System for recommending a reservoir monitoring plan and flow testing plan of wells

Abstract

The present disclosure relates to computer-implemented method (300) for training a model for predicting a flow test of a well as well as a corresponding model for predicting a flow test of a well and a method (100) for determining a flow test plan. In addition, a computer-implemented method (400) for training a model for predicting saturation and pressure within a reservoir is disclosed as well as a corresponding model for predicting saturation and pressure within a reservoir and a method (200) for determining a reservoir monitoring plan. Finally, a corresponding computer program and data- processing system (500) is disclosed.

Description

[0001] SYSTEM FOR RECOMMENDING A RESERVOIR MONITORING PLAN AND FLOW TESTING PLAN OF WELLS

[0002] Field of the invention

[0003] The present invention relates to a computer-implemented method for training a model for predicting a flow test of a well and a corresponding model for predicting a flow test. In addition, a computer-implemented method for determining a flow test plan is disclosed. Additionally, a computer-implemented method for training a model for predicting saturation and pressure with a reservoir and a corresponding model for predicting saturation and pressure within a reservoir is disclosed as well as corresponding computer-implemented method for determining a reservoir monitoring plan. Finally, the present invention relates to a corresponding computer program and data-processing system.

[0004] Background

[0005] Efficient reservoir management includes continuous optimization of field development opportunities (e.g., with respect to estimated oil remaining (EOR) or Infill), identifying or avoiding abnormal events (e.g., inactive strings) while at the same time meeting certain reservoir management objectives (e.g., workovers or rate changes).

[0006] The corresponding decision making thus requires a solid data basis for assessing the reservoir state. However, acquiring reservoir data (e.g., surface and wellbore data for multiple wells of the reservoir) can become quite challenging in mature assets comprising a large number of wells, because such data acquisition implies high operating effort (e.g., with respect to required measurement equipment, measurement specialists etc.). These factors limit the data acquisition, which is why a tradeoff between data maturity and available resources is to be found.

[0007] Against this background, there is a need for methods and system for improving reservoir monitoring and flow testing of wells.

[0008] Summary

[0009] The above-mentioned problem is at least partly solved by the aspects presented in the accompanying independent and dependent claims. Combinations of features from the dependent claims may be combined with features of the independent claims as appropriate and not merely as explicitly set out in the claims.

[0010] An aspect of the present disclosure relates to a computer-implemented method for training a model for predicting a flow test of a well. The method may comprise the step of obtaining a data sample. The data sample may comprise a first set of parameters associated with at least a first flow test of the well and / or a reference parameter associated with a second flow test of the well. The method may comprise the step of providing the first set of parameters associated with at least the first flow test of the well as input to the model. The method may comprise the step of receiving a predicted parameter associated with the second flow test of the well as output from the model. The method may comprise the step of determining an error between the predicted parameter and the reference parameter. The method may comprise the step of adjusting one or more training parameters of the model at least based on the error between the predicted parameter and the reference parameter.

[0011] This way the model is able to learn the relationship between conducted flow tests with respect to reference parameters. Accordingly, when using the model for prediction, the model is able to simulate a flow test and to predict the corresponding parameter.

[0012] In a further aspect, the first flow test of the well may be associated with a first time stamp. The second flow test of the well may be associated with a second time stamp. The first time stamp may be smaller than the second time stamp.

[0013] This way the model is able to learn temporal aspects between conducted flow tests, e.g., between the first flow test and the second (subsequent) flow test This way the model is able to simulate a flow test for a given time stamp in the future and to predict the corresponding parameter.

[0014] In a further aspect, each parameter of the first set of parameters associated with the first flow test may be a measurement value resulting from the first flow test of the well.

[0015] In a further aspect, the reference parameter associated with the second flow test of the well may be a measurement value resulting from the second flow test of the well. The predicted parameter maybe an estimation of the measurement value.

[0016] Using real measurement parameters of a conducted flow tests as training data instead of for example synthetic data provides a higher data quality. Accordingly, the model is better able to learn and thus represent the real behavior of the well. In a further aspect, the first set of parameters may comprise at least one of: a choke opening of the well (CHK), a separator pressure of the well (PSEP), a well-head pressure (WHP) of the well, a well-head temperature (WHT) of the well, a water indicative quantity (e.g., one or more of a water-cut (WCT) of the well, a water oil ratio or a water rate), a gas indicative quantity (e.g., one or more of a gas-oil ratio (GOR) of the well, a water gas ratio or a water rate), an oil rate (e.g., an amount of oil produced (OIL) of the well), an amount of gas-lift rate injected (GLR), a rotating speed of a pump (RPM) or any combination thereof.

[0017] In a further aspect, the reference parameter and the predicted parameter may be one of: GOR, WCT, WHP, WHT or OIL.

[0018] It was found that the presented parameters are best to precisely represent the well performance and are thus selected as the parameters which are measured during flow tests. However, the present method(s) are not limited to these parameter as they are also applicable to other (direct or indirect) measurement parameters such as downhole pressure, downhole temperature, flowline pressure or differential pressure across choke or pipeline.

[0019] Another aspect of the present disclosure relates to a model for predicting a flow test of a well. The model maybe trained to receive as input a time stamp for predicting a flow test of the well. The model maybe trained to output a predicted parameter associated with the flow test of the well at the time stamp.

[0020] In a further aspect, the model may be trained according to a method for training a model for predicting a flow test of a well according to the aspects of the present disclosure.

[0021] In a further aspect, the model may be an autoregressive integrated moving average with exogenous inputs (ARIMAX) model.

[0022] In a further aspect, the training parameters of the model may comprise one or more covariates of the ARIMAX model.

[0023] In a further aspect, the model may comprise a plurality of sub models. Each sub model of the plurality of sub models may be trained according to the method for training a model for predicting a flow test of a well according to the aspects of the present disclosure. Each sub model of the plurality of sub models may be trained to output a different predicted parameter. Defining the model as a sort of ensemble of sub models in which each sub model is trained to predict a different parameter instead of having one large model predicting a plurality of parameter decreases complexity of the training.

[0024] Another aspect of the present disclosure relates to a computer-implemented method for determining a flow test plan. The method may comprise determining a first flow test frequency of a plurality of flow test frequencies for a well. The method may comprise providing to a model a first time stamp as input to predict a first flow test at the first time stamp. The method may comprise receiving from the model a first predicted parameter associated with the first flow test at the first time stamp. The method may comprise determining a second flow test frequency of the plurality of flow test frequencies for the well based on the first predicted parameter.

[0025] Determining the flow test plan based on a data-driven decision as provided by the first predicted parameter allows for a more sophisticated decision whether the flow test frequency is to be adjusted for the well without compromising well health performance.

[0026] In a further aspect, the model may be a model for predicting a flow test of a well according to the aspects of the present disclosure. The model may be trained according to the method for training a model for predicting a flow test of a well according to the aspects of the present disclosure.

[0027] In a further aspect, determining the first flow test frequency of the well may comprise assigning a test criticality level of a plurality of test criticality levels to the well. Additionally, determining the first flow test frequency of the well may comprise determining the first flow test frequency based on the test criticality level.

[0028] In a further aspect, the plurality of test criticality levels may comprise a low test criticality level, a medium test criticality level and / or a high test criticality level. The plurality of flow test frequencies may comprise a low testing frequency, preferably one test every 6o days, a medium testing frequency, preferably one test every 30 days, and a high testing frequency, preferably one test every 15 days.

[0029] In a further aspect, assigning the test criticality level to the well may comprise determining that a set of parameters associated with the first flow test fulfills a first predefined set of rules and assigning a low test criticality level as the test criticality level. Additionally or alternatively, assigning the test criticality level to the well may comprise determining that the set of parameters associated with the first flow test fulfills a second predefined set of rules and assigning a medium test criticality level as the test criticality level. Additionally or alternatively, assigning the test criticality level to the well may comprise determining that the set of parameters associated with the first flow test fulfills a third predefined set of rules and assigning a high test criticality level as the test criticality level.

[0030] In a further aspect, determining the first flow test frequency based on the test criticality level may comprise determining that the first flow test frequency is a low testing frequency if the assigned test criticality level is low. Additionally or alternatively, determining the first flow test frequency based on the test criticality level may comprise determining that the first flow test frequency is a medium testing frequency if the assigned test criticality level is medium. Additionally or alternatively, determining the first flow test frequency based on the test criticality level may comprise determining that the first flow test frequency is a high testing frequency if the assigned test criticality level is high.

[0031] Determining the first flow test frequency according to one of the above-mentioned aspects provides an efficient way of determining a test frequency for the well without requiring further information about the well than the actual information such as set of parameters of the well which are already available. Accordingly, a computational less expensive way is presented to determine a first estimation of the test frequency.

[0032] In a further aspect, determining the second flow test frequency may comprise determining a first error between the first predicted parameter associated with the first flow test at the first time stamp and a first reference parameter.

[0033] Using the prediction error as an indication whether the flow test frequency is to be adjusted is an efficient way of determining whether the well shall be tested as frequent as determined indicated by the first flow test frequency. In case the prediction error is high, the model is not able to predict the well behavior which may indicate that the data input is inconsistent with the model (e.g., additional flowtest(s) are required).. Thus, the flow test frequency may be adjusted accordingly.

[0034] In a further aspect, the method may further comprise providing to the model at least a second time stamp as input to predict a second flow test at the second time stamp. The second time stamp may be smaller than the first time stamp. The method may further comprise receiving from the model a second predicted parameter associated with the second flow test at the second time stamp. Additionally, determining the second flow test frequency may comprise determining a second error between the second predicted parameter associated with the second flow test at the second time stamp and a second reference parameter.

[0035] Extending the time range for which the prediction is to be made allows to better assess whether the flow test requirements of the well can be reduced.

[0036] In a further aspect, determining the second flow test frequency may further comprise determining that the first error, the second error and / or a combination of the first error and the second error is above or equal to a predefined error threshold and increase the flow test frequency. Additionally or alternatively, determining the second flow test frequency may comprise determining that the first error, the second error and / or a combination of the first error and the second error is below the predefined error threshold and decrease the flow test frequency.

[0037] Providing a reasonable predefined error threshold allows to efficiently determine whether an increase or decrease of testing frequency is required. This way an optimized tradeoff between well testing resources and well assessment accuracy is provided.

[0038] In a further aspect, the set of parameters associated with the first flow test of well and / or a set of parameters associated with the second flow test of the well may be unknown to the model.

[0039] When the model predicts parameters for an unknown situation (i.e., flow tests not used during model training), it can be ensured that the corresponding prediction error is less biased than when the model was trained on the data. When the model predicts parameters for an unknown situation, it is known as the blind test prediction error.

[0040] In a further aspect, determining the second flow test frequency may be further based on determining an urgency for performing a flow test on the well.

[0041] In a further aspect, determining the urgency may comprise at least one of: determining the blind test prediction error, determining a flow test compliance of the well, determining an uptime of the well, determining a well priority of the well, determining an allocated production rate of the well or any combination thereof.

[0042] Another aspect of the present disclosure relates to a computer-implemented method for training a model for predicting saturation and pressure within a well-reservoir pair.

[0043] The method may comprise the step of providing a representation of the reservoir as input to the model. The representation may include measurement data of at least one or more of the saturation, one or more flow rates, one or more temperatures and the pressure(s) within the reservoir over a period of time. The method may comprise the step of receiving as output from the model a prediction of measurement data for a point in time within the period of time. The method may comprise the step of adjusting one or more training parameters of the model based on an error between the prediction of the measurement data for the point in time and measurement data at the point in time.

[0044] By training the model using measurement data over the period of time, the model once trained is able to predict measurement data and thus the saturation and pressure within a reservoir for a given - potentially future (e.g., by extrapolation) or past (e.g., by interpolation)- time point.

[0045] In a further aspect, the representation of the reservoir may be a spatiotemporal matrix.

[0046] In a further aspect, a spatial dimension of the spatiotemporal matrix may represent a plurality of locations within the reservoir.

[0047] In a further aspect, the temporal dimension may represent for each location of the plurality of locations within the reservoir an associated timeseries.

[0048] In a further aspect, each timestamp of a time series associated with a location of the plurality of locations within the reservoir maybe associated with measurement data.

[0049] Implementing the representation of the reservoir as a spatiotemporal matrix as described in one of the above-mentioned aspects provides the advantages that the overall development of the reservoir can be accurately described while providing a data structure which is efficiently processable. For example, entries of the matrix are directly accessible via its corresponding index, whereas the index is also an efficient way of representing the spatial and temporal characteristics of a location.

[0050] In a further aspect, the associated measurement data may include at least one of: a saturation measured at the location at the timestamp, a flow rate measured at the location at the timestamp, a temperature measured at the location at the timestamp and / or a pressure measured at the location at the timestamp. The associated measurement data maybe direct measurement data (i.e., directly measured values) or indirect measurement data (i.e., derived from other direct measurement data). For example, a saturation and / or a flow rate maybe regarded as indirect measurement data as they can be derived from other direct measurement data (e.g., electric, magnetic and / or mechanical measurements).

[0051] Rather than using synthetic data for training, real measurement data realistically represents the state (e.g., saturation and pressure) of a reservoir.

[0052] In a further aspect, adjusting one or more training parameters of the model may comprise adjusting the one or more training parameters of the model based on an error between a prediction of measurement data for a given timestamp and the measurement data associated with the given time stamp.

[0053] Another aspect of the present disclosure relates to a model for predicting saturation and pressure within a reservoir. The model may be trained to receive as input a time stamp for predicting the saturation and pressure within the reservoir at the time stamp. The model may be trained to output a prediction of measurement data of saturation and pressure within the reservoir at the time stamp.

[0054] In a further aspect, the model may be trained according to a method for training a model for predicting saturation and pressure within a reservoir according to the aspects of the present disclosure.

[0055] In a further aspect, the model maybe based on non-linear approximators (NLAs).

[0056] In a further aspect, the model may include an additional output function. The additional output function may be configured to output a confidence interval for the prediction of measurement data of saturation and pressure within the reservoir at the time stamp.

[0057] The confidence interval allows to assess the uncertainty of the model prediction, which can serve as valuable information when using the prediction for decision purposes such as determining a reservoir monitoring plan.

[0058] Another aspect of the present disclosure relates to a computer-implemented method for determining a reservoir monitoring plan (RMP). The method may comprise the step of selecting a plurality of well-reservoir pairs for which the RMP is to be determined. The method may comprise the step of determining for each well-reservoir pair of the plurality of well-reservoir pairs a ranking indicator associated with the well-reservoir pair. The method may comprise the step of determining the RMP for the plurality of well-reservoir pairs according to the ranking indicator associated with each wellreservoir pair of the plurality of well-reservoir-pairs.

[0059] The ranking indicator allows to efficiently rank the well-reservoir pairs such that pairs with a high rank are prioritized over pairs with a low rank with respect to value of monitoring. This way resource limitations maybe overcome.

[0060] In a further aspect, determining the ranking indicator associated with the wellreservoir pair may comprise providing a time stamp as input to a model for predicting saturation and pressure within the reservoir at the time stamp. Additionally, determining the ranking indicator associated with the well-reservoir pair may comprise receiving as output from the model, a prediction of measurement data of saturation and pressure within the reservoir at the time stamp. Additionally, determining the ranking indicator associated with the well-reservoir pair may comprise assigning the ranking indicator to the well-reservoir pair based on the prediction of measurement data of saturation and pressure within the reservoir at the time stamp.

[0061] In a further aspect, the method may further comprise deriving an amount of expected remaining oil (ERO) from the prediction of measurement data of saturation and pressure within the reservoir at the time stamp. Additionally, assigning the ranking indicator may be further based on the amount of ERO.

[0062] In a further aspect, assigning the ranking indicator may comprise assigning a high ranking indicator if the amount of ERO is larger than or equal to a predefined ERO threshold. Additionally or alternatively, assigning the ranking indicator may comprise assigning a low ranking indicator if the amount of ERO is smaller than the predefined ERO threshold.

[0063] Ranking the well-reservoir pairs according to their predicted or expected properties may allow to prioritize well-reservoir pairs having a higher amount of ERO.

[0064] Additionally or alternatively, the predicted properties for example the pressure maybe used to assess the fragility of a well-reservoir pair and to prioritize the more fragile well-reservoir pairs which improves the production safety.

[0065] In a further aspect, the output from the model further may comprises a confidence interval for the prediction of measurement data of saturation and pressure within the reservoir at the time stamp. The method may further comprise calculating an uncertainty value based on the confidence interval. Assigning the ranking indicator may be further based on the uncertainty value.

[0066] In a further aspect, the uncertainty value may be a difference between an upper limit of the confidence interval and a lower limit of the confidence interval.

[0067] In a further aspect, assigning the ranking indicator may comprise assigning a high ranking indicator if the uncertainty value is larger than or equal to a predefined uncertainty threshold. Additionally or alternatively, assigning the ranking indicator may comprise assigning a low ranking indicator if the uncertainty value is smaller than the predefined uncertainty threshold.

[0068] Using the uncertainty of the model prediction for ranking improves the decision quality. As a high uncertainty indicates that new information about the well-reservoir pair is more valuable than if the uncertainty is low, the corresponding well-reservoir- pair for which the uncertainty was high may be prioritized.

[0069] In a further aspect, the ranking indicator may be determined based on a combination of the amount of ERO and the uncertainty value.

[0070] In a further aspect, the model maybe a model for predicting saturation and pressure within a reservoir according to the aspects of the present disclosure.

[0071] Another aspect of the present disclosure relates to a computer program comprising instructions which when executed by a computer causes the computer to execute the method(s) and / or the model(s) according to the aspects of the present disclosure.

[0072] Another aspect of the present disclosure relates to a data-processing system comprising means for executing the method(s) and / or model(s) according to the aspects of the present disclosure.

[0073] Brief description of the figures

[0074] Various aspects of the present invention are described in more detail in the following by reference to the accompanying figures without the present invention being limited to the embodiments of these figures.

[0075] Fig. 1 illustrates an overview of a computer-implemented method for determining a flow test plan according to embodiments of the present invention. Fig. 2 illustrates an overview of a computer-implemented method for determining a reservoir monitoring plan according to embodiments of the present invention.

[0076] Fig. 3 illustrates an overview of a computer-implemented method for training a model for predicting a flow test of a well according to embodiments of the present invention.

[0077] Fig. 4 illustrates an overview of a computer-implemented method for training a model for predicting saturation and pressure within a reservoir according to embodiments of the present invention.

[0078] Fig. 5 illustrates an overview of a data-processing system according to the embodiments of the present invention.

[0079] Detailed description

[0080] In the following, certain aspects of the present invention are described in more detail.

[0081] Fig. 1 illustrates an overview of a computer-implemented method 100 for determining a flow test plan according to embodiments of the present invention.

[0082] In step no, a first flow test frequency of a plurality of flow test frequencies for a well is determined. This first step may be seen as a step for determining a baseline frequency which is used for subsequent evaluation with respect to recommended frequency changes.

[0083] Determining the first flow test frequency of the well may comprise assigning a test criticality level of a plurality of test criticality levels to the well. Additionally, determining the first flow test frequency of the well may comprise determining the first flow test frequency based on the test criticality level.

[0084] The plurality of test criticality levels may comprise a low test criticality level, a medium test criticality level and / or a high test criticality level. The plurality of flow test frequencies may comprise a low testing frequency, preferably one test every 6o days, a medium testing frequency, preferably one test every 30 days, and a high testing frequency, preferably one test every 15 days. Accordingly, the test criticality level may indicate how often a well should be tested. For example, a low test criticality level may indicate that a well should be tested less frequent than a well for which a medium or high test criticality level is indicated. Assigning the test criticality level to the well may comprise determining that a set of parameters associated with the first flow test fulfills a first predefined set of rules and assigning a low test criticality level as the test criticality level. Additionally or alternatively, assigning the test criticality level to the well may comprise determining that the set of parameters associated with the first flow test fulfills a second predefined set of rules and assigning a medium test criticality level as the test criticality level. Additionally or alternatively, assigning the test criticality level to the well may comprise determining that the set of parameters associated with the first flow test fulfills a third predefined set of rules and assigning a high test criticality level as the test criticality level. The first flow test may test may be the latest flow test performed on the well (i.e., the set of parameters may be the latest information about the well).

[0085] The first, second and / or third predefined set of rules may be defined by a predefined protocol (e.g., a quality standard specification or a common best practice for test frequencies). The first predefined set of rules may require:

[0086] GOR < fcoR x Rs W CT < tjycj’ wherein GOR is the GOR of the well according to the latest measurement (i.e., the GOR comprised within the set of parameters), / GOR is a factor (e.g., 1.15), Rs is a reference value referring to the initial GOR of the well (i.e., the GOR determined by an initial measurement of the well, or assigned by offset well analogues, or calculated from industry standard methods), WCT is the WCT of the well according to the latest measurement (i.e., the WCT comprised within the set of parameters) and twcris a WCT threshold (e.g., 5%). Should the set of parameters fulfill these rules, than a low test criticality level may be assigned as the test criticality level of the well (i.e., the well may be seen as non-critical which respect to necessity of a next flow test).

[0087] The second predefined set of rules may require:

[0088] GOR > fcoR x Rs WCT > tjycj’ wherein the same notations as for the first predefined set of rules applies, wherein the values of / GOR and / or twcrmay be the same or different. Should the set of parameters fulfill these rules, than a medium test criticality level may be assigned as the test criticality level of the well (i.e., the well may be seen as problematic which respect to necessity of a next flow test).

[0089] The third predefined set of rules may require:

[0090] GOR > fcoR x Rs WCT > tjycj’ wherein the same notations as for the first and / or second predefined set of rules applies, wherein the values of / GOR and / or twcr may be the same or different (e.g., it may be preferred to define larger values such as 2.5 for / coR and 15% for twcr). Should the set of parameters fulfill these rules, than a high test criticality level may be assigned as the test criticality level of the well (i.e., the well maybe seen as critical which respect to necessity of a next flow test).

[0091] Determining the first flow test frequency based on the test criticality level may comprise determining that the first flow test frequency is a low testing frequency if the assigned test criticality level is low. Additionally or alternatively, determining the first flow test frequency based on the test criticality level may comprise determining that the first flow test frequency is a medium testing frequency if the assigned test criticality level is medium. Additionally or alternatively, determining the first flow test frequency based on the test criticality level may comprise determining that the first flow test frequency is a high testing frequency if the assigned test criticality level is high.

[0092] Alternatively, determining the first flow test frequency of a plurality of flow test frequencies for the well may comprise accessing information about the first flow test frequency of a plurality of flow test frequencies. Accessing may refer to receiving the information from a data source (e.g., a date base which stores the first flow test frequency). In this case, the first flow test frequency may have already been determined and thus only retrieving of said information is required.

[0093] In step 120, a first time stamp may be provided as input to a model to predict a first flow test at the first time stamp.

[0094] In addition, at least a second time stamp may be provided to the model as input to predict a second flow test at the second time stamp. The second time stamp may be smaller than the first time stamp. The set of parameters associated with the first flow test of well and / or a set of parameters associated with the second flow test of the well may be unknown to the model. Unknown means that the corresponding set(s) of parameters were not used during training (i.e. , a blind testing is performed). It may be possible that further sets of parameters are unknown to the model. For example, there may be 15 flow tests performed on the well (i.e., an amount of flow tests performed for the well is 15). In the example, that only the first flow test at the first time stamp is predicted, the model may have been trained on the flow tests 1 to 14 and the 15th(i.e., the first flow tests and also the latest) is unknown to the model. In the example, that the first flow test at the first time stamp and the second flow test at the second time stamp are predicted, the model may have been trained on the flow tests 1 to 13 and the 14th(i.e., the second flow tests and also the second latest) and the 15th(i.e., the first flow tests and also the latest) is unknown to the model. Accordingly, the second time stamp maybe smaller (i.e., older) than the first time stamp.

[0095] The model maybe a model for predicting a flow test of a well according to the aspects of the present disclosure. The model may be trained according to the method for training a model for predicting a flow test of a well according to the aspects of the present disclosure. Examples for the model and / or the training method used are explained with respect to Fig. 3.

[0096] This step 120 of method too maybe skipped if it is determined that an amount of flow tests performed for the well is below a required flow test count threshold. For example, the required flow test count threshold maybe 15. In this case, if less than inflow tests were performed on the well, the method step 120 and the method step 130 maybe skipped (i.e., are not executing and / or are not part of the method too). The reason for this is that it is assumed that the model has not enough reliable data available for training (e.g., not enough sets of parameters associated with the flow tests of the well) to provide an efficient (i.e., accurate and reliable) prediction.

[0097] In step 130, a first predicted parameter associated with the first flow test at the first time stamp is received from the model. If method step 120 was skipped, method step 130 may also be skipped.

[0098] In case that at least a second time stamp was provided in step 120, a second predicted parameter associated with the second flow test at the second time stamp is received from the model. The model maybe a model for predicting a flow test of a well according to the aspects of the present disclosure. The model may be trained according to the method for training a model for predicting a flow test of a well according to the aspects of the present disclosure. Examples for the model and / or the training method used are explained with respect to Fig. 3.

[0099] Accordingly, it may also be possible that the model outputs more than one predicted parameter associated with the first, the second or further flow tests. In this case, the model maybe seen as an ensemble of a plurality of sub models (i.e., the model comprises the plurality of sub models). Each sub model maybe a model for predicting a flow test of a well according to the aspects of the present disclosure (e.g., as explained with respect to Fig. 3). Each sub model may be trained according to the training method for training a model for predicting a flow test of a well according to the aspects of the present disclosure. In these aspects, each sub model predicts a different predicted parameter. I.e., a first sub model of the plurality of sub models maybe trained to output a first predicted parameter and a second sub model of the plurality of sub models may be trained to output a second predicted parameter, wherein the first predicted parameter and the second predicted parameter are different. As a result, a plurality of predicted parameters associated with each flow test (e.g., the first flow test) at the corresponding time stamp (e.g., the first time stamp) may be received and further processed as explained in step 140 with respect to the first predicted parameter.

[0100] In step 140, a second flow test frequency of the plurality of flow test frequencies is determined for the well based on the first predicted parameter. If the method steps 120 and 130 were skipped, determining the second flow test frequency may comprise increasing the flow test frequency. For example, if the first flow test frequency is determined as being low, the second flow test frequency may be increased and thus determined to be medium. If the first flow test frequency is determined as being high a further increase may not be necessary and / or possible.

[0101] Determining the second flow test frequency may comprise determining a first error between the first predicted parameter associated with the first flow test at the first time stamp and a first reference parameter. The error(s) as explained within this disclosure may refer to a mean-squared error, a mean absolute error or any other suitable arithmetic way of determining an error.

[0102] In case a plurality of predicted parameters associated with the first flow test is received, the first error maybe determined between each predicted parameter of the plurality of predicted parameters and the corresponding reference parameter. For example, if a predicted parameter of the plurality of predicted parameters is WHP, an error may be determined based on the predicted WHP and the reference WHP. If another predicted parameter of the plurality of predicted parameters is WCT, an error may be determined based on the predicted WCT and the reference WCT. The first error used for further processing (e.g., used for determining the second flow test frequency) may then be determined inter alia based on a combination of the errors determined between each predicted parameter of the plurality of predicted parameters and their corresponding reference parameters. The combination maybe a mean, a median, a minimum value, a maximum value or any other suitable arithmetic comparison. For example, the combination may be a mean of these errors (i.e., the first error may be the mean error of the error between the predicted WCT and the reference WCT and the error of the predicted WHP and the reference WHP).

[0103] In case a second predicted parameter is received in step 130, determining the second flow test frequency may comprise determining a second error between the second predicted parameter associated with the second flow test at the second time stamp and the second reference parameter.

[0104] In case a plurality of predicted parameters associated with the second flow test is received, the second error maybe determined between each predicted parameter of the plurality of predicted parameters and the corresponding reference parameter. For example, if a predicted parameter of the plurality of predicted parameters is WHP, an error may be determined based on the predicted WHP and the reference WHP. If another predicted parameter of the plurality of predicted parameters is WCT, an error maybe determined based on the predicted WCT and the reference WCT. The second error used for further processing (e.g., used for determining the second flow test frequency) may then be determined inter alia based on a combination of the errors determined between each predicted parameter of the plurality of predicted parameters and their corresponding reference parameters. The combination may be a mean, a median, a minimum value, a maximum value or any other suitable arithmetic comparison. For example, the combination maybe a mean of these errors (i.e., the second error may be the mean error of the error between the predicted WCT and the reference WCT and the error of the predicted WHP and the reference WHP).

[0105] Determining the second flow test frequency may further comprise determining that the first error, the second error and / or a combination of the first error and the second error is above or equal to a predefined error threshold and increase the flow test frequency (e.g., increase the frequency from low to medium, from medium to high, or from low to high). Additionally or alternatively, determining the second flow test frequency may comprise determining that the first error, the second error and / or a combination of the first error and the second error is below the predefined error threshold and decrease the flow test frequency (e.g., decrease the frequency from medium to low, from high to medium, or from high to low). The combination may be a mean, a median, a minimum value, a maximum value or any other suitable arithmetic comparison.

[0106] Even though not explicitly shown in Fig. 1, the method may comprise further steps. For example, the method may comprise determining an urgency for performing a flow test on the well. The urgency may for example indicate a necessity for performing the flow test on the well.

[0107] Determining the urgency may comprise at least one of: determining a blind test prediction error of the well, determining a flow test compliance of the well, determining an uptime of the well, determining an allocated production rate of the well, determining a well priority or any combination thereof. Additionally, or alternatively, the method may further comprise assigning a rank to the well based on the urgency.

[0108] The uptime of the well may be determined based on a ratio between flowing days of the well and monthly calendar days. An example for such a ratio maybe Uptime =

[0109] Flowing days

[0110] Monthly Calender days'

[0111] The compliance of the well may be determined based on a ratio between a preset amount of days required between tests and an amount of days between a current day and the days of the last test (i.e., the first flow test). An example for such a ratio may be , . required days between tests

[0112] Compliance = - . actual days since last test

[0113] The allocated production rate of the well may be determined based on a ratio of the production rate of the well and the overall production rate of all wells in production. An example tor such a ratio may be Allocated Rate = - - - . overall production

[0114] The urgency may be determined based on a combination of the uptime of the well, the allocated production rate of the well, the compliance of the well, a priority value of the well, or any combination thereof. The priority value may be based on or refer to a ranking indicator of a reservoir associated with the well. Additionally or alternatively, the priority value may be based on or refer to a blind test prediction error of a well. For example, if the bind test prediction error of the well is above a certain error threshold, then the priority value is equal to one. For example, if the blind test prediction error of the well is below a certain error threshold then then the priority value is greater than one. The ranking indicator may be determined as part of a method for determining a reservoir monitoring plan (RMP) according to the aspects of the present disclosure

[0115] (e.g., as explained with respect to Fig. 2.). The combination may for example be a ratio of the uptime, the allocated rate and the compliance and priority. An example of such a

[0116] Uptime xAllocated Rate ratio maybe Urgency = Compliance xPriority

[0117] The method may further comprise executing the steps of the method 110 for at least a second well (i.e., a well different than the one referred to in the above explained steps of method too) and ranking the well and the second well according to their urgency. In an example, in which the urgency of the second well may be larger than the urgency of the well, the second well may ranked higher than the well. As a result, the flow test plan may indicate to prioritize the second well over the well for the next flow test. This way limitation of available resources is overcome.

[0118] Once a new flow test was conducted for the well, the model for predicting a flow test of a well may be retrained using the newly available information using the training method for training a model for predicting a flow test according to the aspects of the present disclosure (e.g., Fig. 3).

[0119] Fig. 2 illustrates an overview of a computer-implemented method for determining a reservoir monitoring plan (RMP) according to embodiments of the present invention.

[0120] In step 210, a plurality of well-reservoir pairs for which the RMP is to be determined is selected. Selecting the plurality of well-reservoir pairs may be based on one or more criteria. A criterion maybe that a reservoir is only selected if required resources (e.g., monitoring equipment required for monitoring the reservoir) are available. Another criterion may be that a reservoir is only selected if a decision support is required (e.g., saturation uncertainty reduction is required for developing the reservoir). Should the criteria not be fulfilled, the reservoir may not be selected as one of the plurality of reservoirs for which the RMP is to be determined.

[0121] In step 220, for each well-reservoir pair of the plurality of well-reservoir pairs a ranking indicator associated with the well-reservoir pair is determined. Determining the ranking indicator associated with the well-reservoir pair may comprise providing a time stamp as input to a model for predicting saturation and pressure within the reservoir at the time stamp. Additionally, determining the ranking indicator associated with the reservoir may comprise receiving as output from the model, a prediction of measurement data of saturation and pressure within the reservoir at the time stamp. Additionally, determining the ranking indicator associated with the reservoir may comprise assigning the ranking indicator to the reservoir based on the prediction of measurement data of saturation and pressure within the reservoir at the time stamp.

[0122] An amount of expected remaining oil (ERO) may be derived from the prediction of measurement data of saturation and pressure within the reservoir at the time stamp. Additionally, assigning the ranking indicator may be further based on the amount of ERO.

[0123] Assigning the ranking indicator may comprise assigning a high ranking indicator if the amount of ERO is larger than or equal to a predefined ERO threshold. Additionally or alternatively, assigning the ranking indicator may comprise assigning a low ranking indicator if the amount of ERO is smaller than the predefined ERO threshold. This way reservoirs having a small ERO will be lower ranked than reservoirs having a high ERO. Accordingly, the RMP may indicate that the reservoirs having a higher ERO shall be prioritized over the reservoirs having a lower ERO.

[0124] The output from the model further may comprises a confidence interval for the prediction of measurement data of saturation and pressure within the reservoir at the time stamp. An uncertainty value maybe calculated based on the confidence interval. Assigning the ranking indicator may be further based on the uncertainty value.

[0125] The uncertainty value may be a difference between an upper limit of the confidence interval and a lower limit of the confidence interval.

[0126] Assigning the ranking indicator may comprise assigning a high ranking indicator if the uncertainty value is larger than or equal to a predefined uncertainty threshold.

[0127] Additionally or alternatively, assigning the ranking indicator may comprise assigning a low ranking indicator if the uncertainty value is smaller than the predefined uncertainty threshold. This is because a high uncertainty value may indicate that the value of additional / new information about the reservoir maybe high (e.g., with respect to keeping the reservoir state updated), wherein a low uncertainty value may indicate that the value of additional / new information about the reservoir may be low.

[0128] The ranking indicator may be determined based on a combination of the amount of ERO and the uncertainty value. For example, a reservoir having a high amount of ERO and a high uncertainty value may be assigned with a higher uncertainty value than a reservoir having a low amount of ERO and a low uncertainty value.

[0129] The model maybe a model for predicting saturation and pressure within a reservoir according to the aspects of the present disclosure. The model may be trained according to a method for training a model for predicting saturation and pressure within a reservoir according to the aspects of the present disclosure. Examples for the model and / or the training method used are explained with respect to Fig. 4.

[0130] In step 230, the RMP for the plurality of reservoirs is determined according to the ranking indicator associated with each well-reservoir pair of the plurality of wellreservoir pairs. Accordingly, the RMP may determine which well-reservoir pair is to be prioritized over others with respect to monitoring (e.g., collecting new / additional information about the reservoir).

[0131] Fig. 3 illustrates an overview of a computer-implemented method 300 for training a model for predicting a flow test of a well according to embodiments of the present invention.

[0132] In step 310, a data sample is obtained. The data sample may comprise a first set of parameters associated with at least a first flow test of the well. The data sample may comprise a reference parameter associated with a second flow test of the well.

[0133] The first flow test of the well may be associated with a first time stamp. The second flow test of the well may be associated with a second time stamp. The first time stamp may be smaller than the second time stamp.

[0134] Each parameter of the first set of parameters associated with the first flow test may be a measurement value resulting from the first flow test of the well. The reference parameter associated with the second flow test of the well may be a measurement value resulting from the second flow test of the well.

[0135] The first set of parameters may comprise at least one of: a choke opening of the well (CHK), a separator pressure of the well (PSEP), a well-head pressure (WHP) of the well, a well-head temperature (WHT) of the well , a water-cut (WCT) of the well, a gasoil ratio (GOR) of the well, an amount of oil produced (OIL) of the well, an amount of gas-lift rate injected (GLR), a rotating speed of the pump (RPM) or any combination thereof.

[0136] In step 320, the first set of parameters associated with at least the first flow test of the well is provided as input to the model.

[0137] In step 330, a predicted parameter associated with the second flow test of the well is received as output from the model. The predicted parameter may be an estimation of the measurement value (i.e., the measurement value resulting from the second flow test of the well, which is also referred to as reference parameter). The reference parameter and the predicted parameter maybe one of: GOR, WCT, WHP, WHT or OIL.

[0138] In step 340, an error between the predicted parameter and the reference parameter is determined.

[0139] In step 350, one or more training parameters of the model are adjusted at least based on the error between the predicted parameter and the reference parameter.

[0140] As a result of the method 300, a model for predicting a flow test of a well may be created. The model may be trained to receive as input a time stamp for predicting a flow test of the well. The model may be trained to output a predicted parameter associated with the flow test of the well at the time stamp (e.g., past or future time stamp). Creating the model is done by training the model according to a method for training a model for predicting a flow test of a well according to the aspects of the present disclosure such as the method 300.

[0141] The model may be an autoregressive integrated moving average with exogenous inputs (ARIMAX) model. The training parameters of the model may comprise one or more covariates of the ARIMAX model.

[0142] The model may comprise a plurality of sub models. Each sub model of the plurality of sub models may be trained according to the method for training a model for predicting a flow test of a well according to the aspects of the present disclosure such as the method 300. Each sub model of the plurality of sub models maybe trained to output a different predicted parameter. Fig. 4 illustrates an overview of a computer-implemented method 400 for training a model for predicting saturation and pressure within a reservoir according to embodiments of the present invention.

[0143] In step 410, a representation of the reservoir is provided as input to the model. The representation may include measurement data of the saturation and the pressure within the reservoir over a period of time.

[0144] The representation of the reservoir maybe a spatiotemporal matrix. A spatial dimension of the spatiotemporal matrix may represent a plurality of (well) locations within the reservoir. A temporal dimension of the spatiotemporal matrix may represent for each location of the plurality of locations within the reservoir an associated timeseries.

[0145] Each timestamp of a time series associated with a location of the plurality of locations within the reservoir may be associated with measurement data. The associated measurement data may include a saturation measured at the location at the timestamp. Additionally or alternatively, the associated measurement data may include a flow rate measured at the location at the timestamp. Additionally or alternatively, the associated measurement data may include a pressure measured at the location at the timestamp. Additionally or alternatively, the associated measurement data may include a temperature measured at the location at the timestamp.

[0146] In step 420, a prediction of measurement data for a point in time within the period of time is received as output from the model.

[0147] In step 430, one or more training parameters of the model are adjusted based on an error between the prediction of the measurement data for the point in time and measurement data at the point in time.

[0148] Adjusting one or more training parameters of the model may comprise adjusting the one or more training parameters of the model based on an error between a prediction of measurement data for a given timestamp and the measurement data associated with the given time stamp.

[0149] As a result of the method 400, model for predicting saturation and pressure within a reservoir may be created. The model may be trained to receive as input a time stamp for predicting the saturation and pressure within the reservoir at the time stamp. The model may be trained to output a prediction of measurement data of saturation and pressure within the reservoir at the time stamp. Creating the model is done by training the model according to a method for training a model for predicting saturation and pressure within a reservoir according to the aspects of the present disclosure such as the method 400.

[0150] The model maybe based on non-linear approximators (NLAs). The model may include an additional output function. The additional output function may be configured to output a confidence interval for the prediction of measurement data of saturation and pressure within the reservoir at the time stamp.

[0151] Fig. 5 illustrates an overview of a data-processing system 500 according to the embodiments of the present invention. The data-processing system 500 maybe a computing device (e.g., a user device, smartphone, laptop, personal computer etc.) or a system comprising one or more processors, computing devices or the like, which may be distributed or on premise.

[0152] The data-processing system 500 may be designed in a modular architecture comprising one or more modules (e.g., Module A - Module D). Each module maybe capable of executing a method, a model and / or a computer program according to the aspects of the present disclosure independently. Even though only four modules are illustrated, it is to be understood that the data-processing system 500 may of course comprise more than these modules. In the example shown, module A maybe capable of executing the method too as explained with respect to Fig. 1. In the example shown, module B may be capable of executing the method 200 as explained with respect to Fig. 2. In the example shown, module C may be capable of executing the method 300 as explained with respect to Fig. 3. In the example shown, module D may be capable of executing the method 400 as explained with respect to Fig. 4. In certain implementations, the one or more modules may communicate with each other via corresponding communication channels, interfaces or the like. For example, a module may require as input the output of another module. Accordingly, the module may request the corresponding information from the other module via the communication channel, interface or the like.

[0153] The data-processing system 500 may be designed as a client-server architecture, wherein the one or more modules are stored at one or more servers (e.g., cloud server, remote server, local server etc.), and wherein a user device can request the execution of the corresponding method and / or model of one or more modules via an interface (e.g., web-interface, an API etc.). In this case, after the one or more servers receive the request, the one or more servers will execute the requested one or more modules and return the result or the output of the one or more modules to the user device on which the result or the output may then be displayed. It will be apparent to those skilled in the art that numerous modifications and variations of the described examples and embodiments are possible in light of the above teaching. The disclosed examples and embodiments are presented for purposes of illustration only. Other alternate embodiments may include some or all of the features disclosed herein. Therefore, it is the intent to cover all such modifications and alternate embodiments as may come within the true scope of this invention.

[0154] The aspects according to the present invention may be implemented in terms of a computer program which may be executed on any suitable data processing device comprising means (e.g., a memory and one or more processors operatively coupled to the memory) being configured accordingly. The computer program may be stored as computer-executable instructions on a non-transitory computer-readable medium.

[0155] Embodiments of the present disclosure maybe realized in any of various forms. For example, in some embodiments, the present invention maybe realized as a computer- implemented method, a computer-readable memory medium, or a computer system. The steps described within this disclosure maybe automatically performed.

[0156] In some embodiments, a non-transitory computer- readable memory medium may be configured so that it stores program instructions and / or data, where the program instructions, if executed by a computer system, cause the computer system to perform a method, e.g., any of the method embodiments described herein, or, any combination of the method embodiments described herein, or, any subset of any of the method embodiments described herein, or, any combination of such subsets.

[0157] In some embodiments, a computing device or system may be configured to include a processor (or a set of processors) and a memory medium, where the memory medium stores program instructions, where the processor is configured to read and execute the program instructions from the memory medium, where the program instructions are executable to implement any of the various method embodiments described herein (or, any combination of the method embodiments described herein, or, any subset of any of the method embodiments described herein, or, any combination of such subsets). The device maybe realized in any of various forms. Although specific embodiments have been described above, these embodiments are not intended to limit the scope of the present disclosure, even where only a single embodiment is described with respect to a particular feature. Examples of features provided in the disclosure are intended to be illustrative rather than restrictive unless stated otherwise. The above description is intended to cover such alternatives, modifications, and equivalents as would be apparent to a person skilled in the art having the benefit of this disclosure.

[0158] The scope of the present disclosure includes any feature or combination of features disclosed herein (either explicitly or implicitly), or any generalization thereof, whether or not it mitigates any or all of the problems addressed herein. In particular, with reference to the appended claims, features from dependent claims may be combined with those of the independent claims and features from respective independent claims may be combined in any appropriate manner and not merely in the specific combinations enumerated in the appended claims.

Claims

CLAIMS1. A computer-implemented method (300) for training a model for predicting a flow test of a well, wherein the method comprises the steps of: obtaining (310) a data sample comprising: a first set of parameters associated with at least a first flow test of the well; and a reference parameter associated with a second flow test of the well; providing (320) the first set of parameters associated with at least the first flow test of the well as input to the model; receiving (330) a predicted parameter associated with the second flow test of the well as output from the model; determining (340) an error between the predicted parameter and the reference parameter; and adjusting (350) one or more training parameters of the model based at least on the error between the predicted parameter and the reference parameter.

2. The method of the preceding claim, wherein the first flow test of the well is associated with a first time stamp; wherein the second flow test of the well is associated with a second time stamp; and wherein the first time stamp is smaller than the second time stamp.

3. The method of any one of the preceding claims, wherein each parameter of the first set of parameters associated with the first flow test is a measurement value resulting from the first flow test of the well.

4. The method of any one of the preceding claims, wherein the reference parameter associated with the second flow test of the well is a measurement value resulting from the second flow test of the well; and wherein the predicted parameter is an estimation of the measurement value.

5. The method of any one of the preceding claims, wherein the first set of parameters comprises at least one of: a choke opening of the well, a separator pressure, PSEP, of the well, a well-head pressure, WHP, of the well, a well-head temperature, WHT, of the well, a water-cut, WCT, of the well, a gas-oil ratio, GOR, of the well, an amount of oil produced, OIL, of the well, an amount of gaslift rate injected (GLR), a rotating speed of a pump (RPM) or any combination thereof.

6. The method of any one of the preceding claims, wherein the reference parameter and the predicted parameter are one of: GOR, WCT, WHP, WHT or OIL.

7. A model for predicting a flow test of a well, wherein the model is trained to: receive as input a time stamp for predicting a flow test of the well; and output a predicted parameter associated with the flow test of the well at the time stamp.

8. The model of the preceding claim, wherein the model is trained according to anyone of the preceding claims 1 to 6.

9. The model of any one of the preceding claims, wherein the model is an autoregressive integrated moving average with exogenous inputs, ARIMAX, model.

10. The model of the preceding claim, wherein the training parameters of the model comprise one or more covariates of the ARIMAX model.

11. The model of any one of the preceding claims, wherein the model comprises a plurality of sub models, wherein each sub model of the plurality of sub models is trained according to the method of any one of the preceding claims; and wherein each sub model of the plurality of sub models is trained to output a different predicted parameter.

12. A computer-implemented method (too) for determining a flow test plan, the method comprising:determining (no) a first flow test frequency of a plurality of flow test frequencies for a well; providing (120) to a model a first time stamp as input to predict a first flow test at the first time stamp; receiving (130) from the model a first predicted parameter associated with the first flow test at the first time stamp; and determining (140) a second flow test frequency of the plurality of flow test frequencies for the well based on the first predicted parameter.

13. The method of the preceding claim, wherein the model is one according to any one of the preceding claims 7-11.

14. The method of any one of the preceding claims 12-13, wherein determining the first flow test frequency for the well comprises: assigning a test criticality level of a plurality of test criticality levels to the well; and determining the first flow test frequency based on the test criticality level.

15. The method of the preceding claim, wherein the plurality of test criticality levels comprises: a low test criticality level, a medium test criticality level and a high test criticality level; and wherein the plurality of flow test frequencies comprises: a low testing frequency, preferably one every 60 days; a medium testing frequency, preferably one every 30 days; and a high testing frequency, preferably one every 15 days.

16. The method of the preceding claim, wherein assigning the test criticality level to the well comprises: determining that a set of parameters associated with the first flow test fulfills a first predefined set of rules and assigning a low test criticality level as the test criticality level; ordetermining that the set of parameters associated with the first flow test fulfills a second predefined set of rules and assigning a medium test criticality level as the test criticality level; or determining that the set of parameters associated with the first flow test fulfills a third predefined set of rules and assigning a high test criticality level as the test criticality level.

17. The method of the preceding claims 15-16, wherein determining the first flow test frequency based on the test criticality level comprises: determining that the first flow test frequency is a low testing frequency if the assigned test criticality level is low; or determining that the first flow test frequency is a medium testing frequency if the assigned test criticality level is medium; or determining that the first flow test frequency is a high testing frequency if the assigned test criticality level is high.

18. The method of any one of the preceding claims 12-17, wherein determining the second flow test frequency comprises: determining a first error between the first predicted parameter associated with the first flow test at the first time stamp and a first reference parameter.

19. The method of any one of the preceding claim, wherein the method further comprises: providing to the model at least a second time stamp as input to predict a second flow test at the second time stamp; wherein the second time stamp is smaller than the first time stamp; receiving from the model a second predicted parameter associated with the second flow test at the second time stamp; and wherein determining the second flow test frequency further comprises: determining a second error between the second predicted parameter associated with the second flow test at the second time stamp and a second reference parameter.

20. The method of any one of the preceding claims 18-19, wherein determining the second flow test frequency further comprises: determining that the first error, the second error and / or a combination of the first error and the second error is above or equal to a predefined error threshold and increase the flow test frequency; or determining that the first error, the second error and / or a combination of the first error and the second error is below the predefined error threshold and decrease the flow test frequency.

21. The method of any one of the preceding claims 12-20, wherein the set of parameters associated with the first flow test of well and / or a set of parameters associated with the second flow test of the well is unknown to the model.

22. The method of any one of the preceding claims 12-21, wherein the method further comprises: determining an urgency for performing a flow test on the well.

23. The method of the preceding claim, wherein determining the urgency comprises at least one of: determining a flow test compliance of the well; determining an uptime of the well; determining an allocated production rate of the well, determining a well priority or any combination thereof; and / or wherein the method further comprises: assigning a rank to the well based on the urgency.

24. A computer-implemented method (400) for training a model for predicting saturation and pressure within a reservoir, wherein the method comprises the steps of: providing (410) a representation of the reservoir as input to the model; wherein the representation includes measurement data of the saturation and the pressure within the reservoir over a period of time; receiving (420) as output from the model a prediction of measurement data for a point in time within the period of time; andadjusting (430) one or more training parameters of the model based on an error between the prediction of the measurement data for the point in time and measurement data at the point in time.

25. The method of the preceding claim, wherein the representation of the reservoir is a spatiotemporal matrix.

26. The method of the preceding claim, wherein a spatial dimension of the spatiotemporal matrix represents a plurality of locations within the reservoir.

27. The method of the preceding claim, wherein a temporal dimension of the spatiotemporal matrix represents for each location of the plurality of locations within the reservoir an associated timeseries.

28. The method of the preceding claim, wherein each timestamp of a time series associated with a location of the plurality of locations within the reservoir is associated with measurement data.

29. The method of the preceding claim, wherein the associated measurement data includes at least one of: a saturation measured at the location at the timestamp; a flow rate measured at the location at the timestamp; and / or a temperature measured at the location at the timestamp; and / or a pressure measured at the location at the timestamp.

30. The method of the preceding claim, wherein adjusting one or more training parameters of the model comprises: adjusting the one or more training parameters of the model based on an error between a prediction of measurement data for a given timestamp and the measurement data associated with the given time stamp.

31. A model for predicting saturation and pressure within a reservoir, wherein the model is trained to:receive as input a time stamp for predicting the saturation and pressure within the reservoir at the time stamp; and output a prediction of measurement data of saturation and pressure within the reservoir at the time stamp.

32. The model of the preceding claim, wherein the model is trained according to any one of the preceding claims 24-30.

33. The model of any one of the preceding claims 24-32, wherein the model is based on non-linear approximators, NLAs.

34. The model of any one of the preceding claims 24-33, wherein the model includes an additional output function, wherein the output function is configured to: Output a confidence interval for the prediction of measurement data of saturation and pressure within the well-reservoir pair at the time stamp.

35. A computer-implemented method (200) for determining a reservoir monitoring plan, RMP, the method comprising the steps of: selecting (210) a plurality of well-reservoir pairs for which the RMP is to be determined; determining (220) for each well-reservoir pair of the plurality of well-reservoir pairs a ranking indicator associated with the well-reservoir pair; and determining (230) the RMP for the plurality of well-reservoir pairs according to the ranking indicator associated with each well-reservoir pair of the plurality of well-reservoir pairs.

36. The method of the preceding claim, wherein determining the ranking indicator associated with the well-reservoir pair comprises: providing a time stamp as input to a model for predicting saturation and pressure within the reservoir at the time stamp; receiving as output from the model, a prediction of measurement data of saturation and pressure within the reservoir at the time stamp; andassigning the ranking indicator to the well-reservoir pair based on the prediction of measurement data of saturation and pressure within the reservoir at the time stamp.

37. The method of the preceding claim, further comprising: deriving an amount of expected remaining oil, ERO, from the prediction of measurement data of saturation and pressure within the reservoir at the time stamp; and wherein assigning the ranking indicator is further based on the amount of ERO.

38. The method of the preceding claim, wherein assigning the ranking indicator comprises: assigning a high ranking indicator if the amount of ERO is larger than or equal to a predefined ERO threshold; or assigning a low ranking indicator if the amount of ERO is smaller than the predefined ERO threshold.

39. The method of any one of the preceding claims 36-38, wherein the output from the model further comprises a confidence interval for the prediction of measurement data of saturation and pressure within the reservoir at the time stamp; and wherein the method further comprises: calculating an uncertainty value based on the confidence interval; and wherein assigning the ranking indicator is further based on the uncertainty value.

40. The method of the preceding claim, wherein the uncertainty value is a difference between an upper limit of the confidence interval and a lower limit of the confidence interval.

41. The method of any one of the preceding claims 39-40, wherein assigning the ranking indicator comprises: assigning a high ranking indicator if the uncertainty value is larger than or equal to a predefined uncertainty threshold; orassigning a low ranking indicator if the uncertainty value is smaller than the predefined uncertainty threshold.

42. The method of any one of the preceding claims 37-38 and 39-41, wherein the ranking indicator is determined based on a combination of the amount of ERO and the uncertainty value.

43. The method of the preceding claim, wherein the model any one according to the preceding claims 31-34.

44. A computer program comprising instructions, which when executed by a computer, causes the computer to execute the method of any one of the claims 1- 6, the model of any one of the claims 7-11, the method of any one of the claims 12-23, the method of any one of the claims 24-30, the model of any one of the claims 31-34 and / or the method of any one of the claims 35-43.

45. A data-processing system (300) comprising means for executing the method of any one of the claims 1-6, the model of any one of the claims 7-11, the method of any one of the claims 12-23, the method of any one of the claims 24-30, the model of any one of the claims 31-34 and / or the method of any one of the claims 35-43-

Images