Probabilistic seismic inversion based on physics-guided deep mixture density network

Deterministic inversion based on deep learning has been widely utilized in model parameters estimation.Constrained by logging data,seismic data,wavelet and modeling operator,deterministic inversion based on deep learning can establish nonlinear relationships between seismic data and model parameters.However,seismic data lacks low-frequency and contains noise,which increases the non-uniqueness of the solutions.The conventional inversion method based on deep learning can only establish the deterministic relationship between seismic data and parameters,and cannot quantify the uncertainty of inversion.In order to quickly quantify the uncertainty,a physics-guided deep mixture density network(PG-DMDN)is established by combining the mixture density network(MDN)with the deep neural network(DNN).Compared with Bayesian neural network(BNN)and network dropout,PG-DMDN has lower computing cost and shorter training time.A low-frequency model is introduced in the training process of the network to help the network learn the nonlinear relationship between narrowband seismic data and low-frequency impedance.In addition,the block constraints are added to the PG-DMDN framework to improve the horizontal continuity of the inversion results.To illustrate the benefits of proposed method,the PG-DMDN is compared with existing semi-supervised inversion method.Four synthetic data examples of Marmousi II model are utilized to quantify the influence of forward modeling part,low-frequency model,noise and the pseudo-wells number on inversion results,and prove the feasibility and stability of the proposed method.In addition,the robustness and generality of the proposed method are verified by the field seismic data.

A Harmonic Approach to Handwriting Style Synthesis Using Deep Learning

The challenging task of handwriting style synthesis requires capturing the individuality and diversity of human handwriting.The majority of currently available methods use either a generative adversarial network(GAN)or a recurrent neural network(RNN)to generate new handwriting styles.This is why these techniques frequently fall short of producing diverse and realistic text pictures,particularly for terms that are not commonly used.To resolve that,this research proposes a novel deep learning model that consists of a style encoder and a text generator to synthesize different handwriting styles.This network excels in generating conditional text by extracting style vectors from a series of style images.The model performs admirably on a range of handwriting synthesis tasks,including the production of text that is out-of-vocabulary.It works more effectively than previous approaches by displaying lower values on key Generative Adversarial Network evaluation metrics,such Geometric Score(GS)(3.21×10^(-5))and Fréchet Inception Distance(FID)(8.75),as well as text recognition metrics,like Character Error Rate(CER)and Word Error Rate(WER).A thorough component analysis revealed the steady improvement in image production quality,highlighting the importance of specific handwriting styles.Applicable fields include digital forensics,creative writing,and document security.

Improved pruning algorithm for Gaussian mixture probability hypothesis density filter

With the increment of the number of Gaussian components, the computation cost increases in the Gaussian mixture probability hypothesis density(GM-PHD) filter. Based on the theory of Chen et al, we propose an improved pruning algorithm for the GM-PHD filter, which utilizes not only the Gaussian components’ means and covariance, but their weights as a new criterion to improve the estimate accuracy of the conventional pruning algorithm for tracking very closely proximity targets. Moreover, it solves the end-less while-loop problem without the need of a second merging step. Simulation results show that this improved algorithm is easier to implement and more robust than the formal ones.

An Aircraft Trajectory Anomaly Detection Method Based on Deep Mixture Density Network

The timely and accurately detection of abnormal aircraft trajectory is critical to improving flight safety.However,the existing anomaly detection methods based on machine learning cannot well characterize the features of aircraft trajectories.Low anomaly detection accuracy still exists due to the high-dimensionality,heterogeneity and temporality of flight trajectory data.To this end,this paper proposes an abnormal trajectory detection method based on the deep mixture density network(DMDN)to detect flights with unusual data patterns and evaluate flight trajectory safety.The technique consists of two components:Utilization of the deep long short-term memory(LSTM)network to encode features of flight trajectories effectively,and parameterization of the statistical properties of flight trajectory using the Gaussian mixture model(GMM).Experiment results on Guangzhou Baiyun International Airport terminal airspace show that the proposed method can effectively capture the statistical patterns of aircraft trajectories.The model can detect abnormal flights with elevated risks and its performance is superior to two mainstream methods.The proposed model can be used as an assistant decision-making tool for air traffic controllers.

Determination of physical properties for the mixtures of [BMIM]Cl with different organic solvents

Physical properties including refractive index, density, viscosity and conductivity for binary mixtures of l-butyl- 3-methyl imidazolium chloride （[BMIM]CI） and different organic solvents at 298.15 K have been investigated. Ex- cess molar volumes have been calculated and obtained data has been fitted by the Redlich-Kister equation. The density and refractive index were found to increase with increasing concentration of [BMIM]CI, however, excep- tions do exist as in the case of dimethyl sulfoxide （DMSO）/[BMIM]CI. For DMSO/[BMIM]CI, the density decreases with increasing concentration. The addition of different organic solvents was able to disrupt the interactions within mixtures, leading to free mobility of ions. The free mobility of ions has been found to enhance conductivity and decrease viscosity to varying extents in all mixtures studied. It has been observed that solubility parameters, dielectric constants and composition of the solvents used play a vital role in determining the resultant properties. The data obtained will play an important role in understanding the effect of the addition of organic solvents in ILs to enhance their applicability.

Disordered packing density of binary and polydisperse mixtures of curved spherocylinders

Particle elongation is an important factor affecting the packing properties of rod-like particles. However, rod-like particles can be easily bent into non-convex shapes, in which the effect of bending should also be of concerned, To explore the shape effects of elongation and bending, together with the size and volume fraction effects on the disordered packing density of mixtures of non-convex particles, binary and polydisperse mixtures of curved spherocylinders are simulated employing sphere assembly models and the relaxation algorithm in the present work. For binary packings with the same volume, curves of the packing density versus volume fraction have good linearity, while densities are plotted as a series of equidistant curves under the condition of the same shape. The independence of size and shape effects on the packing density is verified for mixtures of curved spherocylinders. The explicit formula used to predict the density of binary mixtures, by superposing the two independent functions of the size and shape parameters, is extended to include a non-convex shape factor. A polydisperse packing with the shape factor following a uniform distribution under the condition of the same volume is equivalent to a binary mixture with certain components. The packing density is thus predicted as the mean of maximum and minimum densities employing a weighing method.