結合基因演算法與類神經網路預測油氣田蘊藏量

Reserves Estimation from Neuro-Genetic Approach

張峻彬(Chun-Pin Chang)

《鑛冶：中國鑛冶工程學會會刊》 47卷1期 (2003/03) Pp. 116-124

https://doi.org/10.30069/MM.200303_47(1).0010

摘要

油氣在生產的時候會受到許多內部或外部因素的影響，而使得產率呈現波動的變化狀況，為了能更正確的預測油氣產率，在進行產率預測時應將各類影響產率變化的因素均結合進預測模式中一併考慮。本研究的主要目的是要結合基因演算法尋求最佳解與類神經網路建構兩組資料間關係模式的特性，建立油氣田產率的預測模式，此模式能整合既有影響產率變化的各類因素資料，計算未來油氣田產率的預測值，估算決定性與機率性蘊藏量值。本文實例分析中的結果顯示，利用基因演算法與類神經網路相結合，可提高生產率預測的準確率，改進規則疊代方式的缺點，並且增進機率性蘊藏量估算的便利性。而且由各種不同的方法所預測出的未來產率與真實生產資料相比較，可以看出基因演算法與類神經網路的結合預測出來的未來產率數值，最能符合真實生產資料的變化趨勢與波動特性。

關鍵字

基因演算法 ； 類神經網路 ； 蘊藏量

並列摘要

The production rate of oil/gas field is affected by many external and internal factors. For the more precise prediction of production rate, these influencing factors must be considered in any prediction model. The main purpose of this research is to combine both characteristic of genetic algorithm (optimize solution) and neural network (relationship between two sets of data) for constructing the prediction model (G-N model) for production of the oil/gas field. The G-N model integrates various data of influening factors to predict production rate, deterministic and probabilistic reserves. The results from the analysis of example by production decline curve analysis, rule-iteration method and G-N model show that the G-N model can increase the accuracy of prediction, improve the defect of rule-iteration method, and promote the convenience of probabilistic reserve estimation. Comparing difference of the production rates from various methods and from real data, it is easy to get the conclusion that G-N model can match most the producing trends and the various characteristic of real data.

並列關鍵字

Genetic Algorithm ； Neural Network ； Reserve