Laboratoire des systèmes électriques et industriels Institute of Energy System Technology Genetic Algorithms-based Battery Predictive Management in INES Smart Grid System Merci monsieur le président بسم الله الرحمان الرحيم Monsieur le président , madame, messieurs les membres de jury, honorable assistance, j’ai le plaisir de vous présenter mon travail dans le cadre d’une soutenance du mémoire de magistère intitulé par: Stockage et récupération d’énergie dans un système multi-sources, application au véhicule électrique Supervisors Dr. A A LADJICI Pr. E BOLLIN PhD student Mustapha HABIB

PhD main axis Forecasting tools for PV power and building power demand: ANN, ANFIS Short-term power management: frequency/voltage control Long-term power management: Daily power dispatching of grid connected hybrid system Experimental validation of battery predictive management: Developing a control algorithm for XTM 4000-48 Xtenders for batteries daily power management

Predictive controller Forecasting tools: ANN and fitting tools Control variables: grid energy and battery SOC System model: power flow equation Actuator variables: battery current and grid relay Cost function and solver: genetic algorithms

Forecasting tools

PV power forecasting with ANN Time Series tool Forecasting tools PV power forecasting with ANN Time Series tool Prediction horizon Prediction precision 6 past steps 1 step ahead

Forecasting tool PV power forecasting with ANN Time Series tool

ANN forecasting system Forecasting tools ANN-based building forecasting tool ANN forecasting system Hour of day Day of week Building power demand Temperature Holyday indicator

Forecasting tools ANN-based building forecasting tool

Forecasting tools Building power demand forecasting with fitting tool

Building power demand forecasting with fitting tool Forecasting tools Building power demand forecasting with fitting tool Polynomial Sum of sin Coefficient Value a1 1340 b1 0.1405 c1 -0.1949 a2 624.8 b2 0.2942 c2 2.055 a3 239 b3 0.5406 c3 2.274 a4 87.47 b4 0.9153 c4 -2.906 a5 55.9 b5 1.773 c5 -3.962 a6 48.87 b6 1.346 c6 0.1323 Coefficient Value a9 1.472e-06 a8 -0.0001152 a7 0.003223 a6 -0.03179 a5 -0.1718 a4 6.069 a3 -46.94 a2 163 a1 -290.3 a0 511.4 RMSE=74.96 R=0.9714 RMSE=51.91 R=0.9863

Control variables

3. Battery charging mode: the Xtender works as charger Control variables Grid energy 1. Power network energy 1. Grid feeding mode: the Xtender works as inverter SCI (current source inverter) 2. Island mode: the Xtender works as inverter VSI (voltage source inverter) 3. Battery charging mode: the Xtender works as charger

Control variables SOC (%) Battery SOC 90 2. Battery SOC 50 1. Maximum limit (90%): keep enough capacity to absorb any excess PV power in island mode 2. Minimum limit (50%): to save batteries life time by minimizing DOD level

System model

Genetic Algorithms-based Predictive Management System model Power flow equation in grid-connected mode Power flow equation in island mode Battery current calculation Battery SOC estimation

Actuator variables

Genetic Algorithms-based Predictive Management Power topology and the role of GA-PM

Cost function and solver

Genetic Algorithms-based Predictive Management Cost function Actuator variables : grid relay and battery current Battery and power converter efficiency Weighting coefficients Battery voltage Reduce the used grid energy/increase the energy injected into the grid Prediction of the evolution of battery current (SOC) in island mode Giving priority to the island mode compared to grid-connected mode

Genetic Algorithms-based Predictive Management System constraints

Global GA-PM Algorithm

Genetic Algorithms-based Predictive Management Global control algorithm MATLAB program

Experimental results

INES smart grid parameters Experimental results INES smart grid parameters Component Subcomponent Parameter Value PV system PV module Power at MPP 240 Wp Voltage at MPP 30.0 V Current at MPP 8.1 A Open circuit voltage 37.4 V Short circuit current 8.6 A Temperature coefficient -0.46 %/K Module model Bosch solar module c-Si M 60 PV power plant Number of modules 27 Inclination 9 x 35° 18 x 30° Alignment 180° south Power 6.3 KWp Batteries system Battery cell Voltage 4 V Nominal capacity 546 Ah Battery model Rolls Battery 4CS17P Batteries bank Number of cell in series 12 Number of cells in parallel 1 4.5 KW Programmable load - Nominal power 3.6 KW Load mode Constant power Control mode Remote Model Chroma 63803

Experimental results Reference method: rules-based power management

Experimental results Rules-based power management June 8th (sunny day)

June 14th (sunny day) Experimental results GA Predictive power management June 14th (sunny day)

21.36 % !! ΔE=580 Wh Economy of 6.23 % for just 7 hours Simulation results with MATLAB/Simulink Comparison Power management technique Produced PV energy (EPV) Consumed Energy (ELoad) Exchanged energy with grid (Eg) EPV/ELoad Estimated exchanged energy with grid when EPV/Eload = 1 Rules management 34.11 KWh 27.88 KWh -10.65 KWh 1.22 -8.72 Predictive management 31.9 KWh 27.88 -10.61 1.14 -9.30 ΔE=580 Wh Economy of 6.23 % for just 7 hours Estimated economy in the case of 24 hours 21.36 % !!

June 9th (cloudy day) Experimental results GA Predictive power management June 9th (cloudy day)

System model is not required Advantages and disadvantages Rules-based power management Simple algorithm System model is not required The control behavior is not on real time

On-line control behavior Advantages and disadvantages Predictive-based power management On-line control behavior Energy economy and battery SOC control performance The effectiveness depends highly on system model exactitude and on weather forecasting accuracy

Fuzzy-based power management

PLoad-PPV Grid relay SOC Battery current Electricity price Fuzzy-based power management Controller structure PLoad-PPV Grid relay SOC Battery current Electricity price MATLAB program

45 Rules ! Fuzzy-based power management Fuzzy rules PLoad-PPV Controller response dP++ dP+ dP0 dP- dP--   EP-L SOC-H P+ R1 R0 SOC-M P- SOC-L P-- EP-M P++ P0 EL-H Energy price SOC Rules Battery charging mode 18 Rules ! 45 Rules ! Grid feeding mode 18 Rules ! 09 Rules ! Island mode

Fuzzy-based power management Fuzzy rules 1. Inject more power into the grid when the energy price is high 2. Using the grid power to charge batteries when the energy price is low 3. When a low difference between building power demand and supplied PV power is detected an island mode is to be applied

Fuzzy-based power management Surfaces

Forecasted energy price in Germany for July 27th 2017 Fuzzy-based power management Energy price profile (€/MWh) Forecasted energy price in Germany for July 27th 2017 © http://www.nordpoolspot.com

Fuzzy-based power management Experimental results

Perspective works

Perspective works Repeat the same tests by introducing wind energy in the local network Improve the PV and building power forecasting tools: using on-line weather forecast with ANN Taking the batteries aging in consideration Taking the frequency/voltage control in consideration Using other solvers for the cost function: PSO Using INES offices as a load when doing new tests

Thank You for your Attention Any Question !