Smart Detection of Natural Gas Pipeline Defects with Enhanced Pollination Algorithm

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Aug 8 2023

In a paper published in the journal PLOS One, researchers proposed an enhanced flower pollination algorithm (FPA) with adaptive adjustments and Gaussian mutation, effectively improving its search capabilities. The improved flower pollination algorithm (IFPA) surpassed benchmarks, aided by a Gaussian mutation strategy optimizing input weights for the extreme learning machine (ELM). The IFPA-ELM model effectively addresses the limitations of FPA and ELM, achieving impressive defect identification rates of 97% and 96%. The model holds potential for intelligent pipeline defect diagnosis, while further optimization could solve complex engineering issues.

Background

The role of natural gas in delivering superior, clean energy has global environmental and economic implications, supplying a significant quarter of the energy demand of the world. The safe transportation of natural gas primarily relies on pipelines, though potential defects stemming from various causes threaten their operation. Recognizing the urgency of this issue is particularly crucial due to the extensive network of pipelines, often spanning long distances.

Traditional techniques like BP neural networks prove inadequate in accurately categorizing natural gas pipeline defects due to their lengthy training periods and intricate parameter demands. In contrast, the ELM, a novel algorithm rooted in neural networks, has found utility in classification, regression, clustering, and feature learning. While the ELM exhibits accelerated learning, particularly in guaranteeing accuracy, it presents certain limitations like inadequate prediction precision and weak convergence accuracy. Despite its application in various defect identification fields, the identification of pipeline defects by ELM remains underutilized.

Studies have reported approaches centered on improving ELM performance through intelligent optimization algorithms. Such algorithms are divided into classical and heuristic categories, with the latter proving more adept at solving intricate problems. In the quest for effective and resilient solutions, meta-heuristic algorithms like swarm intelligence and evolutionary algorithms stand out, exemplifying their adaptability and power.

Among these algorithms, FPA stands out as a significant meta-heuristic swarm intelligence optimization technique, introduced by Yang in 2012. The present study introduces a novel framework for detecting pipeline defects, proposing the IFPA. IFPA leverages an adaptive approach utilizing transformation probability and Gaussian mutation techniques to enhance the FPA's effectiveness. The optimization prowess of IFPA stems from both FPA and Particle Swarm Optimization (PSO). Moreover, IFPA extends its utility to optimizing the ELM. In comparison to ELM and FPA-ELM, IFPA-ELM notably excels in pinpointing various pipeline states, including normal conditions, pit defects, crack defect signals, and beyond, with heightened accuracy. 

Proposed methodology

The innovation of the present study lies in integrating an improved pollination algorithm with the extreme learning machine model to enhance defect identification accuracy. This approach addresses challenges such as insufficient data and imbalanced samples that hinder the robustness of defect identification models. The intricate interplay of the FPA, adaptive adjustments, Gaussian mutation, and IFPA is comprehensively detailed, laying the groundwork for an enhanced solution.

The FPA that simulates natural flower pollination offers a promising optimization strategy. Its key mechanism involves a transition probability, balancing global and local pollination for effective problem-solving. The inclusion of transformation probability P within FPA is pivotal, preventing premature convergence and finely tuning the algorithm's exploration and development capabilities. By regulating the interplay between global and local pollination, P ensures optimal convergence rates. In the context of local pollination, introducing a Gaussian mutation strategy addresses challenges arising from the absence of a built-in mutation mechanism. This strategy injects Gaussian-distributed disturbances into the original, facilitating the exploration of better solutions.

The heart of the study lies in the optimization prowess of the IFPA algorithm, which hinges on adaptive transition probability adjustments and Gaussian mutation strategies. These enhancements delicately balance developmental growth and exploratory leaps. Rigorous comparisons with classical algorithms such as IFPA and the evaluation across benchmark functions confirm its efficacy. Notably, IFPA excels in evading local optima, as demonstrated through convergence curve comparisons. The research further leverages the ELM, a neural network training algorithm, to enhance the robustness of intelligent diagnostics. Harnessing IFPA to optimize the initial weights of ELM and thresholds reduces randomness, thereby elevating the diagnostic capabilities of the IFPA-ELM amalgamation.

Experimental results

As the gas pipelines of Chine age over decades, ensuring smooth gas transportation becomes vital. This study focuses on intelligent defect identification for reliable operations, using the Shaanxi-Beijing gas pipeline as an example. Employing a magnetic flux leakage detector, defects on the pipeline's outer surface are assessed, with dimensions like groove width, depth, and length considered. By analyzing collected signals, the method achieves intelligent defect identification and intensity classification.

IFPA demonstrates remarkable performance across six benchmark functions. Combining IFPA with ELM maximizes model accuracy and speed, confirming its suitability for gas pipeline defect detection. Defect features were extracted from 140 training samples, and 50 neurons in the hidden layer were used for simulation. Pipeline defects were categorized into normal, pit defect, and crack defect signals. ELM, FPA-ELM, and IFPA-ELM algorithms were employed to distinguish pipeline states. The FPA-ELM algorithm notably enhances recognition accuracy, with only minor errors observed in the IFPA-ELM algorithm's results.

Conclusion

The IFPA-ELM model combines advanced IFPA with ELM for pipeline defect identification. Performance comparison with ELM, FPA-ELM, and IFPA-ELM algorithms confirms IFPA-ELM's superior accuracy, achieving recognition rates of 97% and 96%. This represents an increase of 34% compared to FPA and 13% compared to FPA-ELM. However, the model's time cost and the impact of conversion probability's constant value require optimization. Future work could explore wider applications in diverse fields for complex engineering optimization problems.