A detailed analysis of the physical system should guide you to determine what are the most relevant parameters to input into the model. Then, at the validation stage, you can try some of those parameters and observe model results. Assessment of the model quality is carried out by comparing model output against known results (observations). The combination of input parameters and ANN architecture which gives an output closest to the observations is obviously the answer to your question. See: Article Electricity price short-term forecasting using artificial ne...
The number of input parameters of the artificial neural network, the same as for any modeling procedure depends on the characterístics of every concrete problem and it is determined by the systemic analysis of the task you need to solve. The systemic analysis leads to the conceptual mathematical model of your task. The conceptual mathematical model includes all the inputs and outputs. Those entrances and exits are corresponde to the ANN or other model. I am attaching you a work where you can find an analysisand synthesis methodology so you would be able to formulate and to solve the task you could to face.
If you are dealing with experimental data, a preliminary regression model in coded form will help identify most affecting variables based on relative statistical significance. The analysis should help identify input variables thereby simplifying your ANN network.