GIS based decision support for gas pipe maintenance

4.3 Cost component
Cost calculation is one of the standard tasks of controlling. Most of the cost influencing factors for all technologies can be taken directly from the GIS.

Cost can be divided into direct and indirect cost. Direct cost depend on the reconditioning technique, side characteristics and engineering work. A lot of data for calculating prices can taken from data stored in the spatial database.

In practice, indirect cost are often not considered by the utility companies, because they do not have to pay it. These are costs for traffic obstruction – coming to late to office and need more petrol – or lower sales in shops cause people cannot reach a shop because of the construction side. Indirect cost are even higher with open pit technology and so it is important to use an estimation for these cost for the maintenance technique decision process.

4.4 Cluster analysis
Pooling of construction sides helps to reduce fix costs. Standard cluster methods will not work because of two reasons:

  1. Savings are largest when using a single technology for the amalgamated sites. That is why the technology choice may be revised.
  2. Large clusters are often undesirable because of traffic obstruction and duration of civil works.
To tackle problem 1), a saving matrix is derived and stored which gives rules for calculating what cost components a pooling of side 1 using technology A and side 2 using method B creates (see figure 6 for an example). The matrix is only technology dependent and has to be derived only once to be stored in the GIS database.


Figure 6: Example for saving matrix


Considering the problem 2), the pooling process should be guided by the quest to maximize cost savings directly or by upgrading the priority metric as indicated in figure 5 to

priority metric, if sites combined = scoreSite1 + scoreSite2 / costSite1 + costSite2 – savings

The problem of pooling construction sites is totally different from typical clustering problems, where the majority of objective functions are based on proximity of points as measured by Euclidian distance.

DBScan (Density based spatial clustering of applications with noise, Ester et. al. 1996) is a powerful density based cluster algorithm. In its basic version points are in an Euclidian space. Cluster nuclei are formed by regions of high density (MinPts) inside an e-Neighbourhood. Upon termination of the algorithm, each point is either

  • a core point inside a cluster with at least MinPts objects in its e-Neighbourhood
  • a boundary point which lies in the e-Neighbourhood of a core point
  • an outlier, which is a point in a low density region (see figure 7)

Figure 7: Algorithm DBScan with MinPts=5 and e=1 unit


DBScan works well in 2- and 3-dimensional space and has a good running time of O(nlogn) where n is the number of points.

GDBScan (Sander, 1998) generalizes the concept of e-Neighbourhood by replacing the Euclidian distance with an arbitrary quasi-distance function. For the MinPts parameter the points are not simply counted. Other similarity functions can be used which may depend on spatial and non-spatial characteristics of the objects.

For the problem of gas pipe maintenance the travel distance along the pipe network can be used as pseudo-distance function. The parameter e is set to 1 km, a typical threshold for maintenance techniques. The priority metric defined above can serve as similarity function.

Preliminary results with small networks in a prototypical implementation show very promising and plausible results (see figure 8), but some work needs to be done like the choice of e and MinPts will seriously influence the solution.


Figure 8: Exemplary cluster solution in the prototypical implementation


5. Summary and Outlook
Extending GIS functionality through the model base described will assure the transition from a information to a decision support system. Maintenance of gas pipes has been used as prototype. This is an application of medium complexity with immediate utility to the end user. The problem can only be solved through close interaction with quite different fields of research: the technicalities of pipe refurbishing, economic considerations to identify key cost drivers and the algorithmic side, where new cluster algorithms had to be developed.

Gas Pipe Maintenance is only a small part of the work a public utility company faces. Next planned steps are Decision Support Systems for water, sewage and electric power companies. Water and sewage need only minimal adaptations while electricity is more complicated because of different techniques and voltage regimes.

References
  • AME Info FZ LLC (Ed.): Utilisation of natural gas in the Middle East to rapidly overtake oil as governments commit to massive investment in the industry, www.ameinfo.com/60285.html, May 2005.
  • Armstrong, Densham: Organization Strategies for Spatial Decision Support Systems, IJGIS, vol. 4, 1, p. 3-20, 1990.
  • DVGW (Ed.): Technical Standard G 401. Entscheidungshilfen fuer die Rehabilitation von Gasverteilungsnetzen. DVGW – German Technical and Scientific Association for Gas and Water, 1999.
  • Ester, Kriegel, Sander and Xu: A density based algorithm for discovering clusters in large spatial databases with noise, Proc. 2nd Int. Conf. on Knowledge Discovery and Data Mining, Portland, Oregon, 1996.
  • Keenan: Spatial Decision Support Systems, in: Mora, Forgionne and Gupta (Ed.): Decision Making Support Systems. Achievements and challenges for the new decade, Idea Group Publishing, p. 28-29, 2003.
  • Sander: Generalized density-based clustering for spatial data mining, Ph.D. Thesis, Munich, 1998.
Page 3 of 3
| Previous |