Pipeline corrosion rate prediction model using BP neural network based on improved sparrow search algorithm

Objective Accurate corrosion rate prediction of oil and gas pipelines is critical to ensure the operational safety of oil and gas storage and transportation systems. However, most of the existing prediction models are based on the BP neural network, which has drawbacks such as slow convergence rates and a tendency to fall into local optima.

Methods This paper proposes a pipeline corrosion rate prediction model using an optimized BP neural network based on an improved Sparrow Search Algorithm to address the aforementioned disadvantages. The improvement process involved initializing the population through a reverse learning strategy. Additionally, a hybrid sine-cosine algorithm was introduced to update the location of discoverers, and a Levy flight strategy was incorporated to update the location of followers within the Sparrow Search Algorithm. Based on the enhanced Sparrow Search Algorithm, the weights and thresholds of the BP neural network were optimized, leading to a more scientifically selected set of parameters.

Results Incorporating sample data of uniform corrosion rates and pitting corrosion rates from 100 rounds of 20 steel tests, pipeline corrosion rate prediction models were established respectively utilizing BP, SSA-BP, and MIS-SSA-BP neural networks. These models were applied to train, predict, and compare uniform corrosion rates and pitting corrosion rates of oil and gas pipelines. The MIS-SSA-BP neural network prediction model exhibited very low mean absolute error, mean square error, root mean square error, and mean absolute percentage error. Its relative errors between the predicted and measured values of both uniform corrosion rates and pitting corrosion rates were all below 5%. Furthermore, this model showcased superior evaluation indexes and prediction accuracy compared to the BP and SSA-BP neural network prediction models.

Conclusion After additional study efforts, the strong prediction performance of the pipeline corrosion rate prediction model based on the MIS-SSA-BP neural network has been further verified. The research findings offer new approaches and insights for future investigations into the prediction of corrosion rates in oil and gas pipelines.