An Optimized Neural Network Prediction Model for Reservoir Porosity Based on Improved Shuffled Frog Leaping Algorithm

Abstract Efficient and accurate porosity prediction is essential for the fine description of reservoirs, which an optimized BP neural network (BPNN) model proposed. Aiming at problem that BPNN sensitive to initialization converges local optimum easily, improved shuffled frog leaping algorithm (ISFLA) proposed based on roulette genetic coding. Firstly, a mechanism introduced improve selection probability elite individuals, thus enhancing global optimization ability. Secondly, coding method carried out by making full use effective information such as optimal solutions boundary values subgroups. Subsequently, ISFLA verified 12 benchmark functions compared with four intelligent algorithms, experimental results show its good performance. Finally, applied initial weights thresholds BPNN, new named ISFLA_BP study problem. The logging data preprocessed grey correlation analysis deviation normalization, then achieved natural gamma, density other relevant parameters. performance standard three-layer parameter methods swarm intelligence algorithms. Experimental has higher training accuracy, stability faster convergence speed, mean square error 0.02, accuracy than five methods.