This dissertation develops a novel approach named Physics-based Residual Kriging (Phy-RK) for statistical prediction of spatially dependent functional data. This methodology assumes that prior information about the problem under study is available in terms of a physical model which expresses the governing physical laws of the phenomenon under study. The Phy-RK predictor incorporates this physical model within a Universal Kriging setting through a geostatistical modelisation of the residuals with respect to the physical model. The presented methodology has been motivated by the problem of forecasting production curves, which are modelled as time-dependent functional data, of wells operating in a mature hydrocarbon reservoir according to a given drilling schedule. Indeed, production forecast is a crucial tool for the reservoir management. Here, we aim at predicting future production curves of existing wells and wells yet to be drilled at a set of locations in the reservoir, based on the observed past production historical data. The peculiarity of this problem is that drilling new wells perturbs the pressure of the whole reservoir. This aspect has motivated the formulation of the Phy-RK predictor in a sequential framework, through an incremental updating of the predictive model in order to account for both the reservoir dynamics and the misfit between previous predictions and actual observations. The Phy-RK predictor represents a paradigm for the development of new surrogate models for the prediction of production curves. These surrogate models are often a convenient alternative to three-dimensional high-fidelity reservoir simulations, which involve the solution of flow equations on a fine discretised reservoir model. Indeed, the main drawbacks of these reservoir simulators are their high computational burden and the intensive process for setting their parameters, preventing their application in a context where their repeated solution is needed or a detailed geological model is not available. Starting from a pure geostatistical formulation, we then move to the Phy-RK predictor, presenting its application in problems of increasing complexity, from synthetic to realistic reservoirs, from single-phase to two-phase problems. Finally, we present the application of the Phy-RK predictor for the optimisation of the drilling schedule in order to maximise the economical return.
La tesi sviluppa un nuovo approccio chiamato Physics-based Residual Kriging (Phy-RK) per la previsione statistica di dati funzionali spazialmente dipendenti. Questa metodologia assume che le informazioni a-priori sul problema in esame siano disponibili in termini di un modello fisico, che esprime le leggi fisiche che governano il fenomeno. Il predittore Phy-RK incorpora questo modello fisico all'interno del predittore Universal Kriging, attraverso una modellizzazione geostatistica dei residui rispetto al modello fisico. La metodologia presentata è stata motivata dal problema della previsione delle curve di produzione, modellate come dati funzionali dipendenti dal tempo, di pozzi che operano in un giacimento maturo di idrocarburi, secondo una determinata campagna di perforazione. Infatti, la previsione della produzione è uno strumento cruciale per la gestione del giacimento. In questa tesi, l'obiettivo è prevedere le curve di produzione future dei pozzi esistenti e dei pozzi ancora da perforare in determinate posizioni nel giacimento, basandoci sui dati storici di produzione osservati nel passato. La peculiarità di questo problema è che la perforazione di nuovi pozzi perturba la pressione dell'intero giacimento. Questo aspetto ha motivato la formulazione sequenziale del predittore Phy-RK, attraverso un aggiornamento incrementale del modello predittivo per tenere conto sia della dinamica del giacimento che della discrepanza tra le previsioni e le osservazioni. Il predittore Phy-RK rappresenta un paradigma per lo sviluppo di nuovi modelli surrogati per la previsione delle curve di produzione. Questi modelli surrogati sono spesso un'alternativa conveniente alle simulazioni tridimensionali ad alta fedeltà del giacimento, che comportano la soluzione delle equazioni di flusso su un modello di giacimento finemente discretizzato. Infatti, i principali svantaggi di questi simulatori di giacimento sono il loro alto carico computazionale e il processo intensivo per determinare i loro parametri, impedendo la loro applicazione in un contesto in cui sia necessario calcolare la soluzione per molte configurazioni o un modello geologico dettagliato non sia disponibile. Partendo da una pura formulazione geostatistica, passiamo poi al predittore Phy-RK, presentando la sua applicazione in problemi di complessità crescente, dai giacimenti sintetici a quelli realistici, dai problemi monofase a quelli bifase. Infine, presentiamo l'applicazione del predittore Phy-RK per l'ottimizzazione della campagna di perforazione al fine di massimizzare il ritorno economico.
Physics-based residual Kriging : a new paradigm for the prediction of production curves in mature hydrocarbon reservoirs
RICCARDO, PELI
2022
Abstract
This dissertation develops a novel approach named Physics-based Residual Kriging (Phy-RK) for statistical prediction of spatially dependent functional data. This methodology assumes that prior information about the problem under study is available in terms of a physical model which expresses the governing physical laws of the phenomenon under study. The Phy-RK predictor incorporates this physical model within a Universal Kriging setting through a geostatistical modelisation of the residuals with respect to the physical model. The presented methodology has been motivated by the problem of forecasting production curves, which are modelled as time-dependent functional data, of wells operating in a mature hydrocarbon reservoir according to a given drilling schedule. Indeed, production forecast is a crucial tool for the reservoir management. Here, we aim at predicting future production curves of existing wells and wells yet to be drilled at a set of locations in the reservoir, based on the observed past production historical data. The peculiarity of this problem is that drilling new wells perturbs the pressure of the whole reservoir. This aspect has motivated the formulation of the Phy-RK predictor in a sequential framework, through an incremental updating of the predictive model in order to account for both the reservoir dynamics and the misfit between previous predictions and actual observations. The Phy-RK predictor represents a paradigm for the development of new surrogate models for the prediction of production curves. These surrogate models are often a convenient alternative to three-dimensional high-fidelity reservoir simulations, which involve the solution of flow equations on a fine discretised reservoir model. Indeed, the main drawbacks of these reservoir simulators are their high computational burden and the intensive process for setting their parameters, preventing their application in a context where their repeated solution is needed or a detailed geological model is not available. Starting from a pure geostatistical formulation, we then move to the Phy-RK predictor, presenting its application in problems of increasing complexity, from synthetic to realistic reservoirs, from single-phase to two-phase problems. Finally, we present the application of the Phy-RK predictor for the optimisation of the drilling schedule in order to maximise the economical return.File | Dimensione | Formato | |
---|---|---|---|
Tesi_dottorato.pdf
accesso solo da BNCF e BNCR
Dimensione
6.66 MB
Formato
Adobe PDF
|
6.66 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/204742
URN:NBN:IT:POLIMI-204742