基于神经模块网络的可迁移凝汽器真空度预测方法

TransferableCondenserVacuumLevel PredictionMethodBasedon NeuralModuleAlgorithm

本文旨在探索一种基于神经模块网络的具有泛化性的可迁移的凝汽器真空度预测方法。首先,采用神经模块网络对凝汽器真空度进行预测。其次,手动将数据分为低负荷,普通和高负荷工况。其中高斯噪声波动范围为80%到130%,以充分模拟真实世界中的低负荷和高负荷数据。最后,利用神经模块网络在正常数据集上训练的模型,对变工况情况下的低负荷和高负荷数据进行预测。实验证明神经模块网络训练的模型具有泛化性,在不同负荷情况下均能有效预测凝汽器真空度。

This study aims to explore a generalized and transferable method for predicting condenser vacuum based on Neural Module Network.Firstly,the neural module algorithm is employed to predict the condenser vacuum.Secondly,divide the original data,where the Gaussian noise fluctuates between 80%and 130%,to adequately simulate low-load and high-load data in the real world.Finally,utilizing the model trained with the neural module algorithm on the normal data set to predict low-load and high-load data under variable operating conditions.The experiments demonstrate that the model trained by the neural module algorithm exhibits generalization and can effectively predict condenser vacuum under different load conditions.