Research On Compressor Characteristics Based On Field Data And Gas Pipeline Network Optimization

With the rapid development of national industry,natural gas as a clean and efficient energy source,its demand is increasing year by year,natural gas transmis s io n pipeline has been continuously under construction,the energy consumption of natural gas long-distance transportation is not a small number.Therefore,it is of great significance to study the energy consumption optimization of the gas pipeline.In the gas pipeline,the energy consumption cost of the gas station work will directly affect the energy consumption of the gas pipeline,and the energy consumption of the gas station must be carried out.Optimization,and if it is necessary to accurately calculate the energy consumption of the compressor,it is necessary to accurately grasp the characteristics of each compressor in the compressor station.The traditional method of obtaining the actual characteristics of the compressor is to obtain a real characteristic prediction model by in-depth understanding of the physical mechanism of the characteristic deviation,but the cause of the characterist ic deviation is complicated,there are many difficulties in analyzing the physical mechanism.Based on the field operation data of compressor,the difference between the real operation characteristics and the original test characteristics of the compressor is analyzed.Based on the calculation of knowledge and some real characteristics of the compressor characteristics,a prediction model of the real head characteristic of compressor based on deep learning network.A large number of compressors' real head data under different operating conditions are used as learning samples for deep learning networks.The stacked encoder is used to pre-training and the BP algorithm is used to fine-tune are the two training phases to build the predictive network.Finally,the untrained samples are used to test model accuracy.The established deep learning network model has a good prediction accuracy and generalized computing abilit y,which provides a new method for compressor performance evaluation and prediction,which will help to guide the compressor's safe and stable operation.In this paper,the real energy head data of compressors under a large number of different working conditions is used as the learning sample of deep learning network.The prediction network is established by stacking self-encoder pre-training and BP algorithm overall fine-tuning two training stages,and finally using untrained samples to model accuracy.The test was carried out.The operating point between the corrected outlet pressure and the actual outlet pressure value between +-0.1 MPa accounted for 77.04% of all working points,and the working point between the corrected and actual outlet pressure values was between +-0.3 MPa.Accounted for 93.12% of all work points.The established deep learning network model has good prediction accuracy and generalization calculation ability,and provides a method different from the traditiona l model for the evaluation and prediction of compressor performance,which is helpful for guiding the safe and stable operation of the compressor.In addition,based on the calculation model of the true characteristics of the compressor,the steady-state operation optimization model of the natural gas transmission pipeline network is built.The western natural gas transmission pipeline network is taken as an example to calculate the optimization scheme under various conditions,and the compressor built in this paper is successfully realized.The application of the real property calculation model.