| Title | GeoLingo: Defining Optimal Data Collection Requirements for Geothermal Operations and Advanced Analytics |
|---|---|
| Authors | Andrew MARSH, Paul SIRATOVICH, Nicole TAVERNA, Grant BUSTER, Jon WEERS |
| Year | 2023 |
| Conference | Stanford Geothermal Workshop |
| Keywords | data, instrumentation, compression, conditioning, machine learning, advanced analytics, integrated modeling, GOOML |
| Abstract | This paper sets out the optimum data collection methodologies to enable effective data-driven decision making in geothermal operations. The focus is on the performance of the overall geothermal system comprising reservoir, steam field gathering, power plant units, and injection rather than the sub-systems of the power plants. An overall system dataset including power plant performance at the unit level is sufficient to feed models that can give insights for real-world operations. The data collection requirements have been set out for different types of geothermal systems including steam, two-phase, and liquid-dominated wells. The two main types of power plant, flash and binary, are covered. Suggested resolution and intervals of measurement are provided with rationale for how such volumes of collected data would be useful. The intent of this paper is to offer guidance to future or expanding geothermal developments by providing a minimum set of instrumentation and monitoring requirements that will allow operators to use advanced analytics. It may also serve as a useful guide when considering retrofitting and upgrading existing plant. Retrofitting instrumentation will almost certainly yield gains where any previous efforts to optimise existing facilities have been frustrated through lack of data. Improving sensor technologies, analytics algorithms and software make retrofitting any gaps in data increasingly attractive over time. This exercise leverages off the GOOML project (Geothermal Operations Optimisation through Machine Learning (see Buster et al, 2021) where experimentation with Machine Learning algorithms has informed what is and is not possible to optimise given certain data inputs. Data treatment and conditioning pitfalls are also explored to facilitate upfront design of analytical routines. The benefits from big-data analytics can only be realised if data is collected from the right places with sufficient resolution to allow the algorithms to develop an accurate representation of the state of the system and how it is evolving. |