Information retrieval models are widely used in search engines. In the task of information retrieval, these models focuses on the different semaphores, which leads to great differences in model performance. At present, most models are based on part or all of the following information: exact signals, similar signals, signals differentiation, query word weight, proximity, text structure, and different distribution assumptions. This paper introduces the specific meaning of each modeling factor, and exemplifies the positive effect of this factor on modeling through relevant experiments. Based on the above experiments and analysis, this paper finally discusses and analyzes the future development and the trend of information retrieval model.

In recent years, neural machine translation model has become the mainstream model in the field of machine translation. How to learn translation knowledge quickly and accurately from a large amount of training data is a problem worthy of discussion. Different training samples have different degrees of difficulty. Some training samples are simpler and easy for model to learn, while others are more difficult and not easy for model to learn. The difficulty of the samples has a great influence on the convergence of the model, but the traditional neural machine translation model does not consider this difference in the training process. Therefore, this paper explores the influence of the difficulty of the samples on the training process of the neural machine translation model. Considering the sample difficulty for the neural machine translation mode, a dynamic learning method is proposed based on the idea of “curriculum learning”. The difficulty degree of the training samples is quantified from the aspects of the translation effect of the neural machine translation model and the sentence length of the training samples, respectively, then, two learning strategies are designed from-easy-to-difficult and from-difficult-to-easy to train the model. Finally, the translation effects of the model are compared. The experimental results show that both from-easy-to-difficult and from-difficult-to-easy dynamic learning methods can improve the translation effect of the neural machine translation model.

In next-generation networks, Spatial Information Networks (SINs) are expected to play an increasingly important role in providing Internet services over long distances and full coverage in an efficient manner. Most future networks will be hybrid-connecting nodes in space, near space, and on land through satellite links. Security is an essential concern in such networks since it is susceptible to a host of attacks, including eavesdropping, session hijacking, data corruption, and split attack. In this article, various security attacks that are possible in spatial information networks are addressed and different solutions proposed to existing invulnerability analysis in these networks are outlined. Setting in the split attack, a scheme is pointed out to evaluate network invulnerability based on the importance of the node in the complex network and graph convolutional networks classification nodes. Through experiments on the simple network and real spatial information networks, it is verified that the suggested approach has good discrimination and accuracy.

The convolution operators are the core building blocks of convolutional neural network, which enable the network to fuse the information of various layers of space and channels according to a certain perception field of view, and extract the characteristics of the information. However, adjacent pixels often have similar values in an image, which results in a large amount of redundant information in the output of the convolutional layer. In order to reduce redundant information and speed up model inference, many pooling layers are added to the convolutional neural network for reducing information dimensionality. Pooling has better dimensionality reduction effect on image features with the invariance of translation and rotation. And end-to-end model can be maintained compared with traditional dimensionality reduction methods. Therefore, a dimensionality reduction method is proposed based on the attention mechanism by using the pooling layer. In the process of feature extraction, the dimensionality reduction information from each layer’s are reused nonlinearly, so that the potential connections of information in different layers after dimensionality reduction can be learned. In order to obtain the characteristics of the input information, the proposed method focuses on the main texture of the target in the image, and then the low texture and background information of the target are combined. Based on the DLA-34 (deep layer aggregation) neural network, the dimensionality reduction method proposed in this paper and the others dimensionality reduction methods based on the maximum value and the average value are compared to deal with multiple sets on the CIFAR10 and CIFAR100 datasets, which proves the effectiveness of the new method.

As a new signal transformation algorithm, Empirical Mode Decomposition (EMD) can solve the limitation of some existing methods such as Fourier transform that are limited to specific basis functions. Aiming at the problem of insufficient prediction accuracy of artificial neural networks for high-frequency financial time series, this paper combines EMD and Weibull distribution to preprocess financial time series. A classification model based on EMD and multi-branch long short-term memory network is proposed in this paper. The multi-branch LSTM network based on EMD is used to extract information about price movements from high-frequency financial time series and make predictions about future price movements. By predicting the FX time series of EURUSD from 2009 to 2012, the experimental results show that the proposed model can obtain higher prediction accuracy and calculation speed. Compared with ordinary LSTM network, the generalization ability and model stability are improved.

Aiming at the shortcomings of the existing subjective question marking, this paper proposes an automatic scoring method for subjective questions using semantic technology and LSTM. Firstly, the LSTM neural network classification model is constructed to realize the simultaneous double classification of the title type and question type of the subjective question corpus. It provides the early guarantee for the realization of different question types using different automatic scoring methods; Then, syntax analysis and dependencies analysis are used to analyze questions and answers (standard answers and student’s answers), the components of interrogative words in questions are obtained through the dependencies analysis of title, and the core semantic, the non-core semantic and the negative tone of the answers are obtained through the dependencies analysis of the answers. Finally, for different types of questions, different automatic marking methods of subjective questions are adopted to realize the self-adaptive marking of different students with the same standard answers, which further improves the flexibility and accuracy of the subjective question scoring system.

The maximum Lyapunov exponent of fractional order FOSRM system is calculated and 0-1 test for time series of status values is made by using numerical methods. The minimum order of chaotic motion of fractional order FOSRM system is about 2.01.The fractional order synchronous reluctance motor (FOSRM) has chaotic oscillation under the condition of specific order and parameters, which seriously affects the dynamic performance and stability of the system. In this paper, a controller designed based on the infinite state method and sliding mode control theory is used. The stabilization control of the system is realized under the conditions of no disturbance or disturbance. The global asymptotic stability of the system can be achieved rapidly, and the effectiveness of the method is verified by numerical simulation. The evaluation index of state stabilization control is that whether the system can quickly reach the global stability of zero value. Numerical simulation verifies the effectiveness of the method in this paper. The study on the stabilization control of fractional synchronous reluctance motors provides a reference for the research of excellent control methods in engineering practice.

Photovoltaic microgrid has two working modes: off-grid modes and grid-connected modes. In off-grid mode, the DC bus voltage fluctuates due to the power changes of load and renewable energy. In grid-connected mode, the grid-connected current fluctuates greatly due to the changes of input power and the low harmonics generated by the nonlinear load, thus affecting the power quality. This paper uses two kinds of energy storage elements of super capacitors and battery packs to form a microgrid hybrid energy storage scheme. When the microgrid runs off-grid, it can make use of the quick response of super capacitor to compensate the transient power, so as to stabilize the DC bus voltage faster. During grid-connected operation, in the traditional grid-connected voltage and current double closed-loop control strategy, the super capacitor current control loop is added to reduce the current ripple during grid-connected operation. The proposed hybrid energy storage control scheme is simulated and verified. The results show that the proposed hybrid energy storage control strategy can control the fluctuation of DC microgrid bus voltage within 5%, and maintain the total harmonic distortion of grid-connected current at 7%, so as to achieve the purpose of control.

The interference of harmonic aliasing in N-path band-stop filter is a common problem, resulting in higher requirements for its interferer-tolerant performance. Firstly, the harmonic transfer functions relating the input to the voltage sampled on the capacitor are derived using the adjoint network, which simplifies the calculation of the harmonic transfer functions. Then the complete harmonic transfer functions of N-path band-stop filter are derived which are verified by the simulation of the circuit. The influence of filter parameters on harmonic aliasing is analyzed in UMC110 CMOS technology. It is concluded that the aliasing interference can be effectively suppressed by appropriately increasing the number of paths or the capacitance. The simulation result shows, that the max harmonic folding of the filter can be controlled below -27 dB, which has good anti-aliasing effect when the channel number is 16 and the capacitance is 50 pF. A cascaded design scheme of N-path band-stop filter and pre-frequency-selected device is proposed to eliminate harmonic folding. The cut-off frequency should be less than N-1 times the central frequency of N-path filter.

There are many different dynamic behaviors when droplets impact different wettability surfaces. Numerical simulation method is a very simple and effective way to study this phenomenon. Based on the chemical potential lattice Boltzmann method, the wettability of solid surfaces is changed by adjusting the contact angle, and the dynamic behavior that droplets impact the hydrophobic surfaces of single contact angle and different contact angle distributions are simulated and analyzed. When a single contact angle is applied to the hydrophobic surface, the bounce height of the droplet increases as the contact angle of the surface increases, and the second bounce will occur when the contact angle increases to 160°, while the multiple bounce will occur when the contact angle increases to 170°. When differently distributed contact angles are applied to the surface, the droplets bounce phenomenon will be varied correspondingly. The results show that when a droplet impacts the junction of two kinds of contact angles, the droplet always bounces laterally to the side with a small contact angle, and its bounce height and distance depend on the difference between the two contact angles. Besides, the bounce height and distance are changing with the differences of the contact angle respectively. This characteristic is very helpful to effectively control the dynamic behavior of droplets and it can be widely used in self-cleaning, inkjet printing and other fields.

The pricing of the European compound power options is considered under double stochastic volatility jump diffusion model. By using multivariate characteristic function, partial differential-integral equation and the Fourier inversion transform, the analytic formulas for the European compound power options are obtained. Comparison on prices of this option under some differential models is discussed and impact of the main parameters for the proposed model on option price is investigated with some numerical examples. Numerical results show that both the stock’s volatility and the jump intensity can produce considerable effect on option price, and the compound power option has not only better risk management flexibility but also greater returns to investors.

Let G=(V,E) be a graph. A real-valued function f∶V→[0,1] is said to be a fractional dominating function (FDF) if f(N[u])≥1 holds for every vertex u∈V(G). The fractional domination number γ_f(G) of G is defined as γ_f(G)=min{f(V)|f is a FDF of graph G}. In this paper, the exact values of γ_f(K_m×P_n),γ⁰_f(K_m×P_n) for all integers m≥3, n≥2, andγ_f(K_m∨P_n) for all integers m≥5, n≥3 are given.

In this paper, the φ-flat cotorsion theory which is showed to be a perfect cotorsion theory is introduced and studied. Assume R is a φ-ring. It is proved that the φ-flat cotorsion theory coincides with the classical flat cotorsion theory if and only if R is a domain. As an application, new characterizations of Nonnil-coherent rings and φ-von Neumann rings are given. Finally, the envelope property of φ-flat modules is investigated and showing that every R-module has a surjective (pre)envelope if and only if R is a Nonnil-coherent ring and all φ-flat R-modules are closed under submodules.

In this paper, using the energy estimates and the bootstrap arguments, the global existence of solutions for a three species predator-prey model with cross-diffusion is proved when the space dimension is less than 10. Under certain conditions for the coefficients of the reaction functions, the convergence of solutions for the system is discussed by constructing Lyapunov function.

Oleanolic acid, ursolic acid and glycyrrhetinic acid were esterified with corresponding α,ω-dibromoalkane in the presence of potassium carbonate to prepare intermediates. Then, the intermediates were treated with p-hydroxybenzonitrile under potassium carbonate by means of a nucleophilic substitution reaction to generate six oleanolic acid/ursolic acid/glycyrrhetinic acid-p-hydroxybenzonitrile derivatives. Anti-tumor activity was tested by the MTT assay. The results showed that glycyrrhetinic acid-p-hydroxybenzonitrile conjugates [n-(4-cyanophenoxy)-hexyl]-3β-hydroxy-11-oxo-olean-12-en-29-oate and [n-(4-cyanophenoxy)-octyl]-3β-hydroxy-11-oxo-olean-12-en-29-oate had certain anti-proliferative effect on chronic myeloid leukemia cell line K562. Moreover, [n-(4-cyanophenoxy)-hexyl]-3β-hydroxy-11-oxo-olean-12-en-29-oate also had moderate anti-proliferative effect on human liver tumor cell line Bel-7402. Other compounds had no antiproliferative activity on the tested cell lines. [n-(4-Cyanophenoxy)-hexyl]-3β-hydroxy-11-oxo-olean-12-en-29-oate, which has less damage to normal liver cells, expressed almost the same level as the positive control drug 5-FU on the anti-proliferative activity of the tested cells. This indicated that [n-(4-cyanophenoxy)-hexyl]-3β-hydroxy-11-oxo-olean-12-en-29-oateand had the potential value for further research in anti-tumor field.

Ten compounds were isolated from the metabolites of endophytic fungus GDG-178(Fusarium equiseti)by using silica gel column chromatography, ODS chromatography, SephadexLH-20 and semi-prepared HPLC. The structures of these compounds, beauvericin (1), fusaproliferin (2), 11-epiterpestacin (3), 3-indoleacetic acid (4), indole-3-methylethanoate (5), 4-methyl-5,6-dihydro-2H-pyran-2-one (6), sterigmatocystin (7), 2′-hydroxydecanylpentadec-5,8,10,12-tetraneoate (8), 19-norergosta-5,7,9,22-tetraene-3β-ol (9) and p-hydroxy benzaldehyde (10), were identified by spectral analysis. Compounds 7-10 were obtained from Fusarium sp. for the first time. And compounds 5 and 6 were firstly obtained from Fusarium equiseti.

The effects of urea (NS), ammonium chloride (LA) and alanine (BA) on nitrogen form, physical and chemical characteristics and microbial community of aerobic granular sludge were investigated. The results showed that LA group had less removal of dissolved ammonia nitrogen, while the total nitrogen removal rate was the highest in NS group. The ratio of adsorptive organic nitrogen to adsorptive nitrogen in NS, LA and BA groups were 73.28%, 44.6% and 77.3% at the end of the experiment (40 d), indicating that the content of organic nitrogen in the sludge adsorbed state with ammonium chloride as the nitrogen source was less; The fluorescence intensity of tryptophan-like and coenzyme F₄₂₀ peak in theextracellular polymeric substances (EPS) of the sludge was BA＞NS＞LA. At the end of the experiment, the nitrite oxidase of BA group increased by 91.8%, while that of NS and LA group increased by 2.8% and 40.8%, indicating that the use of ammonium chloride as the nitrogen source increased the activity of nitrite oxidase in the sludge. Microbial community showed that the predominant bacteria in the three groups were Proteobacteria, Candidatus Saccharibacteria and Bacteroidetes at phylum level. The abundance of Candidatus Saccharibacteria using ammonium chloride as a nitrogen source was the highest. The addition of alanine could significantly increase the abundance of denitrifying bacteria Proteobacteria and Bacteroidetes, which was the lowest in the LA group, indicating that taking alanine as the nitrogen source could make the denitrifying bacteria accumulate well. Kyoto Encyclopedia of Genes and Genomes (KEGG) secondary pathway diagram showed that amino acid metabolism, carbohydrate metabolism and membrane transport were the main pathways, among which the amino acid metabolism of group BA was the highest (13.11%). In general, the sludge with alanine as the nitrogen source had better denitrification performance and higher denitrification functional bacteria abundance.

Taking Xinghai Lake, a typical urban lake in arid area of Northwest China, as the research area, and based on remote sensing images in 2005, 2011 and 2016 as data sources, the wetland ecological risk index was constructed by using landscape pattern index and landscape vulnerability index. Based on ArcGIS and other software platforms, spatial interpolation and spatial autocorrelation methods were used to analyze the spatial and temporal characteristics of ecological risk in Xinghai Lake. The results showed that: (1) The landscape pattern of Xinghai Lake wetland changed significantly from 2005 to 2016. The area of all landscape types fluctuated in varying degrees. The trend of fragmentation of construction land, grassland and wetland vegetation, which were greatly influenced by human activities, was increasing while the stability of regional landscape was decreasing. (2) The distribution of ecological risk in the study area was closely related to landscape types. The main ecological risk of Xinghai Lake wetland in the three periods included moderate risk. The proportion of low-risk area decreased from 17.6% to less than 1%. The higher risk showed an obvious upward trend, and experienced a transition from below medium risk to medium risk and high risk. (3) The ecological risk of the study area had spatial aggregation effect, and the distribution of the aggregation area was related to the distribution of landscape types. Urban lake wetlands are significantly affected by human activities, so it is urgent to strengthen the management and protection of ecological risk of wetlands.

Tourism flow, as the foundation of tourism, is the essential areas of tourism geography research. Chinese scholars have made abundant achievements in the field of tourism flow research over the past 30 years. However, most of them focus on the tourist flow. However, the research on the tourism information flow which is the key factor that affects the tourist decision and then affects the tourism flow is rarely involved. Based on a universal scheme for quantitative research on the flow space structure features of tourism information flow, and taking the Pearl River-West River economic belt as the typical research area. 12 739 valid data of tourist information flow of 11 core cities in the Pearl River-West River economic belt (Guangzhou, Foshan, Yunfu, Zhaoqing, Nanning, Liuzhou, Laibin, Wuzhou, Baise, Chongzuo, Guigang) were obtained through 5 major online tourism platforms (Ctrip.com, Qunar.com, fliggy.com, ly.com and tuniu.com) in 2019. Firstly, the correlation matrix of tourism information flow is constructed. Then, this paper describes the flow space structure features of tourism information flow of 11 core cities in this region from 3 aspects: characteristics of direction and its rate, scale and principal-subordinate relationship of tourism information flow. Therefore, it provides a scientific basis for the construction and development of tourism integration of this area and a new perspective and method for the study of tourism flow. The results are obtained as follows: ①The provincial capital cities of Guangzhou and Nanning are not only the main gathering cities of tourism information flow in this region, but also the main export cities, and the aggregation and output of tourism information flow are 1.30, 2.55 and -0.13, 0.76 respectively. The non-provincial capital cities with special tourism resources of Chongzuo and Baise are the main gathering cities in this region, and the aggregation of tourism information flow are 1.74 and 1.55 respectively. The forward and backward flow of tourism information among the provincial capital cities and non-provincial capital cities with distinctive tourism resources is basically the same. Namely, the forward and backward flow of tourism information of Guangzhou$\leftrightarrows$Nanning and Chongzuo$\leftrightarrows$Baise are 0.55, 0.47, -0.25, -0.25. The difference of the forward and backward flow of tourism information between provincial capital cities and non-provincial capital cities with distinctive tourism resources is vast, such as Guangzhou$\leftrightarrows$Chongzuo (6.11, 0.11) and Guangzhou$\leftrightarrows$Baise (5.92, 0.26). ②Among 11 cities of this region, the scale of tourism information flow in Guangzhou (2.39) is far larger than that of other cities. Followed by Chongzuo (0.72) and Baise (0.62), with distinctive tourism resources, and their scales of tourism information flow are basically the same. Nanning (0.40), Foshan (0.22) take the fourth and fifth place. And other cities’ scales of tourism information flow are lower than the average level of this region. ③There is significant polarization of flow space for the tourism information flow in this region, and Guangzhou is the dominant city and the largest flow space node. Nanning, Chongzuo and Baise transformed from the sub-dominant divergent, gathering city and subordinate city of the first tourism information dominant flow to the sub-dominant gathering city, dominant city and sub-dominant gathering city of the second and third dominant flow respectively. It shows that the flow barycenter of tourism information dominant flow gradually shifts to Guangxi. At the same time, Baise and Chongzuo have obvious sub-regional characteristics. Other cities are subordinate cities of tourism information flow.