�ִ������ݼ������Զ���������

现代产量递减曲线分析方法是研究油、气田动态规律的重要方法之一，推广运用中无法回避曲线拟合中的多解性问题，而克服曲线拟合中多解性问题的办法是利用计算机自动分析技术。目前，用理论模型自动拟合实测数据的自动分析是建立在最小平方非线性回归方法基础之上的，在解决多解性问题上尚存在三个弊端：①最后收敛值不一定代表的是最小值；②在控制参数相关性较好或者受某些参数的影响的实测数据不足时，其Hessian矩阵呈病态，最终导致计算结果不收敛，且其初值 的影响因素不大；③问题本身的复杂性（模型、数据精度、干扰等）和求解方法优劣，都关系到求解参数估计问题的成败。文章运用人工神经网络模式实现现代产量递减曲线自动分析，采用BP 网络模型对现代产量递减理论曲线进行训练学习。采用Pi-Sigma网络模型对实际产量曲线进行异联想恢复，解决曲线拟合的多解性问题。结合压力恢复曲线分析来检验现代产量递减曲线自动分析的人工神经网络方法的计算结果表明：现代产量递减曲线自动分析方法及其所运用的人工神经网络方法对解决多解性问题都能取得很好的效果，为油气藏工程中自动分析问题提供了新的现代途径。

Down-to-date Production Decline Curve Automatic Analysis Method

??Down-to-date production decline curve analysis method is one of the important methods of studying oil and gas field performance.In its popularization and application,multiple solubility problem can 't be avoided in curve fitting,which,however,may be overcome by applying computer automatic analysis technique.At present,the automatic analysis of fitting meas ured data by a theoretic model is set up on the basis of least square nonlinear regression method,in which there are still three shortcomings:??the last converg ent value is uncertainly representive of the minimum one;??when the correlativity of control parameters is satisfactory and the measured data influenced by some parameters are insufficient,its Hessian matrix appears as ill condition,thus causing the calculation result to be non convergent,however the influence of its initial value is not obvious;and ??the problem's own complexity (model,data's accuracy,interference,etc.)and the superiority and inferiority of the solving proc esses are related to the success or failure of solving parameters.In the paper,the down-to-date production decline curve automatic analysis is realized by applying artificial nerve network model and the training and learning are carried out for the down-to-date production decline theoretical curve by use of BP network model.The multiple solubility problem in curve fitting can be overcome through an iso-association build-up of measured production curve by Pi-Sigma network model.By means of examining the results calculated by artificial nerve network method of down-to-date production decline curve automatic analysis in combination with pressure build-up curve,it is shown that the down-to-data production decline curve automatic analysis method and the artificial nerve network method applied to the former are well suitable for overcoming the multiple solubility problem,thus providing a new path for automatically analyzing problems in rese rvoir engineering.