Effect Of Different Numbers Research Articles

Study on the transient overvoltage suppression effect of the synchronous condenser of high percentage renewable energy in the sending-end system

Abstract This paper analyzes the impact of configuring a synchronous condenser on the characteristics of the Inner Mongolia power grid following the commissioning of the first grid-to-grid DC project. Simulation results show that, after the configuration of synchronous condenser, the multiple renewable energy stations short circuit ratio (MRSCR) rises, and the DC sending end of the grid is a strong system, which is conducive to the whole scale of renewable energy transmission of Inner Mongolia power grid. DC bipolar blocking faults and transient overvoltage problems at the sending end occur for different operating conditions, such as reduced power and bipolar operation. Then, the effect of different numbers of synchronous condensers on transient overvoltages is studied. The study shows that it is necessary to reasonably arrange the DC project supporting thermal power unit start-up mode and configure large-capacity synchronous condenser or a certain number of distributed regulators, which can help to alleviate the problem of transient overvoltage during faults.

Long-term effects of different number of botulinum toxin injections into the gastrocnemius muscle on function and muscle morphology in children with cerebral palsy

BackgroundBotulinum toxin (BoNT) injections are used to reduce spasticity, and sometimes repeated injections of the toxin are required. The aim of this study is to investigate the effects of the number of injections of BoNT into the gastrocnemius muscle on function, muscle morphology, muscle stiffness and muscle vascularization in children with cerebral palsy (CP). MethodsThe study included 22 children with unilateral or bilateral spastic type CP aged 5-13 years who had previously received one, two, or three BoNT injections into the gastrocnemius muscle. A total of 29 gastrocnemius muscles were evaluated. Gastrocnemius muscle morphology (muscle cross-sectional area, thickness, fascicle length and pennation angle) was examined by ultrasonography, muscle stiffness by shear wave ultrasound elastography and muscle vascularization by Superb Microvascular Imaging. The functional status of the children was evaluated by the Gross Motor Function Measurement, the Timed-Up-and-Go Test, the Squat Test, the Vertical Jumping Test and the Sit-and-Reach Test. ResultsIn the motor functions and muscle morphology values were similar in all three groups (p>0.05). Muscle stiffness values were higher (p=0.035) and vascularization values were lower (p=0.03) in the group that received three injections compared to the groups that received one injection. There was a moderate positive correlation of the number of BoNTs with muscle stiffness (p<0.05, r=0.454) and a moderate negative correlation with muscle vascularization (p<0.05, r=-0.497). ConclusionsRepeated BoNT applications have no effect on motor functions and muscle morphology in children with CP, whereas BoNT injections administered three times increase muscle stiffness and decrease vascularization.

The dynamical behavior effects of different numbers of discrete memristive synaptic coupled neurons

The dynamical behavior effects of different numbers of discrete memristive synaptic coupled neurons

A comprehensive evaluation of lightning location accuracy using a weighted gridding method

For the newly built very low-frequency long-range lightning location network (VLF-LLN) in China based on the equivalent propagation velocity method, we have evaluated the probability of deviation distance between the locations results and the reference points by using a weighted gridding method, and have drawn some conclusions. (1) By analyzing the effects of different numbers and configurations of stations on the location deviations, it is found that when the number of participating stations increase, the location deviation relatively decreases, and the numbers and configurations of participating stations has much more effect on the probability of location deviation of the lightning flashes outside the network than those in the network; (2) We statistically analyzed the location deviation for each grid of the network by using a weighted gridding method. The deviation probability using 11-station synchronization as a reference is similar to that for 13-station synchronization per grid. Assuming that the deviation distance is less than 3–5 km, the average location results are almost unchanged when the station number exceeds 7 or 8; (3) Comparing with the Advanced Direction and Time-of-Arrival Detecting network (ADTD), our VLF-LLN has an average deviation value of 3.47 km and a median value of 1.81 km for 63767 lightning flashes, and 90% samples have the location deviation less than 8 km and 54% have the location deviation less than 2 km.

Electrical conductivity of graphene/copper composites at lattice scale

Graphene/copper composites are potential structure-function integrated materials. Different from previous micro-nano scale composites, graphene/copper lattice-scale composites and their effects on electrical conductivity are investigated in this work. Using the first-principles calculation to explore the effects of different numbers of graphene layers on the properties of graphene/copper composites, it is found that with the increase of graphene layer numbers, the interfacial bonding strength improves, but the electrical properties show a decreasing trend due to the electron scattering and increasing Cu–C layer distance. The single-layer graphene/copper composite has the most excellent electrical properties, with the highest density of states at the Fermi level and the most charge transfer. A series of layered-structural composites at lattice scale are screened, and their electrical properties and dynamic stability are also predicted. These composites rely on metallic bonding to connect copper and carbon atoms, most of which show obvious vertical charge transfer and high conductivity, with short Cu–C layer distance and alternating arrangement of metal layers and hexagonal carbon atoms. CuC2 (derived from MgB2) has the highest average conductivity and dynamic stability and can be considered an ideal lattice-scale composite. The development of new lattice-scale composites is significant for the future research of graphene/copper composites.

Using geopolymer coated and uncoated geotextile as a hybrid method to improve uplift capacity of screw piles in cohesionless soil

In this research, experimental studies and numerical analyses were carried out to investigate how the usage of geotextile and geopolymer coated geotextile as a hybrid method changes the uplift behavior of the screw piles in cohesionless soil. In this context, traditional pile behavior, the effect of different number of helixes and embedment depths on screw piles, the mechanism of geotextile and the effects of geopolymer coating process were investigated. In addition, experimental studies were modeled by using Plaxis 3D and parametric studies were carried out after verification between the results of experimental study and numerical analysis. In the numerical analysis, a segmented helix model consisting of four 90-degree slices was developed instead of the planar helixes commonly used in the literature. For further investigation of the effectiveness of hybrid method, parameters such as improvement ratios and breakout factors were calculated. When the results obtained within the scope of the study were evaluated, the geopolymer coating process increased the bearing capacity of the geotextile by 24 % at 27 % less elongation. It was also seen that uncoated and geopolymer coated geotextile increased the screw pile performance in terms of improvement ratios by 294 % and 364 %, respectively.

DFT study of Cun (n = 1–3) cluster-modified WSe2 monolayers for the detection of SF6 decomposition gases (H2S, SO2, SOF2 and SO2F2)

The role of SF6 in gas-insulated equipment is crucial for insulation and arc extinction. When subjected to electrical faults, thermal faults, and mechanical failures in insulation devices, SF6 can decompose into H2S, SO2, SOF2, and SO2F2. This study conducted first-principles simulations to analyze the effect of Cun (n = 1–3) cluster-modified monolayer WSe2 on sensing H2S, SO2, SOF2, and SO2F2 gases during SF6 decomposition. The optimal doping positions of Cun-WSe2 clusters at n = 1, 2, and 3 were determined, showcasing their superiority over pure WSe2 in terms of band gap, frontier orbitals, and density of states. Subsequently, a detailed investigation was carried out on the adsorption effects of the four gases at the optimal doping positions, including parameters such as adsorption distance, adsorption energy, charge transfer, differential charge density, and density of states. It was found that when the number of copper atoms in the cluster increased to 3, the banding energy sharply rose to -2.395 eV. Simulation results demonstrated the adsorption behavior of H2S on Cun (n = 1–3)-WSe2 demonstrates remarkable affinity, with an increasing adsorption energy as the number of Cu atoms rises. Notably, Cun (n = 1–3)-WSe2 exhibits favorable desorption times for SO2. At 378 K, the desorption time for SO2@Cu3-WSe2 is 14.7 s. At 498 K, desorption times for SO2@Cu1-WSe2 and SO2@Cu2-WSe2 are 7.59 s and 2.85 s, respectively. When the number of copper atoms is 2, the cluster-modified system desorbs SOF2 at room temperature in only 0.89 s. Moreover, Cun (n = 1–3)-WSe2 displays strong adsorption towards SO2F2, characterized by prolonged desorption times, rendering it a promising adsorbent for SO2F2 removal. This study aims to highlight the adsorption effects of different numbers of Cu atomic clusters in the Cun-WSe2 system on SF6 decomposition gas, further advancing research for developing high-performance SF6 decomposition gas sensors.

A new model for multi-port berth allocation problem based on an improved genetic algorithm

Port integration can enhance operational efficiency and service level in response to heightened competition among container ports. Nevertheless, container ports are still considered independent systems separately in formulating scheduling plans. This research explores a new multi-port berth allocation problem (MPBAP) with continuous berth layout considering the mutual influence between ports. The MPBAP is compared with the single port berth allocation, and effects of different numbers of vessels continuously berthing at all ports on a single route on the MPBAP are discussed through vessel operation rate. The problem is formulated by a mixed integer programming model by minimising the total port stay time of all vessels. The quay is highly discretised and a GA coupled with parallel computing is applied to accelerate the search for the optimal model solution. A case study is conducted on two adjacent container ports located along the Yangtze River. Results show that multi-port coordinated scheduling is efficiency and the total port stay time of all vessels reduces by an average of 2.63% under the MPBAP. With more than four vessels berthing continuously at two ports on a single route, the MPBAP shows a higher vessel operation rate and the rate gap between the two models exceeds 2%. The approach suggested in this study can enhance the port system’s efficiency and provide a new idea for coordinated scheduling in other port groups.

Once or twice? The effect of different number of video assistant referees intervention

The Chinese Football Super League (CSL) introduced the video assistant referee (VAR) in the 2018 season, but the effect of VAR intervention has not yet been analyzed. The purpose of this study is to explore how varying frequencies of VAR interventions affect performance indicators in the CSL. VAR interventions were categorised into three groups, none (VAR0 = 198), one (VAR1 = 184) and two or more (VAR2 = 194). The performance variables included scoring-related variables (goals, shots, shots in the penalty area and penalties), attacking and passing-related variables (passes, crosses and free kicks) and defensive-related variables (tackles, fouls, yellow cards and red cards). The Kruskal-Wallis tests were used to examine the differences in performance variables among the different groups. The findings of these tests revealed that VAR was used in 66% of matches in the 2019–2021 season of the CSL. There were significant differences in penalties ( p &lt; 0.01, η2 = 0.11), shots ( p &lt; 0.05, η2 = 0.03), shots in the penalty area ( p &lt; 0.01, η2 = 0.01), fouls ( p &lt; 0.01,η2 = 0.02), passes ( p &lt; 0.01,η2 = 0.05), free kicks (p &lt; 0.05, η2 = 0.01) yellow cards ( p &lt; 0.01, η2 = 0.02) and red cards ( p &lt; 0.01, η2 = 0.05), but there were no significant differences in the number of goals ( p &gt; 0.05), crosses ( p &gt; 0.05) and tackles ( p &gt; 0.05). Therefore, VAR intervention was frequently applied in the CSL during the 2019–2021 season, and the VAR interventions had various impacts on performance indicators.

Bioconvective reversible (an esterification process) Casson nanofluid flow over a radiative spinning disc with microorganism

This study describes the Casson (a non-Newtonian fluid) nanofluid bioconvection flow across a spinning disc in the presence of gyrotactic microorganisms, many slips and thermal radiation. Also, the flow is considered as a reversible flow. The esterification process is taken into account. Using the proper variables, a system of extremely nonlinear PDEs is converted into a system of ODEs. To arrive to the solution of such equations, a numerical approach is used. Using the bvp4c approach, nonlinear flow equations can be numerically solved. Investigated are the effects of different numbers on the thermal field, volumetric concentration of nanoparticles and microbiological field. The key characteristics of the parameters in relation to the profiles of the velocity, temperature, concentration and microorganisms are graphically assessed with appropriate physical effects. A graphical explanation is provided for the wall shear stress, local Nusselt number, local Sherwood number and local motile density number. Rate of motile density number shows a prominent difference between reversible and irreversible flows for Brownian motion and Peclet number. The results of the theoretical simulations have dynamic applications in the fields of biotechnology and thermal engineering.

Effect of the Second-Shell Coordination Environment on the Performance of P-Block Metal Single-Atom Catalysts for the Electrosynthesis of Hydrogen Peroxide

Hydrogen peroxide (H2O2) is an important chemical with a diverse range of industrial applications in chemical synthesis and medical disinfection. The traditional anthraquinone oxidation process, with high energy consumption and complexity, is being replaced by cost-effective and environmentally friendly alternatives. In order to explore suitable catalysts for the electrocatalytic synthesis of H2O2, the stability of B,N-doped graphene loaded with various p-block metal (PM) single atoms (i.e., PM-NxBy: x and y represent the number of atoms of N and B, respectively) and the effects of different numbers and positions of B dopants in the second coordination shell on the catalytic performance were studied by density functional theory (DFT) calculations. The results show that Ga-N4B6 and Sb-N4B6 exhibit enhanced stability and 2e− oxygen reduction reaction (ORR) activity and selectivity. Their thermodynamic overpotential η values are 0.01 V, 0.03 V for Ga-N4B6’s two configurations and 0.02 V, 0 V for Sb-N4B6’s two configurations. Electronic structure calculations indicate that the PM single atom adsorbs OOH* intermediates and transfers electrons into them, resulting in the activation of the O-O bond, which facilitates the subsequent hydrogenation reaction. In summary, Sb-N4B6 and Ga-N4B6 exhibit extraordinary 2e− ORR performance, and their predicted activities are comparable to those of known outstanding catalysts (such as PtHg4 alloy). We propose effective strategies on how to enhance the 2e− ORR activities of carbon materials, elucidate the origin of the activity of potential catalysts, and provide insights for the design and development of electrocatalysts that can be used for H2O2 production.

Theoretical Insights into the Effect of Different Numbers of Thiophene Groups on Hydrogen Bond Interaction and Excited-State Intramolecular Proton-Transfer Process for Flavonoid Derivatives.

In this study, we systematically explored the impact of varying the number of thiophene groups on the hydrogen bond interaction and excited-state intramolecular proton-transfer (ESIPT) processes in flavonoid derivatives (STF, DTF, and TTF) using the density functional theory and time-dependent density functional theory methods. Initially, a thorough analysis of the optimized geometric structures revealed that the intramolecular hydrogen bond in the S1 state is enhanced and gradually weakened as the number of thiophene groups increases. To gain a deeper understanding of the hydrogen bond interaction, topological analysis, interaction region indicator scatter plots, and isosurface plots were employed. These images provide further insights that align with the structural analysis. Additionally, we observed a red-shift in the electronic spectra (absorption and fluorescence spectra), which is primarily attributed to the narrowing of the energy gap between the highest occupied molecular orbital and the lowest unoccupied molecular orbital, as elucidated by the frontier molecular orbitals. Furthermore, a combined analysis between the hole-electron distribution and the transition density matrix heat map shows that electron excitation involves the unidirectional charge-transfer mechanism. In the end, by conducting relaxed potential energy curve scans, we found that an increase in the number of thiophene groups elevates the energy barrier for ESIPT, making it more challenging for the reaction. In summary, our study underscores the vital effect of thiophene-substituted numbers in modulating the ESIPT process, which is able to provide valuable insights for the design and synthesis of desired organic fluorescent probes in the future.

Investigating the Impact of Friction Stir Processing on the Hydrogen Embrittlement in AA6082-T6 Heat-Treatable Aluminum Alloy

This study investigates the impact of friction stir processing (FSP) on the hydrogen embrittlement (HE) in AA6082-T6 heat-treatable aluminum alloy. The effects of different number of FSP passes and different hydrogen cathodic charging (HCC) conditions on the material’s response to HE are examined through comprehensive mechanical testing, microhardness analysis, and microstructural characterization. The results revealed that FSP leads to a decrease in yield strength, ultimate tensile strength, and microhardness, accompanied by an increase in energy absorption. The introduction of hydrogen through HCC significantly reduces mechanical properties, particularly in non-FSPed specimens. Notably, specimens with 8 FSP passes exhibit an interesting behavior with a slight increase in energy absorption and microhardness values after HCC. Microstructural analysis shows that FSP refines the microstructure, resulting in enhanced resistance to hydrogen-induced blistering effects. These findings contribute to the understanding of hydrogen embrittlement in FSPed aluminum alloys, providing insights for developing surface-modified materials suited for hydrogen-rich applications.Graphical

Forecasting price in a new hybrid neural network model with machine learning

A key aspect of asset investment and risk management is the study of forecasting stock prices. We investigate the machine learning stock price prediction in a new hybrid neural network model and put forth a forecasting method based on machine learning, composite data preprocessing method and the proposed new neural network model. To address the challenge of predicting stock prices in the face of market complexity and noise, we use the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm and the Savitzky–Golay (SG) filter to de-noise and enhance the data, and employ a neural network with three convolutional layers and a long short-term memory (LSTM) layer that enables it to capture complex temporal patterns in the data. We propose a new hybrid neural network prediction model (CEEMDAN-S-C-LSTM) and adopt a machine learning approach to compare it with the benchmark model using CSI 300 index data. The empirical results validate the effectiveness of the frequency decomposition algorithm and the convolutional layer, and demonstrate that our proposed model outperforms the benchmark model. Compared to the best benchmark model, the CEEMDAN-S-C-LSTM model proposed in this study demonstrates a significant improvement in performance. Specifically, it shows a 45.33% reduction in mean absolute error (MAE), 43.44% reduction in root mean square error (RMSE), 45.01% reduction in mean absolute percentage error (MAPE), and 3.90% improvement in coefficient of determination (R2). The study also explores the effect of different numbers of convolution layers and SG filters on the hybrid model. Our research expands on the use of neural networks and machine learning, offering a novel technical approach to making investment decisions and managing risks in financial systems.

Reduced Channel Interaction Improves Timbre Recognition Under Vocoder Simulation of Cochlear Implant Processing.

This study aimed to investigate the influence of the number of channels and channel interaction on timbre perception in cochlear implant (CI) processing. By utilizing vocoder simulations of CI processing, the effects of different numbers of channels and channel interaction were examined to assess their impact on timbre perception, an essential aspect of music and auditory performance. Fourteen CI recipients, with at least 1 year of CI device use, and two groups (N = 16 and N = 19) of normal hearing (NH) participants completed a timbre recognition (TR) task. NH participants were divided into two groups, with each group being tested on different aspects of the study. The first group underwent testing with varying numbers of channels (8, 12, 16, and 20) to determine an ideal number that closely reflected the TR performance of CI recipients. Subsequently, the second group of NH participants participated in the assessment of channel interaction, utilizing the identified ideal number of 20 channels, with three conditions: low interaction (54 dB/octave), medium interaction (24 dB/octave), and high interaction (12 dB/octave). Statistical analyses, including repeated-measures analysis of variance and pairwise comparisons, were conducted to examine the effects. The number of channels did not demonstrate a statistically significant effect on TR in NH participants ( p > 0.05). However, it was observed that the condition with 20 channels closely resembled the TR performance of CI recipients. In contrast, channel interaction exhibited a significant effect ( p < 0.001) on TR. Both the low interaction (54 dB/octave) and high interaction (12 dB/octave) conditions differed significantly from the actual CI recipients' performance. Timbre perception, a complex ability reliant on highly detailed spectral resolution, was not significantly influenced by the number of channels. However, channel interaction emerged as a significant factor affecting timbre perception. The differences observed under different channel interaction conditions suggest potential mechanisms, including reduced spectro-temporal resolution and degraded spectral cues. These findings highlight the importance of considering channel interaction and optimizing CI processing strategies to enhance music perception and overall auditory performance for CI recipients.

Intermolecular hydrogen bonding in DNA base pairs interacting with different numbers of bare and hydrated Li+: NBO, QTAIM, and computational spectroscopic studies

In this study, the effect of different numbers of Li+ interacting with various sites of DNA base pairs (adenine–thymine (AT) and cytosine-guanine (GC)) on the base pair structures, the strength of hydrogen bonding between the bases, and spectroscopic properties (IR and absorption spectra) of the base pairs was investigated. Two quantum computational analyses, the natural bonding orbitals (NBO) and the quantum theory of atoms in molecules (QTAIM), were used to follow the change in the strength of hydrogen bonds between the bases in each pair. The type of base pair’s site interacting with Li+ showed different effects on the change in the strength of the hydrogen bonds between the bases. The IR and absorption spectra of the lithiated base pairs were calculated and compared with those of bare base pairs. This comparison provided the changes in the spectra as a fingerprint for the structural identification of different lithiated base pairs. Also, the determination of the change in the strength of hydrogen bonds in the lithiated base pairs compared to their bare base pairs. In the other part of this study, the effect of the hydration of the attached Li+ in the structure of lithiated base pairs on the strength of their hydrogen bonds and spectra was investigated.

Ballistic and Charpy impact performance of basalt fiber reinforced polymer composites

AbstractIn this study, the behavior of composite material produced using basalt, a natural fabric, under ballistic and Charpy impact loads was investigated. The damage resistance of target plates under ballistic loads is an important design parameter in the structures that will be exposed to these loads. Cross‐ply fiber‐reinforced composite plates were produced with different numbers of layers. The composite specimens are produced using the vacuum infusion method, cross‐ply laminated with 3, 6, 9, and 12 layers ([0/90]3, [0/90]6, [0/90]9, and [0/90]12). Carbon/Epoxy specimens were also produced in identical sequences to be used as comparison material. The study aimed to understand the absorbed energy under ballistic loads and the required energy level under the Charpy impact loads of Basalt/Epoxy composite specimens with different thicknesses. Ballistic loads are applied to the specimens using a test set‐up including the sample holder mechanism and a firing system. Charpy impact tests were conducted using a standard test machine. According to the results, for 12‐layer specimens, the residual velocity difference between Basalt/Epoxy and Carbon/Epoxy is approximately 1.5 times. For the Basalt/Epoxy plate, increasing number of layers from 3 to 12 increases the amount of energy absorbed approximately 13 times. Although this increase is approximately 19 times for the Carbon/epoxy plate, when Basalt/Epoxy and Carbon/Epoxy plates are compared, Basalt/Epoxy absorbs more energy for all layer numbers. For the Charpy impact cases, it has been observed that Basalt/Epoxy absorbs approximately 34% more energy for flat wise impact and approximately 24% more energy for edgewise impact.Highlights The behavior of composite material produced using basalt, a natural fabric. Cross‐ply fiber‐reinforced composite plates. Effect of different number of layers considered. The damage resistance of target plates under ballistic loads. Comparison of basalt and carbon‐reinforced composites.

Experimental investigation of thermal performance and entropy analysis of sphere turbulators

In this study, the effects of different numbers of sphere turbulators placed in a circular-shaped test pipe under turbulent flow conditions on heat transfer, friction factor, thermal performance factor and entropy production were experimentally investigated. Water was used as the working fluid in the experiments carried out in the number range Re=8379-16701. Initially, plain pipe and supported plain pipe experiments were carried out by keeping the fluid inlet temperature and test pipe surface temperature constant. Then, 1, 2 and 3 sphere turbulator experiments were carried out under the same conditions, respectively. According to the findings, the maximum Nusselt number increase compared to the supported plain pipe was 83.34% for the test pipe with 3 sphere turbulators. The minimum and maximum Nu number increase rates of 3 sphere turbulators compared to 1 and 2 sphere turbulators were obtained as 42.61%-46.93% and 13.99%-17.11%, respectively. The increase rates of maximum friction factors of 1, 2 and 3 sphere turbulators compared to the supported plain pipe were 174.42%, 226.70% and 306.53%, respectively. In addition, the maximum total entropy rate for sphere turbulators was obtained as 0.058 W.m-1.K-1 for 1 sphere turbulator at Re = 8379. In terms of the second law, the maximum entropy generation of sphere turbulators compared to the supported plain pipe decreased by 79.39% for 3 sphere turbulators at Re = 8379. As a result, it was concluded that the optimum number of sphere turbulators is 3 sphere turbulators with the lowest entropy production number and the highest TPF value (1.28).

Simulation and research of chip-level micromixer with T-type based on ANSYS

A T-shaped chip-scale micromixer with built-in baffles is simulated and investigated based on ANSYS. This paper highlights on a comparative analysis of the effect of different numbers, rows, positions, angles and different Reynolds numbers (Re) of baffles on the mixing efficiency within a chip-scale micromixer. The simulation results demonstrate that: increase in the amount of baffles improved mixing of two fluids; the double rows of baffles significantly improve the mixing efficiency compared with the single rows of baffles; the position of the baffles relocates a certain distance to the outlet, which accelerates mixing of fluids in the chip-scale mixer; the mixing efficiency of baffle angle of 120° is also superior to that of 60°; the Re is between 0.1 and 2, which results in a high mixing efficiency; and the mixing efficiency slowly becomes lower when the Re is between 2 and higher mixing efficiencies with Re values between 0.1 and 2; Re between 2 and 40, the mixing efficiency slowly becomes lower; Re between 40 and 100, the mixing efficiency gradually rises.

Estimation of soil organic matter by in situ Vis-NIR spectroscopy using an automatically optimized hybrid model of convolutional neural network and long short-term memory network

It is attractive nowadays to estimate soil information based on in situ Visible and Near-infrared (Vis-NIR) spectroscopy. However, there exist a lot of errors, mainly due to modeling between soil properties and spectra. A hybrid of deep learning (DL) models seems an ideal approach for the modeling, e.g., convolutional neural network (CNN) and long short-term memory (LSTM). Nevertheless, it is not easy for a user, or not efficient for an automatic technique, such as Bayesian optimization (BO), to select hundreds of parameters of a hybrid of DL models. Recently, tree Parzen Estimator (TPE) based HyperBand within BO (BOHB) has been widely used in other fields for establishing a hybrid of DL models efficiently and automatically. Further, it is not quite clear if a DL model or a hybrid DL model performs well for different sample sizes and different numbers of spectral data. This study aimed to investigate these issues by evaluating performances of two hybrids of DL models (CNN-LSTM and LSTM-CNN) established using BOHB, based on soil spectra and organic matter contents of 670 soil samples collected from a forest in Nanning, southwest China. The performances were also compared with those of separate DL models and two commonly used methods, i.e., partial least square regression (PLSR) and random forest (RF), while the effect of sample sizes on the performances was also investigated. Besides, the effect of different numbers of spectra was also investigated, and the spectra were augmented using two methods, one by stacking scans of all measurement points of a sample and one by stacking preprocessed spectra using several methods. Results showed that, given less than 600 samples, accuracies of CNN-LSTM and LSTM-CNN were not higher than those of separate DL models, PLSR and RF. However, BOHB optimization largely improved the accuracies of the two hybrid models which were close to or higher than those of separate DL models, PLSR and RF. With the increase of sample sizes, accuracies of all models used in this study gradually increased and the DL models seemed to be more promising for large sample sizes. Furthermore, augmenting spectra with preprocessed data provided much more benefits than sample sizes, modeling methods, and optimization methods. Thus, it is promising to estimate soil using in situ Vis-NIR spectra based on a hybrid of DL models optimized using BOHB, particularly for large datasets or augmented spectra.