Research On The Rheological Properties Of Super Heavy Oil Based On BP Neural Networks.

The super heavy oil reserves in China was large, while because of the characteristics of the high condensation point, high viscosity and poor liquidity, the super heavy oil had special rheological properties, which caused a lot of troubles in oil recovery, pipeline transport and storage operation, so the rheological characteristics of super heavy oil had been one of the key scientific research topic in oil-gas storage and transportation field. Before this, the study of rheological properties mostly used the traditional methods of experimental or numerical fitting and used the inherent models and formulas to analyze. This kind of methods which had big workload and low accuracy were not completely satisfied with the requirement of engineering, so it needed a new idea and method to study the rheological properties of super heavy oil. The artificial neural network as an information processing system of simulating brain structures and functions not only had the characteristics such as learning and fault tolerance, but also had the characteristic which can approximate arbitrary nonlinear continuous mapping and can simulate the inner relations between causal variables under the influence of various uncertain factors.Aiming at the characteristics of artificial neural network, taking the du 32 block super heavy oil in Liaohe oilfield for example, basing on a large number of rheological experiments, this paper analysed the rheological properties and thixotrotropy in detail according to the experimental data and respectively determined the influencing factors of the viscosity and thixotropic shear stress of super heavy oil. Introducing the artificial neural network theory, using three layers BP neural network which was relatively mature, we established the artificial neural network models which can calculate the viscosity and thixotropic shear stress of super heavy oil. It used the MATLAB neural network toolbox to solve the models. Through the comparison of model calculation results with measured results we can get that, the calculating precision of the models was higher than others and was up to 1% and 3%,which can satisfy the requirement of actual production. The structures of the models were simple and not standing on concrete forms, the calculation amount was small and may save a lot of manpower and material resources. The quotation of BP neural network model will provide a new tool and method to calculate the complex rheological properties of oil.