Abstract
The electrical energy consumption forecasting for crude oil pipeline is critical in many aspects, such as energy consumption target setting, batch scheduling, unit commitment, etc. For actual crude oil pipelines, the nonlinearity of the sample is strong. The electrical energy consumption of crude pipeline is affected by many parameters, including oil physical property parameter, pipe parameter, station parameter, environmental parameter and operating parameter. At the same time, the whole process has the characteristics of intermittency and complex fluctuations. The above three main reasons make the energy consumption forecasting of crude oil pipeline complicated. In the past few years, several intelligence-based models have been introduced to accurately forecast energy consumption. Among them, back-propagation neural network (BPNN) seems to be more effective and can handle the nonlinear energy behavior and achieve accurate forecast results. However, due to its over-fitting problem, the accuracy of energy consumption forecasting will be reduced. To overcome this problem, the paper proposes a hybrid method for short-term energy consumption forecasting, namely PSO-BPNN. Back propagation neural network is integrated with particle swarm optimization to find optimal network weight. In this research, an effective technique called principal component analysis is applied to eliminate redundant noise and extract the primary characteristics of transportation data. The stratified sampling method is used to divide the training set and the test set to avoid large deviations caused by the randomness of sampling. Taking a crude oil pipeline in northeast china as a case study, SCADA system data are collected daily from December 31, 2016 to June 18, 2019. Comparing the evaluation indicators of PSO-BPNN with that of five state-of-the-art forecasting methods of GA-BPNN, SA-BPNN, DE-BPNN, FOA-BPNN, BPNN, the effectiveness of PSO-BPNN algorithm is evaluated. Compared with other five forecasting methods, the forecast results of PSO-BPNN are in best agreement with the actual data. The results indicate that the proposed PSO-BPNN model outperforms all five models used for comparison, which demonstrates its superior ability to generate forecasts in terms of forecasting accuracy.
Electrical Energy Consumption Prediction of the Federal District of Russia on the Based of the Reccurent Neural Network
