CLIENT
The client, a prominent energy company, strategically operates in the upstream, midstream, and power generation sectors. With concessions spanning across Bahia, Espirito Santo, Rio Grande do Norte, and Alagoas in Brazil, the client took a significant step in 2017 by acquiring marginal fields from Petrobras, driven by the goal of enhancing overall gains. As a company committed to addressing Brazil's energy needs, the client invests in integrated solutions, emphasizing the transformation of the energy matrix towards a low-carbon economy.
CHALLENGES
- Lack of Physical Measurements | The client faced challenges with limited flowmeter measurements in all producing wells. Only some naturally flowing wells had the gas phase measured albeit with a notable degree of uncertainty. The scarcity of physical measurements posed a significant hurdle.
- Inconsistencies in Measurements | The well test data presented raw information containing invalid measurements, while the unreliability of results stemmed from the nature and age of the sensors. These inconsistencies added complexity to daily operations.
- Lack of Automation | Automation was lacking, requiring manual verification of well test data, manual processing of sensor data, and a manual analysis for the estimation of production rates. This manual approach hindered efficiency in daily operations.
OBJECTIVE
The primary objective was to transition from manual processes to automated flow rate predictions by leveraging Machine Learning (ML) algorithms and SCADA data.
SOLUTION
- Augmenting Physical Measurements | To address the limited physical measurements, a virtual flow meter utilizing both multivariate and univariate ML models was implemented in PetroVisor. This approach enabled the calculation of gas rates for all wells based on the available data.
- Reliable Measurements | Unsupervised ML techniques and rule-based logics were employed for data cleansing. Python code and Azure functions were incorporated for automating cleansing routines. A dedicated data integrity module was introduced to summarize signal quality and availability both for SCADA and well test data. This module efficiently highlighted issues, allowing for the prompt resolution of malfunctioning sensors and prioritizing well tests for wells that were lacking sensor data.
- Workflow Automation | The lack of automation was remedied through a standard workflow module included in PetroVisor. It involved the automated verification of well test results and SCADA data. Workflows were designed to automate the entire process of data ingestion and verification, with flow rates calculated on a daily frequency.
RESULT
The implementation of both Multivariate and Univariate ML successfully alleviated doubts about the solution's applicability to all wells, even the ones that were lacking sensor data. The initial Proof of Concept (POC) not only demonstrated the feasibility of building a Virtual Flow Meter (VFM) with complex ML but also significantly bolstered PetroVisor's credibility.
The data integrity tool swiftly identified and addressed data-related issues, contributing to higher accuracy in well measurements. The success of the project was reflected in the client’s decision to commit to a 3-year SaaS subscription.
In addition to the successful implementation, the client sought further empowerment. They requested Power User Training for one manager and three engineers, enabling them to use PetroVisor and independently build their own dashboards. At the time of writing a total of more than sixty client users are registered in PetroVisor, with a high collective weekly usage.
RESOURCES
The client commented publicly on the success of the project in the media and on LinkedIn.
