Stratum total organic carbon prediction method based on parameter fusion

Abstract

The invention discloses a stratum total organic carbon prediction method based on parameter fusion, and relates to the technical field of oil exploration and development. Comprising the steps that a stratum total organic carbon measured value and a plurality of logging parameters of an exploration well are obtained, and the logging parameters with correlation coefficients with the stratum total organic carbon measured value larger than a set threshold value in all the logging parameters serve as characteristic logging parameters; an improved deep neural network is constructed, the characteristic logging parameters are input into the improved deep neural network, and a stratum total organic carbon predicted value is obtained; training the improved deep neural network according to an error between the stratum total organic carbon predicted value and the stratum total organic carbon measured value; in the stratum total organic carbon prediction process, characteristic logging parameters are selected from the multiple logging parameters of the to-be-detected exploration well and input into the trained improved deep neural network, and a stratum total organic carbon prediction result is obtained. According to the method, the decoupling capability for the multi-factor coupling effect is enhanced, abnormal values are effectively removed, and the data quality is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil exploration and development, and particularly relates to a method for predicting total organic carbon of a formation based on parameter fusion. Background Art

[0002] The prediction of total organic carbon (TOC) of a formation is necessary in multiple fields. It can improve the efficiency of oil and gas exploration, optimize the development plan, assist in geological research and resource assessment, while enhancing the operation efficiency and management level of enterprises, and also promote scientific research and technological progress, providing strong support for the development of related fields.

[0003] Traditionally, the prediction of TOC relies on complex pyrolysis experiments on core or cuttings samples, which is not only time-consuming and laborious but also costly. In existing research, Patent CN117192613A discloses a TOC prediction method and device based on lithology control. Patent C116859467A discloses a method for predicting TOC of hydrocarbon source rocks. Patent CN15877463A discloses a method for predicting TOC of shale gas reservoirs. Patent CN114943060A discloses a method for predicting total organic carbon of shale gas based on deep learning and interpolation regression. Patent CN13835138A discloses a method for predicting the total organic carbon content of shale based on a deep encoding-decoding network. Patent CN13050191A discloses a method and device for predicting TOC of shale oil based on two parameters. Patent CN12147713A discloses a method for segmentally predicting the total organic carbon content of mud shale. In the 4th issue, Volume 41, 2017, Journal of China University of Petroleum, Chen Haifeng et al. carried out a fine evaluation of TOC of hydrocarbon source rocks based on the variable coefficient resistivity logarithm change method. The variable coefficient resistivity logarithm change method was used to evaluate the TOC of hydrocarbon source rocks in this area. The empirical parameters in the resistivity logarithm change method were regarded as undetermined coefficients. Starting from the geological significance of model parameters, the undetermined parameters in the single-well and "zone and phase" prediction models were targeted to complete the TOC logging evaluation of 120 wells in the whole area. The hydrocarbon source rocks were classified and evaluated by calculating TOC with logging. By the method of "determining the frequency by drilling and determining the distribution by sequence stratigraphy", the thickness of hydrocarbon source rocks with different abundance levels in the first member of the southern section was obtained. In the 4th issue, Volume 25, 2018, Petroleum Geology and Recovery Efficiency, Bian Leibo et al. optimized the resistivity logarithm change method and its application in predicting the total organic carbon content of medium-deep hydrocarbon source rocks. The resistivity logarithm change method was optimized by the normalization method, and the natural gamma parameter was introduced to establish an optimized total organic carbon content prediction model. In the 3rd issue, Volume 34, 2019, Progress in Geophysics, Chen Hao et al. optimized the logging prediction method and application of TOC of hydrocarbon source rocks in the Juyan Sea Depression of the Yin'e Basin. Combining the geological characteristics of the Juyan Sea Depression and relevant logging parameters, the error comparison and analysis of the prediction results of different methods were carried out. It was preferably considered that the multiple regression method could be better applied to the prediction of TOC content of hydrocarbon source rocks in the study area, and a multiple regression prediction model with ∆logR parameter (the relevant variables and coefficients in the method of identifying and calculating the total organic matter carbon in strata using logging data), deep lateral resistivity and natural gamma as basic parameters suitable for the study area was proposed. In the 1st issue, Volume 36, 2021, Progress in Geophysics, Lu Pengyu et al. predicted the organic carbon content of the Lunpola Basin based on the neural network method.

[0004] In the prior art, the implementation process of the TOC prediction method based on well logging data includes: using a well logging instrument to measure in the target well section to obtain various well logging data including natural gamma ray, resistivity, acoustic travel time, etc.; performing processing such as correction and filtering on the collected original well logging data to remove noise and outliers; based on existing geological understanding and sample data, using methods such as statistical analysis and machine learning to establish a relationship model between well logging data and TOC; inputting the preprocessed well logging data into the established model and calculating the predicted value of TOC through calculation.

[0005] The formation lithology is diverse and highly heterogeneous, and different lithologies have different degrees of influence on well logging responses, making the relationship between well logging data and TOC complex and difficult to accurately establish a model. For example, in shale formations, the mineral composition, organic matter type and content vary greatly, affecting the well logging response characteristics and resulting in prediction errors. Summary of the Invention

[0006] Based on this, it is necessary to provide a method for predicting total organic carbon in formations based on parameter fusion to address the above technical problems.

[0007] An embodiment of the present invention provides a method for predicting total organic carbon in formations based on parameter fusion, including: Obtaining the measured value of total organic carbon in the formation of the exploration well and multiple well logging parameters, and taking the well logging parameters with a correlation coefficient greater than a set threshold between each well logging parameter and the measured value of total organic carbon in the formation as the characteristic well logging parameters of total organic carbon in the formation; Connecting a feature mapping layer and a fusion layer between the input layer and the first hidden layer of the deep neural network to obtain an improved deep neural network; Inputting the characteristic well logging parameters into the improved deep neural network, calculating the ∆logR parameter through the input layer; performing a non-linear transformation on the ∆logR parameter through the feature mapping layer to map the ∆logR parameter into well logging features, and performing weighted summation fusion on the well logging features and the characteristic well logging parameters through the fusion layer to obtain fusion data; making a prediction based on the fusion data to obtain the predicted value of total organic carbon in the formation; Training the improved deep neural network according to the error between the predicted value of total organic carbon in the formation and the measured value of total organic carbon in the formation to obtain the trained improved deep neural network; During the process of predicting total organic carbon in the formation, obtaining multiple well logging parameters of the exploration well to be measured, selecting characteristic well logging parameters from the multiple well logging parameters and inputting them into the trained improved deep neural network to obtain the prediction result of total organic carbon in the exploration well to be measured.

[0008] Optionally, calculating the ∆logR parameter through the input layer of the improved deep neural network specifically includes: Calculating the ∆logR parameter through the input layer of the improved deep neural network based on the following formula: Among them, RT is the resistivity, is the resistivity baseline, AC is the acoustic travel time, is the acoustic travel time baseline.

[0009] Optionally, the ∆logR parameter is non-linearly transformed through a feature mapping layer to map the ∆logR parameter into a logging feature, specifically including The feature mapping layer adopts a multi-layer perceptron structure and processes the ∆logR parameter using non-linear transformation; The weighted sum and activation function operation are performed on the ∆logR parameter after non-linear transformation through the hidden layer neurons of the multi-layer perceptron structure to map the ∆logR parameter into a logging feature.

[0010] Optionally, the improved deep neural network includes multiple hidden layers, and an output layer is connected in series after the last hidden layer; the activation function of the output layer of the improved deep neural network is the scaled exponential linear unit; The fusion data is sequentially processed layer by layer through the multiple hidden layers of the improved deep neural network, and the extracted features are combined at the last hidden layer to obtain global features; the scaled exponential linear unit operation is performed on the global features through the output layer of the improved deep neural network for prediction to obtain the predicted value of the total organic carbon of the formation.

[0011] Optionally, according to the error between the predicted value of the total organic carbon of the formation and the measured value of the total organic carbon of the formation, the improved deep neural network is trained, specifically including: An adaptive moment estimation optimization function is constructed according to the predicted value of the total organic carbon of the formation and the measured value of the total organic carbon of the formation to train the improved deep neural network.

[0012] Optionally, it further includes: for each characteristic logging parameter, the SVM parameters of the support vector machine algorithm are determined according to the parameter distribution characteristics of the characteristic logging parameter to train the classifier of the support vector machine algorithm; The outliers deviating from the normal data distribution pattern are identified and removed through the trained classifier of the support vector machine algorithm.

[0013] Optionally, it further includes that before inputting the characteristic logging parameter into the improved deep neural network, exponential weighted moving average is used to smooth the characteristic logging parameter, and the formula is expressed as: Among them, y[t] represents the smoothed value at the t-th moment, x[t] represents the observed value at the t-th moment, is the smoothing coefficient.

[0014] The above-mentioned method for predicting total organic carbon in formation based on parameter fusion provided by the embodiments of the present invention has the following beneficial effects compared with the prior art: According to the correlation coefficients between various logging parameters and the measured values of total organic carbon in the formation, the present invention uses the logging parameters with correlation coefficients greater than the set threshold as the characteristic logging parameters of total organic carbon in the formation, and fuses the ∆logR parameter and the characteristic logging parameters through parameter mapping and weighted summation fusion, enhancing the decoupling ability of the multi-factor coupling effect, and being able to remove the logging parameters with weak or no correlation with the prediction of total organic carbon in the formation, and solving the problem that different lithologies have different influences on logging responses in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of a method for predicting total organic carbon in formation based on parameter fusion provided in an embodiment Figure 2 It is an implementation flowchart of a method for predicting total organic carbon in formation based on parameter fusion provided in an embodiment; Figure 3 It is a DNN model framework diagram of a method for predicting total organic carbon in formation based on parameter fusion provided in an embodiment; Figure 4 It is a TOC prediction scatter plot of a DNN model of a method for predicting total organic carbon in formation based on parameter fusion provided in an embodiment; Figure 5 It is a TOC prediction scatter plot of a DNN that fuses the ∆logR parameter of a method for predicting total organic carbon in formation based on parameter fusion provided in an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] Currently, the following problems exist in the field of TOC prediction: (1) Traditional methods are complex and costly: Traditional complex pyrolysis experimental methods that rely on core or cuttings samples not only take time and effort but also come with high costs, which greatly limits their application in extensive exploration activities.

[0018] (2) Limited prediction accuracy: Although many patents and research results have proposed TOC prediction methods based on logging data, these methods still have difficulty meeting the requirements of high-precision exploration in terms of prediction accuracy, especially under complex geological conditions.

[0019] (3) Single parameter and strong subjectivity: Existing prediction methods mostly rely on single or limited logging parameters, such as ∆logR parameter, deep lateral resistivity, and natural gamma, etc. This may lead to the prediction results being greatly affected by parameter selection, and there is a lack of systematic standards for dealing with outliers, with strong human subjectivity, which affects the stability and reliability of the prediction results.

[0020] In one embodiment, a method for predicting total organic carbon in formation based on parameter fusion is provided, as Figure 1 shown. This method includes: 1. Obtain the measured value of total organic carbon in the formation of the exploration well and multiple logging parameters, and use the logging parameters with a correlation coefficient greater than the set threshold between each logging parameter and the measured value of total organic carbon in the formation as the characteristic logging parameters of total organic carbon in the formation.

[0021] For each characteristic logging parameter, determine the SVM parameters of the support vector machine algorithm according to the parameter distribution characteristics of the characteristic logging parameter to train the classifier of the support vector machine algorithm. Identify and remove outliers deviating from the normal data distribution pattern through the trained classifier of the support vector machine algorithm.

[0022] 2. As Figure 3 shown, connect a feature mapping layer and a fusion layer between the input layer and the first hidden layer of the deep neural network to obtain an improved deep neural network.

[0023] The feature mapping layer adopts a multi-layer perceptron structure and processes the ∆logR parameter using a non-linear transformation. Perform weighted summation and activation function operations on the ∆logR parameter after non-linear transformation through the hidden layer neurons of the multi-layer perceptron structure to map the ∆logR parameter into logging features.

[0024] 3. Input the characteristic logging parameters into the improved deep neural network, calculate the ∆logR parameter through the input layer; perform non-linear transformation on the ∆logR parameter through the feature mapping layer to map the ∆logR parameter into logging features, and perform weighted summation fusion on the logging features and the characteristic logging parameters through the fusion layer to obtain fusion data. Make a prediction based on the fusion data to obtain the predicted value of total organic carbon in the formation.

[0025] The improved deep neural network includes multiple hidden layers, and an output layer is connected in series after the last hidden layer. The activation function of the output layer of the improved deep neural network is a scaled exponential linear unit. Process the fusion data layer by layer through the multiple hidden layers of the improved deep neural network, and combine the extracted features in the last hidden layer to obtain global features. Perform scaled exponential linear unit operations on the global features through the output layer of the improved deep neural network for prediction to obtain the predicted value of total organic carbon in the formation.

[0026] 4. Train the improved deep neural network according to the error between the predicted value and the measured value of the total organic carbon in the formation, and obtain the trained improved deep neural network.

[0027] Construct an adaptive moment estimation optimization function based on the predicted value and the measured value of the total organic carbon in the formation, and train the improved deep neural network.

[0028] 5. During the prediction of the total organic carbon in the formation, obtain multiple logging parameters of the exploration well to be measured, select characteristic logging parameters from the multiple logging parameters and input them into the trained improved deep neural network to obtain the prediction result of the total organic carbon in the exploration well to be measured.

[0029] As Figure 2 shown, a specific embodiment of the present invention is provided: Step 1: Data selection and preprocessing Data collection: First, collect detailed logging data from multiple exploration wells, including but not limited to logging parameters such as acoustic travel time (AC), natural gamma ray (GR), resistivity (RT), density (DEN), and neutron lifetime (CNL). At the same time, ensure that each well has the corresponding measured TOC value as the labeled data.

[0030] Correlation analysis: Use statistical analysis methods (such as Pearson correlation coefficient) to calculate the correlation between each logging parameter and the TOC value. Based on the correlation analysis results, screen out the logging parameters that have a significant correlation with TOC (such as AC, GR, RT, DEN, and CNL in this example) as the core input data for constructing the machine learning model. Specifically, the correlation coefficients of AC and GR both exceed 0.5, showing a strong correlation; RT follows closely with 0.47, indicating a strong association; the 0.33 of DEN shows a moderate correlation; while the -0.34 of CNL reveals its negative correlation with TOC, and this unique feature is also taken into consideration to comprehensively capture the multi-dimensional relationships between data.

[0031] Data preprocessing: Use the Support Vector Machine (SVM) algorithm to detect and process outliers in the selected logging data. Specifically, train an SVM classifier to identify and remove those outliers that significantly deviate from the normal data distribution pattern. This process can be achieved by setting appropriate SVM parameters (such as kernel function, penalty coefficient, etc.) to ensure the accuracy and efficiency of outlier detection.

[0032] After removing outliers using the SVM algorithm, the performance of the XGBoost model (Extreme Gradient Boosting) has been significantly improved. The MAE has decreased from 1.6911 to 1.4271, the RMSE has decreased from 3.4876 to 2.4503, and the R 2 has increased from 0.5484 to 0.7609. These metrics all indicate that the prediction accuracy and stability of the model have been significantly improved.

[0033] Similarly, on the Deep Neural Network (DNN) model, the effect of removing outliers using the SVM algorithm is also significant. The MAE of the DNN model has slightly decreased from 1.0551 to 0.9806, the RMSE has slightly decreased from 1.8857 to 1.81, and the R 2 has slightly increased to 0.8695. Although the improvement in some metrics is small, overall, the prediction performance of the model remains at a high level and is more stable and reliable. The present invention also uses exponentially weighted moving average (EWMA) to smooth the data. The EWMA formula is as follows:

[0034] where y[t] represents the smoothed value at the t-th moment. x[t] represents the observed value at the t-th moment. is the smoothing coefficient, which determines the weight of the observed value in the calculation. The value range of is [0,1], and the value in this example is 0.3.

[0035] Step 2: Construction and training of the DNN model Construct the DNN model: This model is carefully designed and consists of an input layer, four hidden layers, and an output layer. Among them, each hidden layer contains 64 and 128 neurons respectively. Such a structure provides the model with rich processing capabilities. The input layer is responsible for receiving the preprocessed logging data, which is the basis for the model to learn and predict.

[0036] The hidden layer is the core part of the DNN model. They process the input data layer by layer through non-linear transformations and extract the features in the data. These features start from low-level and local information, such as the most basic and direct elements in the data, and then gradually deepen layer by layer. Through continuous learning and processing, they are gradually combined and refined into high-level and global features. This process of feature extraction and combination is one of the unique advantages and characteristics of the DNN model.

[0037] At the output layer of the model, the high-level features processed by the hidden layer are used to predict the TOC value. To enhance the model's non-linear expression ability, SELU (Scaled Exponential Linear Unit) is selected as the activation function, which can help the model better capture the complex patterns in the data.

[0038] In terms of optimization, Adam (Adaptive Moment Estimation) is selected as the optimization function. It combines the advantages of the momentum method and the RMSProp method, and can efficiently update the model's weights, enabling the model to converge quickly during training.

[0039] The multi-layer architecture of the DNN not only enables the model to extract and combine features layer by layer and systematically, but also greatly enhances the model's expression ability and generalization ability. This transformation of features from local to global and from low-level to high-level enables the DNN to demonstrate excellent performance when dealing with complex tasks. Whether it is for in-depth data mining or accurate prediction of unknown data, the DNN has shown strong strength and potential.

[0040] Comparison model construction: Four comparison machine learning models are adopted, including K-Nearest Neighbors (KNN), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and XGBoost. Each model uses the preprocessed logging data as input and the measured TOC value as output for training. Set the number of neighbors in the KNN model; set the number of trees and the tree depth to 8 in the RF model; set the learning rate to 0.01 and the tree depth to 10 in the GBDT and XGBoost models, etc. The setting of these parameters can be optimized by methods such as cross-validation or grid search.

[0041] Cross-validation: The above models are trained and validated using a ten-fold cross-validation strategy. The specific operation is to randomly divide the preprocessed data into ten subsets. Each time, nine of these subsets are selected as the training set, and the remaining one subset is used as the validation set for model evaluation. Repeat this process ten times, each time selecting a different validation set to comprehensively evaluate the generalization ability and prediction performance of each model.

[0042] Step 3: ∆logR parameter fusion and prediction ∆logR parameter calculation: Calculate the traditional ∆logR parameter based on the logging data, which reflects the abundance information of organic matter in the formation. The calculation formula of the ∆logR parameter is as follows:

[0043] Where RT is the resistivity, is the resistivity baseline, AC is the acoustic time difference, The acoustic time difference baseline Feature fusion and model improvement: Construct a feature mapping layer based on ∆logR: Design a special feature mapping layer and input the calculated ∆logR parameters into this layer. In the feature mapping layer, nonlinear transformation is used to process the ∆logR parameters. The Multilayer Perceptron (MLP) structure is used to map the ∆logR parameters into a set of new features through the weighted summation of hidden layer neurons and activation function operations. This can enhance the nonlinear relationship between the ∆logR parameters and other logging parameters and tap into their deeper feature information, rather than simply using them as additional inputs.

[0044] Improved DNN model structure: After the input layer of the DNN model, the feature mapping layer based on ∆logR constructed above is connected. Then, the output of the feature mapping layer is fused with other pre-processed logging parameters (acoustic time difference AC, natural gamma GR, resistivity RT, density DEN and neutron lifetime CNL) in a new fusion layer. The fusion layer adopts a weighted sum fusion method to enable the features of the ∆logR parameter after feature mapping and other logging parameters to achieve more effective information interaction and integration within the model, and together serve as the input of the subsequent hidden layer.

[0045] Joint training and optimization: Joint training is performed on the improved DNN model, and the parameters of all layers in the model are optimized at the same time (including the parameters of the feature mapping layer and the fusion layer, as well as the parameters of the hidden layer, input layer, and output layer of the original DNN model). During the training process, the back propagation algorithm is used to adjust the values ​​of each parameter in the model according to the error between the predicted value and the measured value to minimize the error. A suitable optimizer (such as the Adam optimizer) is used to accelerate model convergence and ensure that the model can fully learn the effective information in the fusion features, thereby improving the prediction performance of the model.

[0046] Through the above-mentioned improved fusion process, the ∆logR parameter can be better integrated with the DNN model, giving full play to its role in improving the prediction accuracy of the model and providing more accurate and reliable support for the prediction of total organic carbon content in the formation.

[0047] Figure 4 is the TOC prediction scatter plot of the DNN model, and Figure 5 This is a TOC prediction scatter plot of DNN fusion ∆logR parameters. It can be seen that Figure 5 The scatter distribution is higher than Figure 4The scatter distribution degree, so after integrating the ∆logR parameter, the prediction performance of the model has been significantly improved. Taking the DNN model as an example, its mean absolute error MAE has been reduced to 0.4193, and the root mean square error RMSE has also been reduced to 1.01196. At the same time, the coefficient of determination R 2 is as high as 0.9457. The changes in these indicators fully illustrate the significant improvement effect of the fusion strategy on the prediction accuracy of the DNN model.

[0048] In addition, the prediction effect of the KNN model is also remarkable. After integrating the ∆logR parameter, the MAE of the KNN model has been significantly reduced from 1.6772 to 0.9762, the RMSE has also been reduced from 3.1395 to 1.7875, and at the same time, R2 has been increased from 0.6427 to 0.8307. This series of improvements fully proves the significant effect of integrating the traditional ∆logR parameter on improving the model performance.

[0049] Through the above specific implementation manners, the TOC content prediction method based on multi-parameter logging data and machine learning algorithms of the present invention can be effectively implemented, realizing high-precision prediction of the total organic carbon content in the formation, and providing reliable and efficient geological parameter support for oil exploration and development. By constructing a variety of machine learning models and performing cross-validation, the model with the best performance is selected for prediction, significantly improving the prediction accuracy of the TOC content. During the cross-validation process, the ten-fold cross-validation strategy is adopted, which not only comprehensively evaluates the generalization ability of each model, but also improves the adaptability and universality of the model under different geological conditions through model optimization and fusion, providing strong support for high-precision exploration under complex geological conditions.

[0050] The above embodiments only represent several implementation manners of the present invention, and their descriptions are relatively specific and detailed, but should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention.

Claims

1. A method for predicting total organic carbon in formation based on parameter fusion, characterized in that Including: Obtain the measured total organic carbon (TOC) value of the formation of the exploration well and multiple logging parameters, and use the logging parameters with a correlation coefficient greater than a set threshold with the measured TOC value of the formation as the characteristic logging parameters of the formation TOC; Connect a feature mapping layer and a fusion layer between the input layer and the first hidden layer of the deep neural network to obtain an improved deep neural network; Input the characteristic logging parameters into the improved deep neural network, calculate the ∆logR parameter through the input layer; perform a non-linear transformation on the ∆logR parameter through the feature mapping layer to map the ∆logR parameter into a logging feature, and perform weighted sum fusion on the logging feature and the characteristic logging parameters through the fusion layer to obtain fusion data; make a prediction based on the fusion data to obtain the predicted value of the formation TOC; Train the improved deep neural network according to the error between the predicted value of the formation TOC and the measured value of the formation TOC to obtain the trained improved deep neural network; During the prediction of the formation TOC, obtain multiple logging parameters of the exploration well to be measured, select the characteristic logging parameters from the multiple logging parameters and input them into the trained improved deep neural network to obtain the prediction result of the formation TOC of the exploration well to be measured.

2. The method for predicting total organic carbon of formation based on parameter fusion according to claim 1, characterized in that The calculation of the ∆logR parameter through the input layer of the improved deep neural network specifically includes: Calculating the ∆logR parameter through the input layer of the improved deep neural network based on the following formula: where RT is the resistivity, is the resistivity baseline, and AC is the acoustic travel time difference, is the acoustic travel time difference baseline.

3. The total organic carbon prediction method of a formation based on parameter fusion according to claim 1, wherein The non-linear transformation of the ∆logR parameter through the feature mapping layer to map the ∆logR parameter into a logging feature specifically includes The feature mapping layer adopts a multi-layer perceptron structure and processes the ∆logR parameter using non-linear transformation; Perform weighted sum and activation function operations on the ∆logR parameter after non-linear transformation through the hidden layer neurons of the multi-layer perceptron structure to map the ∆logR parameter into a logging feature.

4. A method for predicting total organic carbon in a formation based on parameter fusion according to claim 1, characterized in that The improved deep neural network includes multiple hidden layers, and an output layer is connected in series after the last hidden layer; the activation function of the output layer of the improved deep neural network is a scaled exponential linear unit; Process the fusion data layer by layer through multiple hidden layers of the improved deep neural network, and combine the features extracted in the last hidden layer to obtain global features; Perform an operation of the scaled exponential linear unit on the global features through the output layer of the improved deep neural network for prediction to obtain the predicted value of the formation TOC.

5. The method for predicting total organic carbon of formation based on parameter fusion according to claim 1, characterized in that, Training the improved deep neural network according to the error between the predicted value of the formation TOC and the measured value of the formation TOC specifically includes: Construct an adaptive moment estimation optimization function according to the predicted value of the formation TOC and the measured value of the formation TOC, and train the improved deep neural network.

6. The method for predicting total organic carbon of formation based on parameter fusion according to claim 1, characterized in that, It also includes: for each characteristic logging parameter, determine the SVM parameters of the support vector machine algorithm according to the parameter distribution characteristics of the characteristic logging parameter to train the classifier of the support vector machine algorithm; Identify and eliminate outliers deviating from the normal data distribution pattern through the trained classifier of the support vector machine algorithm.

7. The method for predicting total organic carbon of formation based on parameter fusion according to claim 1, characterized in that, It also includes that before inputting the characteristic logging parameters into the improved deep neural network, the exponential weighted moving average is used to smooth the characteristic logging parameters, and the formula is expressed as: Among them, y[t] represents the smoothed value at the t-th moment, and x[t] represents the observed value at the t-th moment. is the smoothing coefficient.

Images