BITS Faculty Publications
Permanent URI for this communityhttp://localhost:4000/handle/123456789/1867
Browse
70 results
Search Results
Item Deep q-learning framework based solar energy generation prediction(2022) Gupta, Karunesh KumarIn the recent years, the power system market has seen a huge shift towards the utilisation of Renewable energy sources (RES) as candidates of power generation since they proved to be a great alternative of conventional sources due to its low carbon footprints and less dependency over fossil fuels, thereby increasing the penetration of RES in microgrids. RES sources like Wind, Tidal, Hydro, Solar etc. are widely available today, among which Solar is the most popular source of energy due to its cheap running cost and easy installation. However, Solar faces complications such as intermittency which is a very big drawback to its applicability and reliability thus requiring additional strategies to increase its resiliency. A short term forecasting of solar generation might be a great solution to observe the intermittency and predict the future generation based on various factors. In this research work a Deep Q-learning framework was proposed to predict the Solar Generation and provide predictive results for special months and days like Spring equinox, Summer Solstice, Autumn Equinox and Winter Solstice. The Deep Q-learning (DQNN) framework is an amalgamation of Deep Learning networks and Q-learning technique that exploits the properties of Deep Learning networks and Q-learning technique to map the state-Q-value pair and perform the prediction process. The simulations of DQN network was performed in an open-source platform known as Keras and the prediction results were compared both in simulated and experimental datasets with other well known Deep learning networks such as CNN, LSTM, GRU and CNN-LSTM.Item A detailed comparative analysis of automatic neural metrics for machine translation: bleurt & bertscore(IEEE, 2025-04) Chamola, Vinay; Gupta, Karunesh KumarBleurt a recently introduced metric that employs Bert, a potent pre-trained language model to assess how well candidate translations compare to a reference translation in the context of machine translation outputs. While traditional metrics like Bleu rely on lexical similarities, Bleurt leverages Bert's semantic and syntactic capabilities to provide more robust evaluation through complex text representations. However, studies have shown that Bert, despite its impressive performance in natural language processing tasks can sometimes deviate from human judgment, particularly in specific syntactic and semantic scenarios. Through systematic experimental analysis at the word level, including categorization of errors such as lexical mismatches, untranslated terms, and structural inconsistencies, we investigate how Bleurt handles various translation challenges. Our study addresses three central questions: What are the strengths and weaknesses of Bleurt, how do they align with Bert's known limitations, and how does it compare with the similar automatic neural metric for machine translation, BERTScore? Using manually annotated datasets that emphasize different error types and linguistic phenomena, we find that Bleurt excels at identifying nuanced differences between sentences with high overlap, an area where BERTScore shows limitations. Our systematic experiments, provide insights for their effective application in machine translation evaluation.Item Qualitative and quantitative analysis of Indian honey samples and various adulterants using near infrared spectroscopy with water as a sensing probe (aquaphotomics) and chemometrics(Elsevier, 2025) Gupta, Karunesh KumarVarious honey samples and their adulterants from C3 plants (rice syrup) and C4 based plants (sugar syrup, corn syrup, and jaggery syrup) were analyzed using Near-Infrared Spectroscopy (NIRS) coupled with aquaphotomics and chemometric algorithms for qualitative and quantitative assessment. To validate the authenticity of the collected samples, stable carbon isotope ratio analysis (SCIRA) was performed. Spectral data for honey samples were acquired using NIRS (600–2600 nm, 1 nm resolution) and optimized using aquaphotomics by selecting wavelengths associated with water characteristics. Additionally, the aquaphotomics wavelength range was expanded by including spectral variables related to the O-H bend second overtone (1940 nm), O-H stretch/O-H bend combination (1960 nm), and O-H bend/C-O stretch combination band (2100 nm). The performance of these optimized variables was evaluated using qualitative (PCA, k-means, LDA, SVM) and quantitative (PCR, SVR, PLS) analyses, achieving a maximum classification accuracy of 100 % and a regression coefficient (R²) of 0.999. This study provides a rapid, non-destructive, and highly accurate method for honey authentication, offering significant applications in food quality control and combating fraudulent honey adulteration. The proposed approach can be effectively implemented in the honey industry and regulatory bodies to ensure product authenticity, protect consumer health, and maintain market integrity.Item Active Power Filter Control Algorithm using Wavelets(IEEE, 2006) Kumar, Rajneesh; Gupta, Karunesh KumarThis paper presents a wavelet transform (WT) based technique to extract fundamental frequency component from a nonsinusoidal and unbalanced load current in a three phase system. The fundamental frequency component is extracted using multiresolution analysis (MRA). The remaining harmonics can be used by the active filter for compensation. Simulation result obtained for a rectifier load current shows the usefulness of the proposed method.Item Wavelet Based Switching Loss Analysis of MOSFET(IEEE, 2007) Gupta, Karunesh Kumar; Kumar, RajneeshSwitching loss calculation in MOSFET requires device parameters like turn-on and turn-off time, input and output capacitances, parasitic inductances and circuit parameters like voltage, current and operating frequency. Using these parameters switching loss is calculated with given approximate mathematical formulae. This paper presents a wavelet based method for switching loss calculation. It requires only the voltage and current waveforms during switching and calculates the power loss and also provides the frequency content during switching. The information regarding frequency content can be utilized for designing snubber as well as for EMI analysis. Multi Resolution Analysis (MRA) is used to decompose signals in wavelet domain and the signals are transformed in different frequency bands. The power is calculated in each band by multiplication of current and voltage wavelet coefficients. Simulation results are presented for a MOSFET with inductive load to support the method described.Item Adaptive Shrinkage Function Optimization by Differential Evolution(IEEE, 2008) Gupta, Karunesh Kumar; Gupta, RajivIn this paper, a new wavelet shrinkage denoising algorithm is presented. The algorithm uses wavelet transform (WT) to extract information about sharp variation in multiresolution images and applies shrinkage function adapting the image features. The features are detected by energy of neighboring pixels, whereas in standard wavelet methods, the empirical wavelet coefficients shrink pixel by pixel, on the basis of their individual magnitude. The shrinkage function is optimized by differential Evolution (DE)Item Feature Adaptive Wavelet Shrinkage for Image Denoising(IEEE, 2007) Gupta, Karunesh Kumar; Gupta, RajivIn this paper, a new wavelet shrinkage denoising algorithm is presented. The algorithm uses wavelet transform (WT) to extract information about sharp variation in multiresolution images and applies shrinkage function adapting the image features. The shrinkage function depends on energy of neighboring pixels, whereas in standard wavelet methods, the empirical wavelet coefficients shrink pixel by pixel, on the basis of their individual magnitude. Experiments show that wavelet shrinkage algorithm which uses neighboring pixels energy improves the denoising performance and achieves better peak signal to noise ratio compared to other thresholding algorithms. Due to its low complexity, the proposed algorithm is very suitable for hardware implementationItem Hybrid LDPC and STBC Algorithms to Improve BER Reduction in OFDM Systems(Scope, 2013-11) Gupta, Karunesh KumarTHE SEARCH for a good coding algorithm is motivated by the fact that various communication channels require their optimal performance. Orthogonal Frequency Division Multiplexing (OFDM) WiMAX transmission system is used over different fading channels with adaptive modulation and coding (AMC). The computer simulation result shows improved error-correction capability in LDPC codes and STBC codes. The paper proposed hybrid LDPC and STBC method and proved its good bit error performance.Item Wavelet denoising: Comparative analysis and optimization using machine learning(IEEE, 2014) Gupta, Karunesh KumarEven after a phenomenal progress in the quality of image denoising algorithms over the years, there is yet a vast scope of improving the standard of denoised images. This paper presents a new methodology for denoising by integrating the wavelet denoising technique with regression boosted trees. Based on ensemble learning by regression boosted trees, an optimal threshold value is obtained. Its denoising performance is better than Stein's unbiased risk estimator-linear expansion of thresholds (SURE-LET) method which is an up to date denoising algorithm. We have also compared its performance with the other current state of art wavelet based denoising algorithms like ProbShrink, and BiShrink on the basis of their Peak Signal to Noise Ratio (PSNR). Simulations and experimentation results demonstrate that PSNR of our proposed method outperforms the other methods. Extension to Dual Tree-Complex Wavelet Transform (DT-CWT) is also presented.Item Vibro acoustic signal analysis in fault finding of bearing using Empirical Mode Decomposition(IEEE, 2013) Gupta, Karunesh KumarBearing fault is an issue in process and control industries, and has significant impact in the production flow. The behaviour of the machinery can be well understood from the frictional forces of the bearing due to load, and also the wear and tear of the ball bearings. The characteristic of this ball bearing can predict the exact nature of the load and any future malfunction in the operating equipments. The signals generated from these bearings can be of any types i.e., sound or vibration. The acoustic phenomenon is tough to predict in noisy environment, where as the vibration data can be used when the acoustic cannot be the source of information. In general the fault diagnosis in bearing is done by comparing the mathematical interpreted data with vibration signal. This method can only be applicable to those system where the complete information about the ball bearing is known. But, this paper predict the fault in the ball bearing using acoustic and vibration signatures without knowing complete bearing information. Signal processing is used rather than using both signal processing and mathematical formulation all together to predict the fault in the bearing under different states. The signal analysis using FFT fails to analyse the signals of transient and non-stationary in nature. The extraction and analysis of the transient signal can be better done using Empirical Mode Decomposition (EMD) technique.