Search

The density regression fitting method is generally used to predict total organic carbon(TOC) content of shale reservoir with seismic data. Because it only considers the linear relationship of single factor, the prediction error is large. Aiming at the shortcomings of this method, a TOC content prediction method based on deep learning was proposed. Firstly, the elastic parameter curves with the highest correlation with the TOC content curve were selected from the logging data as the input data of the sample set, and the TOC content curve was used as the output data of the sample set. Based on the sample set, a deep feedforward neural network model for TOC content prediction was constructed. Then, the structure of the network model was adjusted and the parameters of the network were optimized by conjugate gradient method. Finally, the elastic parameter data volume obtained from prestack amplitude versus offset (AVO) inversion was input into the deep feedforward neural network model to predict the final TOC content volume. Through the application of actual logging and seismic data of shale reservoir in Sichuan Basin, the advantages of the new method over the conventional regression fitting method were compared, the practicability and feasibility of the proposed method were verified, which provides a new idea for TOC content prediction of shale reservoir. [ABSTRACT FROM AUTHOR]

TeraHertz wireless communication has been regarded as an effective technology to satisfy the ever-increasing requirements of high-rate services, where the imperfections of TeraHertz devices, including in-phase/quadrature imbalance, phase noise and nonlinearity, regarded as hybrid distortions, have to be investigated. Due to the hybrid distortions, the explicit system model could not be mathematically derived, which limits the performance of the state-of-the-art demodulation methods. Here, a deep learning-assisted demodulation methodology is proposed to improve the bit error rate performance. Specifically, a multiple-output deep feedforward neural network is designed to fit the mapping between the received signal and likelihood information, where the number of outputs equals to the size of the modulation set, thus enabling the demodulation of the overlapped received signals corresponding to different constellation points. Besides, a training set generation method is proposed to generate the training examples without the prior knowledge of likelihood information. Simulation results validate that the proposed learning-assisted methodology could improve the demodulation performance of the TeraHertz wireless systems under severe hybrid distortions. [ABSTRACT FROM AUTHOR]

TeraHertz (THz) wireless communication constitutes a promising technique of satisfying the ever-increasing appetite for high-rate services. However, the ultra-wide bandwidth of THz communications requires high-speed, high-resolution analog-to-digital converters, which are hard to implement due to their high complexity and power consumption. In this paper, a deep learning-assisted THz receiver is designed, which relies on single-bit quantization. Specifically, the imperfections of THz devices, including their in-phase/quadrature-phase imbalance, phase noise and nonlinearity are investigated. The deflection ratio of the maximum-likelihood detector used by our single-bit-quantization THz receiver is derived, which reveals the effect of phase offset on the demodulation performance, guiding the architecture design of our proposed receiver. To combat the performance loss caused by the above-mentioned distortions, a twin-phase training strategy and a neural network based demodulator are proposed, where the phase offset of the received signal is compensated before sampling. Our simulation results demonstrate that the proposed deep learning-assisted receiver is capable of achieving a satisfactory bit error rate performance, despite the grave distortions encountered. [ABSTRACT FROM AUTHOR]

Generalization ability is known as an important performance index of artificial neural networks (ANNs). The generalization ability of an ANN usually refers to its ability to recognize untrained samples, but it lacks quantitative analysis. A method is designed by using frequency-domain signals to observe the generalization ability of deep feedforward neural networks (DFFNNs) which are popular ANN models. This method allows us to observe that the generalization ability of DFFNNs is limited to a small neighborhood around the trained samples. Then, the relationship between sample similarity and the DFFNN's generalization performance is further analyzed. The analysis results show that the correlation coefficient between samples has a certain positive correlation with the DFFNN's generalization performance. Based on this new understanding, an algorithm in which shadows of the trained samples are added into the training set is proposed to improve the generalization ability of DFFNNs. The proposed algorithm is tested with some simulated signals and some real-world data. The tests show that the proposed method can indeed improve the DFFNN's generalization ability by only changing the training sample set. [ABSTRACT FROM AUTHOR]

Data-driven sensors are techniques capable of providing real-time information of unmeasured variables based on instrument measurements. They are valuable tools in several engineering fields, from car automation to chemical processes. However, they are subject to several sources of uncertainty, and in this way, they need to be able to deal with uncertainties. A way to deal with this problem is by using soft sensors and evaluating their uncertainties. On the other hand, the advent of deep learning (DL) has been providing a powerful tool for the field of data-driven modeling. The DL presents a potential to improve the soft sensor reliability. However, the uncertainty identification of the soft sensors model is a known issue in the literature. In this scenario, this work presents a strategy to identify the uncertainty of DL models prediction based on a novel Monte Carlo uncertainties training strategy. The proposed methodology is applied to identify a Soft Sensor to provide a real-time prediction of the productivity of a chemical process. The results demonstrate that the proposed methodology can yield a soft sensor based on DL that provides reliable predictions, with precision being proven by its corresponding coverage region. [ABSTRACT FROM AUTHOR]