Predicting a fluid property within a hydrocarbon reservoir

Abstract

A method of generating a model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir. The method comprises a training step where size exclusion chromatogr

Description

Technical field The present invention relates to a method of generating a model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, to a computer-based model generated by the method, and to the use of the computer based-model. The present invention also relates to a method and system for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir. Background In the oil and gas industry, there is frequently a need to acquire information on the properties of fluid at a sample location within a hydrocarbon reservoir. Knowing such properties is important in the context of, for example, well placement, reservoir management, production optimization, and flow assurance. Measuring these properties typically requires direct access to a reservoir fluid sample. However, it is often the case that no or insufficient reservoir fluid sample is available for such measurements. In reservoir management and production optimization for example, the necessary reservoir fluid samples are often unavailable in the new development strategy of producing mature and near-field fields. For flow assurance studies, a large quantity of reservoir fluid samples is required to measure key properties of wax and asphaltene to assist a good production system design. Such large samples are difficult to obtain for expensive deep-water projects or small tie-back projects. For well placement, the existing technology is limited to petrophysical logging tools. For development wells, the logging tools are limited to basic gamma ray, resistivity, and density / neutron, which are insufficient to identify the reservoir fluids accurately. There is therefore a need for a method by which properties of a fluid at a sample location within a hydrocarbon reservoir can easily be predicted. The method should be able to be performed using only a small amount of fluid sample, be able to easily and reliably predict properties, e.g. a plurality of properties, of interest with easy high throughput, and be adaptable to account for the presence of contaminants introduced during fluid recovery. The method should have improved predictive power and easier operation. Elias et al., "Assessment of Producible Fluid Quality From the Vaca Muerta Unconventional Play (Neuquen Basin, Argentina)" presented at the SPE / AAPG / SEG Unconventional Resources Technology Conference, Austin, Texas, USA, July 2017. doi: https: / / doi.org / 10.15530 / URTEC-2017-2688044, relates to the correlation of a GPC parameter with API and, to a lesser extent, Hydrogen Index. The method makes use of rock samples. However, the method is limited in that the GPC data comprises values for only one variable (specifically, absorbance at only one particular wavelength) as a function of retention time. This limits the predictive power of the method. Moreover, a calibration curve approach limits the throughput and easy operation of the method. WO2010104728A2 relates to a system and method for determining properties of a downhole oil sample. The method involves downhole analysis based on size exclusion chromatography to produce a chromatography sample. A first property of the chromatography sample is measured, and a second property of the chromatography sample is estimated using a calibration curve approach. However, as already noted, a calibration curve approach adopted limits the throughput and easy operation of the method. Summary Viewed from a first aspect, the present invention provides a method of generating a model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, comprising: providing a training data set comprising input data and target data, the input data comprising input size exclusion chromatography (SEC) data for each of a plurality of sample locations, and the target data comprising the at least one property of the fluid for each of the plurality of sample locations; and instructing a machine learning algorithm to generate the model using the training data set such that the model can be used to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location, wherein the input SEC data and the measured SEC data each comprise values for at least two variables as a function of retention time, the at least two variables in the input SEC data being the same as the least two variables in the measured SEC data. Viewed from a further aspect, the present invention provides a computer-based program for generating a model by a method as hereinbefore described. Viewed from a further aspect, the present invention provides a computer-based model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir based on measured SEC data for that sample location, the computer-based model having been generated by a method as hereinbefore described, wherein the measured SEC data comprises measured values for the at least two variables as a function of retention time. Viewed from a further aspect, the present invention provides a method of predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, the method comprising: receiving measured SEC data for the sample location; and predicting the value of the property of the fluid at the sample location by supplying the measured SEC data to a computer-based model as hereinbefore described, wherein the measured SEC data comprises measured values for the at least two variables as a function of retention time. Viewed from a further aspect, the present invention provides use of SEC to measure values for at least two variables as a function of retention time in a method for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir. The method may be a method as hereinbefore described. Viewed from a further aspect, the present invention provides use of a computer-based model as hereinbefore described to predict at least one property of a fluid at a sample location within a hydrocarbon reservoir. The use may be in a method as hereinbefore described. Viewed from a further aspect, the present invention provides a system for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, the system comprising: a SEC apparatus; and a computer-based model as hereinbefore described, wherein the SEC apparatus is configured to measure values for at least two variables as a function of retention time; and the computer-based model is configured to be supplied with said measured SEC values. Brief Description of the Drawings Figure 1 illustrates a schematic workflow for generating a model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir. Figure 2 illustrates a schematic representation of a method for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir. Detailed description As used herein, the term “size exclusion chromatography”, or “SEC”, refers to a chromatography method which separates particles in a sample according to shape and / or size. The separation causes different particles in the sample to elute with different characteristic retention times. An example type of SEC is gel permeation chromatography (GPC). As used herein, the term “SEC data” collectively refers to both input SEC data and measured SEC data. As used herein, the term “drill cuttings” refers to material recovered from a sample location in a hydrocarbon reservoir following drilling. For example, drill cuttings may be recovered from a well or borehole following drilling. Drill cuttings will typically comprise rock as well as small amounts of fluid derived from the sample location. Drill cuttings are typically recovered from returning drilling mud. As used herein, the term “drilling mud” refers to fluid used to aid the process of drilling to a sample location in a hydrocarbon reservoir. Drilling mud may be used to recover drill cuttings from a sample location in a hydrocarbon reservoir. Types of drilling mud include oil-based mud and water-based mud. As used herein, the term “machine learning algorithm” refers to an algorithm which is configured to generate a model using a training data set. The generated model can be considered to be the result of training using the machine learning algorithm and the training data set. The generated model may therefore equivalently be termed a trained model. As used herein, the term “training data set” refers to a data set used by a machine learning algorithm to generate a model. As used herein, the term “fluid” encompasses oil, gas, and mixtures thereof. As used herein, the term “absorbance” refers to the amount of light of a given wavelength absorbed by a sample. Absorbance values can be obtained by using a spectrophotometer and by employing spectroscopy techniques known in the art. In some embodiments, wavelengths are UV and / or UV-visible (equivalently UV-Vis or UV / vis) wavelengths. Absorbance values at such wavelengths can be obtained using a UV and / or UV-visible spectrophotometer and by employing spectroscopy techniques known in the art. As used herein, the term “refractive index” or “RI” refers to light deflection of a sample in a solvent. Refractive index values are differential between a sample in solvent and a comparative cell with the solvent. Refractive index is obtained by differential refractometer techniques known in the art. Method of generating a model The approach presented here generates a trained model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, the method of generating a model being as hereinbefore described. The reservoir may for example be a gas reservoir, a multiphase reservoir (oil and gas), or an oil reservoir (under-saturated with gas). The fluid sample may for example constitute a sample flowed into a downhole tool (e.g. as in a MDT), a fluid flowed to a separator (e.g. as in a DST), or a fluid extracted by an organic solvent from a reservoir rock sample (drill cuttings, sidewall core or core). In some embodiments, the size exclusion chromatography (SEC) data may comprise absorbance values for at least one, such as a plurality, of wavelengths as a function of retention time and / or refractive index values as a function of retention time. The model may be used to predict a plurality of properties, e.g. at least two properties, of a fluid at a sample location within a hydrocarbon reservoir. The method may be able to generate a model with high predictive power, with easy and high throughput operation, and which enables reliable predications to be obtained even when only a small amount of fluid sample is available. Using at least two variables may increase the predictive power and thus commercial value of the method. Utilising a machine learning algorithm may allow for easy generation and exploitation of the model and for high throughput, facilitating the use of large data sets with a plurality of variables and the possibility of predicting a plurality of properties. The new method may reduce the data acquisition cost significantly, and it may provide the reservoir oil and gas properties when no other tools are available. The method includes a step of instructing a machine learning algorithm to generate the model using the training data set such that the model can be used to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location. The model resulting from this step can be considered a “trained model”, having been trained on the training data set. In other words, the method can equivalently be described as including a step of using a machine learning algorithm and the training data set to obtain a trained model such that the trained model can be used to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location. Use of a machine learning algorithm may allow for easier generation, or - said another way - training, of the model. As the SEC data comprises values for a plurality, i.e. at least two, variables as a function of retention time, an alternative approach, for example using a calibration curve approach, would be highly laborious and difficult to implement. The greater the number of the plurality of sample locations and the greater the number of variables for which values are provided, the more difficult it would be to generate the model without the aid of a machine learning algorithm. The proposed method may allow for the generation of more complex and sophisticated models with greater predictive power. The nature of the machine learning algorithm is not particularly limited, and it will be appreciated that any suitable machine learning algorithm may be used. Examples include Universal Kriging, Support Vector Machine, KMean, Elastic Net, artificial neural networks, Random Forest, and Gaussian Regression algorithms. The machine learning algorithm may be a machine learning algorithm suitable for training a neural network based model to generate a trained neural network model. Suitable machine learning algorithms will be available to the skilled person commercially or from open sources. Bespoke machine learning algorithms could also be devised. Those operating within this field will be familiar with the procedures for selecting and utilising a machine learning algorithm. Machine learning algorithms are currently available quite readily from commercial or open sources, meaning that it may be easy and inexpensive for a model to be generated according to the method. Generating the model may comprise: training a machine learning algorithm with a first subset of the training data set; and testing the machine learning algorithm with a second, disjoint subset of the training data set. The first subset may for example comprise at least 60% of the samples of the training data set. The second subset may for example comprise at least 20% of the samples of the training set. The training data set may be prepared by performing analysis on fluid samples recovered from a sample location. Input data may be obtained by measuring SEC data for a sample, namely recording values for at least two variables as a function of retention time to yield input data. Target data may be obtained by measuring the at least one property of the fluid. The skilled person will know appropriate methods for measuring different properties. Those skilled in the art will typically have access to recovered fluid samples, including historic fluid samples which are often kept in storage facilities for relatively long periods of time. It will also be possible for the skilled person to recover de novo fluid samples from hydrocarbon reservoirs for each of a plurality of sample locations for the purposes of preparing a training data set. The samples may be PVT samples or dead oil samples. The training data set may for example comprise input data and target data for at least 50, at least 100, at least 200, at least 250, e.g. at least 500 different sample locations. The plurality of sample locations may be from a plurality of different hydrocarbon reservoirs. The greater the number and diversity of the sample locations, and the greater the diversity of the fluid properties therein, the greater the adaptability, predictive power, and accuracy of the model. The Data The proposed method makes use of size exclusion chromatography (SEC), for example gel permeation chromatography (GPC), data. SEC enables the skilled person to measure values for different variables as a function of retention time. By using the model generated according to this method, these measurements can be used to predict at least one property of a fluid at a sample location within a hydrocarbon reservoir. The input SEC data and the measured SEC data each comprise values for at least two variables as a function of retention time. It will be understood that the at least two variables in the input SEC data are the same as the at least two variables in the measured SEC data. In other words, the at least two variables are the same two variables in the measured SEC data and in the input SEC data. For example, if the input SEC data comprises absorbance values for a given wavelength as a function of retention time and RI values as a function of retention time, the measured SEC data also comprises absorbance values for that same wavelength as a function of retention time and RI values as a function of retention time. This permits for easier correlation between input and measured SEC data, which may improve the quality of the prediction. Values in the measured SEC data may be as a function of the same retention times as in the input SEC data. The SEC data comprising values for a plurality of, i.e. at least two, variables may increase the predictive power and thus commercial value of the method. For example, this may improve the accuracy of the prediction and / or enable a plurality of different properties to be predicted. The increased complexity associated with providing SEC data comprising values for a plurality of variables is accommodated through the use of a machine learning algorithm. Example embodiments for the SEC data are described below. As noted above, the at least two variables in the input SEC data are the same as the at least two variables in the measured SEC data. Variables may be spectroscopic variables. Such variables may be those related to the fields of absorption spectroscopy, emission spectroscopy (fluorescence), elastic scattering and reflection spectroscopy, impedance spectroscopy, inelastic scattering spectroscopy, Raman spectroscopy, coherent or resonance spectroscopy, and / or quantum spectroscopy. Variables may be those related to the fields of absorption, emission, and / or scattering spectroscopy. For example, the SEC data may comprise absorbance values for at least one wavelength as a function of retention time. In other words, the measured SEC data and the input SEC data may each comprise absorbance values for at least one wavelength as a function of retention time, the at least one wavelength being the same wavelength in both the measured SEC data and in the input SEC data. The SEC data may comprise absorbance values for a plurality of wavelengths as a function of retention time. For example, the SEC data may comprise absorbance values for at least two, at least three, at least four, at least five, or more different wavelengths as a function of retention time. The SEC data may comprise absorbance values for a range, such as a substantially continuous range, of wavelengths as a function of wavelength and retention time. Different constituents of a fluid sample will exhibit different characteristic levels of absorbance at different characteristic wavelengths. When the SEC data comprises values for a plurality of wavelengths as a function of retention time, greater insight may be provided on the composition of the sample fluid. This may improve the power of the model to predict the relative amounts or concentrations of different types of hydrocarbon present in the fluid or to predict other properties related to the composition of the fluid sample. In some embodiments, absorbance values may be for UV and / or UV-visible wavelengths. For example, wavelengths may be selected from 190 to 640 nm, or ranges contained therein. The SEC data may comprise absorbance values for a range, such as a substantially continuous range, of UV and / or UV-visible wavelengths. Such an approach would yield absorbance values as a function of wavelength and retention time. Example ranges of UV wavelengths are as listed above. Absorbance values for a UV or UV-visible wavelength may primarily be influenced by non-saturated hydrocarbons present in the fluid sample. Saturated hydrocarbons are less likely to exhibit UV or UV-visible absorbance and so can be considered somewhat “invisible” to a UV or UV-visible absorbance detector. This may be advantageous when the sample is contaminated with saturated hydrocarbons, for example those contained in oil-based mud which may be used to recover a fluid sample. As the saturated hydrocarbons exhibit little or no UV or UV-visible absorbance, the absorbance values may be primarily attributable to the fluid sample itself, rather than to the contaminants. This is in contrast to other analysis methods such as gas chromatography, where, due to high contamination from oil-based mud, values are dominated by the presence of contaminants. It will be understood, however, that absorbance values are not to be limited to UV or UV-visible wavelengths. Other wavelengths, such as infrared wavelengths and / or microwave wavelengths, may also be used. Variables may be refractive index variables, direct or differential. Such variables are related to the change in direction of light called refraction. Accordingly, the SEC data may comprise refractive index (RI) values as a function of retention time. This may be in addition to the SEC data comprising absorbance values as discussed above. The SEC data may comprise absorbance values for at least one wavelength as a function of retention time and RI values as a function of retention time, for example absorbance values for at least one UV-visible wavelength as a function of retention time and RI values as a function of retention time. The SEC data may comprise absorbance values for a plurality of wavelengths as a function of retention time and RI values as a function of retention time, absorbance values for a plurality of UV or UV-visible wavelengths as a function of retention time and RI values as a function of retention time, and / or absorbance values for a range, such as a substantially continuous range, of UV or UV-visible wavelengths as a function of retention time and RI values as a function of retention time. In some embodiments, the SEC data may therefore comprise absorbance values for a range, such as a substantially continuous range, of wavelengths as a function of wavelength, the refractive index and retention time. When RI values are combined with UV or UV-visible absorbance values, there may be a desirable complementary effect. RI values will detect saturated hydrocarbons, species which can be considered somewhat “invisible” to a UV-visible detector, as discussed above. The effect in combination may therefore be to provide a more complete insight into fluid composition. The SEC data may comprise fluorescence values for at least one absorbed wavelength as a function of retention time. Fluorescence values may include fluorescence intensity, emitted wavelength, fluorescence lifetime, and / or quantum yield. Example embodiments are as discussed above in connection with absorbance values. Similarly, the SEC data may comprise values for other forms of luminescence, such as phosphorescence. Fluorescence values can be obtained using fluorescence spectrometers and by employing spectroscopy techniques known in the art. The SEC data may comprise values for a continuous or substantially continuous range of retention times. For example, the SEC data may comprise values from a retention time of approximately zero or from a retention time at which some component of the sample first begins to elute. The SEC data may, for example, comprise values up to a retention time at which at least 50%, 60%, 70%, 80%, 90%, 95%, or 99%, or substantially 100% of the sample has eluted. The greater the range of retention times for which values are available, the insight provided into the composition of the sample and the predictive power of the model may be the greater. For SEC / GPC, retention time may alternatively be referred to as retention volume (depending on flow rate, e.g. 1 ml / min, 0.5 ml / min, etc.). The retention volume (retention time) provides information on compound and / or oil fraction hydrodynamic volume. The hydrodynamic volume is closely related to molecular weight of the compound and / or oil fraction. The GPC retention volume may be calibrated using a series of known compounds and / or standard oil fraction (saturate, aromatic, resin and asphaltenes). In some embodiments, values may be provided as absolute values as a function of retention time. Alternatively, or additionally, values may be provided as ratios taken across a range of retention times. For example, an integration method may be used to quantify raw data recorded across a range of retention times. Ratios of different integrals may then be used as values for the model. In some embodiments, values as a function of retention time may be normalised across the range of retention times for which values are recorded. This ensures that the values are less dependent, or independent, of sample concentration effects. In this way the values have similar scales which facilitates the learning process and reduces possible converge problems on the machine learning algorithms Values derived from the derivative or second derivative, etc. of the detector signal may also be incorporated as absolute or ratio form relative to retention time. Properties In the proposed method, the identity of the at least one property is not particularly limited. An advantage of generating a model according to the method may be that it may be possible to obtain a high level of predictive power across different properties. This is facilitated by the use of SEC data comprising values for a plurality, i.e. at least two, variables as a function of retention time and by making use of a machine learning to provide easy prediction. Together, these features mean that the at least one property may be predicted with high accuracy, the identity of the at least one property need not be particularly limited, and / or the model may be capable of predicting a plurality, i.e. more than one, different properties. The at least one property may comprise a density of the fluid at the sample location. It will be appreciated that the density may be calculated either at atmospheric conditions or reservoir conditions (e.g. taking into account the oil formation volume factor). The at least one property may comprise a gas-oil ratio. That is to say, a ratio between the quantity of gaseous hydrocarbon and the quantity of liquid hydrocarbon, which is normally determined at surface conditions. The gas-oil ratio may be a volume ratio. The gas-oil ratio may be a single-flash gas-oil measurement. However, any suitable gas-oil measurement may be used. The at least one property may comprise a saturation pressure of the fluid at the sample location. That is to say, the pressure at which a secondary phase will appear with pressure depletion. The at least one property may comprise a formation volume factor of the fluid at the sample location. That is to say, the ratio of the volume of the fluid at reservoir (in-situ) conditions to the volume of the fluid at surface conditions. The at least one property may comprise a concentration of a hydrocarbon within the fluid at the sample location. The at least one property may comprise a concentration of asphaltene within the fluid at the sample location. The at least one property may comprise a concentration of C?+ hydrocarbons within the fluid at the sample location. The at least one property may comprise a concentration of aromatic hydrocarbon within the fluid at the sample location. The at least one property may comprise a concentration of polar hydrocarbon within the fluid at the sample location. The at least one property may comprise SARA data for the fluid at the sample location, that is to say a profile of the respective saturate, aromatic, resin and asphaltene concentrations of the fluid at the sample location. The at least one property may comprise a concentration of asphaltene within the fluid at the sample location and the SEC data may comprise absorbance values for a wavelength, or a plurality or range of wavelengths, selected from 290 to 640 nm, e.g. over an early retention time commensurate with high molecular weight asphaltene compounds. Asphaltenes typically exhibit characteristic absorbance at these characteristic wavelengths. Alternatively, the least one property may not comprise a concentration of asphaltene within the fluid at the sample location. The at least one property may comprise a concentration of porphyrin within the fluid at the sample location. Porphyrin structures are often associated with asphaltene and / or resin fractions. The at least one property may comprise a concentration of porphyrin within the fluid at the sample location and the SEC data may comprise absorbance values for a wavelength in the range of 380 to 420 nm, for example about 400 nm. Porphyrin structures typically exhibit characteristic absorbance at these characteristic wavelengths. The at least one property may comprise a concentration of porphyrin associated with nickel and / or vanadium within the fluid at the sample location. The least one property may comprise a concentration of porphyrin associated with nickel and / or vanadium within the fluid at the sample location and the SEC data may comprise absorbance values for a wavelength in the range of 514 to 550 nm and / or 531 to 570 nm respectively. Nickel and / or vanadium porphyrin complexes typically exhibit characteristic absorbance at these characteristic wavelengths respectively. The at least one property may comprise an API gravity of the fluid at the sample location. Alternatively, the at least one property does not comprise an API gravity of the fluid at the sample location. The at least one property may comprise a stock tank oil viscosity of the fluid at the sample location, that is to say a measurement of the resistance of crude oil to flow at a specific temperature and pressure when it is at surface or "stock tank" conditions. These conditions typically include a temperature of 15.6 °C and atmospheric pressure. The at least one property may comprise a bubble point / dew point pressure of the fluid at the sample location, that is to say the bubble point corresponds to the minimum pressure at which a hydrocarbon mixture remains in a single-phase liquid state and the dew point corresponds to the maximum pressure at which a hydrocarbon mixture remains in a single-phase vapour state. The at least one property may comprise a wax appearance temperature of the fluid at the sample location, that is to say the temperature at which the solubility of wax in the oil phase is exceeded, leading to the nucleation and growth of wax crystals The at least one property may comprise an asphaltene on-set pressure of the fluid at the sample location, that is to say the pressure level at which the concentration of asphaltenes in the fluid reaches a critical threshold and starts to agglomerate and form solid particles. The model may be for predicting a plurality of properties of a fluid at a sample location. For example, the model may be for predicting at least two, at least three, at least four, at least five, or more properties of a fluid at a sample location. The greater the number of properties, the greater the predictive power of the model and the greater the commercial benefit. This high level of predictive power may be facilitated through the use of a machine learning algorithm and by the SEC data comprising values for a plurality, i.e. at least two, variables as a function of retention time. In some embodiments, the number of properties may be less than or equal to the number of variables as a function of retention time for which values are available in the SEC data. This numerical balance of inputs and outputs may improve the accuracy of the prediction produced by the model. Additional Steps The method may further comprise determining a quality for the measured SEC data. The method may further comprise generating an indication of confidence associated with the predicted value of the fluid property. The indication of confidence may be a numerical indication, but other indications may be used, such as colour indications (e.g. red / yellow / green), or word indications (e.g. “good” I “poor”). The method may also make use of non-SEC data. For example, the input data may comprise non-SEC data and the method may comprise instructing a machine learning algorithm to generate the model using the training data set such that the model can be used to predict the at least one property of the fluid at the sample location based on a combination of measured SEC data and measured non-SEC data for the sample location. Non-SEC data may for example include mud-gas data, isotube gas composition and isotope data, geochemical data, and / or petrophysical data. The use of non-SEC data may increase the diversity of data fed to the model, improving the predictive power and accuracy thereof. The proposed method need not be conducted downhole. In some embodiments, the SEC data is not obtained downhole. In other words, samples may be recovered from a hydrocarbon reservoir and SEC data may then be obtained at surface level, rather than downhole. As SEC and the preparation of samples therefor may more easily be conducted at surface level, it may be easier to implement these embodiments. Computer-based programs There is also proposed a computer-based program for generating a model according to the method hereinbefore described and to a tangible computer-readable medium storing such a computer-based program. For example, the program may comprise prompts for inputting a training data set and may be configured to instruct a machine learning algorithm to generate the model using the training data set such that the model can be used to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location. The present invention also relates to a computer-based model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, as hereinbefore described. Example embodiments, e.g. relating to the SEC data and the at least one property, are as discussed above in relation to the method of generating the model. The present invention also relates to a tangible computer-readable medium storing such a computer-based model. These aspects may be embodied as a computer program product for use with a computer system. Such an implementation may comprise a series of computer readable instructions, which may be fixed on a tangible, non-transitory medium, such as a computer readable medium, for example, diskette, CD ROM, DVD, ROM, RAM, flash memory or hard disk. It could also comprise a series of computer readable instructions transmittable to a computer system, via a modem or other interface device, over either a tangible medium, including but not limited to optical or analogue communications lines, or intangibly using wireless techniques, including but not limited to microwave, infrared or other transmission techniques. The series of computer readable instructions embodies all or part of the functionality previously described herein. Those skilled in the art will appreciate that such computer readable instructions can be written in a number of programming languages for use with many computer architectures or operating systems. Further, such instructions may be stored using any memory technology, present or future, including but not limited to, semiconductor, magnetic or optical, or transmitted using any communications technology, present or future, including but not limited to optical, infrared or microwave. It is contemplated that such a computer program product may be distributed as a removable medium with accompanying printed or electronic documentation, for example, shrink wrapped software, pre-loaded with a computer system, for example, on a system ROM or fixed disk, or distributed from a server or electronic bulletin board over a network, for example, the Internet or World Wide Web. The proposal also extends to a computer software carrier comprising such software arranged to carry out the steps of the described methods. Such a computer software carrier could be a physical storage medium such as a ROM chip, CD ROM, DVD, RAM, flash memory or disk, or could be a signal such as an electronic signal over wires, an optical signal or a radio signal such as to a satellite or the like. Use of a computer-based model allows for easier exploitation of the model. As the SEC data comprises values for a plurality, i.e. at least two, variables as a function of retention time, an alternative approach, for example using a calibration curve approach, would be highly laborious and difficult to implement. The greater the number of variables for which values are provided, and / or the greater the number of properties to be predicted, the more difficult it would be to exploit the model without computing support. The method thus allows for the exploitation of more complex and sophisticated models with greater predictive power, and with the possibility of predicting a plurality of properties. Method of predicting a value Example embodiments, e.g. relating to the SEC data and the at least one property, are as discussed above in relation to the method of generating the model. The method of predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir comprises receiving measured SEC data for the sample location. This step may comprise: recovering a sample from the sample location; subjecting the sample to SEC; and measuring values for at least two variables as a function of retention time. It is understood that the sample recovered from the sample location will comprise fluid from the sample location. Recovery may be by any method known to the skilled person for the recovery of fluid from a sample location in a hydrocarbon reservoir. The fluid sample may for example constitute a sample flowed into a downhole tool (e.g. as in a MDT), a fluid flowed to a separator (e.g. as in a DST), or a fluid extracted by an organic solvent from a reservoir rock sample (drill cuttings, sidewall core or core). As discussed below, in some embodiments the sample is recovered through the recovery of drill cuttings. Recovering a sample from the sample location may comprise recovering a drill cuttings sample from the sample location. In other words, an example method comprises: recovering a drill cuttings sample from the sample location; subjecting the sample to SEC; and measuring values for at least two variables as a function of retention time. The use of drill cuttings may be advantageous as these cuttings will frequently be readily available for a sample location in a hydrocarbon reservoir. They will also typically comprise small amounts of fluid from the sample location, which can be subjected to SEC. The sensitivity of SEC as described herein means that measured values can be obtained from even these small amounts of fluid. The method will thus change significantly the way drill cuttings are handled and make them an important low-cost but high-value data source. The method may further comprise a step of treating the drill cuttings sample to obtain a sample suitable for SEC. This may include sieving, washing with water (with or without surfactant), and / or using solvent, e.g. organic solvent, to extract a fluid sample which can suitably be subjected to SEC. An example method is thus a method of predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, the method comprising: recovering a drill cuttings sample from the sample location; optionally, treating the drill cuttings sample to obtain a sample suitable for SEC; subjecting the sample to SEC and measuring values for at least two variables as a function of retention time; and predicting the value of the at least one property of the fluid at the sample location by supplying the measured SEC data to a computer-based model as hereinbefore described. Drill cuttings may be recovered using drilling mud, including oil-based mud or waterbased mud. When the at least two variables include absorbance values for a UV-visible wavelength as a function of retention time, any signal attributable to mud contamination of the fluid sample may be substantially “invisible” or may be reduced or minimised. This may be advantageous as it means the measured values are attributable to the signal from the fluid sample itself, rather than the mud contaminant. Traditional analysis methods such as gas chromatography are particularly vulnerable to such signal contamination, especially when oil based mud is used. In some embodiments, the drill cuttings may be recovered using oil-based mud. The method may comprise: recovering a drill cuttings sample from the sample location, e.g. using drilling mud; optionally, treating the drill cuttings sample to obtain a sample suitable for SEC; subjecting the sample to SEC; measuring values for at least two variables as a function of retention time; and predicting the value of the at least one property of the fluid at the sample location by supplying the measured SEC data to a computer-based model as hereinbefore described, wherein the measured SEC data comprises measured values for at least two variables as a function of retention time, including absorbance values for at least one UV or UV-visible wavelength as a function of retention time. The method may additionally comprise receiving measured non-SEC data for the sample location. Examples of non-SEC data are as hereinbefore described. In such an embodiment, the input data may comprise non-SEC data and the method of generating a model may comprise instructing a machine learning algorithm to generate the model using the training data set such that the model can be used to predict the at least one property of the fluid at the sample location based on a combination of measured SEC data and measured non-SEC data for the sample location The method may be a method of predicting at least one property of a fluid along a length of a well through a hydrocarbon reservoir, comprising: predicting at least one property of a fluid at a plurality of sample locations along a length of a well using the method described. The method may comprise: displaying, using an electronic display screen, a graph plotting the predicting value of the fluid property against a location of the respective sample location for each of the plurality of sample locations along the length of the well. Size Exclusion Chromatography (SEC) The proposed method makes use of SEC data, whether input SEC data or measured SEC data, for example GPC data. SEC may be advantageous because it has a high sensitivity, meaning that data can be obtained even when only small amounts of fluid are available. This, for example, means that the methods can be associated with the use of drill cuttings samples, which comprise only a small amount of fluid. SEC apparatus can also be incorporated or be coupled to different types of detector, meaning that values for different variables as a function of retention time can be obtained. This may be advantageous as it means that the SEC data can be diverse, improving the predictive power of the model. Alternatively, different fractions can be collected as they elute from the column as the method is non-destructive, meaning that a diverse range of analysis can be performed on each fraction. As noted, SEC is a non-destructive analysis method, meaning that the integrity of the sample is maintained and can be subject to further analysis if desired. This contrasts favourably with more aggressive analysis techniques, such as those involving sample ionisation and / or fragmentation. Example embodiments for implementing SEC in any embodiment are described below. These may apply to either or both of measured SEC data (i.e. receiving measured SEC data) and input SEC data (i.e. providing input SEC data). SEC may comprise the use of an organic solvent as a mobile phase. The solvent may be a solvent in which all, or substantially all, of the fluid is soluble, e.g. at room temperature. Example solvents include, xylene, toluene, isopropanol (IPA), tetrahydrofuran, and mixtures thereof. The skilled person will be able to make appropriate adjustments to SEC methodological parameters, including column length, solvent system etc. In some embodiments, substantially the same SEC methodology may be used to provide the input data as is used to receive measured SEC data. This may permit for easier correlation between input and measured SEC data, which may improve the quality of the prediction. For example, substantially the same column length may be used and / or substantially the same solvent system may be used as a mobile phase. In other words, the measured SEC data and the input SEC data may both for example be SEC data obtained using the same solvent system. However, the machine learning algorithm may also be configured to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location independently of SEC methodological parameters, in particular independently of the solvent system. This reduces the benefit of substantially the same SEC methodology being used to provide the input data as is used to receive measured SEC data. In other words, the machine learning algorithm may be configured to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location independently of SEC methodological parameters, in particular independently of the solvent system. When gathering SEC data, the SEC apparatus may be coupled to or may incorporate a detector configured to measure at least one value as a function of retention time. Such detectors are available commercially. Such a detector may be a differential refractometer or refractive index detector. In such a case, the SEC data may comprise RI values as a function of retention time. Such a detector may be an absorbance detector, which may be configured to measure absorbance values at a given wavelength or at a plurality of wavelengths as a function of retention time. In such a case, the SEC data may comprise absorbance values at that wavelength or plurality of wavelengths as a function of retention time. Such a detector may be a UV-visible detector or a UV-visible spectrophotometer, which may be configured to measure absorbance values at a given UV-visible wavelength or at a plurality of UV-visible wavelengths as a function of retention time. In such a case, the SEC data may comprise absorbance values at that wavelength or plurality of wavelengths as a function of retention time. The detector may be a multi-channel detector or a diode array detector (DAD). Such a detector may be advantageous as it enables values for a plurality of variables to be measured at one time. For example, it may enable absorbance values at a plurality of different wavelengths and / or absorbance values for a range, such as a substantially continuous range, of wavelengths to be recorded at the same time. This enables the provision of a large SEC data set, improving the predictive power and accuracy of the methods. The detector may be located at the end of the column, such that values can be recorded as a function of retention time as components of the sample elute. Alternatively, rather than a detector being directly coupled to the SEC apparatus, subjecting the sample to SEC may include collecting fractions at regular retention time intervals and measuring values for at least two variables for each fraction. The skilled person will be able to select appropriate time intervals which may be for example 30 second, one minute, 90 second, or 2 minute intervals etc. The intervals may be chosen based on a noticeable change in value for a measured variable as a function of retention time. In some embodiments, the SEC apparatus may not be located downhole and / or SEC values may not be measured downhole. Use The proposed methods may also relate to the use of size exclusion chromatography, the use being as hereinbefore described. The use may be in a method as hereinbefore described. The proposed methods may also relate to the use of a computer-based model, the use being as hereinbefore described. The use may be in a method as hereinbefore described. System The proposed methods may also relate to a system for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, the system being as hereinbefore described. Example embodiments, e.g. relating to the SEC data and the at least one property, are as discussed above in relation to the methods hereinbefore described. The SEC apparatus may be coupled or to or incorporate a detector, as hereinbefore described. The system may additionally incorporate means for recovering a sample from the sample location, such as means for recovering a drill cuttings sample from the sample location. In the latter case, the system may also comprise optional means for treating the drill cuttings sample to obtain a sample suitable for SEC. The system may also additionally comprise a computer for operating the computer-based model. Example method An example method is: a method of generating a model for predicting a plurality of properties of a fluid at a sample location within a hydrocarbon reservoir, comprising: providing a training data set comprising input data and target data, the input data comprising input size exclusion chromatography (SEC) data for each of a plurality of sample locations, and the target data comprising the plurality of properties of the fluid for each of the plurality of sample locations; and instructing a machine learning algorithm to generate the model using the training data set such that the model can be used to predict the plurality of properties of the fluid at the sample location based on measured SEC data for the sample location, wherein the input SEC data and the measured SEC data each comprise values for at least two variables as a function of retention time, the at least two variables in the input SEC data being the same as the least two variables in the measured SEC data. EXAMPLES The proposed method will now be described with reference to the following non-limiting figures and examples. The following technique seeks to utilise a machine learning algorithm to generate a model that accurately predicts at least one, for example a plurality, of properties relating to the reservoir fluid, based on SEC data. Figure 1 illustrates a workflow 100 for training the machine learning algorithm in order to generate a model for prediction of at least one property of a fluid at a sample location within a hydrocarbon reservoir, based on SEC data. In the following example, an input dataset 101 is used as a training data set and comprises data relating to a plurality of reservoir samples. The input data set comprises reservoir fluid properties data from a large number of reservoir fluid samples, known as target data. Reservoir samples may be obtained by any method known to the skilled person. It is also commonplace for reservoir samples to be available in storage facilities. The input data set further comprises SEC data for each sample, known as input SEC data. The SEC data comprises values for at least two variables as a function of retention time. The skilled person can obtain these values using standard SEC apparatus and methodology, using appropriate detectors or measurement techniques as required. Example variables are as hereinbefore described. The input SEC data may, for example, comprise absorbance values for at least one wavelength as a function of retention time and / or RI values as a function of retention time. Next, a data augmentation pre-processing step 102 and a model generation are performed, described hereinafter with reference to non-limiting example embodiments, in which a model is generated and validated based on the input data set 101. The input data set 101 is first divided into vectorised data 103 and isoabsorbance data 104. Each sub-dataset is further divided into training data sets 105 and testing data sets 106. The input data set 101 may be curated such that at least the testing data set 106 contains data that spans the various classes of the input data set 101 as a whole (e.g. dry gas reservoirs, wet gas reservoirs, oil reservoirs). Typically, at least 60% of the input data set 101 should be used for training, and at least 20% of the input data set 106 should be used for testing. Common ratios include 50:50, 70:30, 75:25, 80:20, 90:10. However, it will be appreciated that other divisions may be used instead. Generally, the larger the training data set, the more accurate the model will be. However, if too small a test data set is used (or indeed if no test data set is used) then it is not possible to confidently verify the accuracy of the model, e.g. making it difficult to detect an over-fitted model (only accurate for the specific training data). To generate a model, a machine learning algorithm is provided with the training data set 105, and a set of training parameters to control the machine learning algorithm. Example algorithms 107 include Support Vector Machine, Gaussian Process Regression, Random Forest, Universal Kriging, KMean, artificial neural networks or Elastic Net algorithms. However, it will be appreciated that any suitable algorithm may be used and the examples shown in Figure 1 are not limiting. Those operating within this field will be familiar with the procedures for selecting and utilising a machine learning algorithm. Therefore, this will not be discussed in detail. Model validation 108 may then then be performed. During the model validation 108, the model is tested to determine how well it predicts new data that was not used in estimating the model, in order to flag problems such as over fitting or selection bias. Model validation 108 is an optional step and depends of each evaluated model. A comparison between the input data set and the augmented data sets performance is recommended. The best model is selected 109 as the model having the best predictive performance. A testing step 110 may be performed in which the model is tested using the training data set as a whole. A final step 111 is then performed, in which the model is tested using the test data set 106. As discussed previously, this is a curated set of data that is broadly representative of the data as a whole, and was not used during the generation of the model. The model is thus configured such that the model can be used to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location. Figure 2 is a schematic representation of a non-limiting method for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir. In the oil and gas industry, there may be a need to predict at least one property of a fluid at a fluid at a sample location within a hydrocarbon reservoir (201). This might be to assist well placement, reservoir management, production optimization, and / or flow assurance. Example properties are as hereinbefore described. The more properties which are predicted, the greater the industrial value of the method. In the method, in order to facilitate the prediction, it is necessary to receive measured SEC data for the sample location. This may involve recovering a fluid sample from the sample location. This may be achieved through the recovery of drill cuttings (202), associated with which will be small amounts of fluid from the sample location in the hydrocarbon reservoir. Drill cuttings may be recovered using drilling mud, including oilbased mud or water-based mud. The drill cutting may be treated to obtain a sample suitable for SEC (203). This may involve isolating the fluid from the drill cuttings sample, e.g. by sieving, washing with water, and / or extracting using solvent. The sample is then subject to SEC to obtain measured SEC data (204). The SEC data comprises values for at least two variables as a function of retention time. The skilled person can obtain these values using standard SEC apparatus and methodology, using appropriate detectors or measurement techniques as required. Example variables are as hereinbefore described. The measured SEC data may, for example, comprise absorbance values for at least one wavelength as a function of retention time and / or RI values as a function of retention time. In some embodiments, if oil-based mud is used to recover the drill cuttings, the measured SEC data may for example comprise absorbance values for at least one, such a plurality of, for example a range of, UV or UV-visible wavelength(s). This may have the advantage that these values are primarily attributable to the fluid sample itself, rather than any containments from the oil-based mud, the latter of which are somewhat “invisible” at UV or UV-visible wavelengths due to saturated hydrocarbons being the primary constituent. The at least two variables in the input SEC data are the same as the at least two variables in the measured SEC data. For example, if the input SEC data comprises absorbance values for a given wavelength as a function of retention time and RI values as a function of retention time, the measured SEC data also comprises absorbance values for that same wavelength as a function of retention time and RI values as a function of retention time. The values may also be provided as a function of the same retention times. This permits for easier correlation between the measured SEC data and the input SEC data, making for easier and more reliable property prediction. The measured SEC data is then supplied to the computer-based model hereinbefore described, which is configured to predict the at least one property. A property prediction (205) is thus obtained, which may be useful in influencing further engineering decisions at the reservoir site.

Claims

4AMENDED CLAIMS:

1. A method of generating a model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, comprising:5 providing a training data set comprising input data and target data, the input datacomprising input size exclusion chromatography (SEC) data for each of a plurality of sample locations, and the target data comprising the at least one property of the fluid for each of the plurality of sample locations; andinstructing a machine learning algorithm to generate the model using the training10 data set such that the model can be used to predict the at least one property of the fluid at the sample location based on measured SEC data for the sample location,wherein the input SEC data and the measured SEC data each comprise values for a plurality of variables as a function of retention time, including at least absorbance values for a plurality of wavelengths as a function of retention time, the wavelengths15 being selected from UV and UV-visible wavelengths; andwherein the plurality of variables in the input SEC data are the same as the plurality of variables in the measured SEC data.

2. A method according to claim 1, wherein the SEC data comprises absorbance 20 values for a range of wavelengths as a function of retention time, the wavelengths being selected from UV and UV-visible wavelengths.

3. A method according to claim 1 or claim 2, wherein the SEC data comprises refractive index (RI) values as a function of retention time.

254. A method according to any preceding claim, wherein the measured SEC data and the input SEC data are both SEC data obtained using the same solvent system.

5. A method according to any preceding claim, wherein the plurality of sample 30 locations are from a plurality of different hydrocarbon reservoirs.

6. A method according to any preceding claim, wherein the algorithm is selected from Universal Kriging, Support Vector Machine, KMean, Elastic Net, artificial neural network, Random Forest, and Gaussian Regression algorithms.

357. A method according to any preceding claim, wherein the at least one property comprises one or more of:a density of the fluid at the sample location;a gas-oil ratio of the fluid at the sample location;a saturation pressure of the fluid at the sample location;a formation volume factor of the fluid at the sample location;a concentration of C?+ hydrocarbons within the fluid at the sample location;an API of the fluid at the sample location;a stock tank oil viscosity of the fluid at the sample location;a bubble point / dew point pressure of the fluid at the sample location;a SARA profile of the fluid at the sample location;a wax content of the fluid at the sample location;a wax appearance temperature of the fluid at the sample location;a concentration of asphaltene within the fluid at the sample location;a concentration of porphyrin within the fluid at the sample location;a concentration of porphyrin associated with nickel and / or vanadium within the fluid at the sample location; and / oran asphaltene on-set pressure of the fluid at the sample location.

8. A method according to any preceding claim, wherein the method is a method for predicting a plurality of properties of a fluid at a sample location within a hydrocarbon reservoir.

9. A method according to any preceding claim, wherein the input data further comprises non-SEC data and the method comprises instructing a machine learning algorithm to generate the model using the training data set such that the model can be used to predict the at least one property of the fluid at the sample location based on a combination of measured SEC data and measured non-SEC data for the sample location.

10. A computer-based model for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir based on measured SEC data for that sample location, the computer-based model having been generated by a method according to any preceding claim,wherein the measured SEC data comprises measured values for the plurality of variables as a function of retention time.

11. A tangible computer-readable medium storing a computer-based model according to claim 10.

12. A method of predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, the method comprising:receiving measured SEC data for the sample location; andpredicting the value of the property of the fluid at the sample location by supplying the measured SEC data to a computer-based model according to claim 10,wherein the measured SEC data comprises measured values for the plurality of variables as a function of retention time.

13. A method of predicting at least one property of a fluid along a length of a well through a hydrocarbon reservoir, the method comprising:predicting at least one property of a fluid at a plurality of sample locations along a length of a well using a method according to claim 12 for each sample location.

14. A method according to claim 12 or claim 13, wherein receiving measured SEC data for the sample location comprises:recovering a sample from the sample location;subjecting the sample to SEC; andmeasuring values for the plurality of variables as a function of retention time.

15. A method according to any one of claims 12 to 14, wherein receiving measured SEC data for the sample location comprises:recovering a drill cuttings sample from the sample location;optionally, treating the drill cuttings sample to obtain a sample suitable for SEC; andsubjecting the sample to SEC and measuring values for the plurality of variables as a function of retention time.

16. A method according to any one of claims 12 to 15, wherein the drill cuttings are recovered using drilling mud.05 12 2417. A method according to claim 16, wherein the drilling mud is oil-based mud.

18. Use of a computer-based model according to claim 10 to predict at least one5 property of a fluid at a sample location within a hydrocarbon reservoir.

19. A system for predicting at least one property of a fluid at a sample location within a hydrocarbon reservoir, the system comprising:a SEC apparatus; and10 a computer-based model according to claim 10,wherein the SEC apparatus is configured to measure values for the plurality of variables as a function of retention time; andthe computer-based model is configured to be supplied with said values.15 20. A system according to claim 19, wherein the SEC apparatus incorporates or iscoupled to at least one detector,wherein the detector is suitable for measuring a value for at least one of the plurality of variables as a function of retention time.20 21. A system according to claim 20, wherein the SEC apparatus incorporates or iscoupled to an absorbance detector and / or an RI detector.

Images