CNN-LSTM model optimized by improved sparrow search algorithm for oil well production prediction

Abstract To address the limitations of existing oil-well production forecasting models, particularly suboptimal hyperparameter selection, difficulties in fine-tuning, insufficient global search capability, and slow convergence, this study proposes a composite prediction model based on the improved sparrow search algorithm (ISSA) optimized the convolutional neural network (CNN), the long short-term memory (LSTM) neural network. The ISSA algorithm uses the cyclic chaotic mapping to initialize the population and introduces the improved producer-predator position update strategy, which can significantly improve the global search ability and convergence speed. It combines the feature extraction advantage of CNN and the processing ability of LSTM for long time series data to build an efficient prediction model. Pearson correlation analysis is applied to identify highly correlated operational parameters, which are used as input variables, while daily oil production serves as the output variable. A corresponding ISSA-CNN-LSTM model is constructed and verified using production data from wells TB987 and TB990. All performance evaluation metrics outperform the comparative models such as ISSA-LSTM and grey wolf optimizer (GWO)-LSTM, and the predictive accuracy satisfies the practical requirements of oilfield engineering applications. This model offers reliable technical support for oil well productivity forecasting and dynamic production performance evaluation.