Oil Well Yield Predict Based On Deep Learning

As a necessary energy for human life and development,it is necessary to make a reasonable prediction of the energy production.In today's world,living energy consumption mainly is oil and natural gas.The amount of oil production is directly related to the development of the national economy,and the oil production is equivalent to the blood of the national economy.As a non-renewable energy source,it is imperative to plan and use oil production reasonably.Making reasonable predictions of oil production is of great significance for guiding the distribution of oil personnel and the planning and use of oil.Aiming at the traditional oil production prediction method,this paper makes a reasonable prediction of oil production based on deep learning methods.Main work contents and achievements of this paper:1.This paper comprehensively analyzes some major factors that affect oil production,and uses Pearson correlation coefficient to quantitatively analyze the factors that affect oil production,and ranks the factors that affect oil production.It can be obtained through the Pearson product moment correlation coefficient.The main factors affecting the oil block production are the verification of monthly oil production,number of water injection wells,geological reserves,total number of water injection wells,monthly oil production at the wellhead,number of oil well openings.There are 11 positive correlation Pearson product moment correlation coefficients,such as production reserves,annual injection volume,total number of wells,cumulative oil production,and annual fluid production,and the correlation coefficient is greater than0.9.In addition,the negative correlation factors of water quality qualification rate and oil production operation cost are very small and have little effect.2.Using blocks as a unit,analyze and predict the single factor's impact on oil production.Using multiple models in deep learning to predict single-factor oil production.The experimental data set was pre-normalized to make the data conform to the smooth data trained by the deep learning model.The single-factor block oil production prediction models were trained using methods such as LSTM,Bi-LSTM,LSTM-seq2 seq,Bi-GRU-seq2 seq,and Bi-LSTM-seq2 seq,which can be obtained through comparative analysis experiments,using Bi-LSTM-seq2 seq.The model trained by the method has the best prediction effect and the accuracy rate reaches 99.5%.3.Multi-factor block oil production prediction,using machine learning to predict multi-factor block oil production.First,pre-process the data.Ridge Regression,Lasso Regression,Random Forest Algorithm Regression,Elastic net and XGBoost algorithms were used to train model to predict the multi-dimensional oil production of the block.First,using an experimental data set that includes all the features,the prediction of oil production can be obtained through comparative analysis experiments.The random forest algorithm has the best prediction effect and the goodness of fit reaches 0.829.Then,using the experimental data to extract the main features,through analysis and comparison experiments can be obtained,the prediction effect of the model trained by the random forest algorithm is closest to the actual output,and the best fit is 0.851.