Backpropagation Research Articles

Aircraft engine fault diagnosis based on fused convolutional Transformer

Aircraft engines will face corrosion and erosion problems when working in harsh gas path environments for a long time, and the fault parameter characteristics are not obvious. Therefore, accurate aircraft engine fault diagnosis methods are of great significance to ensure the safe operation of aircraft. In order to improve the prediction accuracy, an aircraft engine fault diagnosis method based on fused convolutional Transformer is proposed. The self-attention mechanism is used to extract useful features and suppress redundant information. The maximum pooling layer (MaxPool) is introduced into the Transformer model to further reduce the model memory consumption and parameter quantity, and alleviate the overfitting phenomenon. The simulation data set of turbofan engines based on GasTurb modeling is used for verification. The results show that compared with the Transformer network and other traditional deep learning models Back Propagation Neural Networks (BP network), Convolutional Neural Networks (CNN) and Recurrent Neural Network (RNN), the accuracy rates are improved by 6.552%、28.117%、13.189%and respectively 10.29%, which proves the effectiveness of the proposed method and can provide a certain reference for aircraft engine fault diagnosis.

Personalized 3D Printing of Artificial Vertebrae: A Predictive Bone Density Modeling Approach for Robotic Cutting Applications

Robotic vertebral plate cutting poses significant challenges due to the complex bone structures of the lumbar spine, which consist of varying densities in cortical and cancellous regions. This study addresses these challenges by developing a predictive model for robotic vertebral plate cutting force and bone quality recognition through the fabrication of artificial vertebrae with controlled, consistent bone density. To address the variability in bone density between cortical and cancellous regions, CT data are utilized to predict target bone density, serving as a foundation for determining the optimal 3D printing process parameters. The proposed methodology integrates a Response Surface Methodology (RSM), Back Propagation (BP) neural network, and genetic algorithm (GA) to systematically evaluate the effects of key process parameters, such as the filling density, material flow rate, and layer thickness, on the printed vertebrae’s density. A one-factor experimental approach and RSM-based central composite design are applied to build an initial bone density prediction model, followed by Sobol’s sensitivity analysis to quantify the influence of each parameter. The GA-BP neural network model is then employed to rapidly and accurately identify optimal printing parameters for different bone layer densities. The resulting optimized models are used to fabricate personalized artificial lumbar vertebrae, which are subsequently validated through robotic cutting experiments. This research not only contributes to the advancement in personalized 3D printing technology but also provides a reliable framework for developing patient-specific surgical planning models in robot-assisted orthopedic surgery.

Predictions of Peak Discharge of Dam Failures Based on the Combined GA and BP Neural Networks

In this paper, the Artificial Neural Network (ANN) was utilized to predict the peak discharge of dam failures, which was based on the combined Genetic Algorithm (GA) and Back Propagation (BP) neural network. The dataset comprises 40 samples from self-conducted experiments and available literature. To compare the efficiency of the suggested approach, three evaluation metrics, including the coefficient of determination (R2), the root mean square error (RMSE) and the mean absolute error (MAE), were analyzed for both the BP neural network and the GA-BP neural network. The findings suggest that (1) The prediction accuracy of the GA-BP was better than that of the BP; and (2) Compared to BP, GA-BP demonstrated a 9.07% average improvement in R2, a 57.36% average reduction in MAE, and a 57.53% average reduction in RMSE. In addition, the results of GA-BP and semi-empirical formulas were compared and the effect of three parameters on the peak discharge was analyzed. The results showed that the GA-BP model could effectively predict the peak discharge of dam failures.

Adaptive control of an amplified piezoelectric-driven micropositioning stage based on notch filter structure inspiration and neural network online tuning

Abstract Piezoelectric-driven flexure micro/nanopositioning stages often exhibit a low-damping resonance mode, which can easily excite mechanical resonance during high-speed movement, and significantly impact the control system's stability, control bandwidth, and trajectory tracking accuracy. To mitigate the reliance on the precise modeling of stage dynamics inherent in current resonant controllers, an adaptive control method based on a back propagation (BP) neural network was designed to suppress resonance in real-time. First, a piezoelectric-driven flexure micro/nanopositioning stage system was constructed. Next, a feedback controller similar to a notch filter was designed, with bilinear transformation applied based on the system's inherent parameters to determine the initial values. Finally, the designed adaptive control method was tested through trajectory tracking experiments using a triangular wave signal. The experimental results showed that, when tracking the triangular wave signal, the maximum tracking error was reduced by 72.66% compared to proportional-integral (PI) control alone and by 68.66% compared to proportional integral control combined with a traditional notch filter. The tracking results demonstrate a significant improvement in the stage's stability and tracking accuracy.

Thriving through innovation: Boosting green tech performance in China's new energy sector

In response to the challenges of sustainable development, green technological innovation (GTI) is considered as a key driver of high-quality growth. However, current research often lacks a comprehensive assessment of GTI's multidimensional performance and rarely explores improvement pathways from a structural perspective. This study evaluates the GTI performance of China's new energy sector using data from 62 listed companies between 2015 and 2021, employing a Back Propagation (BP) neural network model optimized by a genetic algorithm. Additionally, it explores performance enhancement paths through fuzzy set Qualitative Comparative Analysis (fsQCA). The findings indicate that: Industrial structure and human resource levels significantly boost GTI performance in the new energy sector; Optimizing and upgrading industrial structures plays a crucial role in GTI improvement; High Research and development (R&D) investment and human capital strengthen market competitiveness, leading to higher GTI levels; The development of GTI in the new energy sector exhibited a general upward trend, with notable differences across sub-industries and regions from 2015 to 2021. Based on these results, this study recommends differentiated industry support, targeted regional policies, and enhanced talent cultivation to improve GTI performance in China's emerging energy industry.

A Finite-Time Path Tracking Control Scheme for Unmanned Vehicles Based on Multidimensional Taylor Network and Adaptive Filtering

Aiming at the path tracking control problem of unmanned vehicles under the conditions of model uncertainty and measurement noise, a finite-time path tracking control scheme based on multidimensional Taylor network (MTN) is proposed. The MTN model is used to characterize the uncertainty of the unmanned vehicle model, and the improved back propagation (BP) algorithm is used as its learning algorithm; an adaptive MTN filter is designed to filter out the measurement noise, MTN is used as a filter, and the least mean square (LMS) algorithm is used as its learning algorithm; an MTN finite-time controller is designed to perform precise path tracking control on the unmanned vehicle, which can quickly and accurately track the reference path; according to the finite-time control theory, the convergence proof verifies the effectiveness of the proposed method through unmanned vehicle simulation experiments.

Research on the prediction of blasting fragmentation in open-pit coal mines based on KPCA-BAS-BP

The blasting block size of open-pit mines is influenced by many factors, and the influencing factors have a very complex nonlinear relationship. Traditional empirical formulas and a single neural network model cannot meet the requirements of modern blasting safety. To improve the prediction accuracy of blasting block size, the measured data of Beskuduk open-pit coal mine is used as training and testing samples. Seven factors including rock tensile strength, rock compressive strength, and blast hole spacing are selected as input variables of the prediction model. The average size of blasting fragmentation X50 is used as the output variable of the prediction model. The kernel principal component analysis (KPCA) is adopted to reduce the dimensionality of the input variables. The beetle antennae search algorithm (BAS) is selected to optimize the parameters of the initial weights and thresholds of the back propagation (BP) neural network. Finally, prediction model of blasting fragmentation in open-pit coal mine based on KPCA-BAS-BP is established. The results show that the average relative error of the model is 1.77%, and the root mean square error is 1.52%. Compared with the unoptimized BP neural network and the BP neural network optimized by the artificial bee colony algorithm (ABC) model, this model has higher prediction accuracy and is more suitable for predicting the blasting block size of open-pit coal mines, it provides a new method for predicting the fragmentation of blasting under the influence of multiple factors, filling the gap in related theoretical research, and has certain practical application value.

Multi-input Fourier neural network and its sparrow search optimization

In engineering applications, the back-propagation (BP) neural network often encounters many limitations due to its slow convergence and high noise sensitivity, and meanwhile the reported Fourier neural networks have no ability to extract the features of multi-attribute input data. Hereby, This work proposes a gradient descent-based multi-input Fourier neural network after integrating the multi-layer perceptron with an overlapping Fourier neural network. Thereafter, related to the difficulty of deciding the global optimal parameter settings, an improved sparrow search algorithm is developed to optimize the parameter settings and solve high dimensional function optimization problems, after the Cat chaotic map and the mechanisms of population-size adjustment and parameter adaptiveness are designed to promote the sparrow search algorithm's ability to balance global exploration and local exploitation. The theoretical analysis shows that the improved algorithm's computational complexity is decided by its population size and the optimization problem's dimension. Numerically comparative experiments have validated that not only the acquired Fourier neural network can effectively extract the features of multi-attribute data with strong generalization ability, but also the improved algorithm has significant advantages in coping with high dimensional function optimization problems.

Analysis of Low-Frequency Sound Absorption Performance and Optimization of Structural Parameters for Acoustic Metamaterials for Spatial Double Helix Resonators

Low-frequency noise absorbers often require large structural dimensions, constraining their development in practical applications. In order to improve space utilization, an acoustic metamaterial with a spatial double helix, called a spatial double helix resonator (SDHR), is proposed in this paper. An analytical model of the spatial double-helix resonator is established and verified by numerical simulations and impedance tube experiments. By comparing the acoustic absorption coefficients of the spatial double-helix resonator, it is shown that the results of the analytical model, the numerical model, and the experiments are in good agreement, proving the accuracy of the theoretical model. The effects of different structural parameters on the peak sound absorption coefficient and resonance frequency are quantitatively revealed. The impedance variation law of the model is obtained, and the resistance and reactance distributions at the resonance frequency are analyzed. In the optimization model, the Back Propagation (BP) network is used to construct the mapping between the structural parameters and the resonance frequency and sound absorption coefficient, and this is used as the constraints of the equation, which is combined with Wild Horse Optimization (WHO) to establish the BP-WHO optimization model to minimize the volume of the spatial double helix resonator. The results show that, for a given noise frequency, the optimized structural parameters enhance the space utilization without affecting the performance of the space double helix resonator.

Research on the simulation method of a BP neural network PID control for stellar spectrum

This study investigated the multiple correlations among spectral simulation units based on digital micromirror device (DMD) spectral simulation, which leads to the problem that conventional spectral simulation methods such as PID control exhibit a low fitting accuracy or long fitting time in the spectral simulation of various targets. In this paper, a method of stellar spectrum simulation based on back propagation neural network-based PID (BP-PID) control is proposed to achieve high efficiency and high precision simulation of various spectral targets. The topology of the BP neural network was constructed based on the spectral modulation model of a DMD stellar spectrum simulation system, and the algorithm of the BP-PID control was designed. Finally, an experimental platform was built to verify the performance and spectral simulation accuracy of the BP-PID control algorithm. The results show that the overshoot and response time of the BP-PID control algorithm decreased by 79.01% and 30%, respectively compared with those of the PID control algorithm. The maximum spectral simulation accuracies of 2000K, 7000K, and 12000K color temperature increased by a factor of 2.311, 1.871, and 2.254, respectively, and the standard deviations of the spectral simulation error decreased by 56%, 41%, and 54%, respectively. In the range of 2000-12000K color temperature, the spectral simulation error of the BP-PID control algorithm is better than ±3.495%, and the standard deviation of the spectral simulation error is between 1.8255 and 2.2358. The proposed method can improve the spectral simulation accuracy and simulation efficiency of a star simulator, reduce the magnitude and spectrum calibration errors caused by the differential response, improve the star feature recognition accuracy of the orbiting star sensor, and hence, provide a theoretical and technical basis for the development of high-precision star sensors.

Back propagation neural network model for controlling artificial rocks petrophysical properties during manufacturing process

Artificial rocks are increasingly used in experiments, and it is important to make artificial rocks’ petrophysical properties similar to real rocks to obtain more valuable results. The effect of widely used five factors, grain size (GS), grain size distribution (GSD), mass fraction of cementing agent (MC), pressing pressure (PP), and pressing time (PT), on artificial rock petrophysical properties are analyzed. Three factors, GS, MC, and PP, which are suitable for establishing a quantitative model are opted. The relationships of 80 rock plugs’ petrophysical properties and manufacturing factors are studied, which indicates MC and PP are negatively correlated with porosity and permeability, and GS has a significant effect on permeability but little effect on porosity. Furthermore, a novel back propagation (BP) neural network model is proposed, which can be used to determine factor values during artificial rock manufacturing process. A series of artificial core plugs are manufactured relying on manufacturing factor values calculated by the BP neural network model, and their petrophysical properties are tested and compared with design petrophysical properties value to verify the model. Verification results show that comprehensive average error of porosity and permeability of artificial rock, including calculating error and manufacturing error, is 2.38 and 18.68%, respectively.

Splitting behavior of polyvinyl alcohol fiber reinforced iron ore tailings concrete

To promote solid waste reutilization and reduce natural river sand consumption, iron ore tailings (IOT) are expected to be used in concrete production as a potential substitute for river sand under the premise of comparable material properties with ordinary concrete. In this study an IOT concrete reinforced by Polyvinyl Alcohol (PVA) fiber has been proposed and the slump and splitting behavior under various IOT substitution rates and PVA fiber contents has been experimentally investigated. Slump tests reveal that the optimal PVA fiber content is 0.1 % and IOT substitution rate is 60 % from the view of workability. Splitting tensile tests show a monotonous strength growth with PVA fiber content increase. The incorporation of IOT has increased concrete splitting tensile strength but the strength is smaller than that of normal concrete as soon as IOT substitution rates exceed 50 %. A regression model and Back Propagation (BP) neural network model have been adopted to establish splitting tensile strength prediction model. Compared to regression model, the BP model has exhibited better accuracy which is suitable to predict the splitting tensile strength of PVA-IOT concrete with PVA content less than 0.3 %.

A method for optimizing different geometric shields of D-T neutron generators by combining BP neural network and Analytic Hierarchy Process

In this paper, an optimization of shielding structures with different geometries is established for the D-T neutron generator system by combining Back Propagation (BP) neural network and Analytic Hierarchy Process (AHP). The D-T neutron generator (Model NG-9) used in the system was developed independently by Northeast Normal University. After investigating the rule of shielding performance among spherical, cylindrical and cubic geometries, the spherical shield is selected for BP neural network prediction to determine the total dose rate through it. Information about spherical multilayer-shielding structures and properties calculated by MCNP code is used to train the neural network. The predicted result serves as a parameter of the evaluation function, which provides a comprehensive assessment of the dose rate penetrated the shield, the shielding mass, and the shielding volume. Together with AHP, the weight factors are determined for all the optimization objectives to construct the evaluation function. By comparing its values, the optimal shielding structures for spherical, cylindrical and cubic materials are found. Against MCNP simulated values, the total dose rates’ errors of the optimal shielding structures for the sphere, cylinder, and cube are 1.72 %, -4.94 %, and -5.17 %, respectively. This result demonstrates that the combination of BP neural network and AHP is more effective in addressing multi-objective optimization problems related to the design of radiation shielding for various geometries.

Analysis and prediction of energy consumption in office buildings with variable refrigerant flow systems: A case study

Accurately predicting building air-conditioning energy consumption is particularly important for energy management. However, for small datasets, there is a lack of simple and effective prediction methods applicable to multiple buildings. In order to fully explore the energy consumption mode of Variable Refrigerant Flow (VRF) air conditioning system and improve the accuracy of the model. This study uses the weather data of an outdoor meteorological data collection system and the air conditioning system and energy consumption data of an office building in Jinan. Initial analysis revealed that occupant behavior and daily variations in outdoor air temperatures significantly impact the energy consumption of air conditioning systems. Then, by applying the random forest algorithm and Pearson correlation analysis, the study identified daily average outdoor air temperature and VRF indoor unit running time as critical features for prediction. Subsequently, three machine learning models—Multiple Linear Regression (MLR), Back Propagation Neural Network (BPNN), and Long Short-Term Memory (LSTM) neural network—were developed and assessed using a 4:1 training-to-testing set ratio. Comparative analysis revealed that all models performed robustly, with the MLR model exhibited superior predictive accuracy (R2 (Coefficient of Determination) = 0.98, MAE (Mean Absolute Error) = 0.005 kWh/(m2·day), and RMSE (Root Mean Square Error) = 0.007 kWh/(m2·day) for the testing set) and simplicity. Furthermore, the effectiveness of the MLR model in handling small sample sizes was confirmed through cross-validation of data from two additional office buildings, highlighting its suitability for predicting energy consumption of VRF air conditioning systems in office buildings.

Rapid Identification of the Geographical Origin of the Chinese Mitten Crab (Eriocheir sinensis) Using Near-Infrared Spectroscopy.

The Chinese mitten crab (Eriocheir sinensis) is highly valued by consumers for its delicious taste and high nutritional content, including proteins and trace elements, giving it significant economic value. However, variations in taste and nutritional value among crabs from different regions lead to considerable price differences, fueling the prevalence of counterfeit crabs in the market. Currently, there are no rapid detection methods to verify the origin of Chinese mitten crabs, making it crucial to develop fast and accurate detection techniques to protect consumer rights. This study focused on Chinese mitten crabs from different regions, specifically Hongze Lake, Tuo Lake, and Weishan Lake, by collecting near-infrared (NIR) diffuse reflectance spectral data from both the abdomen and carapace regions of the crabs. To eliminate noise from the spectral data, pretreatment was performed using Savitzky-Golay (SG) smoothing, Standard Normal Variate (SNV) transformation, and Multiplicative Scatter Correction (MSC). Key wavelengths reflecting the origin of Chinese mitten crabs were selected using Competitive Adaptive Reweighted Sampling (CARS), Bootstrap Soft Shrinkage (BOSS), and Uninformative Variable Elimination (UVE) algorithms. Finally, Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Back Propagation Neural Network (BP) models were developed for rapid detection of crab origin. The results demonstrated that MSC provided the best preprocessing performance for NIR spectral data from both the abdomen and back of the crabs. For abdomen data, the SVM model developed using feature wavelengths selected by the CARS algorithm after MSC preprocessing achieved the highest accuracy (Acc) of 90.00%, with precision (P), recall (R), and F1-score for crabs from Weishan Lake at 89.29%, 86.21%, and 87.72%, respectively; for crabs from Tuo Lake at 86.96%, 95.24%, and 90.91%; and for crabs from Hongze Lake at 90.00%, 93.10%, and 91.53%. For carapace data, the SVM model based on wavelengths selected by the BOSS algorithm after MSC pretreatment achieved the best performance, with an Acc of 87.50%, and P, R, and F1 for crabs from Weishan Lake at 77.14%, 93.10%, and 84.38%; for Tuo Lake crabs at 100%, 90.47%, and 95.00%; and for Hongze Lake crabs at 92.31%, 80.00%, and 85.71%. In conclusion, NIR spectroscopy can effectively detect the origin of Chinese mitten crabs, providing technical support for developing rapid detection instruments and thereby safeguarding consumer rights.

Classification of green tea and quantitative prediction of catechin content based on indicator displacement array sensors

Green tea, renowned for its health benefits and cultural significance, exhibits variations in quality and biochemical composition across different grades and varieties. Traditional quality assessment methods are often labour-intensive and time-consuming, rendering them impractical for rapid analysis. This study employed a sensitive indicator displacement array sensor along with supervised learning algorithms—Extreme learning machine, Support vector machine, Random forest, and Back propagation neural network—to accurately classify green tea varieties and grades, as well as predict catechin content. Sensor components were initially screened based on their responsiveness to epigallocatechin gallate, and subsequently developed using specific indicator-receptor combinations: Pyrocatechol Violet/Fe2+, Alizarin Red S/Phenylboronic Acid, and Bromopyrogallol Red/3-Nitrophenylboronic Acid. Validation was performed through monitoring changes in RGB values and UV–visible spectroscopy. Discrimination among Anji white tea grades was assessed using principal component analysis and further refined through supervised algorithms, achieving high accuracy rates, especially with Random forest and Extreme learning machine models, which excelled in both calibration and prediction sets. These methodologies were extended to classify ten green tea varieties, demonstrating robustness and high accuracy in a multiclass setting. Independent validation confirmed the models’ efficacy in-class differentiation, with most achieving an accuracy of over 90%. Moreover, the quantitative prediction of catechin content was enhanced using nonlinear models, specifically Support vector regression optimized with the Particle swarm optimization algorithm. These models exhibited superior performance compared to linear models, particularly in effectively predicting individual catechin levels. These findings highlight the potential of combining sensor technologies with supervised learning algorithms to enhance tea quality assessment and categorization, establishing a robust framework for future applications in tea quality control.

Potential of Hyperspectral Data Combined With Optimal Band Combination Algorithm for Estimating Soil Organic Carbon Content in Lakeside Oasis

ABSTRACTAccurate estimation of soil organic carbon (SOC) content is essential for promoting regional sustainable agriculture and improving land quality. Visible and near‐infrared (Vis‐NIR) near‐Earth remote sensing spectroscopy has become an effective alternative to the traditional time‐consuming and costly methods due to its high‐resolution and nondestructive application, but it is vulnerable to the redundancy of spectral information and the overlap between bands. This study delves into the potential of optimal spectral parameters for estimating SOC in arid lakeside oases, using Bosten Lake in Xinjiang, China, as a focal point. Soil samples (0–10 cm, 10–20 cm, 20–30 cm, 30–40 cm) were collected, and their SOC content and hyperspectral reflectance were measured. The spectral data underwent preprocessing techniques, including continuum removal (CR), standard normal variate (SNV), and continuous wavelet transform (CWT). SOC content was predicted using back propagation neural network models constructed based on one‐dimensional (1D), two‐dimensional (2D), and three‐dimensional (3D) correlation coefficients. Results showcased the effectiveness of the CWT method in accentuating potential spectral information and enhancing variable correlation. Among the indices, 3D exhibited the highest performance (R2 = 0.82, RPD = 2.02 for TDI‐1 at 0–10 cm; R2 = 0.85, RPD = 2.28 for TDI‐2 at 10–20 cm; R2 = 0.83, RPD = 2.24 for TDI‐1 at 20–30 cm; R2 = 0.86, RPD = 2.53 for TDI‐4 at 30–40 cm), followed by 2D and then 1D. These insights offer guidance for future strategies in hyperspectral data preprocessing and spectral index determination, facilitating SOC spatial distribution mapping and advancing sustainable agricultural planning. They also have implications for determining the spatial distribution of SOC content based on spatial interpolation, which would contribute to regional agricultural planning and sustainable development.

Soil Salinity Inversion Based on a Stacking Integrated Learning Algorithm

Soil salinization is an essential risk factor for agricultural development and food security, and obtaining regional soil salinity information more reliably remains a priority problem to be solved. To improve the accuracy of soil salinity inversion, this study focuses on the Manas River Basin oasis area, the largest oasis farming area in Xinjiang, as the study area and proposes a new soil salinity inversion model based on stacked integrated learning algorithms. Firstly, we selected four machine learning regression models, namely, random forest (RF), back propagation neural network, support vector regression, and convolutional neural network, for performance evaluation. Based on the model performance, we selected the more effective RF and BPNN as the basic regression models and further constructed a stacking integrated learning model. This stacking integration learning model improved the prediction accuracy by training a secondary model to fuse the prediction results of these two basic models as new features. We compared and analyzed the stacking integrated learning model with four single machine learning regression models. Findings indicated that the stacking integrated learning regression model fitted better and had good stability; on the test set, the stacking integrated learning regression model showed a relative increase of 8.2% in R2, a relative decrease of 14.0% in RMSE, and a relative increase of 6.5% in RPD when compared to the RF model, which was the single most effective machine learning regression model, and the stacking model was able to achieve soil salinity inversion more accurately. The soil salinity in the oasis areas of the Manas River Basin tended to decrease from north to south from 2016 to 2020 from a spatial point of view, and it was reduced in April from a temporal point of view. The percentage of pixels with a high soil salinity content of 2.75–2.80 g kg−1 in the study area had decreased by 19.6% in April 2020 compared to April 2016. The innovatively constructed stacking integrated learning regression model improved the accuracy of soil salinity estimation on the basis of the superior results obtained in the training of the single optimal machine learning regression model. As a consequence, this model can provide technological backup for fast monitoring and inversion of soil salinity as well as prevention and containment of salinization.

TFE3 and TP53 were novel diagnostic biomarkers related to mitochondrial autophagy in chronic rhinosinusitis with nasal polyps.

Chronic rhinosinusitis with nasal polyps (CRSwNP) belongs to a subtype of Chronic rhinosinusitis which is a heterogeneous inflammatory condition. It has been reported that mitophagy may provide a new therapeutic option for CRSwNP. The GSE136825 (training dataset) and GSE179265 (validation dataset) were scoured from the Gene Expression Omnibus database. The candidate genes related to mitophagy were identified by differential expression analysis. Subsequently, the biomarkers were selected from the machine learning, Receiver Operating Characteristic curves, and expression level verification. A backpropagation (BP) neural network was generated to evaluate the diagnostic ability of biomarkers. In addition, the infiltration abundance of immune cells, potential drugs, and related ear-nose-throat (ENT) diseases were analyzed based on the biomarkers. Finally, qPCR analysis was performed to verify these biomarkers. A total of 8 candidate genes were identified by overlapping 3,400 differentially expressed genes (DEGs) and 72 mitophagy-related genes Subsequently, TFE3 and TP53 were identified as biomarkers of CRSwNP, and the area under the curves (AUC) of the BP neural network was 0.74, which indicated that the biomarkers had excellent abilities. TFE3 and TP53 were co-enriched in the cancer pathway, cell cycle, endocytosis, etc. What's more, Macrophage and Immature dendritic cells had significant correlations with biomarkers. The drugs (Doxorubicin, Tetrachlorodibenzodioxin, etc.) and the ear-nose-throat diseases (hearing loss, sensorineural, tinnitus, etc.) related to biomarkers were predicted. Ultimately, qPCR results showed that the expression levels of TFE3 and TP53 in polyp tissue of CRSwNP were increased. Overall, TFE3 and TP53 could be used as biomarkers or potential therapeutic targets to diagnose and treat CRSwNP.

Comparative Study of Water Quality Prediction Methods Based on Different Artificial Neural Network

The prediction of future data using existing data is an effective tool for regional planning and watershed management. The back propagation neural network （BPNN） and convolutional neural network （CNN） were used to construct a prediction model based on the water quality index of Hengyang in Xiangjiang River Basin from April to May 2022 and the results of permanganate index prediction by different models were compared. The prediction results displayed by BPNN could predict the water quality； however, overfitting occurred during the prediction. BPNN modified by particle swarm optimization （PSO） could avoid overfitting, which improved the parameter selection method of the BPNN mode. The CNN model had a better prediction effect, which had a more complex structure and a more scientific fitting method to avoid the model falling into the local extreme value during the fitting process and improve the accuracy of the model prediction results. The evaluation parameters including root-mean-square error （RMSE）, coefficient of determination （R2）, and mean absolute error （MAE） were used to predict the accuracy of the network. Compared with that of the traditional BPNN model, PSO-BPNN reduced the RESM of the test set from 0.278 2 mg·L-1 to 0.210 9 mg·L-1, reduced the MAE of the test set from 0.222 3 mg·L-1 to 0.153 7 mg·L-1 and increased the R2 of the test set from 0.864 0 to 0.921 8, which indicated that PSO-BPNN had more stable fitting ability. RMSE, MAE, and R2 of the test set in the CNN model were 0.122 0 mg·L-1, 0.092 7 mg·L-1, and 0.970 5, respectively, which showed that CNN had a better fitting and prediction effect than that of BPNN.