A Comparison of DSD With a Neural Net Model to State/Parameter Estimation Directly by Neural Nets

摘要

The DSD is a model independent state/parameter estimation software. An important feature of DSD with respect to other estimation schemes is that it can yield a likelihood ranking of possible system states. Neural net system models can reduce the estimation time with DSD compared to system models consisting of differential equations. Using a third order differential xenon evolution model to train the neural net models, accuracy of xenon tracking with DSD using a neural net model of xenon evolution (Method 1) is compared to direct tracking with the neural net model (Method 2). In the estimation process, 2% noise was assumed in one of the inputs. While Method 1 yielded good results, Method 2 results showed significant differences from the actual xenon levels when the nets were trained with no noise in the input. Method 2 showed good agreement with the true xenon levels when the nets were trained with noisy input.