Dachverband der Österreichsichen Sozialversicherungsräger
In your modell you should have done a single parameter sensitivity analysis as well as a probabilistic sensitivity analysis.The observed data - hopefully - are in the area covered by the probabbilistic sensitivity analysis. If so (and also if not) the sensitivity analyses show you which parameters "move" the results of the model in the direction of the observation. Now you can reevaluate the quality of your parametrization or even of your modelling approach.
If your model and reality are totaly different check the model ;-))
It depends on the questions / field of science you are in. Sometimes scale/granularity of the model has to be adjusted. Sometimes the model misses relevant causal connections.
We know that conducting experiments to get observed data is time-consuming and expensive sometimes. That is one of the reason why you made the model using software to obtain simulated data. In order to ensure the accuracy of the model, you should compare the simulated data (from model) with the observed data (from experiment). If the simulated data is close to the observed data, your model is proved to be accurate and reliable for future use. If it is not, you should check your model again to ensure that all conditions (materials, constraints, loadings, ....) of the model describe the experiment setups exactly. Prof. Harald Kirchsteiger already mentioned this in his comment, I just made it a bit more clear.
After successfully verify the model with the experiment by validating the simulated data with the observed data, now you can confidently use your model to simulate with different model parameters. There are some ways to analyse and develop your idea further:
1. Sensitivity analysis as recommended by prof. Gottfried Endel is worth it to consider. By doing so, you can determine the trend/movement of the obtained data (simulated results). You can also determine the most and the less sensitive parameters to your simulated results.
2. When you plot the experiment data or simulated data with respect to a single input parameter, you possibly have a curve. After that, you can use curve fitting functions in MATLAB or use some available Artificial Neural Network (ANN) packages to predict the explicit mathematical expression of the simulated results (the accurate fitting curve is the curve with lowest errors). This expression/formulation can be used directly for practical designs and further. They are reliable is due to your model and simulated results are already verified before. You should consider MATLAB Curve fitting functions or ANN packages are just a tool to be used.
3. The idea in section 2. right above can be developed further. Now you possibly try to consider multiple input parameters, you possibly have a surface. At this point, some MATLAB surface fitting functions or Polynomial Neural Network (PNN) might be helpful. It might be difficult, but I think that if we can understand clearly the data and can see the data trend by naked eyes, we can do something with it.
The first thing you would want to do is to validate your model with the observed (actual data), herein lies the presentation and analysis of the data. This is done using statistical tools (software). You might want to determine statistical parameters such as chi-square, correlation coefficient, etc. in comparison and validation of your data. If your predicted data highly represents the observed value, then your model is a good fit (well describing the phenomenon).
Two-dimensional and three-dimensional tide models were used to simulate the M2 tide in the Gulf of Maine. Model estimates of changes to the tide caused by a tidal barrier in the upper Bay of Fundy were made and compared. Tidal amplitudes in the presence of a barrier increased 30-50 cm for both models. The three-dimensional model uniformly produced...
This paper presents the up-to-date state in computer-aided modelling and simulation of the extrusion. There are pointed out the main aspects of this process, which have been taken into account by researchers and for which computer softwares for modelling, simulation and design are commercially available.