Improving the Rate of Geothermal Drilling

for Mighty River Power

Hannah Norton (MSCI honours) and Georgina Richards (Geology honours)


Company background

Mighty River Power (MRP) is an integrated energy generation, trading, retailing and metering business. MRP operates under three company names, each for a different sector of their operations. The MRP side of the business deals primarily with energy generation, retailing is carried out under the trademark Mercury Energy and metering under Metrix. MRP’s focus is on producing energy from sustainable and renewable sources; hence the Waikato River hydro system makes up the core of their generation capacity. However, geothermal energy is increasingly making up a larger portion of this capacity, as well as contribution from wind farms, co-generation stations and bio-energy production. Our project focuses on the geothermal aspect; this is MRP’s second largest source of energy, representing 20% of their production. Mighty River Power, in partnership with iwi, own and/or operate a total of four geothermal plants all within the Taupo and Bay of Plenty regions; Rotokawa, Nga Awa Purua, Mokai and Kawerau. They also have a fifth under development, Ngatamariki, which is expected to be online in 2014.

Problem Situation

MRP wished to gain insights into the drilling process at their geothermal sites. This process is an integral part of geothermal power generation, but it does however, come with great costs. Therefore ensuring that the efficiency of this process is maximised is of great importance. The design and planning process for drilling new wells is predominantly based on the overall performance and drilling history of nearby wells. MRP were interested in improving this process by more detailed research, modelling and analysis in order to increase efficiency.

Efficient drilling occurs when the rate of penetration (ROP) is increased; various controllable drilling parameters (such as weight on bit, rotary speed and flow volume) can be varied to affect this. The aim of this project was to identify any relationships between the drilling ROP and the drilling parameters used while drilling. This information could then be used to answer the key question; how can the controllable drilling parameters be manipulated, with knowledge of the uncontrollable parameters, so that ROP is maximised? One of the key uncontrollable drilling parameters is the rock type that is being drilled through; this is the main aspect we focussed on when deciding on how to change the controllable parameters.

Literature Review

The first part of this project was to complete a comprehensive literature review to explore the approaches that have been taken to increase drilling speed in geothermal (and similar) drilling processes. The aim of this step was to provide MRP with an overview of what has been done before and the benefit to which these systems have resulted. Limited research on drilling in geothermal settings was found, but a large area of research on oil and gas drilling was reviewed. A variety of methods have been used in the literature, ranging from mathematical models to statistical approaches such as linear regression and to simulations. The majority of the research reported large cost savings.

Data analysis

MRP supplied us with data from previous drill holes at the Kawerau geothermal plant. Using this data we were able to analyse which circumstances produced the faster ROPs. We used three separate statistical methods to analyse the data: multiple regression, association rules and neural networks.

1.      Multiple Linear Regression

We used regression analysis as starting point to gain an initial understanding of the data and to reveal any clear relationships.


2.      Association rules

Association rules were used to gain a better understanding of the relationships. Association rules explain “what goes with what” and its output is in the form of “If, then” statements (i.e “If (parameter) occurs then this implies that (parameter) will also occur”. The output of this analysis was easily transferrable into recommendations. In order to utilise association rules we had to out the data into categorical form. This was done by classifying each parameter as small, moderate, large or very large, and then determining ranges which these categories would cover.


3.      Neural Networks

We used neural networks to establish more complex relationships between ROP and the drilling parameters. The method of linear regression assumes a linear relationship between the dependent and independent variables, whereas neural networks are able to model non-linear relationships. From the association rules analysis, we were able to identify some relationships, but decided it was important to apply another method to provide confirmation of these and to uncover any further relationships. Once again ROP is the dependent variable and the drilling parameters are the independent variables. The output of neural networks does not shed light on the patterns in the data in the same way that regression and association rules do. In order to uncover useful relationships between ROP and the drilling parameters we had to perform some sensitivity analysis of the neural network output.

For each analysis we separated out data for the different rock types and analysed these individually. This way we were able to determine how the controllable parameters affected the ROP between the different rock types; an important piece of knowledge when drilling. The results were compared between methods to determine final results. The validity of each method was also tested by examining various error measurements.


Results were in the form of recommendations of levels of the controllable parameters to apply for drilling through each individual rock type. Due to confidentiality, we are unable to publish results for public viewing.


We would like to thank the team at MRP for the opportunity to complete this project. Of particular mention are Joe Gamman, Ben Pezaro, Frank Walsh, Scott Curran and Tom Powell who gave up their time to provide us with information and data when needed. We would also like to thank our supervisors at the University of Canterbury; Associate Professor Don McNickle and Dr John Raffensperger from the Department of Management, and Dr Darren Gravely and Dr Ben Kennedy from the Department of Geology. Their support and guidance has been invaluable for the completion of this project.