基于灵敏度分析的海洋油气资源BP神经网络预测模型的优化

Structure optimization of ocean oil and gas resources via BP neural network prediction model based on sensitivity analysis

作者针对BP神经网络结构设计中存在的问题,提出利用灵敏度分析方法对BP神经网络预测模型进行优化。通过BP算法与参数灵敏度分析的结合,寻找网络输入属性与输出属性之间的影响因子;在保证精度的前提下优选网络输入属性,简化网络结构,以增强网络的泛化能力,减少人为主观因素对网络设计的影响。最后以海洋油气资源预测为例,结合实测资料建立BP神经网络预测模型并进行了优化及预测精度评价,表明优化后的模型既能有效提高油气资源预测结果的稳定性,又不损失预测精度。

To resolve problems existing in the backpropagation （BP） neural network structure design, we used the sensitivity analysis method to optimize the BP neural network prediction model. First, we investigated the impact factors of the input and output attributes of the network by combining the BP algorithm and parameter sensitivity analysis. Then, based on an accurate premise, we optimized the input attributes of the BP network and simplified the model network structure to improve the network’s generalization ability and to greatly reduce the subjective choice of the structural parameters. Lastly, taking ocean oil and gas resources prediction as an example, we estab-lished the BP neural network prediction model using the measured data, and conducted a sensitivity analysis and prediction accuracy evaluation. The results indicate that the optimized model can effectively improve the stability of the prediction results with no loss in prediction accuracy.