NN Model Research Articles

Real-Time Machine Learning for Multi-User Massive MIMO: Symbol Detection Using Multi-Mode StructNet

In this paper, we develop a learning-based symbol detection algorithm for massive MIMO-OFDM systems. To exploit the structure information inherited in the received signals from massive antenna array, multi-mode reservoir computing is adopted as the building block to facilitate over-the-air training in time domain. In addition, alternating recursive least square optimization method, and decision feedback mechanism are utilized in our algorithm to achieve the real-time learning capability. That is, the neural network is trained purely online with its weights updated on an OFDM symbol basis to promptly and adaptively track the dynamic environment. Furthermore, an online learning-based module is devised to compensate the nonlinear distortion caused by RF circuit components. On top of that, a learning-efficient classifier named StructNet is introduced in frequency domain to further improve the symbol detection performance by utilizing the QAM constellation structural pattern. Evaluation results demonstrate that our algorithm achieves substantial gain over traditional model-based approach and state-of-the-art learning-based techniques under dynamic channel environment and RF circuit nonlinear distortion. Moreover, empirical result reveals our NN model is robust to training label error, which benefits the decision feedback mechanism.

Investigating the Impacts of Tropospheric Parameters on Received Signal Strength of the Mobile Communication System

Abstract Radio communication systems are crucial for information transmission, but they face challenges like interference and fading, which significantly impact network efficiency. Understanding propagation-related issues is crucial for optimizing networks. This study examined the influence of tropospheric parameters on the Received Signal Strength (RSS) of the Globacom communication system in Rivers State. Four tropospheric variables (air temperature, relative humidity, atmospheric pressure, and wind speed) are collected from the River State University of Science and Technology (RSUST) Environmental and Climatic Observatory Program. The results of the correlation analysis indicated a direct linear correlation between air temperature and RSS. In contrast, wind speed displayed an indirect relationship, while the remaining parameters demonstrated an inverse linear relationship. Multiple linear regression (MLR) and multiple linear perceptron (MLP) models were developed to accurately predict the impacts of tropospheric parameters on RSS. A supervised-learning three-layered MLP feed-forward neural network was employed, consisting of 10 distinct MLP NN models. The Levenberg-Marquardt (LM) learning algorithm was used in training and validating the MLP NN models. The MLP 4-40-1 model showed an exceptional understanding of the non-linear relationship between the tropospheric parameters and RSS. The MLR model exhibited a weaker correlation coefficient of 0.2164, while the MLP demonstrated a stronger correlation coefficient of 0.7942. Additionally, the MLP 4-40-1 model outperformed the MLR model in terms of Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), achieving an accuracy of 0.2480 RMSE and 12.9963% MAPE, while the MLR model had 1.6880 RMSE and 27.3787% MAPE. These findings highlight the efficiency of the MLP 4-40-1 model in estimating the exact relationship between the tropospheric parameters and the signal strength of the mobile network in Rivers State. The study provides valuable insights into optimizing network design and emphasizes the significance of including tropospheric parameters in enhancing mobile communication performance.

Gibbon: An Efficient Co-Exploration Framework of NN Model and Processing-In-Memory Architecture

The memristor-based Processing-In-Memory (PIM) architectures have been proven to be a potential architecture to store enormous parameters and execute the complicated computations of Deep Neural Networks (DNNs) efficiently. Existing PIM studies focus on designing high energy-efficient hardware architecture and algorithm-hardware co-optimization for better performance. However, the impacts of the algorithms and hardware architectures on the performance intersect with each other. Only optimizing the algorithms or the hardware architectures can not realize the optimal design. Therefore, the co-exploration of NN models and PIM architecture is necessary. However, for one thing, the co-exploration space size of NN models and PIM architectures is extremely huge, and is challenging to search. For another, during the co-exploration process, time-consuming PIM simulators are needed to evaluate various design candidates and pose a heavy time burden. To tackle these problems, we propose an efficient co-exploration framework of NN models and PIM architectures, named . In, the co-exploration space is carefully designed to adapt both NN models and PIM architectures. Besides, in order to improve search efficiency, we propose an evolutionary search algorithm with adaptive parameter priority (ESAPP). In addition, introduces a multi-level joint simulator to alleviate the problem of time-consuming evaluation. The experimental results show that the proposed co-exploration framework can find better NN models and PIM architectures than existing studies in only six GPU hours (9.8 48.2× speedup). At the same time, can improve the accuracy of co-design results by 15.3% and reduce the energy-delay-product (EDP) by 5.96× compared with existing work.

Topology-Aware Graph Neural Networks for Learning Feasible and Adaptive AC-OPF Solutions

Solving the optimal power flow (OPF) problem is a fundamental task to ensure the system efficiency and reliability in real-time electricity grid operations. We develop a new topology-informed graph neural network (GNN) approach for predicting the optimal solutions of real-time ac-OPF problem. To incorporate grid topology to the NN model, the proposed GNN-for-OPF framework innovatively exploits the locality property of locational marginal prices and voltage magnitude. Furthermore, we develop a physics-aware (ac-)flow feasibility regularization approach for general OPF learning. The advantages of our proposed designs include reduced model complexity, improved generalizability and feasibility guarantees. By providing the analytical understanding on the graph subspace stability under grid topology contingency, we show the proposed GNN can quickly adapt to varying grid topology by an efficient re-training strategy. Numerical tests on various test systems of different sizes have validated the prediction accuracy, improved flow feasibility, and topology adaptivity capability of our proposed GNN-based learning framework.

In vivo-like nearest neighbor parameters improve prediction of fractional RNA base-pairing in cells.

We conducted a thermodynamic analysis of RNA stability in Eco80 artificial cytoplasm, which mimics in vivo conditions, and compared it to transcriptome-wide probing of mRNA. Eco80 contains 80% of Escherichia coli metabolites, with biological concentrations of metal ions, including 2 mM free Mg2+ and 29 mM metabolite-chelated Mg2+. Fluorescence-detected binding isotherms (FDBI) were used to conduct a thermodynamic analysis of 24 RNA helices and found that these helices, which have an average stability of -12.3 kcal/mol, are less stable by ΔΔGo37 ∼1 kcal/mol. The FDBI data was used to determine a set of Watson-Crick free energy nearestneighbor parameters (NNPs), which revealed that Eco80 reduces the stability of three NNPs. This information was used to adjust the NN model using the RNAstructure package. The in vivo-like adjustments have minimal effects on the prediction of RNA secondary structures determined in vitro and in silico, but markedly improve prediction of fractional RNA base pairing in E. coli, as benchmarked with our in vivo DMS and EDC RNA chemical probing data. In summary, our thermodynamic and chemical probing analyses of RNA helices indicate that RNA secondary structures are less stable in cells than in artificially stable in vitro buffer conditions.

An AI-Extended Prediction of Erosion-Corrosion Degradation of API 5L X65 Steel

The application of Artificial Neuronal Networks (ANN) offers better statistical accuracy in erosion-corrosion (E-C) predictions compared to the conventional linear regression based on Multifactorial Analysis (MFA). However, the limitations of ANN to require large training datasets and a high number of inputs pose a practical challenge in the field of E-C due to the scarcity of data. To address this challenge, a novel ANN method is proposed, structured to a small training dataset and trained with the aid of synthetic data to produce an E-C neural network (E-C NN), applied for the first time in the study of E-C wear synergy. In the process, transfer learning is applied by pre-training and fine-tuning the model. The initial dataset is created from experimental data produced in a slurry pot setup, exposing API 5L X65 steel to a turbulent copper tailing slurry. To the previously known E-C scenario for selected values of flow velocity, particle concentration, temperature, pH, and the content of the dissolved Cu2+, new experimental data of stand-alone erosion and stand-alone corrosion is added. The prediction of wear loss by E-C NN considers individual parameters and their interactions. The main result is that E-C ANN provides better prediction than MFA as evaluated by a mean squared error (MSE) values of 2.5 and 3.7, respectively. The results are discussed in the context of the cross-effect between the proposed prediction model and the resulting estimation of relative contribution to E-C synergy, which is better predicted by the E-C NN. The E-C NN model is concluded to be a viable alternative to MFA, delivering similar prediction with better sensitivity to E-C synergy at shorter computation times when using the same experimental dataset.

Hybrid wavelet/Elman NN model for short term cost prediction utilizing developed deer hunting optimizer

The cost signal of electricity in the competitive electrical energy marketplaces is of special importance for all planning and operation activities. Also, the price of electricity has an uncertain nature and various factors affect it in the short and long term. Factors active in the electricity market need to accurately and effectively forecast the electricity price signal to manage risk in the market. For estimating future electricity prices, this research suggests a combined procedure on the basis of Elman neural network model and the wavelet transform. The proposed Elman neural network/wavelet transform forecasted the next hour's power price based on the past 24 h' pricing. This research uses an optimized Elman neural network using a developed deer hunting optimizer and the total model is named Elman neural network/developed deer hunting optimization-wavelet transform. In this paper, Data of Zone Preliminary Billing is used for establishing the training of model and forecasting of performance. The method is then compared with some other published works and the outcomes demonstrate the offered approach superiority toward those for the electrical cost predicting.

An Automatic Non-Destructive External and Internal Quality Evaluation of Mango Fruits based on Color and X-ray Imaging with Machine Learning and Deep Learning Based Classification Models

Quality evaluation of food products, agricultural produce to be specific, has gained momentum from past few decades due to the increased awareness among consumers across the world. This has resulted in the increased emphasis on the development and use of quality assessment techniques in food industry. Moreover, there is a need to automate the quality monitoring of agricultural produce like fruits and vegetables which is otherwise done manually in developing countries hence labor intensive, time consuming and subjective in nature. This paper presents an empirical analysis to build a rapid, robust, real-time, non-destructive computer vision based quality assessment model for mango fruits. The work employs the automatic disease classification of mango fruits based on machine and deep learning models. Firstly, the dataset of colored mango fruits images with 2279 images falling into three classes and another dataset of soft X-ray images of mango fruits with 572 images belonging to two quality classes are developed for detecting external and internal defects, respectively. The multilayer perceptron neural network (MLP NN) with two hidden layers, which may be considered as the starting point for deep learning technique, is proposed as machine learning model to classify the color images of mango fruits into one of three external quality classes with 95.1% accuracy and also to classify the soft X-ray images into two internal quality classes with 97.5% accuracy. In order to step out of feature engineering, actual deep learning convolutional neural network (CNN) models, a customized CNN model and pre-trained CNN models, VGGNet (VGG16) and DenseNet121 were also explored for mango disease classification. The maximum validation accuracy of custom CNN was found to be with 91.52% and 98.7% for color and augmented X-ray images, respectively. The classification accuracy of pre-trained models were found to be reasonably good for the color images but exhibited high variability in results and made it difficult to draw a general conclusion for the proposed datasets. However, the proposed MLP NN model based on few basic intensity and geometric features and also the proposed customized CNN model were found to be the best models and they outperform the state of the art reported in the literature.

Improved SSA‐RBF neural network‐based dynamic 3‐D trajectory tracking model predictive control of autonomous underwater vehicles with external disturbances

AbstractThis paper studies the three‐dimensional (3‐D) dynamic trajectory tracking control of an autonomous underwater vehicle (AUV). As AUV is a typical nonlinear system, each degree of freedom is strongly coupled, so the traditional control method based on the nominal model of AUV cannot guarantee the accuracy of the control system. To solve this problem, we first propose a prediction model based on a radial basis function neural network (RBF‐NN). The nonlinearity of AUV is learned and modeled offline by RBF‐NN based on previous data. This model can reflect the time sequence state and control variables of AUV. Secondly, to avoid the overfitting problem in network training based on the traditional gradient descent method, a new adaptive chaotic sparrow search algorithm (ACSSA) is proposed to optimize the network parameters, to improve the full approximation ability of RBF‐NN to nonlinear systems. To eliminate the steady‐state error caused by external interference during AUV trajectory tracking, a nonlinear optimizer is designed by updating the deviation of the NN model output layer. In each sampling period, the predictive control law is calculated online according to the deviation between the predicted value and the actual value. In addition, the stability analysis based on the Lyapunov method proves the asymptotic stability of the controller. Finally, the 3‐D dynamic trajectory tracking the performance of AUV under different external disturbances is verified by MATLAB/Simulink, and the results show that the proposed controller is more efficient and robust than the standard model predictive controller (MPC) controller and the standard NN model predictive controller (NNPC).

Pragmatic Evaluation of Health Monitoring & Analysis Models from an Empirical Perspective

Implementing and deploying several linked modules that can conduct real-time analysis and recommendation of patient datasets is necessary for designing health monitoring and analysis models. These databases include, but are not limited to, blood test results, computer tomography (CT) scans, MRI scans, PET scans, and other imaging tests. A combination of signal processing and image processing methods are used to process them. These methods include data collection, pre-processing, feature extraction and selection, classification, and context-specific post-processing. Researchers have put forward a variety of machine learning (ML) and deep learning (DL) techniques to carry out these tasks, which help with the high-accuracy categorization of these datasets. However, the internal operational features and the quantitative and qualitative performance indicators of each of these models differ. These models also demonstrate various functional subtleties, contextual benefits, application-specific constraints, and deployment-specific future research directions. It is difficult for researchers to pinpoint models that perform well for their application-specific use cases because of the vast range of performance. In order to reduce this uncertainty, this paper discusses a review of several Health Monitoring & Analysis Models in terms of their internal operational features & performance measurements. Readers will be able to recognise models that are appropriate for their application-specific use cases based on this discussion. When compared to other models, it was shown that Convolutional Neural Networks (CNNs), Masked Region CNN (MRCNN), Recurrent NN (RNN), Q-Learning, and Reinforcement learning models had greater analytical performance. They are hence suitable for clinical use cases. These models' worse scaling performance is a result of their increased complexity and higher implementation costs. This paper compares evaluated models in terms of accuracy, computational latency, deployment complexity, scalability, and deployment cost metrics to analyse such scenarios. This comparison will help users choose the best models for their performance-specific use cases. In this article, a new Health Monitoring Metric (HMM), which integrates many performance indicators to identify the best-performing models under various real-time patient settings, is reviewed to make the process of model selection even easier for real-time scenarios.

Finite element and generalized regression neural network modelling of multiple cracks growth under the influence of multiple crack parameters

PurposeThis study presents finite element (FE) and generalized regression neural network (GRNN) approaches for modeling multiple crack growth problems and predicting crack-growth directions under the influence of multiple crack parameters.Design/methodology/approachTo determine the crack-growth direction in aluminum specimens, multiple crack parameters representing some degree of crack propagation complexity, including crack length, inclination angle, offset and distance, were examined. FE method models were developed for multiple crack growth simulations. To capture the complex relationships among multiple crack-growth variables, GRNN models were developed as nonlinear regression models. Six input variables and one output variable comprising 65 training and 20 test datasets were established.FindingsThe FE model could conveniently simulate the crack-growth directions. However, several multiple crack parameters could affect the simulation accuracy. The GRNN offers a reliable method for modeling the growth of multiple cracks. Using 76% of the total dataset, the NN model attained an R2 value of 0.985.Research limitations/implicationsThe models are presented for static multiple crack growth problems. No material anisotropy is observed.Practical implicationsIn practical crack-growth analyses, the NN approach provides significant benefits and savings.Originality/valueThe proposed GRNN model is simple to develop and accurate. Its performance was superior to that of other NN models. This model is also suitable for modeling multiple crack growths with arbitrary geometries. The proposed GRNN model demonstrates its prediction capability with a simpler learning process, thus producing efficient multiple crack growth predictions and assessments.

Application of Neural Network Algorithms to Implement Drug Abuse Research on Identifying Cannabis Use Disorder through Analysing Personality Traits

Frequent cannabis use will certainly lead to a reasonable mental complication. Daily and long-time period cannabis use is associated with more intellectual loss. This is identified as cannabis use disorder (CUD). The CUD starting at the teen age is the strongest predictors of cognitive impairment. However, it's miles uncertain which comes first. Whether CUD results in early onset cannabis use or cannabisearly use in lives reasons the CUD. The aim of this paper is to investigate the most influential features origins the CUD among the cannabis user’s personality trails. Depending on the self-extracted data collected from 1885 individuals three of the NN models were built. These three NN models apply the 10-Fold cross validation and 30-70 Hold-Out validation methods to provide the reliability of the prediction results. Among the three methods the RNN Hold-Out method achieves 99.19% as best accuracy result.

A study of the effect of printing process parameters on the delamination and surface finish properties of aluminium-infused PLA printed via material extrusion

ABSTRACT Demand for the newer manufacturing techniques to produce materials with difference and specific properties are kept on increasing. The needed specific properties are obtained through composite materials by reinforcing components onto a base material to improve some critical properties. In that way, the polylactic acid (PLA) incorporated with aluminium was used to print the samples through fused filament fabrication (FFF) process. During printing, the printing process parameters have considerable effect on the material properties. At first, specimens are printed as per the Taguchi’s experimental design with L9 Orthogonal array design and the outputs are obtained experimentally. As an extension, this study proposed two neural network-based machine learning models to predict the influence of process parameters like printing temperature (PT), layer height (LH), raster angle (RA) and fill pattern (FP) on the surface finish (SF), surface finish near to the drilled holes (SFNDH), and delamination factor (DF). Both the proposed models recorded the overall prediction accuracy about 99.99%. To ensure the results predicted by the NN models, validation experiments are conducted and compared with the model predicted results. The experimental results are in good agreement with the predicted values which are obtained through numerical modelling.

Adaptive neural network ensemble using prediction frequency

AbstractNeural network (NN) ensembles can reduce large prediction variance of NN and improve prediction accuracy. For highly non-linear problems with insufficient data set, the prediction accuracy of NN models becomes unstable, resulting in a decrease in the accuracy of ensembles. Therefore, this study proposes a prediction frequency-based ensemble that identifies core prediction values, which are core prediction members to be used in the ensemble and are expected to be concentrated near the true response. The prediction frequency-based ensemble classifies core prediction values ​​supported by multiple NN models ​​by conducting statistical analysis with a frequency distribution, which is a collection of prediction values ​​obtained from various NN models for a given prediction point. The prediction frequency-based ensemble searches for a range of prediction values that contains prediction values above a certain frequency, and thus the predictive performance can be improved by excluding prediction values with low accuracy ​​and coping with the uncertainty of the most frequent value. An adaptive sampling strategy that sequentially adds samples based on the core prediction variance calculated as the variance of the core prediction values is proposed to improve the predictive performance of the prediction frequency-based ensemble efficiently. Results of various case studies show that the prediction accuracy of the prediction frequency-based ensemble is higher than that of Kriging and other existing ensemble methods. In addition, the proposed adaptive sampling strategy effectively improves the predictive performance of the prediction frequency-based ensemble compared with the previously developed space-filling and prediction variance-based strategies.

Simulation of motions of a 6DoF unmanned aerial vehicle from the mathematical model in free space

This work prepares a smooth playing ground to modeling and simulation of a Collision Avoidance for Cooperative UAVs with Rolling Optimization Algorithm (ROA) Based on Predictive State Space (PSS) technique using Matlab, Simulink and Aerospace toolboxes of MathsWork. The methodological approach adopted is Computer Aided Software Engineering and to have handle collisions and maneuvers in a mathematical way, normal simulation of the trajectory of both the UAV and obstacles is defined by geometric approach. The goal of the work is to develop a mathematical model of a six degree of freedom (6DoF) of unmanned aerial vehicle (UAV) in free space for further modeling and simulation of a collision avoidance for cooperative UAVs with the required specifications. The motion (modeling of mathematical parameters like velocity, position, body rotational rate and Euler angles) of the 6DoF aircraft was determined by coordinate systems which allow the aircraft’s position and orientation in spaceto be kept tracked. The discrete form of the developed model for simulation was achieved using appropriate transfer functions. The development of the mathematical model will enable the development of an artificial NN model predictive controller (MPC) that can handle the nonlinearities associated with the UAV based on PSS technique.

Volatile organic compounds for early detection of prostate cancer from urine

Prostate cancer (PCa) is one of the most common cancers in men worldwide. Early diagnosis of PCa is extremely challenging due to the lack of effective diagnostic methods. The study presented here aims to evaluate whether urine volatile organic compounds (VOCs) can be used as an emerging diagnostic biomarker for PCa. Gas chromatography-ion mobility spectrometry (GC-IMS) was used to detect VOCs in urine samples from 66 patients with PCa and to comparatively analyze samples from 87 patients with non-cancerous controls (NCs). A total of 86 substance peak heights were detected in urine samples from all patients. Analysis using four machine learning algorithms suggested that the diagnosis of PCa could be effectively facilitated. Ultimately, diagnostic models were constructed based on the four VOCs selected. The AUC for the RF and SVM model were 0.955 and 0.981, respectively. Both the NN and DT diagnostic models also achieved an AUC of 0.8 or more, but their sensitivity or specificity was poor compared to the RF and SVM models.

A machine learning model for orthodontic extraction/non-extraction decision in a racially and ethnically diverse patient population

The purpose of the present study was to create a machine learning (ML) algorithm with the ability to predict the extraction/non-extraction decision in a racially and ethnically diverse sample. Data was gathered from the records of 393 patients (200 non-extraction and 193 extraction) from a racially and ethnically diverse population. Four ML models (logistic regression [LR], random forest [RF], support vector machine [SVM], and neural network [NN]) were trained on a training set (70% of samples) and then tested on the remaining samples (30%). The accuracy and precision of the ML model predictions were calculated using the area under the curve (AUC) of the receiver operating characteristics (ROC) curve. The proportion of correct extraction/non-extraction decisions was also calculated. The LR, SVM, and NN models performed best, with an AUC of the ROC of 91.0%, 92.5%, and 92.3%, respectively. The overall proportion of correct decisions was 82%, 76%, 83%, and 81% for the LR, RF, SVM, and NN models, respectively. The features found to be most helpful to the ML algorithms in making their decisions were maxillary crowding/spacing, L1-NB (mm), U1-NA (mm), PFH:AFH, and SN-MP(̊), although many other features contributed significantly. ML models can predict the extraction decision in a racially and ethnically diverse patient population with a high degree of accuracy and precision. Crowding, sagittal, and vertical characteristics all featured prominently in the hierarchy of components most influential to the ML decision-making process.

Neural Network Compression for Noisy Storage Devices

Compression and efficient storage of neural network (NN) parameters is critical for applications that run on resource-constrained devices. Despite the significant progress in NN model compression, there has been considerably less investigation in the actual physical storage of NN parameters. Conventionally, model compression and physical storage are decoupled, as digital storage media with error-correcting codes (ECCs) provide robust error-free storage. However, this decoupled approach is inefficient as it ignores the overparameterization present in most NNs and forces the memory device to allocate the same amount of resources to every bit of information regardless of its importance. In this work, we investigate analog memory devices as an alternative to digital media – one that naturally provides a way to add more protection for significant bits unlike its counterpart, but is noisy and may compromise the stored model’s performance if used naively. We develop a variety of robust coding strategies for NN weight storage on analog devices, and propose an approach to jointly optimize model compression and memory resource allocation. We then demonstrate the efficacy of our approach on models trained on MNIST, CIFAR-10, and ImageNet datasets for existing compression techniques. Compared to conventional error-free digital storage, our method reduces the memory footprint by up to one order of magnitude, without significantly compromising the stored model’s accuracy.

An Efficient Wildfire Detection System for AI-Embedded Applications Using Satellite Imagery

Wildfire risk has globally increased during the past few years due to several factors. An efficient and fast response to wildfires is extremely important to reduce the damaging effect on humans and wildlife. This work introduces a methodology for designing an efficient machine learning system to detect wildfires using satellite imagery. A convolutional neural network (CNN) model is optimized to reduce the required computational resources. Due to the limitations of images containing fire and seasonal variations, an image augmentation process is used to develop adequate training samples for the change in the forest’s visual features and the seasonal wind direction at the study area during the fire season. The selected CNN model (MobileNet) was trained to identify key features of various satellite images that contained fire or without fire. Then, the trained system is used to classify new satellite imagery and sort them into fire or no fire classes. A cloud-based development studio from Edge Impulse Inc. is used to create a NN model based on the transferred learning algorithm. The effects of four hyperparameters are assessed: input image resolution, depth multiplier, number of neurons in the dense layer, and dropout rate. The computational cost is evaluated based on the simulation of deploying the neural network model on an Arduino Nano 33 BLE device, including Flash usage, peak random access memory (RAM) usage, and network inference time. Results supported that the dropout rate only affects network prediction performance; however, the number of neurons in the dense layer had limited effects on performance and computational cost. Additionally, hyperparameters such as image size and network depth significantly impact the network model performance and the computational cost. According to the developed benchmark network analysis, the network model MobileNetV2, with 160 × 160 pixels image size and 50% depth reduction, shows a good classification accuracy and is about 70% computationally lighter than a full-depth network. Therefore, the proposed methodology can effectively design an ML application that instantly and efficiently analyses imagery from a spacecraft/weather balloon for the detection of wildfires without the need of an earth control centre.

TransFNN: A Novel Overtemperature Prediction Method for HVDC Converter Valves Based on an Improved Transformer and the F-NN Algorithm.

Appropriate cooling of the converter valve in a high-voltage direct current (HVDC) transmission system is highly significant for the safety, stability, and economical operation of a power grid. The proper adjustment of cooling measures is based on the accurate perception of the valve's future overtemperature state, which is characterized by the valve's cooling water temperature. However, very few previous studies have focused on this need, and the existing Transformer model, which excels in time-series predictions, cannot be directly applied to forecast the valve overtemperature state. In this study, we modified the Transformer and present a hybrid Transformer-FCM-NN (TransFNN) model to predict the future overtemperature state of the converter valve. The TransFNN model decouples the forecast process into two stages: (i) The modified Transformer is used to obtain the future values of the independent parameters; (ii) the relation between the valve cooling water temperature and the six independent operating parameters is fit, and the output of the Transformer is used to calculate the future values of the cooling water temperature. The results of the quantitative experiments showed that the proposed TransFNN model outperformed other models with which it was compared; with TransFNN being applied to predict the overtemperature state of the converter valves, the forecast accuracy was 91.81%, which was improved by 6.85% compared with that of the original Transformer model. Our work provides a novel approach to predicting the valve overtemperature state and acts as a data-driven tool for operation and maintenance personnel to use to adjust valve cooling measures punctually, effectively, and economically.