Gas Distribution Networks
- Infrastructure Data Cleansing
Cleansing and rationalisation of pipe and failure datasets
Sorting by material and diameter (rationalisation), attributing failures to pipes. Sense-checking.
- Environmental Data preparation
The following environmental datasets will be sourced and prepared.
Soils, Hydrology, Ground movement potential models (clay, sand, silt, peat), Soil corrosivity, etc
Weather - MORECS – Soil Moisture Deficit, Temperature, Rainfall, Days Air Frost. These raw data will be used, as well as some new, derived values (accumulated temperature below x degrees, delta SMD etc) which we will create:
- Intersection of pipes and failures datasets with spatio-temporal environmental data, (above)
- Modelling of number of pipe failures.
- Development of reporting graphs, maps and tables. In consultation with NGN develop a suite of reporting outputs inform decision making.
In brief, our modelling strategy is likely to involve fitting generalised linear models (specifically, Poisson regression) to the number of bursts for each week in each pipe-material-diameter-soil combination in each MORECS square. The length of pipework within that class will be used as an offset in the model, which can therefore be thought of as a model for the average rate of bursts in a unit length of pipe. We will divide possible explanatory factors and conditions (for example: shrink-swell potential, temperature, SMD) into groups, within which we will seek the best possible member of that group. For instance, when considering accumulated temperature below a certain threshold over a given preceding number of days, we will compare models that include a number of different thresholds, and number of different antecedent periods. Assuming a global optimum is
- desired, we will consider all possible combinations of explanatory variables from each group. Model choice will be made on the basis of Akaike’s Information Criterion (AIC).
This should provide NGN with a set of predicted and observed number of leaks / bursts per month from which we can test the suitability of the models to possibly identify more local areas which need pipe replacement prioritisation. (This prioritisation itself would lie outside the scope of this project.)
By the end of this project we hope to have proven models that, by the addition of weather / temperature variables, give robust data to better inform NGN’s iron pipe replacement programme.
The models will be for different pipe materials and diameters, which predict the number of leaks / failures that should be expected each month, based on the soil in which they are buried and the observed weather