Decisions Scaps Energy (x k ) [kJ] Mission (k) DP Control Strategy 0 20 40 60 1 2 3 4 5 6 7
Grid Energy Reference (u
k ) [kJ] 0 10 20 30 40 Rules-based EMS
Traction mode: the maximum grid power level is limited. PESS =−(PElevator−PGrid Limit) ∀PESS <0
Regeneration mode: the energy is stored in the ESS. PESS =−PElevator
Simulation Tests
Grid Power Profile
-4 -2 0 2 4 6 Po w e r [ k W ] -4 -2 0 2 4 6 Po w e r [ k W ] 0 10 4 2 0 2 4 6 DP 0 10 4 2 0 2 4 6 Rul Peak Peak 20 Time [s] P Power Profi 20 Time [s] es Power Pro k reduction ESS ch k reduction 30 35 40 ile Grid Elevator ESS 30 40 ofile Grid Elevator ESS harging -4 -2 0 2 4 6 Po w e r [ k W ] -4 -2 0 2 4 6 Po w e r [ k W ] 0 10 4 2 0 2 4 6 DP 0 10 4 2 0 2 4 6 Rul Peak Peak 20 Time [s] P Power Profi 20 Time [s] es Power Pro k reduction ESS ch k reduction 30 35 40 ile Grid Elevator ESS 30 40 ofile Grid Elevator ESS harging
Braking Resistor Energy Profile
Simulation of a Mission SequenceRandom sequence of 80 missions. Elevator (without ESS): 30kJ. DP based EMS: 1.5kJ (-95%). Rules-based EMS: 6.4kJ (-79%).0 10 20 30 40 En e rg y [ k J ] 0 10 20 30 40 En e rg y [ kJ ] 0 20 0 0 0 0 0 Br Elevato Rules DP 0 20 0 0 0 0 0 Braking R Elevator Rules DP 40 Mission raking Resisto or 40 Mission (k) Resistor Energ r 60 80 or 60 80 gy Profile
Introduction Statement Dynamic Programming Implementation and Validation Conclusions
Testbench and Experimental Tests Description
Test Tower Main Characteristics Elevator model Orona M34 Elevator shaft length 18 [m] Number of floors 5
Cabin mass 800 [kg]
Load mass (emulated
[0:78.75:630] [kg] by different loads)
Number of passengers 8 Counterweight Medium
ESS energy capacity 20 [Wh]
5. IMPLEMENTATION AND EXPERIMENTAL VALIDATION
Initially the proposed DP based EMS has been implemented and tested in simulation, using a basic elevator model and the GESD description presented in Figure 2. The cost function parameters implemented on the EMS are c = 1, h = 55 and p =5. Simulations show that it is possible to design the DP algorithm for a limited number of mission sequences instead of the whole set of missions, in order to improve computational cost without degrading the quality of the results. Therefore a seven mission sliding window has been implemented. Figure 4 shows the power peak reduction from the grid for a particular mission. Note that even if the elevator power is zero after the mission (between t =20s and t =35s), some power is being absorbed from the grid in order to charge the ESS. Figure 5 shows that braking losses can be reduced by 95% (from 30kJ to 1.5kJ); comparing an elevator without ESS and the improved elevator with a DP based EMS during the same random sequence of 80 missions.
Figure 4: Power profiles in a mission (simulation).
Figure 5: Braking resistor losses (simulation).
The energy management algorithm has been also implemented and validated experimentally in a real test tower: 18m, 8 passengers = 630kg (emulated by different loads) and 20Wh ESS [3], (Figure 6). Three kinds of tests have been carried out with the same sequence of 80 missions in order to evaluate and quantify the real contribution of the proposed algorithm: (a) an elevator without ESS (Elevator); (b) an elevator with ESS and a simple EMS based on rules (Rules) [3] and (c) an elevator with ESS and an the proposed EMS based on DP (DP).
Figure 6: Test tower with ESS.
0 10 20 30 35 40 -2 0 2 4 6 Time [s] P o w e r [k W ] DP Power Profile Grid Elevator ESS 0 20 40 60 80 0 10 20 30 40 Mission E n e rg y [ k J ] Braking Resistor Elevator DP Peak reduction ESS charging ESS Loads Elevator Controller Cabin
Experimental Tests of a Random Sequence of 80 Missions
(a) An elevator without ESS (Elevator).
(b) An elevator with ESS and the EMS based on rules (Rules).
Introduction Statement Dynamic Programming Implementation and Validation Conclusions
Experimental Validation: Grid Power Smoothing
Single Mission
0 10 20 30 40 -4 -2 0 2 4 6 Time [s] Po w e r [ k W ] DP Power Profile Grid Elevator ESS 0 10 20 30 40 -4 -2 0 2 4 6 Po we r [ k W ]Rules Power Profile Grid Elevator ESS ESS charging Peak reduction Peak reduction 0 10 20 30 40 -4 -2 0 2 4 6 Time [s] Po w e r [ k W ] DP Power Profile Grid Elevator ESS 0 10 20 30 40 -4 -2 0 2 4 6 Time [s] Po we r [ k W ]
Rules Power Profile Grid Elevator ESS ESS charging Peak reduction Peak reduction
Random Sequence of Missions
Summary of Experimental ResultsParameter Elevator Rules DP
Max. power 5.5 [kW] 3.7 [kW] 1.9 [kW] Smoothing level 0 [kW] 1.8 [kW] 3.6 [kW] 0 % 33 % 65 % 0 10 20 30 40 -4 -2 0 2 4 6 Time [s] Po w e r [ k W ] DP Power Profile Grid Elevator ESS 0 10 20 30 40 -4 -2 0 2 4 6 Time [s] Po we r [ k W ]
Rules Power Profile
Grid Elevator ESS 0 20 40 60 80 0 1 2 3
Braking Resistor Power Profile
Po w e r [k W ] Elevator 0 20 40 60 80 0 1 2 3 Po we r [ k W ] Rules 0 20 40 60 80 0 1 2 3 Po w e r [k W ] Mission DP 0 20 40 60 80 0 3 6
Grid Power Profile
Po w e r [ k W ] Elevator 0 20 40 60 80 0 3 6 Po w e r [ k W ] Rules 0 20 40 60 80 0 3 6 Po w e r [ k W ] Mission (k) DP ESS charging Peak reduction Peak reduction
Optimal Energy Management of an Improved Elevator with Energy Storage Capacity based on Dynamic Programming
Experimental Validation: Braking Resistor Energy Losses
Reduction
Random Sequence of Missions
Summary of Experimental ResultsParameter Elevator Rules DP
Energy losses 228 [kJ] 43 [kJ] 37 [kJ] Reduction level 0 [kJ] 185 [kJ] 191 [kJ] 0 % 81 % 84 % 0 20 40 60 80 0 1.5 3
Braking Resistor Power Profile
Po w e r [ k W ] Elevator 0 20 40 60 80 0 1.5 3 Po w e r [ k W ] Rules 0 20 40 60 80 0 1.5 3 Po w e r [ k W ] Mission (k) DP 100 200 300
Braking Resistor Energy Profile
En e rg y [ kJ ] Elevator Rules DP 0 20 40 60 80 0 1.5 3
Braking Resistor Power Profile
Po w e r [ k W ] Elevator 0 20 40 60 80 0 1.5 3 Po w e r [ k W ] Rules 0 20 40 60 80 0 1.5 3 Po w e r [ k W ] Mission (k) DP 0 20 40 60 80 0 100 200 300
Braking Resistor Energy Profile
En e rg y [ kJ ] Mission (k) Elevator Rules DP 16 / 19
Outline
1 Introduction
2 Objectives and Problem Statement
3 Dynamic Programming Based Energy Manager
4 Implementation and Validation
Conclusions
An optimal Dynamic Programming based Energy Management Strategy has been proposed, developed, implemented and validated in real test tower with energy storing capacity. The optimized control strategy for stochastic applications:
Proposal of a General Energy and Statistical Description. The cost function is based on the stock management theory. A rules-based and Dynamic Programming based EMS have been simulated and validated experimentally in a sequence of 80 missions.
The maximum grid power peak has been reduced by 33% (Rules) and by 65% (DP).
The braking resistor energy losses have been reduced by 81% (Rules) and by 84% (DP).