MATLAB Neural Network Toolbox Research Articles

AN ANN MODEL FOR PREDICTING SLUMP OF CONCRETE CONTAINING CRUSHED GLASS AND NATURAL GRAVEL

This study modelled the slump of concrete containing crushed glass and Bida Natural Gravel (BNG) based on deep learning algorithm using the MATLAB neural network toolbox. A total of 240 (150mm × 150mm × 150mm) cubes were cast from 80 mixes generated randomly using Scheffe’s simplex lattice approach. Slump was measured for each of the experimental points of fresh concrete before filling in the moulds. The resulting batch for each mix was used as input data while the laboratory results for slump was used as output data for the ANN-model. Hence a shallow multilayer supervised Neural Network was developed to model these data. The developed model would be able to predict concrete slump containing 0% - 25% crushed glass as partial replacement for fine aggregate, water- cement ratio ranging from 0.45 – 0.65 and concrete grade M15 – M25. The architecture of the network contained 6 input parameters: water to cement ratio, water, cement, sand, crushed glass and BNG, 20 neurons in the hidden layer and slump in the outer layer. The adequacy of the developed model was measured using Mean Square Error (MSE) and Correlation Coefficient (R). Results showed that 6:20:1 model architecture for slump model had an MSE values for training, validation and testing as: 1.84e-2, 5.81e-3, 3.64e-3, 1.73e-3 respectively. Regression values for training, validation and testing are: 79%, 94%, 96% and 79%. The study concluded that a shallow multilayer Neural Network architecture with 20 neurons in the hidden layer is sufficient for predicting concrete slump.

Investigating Feed-Forward Back-Propagation Neural Network with Different Hyperparameters for Inverse Kinematics of a 2-DoF Robotic Manipulator: A Comparative Study

Inverse kinematics is a significant challenge in robotic manipulators, and finding practical solutions plays a crucial role in achieving precise control. This paper presents a study on solving inverse kinematics problems using the Feed-Forward Back-Propagation Neural Network (FFBP-NN) and examines its performance with different hyperparameters. By utilizing the FFBP-NN, our primary objective is to ascertain the joint angles required to attain precise Cartesian coordinates for the end-effector of the manipulator. To accomplish this, we first formed three input-output datasets (a fixed-step-size dataset, a random-step-size dataset, and a sinusoidal-signal-based dataset) of joint positions and their respective Cartesian coordinates using direct geometrical formulations of a two-degree-of-freedom (2-DoF) manipulator. Thereafter, we train the FFBP-NN with the generated datasets using the MATLAB Neural Network Toolbox and investigate its potential by altering the hyperparameters (e.g., number of hidden neurons, number of hidden layers, and training optimizer). Three different training optimizers are considered, namely the Levenberg-Marquardt (LM) algorithm, the Bayesian Regularization (BR) algorithm, and the Scaled Conjugate Gradient (SCG) algorithm. The Mean Squared Error is used as the main performance metric to evaluate the training accuracy of the FFBP-NN. The comparative outcomes offer valuable insights into the capabilities of various network architectures in addressing inverse kinematics challenges. Therefore, this study explores the application of the FFBP-NNs in tackling the inverse kinematics, and facilitating the choice of the most appropriate network design by achieving a portfolio of various experimental results by considering and varying different hyperparameters of the FFBP-NN.

Estimation of SARA composition of crudes purely from density and viscosity using machine learning based models

Accurate characterization of crude oils by determining the composition of saturates, aromatics, resins and asphaltenes (SARA) has always been a challenging task in the petroleum industry. However, conventional experimental methods for determination of SARA composition are labour intensive, time-consuming and expensive. In the present study, artificial neural network (ANN) models were developed to predict the SARA composition from easily measurable parameters like density and viscosity. A dataset of 216 crude oil samples covering wide range of geographical locations was compiled from various literature sources. The ANN models with one hidden layer and six neurons are trained, tested and validated using MATLAB neural network toolbox. Results obtained on analysis revealed reasonably good accuracy of prediction of SARA components except for aromatics. The performance of developed ANN models was compared with various correlations reported in literature and found to be better in terms of mean squared error and coefficient of determination. The developed models hence provide a cost-effective and time-efficient alternative to the conventional SARA characterization techniques.

Comparison of the Artificial Neural Network's Approximations for the Levenberg-Marquardt Algorithm and the Gradient Descent Optimization on Datasets

The approximations obtained by gradient descent optimization on a set of datasets were compared with the results obtained with the Levenberg-Marquardt Optimization Method (LMOM) on the same datasets. The datasets, which comprised three orthogonal databases obtained from MATLAB's Neural Network toolbox accompanying databases, were normalized and serially loaded to the artificial neural network Graphical User Interface (GUI) designed by the researchers. The GUI built with Visual Studio Programming Language (VSPL) implements a gradient descent optimization scheme of the back-propagation algorithm. The characteristics of each database for determination of the termination criteria were approximated from the developed feature extractive iteration algorithm. Revalidation sessions of the LMOM on the sampled datasets showed significant spuriousness in outputted results when compared with the gradient descent optimization results which although slow in attaining convergence produced results that can be closely replicated. Analysis of the F-statistics and the Receiver Operating Characteristics (ROC) for the sampled datasets results of both methods also showed that the gradient descent method demonstrated significant accuracy and parsimony in approximating the nonlinear solutions in the datasets when compared with the results from LMOM processing. Additionally, in this research, an algorithm for deducing and producing the ROC for analyzed Artificial Neural Network (ANN) sessions was also developed and implemented using VSPL.

Machine learning-driven approach for predicting the condensation heat transfer coefficient (HTC) in the presence of non-condensable gases

This research focuses on developing a predictive model that can effectively estimate the condensation heat transfer coefficient (HTC) on vertical tube surfaces during free-fall. The model takes into consideration the presence of non-condensable gases. The main objective is to assemble an extensive and varied database that encompasses different geometric parameters and operational conditions. The motivation here stems from the urgent need to develop practical models and/or correlations that can accurately predict the condensation heat transfer efficiency for industrial heat exchangers affected by non-condensable gases, with a particular focus on passive containment cooling systems (PCCS) employed in nuclear power plants (NPPs). This innovative cooling system utilizes the force of gravity to condense water steam, effectively removing heat from the containment vessel in the event of an accident. For the development of the model, we employed the neural network toolbox in MATLAB to construct an Artificial Neural Network (ANN) known as a Multi-Layer Perceptron (MLP) network. This model was designed to predict the HTC during the condensation process involving two different types of Non-Condensable Gases (NCG): Air and Nitrogen, along with Helium as a representative light gas. To build the model, we utilized a dataset comprising 1,888 data points sourced from 21 experimental references. The input layer of our model receives a range of parameters, encompassing total pressure (Ptot), subcooling temperature (ΔTsub), mass fraction of non-condensable gases (wnc), wall length (L), hydraulic diameter (Dh), the ratio of the Helium to non-condensable gases (RHe/nc), liquid density (ρl), gas density (ρg), liquid viscosity (μl), gas viscosity (μg), saturation temperature (Tsat), latent heat (hfg), liquid thermal conductivity (kl), and gas specific heat (cp,g). The output data of our model corresponds to the condensation HTC (h). To ensure consistent scaling, the input data was normalized by scaling each feature within the range of 0 to 1. On the other hand, the output data underwent a transformation using the natural logarithm. By utilizing Pearson correlation coefficient analysis and importance analysis for parameter selection, a new correlation for heat transfer was developed and evaluated. The developed machine learning model exhibited outstanding effectiveness in precisely forecasting the condensation HTC with an accuracy range of ± 10 % and ± 20 %. The outcome of this investigation represents the notable progress in analyzing extensive datasets gathered through experiments. Specifically, it addresses the previously unexplored influence of Helium gas by harnessing the capabilities of machine learning.

Characterization of various type of cognitive impairment using multiomics – a pilot imaging study from a local institution

AbstractBackgroundIn addition to clinical diagnosis, multi‐modal imaging has been employed for early detection of various types of dementia. Image features from various types of images provided information for accurate diagnosis. This study aimed to evaluate the role of multiple imaging, including MRI T1W, T2W, resting state functional MRI (rs‐fMRI) and 18‐F Flutemetamol PET image in diagnosis of various types of cognitive impairments.Methods56 participants (16 Alzheimer’s Disease (AD), 10 AD type Mild Cognitive impairment (A‐MCI), 18 vascular type MCI (V‐MCI), 12 Vascular dementia (VD) and 25 Healthy Control (HC) were recruited in an local memory clinic. All types of images were re‐orientated, segmentation and normalized to the whole brain mask template. 45 brain regional volumes (Vol) were calculated from T1w using Freesurfer. Brain regional white matter hyperintensities (WMH) were obtained from T2w images. Interhemispheric functional connectivity (IFC) of brain regions were measured from rs‐fMRI. Standard uptake value (SUV) were measured and calculated from the PET images. Also, patient demographics including age, sex and HK‐MoCa score (Demo) were recorded. 70% of each group of participants were assigned to the training group, whereas the rest 30% were assigned to test group. The neural network toolbox of MATLAB was employed to build four characterization models: Model 1 based Vol and WMH; Model 2 based on Vol, WMH and IFC; Model 3 based on Vol, WMH, IFC, and demo; Model 4 based on Vol, WMH, IFC, demo and SUV.ResultsThe receiver operating characteristics (ROC) curves showed that the Model 1, 2 and 3 achieved satisfactory accuracy, with 79%, 68% and 70% respectively. While Model 4 achieved excellent accuracy of 100%. The sensitivity and specificity of each model for each group was over 75% to 100%. The model with more image features attained higher accuracy in dementia diagnosis.ConclusionArtificial neural network model with multiomics can accurately differentiate AD, A‐MCI, V‐MCI, VD and HC. Further study is suggested with more participants for models verification.

Artificial neural network application to characterize patients with Alzheimer ’s disease and Mild Cognitive Impairment from Cognitive Normal.

AbstractBackgroundThis study aimed to build an artificial neural network model to differentiate patients with various diagnosis, including Alzheimer’s disease (AD), mild cognitive impairment (MCI) from cognitive normal (CN) using structural MRI images.MethodsMRI T1 image of 580 participants (97 AD, 291 MCI and 192 CN) were downloaded from ADNI, which is a publicly available database. The MPRAGE T1W MRI were pre‐processed by SPM software, including re‐orientation, segmentation and normalized to the whole brain mask template. 3D slicer software with pyradiomics extension were employed to extract the 107 radiomics features within the brain. Also, 45 brain regional volumes were segmented and calculated using Freesurfer. The AD, MCI and CN groups were divided into a training group and a validation group based on 70:30 ratio. Ten‐fold cross validation were performed to minimize the chance of overfitting. Three feed forward neural network model were built using the neural network toolbox of MATLAB, with one model using 107 radiomics alone (R model), one with 45 brain regional volumes (V Model) and the last one with both radiomics and volumes features (R+V model).ResultsThe receiver operating characteristics (ROC) curves showed that the V model and the R+V model achieved excellent accuracy, with 96.6% and 94.8% respectively. The area under the curve (AUC) was 99.6% and 97.9% respectively.ConclusionThe neural network model can accurately differentiate AD, MCI from HC. The application is practical and can be fully automatic, which can be used for dementia screening.

Prediction of deformation and damage of foundation support system based on BP neural network

The pit in the ground that needs to be excavated during the construction of a building is known as a foundation pit. In this paper, a MATLAB neural network toolbox is proposed to establish a three-layer BP neural network to predict building settlement. The relative error of the predicted values obtained after training is between -2% and 1%, and the maximum absolute error is 0.049 mm. this error is acceptable for rock-on-rock engineering problems. It shows that the building settlement prediction using artificial neural network method is feasible and the prediction results are credible. It automatically identifies the complex nonlinear relationships between the dependent and independent variables through self-learning of the optimal operation strategy, and considers the interactions among the factors in a comprehensive and integrated manner, which has good application prospects for solving nonlinear problems in the field of geotechnical engineering.

Performance Model Identification of the General Electric CF34-8C5B1 Turbofan Using Neural Networks

This paper presents a methodology developed at the Laboratory of Applied Research in Active Controls, Avionics, and Aeroservoelasticity to identify a performance model of the CF34-8C5B1 turbofan engine powering the CRJ-700 regional jet aircraft from simulated flight data using artificial neural networks (ANNs). For this purpose, a qualified virtual research simulator was used to conduct different types of flight tests and to collect engine data under a wide range of operating conditions. The collected data were then used to create a comprehensive database for the training of the ANN model. This process was performed using the Bayesian regularization algorithm available in the MATLAB Neural Networks Toolbox, followed by a study to identify the optimal network architecture, namely, the number of layers and the number of neurons. The validation of the methodology was accomplished by comparing the model predictions with a set of flight data collected with the flight simulator for different flight conditions and flight regimes including takeoff, climb, cruise, and descent. The results showed that the model was able to predict the engine performance in terms of fan speed, core speed, inlet turbine temperature, net thrust, and fuel flow with less than 5% relative error.

Empirical examples based on neuro-fuzzy systems

Neural networks are well suited for pattern recognition and classification. Matlab provides an extensive library and toolbox in neural networks. In the study, I designed a neural network using the Matlab Neural network toolbox. I carried out this process by changing various settings to get better performance and result. Recently, the processing and analysis of large volumes of data has become an urgent task. "Big Data" can be defined as "a large amount of unstructured data". This concept is the object of many scientific studies and developments in the field of information technology. The main characteristics of the concept "Big Data" are their unstructured nature.

Voting-Based Deep Convolutional Neural Networks (VB-DCNNs) for M-QAM and M-PSK Signals Classification

Automatic modulation classification (AMC) using convolutional neural networks (CNNs) is an active area of research that has the potential to improve the efficiency and reliability of wireless communication systems significantly. AMC is the approach used in a communication system to detect the type of modulation format at the receiver end. This paper proposes a voting-based deep convolutional neural network (VB-DCNN) for classifying M-QAM and M-PSK signals. M-QAM and M-PSK signal waveforms are generated and passed through the fading channel in the presence of additive white Gaussian noise (AWGN). The VB-DCNN extracts features from the input signal through convolutional layers, and classification is performed on these features. Multiple network instances are trained on different subsets of training data in the VB-DCNN. A network instance predicts the input signal during testing. Based on the votes, the final prediction is made. Different simulation experiments are carried out to analyze the performance of the trained network, and the DCNN is designed with the Deep Neural Network Toolbox in MATLAB. The generated frames are divided into training, validation, and test datasets. Lastly, the classification accuracy of the trained network is determined using test frames. The proposed model’s accuracy is near to 100% at lower SNRs. The simulation results show the superiority of the proposed VB-DCNN compared to existing state-of-the-art techniques.

Prediction of corrosion failure probability of buried oil and gas pipeline based on an RBF neural network

Risk assessment is critical to ensure the safe operation of oil and gas pipeline systems. The core content of such risk assessment is to determine the failure probability of the pipelines quantitatively and accurately. Hence, this study combines the MATLAB neural network toolbox and adopts an Radial Basis Functions (RBF) neural network with a strong non-linear mapping relationship to build a corrosion failure probability prediction model for buried oil and gas gathering and transmission pipelines. Based on the hazard identification of pipeline corrosion failure, the model summarizes the causes of corrosion failure and determines the input and output vectors of the neural network based on the fault tree. According to the selected learning samples, through the design and training of network parameters, the RBF neural network that can predict the system failure probability is finally obtained. Taking the failure probability of 30 groups of high-pressure gathering and transmission pipelines of gas storage as an example, the capability of inputting the probability of the bottom event and outputting the probability of the top event is demonstrated through training data. Our results show that the calculated failure probability based on the fault tree analysis model is consistent with the predicted failure probability based on the RBF neural network model. Hence, the RBF neural network model is shown to be reliable in predicting the corrosion failure probability of buried pipelines.

Modeling Lasolo Watershed Sedimentation and Mangrove Root Growth at the Lasolo Coast in North Konawe, Indonesia

This research sought to simultaneously measure sedimentation and mangrove root growth rates at the Lasolo River estuary and construct a predictive sedimentation model from the measured rates. The applied methods were descriptive quantitative analyses using cloud-based computation to obtain sea surface temperatures, Landsat 8 and 9 OLI images to identify land cover change, field-derived data, and rain data. The neural network toolbox in MATLAB was used to perform the modeling. Results showed that sediments accumulated at the highest rate of 6.10 cm, and mangrove roots grew at an average of less than 1 cm per two weeks, meaning that the former always surpasses the latter. The forecasting model’s training and validation each produced a regression value (R) of 0.92 and 0.69 for sedimentation rates and R=0.83 and R=0.88 for root growth rates. Sedimentation rates continued to vary until the end of the observation, with a maximum of 3.91 cm. Although the root growth rates had a similar pattern (fluctuating), an upward trend with signs of acceleration to 0.62 cm was observed. This study also detected a conversion of vegetated land into open mining areas, which expanded, tentatively, from 645.8 ha in 2013 to 2,112.6 ha in 2022.

Evaluation of best value of wind speed for maximum wind energy output in Nigeria using neural network

The outcome of the study, evaluation of best value of wind speed for maximum wind energy output in Nigeria using neural network was successfully achieved. Wind speed for various months and elevation data gotten from NASA were prepared in excel and imported into neural network toolbox in MATLAB for training and analysis using levenberg maquardt and scaled conjugate gradient algorithms to determine the best performance value. Furthermore, training of the data using levenberg marquardt algorithm at 70% training data, 15% test data and 15% validation data respectively revealed that the best performance level was 1.9046 m/s at 7 epochs and this matches elevation of 7. 24meters. In addition, training of the same data using scaled conjugate gradient algorithm with the stated conditions above, revealed that the best performance level was 9.0376 m/s at 182 epochs and this matches elevation of 34.35meters. Also, training fit coefficients were found to be 0.99984 and 0.99804 respectively which indicated that there is a close and positive relationship between wind speed and wind turbine hub height. Results revealed that the best value of wind speed and elevation for maximum wind energy output in Nigeria ranges from 1.9046 m/s to 9.0376 m/s and 7.2 meters to 34.35 meters respectively. The researchers made the following recommendations: Wind turbine tower and blades should be designed to have wind impact speed capacity beyond the estimated values; further research can also be done in future to establish a more accurate mathematical model between wind speed and wind energy output with other advanced program for generalization, etc.

Global Solar Radiation Modelling using an Artificial Neural Network for Kazaure, Jigawa State, Nigeria

This research presents an algorithm based on Artificial Neural Networks (ANN), for estimating monthly mean daily and hourly values of solar global radiation. To effectively investigate solar energy consumption and estimate solar renewable energy resources, the Hourly Global Solar Radiation measurements are necessary. In order to predict monthly average daily global sun irradiance on a horizontal area of Kazaure- Nigeria, this study creates a model utilizing ANN to solve the problem of solar energy distribution. Five empirical correlations are developed using the data from 42 months to aid in the prediction of the solar energy distribution pattern. The software is constructed around the Multilayer Perceptron under categorized tabs, with Multilayer perception in neural network Toolbox in MATLAB 9.7 version as a feed forward ANN that maps sets of input data into a set of suitable output. It differs from conventional linear perception by employing three or more layers of neurons (nodes) with nonlinear activation functions. It is also more effective than perceptrons in identifying input that is not linearly separable by a linear hyper-plane. Results obtained utilizing the suggested structure reveals good agreement between the calculated and measured levels of global solar irradiation. The ANN model is shown to be superior when compared to empirical models, due to negligible noise margin.

Early warning of enterprise financial risk based on improved BP neural network model in low-carbon economy

The concept of low-carbon economic development has led to changes in the business environment and financial environment of enterprises, leading to increased financial risks faced by enterprises. How to help enterprises better warn, prevent and control financial risks from the perspective of low-carbon economy has become a hot issue worth studying. Based on this, this paper is based on the perspective of low carbon economy, on the basis of analyzing the financing risk, investment risk, capital operation risk and growth risk faced by enterprises under the requirements of low carbon economy development. A set of financial risk management framework with clear hierarchy and strict vertical logic has been constructed. Ten financial early-warning indicators are constructed from four aspects. The risk prediction model of the indicator system is established using the research method of BPNN (Back Propagation Neural Network). The model is trained and simulated through the MATLAB neural network toolbox. After 10 indicators passed Bartlett’s correlation test, the BPNN financial early warning model was programmed using MATLAB software. The accuracy rate was 84.3%. The neural network training results show that when the layer node is 8, the best correct recognition rate can be obtained. Incorporate “low carbon” into the financial risk early warning indicator system that meets the requirements of low carbon economic development in the design of enterprise financial risk early warning indicators. This paper is expected to provide reference and reference for low-carbon economy enterprises to deal with financial risks under the new situation.

Solar assisted air source heat pump systems for campus water heating in China: Economic optimization of solar fraction design

• TRNSYS software is to simulate the solar assisted air source heat pump system . • The solar fraction design values of solar water heating systems are not economically optimal in China. • Long-term economic optimization of solar fraction design values for solar assisted air source heat pump systems is performed. • An ANN-generated proxy model is to predict the most economical solar fraction design values of solar assisted air source heat pump systems. The solar assisted air source heat pump (SAASHP) system is widely used for campus water heating owing to its cost-effectiveness and cleanliness. In the current design process of the SAASHP system, the solar collector area is determined by the solar fraction ( f ). However, it is difficult to design a cost-optimal f that can comprehensively take the atmospheric factors, device model, electricity price and initial investment into consideration. To address this issue, a SAASHP system that meets the bathing hot water demand of 480 students was established by TRNSYS software, and the cost-optimal solar collector areas and f values for 12 cities across 4 different solar resource regions in China were obtained by the TRNOPT tool of TRNSYS. Moreover, the effects of SC price, electricity price, bathing hot water demand and rated COP of ASHP on cost-optimal f values were investigated by sensitivity analysis. Finally, a proxy model was generated by the artificial neural network toolbox of MATLAB to predict the cost-optimal f values for different scenarios. The results showed that compared to the economically optimized SAASHP system, the designed SAASHP system according to the national design standards for solar heating systems in China has a larger SC area value, leading to an increase in the annual life cycle cost of the system. The difference between the optimized and designed values of solar collector area and f for the SAASHP system was minor in rich resource and sub-rich resource regions but was larger in general resource and poor resource regions. The solar collector price, electricity price, bathing hot water demand and the rated COP of the SAHP were significant factors affecting the annual life cycle cost, the cost-optimal f and solar collector area of the SAASHP system. The proxy model generated by the artificial neural network toolbox of MATLAB can accurately predict the economically optimal solar collector area and f values for SAASHP systems in different cities and system factors, providing a method for the economic design of SAASHP systems in China.

Innovation of Platform Economy Business Model Driven by BP Neural Network and Artificial Intelligence Technology.

In order to enhance the competitiveness of enterprises, how to evaluate and enhance the competitiveness of B2B e-commerce enterprises and promote the orderly and healthy development of B2B e-commerce industry are discussed. This paper puts forward the research on the innovation of platform economic business model driven by BP neural network and artificial intelligence technology. BP neural network is used to study and evaluate the competitiveness of B2B e-commerce companies. According to the B2B e-commerce company competitiveness theory and BP neural network algorithm, combined with BP neural network and B2B e-commerce company competitiveness evaluation index system, a BP neural network model is designed to analyze the competitiveness of B2B e-commerce enterprises. Determine the expected value of network samples, select G1 method to determine the subjective weight, and select entropy weight method to determine the objective weight. With the help of the function in the MATLAB neural network toolbox, the neural network is trained. The results show that when the training times reach 3297 times, the sample mean square error is 9.9869e − 06, and the training network reaches convergence. The samples of three enterprises test the trained neural network and input the data of three test samples into the trained BP neural network, and the output results are 0.1531, 0.1371, and 0.1557, respectively. The network model constructed in this paper is effectively close to the training samples. The established BP neural network has good performance and can be used to evaluate the competitiveness of B2B e-commerce companies. Accelerate technological change and realize innovation. Technological capability is the inexhaustible driving force for the development of enterprises. Only with the innovation of keeping pace with the times can application-oriented e-commerce enterprises meet the needs of customers and the market, form the difference between goods or services, and then enable enterprises to win more customers and market share.

Application of Temperature Compensation Combined with Neural Network in Infrared Gas Sensor

Due to the state of the gas to be measured, the detection mechanism of the pyroelectric detector and the temperature drift of the peripheral circuit components and the detection of the ambient temperature will interfere with the measurement accuracy of the nondispersive infrared gas sensor from many aspects. This paper proposes a temperature compensation method based on the BP neural network. The compensation function of the gas sensor is realized by programming the various functional parameters in the neural network through the program provided in the Matlab neural network toolbox. Experimental simulation results show that the proposed method effectively reduces the influence of external temperature on the gas sensor output and improves its accuracy and stability.

Prediction of Purchase Volume of Cross-Border e-Commerce Platform Based on BP Neural Network

As a new form of foreign trade, cross-border e-commerce has huge development potential. Although the development prospect of cross-border e-commerce is good, the management of global supply chain is very important in order to gain a place in the fierce competition and develop steadily. The traditional forecasting of purchasing volume adopts time series, and the forecasting model is relatively simple. The purchase volume of the platform is related to the various consumption behaviors of consumers, such as the number of product reviews, the number of product collections, and whether there are tax subsidies. The sales volume in the next few days is predicted by the item number, time, and sales quantity. The four-layer BP neural network model is used, and the MATLAB neural network toolbox is used to draw the training error curve and the correlation coefficient curve. After network training, the training correlation coefficient R reaches 95.823%, and the prediction accuracy obtained at this time is higher. Further, using the established model based on BP algorithm, the traditional BP algorithm is optimized to obtain the purchase quantity of commodities. The method is applied to the forecast of commodity purchase volume of a cross-border e-commerce platform, and the results show that the average error rate of this method is 5.9%, which has high practical application value. The research results show that this paper considers multiple influencing factors and selects an appropriate forecasting method, which can effectively improve the accuracy of the company's commodity sales forecast, so as to better formulate procurement plans and optimize inventory structure, which has certain implications for the actual operation of cross-border e-commerce platforms.