用BP神经网络预测延迟焦化产品收率

摘要

针对传统的焦化产品收率预测方法准确性较差的实际情况,用Matlab编程构造了3层前馈BP神经网络,采用带动量的批处理梯度下降法来训练网络,并用所得模型对已知样本数据进行预测.结果表明,运用BP神经网络对焦化产品收率能够进行准确预测,最大相对误差为3.33%.与传统的预测模型相比,该网络模型的预测精准度更高.%The accuracy is very low in the prediction of product yields of delayed coking process with conventional mathematic model. A 3 forward-feed BP neural network was constructed with Matlab, the network was trained based on train gradient descent method, and the modeling was performed on the known sample data. The result shows: the BP neural network can efficiently predict the coke make with high accuracy. The highest relative error is 3.33％. As compared with the conventional prediction model, the BP model offers a higher prediction accuracy.