Multifactor LSTM Regional Production Forecasting Method Based on SM-PSO Optimization

Abstract

Accurate and rapid regional production prediction in oil and gas fields is crucial for production management, workload allocation, and investment planning. Currently, oil companies primarily rely on the production composition method for regional oil production planning. However, this method suffers from poor timeliness and consumes substantial human and material resources. Unlike oil field production prediction, regional production planning is influenced by a greater number of macro factors, such as regional exploration resources, development strategies, and market conditions. Therefore, we have established a multidisciplinary sample set that comprehensively considers exploration indicators, development indicators, production indicators, and economic indicators. Additionally, we innovatively propose an SM-PSO-RF-LSTM production prediction model. This model optimizes hyperparameters based on an innovative SM-PSO hybrid algorithm and initializes feature indicators based on importance weights derived from random forests. We designed three sets of comparative studies: Study 1 demonstrates that, in terms of regional production prediction, the new method outperforms previous approaches in prediction performance; Study 2 proves that hyperparameter optimization using the SM-PSO algorithm can significantly enhance the prediction accuracy of the LSTM model; and Study 3 establishes that the regional production prediction method based on planning strategies is more consistent with actual planning results.