Neural Net few traning data

Hi everyone,

First of all i want to mention that i recently started working with NN's with no previous experience. For my diploma thesis i have the following problem. i want to built a nn that will have 7 input values and will return 1 output. Now, 5 of my input values represent a certain working condition of a power plant, from now i will mention these values as ''case''. The other two inputs are parameters that change within the same case. Totally i have 21 defferent cases (5 inputs). for the two inputs that are parameters i can generate as many data as i want with curvfitting tooll. Initially i built a nn using nnfitting tool, just for one case, that had 2 inputs and one output and it worked fine. Now i am trying to built a nn that will also take as inputs different cases (5+2 inputs) but due to few training data (cases) my nn does not perform well at all when i am trying to make a prediction for a case that it has not been trained. My question is, i can not get any more case data (5 inputs) i can only dense my parameter inputs (2 inputs) for each case. Is there any way i can train a nn with just these few 21 case data?? (for each case i have 300 data of 2 parameter inputs + output which i can increase if nessecery). Please if you are going to answer to this topic consider my low experience.