A Time Series Approach for Pipe Network Simulation

We applied time series predicting tools for the simulation of the temporal behavior of large pipeline networks submitted to timely changing inputs. The inputs may consist of a set of specified flow rates at client or supply nodes, while the outputs are another set of nodal pressures and internal flow rates. According to the topology, size, age and history of the network, the continuous generation of phenomenological dynamic simulations may be impossible, imprecise or numerically expensive, demanding thus alternative approaches. Our methodology is particularly oriented to this kind of demand. From recorded network past data covering relevant history of inputs and selected outputs, ARX-MIMO predictors are built with identification methods and launched for continuous estimation of the network outputs one time step ahead. Results are precise enough for engineering, training and monitoring applications.