Backpropagation Learning Research Articles

Prediction of solid circulation rate in an internal circulating fluidized bed: An empirical and ANN approach

Internal Circulating Fluidized beds (ICFBs) are interested in various industries because of their higher thermal efficiency and high reaction rates. However, understanding the complex flow hydrodynamics inside an ICFB is challenging. Also, the experiments needed are expensive and time-consuming. Therefore this study aims to predict the solids circulation rate in an in-house ICFB (0.3 m internal diameter x 3.0 m height) using an empirical model and an Artificial Neural Network (ANN) technique. The solid circulation rates measured at different operating and design conditions using a high-speed video camera are utilized to develop the models mentioned earlier. A dimensionless approach and nonlinear regression models are adopted to derive the empirical model. The Analysis of Variance (ANOVA) technique calculates F-number and their corresponding probabilities (P-values). The ANN model is developed with four input variables: particle size, static bed height, gap height, and gas superficial velocity. Multi-layer Perception model (MLP) with the Feedforward Back Propagation learning rule is employed to build the ANN model. A single hidden layer with nine neurons predicted a close solid circulation rate with minimal error over the empirical model compared to experimental data. Further, the empirical and ANN model's predicting capability is tested against literature data.

Application of artificial neural network for prediction of fluoride removal efficiency using neutralized activated red mud from aqueous medium in a continuous fixed bed column.

The present research work approaches the removal of fluoride from aqueous medium using neutralized activated red mud (NARM) in a continuous fixed bed column. Artificial neural network (ANN) technique was applied effectively for optimization of the model for the practicability of the removal process. The consequences of various experimental variables, like bed length, adsorbate concentration, experimental time, and adsorbate solution flow rate are studied to know the breakthrough point and saturation times. The highest removal potentiality of NARM was considered to be 3.815mgg-1 of F- in the bed height of 15cm, starting concentration 1ppm, susceptible time 120min, adsorbate solution flow rate 0.5mLmin-1, and constant room temperature, respectively. Bohart-Adams and Thomas models were considered to describe the fixed bed column effect to the bed height and adsorbate concentrations. The experimental data were applied to a back propagation (BP) learning algorithm programme with a four-seven-one architecture model. The artificial neural network model was considered to be functioning correctly as absolute relative percentage error throughout the learning period. Differentiation between the predicted outcomes from ANN model and actual results from experimental analysis affords a high degree of correlation (R2 = 0.998) stipulating that the model was able to predict the adsorption efficiency. Experimented adsorbent materials were characterized using different instrumental analysis that is scanning electron microscopy-energy dispersive X-ray spectroscopy (SEM-EDS), Fourier transform infrared spectroscopy (FTIR), and X-ray diffraction (XRD).

Multi-objective optimisation of mild steel embossed plate shear connector using artificial neural network-integrated genetic algorithm

A hybrid artificial neural network (ANN), integrated with a genetic algorithm (GA), was used for the multi-objective optimisation of mild steel embossed plate shear connector and the optimisation responses were examined by the push-out tests in accordance with BS 5400-5:2005 code. This innovative shear connection enables a greater frictional interlocking with the composite girder, allowing it to sustain a higher transverse load and gain a better bearing capacity at the same time retaining ductility. To obtain the ultimate load and relative slip data, the Taguchi experimental design was used, which took into account the three most influencing parameters: length, height, and thickness of the shear connector, which are the input elements for optimisation. The ANN-integrated GA approach is used for multi-objective optimisation, in which optimisation begins with modelling of the ANN and its output is optimised using GA. Five multi-layered (three) perceptron models that have different number of neurons in the hidden layer are developed using a feed-forward backpropagation learning algorithm. Logarithmic sigmoid (Logsig) and linear (Purelin) transfer functions from the Levenberg–Marquardt approach are adopted in this model. Using the best ANN model response values (3−11−2), multi-objective optimisation with GA was performed. The following optimum process parameters are established in this study: length, 40 mm; height, 124.97 mm; thickness of the connector, 12 mm for the maximum ultimate load of 431.991 kN and minimised relative slip of 2.056 mm, both with more than 95 % confidence level.

A Novel Bidirectional Gated Recurrent Unit-Based Soft Sensor Modeling Framework for Quality Prediction in Manufacturing Processes

Quality prediction is very important for improving the accuracy of quality control and the stability of product quality in manufacturing processes. However, the complex time series with high dimension, nonlinearity, and dynamics brings great challenges to the traditional quality prediction methods. To address this issue, a new soft sensor modeling framework is proposed for quality prediction. Specifically, the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${k}$ </tex-math></inline-formula> -nearest neighbor mutual information is first used to mine the inherent relationship between process and quality variables for dimension reduction and variable selection. Then, a bidirectional gated recurrent unit structure is designed for dynamic, nonlinear soft sensor modeling, where the historical and future information inside industrial time series and relevant features have been fully used for quality prediction, and the backpropagation learning algorithm is given in detail. Finally, a typical manufacturing process, the hot rolling process, is used for verification, and the simulation and comparison results show that the new method is able to predict the final product quality with higher accuracy and efficiency.

ANN AND ANALYTICAL SOLUTIONS TO RELATIVISTIC ISOTHERMAL GAS SPHERES

Relativistic isothermal gas spheres are a powerful tool to model many astronomical objects, like compact stars and clusters of galaxies. In the present paper, we introduce an artificial neural network (ANN) algorithm and Taylor series to model the relativistic gas spheres using Tolman-Oppenheimer-Volkoff differential equations (TOV). Comparing the analytical solutions with the numerical ones revealed good agreement with maximum relative errors of 10−3. The ANN algorithm implements a three-layer feed-forward neural network built using a back-propagation learning technique that is based on the gradient descent rule. We analyzed the massradius relations and the density profiles of the relativistic isothermal gas spheres against different relativistic parameters and compared the ANN solutions with the analytical ones. The comparison between the two solutions reflects the efficiency of using the ANN to solve TOV equations.

A Hybrid ARFIMA Wavelet Artificial Neural Network Model for DJIA Index Forecasting

This paper proposes a hybrid modelling approach for forecasting returns and volatilities of the stock market. The model, called ARFIMA-WLLWNN model, integrates the advantages of the ARFIMA model, the wavelet decomposition technique (namely, the discrete MODWT with Daubechies least asymmetric wavelet filter) and artificial neural network (namely, the LLWNN neural network). The model develops through a two-phase approach. In phase one, a wavelet decomposition improves the forecasting accuracy of the LLWNN neural network, resulting in the Wavelet Local Linear Wavelet Neural Network (WLLWNN) model. The Back Propagation and Particle Swarm Optimization (PSO) learning algorithms optimize the WLLWNN structure. In phase two, the residuals of an ARFIMA model of the conditional mean become the input to the WLLWNN model. The hybrid ARFIMA-WLLWNN model is evaluated using daily returns of the Dow Jones Industrial Average index over 01/05/2010 to 02/11/2020. The experimental results indicate that the PSO-optimized version of the hybrid ARFIMA-WLLWNN outperforms the LLWNN, WLLWNN, ARFIMA-LLWNN, and the ARFIMA-HYAPARCH models and provides more accurate out-of-sample forecasts over validation horizons of one, five and twenty-two days.

Research on Financial Early Warning Based on Combination Forecasting Model

Since entering the 21st century, “economic globalization” has become a hot topic. Under the impact of “economic globalization”, the competition of the Chinese domestic market continues to intensify, and Chinese enterprises are facing enormous pressure for survival and development. Among them, there are many cases of poor business operation caused by financial crisis which have directly put these companies in trouble, even causing them to go bankrupt. Therefore, it is very practical to establish a scientific data model to analyze and predict the financial data of enterprises. It can not only monitor the financial status of the enterprise in real time, but also play an effective financial early warning role. This research focuses on using the combined forecasting method to establish a more comprehensive financial early warning model to solve the related financial crisis forecasting problem. Specifically, two different forecasting methods are first adopted in this study to conduct financial early warning research. The first is time series forecasting. It is a dynamic data processing statistical method, which is often used in forecasting research in the business field. The second is the BP neural network algorithm (referred to as BP), which is an error back-propagation learning algorithm, which is often used in the field of artificial intelligence. Then, the prediction error values of the two methods are compared and they are applied to the combined prediction method. Finally, a new error prediction formula is obtained. The result shows that the BP method provides the best performance over others, while the combinational forecasting method offers better performance than any single method.

Online intelligent fault diagnosis of redundant sensors in PWR based on artificial neural network

Sensors in the primary circuit of a pressurized water reactor (PWR) are normally designed with redundant structures to improve system safety and reliability. However, reliability of the actual system is often lower than that obtained by theoretical calculation due to the inevitable occurrence of common mode fault (CMF), which is a dependent failure event that can cause multiple failures in redundant channels. CMF may increase the reliability deviation of the system by orders of magnitude and, hence, seriously affects the reliability of the system. To mitigate the CMF of redundant sensors in nuclear power plants, an artificial neural network (ANN) can serve as a data-driven analytic model to monitor sensor parameters, to identify any possible abnormal status of the sensors, and provide an early warning. In this study, by using the high-fidelity dataset obtained in a full-scope PWR simulator as training, validation, and test data, a relevant parameter-based ANN black-box model (RPANN) was established by employing the back-propagation (BP) learning algorithm, which was then defined as an analytic redundancy. Time series-based ANN checking models (TSANNs) were also established for each of the input and output parameters of the RPANN in order to identify its abnormal state based on historical data in the past. When combined with the existing hardware redundancy, the ANN-based analytic redundancy can serve as an online monitoring tool of the hardware status and an online diagnosis strategy for sensor faults. Furthermore, ANN-based analytic redundancy can replace faulty hardware sensors to analytically reconstruct the reading of the monitored sensor parameter without having to reduce the reactor output power or even shut down the reactor for emergency maintenance so that the on-site calibration frequency of hardware sensors in redundant channels can be effectively reduced. This is not only of vital importance in reducing operation and maintenance costs of existing PWR power plants but also plays an important role in building reactor operation schemes with rapid and frequent changes in power output in the future. Simultaneously, the diverse redundancy combining analytic software redundancy and physical hardware redundancy can effectively reduce the threat of CMF of hardware sensors on the operation safety of reactor systems.

Ultralow Power Always-On Intelligent and Connected SNN-Based System for Multimedia IoT-Enabled Applications

The recent advances in artificial neural networks (ANNs) have created immense opportunities to achieve excellent results on the Internet of Things (IoT), which serve the uses of various real-time smart applications, such as in computer vision and speech recognition. However, the efficiency of ANNs comes at the expense of a huge number of computational resources. This tends to necessitate larger and wider ANNs unapplicable for embedded systems with limited hardware resources, e.g., mobile and wearable devices. To that end, biologically realistic spiking neural networks (SNNs) are first employed to build an always-on intelligent and connected integrated IoT-enabled system with ultralow power consumption, where we encode the temporal dynamic stimuli into effective, efficient, and reconstructable spike patterns to facilitate the subsequent processing. Herein, neural encoding plays a key role in faithfully describing the temporally rich patterns for downstream cognitive tasks. Therefore, a novel nonlinear piecewise latency coding approach for a fully event-driven SNN system is developed. Moreover, a surrogate postsynaptic potential kernel function is utilized to address the nondifferential nature of the spike generation scheme when using the error backpropagation learning method. The effectiveness of the proposal tandem with SNNs has been corroborated by indicative empirical results on different data sets serving cognitive tasks.

VLA-SMILES: Variable-Length-Array SMILES Descriptors in Neural Network-Based QSAR Modeling

Machine learning represents a milestone in data-driven research, including material informatics, robotics, and computer-aided drug discovery. With the continuously growing virtual and synthetically available chemical space, efficient and robust quantitative structure–activity relationship (QSAR) methods are required to uncover molecules with desired properties. Herein, we propose variable-length-array SMILES-based (VLA-SMILES) structural descriptors that expand conventional SMILES descriptors widely used in machine learning. This structural representation extends the family of numerically coded SMILES, particularly binary SMILES, to expedite the discovery of new deep learning QSAR models with high predictive ability. VLA-SMILES descriptors were shown to speed up the training of QSAR models based on multilayer perceptron (MLP) with optimized backpropagation (ATransformedBP), resilient propagation (iRPROP‒), and Adam optimization learning algorithms featuring rational train–test splitting, while improving the predictive ability toward the more compute-intensive binary SMILES representation format. All the tested MLPs under the same length-array-based SMILES descriptors showed similar predictive ability and convergence rate of training in combination with the considered learning procedures. Validation with the Kennard–Stone train–test splitting based on the structural descriptor similarity metrics was found more effective than the partitioning with the ranking by activity based on biological activity values metrics for the entire set of VLA-SMILES featured QSAR. Robustness and the predictive ability of MLP models based on VLA-SMILES were assessed via the method of QSAR parametric model validation. In addition, the method of the statistical H0 hypothesis testing of the linear regression between real and observed activities based on the F2,n−2 -criteria was used for predictability estimation among VLA-SMILES featured QSAR-MLPs (with n being the volume of the testing set). Both approaches of QSAR parametric model validation and statistical hypothesis testing were found to correlate when used for the quantitative evaluation of predictabilities of the designed QSAR models with VLA-SMILES descriptors.

Digital twin for rapid damage detection of a fixed net panel in the sea

The fishing net is the most important component of aquaculture net cages. Once damaged, it can cause substantial economic losses and ecological problems to the aquaculture industry. To avoid these subsequent issues caused by the damage to the fishing net, seeking a computerized, labor-saving approach and detecting damage in real time becomes the primary task of this paper. Inspired by recent development of the artificial neural networks and machine learning, this study proposes a fast and accurate approach for fishing net damage detection based on digital twin. Firstly, time-domain numerical simulations of the fishing net are conducted in a series of wave and current conditions to develop the artificial neural network-based digital twin. Then, the significant wave height, the spectral peak period, and tensions of vertical and horizontal ropes are used as input variables during the artificial neural network training; the intact and damaged states of the fishing net are considered outputs. Alongside this, the back-propagation learning algorithm is used for training to maximize damage detection performance. The results highlight that the digital twin model can effectively identify the fishing net damage using the sensor data, and the accuracy of the damage detection reaches above 93%. When the damage occurs at different net positions, the prediction accuracies of the artificial neural network model for training and testing sets are 93.97% and 93.00%, respectively. Regarding different wave-current directions, the prediction accuracies of the artificial neural network model are 99.87 % and 99.34 % for training and testing sets. Moreover, the developed digital twin can accurately detect the damage of the net even sea conditions and sensor data are not included in the training. The digital twin model proposed in the present study can potentially be used for damage detection in other aquaculture components and structures.

Particle Swarm-Assisted Artificial Neural Networks for Making Liquefied Natural Gas Processes Feasible Under Varying Feed Conditions

Natural gas (NG) has been widely recognized as a cleaner fuel compared to other fossil fuels. Reserves of NG are typically located in remote areas, and their conditions and compositions vary geographically. The NG from such areas is transported in the form of liquefied natural gas (LNG). Liquefying NG is highly complex. Generally, the process is designed to be carried out under fixed natural gas (NG) conditions; hence, it may not perform well under variable NG conditions. Considering this issue, the use of an artificial intelligence approach, rather than the conventional optimization one, was investigated to make the LNG process feasible under variable NG conditions. This study is the first in this research area to train an artificial neural network (ANN) using the particle swarm optimization (PSO) algorithm as a learning method. The developed PSO-ANN model was used to predict the decision variables of a single mixed refrigerant (SMR) LNG process for its feasible design under varying NG conditions. The correctness of the predicted set of decision variables (NG conditions) was verified by inputting them into Aspen HYSYS. The output of the SMR-LNG process was the overall power at a constrained minimum internal temperature approach (MITA) value, i.e., 1.0 ≤ MITA ≤3.0. The prediction results of the PSO-ANN model were compared with those of the classical ANN backpropagation learning method. The success rate of the proposed PSO-ANN model was 80%. Furthermore, the proposed model can make the LNG process feasible for a diverse range of temperature and pressure values. A feasible process with a better MITA value can also be achieved by tuning the model parameters.

A comparison between artificial neural network algorithms and empirical equations applied to submerged weir scour evolution prediction

Estimating the time evolution of a local scour hole downstream of submerged weirs can help determine the maximum scour depth and length and is essential to designing submerged weir foundations. In the current study, artificial neural networks with a backpropagation learning algorithm were used to estimate the temporal variation of scour profiles downstream of submerged weirs under clear water conditions. Physical factors, such as the flow condition, weir size, and sediment characteristics, are general scour considerations. Two sets of data combinations, namely original and non-dimensional data, were utilized in developing the backpropagation network (BPN) model. Using the data combinations and a trial-and-error method, the appropriate number of hidden neurons in the model architecture was determined. The results indicated that using non-dimensional variables as inputs in estimating the time evolution of scour holes downstream of a submerged weir is more suitable for the BPN model. Comparisons of the prediction results of the BPN model and an empirical regression formula revealed that the BPN model was more accurate and more direct in predicting the temporal variation of the local scour hole profile. In addition, sensitivity analysis showed that the dimensionless time exerted the most significant impact on the time evolution of the non-dimensional scour hole profile.

Patient's data privacy protection in medical healthcare transmission services using back propagation learning

Medical healthcare services rely on communication technologies for exchanging digital records of patients. The new digitalizing of health records brings a specific change in healthcare services. The Electronic Health Records (EHRs) contains cumulative information about patients, such as medical history, observations, diagnostics, specimens, and reports. EHRs are sensitive information readily available for patient's and healthcare providers’ access while maintaining privacy. Therefore, preserving security and privacy is of utmost importance for healthcare systems since it reduces the impact of adversaries on EHR exchange and transmissions through wearable antennas. This research introduces Healthcare Data Privacy (HDP) through Backpropagation Learning (BL) to improve privacy maintenance in medical healthcare transmission services using wearable devices. The proposed method identifies the need for encrypting and decrypting the accumulated healthcare data based on data integrity verifications. It operates on two levels for verifying the security measures to prevent data losses in successive transmissions of wearable devices.

High-temperature deformation characteristics and constitutive models of Inconel 625 superalloy

The isothermal compressive tests are performed to analyze the hot deformation characteristics of Inconel 625 superalloy over a wide temperature range of 1000–1150 °C and the strain rate scope of 0.01–10 s−1. The hot deformation mechanisms and the impacts of deformation parameters on the thermal deformation features, are analyzed. It reveals that the true stress first rises to a maximum stress level and then progressively drops to a comparatively stable level, demonstrating apparent work-hardening (WH) and dynamic softening (DS) features. Meanwhile, the true stress swiftly rises with the reduced deformation temperature or the increased strain rate. Moreover, the almost fine and uniform grains surround the whole microstructure under relatively large true strain, elevated temperature, and depressed strain rate. Besides, three kinds of constitutive models, including the Arrhenius-type (AT) model, physical-based (PB) model, and artificial-neural-network (ANN) model, complied with the backpropagation learning (BP) and new binary particle swarm optimization (NBPSO) algorithms are developed, respectively. The correlation coefficient (R) and absolute relative error (ARE) distribution are estimated to assess the predictive abilities of the established models. It can be concluded that the maximum absolute relative errors are near 12%, 15%, and 7.2% for AT, PB, and NBPSO-BP ANN models, respectively. Meanwhile, the correlation coefficient of the NBPSO-BP ANN model is as high as 0.9983, while the corresponding ones for the AT and PB models are 0.9951 and 0.9903, respectively. Thus, the established NBPSO-BP ANN model can depict the hot deformation features of the studied superalloy precisely.

Exploration on the Construction of Cross-Border E-Commerce Logistics System Using Deep Learning

The intention of this article is to solve the disadvantages of the current logistics model and promote the healthy development of modern cross-border e-commerce (CBEC) Logistics. First, this paper expounds on and compares the traditional and CBEC logistics models. Then, the CBEC logistics system is constructed and adjusted according to system construction requirements. Further, two key subsystems are designed: the logistics object distribution subsystem and risk detection subsystem, based on the deep learning backpropagation neural network (BPNN) algorithm. The relevant parameters of the object distribution subsystem are calculated and sent to the risk detection subsystem model and tested. It is concluded that the sorting completion rate before 18:00 can reach 95.2%, indicating that the proposed CBEC logistic system can meet the needs of CBEC logistics enterprises. Logistics risk detection’s expected and actual outputs can fit 99%, indicating a tiny deviation. The research has certain reference significance for clarifying the logistics system and service mode of CBEC.

Performance Comparison of PID and NARX Neural Network for Attitude Control of a Quadcopter UAV

ABSTRACT: In this study, two different types of controllers have been designed and tested for altitude and motion control of an autonomous quadrotor to compare the control performance under the influence of parametric uncertainty and disturbances. The first controller is a proportional-integral-derivative (PID) controller which is a conventional linear controller. The closed-loop PID algorithms calculate the results of the system by using the error values that consist of the difference between the sensor values measured by the closed-loop feedback method and the reference inputs. The second method that has been used is artificial neural network (ANN) algorithms, which provide both advantages and convenience in defining and controlling linear systems and non-linear systems with the closed-loop feedback method used in PID. The most important feature of the ANN algorithms is their high performance as a result of training with different input values. Therefore, the ANN control system has been trained with the input data used with Gaussian noise and the desired target data. A dynamic time series non-linear autoregressive with Exogenous input (NARX) neural network has been chosen as an ANN controller because of the time-delayed backpropagation learning performance. In this study, PID, and NARX NN control algorithms to control the maneuvers and altitude of the quadcopter and the mathematical model have been designed on Matlab Simulink. Motion control performances of the PID and NARX controllers are tested on the model. The design was tested on a real-time simulation environment with a one-millisecond fixed-step size. This paper proposes an alternative approach to control attitude and altitude on a quadcopter with the NARX NN algorithm.

A backpropagation learning algorithm with graph regularization for feedforward neural networks

The backpropagation (BP) neural network has been widely used in many fields. However, it is still a great challenge to design the architecture and obtain optimal parameters for BP neural networks. For improving the generalization performance, regularization is the most popular technique to train the BP neural networks. In this paper, we propose a novel BP algorithm with graph regularization (BPGR) to obtain optimal parameters, by imposing the graph regularization term to the error function. The essential idea is to force the latent features of hidden layer to be more concentrated, which enhances the generalization performance. Besides, the proposed modified graph regularization facilitates the calculation of gradient and is more capable to penalize the extreme values of weights. Furthermore, the graph regularization can also be integrated with deep neural networks to improve their generalization performance. In addition, we provide the convergence analysis of our method BPGR under some regularity conditions. By comparison on several datasets with five activation functions, experimental results validate the theoretical analysis and demonstrate outstanding performance of BPGR.

Double Moving Average Control Chart for Autocorrelated Data

The assumption of normality and independence is necessary for statistical inference of control charts. Misleading results are obtained if the traditional control chart technique is applied on the auto-correlated data. When data is correlated, a time series model is employed to produce an optimum output. The objective is to create a new control chart methodology that takes the autocorrelation of observations into account. Charts of Moving Average, Exponentially Weighted and Cumulative Sum better perform in the existence of autocorrelation data for small and moderate changes. The proposed methodology is highly skilled and receptive to minor, moderate and major changes in the process. Propsed DMA chart increases efficiency of average run length (ARL) chart for moving average (MA) to detect the small to medium magnitude shifts in the mean. The simulation also demonstrates that the DMA chart with spans of w=10 and 15 generally performs well in terms of average run length (ARL) as compared to clasical MA. This research may be extended to a multivariate autocorrelated statistical process control, but it can also be used to recognise and categorise seven categories of traditional control chart patterns, such as Downward, Upward Shift, Normal Trend, Cyclic, Systematic patterns, Increasing and Decresing Trend. In order to identify and categorize a set of subclasses of abnormal patterns, this model (multivariate autocorrelated statistical process control chart) should employ a multilayer feed forward Artificial Neural Network (ANN) architecture controlled by a back-propagation learning rule.

Landslide Susceptibility Evaluation Using Different Slope Units Based on BP Neural Network.

Landslides are one of the most widespread natural hazards that cause damage to both property and life every year. Therefore, the landslide susceptibility evaluation is necessary for land hazard assessment and mitigation of landslide-related losses. Selecting an appropriate mapping unit is an essential step for landslide susceptibility evaluation. This study tested the back propagation (BP) neural network technique to develop a landslide susceptibility map in Qingchuan County, Sichuan Province, China. It compared the results of applying six different slope unit scales for landslide susceptibility maps obtained using hydrological analysis. We prepared a dataset comprising 973 historical landslide locations and six conditioning factors (elevation, slope degree, aspect, lithology, distance to fault lines, and distance to drainage network) to construct a geospatial database and divided the data into the training and testing datasets. We based on the BP learning algorithm to generate landslide susceptibility maps using the training dataset. We divided Qingchuan County into six different scales of slope unit: 4,401, 13,146, 39,251, 46,504, 56,570, and 69,013, then calculated the receiver operating characteristic (ROC) curve, and used the area under the curve (AUC) for the quantitative evaluation of 6 different slope unit scales of landslide susceptibility maps using the testing dataset. The verification results indicated that the evaluation generated by 56,570 slope units had the highest accuracy with a ROC curve of 0.9424. Overelaborate and rough division of slope units may not get the best evaluation results, and it is necessary to obtain the slope units most consistent with the actual situation through debugging. The results of this study will be useful for the development of landslide hazard mitigation strategies.