Dissolved Gas Analysis (DGA) is a key diagnostic tool for power transformers that can detect early faults. Traditional heuristic methods (Key Gas, Duval Triangle) are simple but error-prone, achieving roughly 60-75% accuracy on small utility datasets (<1000 samples). Machine Learning (ML) and Deep Learning (DL) approaches have been adopted to overcome these limits: classical ML models (e.g., Support Vector Machines, Random Forests) reach similar to 85-95% accuracy on datasets of similar to 300-1000 samples but face class imbalance and interpretability issues. Larger datasets (>= 5000 samples) and advanced DL architectures including Generative Adversarial Networks (GANs), Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM) networks can exceed 97% fault-classification accuracy, but tend to increase false labeling of complex/mixed faults and suffer from opaque decision-making and high computational cost, limiting industrial uptake. Hybrid Artificial Intelligence (AI) methods that integrate physics-based or rule-governed frameworks (e.g., neuro-fuzzy models, Bayesian networks) offer better interpretability and robustness, typically delivering 95-98% accuracy on medium-sized datasets (similar to 600-1500 samples). Emerging approaches like digital twins combine current DGA data with thermal and electrical simulators to predict Remaining Useful Life (RUL) within +/- 10-15% error margins. Internet of Things (IoT) enabled deployments show potential for scaling but raise interoperability, cybersecurity, and standardization challenges. This review quantitatively compares conventional, ML-based, and digital-twin-supported DGA methodologies, assessing accuracy, requirements, and limitations, and proposes a roadmap emphasizing hybrid explainable models, low-cost online sensors, federated learning, and built-in cybersecurity to transition DGA from a diagnostic tool to an AI-supported predictive engine for grid resilience. The review emphasizes the use of ML, DL, explainable artificial intelligence (XAI), federated learning, and digital twins to diagnose transformer faults, monitor their conditions, predict maintenance needs, and estimate remaining useful life.

ANN-based controller is developed and evaluated for onboard converters for V2V energy transfer in electric vehicles (EVs). The ANN controller is adaptive unlike PI controllers and regulates nonlinear converter dynamics according to the variations in battery voltage levels and load conditions. V2V operating modes explore forward boost, reverse buck. The quantitative results demonstrated that in comparison with PI control, the ANN controller allows an increase of power transfer efficiency by up to 7.7%, reduces the steady-state current error by 67%, and suppresses the voltage and current ripple by more than 60%. The dynamic settling time is improved by 46%, providing faster current tracking and smoother transition of SOC for both EVs. This means less strain on the battery, better reliability on the converters, and more efficient bidirectional energy transfer. It has been shown that the ANN technique for control is a superior alternative to traditional methods in V2E energy management onboard and can ensure decentralization for future V2G and V2H integrations.

Modern power distribution networks are experiencing a great deal of variability and uncertainty due to the explosive growth of electric vehicles. Therefore, precise forecasting and flexible optimization techniques are necessary for effective charging demand coordination. An integrated forecast-guided multi-agent reinforcement learning framework is proposed in this paper, combining real-time decentralized optimization with attention-based long short-term memory forecasting for large-scale electric vehicle charging systems. The forecasting module accurately predicts short-term aggregate electric vehicle load using spatio-temporal features such as historical demand, temperature, and dynamic pricing. Quantitative evaluation takes place on real charging-station data, showing that attention-based long short-term memory achieves root mean square error = 7.65 kW, outperforming classical ones (Autoregressive Integrated Moving Average, Support Vector Regression) and new architectures (Transformer, Temporal Fusion Transformer) by 2%-4% at 15, 30, and 60 min horizons. It was shown that multi-agent reinforcement learning -layer learned coordinated policies, when trained under a centralized-training/decentralized-execution paradigm, that reduced peak demand by 16.8%, lowered total energy costs by 12.9%, and resulted in convergence over 150 episodes, all more than mixed-integer linear programming and non-forecasted multi-agent reinforcement learning baselines. The model really notes on meaningful frangible periods such as evening spikes of charging, thus, verifying the analysis of interpretation concerning attention heatmaps. Scalability and uncertainty checks reveal nearly linear growth in runtime with a robustness of +/- 20% noise on the forecast. The proposed hybrid framework results, in general, in an effective computationally interpretable, and real-time adaptable solution for proactive EV load management in a sustainable manner for smart grid operators.

A hybrid Model Predictive Control (MPC)–Offline Reinforcement Learning (RL) framework for lifetime-optimized State-of-Charge (SoC) regulation in electric vehicle (EV) batteries is presented in this paper. It is forecast- and State-of-Health (SoH)-aware. Achieving lifetime-aware State-of-Charge (SoC) regulation with enhanced tracking precision, decreased degradation, and computational viability under representative drive cycles is the main goal of this work. Hardware-in-the-loop (HIL)-validated physics-based electrothermal model for real-time state forecasting, an offline-trained RL policy for adaptive decision-making, and a constrained MPC safety layer are all integrated into the suggested architecture. With peak temperature deviations under 2.5 °C, experimental validation under Worldwide Harmonized Light Vehicle Test Procedure (WLTP), New European Driving Cycle (NEDC), and Urban Dynamometer Driving Schedule (UDDS) cycles shows strong SoC tracking and thermal safety. In comparison to baseline MPC, the hybrid controller improves forecast accuracy by 12.8

Model order reduction plays a critical role in simplifying complex discrete-time linear time-invariant (LTI) systems while preserving essential system dynamics. Traditional reduction techniques primarily focus on frequency-domain characteristics, often neglecting specific performance criteria crucial to control system design. This manuscript presents a novel method for reducing discrete-time systems with an explicit emphasis on matching key prominent parameters i.e., rise, settling times, over-shoot, and steady-state error. The proposed approach formulates model reduction as a constrained optimization problem, where the reduced model is derived to minimize time-domain response error relative to the original system (OS). Extensive numerical experiments on benchmark systems demonstrate the advantage of the projected method compared to traditional techniques.

In this article, the prominent parameters of lightning channel-base-current (LCBC) function are evaluated for precisely representing measured data of negative first return stroke (NFS), negative subsequent return stroke (NSS), and positive first return stroke (PFS) using Hippopotamus optimization (HOA) technique. Heidler’s function undertaken by the IEEE standard for representing the measured data is considered in this work and its parameters (I01, I02, τ11, τ12, τ21, τ22, n1, and n2) are assessed using HOA for representing all the prominent parameters of the typical (50%) and severe (5% & 1%) cases of NFS, NSS, and PFS reported by IEC Standard 62305-1, 2010. The results of HOA based Heidler’s function constants are illustrated. Moreover the percentage error is also calculated between the desired and computed was calculated and a substantial concurrence was found. We demonstrate that employing an HOA allows for precise detection of HFPs, which in turn simplifies the derivation of the full-wave CBC. In addition, this technique is expected to be beneficial for future research on the LCBC behavior under lightning conditions.

Large dimensional systems are complicated and very tough to control. The effective solution is to reduce the large dimension of the systems to a lower dimension. This paper aims to reduce the dimension of a higher-order fractional commensurate interval system to a low-order fractional commensurate interval system by using evolutionary techniques. Kharitonov's theorem and artificial bee colony optimization technique are used to determine the interval numerator and denominator polynomials for the simplified models. The algorithm is very modest and generates a stable reduced dimensional commensurate fractional interval model preserving the properties of the original system. The efficiency of the suggested strategy is illustrated with a numerical example.

This work relates to the reduction of a noninteger commensurate high dimensional system. The essential objective of this article is to come up with an approximating technique to replace the original high dimensional system with a low dimensional model preserving the properties of the original system in its shortened model. Superiority of the proposed technique is exhibited by correlating the reduced model with the models of other current methods. The simulation results are cited to approve that the recommended technique has high efficiency with a closing value of time-domain specifications. Conclusively, more logical comparisons were made between other existing methods. The performance indices are calculated for both original system and reduced model system and presented in the manuscript.

Three low carbon steel grades with Mn content varying between 1.5 and 2.5% were subjected to a dual phase heat treatment involving slow cooling in the inter critical temperature range followed by water quenching to generate a hard martensite and soft ferritic phase constitution. The same three steels were subjected to two different heat treatments that could generate hard bainitic ferrite and a soft ferritic phase. The three steels were austenitized at 870 °C, well above Ac3 temperature, 850 °C very close to Ac3 temperature and 790 °C in the inter critical temperature range. Post austenitization one set of samples were forced air cooled and another set was subjected to isothermal holding in a salt bath at 450 °C. Compared to dual phase heat treatment, a ferritic bainitic treatment gives a moderate loss in strength but better ductility was obtained in the forced air cooled condition. The strain hardening exponent was significantly lower for the dual phase steel with martensite content. The strain hardening exponent were much higher for ferritic bainitic condition with distinct slope for the initial deformation in the ferrite phase followed by bainitic phase and retained austenite phase. The ferritic-bainitic condition with isothermal bainitic holding condition gave almost double the ductility and the tensile elongation exceeded 20 GPa.%. With some modification dual phase steels can be modified into ferritic bainitic steels with attractive formability properties. The various aspects of structural evolution and properties has been brought out.

Differential evolution optimization algorithm (DEOA) is effectively utilized in this paper to determine the parameters of single-stage impulse generator circuit (SSIGC) (such as R1, R2, C1, and C2) for generating standard lightning impulse (SLI) waveform in the high voltage (HV) testing laboratory. Further, this SLI waveform is sufficient to evaluate dielectric strength of HV equipment used in power system network. The results that are evaluated and illustrated in this paper are beneficial to generate SLI voltage waveforms as given in standards IEC 62305-1, 2010 and IEEE 4-2013. The attempt made in this paper is the first of its kind to use DEOA as searching tool to determine the parameters of SSIGC. This methodology assists to conquer the conventional methods for determining the parameters SSIGC. These results reported in this paper are validated through the SSIGC prototype developed in MATLAB/Simulink.

In this paper, a new approach based on particle swarm optimization (PSO) algorithm is used to evaluate the single-stage impulse generator (SSIG) circuit parameters (R1, R2, C1, and C2) from the known values of rise and tail times. The results reported in this paper provide the standard lightning impulse (LI) signals which meticulously matches with those waveforms illustrated in the IEEE 4-2013, and IEC-62305-1, 2010 standards. This is the first of its kind attempt to use PSO as a tool to evaluate the SSIG circuit parameters. This approach helps to overcome the practice of conventional trial and error methods for finding the impulse circuit parameters. These results are verified by the impulse circuit developed in MATLAB.

The effect of substituting the fossil fuel partly with gaseous fuels on CO2 emission in sinter plants was assessed. Fuel gases such as coke oven gas (COG) and liquefied petroleum gas (LPG) were injected into the sinter pot, to see the effect on sintering speed and quality parameters. Densification was predominant in the case of COG and LPG, leading to an increase in both cold and hot strength. Acceleration of the sintering process was noticed when oxygen was injected. The various phases formed as a result of injecting COG, LPG, and oxygen were identified with the help of X-ray diffractometer (XRD). Dicalcium silicate, a detrimental phase, was found in the range of 16–25% when COG and LPG were injected and was absent when oxygen was injected. Magnetite (Fe3O4) and silico ferrite of calcium and aluminum (SFCA) formation got enhanced with oxygen injection. Injection of oxygen along with COG and LPG reduced dicalcium silicate and resulted in acceleration of the sintering process as well as superior quality parameters. The presence of manganese in the iron ore up to 2% resulted in MnFe2O4, a spinel phase that enriched the magnetite. No adverse effect of goethite was noticed due to gas injection. A reduction of 0.65 million tons in CO2 emissions per annum was envisaged by substituting 10 kg solid fuel with gaseous fuel per ton of sinter. After successful pot tests and encouraging results therein, a COG injection system was designed and installed in an industrial-scale sinter plant.

Analysis of transient response of current along the horizontal buried conductor due to the excitation of standard lightning impulse current in lossy grounding conditions is an important objective for designing the proper grounding system. In this paper, the transient analysis of current along the buried conductor is performed and reported due to the application of standard lightning current of 100 kA, 1.2/50 µs under different grounding conditions (i.e., \(\sigma = 0.01,0.001,0.0001,0.00001\,{\text{s}}/{\text{m}}\)) by using transmission line (TL) approach. This is the first of its kind attempt that the standard lightning current is considered as an exciting current. It is observed from the reported results that the conductivity of ground much effects on peak of transient current along the buried conductor. Further, this approach will be helpful in order to design proper grounding system to diminish the peak current along the horizontal buried conductor.

Estimation and analysis of the lightning channel-base-current (CBC) are essential requirements in the process of designing protection systems against lightning strokes. Thus, a new method is proposed to evaluate the lightning CBC based on the measured magnetic field using the deconvolution method. Engineering return-stroke models, such as transmission line (TL), modified TL with linear current decaying with height (MTLL), and modified TL with exponential current decaying with height (MTLE) models are effectively utilized in the evaluation of lightning CBC from the measured magnetic field. The data of measured magnetic field above the ground surface at four radial distances from the lightning channel, reported in the literature, are considered as input to the proposed method. The peak of lightning CBC is calculated for return-stroke velocities in range one-third to two-third the velocity of light in free space, specifically at 100, 125, 150, 175, and 200 m/mu s for aforementioned measured magnetic field data using MTLE, MTLL, and TL models. The variations in the peak currents, corresponding to the variation in return-stroke velocities, are also observed. Closeness in the peak of simulated CBC between each of the models used in the study is analyzed and reported. Furthermore, this approach will be helpful for better understanding while designing protection systems against lightning.

According to IEC 60060-1, 2010 and IEEE 4-2013 standards, the standard lightning impulse (LI) voltage waveforms are needed for testing the insulation of large equipment. The generation of these impulse waveforms sometimes difficult due to presence of internal inductance (Ls) in impulse generator (IG) circuit, the level of test object capacitance (C2), and the level of test object inductance (Lt). Thus, a new procedure is proposed for identifying the limit of wavefront resistance (R1) to generate required impulse waveforms (i.e., rise time (tr ≤ 1.56 µs) and overshoot rate (β'≤5%)) when the parameters Ls, C2, and Lt of fourth order IG circuit are systematically varies. Another attempt has also made in this paper to determine the limit values of the test object parameters such as load resistance Rt and load inductance Lt for matching the response of fourth order to third order IG circuit. Further, the analysis performed in this paper will be helpful in generating the test impulse voltage waveforms for testing ultra-high voltage (UHV) transformer.

In this paper, a new method is proposed for directly evaluating approximate lightning channel-base-current (CBC) parameters (I-m, (di/dt)(max), t(FD), and t(SD)) and analytical function parameters (AFPs) of the CBC based on the measured lightning magnetic fields using a deconvolution and differential evolution (DE) method. This proposed method comprises two sequential steps. The first step is for determining the lightning CBC based on the measured magnetic field using modified transmission line model with exponential current decaying with height (MTLE) through deconvolution, while the second step is for evaluating the AFPs (Heidler's function) corresponding to the lightning CBC obtained from the first step. Data from the measured magnetic field above ground level (z = 10 m) at different horizontal distances (r = 9 km, 4.6 km, 2 km, and 50 m) from the lightning channel are considered for testing the proposed method. The sum of two Heidler's functions is used to represent the lightning CBC; and its parameters, namely I-01, I-02, tau(11), tau(12), tau(21), tau(22), n(1), and n(2) are identified using DE as a tool. Further, this methodology is helpful in the evaluation of the CBC parameters and mathematical modeling of the lightning CBC and the estimation of the lightning current through the channel using data from the measured magnetic field for different engineering return stroke models.

In this study, particle swarm optimisation (PSO) is applied to determine the Heidler's function parameters (HFPs) for approximating the measured full lightning channel-base-current (CBC) of the severe (5% data in IEC standard 62305-1, 2010) negative first return stroke (NFS) and subsequent return stroke (NSS). The parameters in the sum of two Heidler's functions, namely I01, I02, τ11, τ12, τ21, τ22, n1, and n2, are identified using PSO as a tool. The HFPs for severe NFS and NSS are tuned to keep four parameters constant with systematic variation of one out of all prominent parameters (i.e. peak current (Im), maximum time rate of change of current ((di/dt)max), front duration (tFD), stroke duration (tSD), and total charge (Q)) of full CBC wave and vice versa. The corresponding data is reported, and an independent controlling of HFPs is analysed to achieve the desired full CBC wave. Also, an attempt has been made in this study to examine the interdependence between the wavefront and tail of CBC using PSO. The data thus generated and reported for severe NFS and NSS can be used to generate the suitable full CBC waveforms. This methodology will be helpful in lightning-related simulation study and research.

In this paper, the lightning channel-base-current (CBC) function parameters are identified using four different optimization techniques (such as genetic algorithm (GA), particle swarm optimization (PSO), differential evolution (DE), and sine cosine algorithm (SCA)) to representing the measured data given in IEC standard 62305-1, 2010 for the severe (5%) case negative subsequent return stroke (NSS). The sum of two Heidler’s functions (HFs) is used for representing the desired severe NSS, and its parameters (namely, I01, I02, τ11, τ12, τ21, τ22, n1, and n2) are identified using GA, PSO, DE, and SCA. The data thus generated and reported for approximating the measured severe NSS. During this process, the data obtained by SCA showing a good agreement between the computed and measured CBC as compared to the data obtained by GA, PSO, and DE. However, the data provided by all above mentioned optimization algorithms can be used to observe the behavior of the return stroke current at different heights through a lightning channel, and this will useful in calculation of lightning electromagnetic field (LEMF).

Lightning channel-base-current (CBC) function parameters for negative first return stroke (NFS), negative subsequent return stroke (NSS), and positive first return stroke (PFS) are evaluated using particle swarm optimization (PSO) technique. In this paper, parameters (I-01,I-02, T-11,T-12, T-21, T-22, n(1) and n(2)) in the sum of two Heidler's CBC function are identified using PSO as a tool fOr the typical (50%) and severe (5%) cases of NFS, NSS, and PFS reported by IEC Standard 62305-1, 2010. The result of PSO tuned CBC, the error between computed and measured as given in the standard was calculated and a good agreement was observed. Heidler's function parameters (HFPs) are also tuned to keep four constants with systematic variation of one out of all prominent parameters (I-m, (d(i)/d(t))(max), t(FD),t(SD), and Q) of full (front and tail) CBC wave of NFS and NSS. Interdependence between these parameters is scrutinized during resemblance of the computed and the measured CBC. This is an attempt to show that a PSO can be used for identifying HFPs to straightforwardly obtain the CBC full wave. Thus, the data generated and reported including extreme severe case (1%) can produce a tailor-made full wave of CBC. Furthermore, this approach will be needful in research related to the lightning CBC characteristics.

In this paper, a new approach has been presented for evaluation of the impulse circuit parameters from known physical parameters (rise or peak time and tail time) using Nelder-Mead algorithm (NMA). Results of second order impulse circuit parameters such as C 1 , C 2 , R 1 , and R 2 are reported for generating lightning impulse (LI) and switching impulse (SI) as given in standards (IEEE 4-2013 and IEC 60060-1). The determined circuit parameters are validated for practical applications through the impulse circuit developed in MATLAB. The presented method enables accurate estimation of unknown impulse circuit parameters from the known physical parameters. Impulse wave (LI & SI) attained using analytical (double exponential) expression and through simulation circuits are presented. The percentage error in peak time, rise time, tail time and peak voltage for both standard LI (1.2/50μs) and standard SI (250/2500μs) are also reported.