Shear wave velocity prediction based on adaptive particle swarm optimization optimized recurrent neural network

Shear wave velocity (Vs) is an important parameter for reservoir description and fluid identification and is extensively applied in study of oil and gas reservoir geomechanics and rock physics. However, due to high cost of Vs log, actual well log data generally lacks information about Vs, which doesn't meet the requirements of practical work. Therefore, it is very important to estimate the Vs by using conventional logs. In this study, according to the new achievements in deep learning, a novel hybrid Vs prediction model is established based on long short term memory (LSTM) recurrent neutral network optimized by particle swarm optimization (PSO) algorithm with adaptive learning strategy. For this purpose, firstly, the grey relational analysis method was employed to reduce data dimensions, the sensitive logs were chosen among the inputs to predict Vs. Next, LSTM network is applied to model the nonlinear relationship between the selected sensitive logs and Vs, and through finding the key hyper-parameters in LSTM model by the PSO algorithm with adaptive learning strategy, the log data feature matches the network topology structure, and the model's prediction ccuracy of Vs is improved. Finally, the superiority is verified through real petrophysical log data, and compared with other prediction methods. By this method, Vs can be predicted from series of input log data with consideration of variation trend and context information with depth. It is more suitable for the problem with multiple series data such as Vs prediction. Results of processing real petrophysical log data indicate that compare with conventional methods, the PSO algorithm with adaptive learning strategy based LSTM network model may not only provide higher predictive accuracy and robustness, but also has a promising application prospect in Vs prediction study.