基于麻雀搜索算法与BP神经网络的压裂效果预测

Fracturing effect prediction based on sparrow search algorithm and BP neural network

彭绪涛 王仪 贾程 任俊松

PENG Xutao WANG Yi JIA Cheng REN Junsong

四川轻化工大学计算机科学与工程学院

School of Computer Science and Engineering, Sichuan University of Science & Engineering, Zigong 643000, Sichuan, China

现有工程技术方法对压裂效果的预测精度普遍不高，容易造成经济损失，为此以麻雀搜索算法(Sparrow Search Algorithm，SSA)优化人工神经网络的算法模型，开展以提高压裂效果预测精度为目标的研究。首先以BP神经网络模型对压裂效果进行预测，其次以麻雀搜索算法优化BP神经网络权值后的模型进行预测，通过数据对比发现后者的预测精度更高，且能解决BP神经网络收敛慢、易陷入局部最优解、易产生过拟合现象等问题。研究结果表明，经过麻雀搜索算法调整权值的BP神经网络模型平均相对准确率达到93.85%，不仅比工程方法预测结果的精度更高，还高于未以麻雀搜索算法优化的BP神经网络模型的90.91%，在实际任务中拥有更稳定的性能和更高的精度。

The accuracy of the existing engineering methods for predicting fracturing effects is generally not high, which is likely to cause economic losses. For this reason, an algorithm model was proposed to optimize the artificial neural network by using sparrow search algorithm (SSA), so as to improve the prediction accuracy of fracturing effect. First, the BP neural network model was used to predict the fracturing effect, and then the prediction model was proposed after optimizing the weight of the BP neural network by the sparrow search algorithm. The latter has higher prediction accuracy and can solve the problems, such as slow convergence, easy falling into local optimal solution, and prone to overfitting, in BP neural network. The comparison results show that the average relative accuracy of the results evaluated by BP neural network model whose weights are adjusted by the sparrow search algorithm reaches 93.85%, which is not only higher than the accuracy predicted by the engineering method, but also higher than 90.91% of the BP neural network model before optimized by the sparrow search algorithm. This algorithm model has more stable performance and higher accuracy in actual tasks.

10.13639/j.odpt.2022.04.018