Read the full report here.
Soil resistivity measurements are required for the earth electrode design of many structures like telecommunication towers, photovoltaic power plants, and overhead power lines to comply with design standards. For long lines of high voltage overhead power lines, this may require the laborious task of soil resistivity measurement at every tower site as well as a multistep iterative design and construction process. As part of the initial design phase, time and money can be saved by having a model that predicts the soil resistivity based on freely available geospatial data.
In this study, the feasibility of this soil resistivity prediction was investigated. This was done using descision trees trained on geospacial data such as bedrock depth, runoff, relief index, and biome type. The accuracy of the models produced in this study were between 48% and 54%.
The lack of accuracy was likely due to the limited size of the data set used. This led to the hypothesis space becoming too restricted for the decision trees to make accurate predictions and a need to reduce the amount of predictor variables used to train them.
To improve on this study, the most important aspect is to increase the data set size used for training and testing the decision trees. This would allow the addition of more predictor variables to the models which might lead to increased discrimination power and more accurate predictions. Alternatively, other machine learning methods that might be better suited for use in restricted hypothesis spaces can be tested on this data set.
Soil resistivity is an electrical property of soil which describes to what extent it resists electric current flowing through it [1]. The earth electrode design of many structures like telecommunication towers [2], photovoltaic power plants [3], and overhead power lines [4], [5] are dependent on the soil resistivity at the location where they will be constructed.
In the case of high voltage overhead power transmission lines for example, the tower foot resistance must be below a specific value to comply with design standards. In South Africa the towers of a 400 kV power line must have a tower foot resisance of below 40 Ω [4]. This tower foot resistance depends highly on the local soil resistivity at the tower. Due to high soil resistivity at the location of construction of some of these towers, this design value is sometimes not met after the first design iteration [2], [4].
Towers that are found not to comply with the design standard have to be modified or retrofitted after the initial setup and construction to reduce the tower foot resistance to a satisfactory level. This requires a lot of extra time and money to fix. In some cases the towers need to be addressed on an individual basis and a unique re-design or updated design is required for each one [4].
The iterative design process just described includes many stages: These include the initial design and construction of the towers; the measurement of tower foot resistances for each tower after initial construction to see if it is below the design requirement; the re-design and modification of some of the towers that did not meet the requirement; and finally, after modifications are made, follow-up tower foot resistance measurements to see if the modifications were adequate [2], [4].
This process can be improved and the number of steps in this process can be reduced if reliable soil resistivity information was available without having to do physical measurements. The focus of this paper is to evaluate the feasibility of soil resistivity prediction in South Africa for this purpose. If the soil resistivity can be reliably predicted, the initial route layout and design of overhead power transmission lines can be adjusted so that no re-design and modification is required later on which will save time and money. For this study, prediction of soil resistivity was done by using freely available geospaial data.
The machine learning method selected for relating this geospatial data (input features) to soil resistivity (target variable) was the decision tree. Decision trees were used for this purpose mainly because of the human readability of their output. The hypothesis of a decision tree is represented as a series of if-then rules [6]. This is advantageous from an engineering design point of view since design decisions need to be transparent and clearly supported. At the time of writing, no other methods were found in literature that attempted soil resistivity prediction with machine learning methods. Especially for imporving the design process of overhead power lines.
This work was funded by the DSI-NICIS National e-Science Postgraduate Teaching and Training Platform (NEPTTP). Some of the soil resistivity data which the CDEGS soil models were constructed from was kindly shared by: Barry Reid from Royal Haskoning DHV; Gary Thoresson from The Testing Guys; Ivan Grobbelaar from DEHN Africa; Johann Rossouw from EPCM Solutions; Dr. Pieter Pretorius from Terratech; Theunus Marais from Eskom; and Trevor Manas from LP Concepts.
-
[1] M. Kizlo, A. Kanbergs, ''The Causes of the Parameters Changes of Soil Resistivity, the causes of the parameters changes of soil resistivity.'' Electrical, Control and Communication Engineering 25.25, 2009, pp. 43-46.
-
[2] L. W. Choun, C. Gomes, M. Z. A. Ab Kadir, W. F. Wan Ahmad, ''Analysis of earth resistance of electrodes and soil resistivity at different environments,'' 2012 International Conference on Lightning Protection (ICLP), Vienna, 2012, pp. 1-9, doi: 10.1109/ICLP.2012.6344314.
-
[3] P. H. Pretorius, ''Loss of equipotential during lightning ground potential rise on large earthing systems,'' 2018 IEEE International Symposium on Electromagnetic Compatibility and 2018 IEEE Asia-Pacific Symposium on Electromagnetic Compatibility (EMC/APEMC), Singapore, 2018, pp. 793-797, doi: 10.1109/ISEMC.2018.8393890.
-
[4] P. H. Pretorius, B. Ntshuntsha, S. Ramadhin, ''Application of counterpoise in the reduction of tower footing resistance - low frequency design and case study,'' Cigre Symposium Cape Town - 2015, South Africa, Session 9 – Paper 7, 2015.
-
[5] Jinxi Ma, F. P. Dawalibi and W. Ruan, ''Design Considerations of HVDC Grounding Electrodes,'' 2005 IEEE/PES Transmission & Distribution Conference & Exposition: Asia and Pacific, Dalian, 2005, pp. 1-6, doi: 10.1109/TDC.2005.1546811.
-
[6] T. M. Mitchell, ''Machine learning,'' Singapore: McGraw-Hill, 1997, pp. 52-80.
-
[7] IEEE Guide for Measuring Earth Resistivity, Ground Impedance, and Earth Surface Potentials of a Grounding System, in IEEE Std 81-2012 (Revision of IEEE Std 81-1983), vol., no., pp.1-86, 28 Dec. 2012, doi: 10.1109/IEEESTD.2012.6392181.
-
Department of Agriculture, Fisheries and Forestry - South African Environmental GIS Data, ''Soil and Terrain Database (SOTER) for South Africa,'' March 17, 2020, Distributed by Open Africa. https://africaopendata.org/dataset/department-of-agriculture-fisheries-and-forestry-south-african-environmental-gis-data
-
International Soil Reference and Information Centre (ISRIC) - Soil and Terrain (SOTER) database programme, ''Environmental GIS Data,'' November 17, 2015, Distributed by ISRIC Data Hub. https://data.isric.org/geonetwork/srv/eng/catalog.search#/metadata/c3f7cfd5-1f25-4da1-bce9-cdcdd8c1a9a9
