Arailym-tleubayeva/NK-Oil-Well-Sensor-Monitoring

Oil Well Sensor Monitoring Dataset - NK Field

Dataset Description

This dataset contains hourly sensor readings from 10 oil wells at the NK field.

The monitoring period covers approximately seven months, from January 1, 2026, to July 20, 2026.

The data were provided by Galaz and Company LLP (ТОО «Галаз и Компания») within the research project:

“Development and Implementation of Control Algorithms for Low-Production-Rate Wells in Mechanized Oil Production Systems, Including Plunger Lift and Sucker-Rod Pump Systems, Using Artificial Intelligence Methods.”

The dataset was prepared for research on artificial intelligence methods for monitoring, forecasting, anomaly detection, predictive maintenance, and control of low-production-rate oil wells.

Dataset Summary

• Number of wells: 10
• Monitoring period: January 1, 2026 - July 20, 2026
• Sampling frequency: hourly
• Total number of rows: 186,251
• Fountain wells: 3
• Sucker-rod pump wells: 7
• Data format: long tabular format
• Train and test split: chronological

Wells Included

Well ID	Well type	Recorded parameters
NK-75	fountain	tubing pressure, casing pressure, line pressure
NK-76	fountain	tubing pressure, line pressure
NK-83	fountain	tubing pressure, casing pressure, line pressure
NK-7	sucker-rod pump	SPM, pump fillage, minimum rod weight, maximum rod weight, dynamometer area
NK-68	sucker-rod pump	SPM, pump fillage, minimum rod weight, maximum rod weight, dynamometer area
NK-77	sucker-rod pump	SPM, pump fillage, minimum rod weight, maximum rod weight, dynamometer area
NK-82	sucker-rod pump	SPM, pump fillage, minimum rod weight, maximum rod weight, dynamometer area
NK-86	sucker-rod pump	SPM, pump fillage, minimum rod weight, maximum rod weight, dynamometer area
NK-91	sucker-rod pump	SPM, pump fillage, minimum rod weight, maximum rod weight, dynamometer area
NK-93	sucker-rod pump	SPM, pump fillage, minimum rod weight, maximum rod weight, dynamometer area

NK-76 does not contain casing-pressure measurements because this parameter was not recorded in the original source files.

Dataset Structure

The dataset is provided in long, tidy format.

Each row represents one parameter observation for one well at one timestamp.

Field	Type	Description
well_id	string	Unique well identifier
well_type	string	Well operating type
timestamp	datetime	Hourly observation timestamp
parameter	string	Name of the measured parameter
value	float	Recorded sensor value
unit	string	Unit of measurement

Parameters

Parameter	Unit	Applicable well type
tubing_pressure	atm	fountain
casing_pressure	atm	fountain
line_pressure	atm	fountain
SPM	strokes/min	sucker-rod pump
pump_fillage	%	sucker-rod pump
min_rod_weight	kg	sucker-rod pump
max_rod_weight	kg	sucker-rod pump
dynamometer_area	unitless	sucker-rod pump

Dataset Files

The repository contains the following files:

File	Description
wells_dataset.csv	Complete dataset without splitting
train.csv	Chronological training split
test.csv	Chronological testing split

Data Splits

The dataset was divided chronologically without random shuffling.

This split strategy prevents future observations from leaking into the training data and reflects a realistic forecasting scenario.

Split	Rows	Date range	Approximate share
Train	148,308	2026-01-01 00:00 to 2026-06-09 23:00	80%
Test	37,943	2026-06-10 00:00 to 2026-07-20 00:00	20%

Research Context

The dataset was prepared within the research project:

“Development and Implementation of Control Algorithms for Low-Production-Rate Wells in Mechanized Oil Production Systems, Including Plunger Lift and Sucker-Rod Pump Systems, Using Artificial Intelligence Methods.”

The research focuses on the development of intelligent methods for monitoring and controlling low-production-rate wells.

The main research directions include:

• monitoring of low-production-rate wells;
• analysis of fountain wells;
• analysis of sucker-rod pump wells;
• analysis of mechanized oil production systems;
• detection of abnormal operating conditions;
• prediction of pressure and equipment parameters;
• predictive maintenance;
• artificial intelligence-based well control;
• decision support for oil production systems;
• optimization of low-production-rate well operation.

Data Provider

Organization: Galaz and Company LLP

Russian name: ТОО «Галаз и Компания»

Industry: Oil and gas

Data type: Industrial well-monitoring, SCADA, telemetry, and dynamometer data

Research purpose: Development and evaluation of artificial intelligence algorithms for monitoring and controlling low-production-rate wells

Data Processing

The original data were provided as separate Microsoft Excel files for individual wells and monitoring periods.

The following preprocessing steps were performed:

• consolidation of multiple source files;
• standardization of well identifiers;
• standardization of parameter names;
• conversion into a unified long-format table;
• chronological ordering of observations;
• removal of overlapping duplicate timestamps;
• assignment of well types;
• creation of chronological training and testing subsets.

No automatic missing-value imputation was applied.

No systematic outlier removal or correction was applied.

Potential Tasks

The dataset can be used for:

• time-series forecasting;
• multivariate sensor forecasting;
• pressure trend prediction;
• anomaly detection;
• outlier detection;
• missing-data analysis;
• well shut-in detection;
• operating-mode classification;
• equipment condition monitoring;
• sucker-rod pump monitoring;
• predictive maintenance;
• low-production-rate well optimization;
• industrial artificial intelligence research;
• benchmarking of machine learning models;
• decision-support system development.

Data Quality Notes

Users should consider the following data-quality limitations:

• The timestamps in the original source files did not contain an explicit year.
• The year 2026 was assigned based on the sequential order of monthly source files.
• The assigned year should be verified against the original SCADA or telemetry system before time-sensitive operational use.
• Missing readings are represented by absent rows rather than explicit null values.
• NK-76 has no casing-pressure measurements.
• NK-86 has a multi-day gap in source records from April 29 to May 12, 2026.
• NK-91 has a multi-day gap in source records from January 29 to February 4, 2026.
• The dynamometer_area parameter for NK-7 contains extreme values, including negative readings.
• Some extreme values may represent sensor, telemetry, or export artifacts.
• Overlapping timestamps at monthly file boundaries were deduplicated by keeping the first occurrence.
• The dataset has not undergone automatic outlier correction.
• The dataset has not undergone automatic missing-value imputation.

Intended Use

The dataset is intended for scientific, educational, and experimental research in:

• artificial intelligence;
• machine learning;
• time-series analysis;
• oil and gas engineering;
• industrial monitoring;
• predictive maintenance;
• mechanized oil production;
• low-production-rate well control;
• sucker-rod pump analysis;
• intelligent decision-support systems.

The dataset should not be used as the sole source for real-time production, operational, or safety-critical decisions.

All analytical results, predictions, and control recommendations should be verified by qualified oil and gas specialists and compared with the original company monitoring systems.

Limitations

The dataset does not contain all variables that may influence oil-well performance.

For example, it may not include:

• production rate;
• fluid composition;
• water cut;
• bottom-hole pressure;
• reservoir pressure;
• equipment maintenance history;
• intervention history;
• weather conditions;
• operator actions;
• detailed geological information.

The absence of these variables may limit the interpretation of anomalies and model predictions.

The dataset also does not contain verified labels for all equipment failures, well shutdowns, or abnormal operating modes.

Ownership and Attribution

The original industrial data were provided by Galaz and Company LLP (ТОО «Галаз и Компания») within the stated research project.

Users should acknowledge the data provider and the research context in publications, reports, presentations, and derivative research.

Publication of the dataset does not remove the obligation to comply with applicable confidentiality, data-ownership, industrial-data protection, and contractual requirements.

Example Usage

Load the Complete Dataset

import pandas as pd

df = pd.read_csv(
    "wells_dataset.csv",
    parse_dates=["timestamp"]
)

print(df.head())
print(df.shape)
print(df.columns)
print(df["well_id"].unique())

Load Train and Test Splits

import pandas as pd

train = pd.read_csv(
    "train.csv",
    parse_dates=["timestamp"]
)

test = pd.read_csv(
    "test.csv",
    parse_dates=["timestamp"]
)

print("Train period:")
print(train["timestamp"].min())
print(train["timestamp"].max())

print("Test period:")
print(test["timestamp"].min())
print(test["timestamp"].max())

Select Fountain Wells

fountain_wells = df[
    df["well_type"] == "fountain"
]

print(fountain_wells.head())
print(fountain_wells["well_id"].unique())

Select Sucker-Rod Pump Wells

pump_wells = df[
    df["well_type"] == "sucker_rod_pump"
]

print(pump_wells.head())
print(pump_wells["well_id"].unique())

Convert One Well to Wide Format

nk75 = (
    df[df["well_id"] == "NK-75"]
    .pivot(
        index="timestamp",
        columns="parameter",
        values="value"
    )
    .sort_index()
)

print(nk75.head())

Compare Tubing Pressure Across Fountain Wells

fountain_pressure = df[
    (df["well_type"] == "fountain")
    & (df["parameter"] == "tubing_pressure")
]

pressure_table = fountain_pressure.pivot(
    index="timestamp",
    columns="well_id",
    values="value"
)

pressure_table.plot(
    title="Tubing Pressure of Fountain Wells"
)

Analyze Missing Hourly Observations

well_data = df[
    df["well_id"] == "NK-75"
].copy()

hourly_grid = pd.date_range(
    start=well_data["timestamp"].min(),
    end=well_data["timestamp"].max(),
    freq="h"
)

wide_data = well_data.pivot(
    index="timestamp",
    columns="parameter",
    values="value"
)

wide_data = wide_data.reindex(hourly_grid)

print(wide_data.isna().sum())

License

The dataset is released under the Apache License 2.0.

Users must also comply with applicable data-ownership, confidentiality, industrial-data protection, and contractual requirements associated with the original data.

Citation

Please cite the dataset as follows:

@dataset{tleubayeva_2026_nk_well_monitoring,
  author       = {Arailym Tleubayeva},
  title        = {Oil Well Sensor Monitoring Dataset: NK Field},
  year         = {2026},
  publisher    = {Hugging Face},
  organization = {Galaz and Company LLP},
  note         = {Dataset prepared within a research project on artificial intelligence-based control algorithms for low-production-rate wells},
  url          = {https://huggingface.co/datasets/Arailym-tleubayeva/NK-Oil-Well-Sensor-Monitoring}
}

Acknowledgements

The dataset author acknowledges Galaz and Company LLP (ТОО «Галаз и Компания») for providing the industrial well-monitoring data used within the research project:

“Development and Implementation of Control Algorithms for Low-Production-Rate Wells in Mechanized Oil Production Systems, Including Plunger Lift and Sucker-Rod Pump Systems, Using Artificial Intelligence Methods.”

Contact

For questions, corrections, or research collaboration, please use the Community section of the Hugging Face dataset repository.