Research On Gas-logging Data Processing And Hydrocarbon Reservoir Distinguishing Method

In modern drilling technology, workers know the geological structure condition mainly through gathering and analyzing various information generating by drilling instruments. With the improvement of science and technology and constant perfection of drilling technology, this kind of technology has gradually evolved into well logging technology. So far, well Gas logging technology has become the most important method to gain this kind of data. By detecting gas-logging data, workers can not only monitor down-hole conditions at any time but also distinguish the nature of drilling reservoir fluids in order to judge which are oil reservoirs and which are gas reservoirs. This technology is called hydrocarbon reservoir distinguishing method. In modern hydrocarbon reservoir distinguishing methods。however, there exists two ubiquitous drawbacks: first, the detection of gas data is easy to be influenced by many external factors, thus data adopted was not the original data, which reducing the accuracy rate in distinguishing of reservoirs. Secondly, many modern hydrocarbon reservoir distinguishing methods are artificial distinguishing methods, and active jamming factor will be the main leading factor in discrimination results. Including the gas detection method of Fischer as well as Bayes has the results can not be refined and the modeling time is longer two defects.Therefore,the thesis puts forward the correction processing method of gas logging data processing and hydrocarbon reservoir distinguishing method based on BP neural network. First of all, the thesis studies every factor influencing the accuracy of detection of gas data, in which includes two main classes of geo-environment factors and equipment affecting factors. The thesis puts forward relevant gas logging data correction processing method against each influencing factor in order to restore the gas data recorded by equipment into the original state. Secondly,by combining gas data and BP neural network, the thesis establishes gas logging BP network model for identifying oil and gas.In allusion to local minimum defects appeared in stage of modeling of BP neural network, the thesis proposes an improved adaptive learning method applying to network model, thus reduces the training time of gas logging BP network model for identifying oil and gas and raises up work efficiency.The next is to apply the gas logging BP network model for identifying oil and gas proposed by the thesis into practice. First, takes one group of gas logging data unknown the result as sample data, then normalizes the sample data with the correction processing method of gas logging data processing to improve the accuracy of the data and remove the influence of environment factors. After that, the author applied gas logging BP network model for identifying oil and gas and triangle distinguishing method to the sample data. By comparing the later oil testing conclusion and discrimination result, the author found out that hydrocarbon reservoir distinguishing method proposed by the thesis has higher accuracy rate.