| Chemical flooding is a crucial tertiary oil recovery technology in oilfield development,including alkali flooding,surfactant flooding,polymer flooding,and their binary or ternary mixtures.The displacement properties of these components are conducive to increase oil content.Chemical flooding optimization refers to the technology which optimizes the injection scheme of displacement agents according to displacement mechanism or geophysical data,so as to achieve the maximum economic benefit or oil production with injection limitation satisfied.Nevertheless,a majority of existing research on chemical flooding optimization focuses on the injection strategy of chemical flooding for a single well,with a cold shoulder to the scheduling of working states of multiple wells in a large-scale reservoir.As a direct result,the utilization rate of chemical flooding is reduced,which not only proliferates the development cost,but also accelerates undesired environmental pollution.In consequence,aiming at fully utilizing the effectiveness of displacement agents as well as further improving the profits,synchronous dynamic optimization of chemical flooding injection strategy and oil wells’ states is urgent and of self-evident importance.The above challenging requirement can be summarized as a MIOCP,and the purpose of this paper is to explore advanced solution methods for MIOCP based on mixed-integer optimal control.In this MIOCP,1)the injection concentrations of the displacing agents are continuous controls and the operating states of multiple wells are integer controls,2)the minimum water content,maximum NPV,etc.,are regarded as objective functionals,3)the oil displacement process dynamic description is the governing equation,and 4)the injection limitations of displacing agents and the wells’ states are constraints.Accordingly,efficient and high-precision MIOCP numerical methods,as well as flexible handling methods of uncertain factors or even uncertain first-principle model are the focuses of this study.The specific contents and innovations are as follows.1.For chemical flooding MIOCP with first-principle model,an EMLGR pseudospectral method is proposed as an accurate and efficient tool for discrete numerical solution.In response to the defects of Lavrentiev phenomenon and overlong switching time by existing LGR pseudospectral method when facing MIOCP,EMLGR extends control variables and related constraints at each intersection point of adjacent intervals,so as to improve the approximation degree of non-smoothness positions of the control trajectory.Moreover,an adaptive configurator is designed to independently optimize the polynomial order and interval structure,thereby further improving the accuracy.The simulation tests on five engineering MIOCPs and a core experiment polymer flooding case verify the superiority of EMLGR in both computational efficiency and precision.2.The first-principle model suffers from difficulties such as low generalization ability and uncertain geological parameters.Considering the spatiotemporal characteristics of the reservoir DPS,a data-driven 3-D DPS online spatiotemporal modeling method is presented,which consists of two parts:the IRD algorithm used for identify the spatial dynamics between multiple sensors and the O-LS-SVM algorithm utilized for learning the nonlinear temporal dynamic.The final DPS output can be predicted via time-space integration.In this modeling framework,the autonomously evolved dimensional embedded SDBF set settles the shortcomings of mapping distortion and inflated modeling complexity exposed by existing DPS modeling paradigms.It is also the first attempt to provide data-driven modeling scheme for 3-D time-varying DPSs considering the sensor layout.3.The key of chemical flooding optimization is to obtain the optimal mixed-integer decisions by solving the large-scale multi-interval MINLP.However,the MINLP is an NP-hard issue whose computational complexity increases exponentially with the integer decisions.To bridge the gap of slow efficiency and non-global optimization by conventional mixed-integer optimization algorithms,a QA-DESS algorithm is designed in this study to achieve efficient global optimization of the MINLP.Concretely,the QA optimizer with parameter self-tuning is utilized to search optimal integer decisions,and the DESS devotes to solve continuous optimization.The two heuristic optimizers realize the global searching within solution space via an interactive parallelism mode.Relevant computational complexity analysis and global optimality proof demonstrate that QA-DESS owns strong competitiveness in terms of convergence speed and global optimization.4.Centering around the actual industrial requirements of uncertain information processing and multi-objective optimization,advanced methods for solving uncertain multi-objective chemical flooding MINLP are also the focus of this study.Specifically,1)a CCP-based method for solving uncertain MINLPs is proposed for the case where distribution characteristics of uncertainties are known,and 2)an IP method based on interval credibility is presented to estimate interval fitness functions and screen excellent individuals when faced with the case where the distribution characteristics are unknown and described by interval numbers.For the chemical flooding IMOMINLP with multiple objectives,aiming at achieving a non-dominated solution set closer to the complete Pareto front,a PS-MCMO technique is designed to enhance the evolutionary ability of multiple clusters.The above efforts can provide theoretical support for the formulation of optimal multi-objective oil displacement scheme under uncertain conditions.5.In order to dwindle the computational burden in solving combinational optimization,this paper puts forward a new algorithm named QST-MPC to solve chemical flooding MIOCP.The self-triggered control update strategy is utilized to program the optimal displacement scheme.Owing to the substitution of MINLP solving,the computational complexity dramatically reduces.In addition,the QST-MPC casts a new light on balancing the relationship between triggering frequency and suboptimality,and provides more initiative for policymakers to determine the triggering frequency in advance.The simulation results of an ASP case with one injection well and four production wells also verify the effectiveness of the proposed QST-MPC method. |
