Search

Flexible district heating grids form an important part of future, low-carbon energy systems. We examine probabilistic state estimation in such grids, i.e., we aim to estimate the posterior probability distribution over all grid state variables such as pressures, temperatures, and mass flows conditional on measurements of a subset of these states. Since the posterior state distribution does not belong to a standard class of probability distributions, we use Markov Chain Monte Carlo (MCMC) sampling in the space of network heat exchanges and evaluate the samples in the grid state space to estimate the posterior. Converting the heat exchange samples into grid states by solving the non-linear grid equations makes this approach computationally burdensome. However, we propose to speed it up by employing a deep neural network that is trained to approximate the solution of the exact but slow non-linear solver. This novel approach is shown to deliver highly accurate posterior distributions both for classic tree-shaped as well as meshed heating grids, at significantly reduced computational costs that are acceptable for online control. Our state estimation approach thus enables tightening the safety margins for temperature and pressure control and thereby a more efficient grid operation. • Novel approach combining classic statistical analysis and AI. • The estimated state distributions are not standard distributions. • MCMC and Bayesian calculus can be used to approximate state distributions. • MCMC requires mapping demands to grid states in every step. • Deep Neural Networks can speed this up by orders of magnitude. [ABSTRACT FROM AUTHOR]

Abstract: For a long time, classical ensemble forecasts have concentrated on simulating the uncertainties of the observations by variations of the initial and boundary conditions. To supplement this with inherent contributions from the forecast model itself, stochastic physics has been proposed as an appropriate approach to improve the forecast quality and to increase the ensemble spread of underdispersive ensembles. Concentrating on cloud processes and their temporal development we here introduce a set of stochastic parameters into existing deterministic physical parameterization schemes as for the trigger function, closure assumption and turbulent mixing in the convection scheme, for the initial ice concentration and turbulent collection kernel in the grid scale precipitation/microphysics scheme and for the subgrid scale cloud cover in the radiation scheme. Considering the stochastic parameters to be subject to unresolved turbulent fluctuations, we derive the corresponding stochastic processes from a Langevin equation based on an approach of direct numeric simulation. This allows for the inclusion of a turbulent autocorrelation time calculated from the eddy dissipation rate and the turbulent kinetic energy of the hosting model. Ensemble simulations of two meteorologicaly different real cases are compared to deterministic control simulations and observations of 1h precipitation sums. The turbulent autocorrelation time shows pronounced spatial variability correlated with frontal dynamics and convective activity. Maps of spatial precipitation patterns and continuous ranked probability score (CRPS) improvements show a better forecast quality of the ensemble mean including stochastic physics. Depending on the meteorological situation, the ensemble spread and the CRPS improvement both exhibit a spin up time and reach a saturation. For air-mass convection/postfrontal showers the spread and the CRPS improvement are higher than for frontal precipitation. The stochastic parameterization of subgrid scale cloud cover is found to degrade the forecast quality of the ensemble. This is explained by a too homogeneous spatial distribution of the added variability. Consequently, this part of the stochastic parameterization has been excluded from the ensemble simulations and is left in its operational deterministic version. [Copyright &y& Elsevier]

Abstract: A method is presented to reduce the time-splitting errors of one-dimensional flux form advection schemes. In multidimensional applications deformational correction terms are included in each one-dimensional advection step whereby all correction terms add up to zero. With the correction terms a consistent form of a given one-dimensional advection scheme is achieved when time-splitting is performed. Positive definiteness and shape preservation of the scheme are obtained by nonlinear limitation of the advective fluxes and the deformational correction terms. Large Courant numbers are treated by applying local timestep reduction. The resulting scheme is strictly one-dimensional, mass conservative and positive definite. Shape preservation of the method is nearly achieved. Two- and three-dimensional test runs are presented demonstrating that the method yields satisfactory results. The horizontal flow in a terrain with a steep mountain ridge is investigated by utilizing a terrain following vertical coordinate. It is shown that even in steep orographical terrain the method works quite well, however, small errors occur in the shape preservation of the scheme. These errors increase with increasing spatial variation of the vertical grid mesh. [ABSTRACT FROM AUTHOR]