A Customised LSTM-Based Deep Learning Framework for Transformer Predictive Maintenance: Performance Analysis | Chapter 5 | New Horizons of Science, Technology and Culture Vol. 9

 

Transformers are critical and costly components of power systems whose health deteriorates over time due to factors such as poor cooling and heavy loading. Consequently, predictive maintenance is emerging as an effective alternative to conventional corrective maintenance, enabling continuous monitoring and early fault detection.

 

To enhance the effectiveness of predictive maintenance for power transformers under limited Dissolved Gas Analysis (DGA) data conditions, this study proposes a customised Long Short-Term Memory (C-LSTM) deep learning model. The developed C-LSTM architecture is specifically designed to address the limitations of conventional LSTM networks, which often exhibit higher classification error rates when trained on small datasets and may underperform compared to traditional machine learning approaches.

 

A comprehensive performance evaluation was conducted by comparing the proposed C-LSTM model with several well-established traditional machine learning algorithms using multiple metrics, including validation accuracy, test accuracy, precision, recall, and F1-score. Additionally, the diagnostic capability of the model was rigorously assessed across seven transformer fault categories, including low- and high-energy discharges, partial discharge, electrical and thermal faults, and low-, medium-, and high-temperature thermal faults.

 

The experimental results demonstrate the superior classification and diagnostic performance of the proposed C-LSTM model, achieving a validation accuracy of 100% and a test accuracy of 98.57%, significantly outperforming conventional machine learning techniques. These findings confirm that the proposed C-LSTM framework offers a robust and reliable solution for transformer fault diagnosis and predictive maintenance, particularly in scenarios characterised by scarce DGA datasets.

 

 

Author(s) Details

G.V.S.S.N. Srirama Sarma
Department of Electrical and Electronics Engineering, Matrusri Engineering College, Saidabad, Hyderabad, India.

 

Please see the book here :- https://doi.org/10.9734/bpi/nhstc/v9/6804

 

Using Run-of-mine Tonnage and Grade to Predict Cost and Categorize Gold Mines: Analysing the Application of Logistic Regression Analysis and ANN |Chapter 3 | Current Approaches in Engineering Research and Technology Vol. 5

 

The increase in demand for gold in the world over the past three decades has witnessed an increase in the gold price to accommodate more efforts of mining and exploration in the gold mining industry. Categorization of gold mines based on rom-tonnage and grade would be more useful if it can statistically be used to predict the cost of gold mining. This paper focuses on establishing logistic and ANN models of the cost effect of run-of-mine tonnage and grade in the justification of the categorization of gold mines. Gold mines could be categorized based on the abilities of rom-grade and rom-tonnage to predict cost. The data used in the generation of the logistic and ANN models were cash-cost as the dependent variable vs. rom-grade, and rom-tonnage as independent variables together with the type of mine. The data were obtained from 160 gold mines selected from the top 20 gold-rich countries in the period of 7 years from 2002 to 2008. The first analysis was the Logistic Regression Analysis which was carried out in the SPSS software to determine the probability of occurrence of low cost given rom-grade and rom-tonnage for either an open pit, underground, or both mines together. The second analysis was based on ANN which was carried out to develop the ANN model. Using a Multilayer perceptron neural network and backpropagation, the ANN Model was trained to predict the cost. The main results for both logistic and ANN Models indicated that only rom-grade with a cut-off value of 5.385 g/t can be used to categorize gold mines as low and high grade while there was not enough evidence to categorize gold mines based on their rom-tonnage. The only evidence provided by ANN indicated that the normalized importance of rom-grade was 100% while for rom tonnage was 48.65. The full models established in this study have a percentage correct of 62.9 for logistic and 64 for ANN compared to 57.9 by guess. The relationship between cost vs. rom-grade and rom-tonnage indicated that only 6.9 percent of the cost is accounted for by rom-grade and rom-tonnage. This is a weak relation indicating that rom-grade and rom-tonnage are not the only determinants and therefore on their own must be used with precaution. Validation of the model agrees well with the actual results. Future research should focus on the other determinants which account for the remaining unaccounted 37.6%.

 

 

Author(s) Details

Karim Rajabu Baruti

School of Mines and Geosciences, University of Dar es Salaam, Tanzania.

 

Please see the book here:- https://doi.org/10.9734/bpi/caert/v5/5929E

Machine Learning in Water Management | Chapter 3 | Science and Technology - Recent Updates and Future Prospects Vol. 1

 

Water management is a major issue already addressed in most international forums. Water harvesting and recycling are critical criteria for meeting the per capita availability of water (Krishna et.al., 2008). In this regard, we must focus on water management approaches that can be easily implemented across a wide range of applications (Benos et.al., 2021). Amid population increase and varied challenges, there is an urgent need to establish intelligent water management mechanisms for the effective distribution, conservation, and maintenance of water (Safder et.al., 2022). The present work highlights a few important application areas that are essential for precision water management in which artificial neural network (ANN), recurrent neural network (RNN) and random Forest (RF) are some of the most useful developments in machine learning (ML) models (Mokhtari et.al.,2020) in different aspects of water such as wastewater recycling, water distribution, rainfall estimation and irrigation water management that can be used to predict the future scenario. As a result, there is an urgent need to generate datasets and models/algorithms that can be used to deliver solutions for the above-mentioned applications. Machine learning architecture can aid in the development of a smart model for the sustainable use of natural resources (Lowe et.al.,2022), as well as the usage of AI/ML in conjunction with different neural network models and simple statistical analysis to create an effective water management framework to deal all water-related problems.

Author(s) Details:

Aditya V Machnoor,
Discipline of Water Science and Technology, Water Technology Centre, ICAR-Indian Agricultural Research Institute, New Delhi, 110012, India.

Ajayakumar
Department of Agricultural Economics, University of Agricultural Sciences, Bengaluru-560065, Karnataka, India.

Mallanna Malagatti
Discipline of Seed Science and Technology, ICAR-Indian Agricultural Research Institute, New Delhi 110012, India.


Please see the link here: https://stm.bookpi.org/STRUFP-V1/article/view/14325

Simulation of Dye Synthesized Solar Cell Using Artificial Neural Network: Brief Overview | Chapter 6 | Emerging Trends in Engineering Research and Technology Vol. 1

 The basic goal of present test is to foresee continually worldwide battery powered by the sun efficiency in view of of or in the atmosphere factors, handling distinctive counterfeit neural method (ANN) procedures. In the examination we report the impact of Dye Synthesized solar cell. A three-coating artificial interconnected system (ANN) model was developed to predict the effectiveness of Dye Synthesized solar cell established 100 experimental sets. In the present test we report the impact of Dye Synthesized solar cell. The effect of functional parameters such as avoid current (Jsc), Open circuit energized matter (Voc), Fill factor (FF) were intentional to optimize the conditions to check the adeptness of Dye Synthesized solar cell. Experimental results revealed that the ANN model was able to predict adsorption effectiveness with a touching sigmoid transfer function (tansig) at hidden coating with 20 neurons and a uninterrupted transfer function (purelin) at output layer [1]. The Levenberg–Marquardt treasure (LMA) was used accompanying a minimum mean squared error (MSE) of 0.00350141. The uninterrupted regression middle from two points the network outputs and the corresponding targets were explained to be acceptable with a correlation cooperative of about 0.9993 for six model variables used in this place study.

Author(s) Details:

S. K. Kharade,
Department of Mathematics, Shivaji University, Kolhapur, India.

K. G. Kharade,
Department of Computer Science, Shivaji University, Kolhapur, India.

S. V. Katkar,
Department of Computer Science, Shivaji University, Kolhapur, India.

R. K. Kamat,
Department of Computer Science, Shivaji University, Kolhapur, India.

Please see the link here: http://bp.bookpi.org/index.php/bpi/catalog/book/147

Model Development of Analogue Wavelet Transform Based the Predicted Imaginary Part of the Dynamics of Rational Map Having Zeros | Chapter 02 | Recent Advances in Mathematical Research and Computer Science Vol. 7

Quadratic polynomial dynamics have already attracted a lot of attention. The kinetics and attributes of rational functions are both fascinating. In this chapter, we look at Witten's [1] solution to the classical Yang-Mills problem and use a C++ computer programme to generate computer graphics. We next develop an artificial neural network model based on RMS kind of error from two samples of points extracted from the created images using predictive modelling software. The imaginary component of sample II might be predicted by feeding the real sections of sample I and sample II [2] to the artificial neural network. The real part of sample II is more essential than the actual part of sample I in anticipating the imaginary component of sample II. The projected imaginary component of Sample II is then loaded via Matlab workspace into Matlab Signal Processing Tool (SPTool). To remove noise from the model for analysis, we use a reliable band pass filter. The methodology can be used to explore the attributes of computer generated images formed from the generated wavelet by producing a modulated signal. The projected imaginary component of sample II is then imported into autoSIGNAL software for continuous wavelet transform time-frequency analysis. The goal of this research is to close the gap between the dynamics of rational maps and the dynamics of electric power systems [3,4].


Author(S) Details

Jean-Bosco Mugiraneza
African Centre of Excellence in Energy for Sustainable Development (ACEESD) , University of Rwanda (UR), Kigali, Rwanda.

View Book:- https://stm.bookpi.org/RAMRCS-V7/article/view/5487

A Statistical Analysis and Artificial Neural Network Behavior on Wind Speed Prediction: Case Study | Chapter 4 | Theory and Practices of Mathematics and Computer Science Vol. 6

The quest for renewable and emission-free energy sources has been encouraged by the increased usage of energy and the decline of fossil fuel supplies, along with an increase in environmental pollution. Wind energy is one of these. In the last few years, the wind power industry has seen exponential growth. The increase in wind turbine orders has resulted in a manufacturer's market. This disparity in the market, the relative immaturity of the wind industry and the rapid evolution of data processing technologies have provided opportunities to enhance the efficiency of wind farms and to change the myths surrounding their operations. This study provides data-driven modeling, a new paradigm for the wind power industry. For several parameters, each wind mast produces extensive data, recorded as frequently as every minute. Since the predictive performance approach is new to the wind industry, it is important to build a viable road map for study. This paper proposes a long-term wind forecasting (ANN) predictive analysis and data-mining approach, which is ideal for dealing with broad real-world datasets. The paper provides a case study focused on a real database of five years of wind speed data for a location and addresses wind power density results calculated using the probability density functions of Weibull and Rayleigh. Wind speed predicted using wind speed data using intelligent technologies such as Artificial Neural Networks with Datamining methodology (ANN). The MATLAB R2008a Neural Network Toolbox is designed to measure the monthly and annual mean wind speed for the training of the ANN back propagation algorithm and the PROLOG software. The statistical analysis of wind speed prediction shows that the distribution of Weibull is more acceptable than the distribution of Rayleigh and we can infer that higher values of k mean a sharper limit in the frequency distribution curve and thus a lower density of wind power by seeing the values of k.

Aurthor(s) Details:

K. Mahesh
Department of Electrical and Electronics Engineering Sir M Visvesvaraya Institute of Technology, Bengaluru, India.

View Book :- https://stm.bookpi.org/TPMCS-V6/issue/view/6

An Intelligent Way of Modeling and Simulation of WO3 for Supercapacitor | Chapter 11 | Recent Studies in Mathematics and Computer Science Vol. 3

 In this investigation we studied the different electro-chromic properties of W03 thin films that relates on artificial

neural network. When electro-chromic energy storage devices stores energy for the changes colour, this is useful
in buildings as well as automobiles. Tungsten oxide is also known as tungsten anhydride (WO3). This material is
a combination of oxygen and tungsten. It is oxidative agent. Tungsten oxide or tungsten trioxide is used for
development of many things which are used in daily life for e.g. energy storage, gas sensors, smart window,
photocatalysis etc. The research investigation studies the WO3 thin film for supercapacitor. The simulation
process carried out in MATLAB here researchers checks the results generated by neural network and then that
results are matches with experimental results. For finding optimized supercapacitor we have calculated the error
which found at different hidden neurons in artificial neural network. In the conclusion of this study confirms
that ANN is appropriate tool for modelling of WO3 thin film for supercapacitor.

Author(s)Details

S. V. Katkar
Department of Computer Science, Shivaji University, Kolhapur, India.

K. G. Kharade
Department of Computer Science, Shivaji University, Kolhapur, India.

S. K. Kharade
Department of Mathematics, Shivaji University, Kolhapur, India.

R. K. Kamat
Department of Electronics, Shivaji University, Kolhapur, India.

View Book :- http://bp.bookpi.org/index.php/bpi/catalog/book/233