Research On Prediction Method Of Process Operation Parameters Of Mixed Oil Pipeline Based On Deep Learning

The winter temperature in the northeast of China is extremely low,and the difference between the oil transport temperature and the soil temperature is large,resulting in faster heat dissipation from the pipeline,causing higher energy loss in the pipeline.Temperature and pressure are important operating parameters in the pipeline transmission process,which are the prerequisite and basis for pipeline calculation and optimal operation.The theoretical process calculation model and the process calculation model based on the traditional fitting method to calculate the temperature drop and pressure drop of the oil pipeline have a large deviation from the actual operation data.In order to improve the accuracy,the deep learning method was selected to carry out the research on the prediction model of the operating parameters of the oil mixing pipeline process,and the research results obtained are as follows:Pearson correlation coefficient was used to analyze the temperature drop influence factors based on python language to establish PSO-BP temperature drop prediction model.The root mean square error of PSO-BP temperature drop prediction model was reduced from0.8376 to 0.5556,and the average relative error was reduced from 9.07% to 5.66%,which is3.41% lower than the average relative error of traditional calculation model.To improve the calculation accuracy,the RNN(Recurrent Neural Network)temperature drop prediction model is established considering the time series problem of operation data,and the results show that the root mean square error of RNN temperature drop prediction model is reduced from 0.5556 to 0.2885,and the average relative error is reduced from 5.66% to 2.90%,and the average relative error is reduced by 6.17% compared with the traditional calculation model,which has the highest calculation accuracy.The influencing factors of pressure drop were analyzed,and the RNN pressure drop prediction model was established.Compared with the traditional calculation model,the Root-mean-square deviation of pressure drop was reduced from 0.1738 to 0.1144,and the average relative error was reduced from 4.58% to 2.95%.The RNN pressure drop prediction model has the problem of gradient disappearance when using the BP algorithm to optimize the parameters.For this reason,the LSTM(Long Short-Term Memory Network)pressure drop prediction model is established to realize the long time preservation of input information by using three control gates and cell cell states to effectively solve the gradient disappearance problem.The results show that the root mean square error of the LSTM pressure drop prediction model decreases from 0.1144 to 0.0453,and the average relative error decreases from 2.95% to 1.44%,which is the highest computational accuracy with a 3.14% decrease in the average relative error compared to the traditional computational model.The LSTM integrated prediction model is established to calculate temperature and pressure drop simultaneously,which is more accurate than the traditional calculation model and lower than the single model.A DE-LSTM(Differential Evolutionary Algorithm-Long Short Term Memory Network)integrated prediction model is established to improve the prediction accuracy,and the prediction results are compared with traditional computational models and single models.The results show that the DE-LSTM integrated prediction model temperature drop root mean square error decreases from 0.8376 to 0.4527,and the average relative error decreases from 9.07% to 4.02%,which decreases 5.05% compared with the traditional model error and increases 1.12% compared with the RNN temperature drop model average relative error.The root mean square error of the DE-LSTM integrated prediction model pressure drop decreased from 0.1738 to 0.0836,and the average relative error decreased from 4.58% to 1.84%,which is 2.74% lower than the average relative error of the conventional model and 0.40% higher than the average relative error of the LSTM pressure drop model.The study shows that the DE-LSTM integrated prediction model improves significantly in accuracy compared with the traditional computational model and decreases compared with the single model,but still has some advantages over the traditional computational model and the single model.