基于动量自适应学习率PSO-BP神经网络的钻速预测模型研究

Prediction Model of Penetration Rate Based on PSO-BP Neural Network with Momentum Adaptive Learning Rate

机械钻速(rate of penetration,ROP)是钻井作业优化和减少成本的关键因素,钻井时有效地预测ROP是提升钻进效率的关键。由于井下钻进时复杂多变的情况和地层的非均质性,通过传统的ROP方程和回归分析方法来预测钻速受到了一定的限制。为了实现对钻速的高精度预测,对现有BP (back propagation)神经网络进行优化,提出了一种新的神经网络模型,即动态自适应学习率的粒子群优化BP神经网络,利用录井数据建立目标井预测模型来对钻速进行预测。在训练过程中对BP神经网络进行优化,利用启发式算法,即附加动量法和自适应学习率,将两种方法结合起来形成动态自适应学习率的BP改进算法,提高了BP神经网络的训练速度和拟合精度,获得了更好的泛化性能。将BP神经网络与遗传优化算法(genetic algorithm,GA)和粒子群优化算法(particle swarm optimization,PSO)结合,得到优化后的动态自适应学习率BP神经网络。研究利用XX8-1-2井的录井数据进行实验,对比BP神经网络、PSO-BP神经网络、GA-BP神经网络3种不同的改进后神经网络的预测结果。实验结果表明:优化后的PSO-BP神经网络的预测性能最好,具有更高的效率和可靠性,能够有效的利用工程数据,在有一定数据采集量的区域提供较为准确的ROP预测。

The rate of penetration(ROP) is a key factor in optimizing drilling operations and reducing costs.Effectively predicting the ROP is the key to improve drilling efficiency.Due to the complex and changeable conditions and the heterogeneity of the formation during drilling,the traditional ROP equation and regression analysis methods are limited to predict the ROP.To achieve high-precision prediction of ROP,a new optimized neural network model was proposed,namely,particle swarm optimization back propagation(BP) neural network with dynamic adaptive learning rate,which used logging data to establish a target well prediction model to predict the drilling speed.In the training process,the BP neural network was optimized,and the heuristic algorithm,that is,the additional momentum method and the adaptive learning rate,was used to combine the two methods to form a dynamic adaptive learning rate BP improved algorithm,which improved the training of the BP neural network,speed and fitting accuracy,resulting in better generalization performance.Combining BP neural network with genetic algorithm(GA) and particle swarm optimization(PSO),the optimized dynamic adaptive learning rate BP neural network was obtained.The logging data of well XX8-1-2 was used to conduct experiments and compare the prediction results of three different improved neural networks:BP neural network,PSO-BP neural network,and GA-BP neural network.The experimental results show that the optimized PSO-BP neural network has the best prediction performance,has higher efficiency and reliability,can effectively use engineering data,and provide more accurate ROP prediction in areas with a certain amount of data collection.The research results provide a new prediction method and idea with higher precision and efficiency than the traditional method of ROP equation for the prediction of ROP in drilling operations.