Post-stack seismic inversion through probabilistic neural networks and deep forward neural networks

Seismic inversion methods are routinely used to estimate attributes such as P-impedance, S-impedance, density, P-wave and S-wave velocity, and elastic impedances from seismic and well log data. These attributes help to understand subsurface lithology and fluid content. There are several types of seismic inversion methods available in which model-based inversion has got more attention. In the present study, the model-based inversion and geostatistical methods namely probabilistic neural network (PNN), multi-layer feed-forward neural network (MLFN) and radial basis function neural network (RBFNN) are implemented to estimate acoustic impedance, porosity, density, gamma-ray volume and P-wave velocity in the inter-well region of the F-3 block, Netherlands. The aim of the study is to predict a number of petrophysical parameters and demonstrate how they can be used to interpret seismic reflection data. A comparative study is also performed to estimate the best suitable method to predict various petrophysical parameters. The assessment of inverted results demonstrates that the impedance in the region varies from 2550 m/s*g/cc to 6150 m/s/*g/cc, which is relatively low and indicates the presence of loose formation in the area. The correlation between synthetic seismic trace and initial seismic trace is estimated to be 0.98, and the synthetic relative error as 0.19, indicating excellent algorithm output. Further, the assessment of the geostatistical methods is carried out to predict petrophysical parameters in conjunction with Multi-Attribute Transform Analysis. The analysis shows that among the geostatistical methods, the PNN provides slightly better results. The petrophysical parameters that are estimated in this study are porosity, density, gamma-ray, and P-wave velocity volumes. These parameters reinforce the interpretation of seismic data, which is a significant step in any project of exploration and production. The technique is first implemented to the composite traces near to the well location, and the findings are analysed with well log data. Second, the entire seismic section is inverted for porosity, density, gamma-ray, and P-wave velocity after achieving satisfactory outcomes in the first step. The assessment demonstrates very high-resolution subsurface petrophysical parameters which strengthen the seismic data interpretation.

The deposition and mixing of carbonates and siliciclastics in the Cisco Group of the Eastern Shelf of the Permian Basin are complicated by the temporal overlap between icehouse eustatic sea-level oscillations and fluctuations in sediment influx due to the rejuvenation of the Ouachita fold belt. Previous investigators have used well-log correlation as the primary tool in their interpretations of the area’s reciprocal depositional model, but well-log correlation alone cannot explain the full range of spatial lithology variations in the system. To better understand the lithology variation in the area, we use an integrated technique that combines wireline log information from 17 wells with 625 km2 3D seismic data through poststack seismic inversion, probabilistic neural networks (PNNs), and Bayesian classification. We use deterministic matrix inversion to derive lithology classes from well logs. Crossplot analyses reveal that the acoustic impedance and neutron porosity log pair can be used to differentiate lithologies. We perform model-based poststack inversion to generate a P-impedance volume and use PNNs to generate a neutron porosity volume. We combine these volumes through supervised Bayesian classification to generate lithology probability volumes for each lithology and a most probable lithology volume throughout the seismic data. The lithology volumes highlight the dominant lithologies (carbonate, shale, sand, and mixed) that allowed the interpretation of major carbonate platforms, sand-to-shale ratio variations, carbonate buildups between wells, and channel fill lithologies. Our semiautomated lithology detection workflow applies to regional studies and is also valid for reservoir-scale studies to determine variations in lithologies.

Reservoir facies modeling is an important way to express the sedimentary characteristics of the target area. Conventional deterministic modeling, target-based stochastic simulation, and two-point geostatistical stochastic modeling methods are difficult to characterize the complex sedimentary microfacies structure. Multi-point geostatistics (MPG) method can learn a priori geological model and can realize multi-point correlation simulation in space, while deep neural network can express nonlinear relationship well. This article comprehensively utilizes the advantages of the two to try to optimize the multi-point geostatistical reservoir facies modeling algorithm based on the Deep Forward Neural Network (DFNN). Through the optimization design of the multi-grid training data organization form and repeated simulation of grid nodes, the simulation results of diverse modeling algorithm parameters, data conditions and deposition types of sedimentary microfacies models were compared. The results show that by optimizing the organization of multi-grid training data and repeated simulation of nodes, it is easier to obtain a random simulation close to the real target, and the simulation of sedimentary microfacies of different scales and different sedimentary types can be performed.

Estimation of gas hydrate saturation in the Ulleung basin using seismic attributes and a neural network

Neural Network (NN) system is an information-processing that has characteristics similar to the neural network in living beings. A model of Neural Network is used for classification is Probabilistic Neural Network (PNN). PNN structured by four layers, the input layer, layer pattern, the summation layer and output layer. One of classification problems that can be solved by PNN is a classification of Diabetes Mellitus’s status. Diabetes mellitus is a chronic disease that occurs either when the pancreas does not produce enough insulin or when the body cannot effectively use the insulin produced. To facilitate the classification of diabetes mellitus, it is used a software-based Graphical User Interface (GUI) of Matlab to build a software of PNN. GUI that is formed can do PNN classification and predict the status of one’s Diabetes Mellitus. PNN structure that is formed resulting the highest accuracy 0.9143548 on the training process and 0.919512 on the testing process obtained by the percentage of training data than testing data by 90%:10% with holdout accuracy evaluation method, and a smoothing value of 1. This classification resulting 23 patients were classified as negative diabetes and 18 patients were classified as positive diabetes. Keywords : Neural Network, Probabilistics Neural Network, diabetes mellitus, GUI, holdout, smoothing parameter.

Deep learning systems for automatic diagnosis of infant cry signals

Post-stack seismic inversion utilizes post-stack seismic data along with well log data to estimate acoustic impedance. Post-stack seismic inversion is very fast compared to other pre-stack seismic inversion methods and provides a high-resolution subsurface image. This chapter discusses several types of post-stack seismic inversion methods namely model-based inversion, colored inversion, sparse spike inversion, and band-limited inversion. The chapter also includes the synthetic as well as real data examples of above seismic inversion methods.

Abstract. Time series data in practical applications always contain missing values due to sensor malfunction, network failure, outliers etc. In order to handle missing values in time series, as well as the lack of considering temporal properties in machine learning models, we propose a spatiotemporal prediction framework based on missing value processing algorithms and deep recurrent neural network (DRNN). By using missing tag and missing interval to represent time series patterns, we implement three different missing value fixing algorithms, which are further incorporated into deep neural network that consists of LSTM (Long Short-term Memory) layers and fully connected layers. Real-world air quality and meteorological datasets (Jingjinji area, China) are used for model training and testing. Deep feed forward neural networks (DFNN) and gradient boosting decision trees (GBDT) are trained as baseline models against the proposed DRNN. Performances of three missing value fixing algorithms, as well as different machine learning models are evaluated and analysed. Experiments show that the proposed DRNN framework outperforms both DFNN and GBDT, therefore validating the capacity of the proposed framework. Our results also provides useful insights for better understanding of different strategies that handle missing values.

شکنندگی یکی از مهم‌ترین خصوصیات سنگ می­باشد. تاکنون روشی مستقیم و استاندارد برای به‌دست آوردن شاخص شکنندگی ارائه نشده است. از شاخص شکنندگی بر اساس پارامترهای مقاومت فشارشی تک‌محوره و مقاومت کششی جهت ارزیابی شکنندگی استفاده می­شود. هدف از این مطالعه بررسی مفاهیم شکنندگی ارائه‌شده توسط محققین و استفاده از روش وارون‌سازی لرزه­ای، آنالیز چندنشانگری و شبکة عصبی در میدان ویچررنج (Whicher-Range) واقع در حوضة پرت (Perth Basin) استرالیای غربی می‌باشد. به این منظور دو چاه و یک مقطع لرزه­ای از میدان ویچررنج مورداستفاده قرار گرفت. کمترین شاخص شکنندگی در چاه اول و دوم که با فاصله 1 کیلومتری از هم حفاری شدند، به‌ترتیب با استفاده از مفهوم برابر 69/1 و 67/1 مگاپاسکال که بیانگر شکنندگی کم و بیشترین مقدار آن برابر 78/39 و 15/48 مگاپاسکال که بیانگر شکنندگی زیاد و از نظر حفاری در رتبة حفاری بسیار سخت می­باشد. در مرحلة بعد با استفاده از استخراج موجک و رسم مدل زمین‌شناسی، مدل‌سازی اولیه انجام و سپس امپدانس صوتی حاصل از فرایند وارون‌سازی در کنار دیگر نشانگرها در راستای ساخت ترکیب بهینه از نشانگرها جهت تخمین شاخص شکنندگی استفاده شد. از سه نوع الگوریتم شبکة عصبی جهت تخمین پارامتر هدف از ترکیب بهینه نشانگرهای موجود استفاده می­شود که به‌کمک نرم‌افزار همپسون–راسل و نشانگرهای بهینه­ای، الگوریتم شبکة عصبی تهیه و در نهایت آنالیز چند نشانگری با سه الگوریتم شبکة عصبی مقایسه می‌شود و نتایج بیانگر ضریب همبستگی بیشتر شبکة عصبی احتمالی نسبت به آنالیز چند نشانگری جهت تعیین شاخص شکنندگی است.

This study investigates the generalization performance of deep learning (DL) models for solving the inverse kinematics (IK) problem in a 2-degrees of freedom (DOF) revolute-prismatic (RP) robotic manipulator. The goal is to evaluate how effectively different neural architectures predict joint configurations from end-effector positions across diverse workspace regions. Two training strategies were used: quadrant-based and full workspace training. To improve robustness, k-fold cross-validation (CV) was applied to the deep feedforward neural network (DFNN). The models evaluated include DFNN with k-fold CV using 2-input 1-output and 2-input 2-output formulations and without k-fold CV, long short-term memory (LSTM), and gated recurrent unit (GRU). Performance was tested on predefined Square and Circle paths within each quadrant and the full workspace. The DFNN with k-fold CV (2-input, 1-output) consistently achieved the lowest Cartesian deviation errors- for instance, 0.289mm in Q1, 0.410mm in Q2, 0.508mm in Q3, and 0.715mm in Q4 on Square path. Similar trends were observed on the Circle path, with errors of 0.312mm, 0.366mm, 0.438mm and 0.662mm in Q1 to Q4 respectively. In full workspace testing, it maintained strong performance with 1.594mm (Square) and 2.084mm (Circle) errors. In contrast, DFNN with k-fold CV (2-input, 2-output), without k-fold CV, LSTM and GRU exhibited significantly higher errors. These findings demonstrate that the k-fold CV-based DFNN with single-output formulation, achieves high accuracy and generalization and also capable of handling singularities and ambiguity in joint solutions.

Bayesian neural networks for flight trajectory prediction and safety assessment

Time-history responses of the bridge induced by the moving vehicle provide crucial information for bridge design, operation, maintenance, etc. As inspired by this, this work attempts to provide a new paradigm for vehicle–bridge interaction (VBI) by highlighting the comparison of different deep learning algorithms applied to the prediction of time-history responses of the bridge under vehicular loads. Particularly, three deep learning architectures with few and measurable input features developed by using fully-connected feedforward neural network, long short-term memory (LSTM) network, and convolutional neural network (CNN) are proposed on the basis of the governing equation of bridge vibrations. Three VBI systems with various vehicle models are developed and further validated to produce reliable training data. To examine the accuracy of the predictive models, two advanced metrics are exploited for time-history estimate. Moreover, the proposed deep learning models are comprehensively investigated through a parametric study on the influential factors associated with the VBI system and network architecture. The results show that deep feedforward neural network (DFNN), LSTM network, and CNN can be applied in VBI analysis to estimate the bridge time-history response. The three neural networks have comparable prediction accuracies. When considering the irregularity excitation, CNN is found to be the most efficient predictive model, while DFNN needs the least training time under perfect bridge surface condition.

3D seismic attributes and well logs were used to estimated porosity and water saturation in the Asmari Formation in the Dezful Embayment, SW Iran. For this purpose, at first, the 3D seismic volume was inverted base on the model, to obtain acoustic impedance cube. Afterward, the impedance and other attributes extracted from seismic volume were analyzed by multiple attribute regression transform and neural networks to predict porosity and water saturation between wells. Then linear or non–linear combinations of attributes performed for porosity and water saturation prediction. The result shows that the match between the actual and predicted porosity and water saturation improved; using only a single attribute to the derived multi attribute transforms and neural networks model. Based on the results of neural networks, the highest cross–correlation was observed between seismic attributes and the observed target logs between seven wells in the study area. Based on our study, the cross–correlation between actual and predicted porosity and water saturation increased and reached 93% and 90% respectively in the case of using probabilistic neural networks (PNN). Finally, according to the cross–validation results, PNN neural networks are used for porosity and water saturation prediction. We carry out porosity and water saturation slicing from the Asmari Formation for display lateral and vertical heterogeneities, and the result provided a reliable picture from subsurface formations. Porosity maps distribution shows the western portion of the structure is highly porous and should be considered for further exploration and development purposes. A possible reason for this high porosity in the western portion of the studied formation is the presence of sand layers, especially in zone 2.Note that sand volume increased towards west and northwest in direction of shadegan and Ahvaz fields and decreased towards east and southeast to Rag–e–Sefid field. Based on the result between acoustic impedance and core, changes in acoustic impedance were related to changes in the geological nature of the Asmari reservoir in the field. Accordingly, seismic inversion is a powerful tool for studying the details of lithology and sedimentary facies.

Advanced Probabilistic Neural Network for the Prediction of Concrete Strength Doo Kie Kim1, Seong Kyu Chang1 and Sang Kil Chang1 Summary Accurate and realistic strength estimation before the placement of concrete is highly desirable. In this study, the advanced probabilistic neural network (APNN) was proposed to reflect the global probability density function by summing the heterogeneous local probability density function automatically determined in the individual standard deviation of variables. Currently, the estimation of the compressive strength of concrete is performed by a probabilistic neural network (PNN) on the basis of concrete mix proportions, and the PNN is improved by the iteration method. However, an empirical method has been incorporated to specify the smoothing parameter in the PNN technique, causing significant uncertainty in the estimation results. In addition, the probability density function (PDF) is the sum of homogeneous multivariate Gaussian distribution because only one global smoothing parameter is used. The APNN was applied to predict the compressive strength of concrete using actual test data from two concrete companies, and the estimated results of APNN were compared with those of PNN. APNN showed better results than PNN in predicting the compressive strength of concrete and provided the promising probabilistic viewpoints by using the individual standard deviation in a variable.

Background Chondroitin sulphate proteoglycans (CSPGs) can cause the termination of ocular dominance plasticity in the visual cortex.Recently,protein tyrosine phosphatase σ (PTPσ) has been identified as a receptor that inhibits CSPGs.However,whether PTPσ and its downstream molecules participate in the reactivation of ocular dominance plasticity in adult visual cortex has not been studied.Objective The present study was to investigate the changes in the expression of the PTPσ,probabilistic neural networks (PNNs),and molecules downstream of PNN,such as N-cadherin/β-catenin,after the reactivation of adult visual cortical plasticity.Methods Fifty-four SPF Long Evans rats were grouped according to different postnatal week (PW) as the PW1 (6 rats),PW3 (6 rats),PW5 (6 rats),PW7 (24 rats),and PW9 (12 rats) groups,and the upper and lower eyelids were sutured in the 12 rats from the PW7 group for 14 days to establish the binocular plasticity reactivation models.Expression of PTPσ and PNNs in the rat visual cortex was detected using immunochemistry,and changes of PTPσ mRNA,N-cadherin mRNA and β-catenin mRNA expression in the rat visual cortex with plasticity reactivation were assessed by real-time fluorescence quantitative PCR (RT Q-PCR).The use of animals followed the Regulations for the Administration of Affairs Concerning Experimental Animals by State Science and Technology Commission.Results The expression level of PTPσ mRNA was significantly higher in the PW9 group than that of the binocular plasticity reactivation models and the PW7 group (t =1.965,3.526,P＜0.01).The staining of the rat visual cortex for PTPσwas localized to the cellular membrane,cytoplasm and axon.Cell densities of the PW9 group in the Ⅱ-Ⅲ layer,Ⅳ layer and Ⅴ-Ⅵ layer of the visual cortex were elevated in the PW9 rats compared with the PW7 rats (t =24.593,23.444,13.556,P＜0.01) and rats from the binocular plasticity reactivation model (t =44.111,43.000,16.556,P＜0.01).Cell densities for PNNs in the Ⅳ and Ⅴ-Ⅵ layers were significantly increased in the PW9 rats in comparison with the PW7 rats (t=1.926,P＜0.01 ;t=1.370,P＜0.05),but the cell density in the Ⅱ-Ⅱ layer has no statistical significance (t=0.889,P＞0.05).However,cell densities for PNNs in the Ⅱ-Ⅲ and Ⅳ layers in the binocular plasticity reactivation models were lower than those of the PW9 rats (t =2.556,4.585,P＜0.01).Compared with PW1 rats,the expression levels of the N-cadherin mRNA in the PW3,PW5,PW7,PW9 rats were lower (t =28.932,28.988,27.083,28.908,P＜0.01),but those in the PW7 rats were enhanced in comparison with the PW3 rats,PW5 rats and PW9 rats (t =1.848,1.904,1.825,P＜0.01).No significant difference was seen in the expression of the N-cadherin mRNA between the PW9 rats and rats from the binocular plasticity reactivation model (t =0.072,P＞0.05).A statistically significant increase was found in the β-catenin mRNA expression in the PW1 rats compared with the PW3,PW5,PW7 and PW9 rats (t =3.918,3.534,2.645,4.652,P＜ 0.0 1),as well as between rats from the binocular plasticity reactivation model and the PW9 rats (t =0.570,P＜0.01).Conclusions PTPr,PNNs and β-catenin are involved in the reactivation of ocular dominance plasticity in the adult visual cortex. Key words: Visual cortex; Protein tyrosine phosphatase σ; Plasticity; Binocular form-deprivation