Method and system for predicting well production rate

Abstract

A computer-implemented method for predicting the well production rate based on a machine learning model, the method comprising the steps of: obtaining a predicted well production rate of an existing well by using the machine learning model that is trained by the well group which includes the existing well, or obtaining a predicted well production rate of a future well by using the machine learning model trained based on the wells of the same reservoir.

Description

METHOD AND SYSTEM FOR PREDICTING WELL PRODUCTION RATEFIELD OF THE INVENTIONThe present invention relates to a method and system for predicting well production rate.BACKGROUND OF THE INVENTIONDuring the phase for exploring existing well fields or developing new wells in an existing oil field, it is crucial to have a well production rate forecast. For production forecast of existing active wells and new wells that have not been drilled yet, an accurate prediction of oil, gas or water production rates for individual wells help, for example, the oil, water and gas companies to optimize their production operations and make more informed decisions about reservoir management.In order to predict the well production rate for existing or future wells, companies typically use the simulation models which requires a 3D geological and simulation model which is a time-consuming task that could take years. These typical simulation models require input parameters such as the reservoir properties, well parameters, and operating conditions.In recent years, there has been a growing interest in data-driven approaches for well forecasting, which leverage large amounts of data to train machine learning algorithms to predict oil, water and gas production rates. The most common data driven method is decline curve analysis (DCA), which involves fitting declining a hyperbolic curve to historical production data to estimate future production rates. This method applies to wells in situations where the reservoirs are mature, and production is naturally declining with minimum intervention from the operators to boost the production. Hence, most of the current methods for well production predictions using either DCA and machine learning (ML) focus on mature or declining reservoirs.European patent application EP 3414428 A2 describes a method for estimating oil, water and gas production from individual wells using statistical learning models. It combines DCA with a machine learning model, utilizing various data sources such ascompletion, geologic, engineering, seismic, geochemical, and petrophysical data to predict production behavior.However, DCA has several limitations and drawbacks, including and not limited to its inability to capture complex reservoir behaviors such as changes in well performance due to changes in operating conditions, well interference, or changes in the reservoir properties. Additionally, DCA is not effective in predicting production rates for future wells without historical data.This highlights the need for new approaches to well forecasting that can incorporate a wider range of data and account for more complex reservoir and changing well conditions. There is also need for effective approaches for future well forecasting.SUMMARY OF THE INVENTIONThe present invention describes a method and system for predicting well production rates.A first aspect of the invention is a computer-implemented method for training a machine learning model for predicting well production rate, comprising the steps of: receiving input data representing an aspect of a well; generating three-dimensional well data representing properties of the well at least partially based on the data representing the aspect of the well; converting the three-dimensional well data into two-dimensional well data; and training the machine learning model through inputting at least the two- dimensional well data and historical production data of the well into the machine learning model.The first aspect of this invention enables a more effective and optimal training process for the machine learning model for predicting future well production rate.According to a second aspect of this invention, according to the first aspect, the three- dimensional well data are converted into the two-dimensional well data through aggregation.The second aspect of this invention ensures that the well data for training the machine learning model for future well production rate undergoes effective and lossless compression.According to a third aspect of this invention, according to any one of the preceding aspects, the step of converting the three-dimensional well data into two-dimensional well data comprises the step of: dividing the three-dimensional well data into multiple sets of three-dimensional well data, and converting the multiple sets of three-dimensional well data into the multiple sets of two-dimensional data for training the machine learning model; preferably, converting each set of the multiple sets of three-dimensional well data into a corresponding two-dimensional well data by aggregating each set of the multiple sets of three-dimensional well data.According to a fourth aspect of this invention, according to any one of the preceding aspects, the three-dimensional well data is divided into multiple sets of three- dimensional well data based on a condition according to an aspect of the well.The arrangement of the third and fourth aspects of this invention ensures that representative data is preserved when processing data.According to a fifth aspect of this invention, according to any one of the preceding aspects, the training method further comprises the steps of: inputting an opportunity index, sweep efficiency, and / or the well connection data of the well into the machine learning model to train the machine learning model, wherein the well connection data represents the trajectory of the wellbore of the well in three-dimensional space.The arrangement of fifth aspect of this invention ensures that the machine learning model of this invention can provide a better prediction for the future production rates.According to a sixth aspect of this invention, according to any one of the preceding aspects, the machine learning model training steps comprises the step of:clustering multiple wells into a first number of well groups by using an Al clustering model based on the similarity of an aspect of the multiple wells, and training a first set of different machine learning models for each well group through inputting at least the two-dimensional well data and historical production data of the wells of the corresponding well group into the first set of machine learning models.Instead of training a machine learning model for every individual well, the arrangement of the fifth aspect of this invention ensures that the training process is more efficient, as fewer machine learning models need to be trained.According to a seventh aspect of this invention, according to any one of the preceding aspects, the Al clustering model is an unsupervised machine learning algorithm, preferably a K-means algorithm.The clustering algorithm of the seventh aspect of this invention facilitates the subsequent processing by expediting and simplifying the process.According to an eighth aspect of this invention, according the sixth and seventh aspects, the well groups are clustered by the Al clustering model based on a similarity of the historical production data of the wells.The eighth aspect of this invention allows the different machine models for every well group to have a better understanding of the overall production trends in the reservoir.According to a ninth aspect of this invention, according the sixth to eighth aspects, the machine learning model training step further comprises the step of: evaluating the first set of machine learning models at least based on their Root Mean Squared Error, RMSE, or a combination of their cumulative error and RMSE across N forecast points in a certain period of time.The ninth aspect of this invention ensures that the trained machine learning model is assessed favourably and thoroughly.According to a tenth aspect of this invention, according the sixth to ninth aspects, the machine learning model training step further comprises the step of:clustering the multiple wells into a second number of well groups by using the Al clustering model based on the similarity of the aspect of the multiple wells, training a second set of different machine learning models for each well group, which have been clustered with the second number of well groups, evaluating the second set of different machine learning models at least based on their Root Mean Squared Error, RMSE, or a combination of their cumulative error and RMSE across N forecast points in a certain period of time, comparing results of the evaluations of the first and second set of machine learning models and selecting the set of machine learning models that produces a lower RMSE, or a lower RMSE and cumulative error across the certain period of time.According to an eleventh aspect of this invention, according to the preceding aspect, the machine learning model training step further comprises the steps of: repeating the clustering, training, and evaluating with different numbers of well groups in a multi-iteration manner, until the RMSE or the combination of cumulative error and RMSE of the machine learning models is minimized.The tenth and eleventh aspects of this invention ensure that an optimized clustering number of the well groups is picked for the machine learning model and the prediction of the future production rate.According to a twelfth aspect of this invention, according to any one of the preceding aspects, the historical production data includes monthly production rates of oil, water and gas, choke size, production days, and / or well head flowing pressure; the data representing an aspect of a well comprises the coordinates and the trajectory of the well, the connection status, Kh, permeability, porosity, saturation, pressure, and / or well type; and / or the two-dimensional and three-dimensional data are generated at least based on the porosity, permeability, saturation, reservoir pressure, and / or coordinates and / or trajectory of the well.The input data of the twelfth aspect of this invention ensures that the machine learning model is well trained for the future production rate.According to a thirteenth aspect of this invention, according to the preceding aspect, the machine learning model is a multi-output regression model or time series forecasting ML model.The thirteenth aspect of this invention ensures that optimal Al models are chosen for predicting future production rate for the existing wells or future wells for which no history data is available.According to a fourteenth aspect of this invention, according to any one of the preceding aspects, the method further comprises: generating three-dimensional well data by using a simulation program with the aspect of the well.According to a fifteenth aspect of the invention, the invention provides a computer- implemented method for predicting the well production rate based on a machine learning model trained by the method of any one of first to fourteenth aspects, the method comprising the steps of: obtaining a predicted well production rate of an existing well by using the machine learning model that is trained by the well group which includes the existing well, or obtaining a predicted well production rate of a future well by using the machine learning model trained based on the historical wells of the same reservoir.According to a sixteenth aspect of this invention, according to the preceding aspect, the step of predicting the well production rate includes a production uncertainty envelope.By quantifying the uncertainty in these predictions, the sixteenth aspect of this invention provides a probabilistic estimate of the possible range of oil, gas or water production for each well, taking into account the variability and uncertainty in the data and models used for forecasting. Hence, the users are furnished with a comprehensive and holistic perspective on the possibility of future well production rate.According to a seventeenth aspect of this invention, according to the fifteenth or sixteenth aspects, the specified percentiles of the production uncertainty envelope correspond to at least io percentile, 50 percentile, and / or 90 percentile of the predicted well production rate.An eighteenth aspect of the invention is a system for predicting a well production rate, comprising an information processing system configured to implement the method according to any one of first to seventeenth aspects of this invention.BRIEF DESCRIPTION OF THE FIGURESFIG. i shows a diagram of a simulated 3D well for predicting the future production rate according to one embodiment of this invention;FIG. 2a shows an illustration of a well with multiple reservoir zones correlation with an aspect of the well in these zones according to one embodiment of this invention;FIG. 2b shows an example three-dimensional (3D) model of reservoir zones according to one embodiment of this invention;FIG. 3a and 3b show examples of two-dimensional (2D) models of a horizontal well and vertical well according to embodiments of this invention;FIG. 4 is a flow chart illustrating an example method for pre-processing of the well data for predicting the future production rate according to embodiments of this invention;FIG. 5 is a flow chart illustrating an example method for integrating the well data for predicting the future production rate according to an embodiment of the invention;FIG. 6 is a diagram illustrating an example method for calculating uncertainty intervals of the predicted future well production rate using quantiles according to an example of the method of this invention;FIG. 7 is a flow chart illustrating an example method for predicting the future production rate of an existing well according to embodiments of this invention;FIG. 8 is a flow chart illustrating an example method for predicting the future production rate of a future well according to embodiments of this invention.DETAILED DESCRIPTION OF THE INVENTIONThe invention will now be described based on the figures. It will be understood that the embodiments and aspects of the invention described herein are only examples and do not limit the protective scope of the claims in any way. The invention is defined by the claims and their equivalents. It will be understood that features of one aspect orembodiment of the invention can be combined with a feature of a different aspect or aspects and / or embodiments of the invention.The method and system for predicting well production rates of this invention (hereinafter the method and system) aim to provide an accurate prediction of oil, gas and water production rates for individual wells, which can help oil, water and gas companies optimize their production operations and make more informed decisions about well management. Specifically, the method and system of this invention help to increase the overall productivity, profitability and sustainability of oil, water and gas fields. Additionally, the ability to predict future production rates can aid in budget planning and risk management of oil, water and gas companies.The method and system of this invention is a data-driven solution for predicting oil, gas and water production rates, namely the volume of produced fluid per unit of time, in subsurface reservoirs.The method and system of this invention can be used with any existing well system. For example, a drilling system with a bottomhole assembly. For data measurement and transmission, a system with logging-while-drilling (LWD) tools and / or measuringwhile-drilling tools (MWD) can be used with the method and system of predicting well production rates.The method and system of this Invention can also be used for any future wells, namely, wells with no history data available or wells still to be drilled. The same well drilling and measuring systems for the existing wells can also be used for future wells.[DATA EXTRACTION]To create accurate and reliable predictions of the production rate, the system and method of this invention use a wide arrange of data sources, including both static (e.g., physical properties of the well, such as porosity and permeability) and dynamic properties, namely time-variable properties of the subsurface reservoir (such as saturation and production history). Any data that are relative to the well’s properties are considered as the aspects of the well. All data that relates to the well may be used as input data for further processing. The data extraction process is the same for both the systems and methods for predicting the production rate of the existing wells and future wells.In preferred embodiments, the production and injection history data are used as input data for the system, which helps for understanding how the wells have performed in the past. In this set of data, monthly production rates, choke size, production days, and well head flowing pressure are used. Herein, the production days is the time of active well production per month; for example, if a well production is 20 days in a month, then the producing days are 20.In the preferred embodiments, the input data used for the system and method can also include various parameters such as the coordinates and trajectory of the well, as well as information on the connection status, Kh, permeability, porosity, and the well type, namely horizontal or vertical wells. The information on the connection status includes the information about if the well is open or closed for production along the entire length of the well. Kh refers to the product of formation permeability, k, and producing formation thickness, h, in a producing well. The percentage of holes or empty spaces in a rock is called porosity. These spaces can be filled with petroleum if it exists. When the rock is buried deeper, it gets squeezed and loses most of its porosity. Only a small fraction of the original porosity remains. The ability of fluids to flow through the spaces between the grains of a rock is called permeability. The container has sand with small openings that connect the large spaces. The size of these openings is measured in millidarcy (md) or Darcy (1000 md). A rock that can store petroleum usually has a permeability between too and 500 mg. Fractures in the rock have no limit to their permeability.In the preferred embodiments, the system and method of this invention may additionally use the opportunity index, which contains a pressure sink 3D grip, and the sweep efficiency. These data can help to identify sweet spots within the subsurface reservoir that are likely to yield the highest oil, gas or water production rates. The opportunity index, for example, is a well-known term in oil, gas or water industry, as referred to in “Well Placement Optimization Under Uncertainty Using Opportunity Indexes Analysis And Probability Maps” by H.M. Mustaphai, D.D. Dias. The opportunity index is comprised of three components - the reservoir quality index, mobile oil volume index, and pressure index - and it provides valuable insights into the potential for oil production in a given area. For example, the sweep efficiency map is used to evaluate the efficiency of water injection. The pressure sink 3D grips are areas within the reservoir that suffers from lower reservoir pressure due to over production and lack of pressure support from injectors.[3D Properties Pre-processing]The method and system of this invention can be employed, for example, in a computer system for real-time simulation system. Examples of this simulation tool can be any current simulation system that are available on the market including, but not limited to, the simulation system from Intersect, CMG, Petrel, RMS etc. With the simulation tool, the system and method of this invention provide a powerful tool combining detailed information provided by the simulated 3D geological and petrophysical models with the well data, production history data, and Artificial Intelligence (Al) model, specifically, machine learning (ML) algorithms, which allows for a comprehensive understanding of the reservoir and can help to optimize oil, gas and water production. Specifically, the pressure sink is created as part of the reservoir management strategy under water injection drive, and it brings the benefit of effectively understanding the reservoir fluid characterization associated with production operation; effectively identifying areas of opportunity to increase production without damaging the reservoir; and, effectively identifying areas of opportunity to increase water injection.Figure 1 is a diagram of a simulated well. In other words, Figure 1 shows a 3D grid of a well, that is simulated by the above-mentioned systems. To have the 3D model simplified, and suited for further processing, e.g., Al training, the 3D model together with its associated well data are converted to 2D well data, specifically, multiple sets of 2D well data, i.e. the 2D grid model together with its associated well data. Herein, the well data may refer to any one of the above mentioned well data, e.g., permeability, porosity, and / or reservoir pressure, or a combination of the above mentioned well data.Figure 2a is a diagram illustrating multiple reservoir zones, and a correlated aspect of the well, herein, for example, is the porosity. Based on a predetermined condition associated with the well levels, seismic is used to interpolate in between wells. Accordingly, the 3D grid is divided into multiple zones, as shown in different shades in Figure 2b. For example, as shown in Figure 2a, the reservoir is divided into 5 zones, i.e., zones 1 to 5, based on the well levels and the seismic are used in between wells / reservoir to divide the well / reservoir into these 5 zones. One possible zone interpretation is described in European patent EP 3 698 018 Al. However, another interpretation, such as interpretation based on seismic wave processing, may also be utilized to determine the reservoir zones. Dividing the 3D model into multiple reservoir zones is common workflow. Zone interpretation is a common activity and well understood by the O&G industry.Once the reservoir zones are determined, the multiple sets of corresponding 3D well data of each reservoir zone, i.e., any petrophysical property like porosity, permeability, saturation over each zone are aggregated, preferably averaged by reservoir zones as shown in Figure 3a for a horizontal well and Figure 3b for a vertical well and are converted into multiple sets of 2D property data (i.e. aggregated well properties by zones). As shown in Figure 2a, the properties like porosity, permeability or saturation may be averaged over the 5 zones so as to generate 5 sets of 2D property data with the averaged properties. Other aggregation methods such as maximum, minimum etc. can also be used.The aggregated property data are then allocated onto a 2D plane so as to generate the 2D grid data / 2D well data according to the well connection data, namely the trajectory of the wellbore of the well in three-dimensional space, more specifically, the trajectory of the corresponding reservoir zones. As shown in Figure 3a, the aggregated properties, for example, averaged values of the porosity of each of the reservoir zones are allocated onto a 2D grid plane according to their relative position / trajectory in the 3D grid. Accordingly, the multiple sets of 2D well data for training the machine learning model is obtained.The optimal radius of the polygon of the 2D well grid is determined in a way so that that all relevant properties around the well are captured. Specifically, this is determined by the spatial distribution of 2D properties (namely the allocated processed data derived from the 3D grid), well connection data, and other relevant information. Preferably, the radius of the polygon of the 2D well grid is determined based on the well density within a reservoir. If too many wells are close to each other in a reservoir, the radius is smaller compared to a reservoir with less wells. This ensures that the resulting polygon accurately captures the spatial extent of the well and its associated properties, providing a solid foundation for further averaging and the analysis of 3D properties within the polygon.Figure 4 shows a flowchart of a process of the preprocessing of the well data. Similarly, this flow of the process may involve the utilization of a CFD 3D model that captures intricate characteristics of the physical and chemical properties in a designated area of the reservoir or well.The process starts with obtaining input data about the well and using the CFD model as the simulation program. The input data includes physical and chemical properties ofthe well, for example, well aspects such as reservoir properties and the trajectory of the wellbore of the well.In block 401, the received data is fed into the simulation program so as to generate 3D well data (i.e., a 3D well grid). In other words, the simulation is executed by applying the received input data to the simulation program. Then, the 3D well data is divided into multiple reservoir zones of 3D well data, i.e. multiple sets of 3D well data, according to the well levels and based on seismic in between wells.In block 402, the multiple sets of 3D well data is averaged by zone.In block 403, the aggregated properties are allocated onto a 2D plane in accordance with the well connection data so as to generate the 2D grid / 2D well data. In other words, the averaged properties are positioned on the 2D plane based on the trajectory of the wellbore of the well or the corresponding position of the corresponding reservoir zone in the 3D grid. The multiple sets of 2D well data is thus obtained and ready for further data-preprocessing before training the machine learning model. With this preprocessing of 3D well data into 2D well data, the quantity of well connection data, property data, and dimensions depicted in the original 3D property are significantly minimized.In this invention, the method and system for oil, water and gas production prediction can be used for both an existing well and a future well. However, considering that for future wells (wells that have not been drilled but it is planned that they are drilled in the future) no history data is available, the method for training and selecting the artificial intelligence model is different from the one for existing wells.Hereinafter, the method and system for training and predicting the oil, water or gas production rate for the existing well and the future wells are explained in detail.[Existing Well Forecast]In the preferred embodiments of the method and system of this invention, a machine learning model (ML) is employed, which is trained based on at least the multiple sets of 2D well data, the opportunity index and the production data. The trained ML model can then be utilized to predict the well production rate by inputting the historical production data. This enables the approximation power and speed of production rate prediction based on the ML with the simulation program to achieve both a quick and highly accurate prediction.Specifically, for each well in the reservoir, a comprehensive dataset is prepared to include exhaustive historical production data, and reservoir static (porosity and permeability) and dynamic (pressure and saturation) properties. Additionally, reservoir quality indices (Opportunity Index, Pressure Sink, etc.) are integrated to the dataset.Data such as the well identifications, the exogenous features (i.e., the properties of the well) together with their values over periods of time is integrated into an input dataset.More specifically, in the preferred embodiments, the data can be pre-processed as shown in Figure 5 for training the machine learning model. First, data that relates to the wells are collected. Herein, the properties of two wells that are collected. Then, the data is integrated by combining the data into a cohesive data set, for example, by organizing the corresponding or related data together. In addition, data that are not relevant to predict the well production rate are removed. Finally, the data is organized in a way such that the exogenous features, i.e., the properties relevant to the well production rate, the timestamp, and the well identifications are aligned with the corresponding well production rates.After the pre-processing of the data, the ML model is trained. In preferred embodiments, a regression ML model, specifically, a time series forecasting ML model, is trained and utilized for the prediction.To evaluate the performance of the trained ML models on well data, the available data are split into a training set and a testing set, where the last N months / years of each well’s data is used for testing and the rest is used for training. Different architectures of the regression ML model are trained and evaluated based on the following metrics (1) and (2):In order to train the model more efficiently, rather than training ML models for every single well, wells having similar production behavior, e.g. similar production rate are grouped together, and each group of wells are trained with different / unique ML data.To cluster the wells, in a preferred embodiment as shown in Figure 7, an unsupervised machine learning algorithm, specifically K-means clustering, may be applied to the historical oil, gas or water production data of each well. This algorithm groups or clusters the wells into different groups based on their production behaviour, which allows a better understanding of the overall production trends in the reservoir.To find the best ML model with the optimized clustering number, i.e. to find the balance between accurate prediction and the clustering scheme (the number of clustering groups for the wells), the ML models are trained with a range of clustering schemes, namely a range of different group numbers for the same wells. The number of ML models corresponds to the number of the clustering groups. The wells are clustered repeatedly into different numbers of groups or group numbers according to a range of the clustering schemes. The different / unique ML models, i.e., ML models with different architectures, are then trained for different clustering groups. The processes are iterated with different clustering schemes. More specifically, the wells are clustered into different numbers of groups, e.g., 3 groups, 4 groups or 5 groups, wherein each group may have different numbers of wells. Multiple sets of ML models are trained accordingly in a multi-iteration manner. In other words, the process is iterative; the number of clusters obtained through K-means clustering is varied until the Total Error minimized.For example, as shown in Table (1), ten wells (wells 1 to 10) are first divided into three groups, with the wells in each groups having similar production rates. Three ML models are then trained for each group. The three ML models are then tested in order to obtain the corresponding error. The error is then averaged: herein, the mean error by well is (o.25+o.i9+o.o4+o.o5+o.4+o.i7+o.o2-i-o.2-i-o.i6-i-o.4) / io = 0.188.Table (i)The above process is iterated again with a different clustering number, which is 4, as shown in Table (2). Accordingly, 4 ML models are trained and tested.The mean error by well is (0.13+0.4+0.09+0.11+0.07+0.13+0.25+0.3+0.1+0.08)710 = 0.166.The clustering scheme of having 4 groups of wells has a lower mean error by well than the clustering scheme of having 3 groups. Hence, in this case with two iterations N=3 and N=4 the optimal number of clusters is N=4 because it minimizes the mean error per well.Preferably it is iterated at least 3 times, from Nmin to Nmax, by running the cluster algorithm for different numbers of clusters, starting from the maximum number Nmax and going down to the minimum number Nmin. The number of iterations and the number of groups are determined based on the total amount of the wells.With the optimized number of clusters described above, the trained ML model can be utilized to generate predictions for each N-month interval of the well’s production, with the input of, for example, the 2D properties (specifically, the 2D well data), the Opportunity index, the well connection data and historical data of the well and future choke size and the well head flowing pressure, which is carried over the last value of the well.In addition, in more preferred embodiments, to quantify the uncertainty in the predictions, the ML model calculates uncertainty intervals using quantiles as shown in Figure 6. The model predicts along with the 50thpercentile (P50), the 90thpercentile(P90) and the 10thpercentile (Pio), the uncertainty interval represents the range within which 80% (P90 - Pio) of the possible outcomes fall, providing a measure of the uncertainty associated with the predictions (i.e. a production uncertainty envelope).By calculating uncertainty intervals using quantiles, the method and system of this invention provide a probabilistic estimate of the possible range of oil, gas or water production for each well, taking into account the variability and uncertainty in the data and models used for forecasting. This will facilitate users to make informed decisions based on the level of risk they are willing to accept.Figure 7 shows a flowchart of the method and system for predicting production rate for existing wells.In block 701, the input data representing the aspects of the wells (well properties such as the 2D properties [specifically, the 2D well data], the opportunity index by year, the well connection data, the sweep efficiency, and / or the histoiy production data by month) are collected. Preferably, the collected data is then pre-processed by integrating it into a more useful dataset as mentioned above.In block 702, the clustering Al model, specifically K-means clustering, is utilized to cluster the wells into an Nmax number of clustering groups based on their histoiy production data (by month) / the similarity of production behaviour.In block 703, an Nmax number of ML models with different architectures are trained with the collected input data.In block 704, the Nmax number of ML models with different architectures are tested with the test set of histoiy data, by which the ML models are evaluated. Specifically, the ML models are evaluated through using a combination of their cumulative error and RMSE across N forecast points in a certain period of time of the test set of history data, namely a period of time (e.g., 2 years) of overlap to check the prediction quality against the historical data.In block 705, blocks 702 to 704 are iterated until: 1) an Nmin number of ML models, which is the predetermined minimum number of clustering group, are trained and evaluated; or 2) the error is minimized, i.e., number of clustering is optimized. The ML models having the lowest error are chosen for prediction.In block 706, the chosen ML model is utilized for predicting the future production rate. Preferably, in order to quantify the uncertainty of these predictions, the future well production rate is predicted with the production uncertainty envelope.[Future Well Forecast]In the future well forecast, a formal approach is followed to create one model per reservoir. In other words, an individual ML model is trained for each reservoir, and the data of all the existing (active) wells in that reservoir are used for training, so that for a future well located in that reservoir the trained ML model can be used for the future well production rate prediction.For pre-processing of the future well forecast, the method and system of this invention is preferably the same as the system and method for the existing wells including the conversion of 3D well data into 2D well data. Specifically, it includes well data such as the drilling location and all the relevant properties derived from existing wells of the reservoir, such as the above mentioned 2D properties around the well, specifically the 2D well data, opportunity index, and / or well level data.For historical production data, preferably the last N months / years (number of months / years to make forecasts) of production data from existing wells of the same reservoir are utilized, which includes choke size, well head flowing pressure, and the number of producing days.A multi-output regression ML model is preferably chosen to be trained and utilized for predicting an N-month forecast, which involves dividing the available wells into a test set and a training set. The test set consists of wells that have been drilled within the past year, while the remaining historical wells are used for training.The multi-output regression models with different architectures are then trained for each future well in the reservoir, and their respective performances are evaluated based on their Root Mean Squared Error (RMSE) across the N forecast points (number of months / years to make forecasts). The multi-output regression model with the architecture that produces the lowest RMSE across the N points is selected for prediction.Figure 8 shows a flowchart of predicting the production rate for the future wells.In block 8oi, the input data, including the data representing the aspects of the existing (active) wells of the same reservoir as the future well(s), (e.g., well properties such as the 2D well data, the opportunity index by year, the well connection data, the sweep efficiency, and / or the history production data by month of the existing wells) are collected for data pre-processing. Preferably, the data are integrated into a more useful dataset by using the data-preprocessing method mentioned above.In block 802, the pre-processed input data are utilized to train a different / unique ML model (i.e., an ML model with a different architecture). The ML models are then evaluated. The ML model with the lowest RMSE is chosen for predicting the future well in the reservoir.In block 803, the chosen ML model is then used to predict the production rate of the future well in the upcoming period of time (N months).In the present invention, the terms 3D grid, 3D properties, 3D model and / or 3D well data all represent a numerical representation of intricate patterns of the physical and chemical characteristics within a defined three-dimensional (3D) domain, such as a 3D area symbolizing a section of a reservoir formation around a wellbore or an area around a drilling rig. While 2D properties may refer to the aggregated properties of the wellbore, 2D well data refers to that the 2D properties are located on a 2D grid plane according to the well connection data of the 3D well data. In other words, the 2D well data is obtained by mapping or projecting the aggregated 2D properties to their relevant positions in a 2D grip plane based on the 3D connection data.The described approach can be implemented through computer-based methods, computer-readable media with instructions, and computer systems with memory and processors.Implementations of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Software implementations of the described subject matter can be implemented as one or more computer programs, that is, one or more modules of computer program instructions encoded on a tangible, non-transitory, computer- readable computer-storage medium for execution by, or to control the operation of, data processing apparatus. Alternatively, or additionally, the program instructions canbe encoded in / on an artificially generated propagated signal, for example, a machinegenerated electrical, optical, or electromagnetic signal that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. The computer-storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of computer-storage mediums. Configuring one or more computers means that the one or more computers have installed hardware, firmware, or software (or combinations of hardware, firmware, and software) so that when the software is executed by the one or more computers, particular computing operations are performed.The term "real-time," "real time," "realtime," "real (fast) time (RFT)," "near(ly) realtime (NRT)," "quasi real-time," or similar terms (as understood by one of ordinary skill in the art), means that an action and a response are temporally proximate such that an individual perceives the action and the response occurring substantially simultaneously. For example, the time difference for a response to display (or for an initiation of a display) of data following the individual's action to access the data may be less than i ms, less than i sec, or less than 5 sees. While the requested data need not be displayed (or initiated for display) instantaneously, it is displayed (or initiated for display) without any intentional delay, taking into account processing limitations of a described computing system and time required to, for example, gather, accurately measure, analyze, process, store, or transmit the data.The terms "data processing apparatus," "computer," or "electronic computer device" (or equivalent as understood by one of ordinary skill in the art) refer to data processing hardware and encompass all kinds of apparatus, devices, and machines for processing data, including by way of example, a programmable processor, a computer, or multiple processors or computers. The apparatus can also be, or further include special purpose logic circuitry, for example, a central processing unit (CPU), an FPGA (field programmable gate array), or an ASIC (application-specific integrated circuit). In some implementations, the data processing apparatus or special purpose logic circuitry (or a combination of the data processing apparatus or special purpose logic circuitry) may be hardware- or software-based (or a combination of both hardware- and software-based). The apparatus can optionally include code that creates an execution environment for computer programs, for example, code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of execution environments. The present disclosure contemplates the use of dataprocessing apparatuses with or without conventional operating systems, for example LINUX, UNIX, WINDOWS, MAC OS, ANDROID, IOS, or any other suitable conventional operating system.A computer program, which may also be referred to or described as a program, software, a software application, a module, a software module, a script, or code can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data, for example, one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, for example, files that store one or more modules, sub-programs, or portions of code. A computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a communication network.

[0106] While portions of the programs illustrated in the various figures are shown as individual modules that implement the various features and functionality through various objects, methods, or other processes, the programs may instead include a number of sub-modules, third- party services, components, libraries, and such, as appropriate. Conversely, the features and functionality of various components can be combined into single components, as appropriate. Thresholds used to make computational determinations can be statically, dynamically, or both statically and dynamically determined.The methods, processes, or logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output. The methods, processes, or logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, for example, a CPU, an FPGA, or an ASIC.Computers suitable for the execution of a computer program can be based on general or special purpose microprocessors, both, or any other kind of CPU. Generally, a CPU will receive instructions and data from and write to a memory. The essential elements of a computer are a CPU, for performing or executing instructions, and one or more memory devices for storing instructions and data. Generally, a computer will alsoinclude, or be operatively coupled to, receive data from or transfer data to, or both, one or more mass storage devices for storing data, for example, magnetic, magneto-optical disks, or optical disks. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, for example, a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device, for example, a universal serial bus (USB) flash drive, to name just a few.Computer-readable media (transitory or non-transitory, as appropriate) suitable for storing computer program instructions and data includes all forms of permanent / nonpermanent or volatile / non-volatile memory, media and memory devices, including by way of example semiconductor memory devices, for example, random access memory (RAM), read-only memory (ROM), phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), erasable programmable read-only memory (EPROM), electrically erasable programmable readonly memory (EEPROM), and flash memory devices; magnetic devices, for example, tape, cartridges, cassettes, internal / removable disks; magneto-optical disks; and optical memory devices, for example, digital video disc (DVD), CD-ROM, DVD+ / -R, DVD- RAM, DVD-ROM, HD-DVD, and BLURAY, and other optical memory technologies. The memory may store various objects or data, including caches, classes, frameworks, applications, modules, backup data, jobs, web pages, web page templates, data structures, database tables, repositories storing dynamic information, and any other appropriate information including any parameters, variables, algorithms, instructions, rules, constraints, or references thereto. Additionally, the memory may include any other appropriate data, such as logs, policies, security or access data, reporting files, as well as others. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.To provide for interaction with a user, implementations of the subject matter described in this specification can be implemented on a computer having a display device, for example, a CRT (cathode ray tube), LCD (liquid crystal display), LED (Light Emitting Diode), or plasma monitor, for displaying information to the user and a keyboard and a pointing device, for example, a mouse, trackball, or trackpad by which the user can provide input to the computer. Input may also be provided to the computer using a touchscreen, such as a tablet computer surface with pressure sensitivity, a multi-touch screen using capacitive or electric sensing, or other type of touchscreen. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedbackprovided to the user can be any form of sensory feedback, for example, visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user's client device in response to requests received from the web browser.The term "graphical user interface," or "GUI," maybe used in the singular or the plural to describe one or more graphical user interfaces and each of the displays of a particular graphical user interface. Therefore, a GUI may represent any graphical user interface, including but not limited to, a web browser, a touch screen, or a command line interface (CLI) that processes information and efficiently presents the information results to the user. In general, a GUI may include a plurality of user interface (UI) elements, some or all associated with a web browser, such as interactive fields, pulldown lists, and buttons. These and other UI elements maybe related to or represent the functions of the web browser.Implementations of the subject matter described in this specification can be implemented in a computing system that includes a back-end component, for example, as a data server, or that includes a middleware component, for example, an application server, or that includes a front-end component, for example, a client computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of wireline or wireless digital data communication (or a combination of data communication), for example, a communication network. Examples of communication networks include a local area network (LAN), a radio access network (RAN), a metropolitan area network (MAN), a wide area network (WAN), Worldwide Interoperability for Microwave Access (WIMAX), a wireless local area network (WLAN) using, for example, 802.11 a / b / g / n or 802.20 (or a combination of 802.1 lx and 802.20 or other protocols consistent with this disclosure), all or a portion of the Internet, or any other communication system or systems at one or more locations (or a combination of communication networks). The network may communicate with, for example, Internet Protocol (IP) packets, Frame Relay frames, Asynchronous Transfer Mode (ATM) cells, voice, video, data, or other suitable information (or a combination of communication types) between network addresses.The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.While this specification contains many specific implementation details, these should not be construed as limitations on the scope of any invention or on the scope of what may be claimed, but rather as descriptions of features that may be specific to particular implementations of particular inventions. Certain features that are described in this specification in the context of separate implementations can also be implemented, in combination, in a single implementation. Conversely, various features that are described in the context of a single implementation can also be implemented in multiple implementations, separately, or in any suitable sub-combination. Moreover, although previously described features may be described as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can, in some cases, be excised from the combination, and the claimed combination may be directed to a sub-combination or variation of a sub-combination.Particular implementations of the subject matter have been described. Other implementations, alterations, and permutations of the described implementations are within the scope of the following claims as will be apparent to those skilled in the art. While operations are depicted in the drawings or claims in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed (some operations may be considered optional), to achieve desirable results. In certain circumstances, multitasking or parallel processing (or a combination of multitasking and parallel processing) may be advantageous and performed as deemed appropriate.Moreover, the separation or integration of various system modules and components in the previously described implementations should not be understood as requiring such separation or integration in all implementations, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.Accordingly, the previously described example implementations do not define or constrain this disclosure. Other changes, substitutions, and alterations are also possible without departing from the spirit and scope of this disclosure.Furthermore, any claimed implementation is considered to be applicable to at least a computer-implemented method; a non-transitory, computer-readable medium storing computer-readable instructions to perform the computer-implemented method; and a computer system comprising a computer memory interoperably coupled with a hardware processor configured to perform the computer-implemented method or the instructions stored on the non-transitory, computer-readable medium.

Claims

Claims1. A computer-implemented method for training a machine learning model for predicting well production rate, comprising the steps of: receiving input data representing an aspect of a well; generating three-dimensional well data representing properties of the well at least partially based on the data representing the aspect of the well; converting the three-dimensional well data into two-dimensional well data; and training the machine learning model through inputting at least the two- dimensional well data and historical production data of the well into the machine learning model.

2. The training method of the preceding claim, wherein the three-dimensional well data are converted into the two-dimensional well data through aggregation.

3. The training method of any one of the preceding claims, wherein the step of converting the three-dimensional well data into two-dimensional well data comprises the step of: dividing the three-dimensional well data into multiple sets of three-dimensional well data, and converting the multiple sets of three-dimensional well data into the multiple sets of two-dimensional well data for training the machine learning model.

4. The training method of any one of the preceding claims, further comprising the step of: inputting an opportunity index, sweep efficiency, and / or the well connection data of the well into the machine learning model to train the machine learning model,wherein the well connection data represents a trajectory of the wellbore of the well in three-dimensional space.

5. The training method of any one of the preceding claims, wherein the machine learning model training step comprises the steps of: clustering multiple wells into a first number of well groups by using an Al clustering model based on the similarity of an aspect of the multiple wells, and training a first set of different machine learning models for each well group through inputting at least the two-dimensional well data and historical production data of the wells of the corresponding well group into the first set of machine learning models.

6. The training method of the preceding claim, wherein the Al clustering model is an unsupervised machine learning algorithm, preferably a K-means algorithm.

7. The training method of any one of claims 5 or 6, wherein the well groups are clustered by the Al clustering model based on a similarity of the historical production data of the wells.

8. The training method of any one of claims 5 to 7, wherein the machine learning model training step further comprises the step of: evaluating the first set of machine learning models at least based on their Root Mean Squared Error, RMSE, or a combination of their cumulative error and RMSE across N forecast points in a certain period of time.

9. The training method of any one of claims 5 to 8, wherein the machine learning model training step further comprises the steps of:clustering the multiple wells into a second number of well groups by using the Al clustering model based on the similarity of the aspect of the multiple wells, training a second set of different machine learning models for each well group, which have been clustered with the second number of well groups, evaluating the second set of different machine learning models at least based on their Root Mean Squared Error, RMSE, or a combination of their cumulative error and RMSE across N forecast points in a certain period of time, comparing results of the evaluations of the first and second set of machine learning models and selecting the set of machine learning models that produces a lower RMSE, or a lower RMSE and cumulative error across the certain period of time. io. The training method of the preceding claim, wherein the machine learning model training step further comprises the steps of: repeating the clustering, training, and evaluating with different numbers of well groups in a multi-iteration manner, until the RMSE or the combination of cumulative error and RMSE of the machine learning models is minimized. n. The training method of any one of the preceding claims, wherein: the historical production data includes monthly production rates, choke size, production days, and / or well head flowing pressure; the data representing an aspect of a well comprises the coordinates and the trajectory of the well, the connection status, Kh, permeability, porosity, saturation, pressure, and / or well type; and / or the two-dimensional and three-dimensional data are generated at least based on the porosity, permeability, saturation, reservoir pressure, and / or coordinates and / or trajectory of the well.

12. The training method of the preceding claim, wherein the machine learning model is multi-output regression model or time series forecasting ML model.

13. The training method of any one of the preceding claims, further comprising: generating three-dimensional well data by using a simulation program with the aspect of the well.

14. A computer-implemented method for predicting the well production rate based on a machine learning model trained by the method of any one of claims 1 to 13, the method comprising the steps of: obtaining a predicted well production rate of an existing well by using the machine learning model that is trained by the well group which includes the existing well, or obtaining a predicted well production rate of a future well by using the machine learning model trained based on the wells of the same reservoir.

15. The computer-implemented method of the preceding claim, wherein the step of predicting the well production rate includes a production uncertainty envelope.

16. The computer-implemented method of the preceding claim, wherein the specified percentiles of the production uncertainty envelope correspond to at least 10 percentile, 50 percentile, and / or 90 percentile of the predicted well production rate.

17. A system for predicting a well production rate, comprising an information processing system configured to implement the method according to any one of claims 1 to 16.

Images