Approximation of a marine ecosystem model by artificial neural networks designed using a genetic algorithm

Data from the Paper: Approximation of a marine ecosystem model by artificial neural networks designed using a genetic algorithm.

Abstract: 

Marine ecosystem models are important to identify the  processes that affects for example the global carbon cycle. Computation of an annually periodic solution (i.e., a steady annual cycle) for these models requires a high computational effort. To reduce this effort, we approximated an exemplary marine ecosystem model by different artificial neural networks. We used a fully connected network, then applied the sparse evolutionary training  (SET) procedure, and finally applied a genetic algorithm (GA) to optimize both the   network topology. With all three approaches, a direct approximation of the  steady annual cycle  was not sufficiently accurate. However, using the mass-corrected prediction of the ANN as initial concentration for additional model runs, the results were in very good agreement.   In this way, we achieved a runtime reduction by about 15 \%. The result from the SET algorithm were comparable to those of the full network. Further application of the GA may lead to an even higher reduction.

Content:

Database sqlite ANN_Database.db

zip-files with data: 

ANN-Data.zip structure and weights of used networks

ANN-Results.zip results obtained with networks

Reference-Results.zip reference results and training data