Battery Parameter Extraction from Data
This example shows optimization of the Battery block's parameters to fit data defined over different temperatures. It uses the MATLAB® optimization function fminsearch. Other products available for performing this type of parameter fitting with Simscape™ Electrical™ models are the Optimization Toolbox™ and Simulink® Design Optimization™. These products provide predefined functions to manipulate and analyze blocks using GUIs or a command line approach.
Strategy
Fit output voltage curves for a Battery to data using a 4 step procedure:
Optimize parameters in the Battery Main dialog tab.
Optimize parameters in the Battery Dynamics dialog tab.
Optimize nominal voltage and internal resistance in the Battery Temperature Dependence dialog tab.
Optimize temperature dependent charge dynamics parameters in the Battery Temperature Dependence dialog tab.
Data and Block Setup
The MATLAB data file, ee_battery_data.mat, stores Battery data as an array of structures. Each structure contains 5 fields: v (voltage) , i (current), t (time), SOC0 (initial state of charge) and T (temperature). Scope save the output voltage as structure data, out.Vo.signals.values.
% Load Battery data load ee_battery_data.mat assignin('base','T1',battery_data(find([battery_data(1:2).T]==25)).T); assignin('base','T2',battery_data(find([battery_data(1:2).T]~=25)).T); % Display the Battery model Model = 'ee_battery'; open_system(Model)
close_system(Model, 0);
Initial Parameter Specification
Starting values for fminsearch can be estimated using a combination of Battery block defaults and data sheet values
List of parameters and initial values prior to optimization
ParsListMain = {'Vnom', 'R1', 'AH', 'V1', 'AH1'};
InitGuessMain = [3.6, 0.045, 2.7, 3.4, 1.4];
ParsListDyn = {'Rp1', 'tau1'};
InitGuessDyn = [0.006, 200];
ParsListTemp = {'Vnom_T2', 'R1_T2', 'V1_T2','Rp1_T2','tau1_T2'};
InitGuessTemp = [3.8, 0.055, 3.6, 0.006, 200 ];
Pars0 = reshape([[ParsListMain ParsListDyn ParsListTemp]; cellstr(num2str([InitGuessMain InitGuessDyn InitGuessTemp]'))'],1,[]);
fprintf('\t%5s = %s\n', Pars0{:});
clear Pars0
Vnom = 3.6 R1 = 0.045 AH = 2.7 V1 = 3.4 AH1 = 1.4 Rp1 = 0.006 tau1 = 200 Vnom_T2 = 3.8 R1_T2 = 0.055 V1_T2 = 3.6 Rp1_T2 = 0.006 tau1_T2 = 200
Since fminsearch is an unconstrained nonlinear optimizer that locates a local minimum of a function, varying the initial estimate will result in a different solution set.
Plot Data Versus Battery Output Using Initial Parameters
Load single cell Battery model and set parameters
load_system(Model); % Enable Fast Restart to speedup the simulation set_param(Model,'FastRestart','on') Pars = reshape([ParsListMain; cellstr(num2str(InitGuessMain'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end Pars = reshape([ParsListDyn; cellstr(num2str(InitGuessDyn'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end Pars = reshape([ParsListTemp; cellstr(num2str(InitGuessTemp'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end % Generate preliminary model curves and plot against data num_lines = length(battery_data)-1; v_model = cell(1, num_lines); t_model = cell(1, num_lines); legend_info_data = cell(1, num_lines); legend_info_model = cell(1, num_lines); for idx_data = 1:num_lines assignin('base','t_data',battery_data(idx_data).t); assignin('base','i_data',battery_data(idx_data).i); assignin('base','T_data',battery_data(idx_data).T*ones(length(t_data),1)); assignin('base','T0',battery_data(idx_data).T); assignin('base','Ts',t_data(2)-t_data(1)); assignin('base','AH0',AH*battery_data(idx_data).SOC0); out = sim(Model); v_model{idx_data} = out.Vo.signals.values; t_model{idx_data} = out.Vo.time; legend_info_data{idx_data} = [ 'Temp = ' ... num2str(battery_data(idx_data).T) '\circC, Data']; legend_info_model{idx_data} = [ 'Temp = ' ... num2str(battery_data(idx_data).T) '\circC, Model']; end plot([battery_data(1:num_lines).t]/3600, [battery_data(1:num_lines).v], 'o', [t_model{:}]/3600, [v_model{:}]) xlabel('Time (hours)'); ylabel('Battery voltage (V)'); legend([legend_info_data legend_info_model], 'Location', 'Best'); title('Model with Initial Parameter Values');
Sum of Squares of Error Calculation
ee_battery_lse is the function to be minimized by fminsearch. This function returns a sum of squares of error for the difference between the Battery output voltage and the data. If an invalid parameter value is supplied by fminsearch, the catch statement returns a large value for the error.
Optimize Main Tab Dialog Parameters Without Charge Dynamics (Step 1)
% Find ambient temperature data index idx_data = find([battery_data(1:num_lines).T]==25); assignin('base','t_data',battery_data(idx_data).t); assignin('base','i_data',battery_data(idx_data).i); assignin('base','T_data',battery_data(idx_data).T*ones(length(t_data),1)); assignin('base','T0',battery_data(idx_data).T); assignin('base','Ts',t_data(2)-t_data(1)); assignin('base','v_data',battery_data(idx_data).v); % Optimize parameters in main dialog tab of Battery assignin('base','ParsList',ParsListMain(1:4)); InitGuess = InitGuessMain(1:4); OptPars = fminsearch(@ee_battery_lse, InitGuess, ... optimset('TolX', 1e-3)); OptParsMain = [OptPars(1:4) InitGuessMain(5)]; % Update Battery block with optimized parameters Pars = reshape([ParsListMain; cellstr(num2str(OptParsMain'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end % Display optimized parameters fprintf(['Optimized parameters for the battery main ' ... 'dialog tab are:\n']); fprintf('\t%5s = %s\n', Pars{:}); clear i_data v_data t_data T_data Ts clear k InitGuess
Optimized parameters for the battery main dialog tab are: Vnom = 3.6999 R1 = 0.050299 AH = 2.6033 V1 = 3.5265 AH1 = 1.4
Optimize Charge Dynamics Parameters (Step 2)
% Use only one current pulse for optimizing the charge dynamics i_pos=battery_data(1).i.*(battery_data(1).i>=0); a=find(diff(i_pos)>0,2); b = find(diff(battery_data(1).i)); c = fix((b(find(b<a(1),1,'last'))+a(1))/2); assignin('base','i_data',battery_data(idx_data).i(c+1:a(2))); assignin('base','v_data',battery_data(1).v(c+1:a(2))); assignin('base','t_data',battery_data(idx_data).t(1:length(i_data))); assignin('base','T_data',battery_data(idx_data).T*ones(length(t_data),1)); assignin('base','T0',battery_data(idx_data).T); assignin('base','Ts',t_data(2)-t_data(1)); % Find Battery initial charge before optimizing charge dynamics parameters assignin('base','ParsList',{'charge'}); InitGuessCharge = OptParsMain(3); OptCharge = fminsearch(@ee_battery_lse, InitGuessCharge, ... optimset('TolX', 1e-3)); assignin('base','AH0',OptCharge); % Optimize Battery charge dynamics parameters assignin('base','ParsList',[ParsListMain(2) ParsListDyn]); InitGuessDyn = [OptPars(2) InitGuessDyn]; OptParsDyn = fminsearch(@ee_battery_lse, InitGuessDyn, ... optimset('TolX', 1e-3)); % Update Battery block with optimized charge dynamics parameters ParsListMainDyn = [ParsListMain ParsListDyn]; OptParsMainDyn = [OptPars(1) OptParsDyn(1) OptPars(3:4) InitGuessMain(5) OptParsDyn(2:3)]; Pars = reshape([ParsListMainDyn; cellstr(num2str(OptParsMainDyn'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end assignin('base','AH0',AH*battery_data(idx_data).SOC0); % Display optimized parameters fprintf(['Optimized parameters for the Battery, ' ... 'including charge dynamics, are:\n']); fprintf('\t%5s = %s\n', Pars{:}); clear i_data v_data t_data T_data Ts clear i_pos a b c clear k clear OptPars OptParsDyn ParsListMainDyn InitGuessMain InitGuessDyn ParsListDyn ParsListMain
Optimized parameters for the Battery, including charge dynamics, are: Vnom = 3.699931 R1 = 0.05019736 AH = 2.603326 V1 = 3.526493 AH1 = 1.4 Rp1 = 0.005029392 tau1 = 109.691
Optimize Temperature Dependent Parameters (Step 3)
idx_data = find([battery_data(1:num_lines).T]~=25); assignin('base','t_data',battery_data(idx_data).t); assignin('base','i_data',battery_data(idx_data).i); assignin('base','T_data',battery_data(idx_data).T*ones(length(t_data),1)); assignin('base','T0',battery_data(idx_data).T); assignin('base','Ts',t_data(2)-t_data(1)) assignin('base','v_data', battery_data(2).v); % Use parameters for T1 as initial guess for T2 parameters InitGuessTemp = [OptParsMainDyn(1:2) OptParsMainDyn(4) OptParsMainDyn(6:7)]; Pars = reshape([ParsListTemp; cellstr(num2str(InitGuessTemp'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end % Optimize Battery temperature dependent parameters assignin('base','ParsList',ParsListTemp(1:3)); OptParsTemp = fminsearch(@ee_battery_lse, InitGuessTemp(1:3), ... optimset('TolX', 1e-3)); % Update Battery block with optimized parameters Pars = reshape([ParsListTemp(1:3); cellstr(num2str(OptParsTemp'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end assignin('base','AH0',AH*battery_data(idx_data).SOC0); % Display optimized parameters fprintf(['Optimized temperature dependent parameters for the Battery ' ... 'are:\n']); fprintf('\t%5s = %s\n', Pars{:}); clear i_data v_data t_data T_data Ts clear k clear OptParsMainDyn
Optimized temperature dependent parameters for the Battery are: Vnom_T2 = 3.9003 R1_T2 = 0.081404 V1_T2 = 3.8133
Optimize Charge Dynamics Parameters for Second Temperature (Step 4)
% Find index into data for non-room temperatures % Use only one current pulse for optimizing the charge dynamics i_pos=battery_data(idx_data).i.*(battery_data(idx_data).i>=0); a=find(diff(i_pos)>0,2); b = find(diff(battery_data(idx_data).i)); c = fix((b(find(b<a(1),1,'last'))+a(1))/2); assignin('base','i_data',battery_data(idx_data).i(c+1:a(2))); assignin('base','v_data',battery_data(idx_data).v(c+1:a(2))); assignin('base','t_data',battery_data(idx_data).t(1:length(i_data))); assignin('base','T_data',battery_data(idx_data).T*ones(length(t_data),1)); assignin('base','T0',battery_data(idx_data).T); assignin('base','Ts',t_data(2)-t_data(1)) % Find Battery initial charge before optimizing charge dynamics parameters assignin('base','ParsList',{'charge'}); InitGuessCharge = OptParsMain(3); OptCharge = fminsearch(@ee_battery_lse, InitGuessCharge, ... optimset('TolX', 1e-3)); assignin('base','AH0',OptCharge); % Optimize Battery charge dynamics parameters assignin('base','ParsList', [ParsListTemp(2) ParsListTemp(4:5)]); InitGuessTempDyn = [OptParsTemp(2) InitGuessTemp(4:5)]; OptParsTempDyn = fminsearch(@ee_battery_lse, InitGuessTempDyn, ... optimset('TolX', 1e-3)); % Update Battery block with optimized parameters OptParsTempDyn = [OptParsTemp(1) OptParsTempDyn(1) OptParsTemp(3) OptParsTempDyn(2:3)]; Pars = reshape([ParsListTemp; cellstr(num2str(OptParsTempDyn'))'],1,[]); for k=1:2:length(Pars) evalin('base',[Pars{k} '=' Pars{k+1} ';']) end assignin('base','AH0',AH*battery_data(idx_data).SOC0); % Display optimized parameters fprintf(['Optimized temperature dependent parameters for the Battery, ' ... 'including charge dynamics, are:\n']); fprintf('\t%5s = %s\n', Pars{:}); clear i_data v_data t_data T_data Ts clear i_pos a b c clear k clear OptCharge OptParsMain OptParsTemp OptParsTempDyn clear Pars ParsList ParsListTemp InitGuessCharge InitGuessTemp InitGuessTempDyn
Optimized temperature dependent parameters for the Battery, including charge dynamics, are: Vnom_T2 = 3.900266 R1_T2 = 0.07979468 V1_T2 = 3.813256 Rp1_T2 = 0.007920818 tau1_T2 = 160.2999
Display Optimized Curves
for idx_data = 1:num_lines assignin('base','t_data',battery_data(idx_data).t); assignin('base','i_data',battery_data(idx_data).i); assignin('base','T_data',battery_data(idx_data).T*ones(length(t_data),1)); assignin('base','T0',battery_data(idx_data).T); assignin('base','Ts',t_data(2)-t_data(1)); out = sim(Model); v_model{idx_data} = out.Vo.signals.values; t_model{idx_data} = out.Vo.time; end plot([battery_data(1:num_lines).t]/3600, [battery_data(1:num_lines).v], 'o', [t_model{:}]/3600, [v_model{:}]) xlabel('Time (hours)'); ylabel('Battery voltage (V)'); legend([legend_info_data legend_info_model], 'Location', 'Best'); title('Model with Optimized Parameter Values');
Validation using a dynamic current cycle
idx_data = 3; assignin('base','t_data',battery_data(idx_data).t); assignin('base','i_data',battery_data(idx_data).i); assignin('base','T_data',battery_data(idx_data).T); assignin('base','T0',battery_data(idx_data).T(1)); assignin('base','Ts',t_data(2)-t_data(1)); assignin('base','AH0',AH*battery_data(idx_data).SOC0) out = sim(Model); subplot(3,1,1) plot(t_data, battery_data(3).v, 'o', out.Vo.time, out.Vo.signals.values) xlabel('Time (s)'); ylabel('Battery voltage (V)'); legend('Data','Model', 'Location', 'Best'); subplot(3,1,2) plot(t_data,i_data) xlabel('Time (s)'); ylabel('Current requirement (A)'); subplot(3,1,3) plot(t_data,T_data) xlabel('Time (s)'); ylabel('Temperature (^oC)'); title('Model validation');
bdclose(Model) clear num_lines legend_info_data legend_info_model out v_model t_model clear battery_data idx_data Ts Model