Nonlinear Neural Network Research Articles

Heat Transfer Model of Natural Gas Pipeline Based on Data Feature Extraction and First Principle Models

The rapid development of natural gas pipelines has highlighted the need to utilize SCADA (supervisory control and data acquisition) system data. In this paper, a heat transfer model of a natural gas pipeline based on data feature extraction and first principle models, which makes full use of the measured temperatures at each end of the pipeline, is proposed. Three methods, the NARX neural network (nonlinear autoregressive neural network with exogenous inputs), time series decomposition, and system identification, were used to model the changes of gas temperatures of the pipeline. The NARX neural network method uses a cyclic neural network to directly model the relationship of temperature between the start and the end of the pipeline. The measured temperature series at the pipeline inlet and outlet were decomposed into trend items, fluctuation items, and noise items based on the time series decomposition method. Then the three items were fitted separately and combined to form a new temperature prediction series. The system identification method constructed the first-order and second-order transfer function to model the temperature. The simulation of the three data-driven models was compared with those of the physics-based simulation models. The results showed that the data-driven model has great advantages over the physics-based simulation models in both accuracy and efficiency. The proposed models are more suitable for applications such as online simulation and state observation of long-distance natural gas pipelines.

Variable-delay feedback control for stabilization of highly nonlinear hybrid stochastic neural networks with time-varying delays

A new highly nonlinear hybrid stochastic neural network (SNN) with variable time delays is formulated, which is usually found to be unstable. In addition, the network coefficients considered in this article, grow polynomially rather than linearly as we always considered before. Since original system is unstable, the purpose of this article is to construct a variable-delay feedback control function to make the augmented system stable. Then, four results on the stabilization of the controlled system are given separately. At last, an example is provided to display the feasibility of theoretical results.

A Multi-Rate Approach for Nonlinear Pre-Distortion Using End-to-End Deep Learning in IM-DD Systems

Modern intra-data center (IDC) interconnects leverage robust and low-cost intensity modulation (IM) and direct detection (DD) optical links, based on multimode fibers (MMFs) and vertical-cavity surface-emitting lasers (VCSELs). Current solutions, based on on-off keying (OOK) modulations, reach up to 25-50 Gbps per lane over nearly 100 meters. The actual target for IDCs is to increase VCSEL-MMF links capacity up to 100 Gbps, using PAM-4 on the same devices. To counteract the consequent linear and nonlinear distortions affecting the transmitted signals, an effective solution is to exploit digital signal processing (DSP). In this manuscript, we propose a novel method to optimize a nonlinear artificial neural network (ANN) digital pre-distorter (DPD), based on End-to-end (E2E) learning, that, trained jointly with a Feed-Forward Equalizer (FFE), fulfills physical amplitude constraints and handles different ratio between the sampling rates incurring along with an optical IM-DD system. We indeed propose an E2E ANN system operating simultaneously at different sampling frequencies. We moreover propose in our training method a substitution to the time-domain injection of the receiver noise in the system with an additive regularization term in the FFE gradient loss. We experimentally show the advantages of our proposed DPD comparing the bit error rate (BER) performance against the same scenario without DPD. We assess the gain in terms of Gross Bit Rate and Optical Path Loss (OPL), at given BER targets, for different fiber lengths.

An Economic Analysis of Energy Consumption at Student Residences in a South African-Based Academic Institution Using NARX Neural Network

One of the issues associated with the supply of electricity is its generation capacity, and this has led to prevalent power cuts and high costs of usage experienced in many developing nations, including South Africa. Historical research has shown that the annual rate of increase for electricity has grown at an alarming rate since 2008 and, in some years, has grown as much as 16%. The objectives of this study are to estimate the cost analysis of electricity usage at the twenty-nine residences of the University of Johannesburg (UJ-Res) and propose a model for our university, as well as other South African universities, to become more energy-efficient. This was achieved by analyzing the tariffs between 2015 and 2021. A forecast was made for a period of five years (2021 to 2026) using a non-linear autoregressive exogenous neural network (NARX-NN) time-series model. From the results obtained, the better NARX-NN model studied has a root mean squared error (RMSE) of 2.47 × 105 and a determination coefficient (R2) of 0.9661. The projection result also shows that the annual cost of energy consumed will increase for the projected years, with the year 2022 being the peak with an estimated annual cost of over ZAR 30 million (USD 2,076,268).

1.2 nJ/classification 2.4 mm2 asynchronous wired-logic DNN processor using synthesized nonlinear function blocks in 0.18 μm CMOS

We have developed a full-digital wired-logic DNN processor that is 5.3 times smaller and 2.6 times more energy efficient than previously developed processors. Our processor is capable of inferring a MNIST classification task with 90.6% accuracy and 1.2 nJ of energy per classification at 3.89 Mfps. We also developed a neuron and synapse-saving neural network using nonlinear neural network technology to reduce the number of processing elements to be implemented. Lastly, we developed a logical compression technique for area and energy-saving neuron cell circuits. Using these techniques, we devised a digital asynchronous wired-logic DNN processor.

Improving the MODIS leaf area index product for a cropland with the nonlinear autoregressive neural network with eXogenous input model

The leaf area index (LAI) is a crucial descriptive parameter of the dynamic change of ground vegetation. The widely used MODIS LAI product, however, does not satisfy the requirements of regional eco-environment modeling. There is an urgent need to improve the product’s overall accuracy. Under this circumstance, this study proposed an improvement scheme based on the nonlinear autoregressive neural network with eXogenous input (NARXNN) model and the high-quality time series LAI inversion result. Case studies were implemented for two seasons a year croplands in Wuzhi, Xinzheng, and Xiangcheng in Henan province. This research acquired 46 periods of the NARXNN model-improved LAI, which went through rigid in situ LAI validation. The in situ measured LAI by LAI-2000 was used to validate the accuracy of NARXNN-enhanced LAI data. The R2 values of the improved LAI of the three research areas are 0.54, 0.41, and 0.51, while the RMSE decreased by 0.07, 0.1, and 0.03, and the bias also decreased to a certain extent. Direct validation using the in situ measured LAI demonstrates that the NARXNN model-enhanced LAI data were more accurate and had a lower bias than MCD15A2H. A comparison of the time series change indicates that the NARXNN-enhanced LAI shows a smoother bimodal change trend and is more conformed to the actual cropland growth than the original MODIS product. The results indicated that the NARXNN neural network further increased the accuracy of the MODIS product and has a particular practical value in future research.

Shaping energy cost management in process industries through clustering and soft sensors

With the ever-increasing growth of energy demand and costs, process monitoring of operational costs is of great importance for process industries. In this light, both financial budget management and local operational optimization supposed to be guaranteed properly. To achieve this goal, a support vector machine recursive feature elimination (SVM-RFE) method together with clustering algorithm was developed to extract features while serving as importance measurements of each input variable for the sequential prediction model construction. Then, the four variants of autoregressive and moving average (ARMA), i.e., ARMA with exogenous input (ARMAX) based on recursive least squares algorithm (RLS), ARMAX based on recursive extended least squares algorithm (RELS), nonlinear auto-regressive neural network (NARNN) and nonlinear auto-regressive neural network with exogenous input (NARXNN), were applied, respectively, to predict the costs incurred in the daily production for process industries. The methods were validated in the Benchmark Simulation Model No.2-P (BSM2-P) and a practical data set about steel industry energy consumption from an open access database (University of California, Irvine (UCI)), respectively. The nonlinear model, NARXNN, was validated to achieve better performance in terms of mean square error (MSE) and correlation coefficient (R), when used for multi-step prediction of the aforementioned datasets with strong nonlinear and coupled characteristics.

Architecture Optimization of a Non-Linear Autoregressive Neural Networks for Mackey-Glass Time Series Prediction Using Discrete Mycorrhiza Optimization Algorithm.

Recurrent Neural Networks (RNN) are basically used for applications with time series and sequential data and are currently being used in embedded devices. However, one of their drawbacks is that RNNs have a high computational cost and require the use of a significant amount of memory space. Therefore, computer equipment with a large processing capacity and memory is required. In this article, we experiment with Nonlinear Autoregressive Neural Networks (NARNN), which are a type of RNN, and we use the Discrete Mycorrhizal Optimization Algorithm (DMOA) in the optimization of the NARNN architecture. We used the Mackey-Glass chaotic time series (MG) to test the proposed approach, and very good results were obtained. In addition, some comparisons were made with other methods that used the MG and other types of Neural Networks such as Backpropagation and ANFIS, also obtaining good results. The proposed algorithm can be applied to robots, microsystems, sensors, devices, MEMS, microfluidics, piezoelectricity, motors, biosensors, 3D printing, etc.

Adaptive Hyperparameter Fine-Tuning for Boosting the Robustness and Quality of the Particle Swarm Optimization Algorithm for Non-Linear RBF Neural Network Modelling and Its Applications

A method is proposed for recognizing and predicting non-linear systems employing a radial basis function neural network (RBFNN) and robust hybrid particle swarm optimization (HPSO) approach. A PSO is coupled with a spiral-shaped mechanism (HPSO-SSM) to optimize the PSO performance by mitigating its constraints, such as sluggish convergence and the local minimum dilemma. Three advancements are incorporated into the hypothesized HPSO-SSM algorithms to achieve remarkable results. First, the diversity of the search process is promoted to update the inertial weight ω based on the logistic map sequence. Then, two distinct parameters are trained in the original position update algorithm to enhance the work efficiency of the successive generation. Finally, the proposed approach employs a spiral-shaped mechanism as a local search operator inside the optimum solution space. Moreover, the HPSO-SSM method concurrently improves the RBFNN parameters and network size, building a model with a compact configuration and higher precision. Two non-linear benchmark functions and the total phosphorus (TP) modelling issue in a waste water treatment process (WWTP) are utilized to assess the overall efficacy of the creative technique. The results of testing indicate that the projected HPSO-SSM-RBFNN algorithm performed very effectively.

Garden Landscape Construction Based on Landscape Urbanism and Dynamic Nonlinear System Thinking

Abstract This paper uses nonlinear dynamic system thinking to evaluate the ecological environment of landscaping. In this paper, the mathematical models of landscape dimension, fragmentation index, diversity index, and aggregation index are constructed using an artificial neural network combined with a fuzzy C-means (FCM) clustering nonlinear thinking system. In the end, this paper plans and evaluates the ecological environment of S city. Through the analysis of the model, it is verified that the nonlinear artificial neural network system combined with fuzzy C-means (FCM) clustering proposed in this paper is an effective method for forest landscape planning and evaluation. The scope and feasibility of this method are high.

Nonlinear autoregressive neural network with exogenous input for an energy efficient non-cooperative target tracking in wireless sensor network

Nonlinear autoregressive neural network with exogenous input for an energy efficient non-cooperative target tracking in wireless sensor network

Prediction of Temperature Behavior in Hydraulic Circuits

The increasing complexity of products and systems in modern engineering production requires higher demands on the operation and maintenance of systems. For this reason, the role of maintenance can hardly be underestimated, as the maintenance of machinery and equipment is an integral part of any manufacturing process. Maintenance is closely related to the precision of the parts produced and the quality of the manufacturing process. A current trend in Industry 4.0 is predictive maintenance, which enables real-time detection of faulty equipment conditions based on condition monitoring. Utilizing artificial intelligence for predictive analytics can considerably contribute to the prediction of potential failures. Consequently, maintenance costs related to production downtime and equipment repair will be decreased. In this paper, a non-linear autoregressive neural network for real-time prediction of working fluid temperature parameters in a hydraulic circuit is presented. A testing device is also developed to collect experimental data, optimize, and test the proposed prediction model.

The Characteristics and Monitoring System of Athletes’ Physical Strength Training Based on Tunable Neuron Mathematical Model

Abstract In this paper, an adjustable neural network model is developed to predict the physical training load of athletes. This paper takes 48 contestants as experimental subjects. Then the average speed prediction method, nonlinear regression model prediction method and neural network model prediction method model are established to predict the training load. This paper uses three different prediction models to compare the training results. The relative error rate calculated by the average rate prediction method is about 23%. The close error rate calculated by the linear regression method is about 32%. The relative error rate calculated by the adjustable neural network is around 8%. The e flexible neural network model has good prediction accuracy.

Nonlinear neural networks adaptive control for a class of fractional-order tuberculosis model.

The problem of nonlinear adaptive control for a class of fractional-order tuberculosis (TB) model is studied in this paper. By analyzing the transmission mechanism of TB and the characteristics of fractional calculus, a fractional-order TB dynamical model is established with media coverage and treatment as control variables. With the help of universal approximation principle of radial basis function neural networks and the positive invariant set of established TB model, the expressions of control variables are designed and the stability of error model is analyzed. Thus, the adaptive control method can guarantee that the number of susceptible and infected individuals can be kept close to the corresponding control targets. Finally, the designed control variables are illustrated by numerical examples. The results indicate that the proposed adaptive controllers can effectively control the established TB model and ensure the stability of controlled model, and two control measures can protect more people from tuberculosis infection.

Global exponential periodicity of nonlinear neural networks with multiple time-varying delays

<abstract><p>Global exponential periodicity of nonlinear neural networks with multiple time-varying delays is investigated. Such neural networks cannot be written in the vector-matrix form because of the existence of the multiple delays. It is noted that although the neural network with multiple time-varying delays has been investigated by Lyapunov-Krasovskii functional method in the literature, the sufficient conditions in the linear matrix inequality form have not been obtained. Two sets of sufficient conditions in the linear matrix inequality form are established by Lyapunov-Krasovskii functional and linear matrix inequality to ensure that two arbitrary solutions of the neural network with multiple delays attract each other exponentially. This is a key prerequisite to prove the existence, uniqueness, and global exponential stability of periodic solutions. Some examples are provided to demonstrate the effectiveness of the established results. We compare the established theoretical results with the previous results and show that the previous results are not applicable to the systems in these examples.</p></abstract>

Remote Sensing and Nonlinear Auto-regressive Neural Network (NARNET) Based Surface Water Chemical Quality Study: A Spatio-Temporal Hybrid Novel Technique (STHNT).

In recent days, the quality of water in inland water bodies has been threatened by various natural and anthropogenic activities. Henceforth, the continuous monitoring of water quality is mandatory to control the pollution level in surface water bodies. In this work, remote sensing technology integrated with an Artificial Intelligence (AI) algorithm, a new technique called Spatio-Temporal Hybrid Novel Technique (STHNT), was used to predict, and monitor the chemical water quality pollution level through the Water Quality Index (WQI). The Two Bands Regression Empirical (TBRE) water quality model has been used to retrieve water quality parameters from multi-resolution satellite imagery (Sentinel-2 MSI). The Nonlinear Auto-regressive Neural Network (NARNET), which is an Artificial Neural Network (ANN), was set up to predict the water quality index. Based on the model performed on the remote sensing water quality data, it is inferred that NARNET (Coefficient of determination-R2:0.9911, Root Mean Square Error-RMSE:1.693 and Sum of Squares of Error-SSE:14.33) provides significant results in predicting WQI. Therefore, the combined remote sensing technology with artificial intelligence plays a pivotal role in water resource management, which helps in reducing the pollution level in surface water bodies.

An Index for Rail Weld Health Assessment in Urban Metro Using In-Service Train

Rail welds are considered as the weak part of a railway track. Their defects and health can directly affect wheel-rail interaction, track safety, and reliability. Current practices for rail welds health assessment are based on 2D vertical and lateral wear measurement which needs time and track blocking. The development of inertia-based condition monitoring methods such as measuring axle box acceleration (ABA) comes with a crucial question on criteria or index for each rail track component health monitoring. In this study, an index for evaluation of rail weld health is proposed through integrated numerical and field experiment data within a metro line using the ABA technique. The relationship between the speed, wheel structural vibration, and acceleration amplitude is investigated using fast Fourier transformation (FFT) and a nonlinear neural network principal component analysis (PCA) model. An index is introduced to assess weld severity level based on the statistical method. This index is simple and applicable for maintenance practice.

Soil-Moisture Estimation Based on Multiple-Source Remote-Sensing Images

Soil moisture plays a significant role in the global hydrological cycle, which is an important component of soil parameterization. Remote sensing is one of the most important methods used to estimate soil moisture. In this study, we developed a new nonlinear Erf-BP neural network method to establish a soil-moisture-content-estimation model with integrated multiple-resource remote-sensing data from high-resolution, hyperspectral and microwave sensors. Next, we compared the result with the single-resource remote-sensing data for SMC (soil-moisture content) estimation models by using the linear-fitting method. The results showed that the soil-moisture estimation model offers better accuracy by using multiple-resource remote-sensing data. Furthermore, the SMC predicted the results by using the new Erf-BP neural network with multiple-resource remote-sensing data and a good overall correlation coefficient of 0.6838. Compared with the linear model’s estimation results, the accuracy of the SMC estimation using the Erf-BP method was increased, and the RMSE decreased from 0.017 g/g to 0.0146 g/g, a decrease of 16.44%. These results also indicate that the improved algorithm of the Erf-BP artificial neural network has better fitting results and precision. This research provides a reference for multiple-resource remote-sensing data for soil-moisture estimation.

Forecasting the concentration of NO2 using statistical and machine learning methods: A case study in the UAE

Nitrogen dioxide (NO2) is the most active pollutant gas emitted in the industrial era and is highly correlated with human activities. Tracking NO2 emissions and predicting their concentrations represent important steps toward controlling pollution and setting rules to protect people's health indoors, such as in factories, and in outdoor environments. The concentration of NO2 was affected by the COVID-19 lockdown period and decreased because of restrictions on outdoor activities. In this study, the concentration of NO2 was predicted at 14 ground stations in the United Arab Emirates (UAE) during December 2020 based on training over a full time period of two years (2019–2020). Statistical and machine learning models, such as autoregressive integrated moving average (ARIMA), seasonal autoregressive integrated moving average (SARIMA), long short-term memory (LSTM), and nonlinear autoregressive neural network (NAR-NN), are used with both open- and closed-loop architectures. The mean absolute percentage error (MAPE) was used to evaluate the performance of the models, and the results ranged from “very good” (MAPE of 8.64% at the Liwa station with the closed loop) to “acceptable” (MAPE of 42.45% at the Khadejah School station with the open loop). The results show that the predictions based on the open loop are generally better than those based on the closed loop because they yield statistically significantly lower MAPE values. For both loop types, we selected stations exhibiting the lowest, medium, and highest MAPE values as representative cases. In addition, we demonstrated that the MAPE value is highly correlated with the relative standard deviation of NO2 concentration values.

Corrosion Inhibition Studies of 8-Hydroxyquinoline Derivatives for N80 Steel in a 1.0 M HCl Solution: Experimental, Computational Chemistry, and Quantitative Structure-Activity Relationship Studies.

Twelve kinds of 8-hydroxyquinoline derivatives were synthesized and characterized. The weight loss method was used to evaluate their inhibition efficiencies (IEs) in a 1.0 M HCl solution at 333 K. The results showed that the alkyl chain length, heteroatoms (S, N, and O), and number of benzene rings significantly affect the IE. Herein, the IE of 5-[(dodecylthio)methyl]-8-quinolinol reached 98.71%. Meanwhile, the potentiodynamic polarization results indicated that all 8-hydroxyquinoline derivatives were mixed-type inhibitors. Electrochemical impedance spectroscopy results revealed that 8-hydroxyquinoline derivatives can increase polarization resistance, supporting their adsorption on the N80 steel surface. Moreover, according to density functional theory (DFT), the frontier orbital distribution and quantum chemical parameters (EHOMO, ELUMO, dipole moment μ, etc.) were calculated, and the results confirmed that the substituents of protonated 8-hydroxyquinoline derivatives significantly influenced the frontier orbital distribution. Molecular dynamics simulation illustrated that all protonated 8-hydroxyquinoline derivatives were adsorbed parallel to the Fe(110) surface, and the interaction energy (Eint) evidenced that the molecular size would affect their strength of adsorption on the Fe(110) surface. The linear and nonlinear quantitative structure-activity relationship models were established by linear regression (LR) methods and BP neural networks (NN), respectively. The LR model was established by using Eint and μ, and the coefficient of determination (R2) was 0.934. In addition, the nonlinear NN model was obtained according to IE and all parameters (DFT parameters and Eint). Then, the two calculation inhibition efficiencies (IEcal) were obtained from the LR and NN models, and the R2 values of the linear correlation between the IEcal and the experimental IE were 0.940 and 0.951, respectively. In addition, the IE of the tested inhibitor was 51.86% and the IEcal values predicted by the LR and NN models were 52.68% and 53.06%, respectively. Our results demonstrate that both the LR and NN models have good fits and predictive ability.