Time-series data historians have been a staple of industrial computing for almost 30 years. Yet today, this flavor of database technology has become more important than ever and is considered to be a linchpin of the industrial internet-of-things (IIOT). The predictive analytics market has seized upon this trend with the idea that intelligent computing programs can drive cash-generating insights out of these databases with little or no human supervision. Seeq Corporation is taking a different approach to this concept by enabling humans to do what they already know how to do—only much faster. So far, more than 100 companies have bought into the 5-year-old firm’s software technology, including a growing number of the oil and gas industry’s most familiar names.
This paper presents a methodology for quantifying uncertainty in production forecasts using Logistic Growth Analysis (LGA) and time series modeling. The applicability of the proposed method is tested by history matching production data and providing uncertainty bounds for forecasts from eight Barnett Shale counties.
In the methodology presented, the trend in the production data was determined using two different non-linear regression schemes. Predicted trends were subtracted from the actual production data to generate two sets of stationary residual time series. Time series analysis techniques (Auto Regressive Moving Average models) were thereafter used to model and forecast residuals. These residual forecasts were incorporated with trend forecasts to generate our final 80% CI.
To check reliability of the proposed method, we tested it on 100 gas wells with at least 100 months of available production history. The CIs generated covered true production 84% and 92% of the time when 40 and 60 months of production data were used for history matching respectively. An auto-regressive model of lag 1 was found to best fit residual time series in each case.
The proposed methodology is an efficient way to generate production forecasts and to reliably estimate the uncertainty. The method is computationally inexpensive and easy to implement. The utility of the procedure presented is not limited to gas wells and can be applied to any type of well or group of related wells.
Land subsidence is a critical issue to be addressed for large cities located near the sea. The monitoring of land subsidence is vital for predicting and managing the disasters that might occur. Many methods have been established to conduct this work, such as using geotechnical monitoring instruments and applying artificial satellite technologies. Those methods can provide highly accurate measurements for small areas. However, it would be expensive and ineffective to apply them to extensive areas. Hence, a monitoring method, that is economical to conduct, can be applied quickly and continuously, and can provide accurate measurements over large areas, is needed. Multi-temporal Differential Interferometry Synthetic Aperture Radar (MT-DInSAR), such as the Small Baseline Subset (SBAS), is a powerful technique for meeting the above demands. And, since the lifespans of current SAR satellites are commonly designed to be around 5-7 years, continuous monitoring for longer periods by the MT-DInSAR technique is important. To deal with these types of issues, a new method is required that can utilize the data from multiple (different) satellites.
In this study, a method for long-term land subsidence monitoring by MT-DInSAR, using multi-sensor data sets, is presented. Firstly, the SBAS method is performed for each time series SAR data set. Secondly, the hyperbolic fitting method is applied to estimate real values from the results of each data set. Finally, the hyperbolic curve is used to connect the results of the unlinked time series data sets. To verify this method, the land subsidence in Semarang City, Indonesia is taken as an example case.
Interferometry Synthetic Aperture Radar
DInSAR is an invaluable tool for observing land surface deformation over vast areas with the high accuracy of centimeter and high-spatial resolution of 3-30 m after spatial averaging and geocoding. Moreover, DInSAR does not require the installation of any devices on the ground, and it has been widely used for detecting horizontal and vertical displacements of the land surface (Hanssen, 2002).
Electrical Submersible Pump (ESP) operation faces new challenges with the advent of unconventional completions. Quick production decline means that ESP operators need proactive methods to deploy equipment for applicable flowrate ranges. The benefit for production forecasting and optimization is not only maximized accumulated oil production, but also improved ESP run life. This paper demonstrates the production forecasting capability of a time series data analysis method called Singular Spectrum Analysis (SSA).
Applying SSA to customer-provided raw, daily production data results in production data historical matching and future production forecasting. The strength of SSA stems from the ability to make a decomposition of the original series into a summation of the principal independent and interpretable components such as slowly varying trends, cycling components and random noise. [
Research proves that SSA can be utilized to forecast daily production rates based on a raw production dataset without any preprocessing or transformation of the original series. The trending component revealed by SSA for production prediction matches the forecasting capability of traditional reservoir production decline curve analysis (DCA), and is a considerable time-saving method. Unlike DCA, SSA is a nonparametric, modeless time series analysis method so no assumption for a certain model is needed to be setup before analysis.
Not only can SSA be used for production forecasting, but it can also be used for ESP operational optimization. Secondary or tertiary decomposed components from SSA can shed light on possible ESP operational issues or wellbore issues that could change the course of the typical production decline behavior for a well. Several case studies are included in this paper to demonstrate the capability of SSA in areas of ESP early faulty detection, evaluating the impact of ESP operation parameters on the reservoir and detection of cycling pattern on production that could lead to further investigation.
A stochastic sea-state model expressed with the Fourier series expansion is proposed. Model parameters are identified from hindcasting sea-state time series data of a few years period. Sea-state time series simulated with the identified model have the same statistical characteristics as original hindcasting sea-state data. Five components of sea-state are dealt, which are significant wave height, mean wave period, primary wave direction, mean wind speed, prevailing wind direction. Sea-state time series is separated into element sinusoidal signals which are individually expressed by amplitude, phase and frequency. Adding a random phase on an original phase enables to simulate a new sea-state time series having the original statistical correlation of variables. As a result of calculation, it is clarified that statistical characteristics of simulated sea-state time series with the proposed model agree with those of the original hindcasting data with respect to probability density function, mean value, variance value, cross-correlation function and persistence duration of sea-state. This technique would be applied to the service performance simulation to evaluate performance of ship and ocean platform in actual sea.
Sea-state time series data is useful for the analysis of the service performance of oceangoing ships. A ship captain usually changes an operation mode according to sea conditions. In a calm sea condition the main engine is usually worked in a mode of the revolution speed constant, but in a rough sea condition would be switched to a mode of the torque constant to prevent the engine overload. In a high sea condition a ship captain would order to decrease the ship speed to reduce violent ship motions, which is called as the voluntary speed loss. The voyage period and fuel efficiency must be evaluated from the history of these operations, based on the time series analysis with a short-term time step at least. This time series analysis is named as the service performance simulation. The service performance of ships is represented by voyage period(reliability of arrival time), fuel efficiency(economy), statistics of ship motions(comfortability, safety) and so on. Sea-state time series data is indispensable for this simulation. This analysis is also applicable for the other problems not only on the service performance analysis of oceangoing ships.
Hayatdavoudi, A. (University of Louisiana at Lafayette) | Boamah, M. A (University of Louisiana at Lafayette) | Tavnaei, A. (University of Louisiana at Lafayette) | Sawant, K. G. (University of Louisiana at Lafayette) | Boukadi, F. (University of Louisiana at Lafayette)
It has been almost 70 years since the first hydraulic fracturing job was carried out. Since then, hydraulic fracturing has made it possible to produce oil and clean burning natural gas from shale where conventional technologies are ineffective. However, in fracking the shale, the mechanism of increased gas production after the
To shed light on this issue we have studied and experimented with Pierre shale in detail. In our experimental work we suspended approximately 15 grams of shale cubic samples in deionized water. We have measured the changes in pH, Eh, (Redox potential), and Temperature and simultaneously have recorded the process of gas bubble flow under microscope, using a video system. We plotted our 1024 data points taken every three seconds. We used the Fourier Transform of the data to construct the Power Spectrum for extracting the hidden information in the data in relation to the release of the
The results of time series analysis in frequency domain reveals the following information: (1) depending on the type of shale, it takes a certain amount of time for water molecules to saturate/activate the shale capillaries. This is analogous to the
The practical application of our paper is (1) we propose a simple and cost effective methodology for determining the post frac optimal shut-in time and (2) once this optimal shut-in time is implemented the industry may benefit from realizing higher gas production from their prospects.
Production forecasting in shale reservoirs is a challenging task because of the complex influences of geology, lithology, stimulation practices, etc. The large well count makes history matching and forward simulation particularly time consuming and laborious. In such a context, it is important to consider alternative methods, and to this end, we have developed two new methods of forecasting production.
The first method uses data mining techniques, which allow the analysis of large quantities of data to discover meaningful pattern and relationships. These can subsequently be used for prediction. Some common data mining tools are neural networks (NN), genetic algorithms (GA), and self-organizing maps (SOM). Our method uses NN for predicting the future performance of a shale gas well based on historical production data of the previous year. The decline in production is captured during the NN training process and applied to the production data during the forecasting phase. The model is simple, elegant and fast and is able to forecast production in an unconventional play with reasonable tolerance.
The second method uses time series analysis. It the trend, changes in value, rate of decline, and correlation with the past to generate a rapid and accurate forecast. The stock markets use this technique, and it is safe to say that if it can predict the stock ticks, then it can yield good results on a fluctuating, but surely declining, production rate.
These methods are elegant and fast and are able to forecast production in an unconventional play with reasonable tolerance. They are not data intensive and can also be automated to be applied to a large number of wells, which makes them particularly useful in integrated operations in which a comparison of actual versus predicted behavior would enable operators to quickly identify problem wells for a more detailed investigation. The methods were applied to wells from the Barnett, Bakken, and Eagle Ford plays.
Data mining techniques allow analyzing large quantities of data to discover meaningful patterns and relationships. These can subsequently be used for prediction. Some common data mining tools are neural networks (NN), genetic algorithms (GA), and self-organizing maps (SOM). The workflow in this paper uses NN for predicting the future performance of a shale gas well based on historical production data. Specifically, the workflow uses NN to predict the production of a well based on its performance in the previous year. The decline in production is captured during the training process and applied to the production data during the forecasting phase. The NN trains itself on the production trend of all the wells in the dataset. It develops the ability predict the decline in the next time period based on the decline in the current time period. The next time period is then used to predict the period after that and so on. The model is simple yet elegant and is able to forecast production in an unconventional play with reasonable tolerance. This paper discusses the process of designing the NN, validating it, and using an advanced technique of clustering with SOM in cases where the tolerance is out of bounds.
Production from a shale gas well carries an underlying signature of the reservoir and the completion characteristic and so this workflow uses production data to predict itself out in the future.
Decline curve analysis (DCA), material balance (MB) and numerical simulation are used to predict reservoir performance. However, these methods have peculiar strengths and limitations. While some are based on analytical observations and theories others are purely stochastic. The objective of this study was to investigate the applicability of Time Series analysis to predict oil cumulative production as a measure of reservoir performance. The Box Jenkins method of Time Series analysis utilizing the Autoregressive Integrated Moving Average (ARIMA) models was used to predict and analyse oil cumulative production of wells and reservoirs. The process involves a robust transformation of production datasets into time series. A best fitting model was chosen using maximum absolute deviations as a measure of fit. Predictions of the ARIMA model were validated with historical data and its performance was compared with Decline Curve analysis (DCA).The coefficient of each model was optimized using the simplex optimization technique.
The results show that ARIMA (1,1,0) model gives the best match with actual cumulative oil production data. The model had the least maximum deviation of 4.24% compared to 178.95%, 182.32% and 24.45% for ARIMA (1,0,0), ARIMA (2,0,0) AND ARIMA (1,2,0) respectively after predicting for 900 days. Analysis of results after 1097 days of production also shows that the ARIMA (1,1,0) predicts better than the DCA method. The maximum deviation for both ARIMA and DCA were 1.81% and 12.79% respectively. The best model coefficients fall between 1.99 and 1.9999999. A model coefficient too high or too low will definitely lead to erroneous predictions. This study has shown that ARIMA (1,1,0) can be used to model and predict reservoir/well cumulative oil production accurately for short to mid-long term periods. Furthermore, the model performs better than the DCA for medium to mid-long term predictions. The accuracy of the model can be improved as more data become available for history matching to further refine the model coefficients. A model coefficient of 1.995 is recommended when few data are available.
Rebeschini, J. (Halliburton) | Querales, M. (Halliburton) | Carvajal, G. A. (Halliburton) | Villamizar, M. (Halliburton) | Md Adnan, F. (Halliburton) | Rodriguez, J. (Halliburton) | Knabe, S. (Halliburton) | Rivas, F. (Halliburton) | Saputelli, L. (Frontender Corporation) | Al-Jasmi, A. (Kuwait Oil Company) | Nasr, H. (Kuwait Oil Company) | Goel, H. K. (Kuwait Oil Company)
Implementing asset-wide intelligent digital oilfield (iDOF) solutions, aiming to optimize oil and gas production system in an "intelligent?? manner, requires integrating concepts from different disciplines, such as artificial intelligence. Neural network- (NN-) based models are a form of artificial intelligence, which is a branch of computer science that generates mathematical models that can be "trained?? to determine relationships between inputs and outputs, recognize patterns, and perform reliable short-term predictions. NN models using real-time data are proven tools for short-term well production forecasting with acceptable accuracy.
An oilfield is a hostile environment for even the most robust instrumentation. As a result, technical outages or anomalies can result in lost or poor quality data. Experience shows that many samples in a real-time database are frozen, missing, corrupted, or incorrect. This fact represents the biggest challenge to creating a reliable NN model. However, the models can be trained to correct or estimate missing real-time data. This paper presents a case study where nodal analysis was used to populate missing data used to train NN model, thus improving the reliability of the model. Because nodal analysis is not suitable for prediction, time-series analysis was used to assess the impact of historical events, and operational conditions were used to forecast trends. The NN trained with nodal analysis can cover a wide variability spectrum and, when trained with a time-lapse series, can predict short-term (30-day) production scenarios by changing highly correlated parameters, such as tubing head pressure (THP) or frequency (Freq).
This paper describes training NNs using nodal analysis and time-series analysis to predict short-term water cut (WC or BS&W) and liquid flow rate. This technique was applied in over 20 wells with electronic submersible pumps (ESPs) and gas lifts (GLs). The NNs made robust estimates of production rate and an acceptable prediction trend for 30 days, even when confronted with flow meter instrumentation failure, lost signals, and out-of-calibration instruments. Hence, the NN served as a "virtual meter,?? providing instantaneous and accurate estimation of production data.