Method investigation on intelligent optimization of high dimension HWMHF parameters
LI Lizhe, ZHOU Fujian, WANG Bo
1 Unconventional Oil and Gas Institute of Science and Technology, China University of Petroleum-Beijing, Beijing 102249, China 2 Petroleum Institute, China University of Petroleum-Beijing at Karamay, Karamay 834000, China
Horizontal well with multistage hydraulic fracturing (HWMHF) optimization is crucial for the economic and efficient development of unconventional oil and gas resources. In this paper, fracture propagation modeling, production simulation, and automatic search algorithms are coupled to establish an integrated workflow for the optimization of HWMHF. Intelligent optimization of high-dimensional parameters of HF is conducted to obtain the best matching set of HWMHF parameters, which maximizes the value of the economic index in the global range. The fracture propagation simulation employed a self-programmed boundary element fracture propagation simulator. The production simulation employed a CMG component simulator. The automatic search algorithm employed the Genetic algorithm (GA) and the Bayesian optimization algorithm. The whole optimization process is automated, i.e., the overall fracture morphology is automatically imported into CMG, the fracture model is automatically established, the oil and gas water production is predicted by the CMG HWMHF model, and the objective function value is calculated automatically. Further, the automatic search algorithm updates the next generation of input by measuring the relationship between the objective function values and the HWMHF parameter values, and re-performs fracturing simulation until the ideal fracturing parameters are obtained. The fracturing optimization process and results show that: (1) the combination of boundary element fracture propagation simulator and CMG component simulator can achieve fast fracturing simulation and satisfy the high simulation times for intelligent search; (2) the economic index has been improved 55% though the intelligent optimization; (3) the GA searches for better HWMHF parameters, while the Bayesian optimization algorithm performs a less number of iterations, while it can better embrace the domain knowledge; (4) both methods are suitable for solving the “black box” question of HWMHF optimization, and each has its advantages and obvious potential for field application promotion.
