Research On Energy Consumption Prediction Method For Long-distance Oil Pipelines

For a long distance oil pipeline, it is significant to improve the operation management level and reduce the energy consumption. It is also the main way to improve the economic efficiency of the enterprises. The measure to reduce the operation cost of oil pipeline is to implement strict energy consumption target monitoring for each pipeline. Therefore, the key to implement target management is to predict the energy consumption value accurately.Analysis the energy components of a long distance oil pipeline, by all types of energy and batch data of Lan Cheng Yu pipeline in each station per month from 2011 to 2015, with the qualitative and quantitative analysis methods. According to history data of pump units energy consumption and all kinds of factors that influence energy consumption data per day in 2012, using IBM SPSS 19 do correlation analysis and principal component analysis. And then, establish an index system of product oil pipeline energy consumption and the corresponding energy-saving measures.By learning various prediction algorithm theories, elected viable pipeline energy consumption prediction method, which based on statistical analysis of historical operating data:Grey prediction, SVM, BP neural network, ELM neural network and RBF neural network algorithm.Establish the pump units and other energy consumption per month prediction model of Lan Cheng Yu pipeline, and programming in MATLAB R2013B through the above five kinds of algorithms. Through the per month energy consumption prediction model of Lu Wan Yi Qi pipeline and the daily pump units energy consumption prediction model of Lan Zhou pumping station, test the prediction accuracy of all the algorithms.The results show that RBF neural network algorithm stand out in terms of the long distance oil pipeline energy consumption prediction:calculating errors close to zero (10"15magnitude), learning ability, highly accurate and easy to converge; The BP neural network prediction results are close to the real condition, which could be used as a simple comparison of energy consumption prediction; SVM prediction results are often larger than the true value, ELM neural network prediction results are often smaller than the true value, these two kinds of algorithm prediction accuracy is poor, do not apply to the field; Gray model prediction errors are large and irregular, does not apply to the long distance oil pipeline energy consumption prediction specially. In addition, the same algorithm prediction accuracy is evident lower in daily energy consumption model than the prediction accuracy of monthly energy consumption model. Recommended consumption prediction intervals take months.