Stochastic gas signal generator: Training, generation and extrapolation of time series related to gas species

摘要

When sampling physical properties of natural gas transported within natural gas distribution grids, like the gas species concentrations or thereof dependent properties like the standard density, alternatively to gas tracking realized by numerical simulations, sensor-based gas tracking becomes feasible. Thus sampled time series are processed to determine the transit times and fractions of gas that is originating from sampled upstream nodes, e.g. gas entry nodes, and that is contributing to the sampled gas of downstream nodes, consumed by gas customers at exit nodes. Such time series appropriate for sensor-based gas tracking are rare, since sensor-based gas tracking is still in an early stage of development. And for the development of signal processing techniques around sensor-based gas tracking more such data is required. To bridge that gap, we are introducing a probabilistic signal model and a system model for the generation of synthetic datasets. We consider the characteristics of the sampling process as well as deterministic transmission properties of natural gas distribution grids. With our approach, the generation of complete sets of data representing gas compositions corresponding to specific types of gas, e.g. high calorific natural gas or upgraded biogas etc., is feasible. The resulting datasets exhibit high statistical and visual similarities to real-world data sets obtained by sampling gas along gas distribution grid nodes. Synthetic data, generated for virtual upstream nodes, is extrapolated to downstream (customer or exit) nodes, taking the transmission characteristics of natural gas distribution grids into consideration. Thus derived upstream and downstream node data sets can be seamlessly taken for the evaluation of signal processing methods around sensor-based gas tracking.