Algorithm 1. MATLAB® code for GRNN parallel training algorithm

% Setting the spread constant population vector with step s Spread = [Spread_min:s:Spread_max]; % Parallel computing loop for calling the built-in GRNN training function parfor ii = 1:length(Spread)   pop_GRNN(ii).net = newgrnn(Inputs_train,Targets_train,Spread(ii)); end % Calling the built-in simulation function of the trained GRNN with % training inputs in parallel computing mode for ii = 1:length(Spread)   Outputs_train = sim(pop_GRNN(ii).net,Inputs_train,...     ‘useParallel’,’yes’);   % Calculation of the absolute error of the trained network   Err = Targets_train − Outputs_train;   % Calculation of the average absolute percentage error of the trained   % network   pre_MAPE = abs(Err./Targets_train);   mean_MAPE = mean(pre_MAPE(isfinite(pre_MAPE))) * 100;   % Conditional storage of the corresponding variables in the pop_GRNN   % structure    if mean_MAPE < max_MAPE     pop_GRNN(ii).Spread = Spread(ii);     pop_GRNN(ii).Outputs = Outputs_train;     pop_GRNN(ii).MAPE = mean_MAPE;   else     % Premature termination of the cycle     break   end end % Identification of the MAPE maximum value index in the pop_GRNN % structure [~,k] = max([pop_GRNN.MAPE]); % Setting the variables of the found values of the corresponding % parameters if ~isempty([pop_GRNN(k).Spread])   spread = pop_GRNN(k).Spread;   net_GRNN = pop_GRNN(k).net;   Outputs_GRNN_train = pop_GRNN(k).Outputs;   % Removing empty fields from the pop_GRNN structure   pop_GRNN = pop_GRNN(1:length([pop_GRNN.Spread]),:); else   % Terminate execution   return end