Abstract: The selection of a proper drilling fluid system is the key to enabling fast and high-quality drilling operations while avoiding or reducing the occurrence of drilling accidents when working in deep wells, ultra-deep wells, and complex formations. When designing the drilling fluid using case-based reasoning (CBR), the drilling fluid system can be derived from multiple attributes such as lithology, well type, and well depth. However, the derivation results can be substantially affected by each attribute's weight assignment. The genetic algorithm suffers from slow convergence and low convergence precision when used for optimization of the attribute weights. Considering this issue, this study proposes an improved genetic algorithm to address the issue of attribute weight assignment in CBR. Initially, the genetic operator is improved using the following techniques. An exponential scale transformation method is used to optimize the selection of the individual operator. A self-adaptive adjustment is performed on the scale factors in the arithmetic crossover. With reference to the mutation operator, the mutation direction of each individual is modified to maintain the diversity of the population. Next, the self-adaptive adjustment of the crossover probability is realized from two aspects, namely the individual fitness and the level of variation between crossover individuals. Finally, by performing comparative experiments on the UCI dataset, we proved that the improved genetic algorithm can enhance the global convergence performance and increase the accuracy of CBR. Experimental results demonstrate that applying the improved genetic algorithm to the CBR-based drilling fluid design can effectively optimize the weight assignments of each attribute and therefore improve the quality of drilling fluid.

Key words: drilling fluid, case-based reasoning, case retrieval, weight of attribute, genetic algorithm