The gated structure of the long short-term memory (LSTM) alleviates the defects of gradient disappearance and explosion in the recurrent neural network (RNN). It has received widespread attention in sequence learning such as text analysis. Although LSTM has good performance in handling remote dependencies, information loss often occurs in long-distance transmission. We propose a new model called ELSTM based on the computational complexity and gradient dispersion in the traditional LSTM model. This model simplifies the input gate of LSTM, reduces some time complexity by reducing some components, and improves the output gate. By introducing the exponential linear unit activation layer, the problem of gradient dispersion is alleviated. Comparing the new model with multiple existing models, when predicting language sequences, the time used by the model has been greatly reduced, and the language confusion has been reduced, showing good performance.

The concentration of dissolved gas in transformers is closely related to their operating status. Aiming at dissolved gas analysis (DGA) in transformer oil, this paper proposes a fault diagnosis method for transformer DGA based on long-term and short-term memory (LSTM) artificial neural networks. The method uses 240 sets of samples collected by China Southern Power Grid Corporation, with 180 sets as training data and the remaining 60 sets as test data. The input consists of five kinds of dissolved gases in oil, and the output is the corresponding fault type. The hyperparameters (H1=H2=50) of the network are determined through experimentation to establish a transformer DGA fault diagnosis model based on LSTM. The research results indicate that the LSTM diagnosis model has higher consistency with actual fault types compared to the traditional neural network diagnosis model. These findings demonstrate the promising application prospects of LSTM in the field of transformer DGA fault diagnosis.

The ‘intention’ classification of a user question is an important element of a task-engine driven chatbot. The essence of a user question’s intention understanding is the text classification. The transfer learning, such as BERT (Bidirectional Encoder Representations from Transformers) and ERNIE (Enhanced Representation through Knowledge Integration), has put the text classification task into a new level, but the BERT and ERNIE model are difficult to support high QPS (queries per second) intelligent dialogue systems due to computational performance issues. In reality, the simple classification model usually shows a high computational performance, but they are limited by low accuracy. In this paper, we use knowledge of the ERNIE model to distill the FastText model; the ERNIE model works as a teacher model to predict the massive online unlabeled data for data enhancement, and then guides the training of the student model of FastText with better computational efficiency. The FastText model is distilled by the ERNIE model in chatbot intention classification. This not only guarantees the superiority of its original computational performance, but also the intention classification accuracy has been significantly improved.

The long short-term memory (LSTM) model trained on the universal language modeling task overcomes the bottleneck of vanishing gradients in the traditional recurrent neural network (RNN) and shows excellent performance in processing multiple tasks generated by natural language processing. Although LSTM effectively alleviates the vanishing gradient problem in the RNN, the information will be greatly lost in the long distance transmission, and there are still some limitations in its practical use. In this paper, we propose a new model called NEWLSTM, which improves the LSTM model, and alleviates the defects of too many parameters in LSTM and the vanishing gradient. The NEWLSTM model directly correlates the cell state information with current information. The traditional LSTM's input gate and forget gate are integrated, some components are deleted, the problems of too many LSTM parameters and complicated calculations are solved, and the iteration time is effectively reduced. In this paper, a neural network model is used to identify the relationship between input information sequences to predict the language sequence. The experimental results show that the improved new model is simpler than traditional LSTM models and LSTM variants on multiple test sets. NEWLSTM has better overall stability and can better solve the sparse words problem.

The LDA (Latent Dirichlet Allocation) model is an unsupervised learning model that can extract potential topics in the corpus. It is widely used in natural language processing applications. The combination of SVM (Support Vector Machines) and LDA has better text classification effect. However, because the mature kernel function of SVM and its parameter selection can only be selected according to experience, with certain randomness, it has a great influence on classification accuracy. This paper proposes an optimized LDA-SVM text classification method using Particle Swarm optimization (PSO). The algorithm (PSO) optimizes the SVM error penalty parameter C to improve the LDA-SVM text classification algorithm. The experimental results show that the LDA-PSO-SVM text classification algorithm proposed in this paper has higher accuracy and better classification performance than other algorithms.

A new digital controller for a single-phase boost power factor correction (PFC) converter operating at a discontinuous conduction mode (DCM), is presented to achieve high input power factor over wide input voltage and load range. A method of duty cycle modulation is proposed to reduce the line harmonic distortion and improve the power factor. The loop regulation scheme is adopted to further improve the system stability and the power factor simultaneously. Meanwhile, a novel digital pulse width modulator (DPWM) based on the delay lock loop technique, is realized to improve the regulation linearity of duty cycle and reduce the regulation deviation. The single-phase DCM boost PFC converter with the proposed digital controller based on the field programmable gate array (FPGA) has been implemented. Experimental results indicate that the proposed digital controller can achieve high power factor more than 0.99 over wide input voltage and load range, the output voltage deviation is less than 3 V. and the peak conversion efficiency is 96.2% in the case of a full load.

The purpose of this paper is to present a novel trajectory prediction method for proximate time-optimal digital control DC–DC converters. The control method provides pre-estimations of the duty ratio in the next several switchingcycles,soas to compensate the computational time delay of the control loop and increase the control loop bandwidth, thereby improving the response speed. The experiment results show that the fastest transient response time of the digital DC–DC with the proposed prediction is about 8 s when the load current changes from 0.6 to 0.1 A.

A novel trajectory prediction control algorithm for digital control DC–DC converters has been presented in this paper. The proposed trajectory prediction control algorithm can provide an accurate prediction of the duty ratio of the next several switching cycles, so as to overcome the inherent time delay of the digital control loop, and to improve the transient response of digital control DC–DC converters, including load response, line response and reference tracking response. A digital control buck DC–DC converter was implemented to verify the effectiveness of the proposed prediction control algorithm. The recovery time is about 8 μs and 4 μs respectively, when the load current changes from a full load to a 17% load and the input voltage changing between 5 V to 6 V. The fastest reference tracking speed is about 26.7 μs/V.

A novel digital controller for boost power factor correction converter is proposed to achieve high power factor and fast dynamic response. A method of direct duty cycle calculation is adopted to improve the power factor. The prediction module is adopted to estimate the track of the output voltage and the inductor current of next switching cycle in advance to improve the dynamic response. A multiplex successive approximation analog-to-digital converter (ADC) is designed to aim high resolution via little hardware resource. Meanwhile, a novel digital pulse width modulator (DPWM) is realized to improve the regulation linearity. The boost PFC converter with proposed digital controller based on the field programmable gate array (FPGA) has been implemented. Experimental results indicate that the proposed digital controller can achieve high power factor more than 0.98, the dynamic response under load variation is about 80ms and the output voltage overshoot is about 6%.

In order to improve some performance of the system under light load conditions, a PWM/LDO dual-mode system controlled by an external synchronous signal is presented here. In the work, we’ve designed a switching circuit based on the detailed analysis of the systematic structure and functions. The whole circuit was verified with Cadence simulations under the CSMC 0.5 μm CMOS process. The results show that when the input voltage V in is 3.5–6 V, the output voltage V O would be kept at 3.3 V and the operating frequency is about 1.2 MHz. In addition, the ripple of output voltage is less than 30 mV in PWM mode, while the ripple voltage is less than 10 μV and the maximum load current is up to 50 mA in LDO mode, and the converter’s overall efficiency is more than 80 %. Finally the whole circuit’s layout and post-simulation results are given. Consequently, this method can improve the overall efficiency of the system evidently.

Aiming for the dynamic voltage scaling technology in SoC(system on chip), a digital controlled DC-DC Buck converter with variable output voltage is designed. First, a trajectory prediction method based on system model is used. Based on the sampling values of the inductor current and the output voltage at the current switching cycle, the inductor current and the output voltage at the next switching cycle are predicted according to the characteristics of energy storage devices such as inductor and capacitor. Then, these predictive values are used to substitute the actual sample values, which can compensate the loop time delays and improve the transient responses. Besides, the multiple-sample technology is adopted to minimize the effect of random noise. The circuit is designed on the SMIC 65 nm process. The experimental results show that when the maximum deviation between the prediction values and the actual ones is less than 15 mV, the steady-state error of the output voltage can be controlled within 5 mV. When the load current and the reference voltage change, the transient response time of the output voltage are 122.3 and 476.1 μs, respectively. The line regulation is 14.37%.

A novel integrated single-inductor dual-output (SIDO) buck converter with peak current common-mode and the two output voltage ripples comparison differential-mode control method in continuous conduction mode is presented. The system can use only one inductor to provide two independent output voltages, 1.2 V and 1.8V, with a maximum total output current 460 mA. The proposed converter has been fabricated in a 0.18μm 1P6M CMOS process. Experimental results show the load transient response time is only 8μs and the cross-regulation is about 0.05 mV/mA when the load current suddenly changes 200 mA. The maximum power conversion efficiency 93.5% is achieved at total output power 240 mW.

A novel digital control algorithm for digital control power factor correction is presented, which is called the prediction algorithm and has a feature of a higher PF (power factor) with lower total harmonic distortion, and a faster dynamic response with the change of the input voltage or load current. For a certain system, based on the current system state parameters, the prediction algorithm can estimate the track of the output voltage and the inductor current at the next switching cycle and get a set of optimized control sequences to perfectly track the trajectory of input voltage. The proposed prediction algorithm is verified at different conditions, and computer simulation and experimental results under multi-situations confirm the effectiveness of the prediction algorithm. Under the circumstances that the input voltage is in the range of 90-265 V and the load current in the range of 20%-100%, the PF value is larger than 0.998. The startup and the recovery times respectively are about 0.1 s and 0.02 s without overshoot. The experimental results also verify the validity of the proposed method.

Based on the research of existing index systems and the analysis of the features of Internet public opinion with E-R model,a new network monitoring and early warning index system for public opinion is built.With four dimensions of popularity,intensity,tendency and inclination,the new index system outlines the full picture of the Internet public opinion,such as dissemination scope,content value,opinion distribution and development status.

A PWM/LDO dual-mode system controlled by external synchronous signal was designed,which improved the system performance at light load.Architecture of the entire system and its functions were described,and a detailed analysis was made on the switchover circuit.The circuit was simulated based on CSMC 0.5 μm CMOS process.Results showed that the system had an input voltage ranging from 3.5 V to 6 V,an output voltage of 3.3 V,and an operating frequency of 1.2 MHz.The ripple of the output voltage was less than 30 mV in PWM mode,while the ripple voltage was less than 10 μV and the max load current was 50 mA in LDO mode.The circuit had a conversion efficiency above 80% over the load range from 20 mA to 1.2 A,and the PWM/LDO switchover time was less than 60 μs.

This paper introduces an approach for the design of an adaptive non-linear digital Fuzzy logic controller for a digital controlled auto-adjustable Buck DC-DC converter, which is validated by Simulink tools of Matlab. The setup time of the output voltage was about 10μs and the steady state error was less than 2mV with the proposed controller. When the supply voltage or load current was disturbed, the output reached stability again after 15μs. And the output voltage tracked the change of the reference voltage in 15μs. The results show that its dynamic response was quick and its sensitivity was very interesting for various supply voltages and load resistance variations.