how to modify this code of loop.

I am doing time series prediction with gaussian regression and get some problem with the code. In the code below, I am trying to do a prediction from today 18:00 until next day 24:00 every 3 hours.

(18-21-24-3-6-9-12-15-18-21-24)

In the code, the first step is to make a prediction of 21:00 from 18:00, so I write the code

case a>=3%21h   
    wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*0.9,1);

The next step is to make a prediciton of 24:00 based on the prediction value of 21:00 we get.

after that, I need to make a prediction of 3:00 based on the prediction value of 24:00 we get.

Then, I need to make a prediction of 6:00 based on the prediction value of 3:00 we get.

So I need keep on doing this way until next day 24:00

Anyone can help me how to modfiy the code ? thank you very much.

    for i=3:3:30% prediction from today 18:00 to next day 24:00 
        ysd = zeros(30,1);
        x_tmp(i,:) = wbgt_tmp(end+1-(input_time(end:-1:1)))';
        [y_pred(i,1) ,ysd(i,1),yint] = predict(regressionGP, x_tmp(i,:),'Alpha',0.05);
        y_tmp(i,1) = predict(linearModel, x_tmp(i,:)) ...
            + y_pred(i,1) + 0.5*ysd(i,1);
        a=rem(i,30)
        switch(1)
            case a>=3%21h   
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*0.9,1);
            case a>=6%24h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*0.7,1);
            case a>=9%3h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*1.5,1);
            case a>=12%6h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*1.8,1);
            case a>=15%9h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)+ysd(i,1)*1.5,1);
            case a>=18%12h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)+ysd(i,1)*1.4,1);
            case a>=21%15h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*0.8,1);
            case a>=24%18h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*0.9,1);
            case a>=27%21h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*0.5,1);
            case a>=0%24h
                wbgt_tmp(end+1,1) = round(y_tmp(i,1)-ysd(i,1)*0.3,1);
        end
    end
    wbgt_predict(:,month_days) = round(0.95*wbgt_tmp(end-23:end)+2.1);