Research on digital system for identifying on-site operations of power infrastructure based on time-stamped measurements (TSM)

Abstract

The effective tracking and management of on-site operations in power infrastructure (PI) is critical for assuring operational reliability and avoiding service interruptions. Traditional monitoring systems are frequently constrained by human error, slow response times, and the inability to give real-time data on system conditions. This research uses a digital system to identify on-site operations of PI, utilizing Time-Stamped Measurements (TSM) to provide real-time monitoring and problem identification. The suggested system architecture is made up of both software and hardware parts that work together to collect, process, and evaluate data concurrently for fault detection and condition estimate in PI. Various types of sensors are installed throughout the power grid to collect data on operating characteristics such as current, temperature, voltage, and performance metrics. The collection of TSM is an important step in the proposed system since it provides precise, time-based data required for reliable problem identification and operational analysis. The sensors communicate TSM to IoT devices or gateways and employ communication protocols such as Wi-Fi to deliver the data to the main server or cloud server. Before the analysis, the raw data is pre-processed, utilizing data normalization and feature extraction using the Fast Fourier Transform (FFT). Intelligent Genetic Energy Valley Optimizer (IntGen-EVO) is used to detect defects in real time using time-stamped data. The framework demonstrates superior accuracy (98.7%), precision (98.3%), recall (98%), and F1-score (98.4%) compared to traditional methods, significantly enhancing fault detection in the PI system. The findings demonstrate the method provides accurate concurrent insights into the infrastructure’s operational state, allowing for preventive maintenance and rapid identification of problems. Therefore, the digital system described in this research provides an effective solution for increasing operational effectiveness and reliability in PI management.

Keywords

power infrastructure (PI)digital system time-stamped measurements (TSM)intelligent genetic energy valley optimizer (IntG-EVO)fault detection

Introduction

Power infrastructure (PI) is one of the most important resources for contemporary society. The significance of digitalization in power infrastructure has become increasingly apparent, particularly with the advent of smart technologies and IoT systems. These advancements provide real-time tracking, data analysis, and improved operational efficiencies, thereby ensuring that electrical energy remains reliable for critical societal functions. By harnessing these digital innovations, the primary objective of this research is to develop a cutting-edge digital system that facilitates real-time monitoring and fault detection in power infrastructure using Time-Stamped Measurements (TSM). This system not only aims to enhance operational reliability but also seeks to pre-emptively address potential service interruptions through proactive management. Broadcast exceeds maintaining high-voltage diffusion lines that transmit the bulk electrical energy across great distances from power generating locations to power distribution locations. These transmission tower collapses frequently cause significant economic losses as well as interruptions in the operation of other infrastructure systems.¹ Infrastructure has gotten smarter due to digitalization. The process of digitization has gradually improved the effective functioning of the power system, the proactive management of users, assets, and processes, and the best possible use of physical space and energy. Nations that have used digitalization in areas other than manufacturing like robotics and the Internet of Things (IoT).² PI supports electrical circuits with anticipated lifespans of 60–80 years and is made up of transmitters and hundreds of entity extremities. The infrastructure is subject to changing environmental conditions as it matures, leading to unanticipated degradation and jeopardizing the system’s security and dependability. The physical condition of PI must be regularly monitored for proper asset management. Trained field personnel perform visual inspections at regular intervals, visiting each pole in PI’s right-of-way.³ Figure 1 shows the classification of the PI system.

Figure 1.

PI system classification.

Nevertheless, conducting routine inspections is not economically viable. The budgets for the TSO are strained even by yearly inspections. Statistical methods for lifespan estimation are inadequate due to the lack of information from traditionally low-conduction transportation faults exacerbating this issue.⁴ A Health Index (HI) organization provided on a broader position of extremity situation is developed to assess the probability of quality failure in a specific instant structure. Several technical publications concerning PI state estimation and determination of significant faults concentrate on the recognition of fundamental failure. The method necessitates data gathering through the foot patrol inspections, even if it is created precise, predefined criteria lists for every PI component.⁵ Over the past 10 years, the usage of aerial vehicles, like drones or helicopters, for inspections has increased. Typically, cameras and Light Detection and Ranging (LiDAR) are used for aerial inspections to obtain 3D models, images, and geographic data, including the PI’s point cloud and right-of-way.⁶ The primary focus of such inspections is on the vegetation and transmitter’s earth consent examination rather than on evaluating the mechanical condition. Unfortunately, input regarding PI problems is typically lacking on the stage that foot rounds offer evidence. While line components are detected in entity recognition based on video and images from aerial analysis, no condition evaluation is carried out.⁷ The research objective is to develop a digital system for real-time monitoring and fault detection in PI using Time-Stamped Measurements (TSM). The aim is to improve the accuracy and consistency of operational insights, facilitating efficient fault identification, preventive maintenance, and improved overall management of the PI system.

The effective management and monitoring of PI are crucial for ensuring uninterrupted service and operational reliability. The main contributions of this work are:

(1) The objective is to develop a digital system for real-time monitoring and fault detection in PI, utilizing TSM to enhance operational reliability and identify issues promptly.

(2) Data is gathered from various sensors installed across the power grid to monitor operating features like current, voltage, temperature, and performance metrics.

(3) The data is pre-processed using normalization to standardize values and feature extraction using Fast Fourier Transform (FFT) to ensure the data.

(4) The research introduces the Intelligent Genetic Energy Valley Optimizer (IntGen-EVO) to execute concurrent defect detection and condition estimation on the pre-processed TSM data, providing accurate insights into the operational state of the PI.

(5) The system successfully delivers real-time monitoring and defect detection, offering precise insights into the PI operational state. This enables preventive maintenance and timely problem identification, improving the overall effectiveness and reliability of PI management.

The proposed system effectively improves operational monitoring and fault detection in PI, enabling timely maintenance and problem resolution. Its real-time capabilities significantly boost the efficiency and reliability of PI management.

Related work

This section provides a summary of previous research and methodologies in the field of PI monitoring and defect detection, focusing on the use of advanced technologies and highlighting the strengths and limitations of the existing approaches.

Research⁸ created a new Digital Twin (DT) structure to improve the Electric Power Network’s (EPN) hurricane resilience. The technique allows for real-time failure estimation by combining data-driven and physics-based models with a dynamic Bayesian network (DBN). Results demonstrated precise performance validation; nonetheless, scalability issues still exist.

The objective of the goal⁹ focused on cybersecurity in the electric influence industry and used a DT to create a cyber-resilience system for Critical Cyber Infrastructures (CCI). Minimizing response time and impact, the approach improved common understanding, situational awareness, and response capability. Limitations include possible problems with generalization and dependence on validation particular to the EU.

Research¹⁰ evaluated the Broadband Pilot Policy (BCPP) provided an exogenous shock to boost energy efficiency in developed countries that were more energy efficient than China. According to a difference-in-differences (DID) model, BCPP improved urban energy efficiency and was fueled by digital dividends, such as technology and talent. Targeted interventions were suggested by the fact that the effect was less pronounced in resource-based and historically developed cities.

The goal¹¹ described the physical, cyber, combination, and examination stages of a bridge’s DT implementation and provided an outline of DT infrastructures. Interoperability was emphasized by the findings and also highlighted the importance of data management in multidisciplinary designs. Scalability issues and misaligning organizational viewpoints were the limitation.

Research¹² investigated how China’s innovation gap was impacted by the digital infrastructure disparity. Provincial panel data from 2013 to 2018 was examined using the entropy weight approach. Research showed that while upgrading the industrial structure reduced the innovation gap, the split worsened it. Regional variances and the limited coverage of the data were the limitations.

Assessment¹³ investigated the detection and classification of Smart Grid (SG) faults, tackling important issues such as low-latency connectivity, data management, and real-time measurements. In addition, a thorough fault classification methodology for different SG applications was evaluated. It explored how monitoring and fault detection advance in the future. Among the drawbacks were insufficient data on developing SG technology.

Exploration¹⁴ addressed the problem of monitoring power transformers that were situated far from electrical plants by combining Deep Learning (DL) and an IoT approaches to identify malfunctions and cyber-attacks. With ±5% uncertainty, it obtained accuracy rates of 94.36% in normal situations and 92.58% in cyber-attacks using a 1D-conventional Neural Network (CNN). Potential vulnerability to high levels of uncertainty and a variety of attack kinds were among the limitations.

Research¹⁵ emphasized data-driven condition monitoring to lower Operations and Maintenance (O&M) expenses and improve wind turbine reliability. Twin Support Vector Machine (TWSVM) and adaptive threshold identify abnormalities in gearboxes. Scalability and real-time application were still limited, despite the demonstrated superior performance over other methods.

The goal¹⁶ suggested Industry 4.0 by identifying and addressing sensor irregularities for trustworthy DTs. A classifier and neural network estimators were used in an ML-based framework for data reconstruction and fault identification. Results demonstrated resilience to artificial errors when validated on three real-world datasets; nonetheless, environmental dependence and model complexity were drawbacks.

Research¹⁷ presented a Global System for Mobile Communications (GSM) based electric traction power supply examination and defense system. It makes difficulty recognition, wear examination of circuit breakers, and preservation alert probable through the use of complicated sensors and apparatus. It enhanced the constancy of railway infrastructure and it has been authorized in labs and field tests; nonetheless, it has restrictions in terms of scalability and concurrent receptiveness when dealing with large data loads.

The objective¹⁸ employed early ignition risk detection to stop wildfires brought by High Impedance Faults (HIFs). With the help of a sizable experimental dataset and features from time- and frequency-domain analysis, the approach showed excellent accuracy in anticipating dangers before materializing. However, depending on the circumstance, its performance was changed.

The goal¹⁹ suggested an ML-based microgrid fault detection system that would permit independent fault identification through an Intelligent Electronic Device (IED) without the required for dependable contact. When tested on a customized IEEE 13-node feeder, it is established to be precise, flexible, and easy to use. For outsized real-world requests, scalability and elevated prices present difficulties.

Probe²⁰ utilized DL to mechanize the assessment of physical extremity situations in electrical program systems. Nine diverse kinds of defects were found by testing three object detection systems and translating the outcomes to HI. Results were designated in reduced manual inspection efforts and good concert for serious flaws, while accuracy for small flaws was low.

Research²¹ suggested a cloud-based IoT solution that used a novel fault localization method to reduce the amount of time it takes to restore power in a Power Distribution System (PDS). Using a zone-based methodology, the system finds single or multiple problems as well as sensor failures. Results indicated that data reduction and fault localization were successful. One limitation is that complex systems face issues with sensor accuracy.

Assessment²² introduced an affordable Long Range Wide Area Network (LoRaWAN) based IoT platform to improve problem identification and detection in sub-Saharan African power distribution systems. The technology was in use in Nakuru, Kenya, detects faults in 100 milliseconds. Budgetary restrictions and the difficulty of modernizing legacy grids in poor nations were limitations.

Goal of the research²³ discussed high-resistance fault sensitivity and fixed thresholds were two issues in DC grid fault protection. Using disjoint-based bagging and Bayesian optimization for Ensemble Artificial Neural Networks (EANN), it suggested a unique fault detection method. According to simulation results, accuracy reaches 400 Ω. Reliance on offline data and the requirement to account for noise influence were among the limitations.

Probe²⁴ addressed the shortcomings of conventional approaches for fault detection in microgrids by proposing Brownboost, a sequential ensemble of intelligence-based techniques. Nonconvex optimization was used by the ensemble to make it resistant to overfitting. The Hilbert-Huang transform lessens the sensitivity to noise. Its efficiency was confirmed by the results, albeit it worked well in extremely noisy environments.

Research²⁵ examined the difficulties and methods electric companies confront in reducing the risk of wildfires, especially in the wildland-urban interface. Techniques including enhanced protective systems, vegetation management, and structural hardening were presented. Limitations included the need for more research on incorporating wildfire mitigation strategies into power systems and operational conflicts with reliability assumptions.

In power systems, protection algorithms were generally separated into AI-based techniques and traditional protective devices. Conventional methods have drawbacks such as signal distortion and fault resistance, resulting in imprecise fault location estimates. Faster power restoration and more precise fault identification were offered by AI algorithms. ANN, communication between protective devices, directional relays, and Multi-Agent Systems (MAS) have been employed in recent studies on protection in ring grids to identify problems. Nevertheless, the effects of Distributed Generation (DG)-based inverters were frequently ignored in these researches.²⁶ The research creates a digital system for real-time monitoring and fault detection in PI to improve operational reliability and quickly identify problems utilizing TSM.

Methodology

Research gathers sensor data that measures temperature, voltage, current, and performance metrics and then it is pre-processed using Min-Max normalization. The FFT is used to extract features from frequency-domain signals to detect faults. To improve operational dependability and the accuracy of real-time monitoring in PI, fault identification and condition estimation are optimized using the IntGen-EVO. Figure 2 shows the methodological flow.

Figure 2.

Methodological flow.

Data collection

The data is gathered from TSM through a variety of sensors positioned across the electrical grid to track operational parameters like voltage, current, temperature, and performance indicators. Every element contains a timestamp, sensor type, sensor identifier, and related measurements. The communication protocol and the IoT gateway ID used for data transmission are also recorded. The status code field shows the model’s existing condition of functioning. To enable preventative preservation and prompt issue identification, this data is essential for real-time monitoring, fault detection, and condition estimates in the power infrastructure.

Data preprocessing

The gathered data is pre-processed using Min-Max Normalization to ensure standardization across different measurement scales. Normalization rescales the input variables to a uniform range, which significantly contributes to reducing biases in the dataset and improving the overall model performance. This step is crucial, especially when integrating measurements from various sensors that may operate under different units or scales. Following normalization, the Fast Fourier Transform (FFT) is employed for feature extraction, converting time-domain signals into frequency-domain representations. The use of FFT enables the identification of dominant frequencies and harmonics related to fault conditions, thus enhancing the model’s ability to accurately predict and detect anomalies. By isolating significant patterns in the data while minimizing noise, the combination of normalization and FFT results in a robust preprocessing framework that underpins effective fault detection. This normalization scales features to the [0, 1] range using equation (1).

y_{j} = \frac{w_{j} - w_{\min j}}{w_{\max j} - w_{\max j}}

(1)

The features are $w_{\min j}$ minimum and $w_{\max j}$ maximum values, $w_{j}$ is the component of the vector $w$ ensuring all features contribute equally. Normalization transforms each data point to have a unit Euclidean length using equation (2).

y_{j} = \frac{w_{j}}{| | w | |}

(2)

where

y_{j}

normalize the vector

w

to a unit vector. Each component

y_{j}

represents the proportion of the

j^{t h}

component of the original vector

w

. The norm of the feature vector preserves the direction of the data points. These normalization techniques are essential for ensuring that features such as waste type, quantity, collection frequency, and data with features representing different waste characteristics thereby improves the performance of methods used for classification.

Feature extraction using FFT

The standardized data features are extracted using FFT. The output patterns are analyzed in the frequency range to produce a potent fault analysis technique. Different output signal parameters have been extracted using the FFT approach. The inability to measure output parameter signals is a key characteristic in classifying incorrect assumptions. FFT is used in the execution of a single likely approach with a Digital Signal Processing (DSP) microprocessor. Starting with the Discrete Fourier Transform in equation (3), the FFT technique is described in equations (4) and (5) using a combination of decimation in time decomposition algorithms.

F_{l} = \sum_{m = 0}^{M - 1} e_{m} Z_{M}^{m l} f o r l = 0, \dots ., M - 1

(3)

where

Z_{M} = f^{- i \frac{2 π}{M}}

F_{l} = H_{l} + Z_{M}^{l} G_{l} f o r l = 0,, \frac{M}{2} - 1

(4)

F_{l + \frac{M}{2}} = H_{l} + Z_{M}^{l} G_{l} f o r l = 0,, \frac{M}{2} - 1

(5)

The even-numbered elements of $e_{m}$ should be represented by $H_{l}$ , and the odd-numbered essentials of $e_{m}$ by $G_{l}$ . Equations (6) and (7) require that $H_{l}$ , and $G_{l}$ be evaluated as exposed.

H_{l} = \sum_{m = 0}^{\frac{M}{2} - 1} e_{2 m} Z_{\frac{M}{2}}^{m l}

(6)

G_{l} = \sum_{m = 0}^{\frac{M}{2} - 1} e_{2 m + 1} Z_{\frac{M}{2}}^{m l}

(7)

By transforming time-domain signals into frequency-domain signals, important patterns, such as dominant frequencies and harmonics are seen. Signal analysis applications benefit from its ability to streamline features, reduce noise, facilitate efficient computing, assist real-time defect detection, and improve model performance.

Fault detection using intelligent genetic energy valley optimizer (IntGen-EVO)

The IntGen-EVO integrates the advantages of EVO and IntGen algorithms for fault identification and condition estimation in PI.

Intelligent genetic (IntGen) optimization

The frequency-removed data is used for identifying Faults in PI through IntGen. The fitness ratio method and the best individual preservation strategy are frequently used by conventional genetic algorithms to choose individuals. Even though, this approach has a high chance of selecting suitable parents, it has random error means that people with great fitness are disqualified indicating a lack of competition. Consequently, this research suggests a novel approach to selection operations that efficiently chooses the best members of the population. Equation (8) used to calculate the cumulative probability for each individual.

O (y_{j}) = \frac{e (y_{j})}{\sum_{i = 1}^{M} e (y_{j})}

(8)

where

\sum_{i = 1}^{M} e (y_{j})

is the total of the selection probabilities of every individual in the population,

y_{j}

is the

j^{t h}

character in the residents, and

e (y_{j})

is the fitness value of

y_{j}

. To determine each person’s cumulative probability using equation (9).

r_{j} = \sum_{j = 1}^{j} e (y_{j})

(9)

where

\sum_{j = 1}^{j} e (y_{j})

is the total of the decision rates of every person before the

j^{t h}

candidate to produce a random number,

q

that falls between

0 a n d 1

. A single

y_{j}

is chosen if

r_{j} > q

. By exchanging their partial genes based on a predetermined crossover probability, two matched chromosomes create two new persons through the overlap function of the IntGen. The following is how the cross equations are applied in equation (10).

{\begin{cases} z_{n j} = z_{n j} (1 - a) + z_{m j} \times a \\ z_{m j} = z_{m j} (1 - a) + z_{n j} \times a \end{cases}

(10)

where

a

is a random number between

[0, 1]

, and

z_{n j}

and

z_{m j}

stand for the

j^{t h}

bit of the

n

and

m

genes, respectively. According to a specific mutation probability, the mutation operation of the IntGen is the procedure of creating an entirely novel person from the mutation of an individual selected at random from the population. These are the mutation equations displayed in equations (11) and (12).

z_{n m} = {\begin{cases} z_{n m} + (z_{n m} - z_{\max}) \times e (h), q > 0.5 \\ z_{m n} + (z_{m n} - z_{\min}) \times e (h), q \leq 0.5 \end{cases}

(11)

e (h) = q^{2} \times (1 - \frac{h}{H_{\max}})

(12)

where the present repetition number is

h

, the maximum iteration number is

H_{\max}

, the highest and lowest amount of the gene

z_{n m}

are represented by

z_{\max}

and

z_{\min}

, respectively, and

q

and

q^{2}

are arbitrary numbers between

[0, 1]

. By analyzing complex, dynamic data from PI effectively, it maximizes real-time defect detection. In the end, it enhances operational performance and decreases system downtime by improving system precision, adaptability, and reliability making precise issue diagnosis and predictive maintenance possible.

Energy valley optimization (EVO)

The fault identified data is optimized using EVO for better accuracy and reliability. The EVO method is presented as a method of optimization that draws inspiration from sophisticated physics concepts, especially those about stability and different particle decay modes. The EVO is primarily inspired by the basic ideas of the decay mechanism affecting additional particles in physics. Excess energy is released during decay processes when lower-energy particles are formed. The $N / Z$ ratio (the neutrons to protons), relations, and binding electricity all affect particle stability. Though larger atoms show durability with a greater $N / Z$ proportion, secure, tiny particles have $N / Z = 1$ . Patterns found in measurements help to comprehend particle decay. Determining criteria for stability utilizing $N, Z$ , and the $N / Z$ percentage is the main obstacle. One important component of electron durability is neutron enhancement. Further neutrons are needed for stabilization in neutron-rich atoms above durability deteriorates, whereas capture of electrons or positron release occurs in neutron-poor atoms. Decay releases extra energy and produces $α$ particle with less energy. Decay is controlled by three emissions: $γ$ , $β$ and $α$ particles. $α$ decay results in the loss of a $α$ particle and lowers $Z$ and $N$ values. A $β$ particle is ejected during $β$ decay, raising the $\frac{N}{Z}$ fraction. A $γ$ photon is released during $γ$ decay, although $N / Z$ values remain unchanged.

The initialization process takes place in the first phase, and treats solution candidates ( $Y_{j}$ ) as elements with diverse stages of constancy inside a specified exploration space region comparable to a certain area of the space using equations (13) and (14).

Y = [\begin{array}{l} \begin{array}{l} Y_{1} \\ Y_{2} \\ ⋮ \end{array} \\ Y_{j} \\ ⋮ \\ Y_{m} \end{array}] = [\begin{array}{c} \begin{array}{c} Y_{1}^{1} & Y_{1}^{2} & \dots \end{array} \begin{array}{l} Y_{1}^{i} & \dots & Y_{1}^{c} \end{array} \\ \begin{array}{l} Y_{2}^{1} & Y_{2}^{2} & \dots \end{array} \begin{array}{l} Y_{2}^{i} & \dots & Y_{2}^{c} \end{array} \\ \dots ⋱ ⋮ \\ \dots \\ \begin{array}{l} Y_{j}^{1} & Y_{j}^{2} & \dots \end{array} \begin{array}{l} Y_{j}^{i} & \dots & Y_{j}^{c} \end{array} \\ \dots \dots \\ \dots \\ \begin{array}{l} Y_{m}^{1} & Y_{m}^{2} & \dots \end{array} \begin{array}{l} Y_{m}^{i} & \dots & Y_{m}^{c} \end{array} \end{array}], {\begin{cases} j = 1, 2, \dots . ., m . \\ i = 1, 2, \dots . ., n . \end{cases}

(13)

Y_{j}^{i} = Y_{j, \min}^{i} + r a n d \cdot (Y_{j, \max}^{i} - Y_{j, \min}^{i}), {\begin{cases} j = 1, 2, \dots . ., m . \\ i = 1, 2, \dots . ., n . \end{cases}

(14)

where the variables

m

c

, and

Y_{j}

denote the total amount of particles that serve as remedy applicants in the universe, indicate the complexity of the specified difficulty, and, respectively, correspond to the

i^{t h}

selection features that determine the

j^{t h}

candidate’s original location. Additionally, the

j^{t h}

aspirant’s maximum and minimum limits for the

j^{t h}

variable are indicated by

Y_{j} Y_{j, \min}^{i}

and

Y_{j, \max}^{i}

. Random numbers in the interval

[0, 1]

that are uniformly distributed are referred to as

r a n d

. The Enrichment Bound (EB) for electrons is calculated in the instant period of the procedure, accounting for the distinctions among neutron-rich and neutron-poor atoms. In this procedure, each particle’s objective function measured by the particles’ Neutron Enrichment Level (

N E L

) is evaluated. The following equation (15) describes how these processes are represented mathematically.

F A = \frac{\sum_{j = 1}^{m} {N E L}_{j}}{m}, j = 1, 2, \dots ., m

(15)

where

{N E L}_{j}

stands for the

j^{t h}

particle’s NEL. Moving on to the third phase, the following equation (16) is used to ascertain the particles’ stability levels presented on the intent role assessments.

{T K}_{j} = \frac{{N E L}_{j} - A T}{Z T - A T}, j = 1, 2, \dots ., m

(16)

where

A T

and

Z T

and for the atoms with the finest and worst constancy stages in the creation, and

{T K}_{j}

represents the constancy stage of the

j^{t h}

atom. Latent

γ

β

α

decay is taken into consideration in the EVO’s primary exploration round if an element’s

{N E L}_{j}

beats EB, suggesting a larger

N / Z

ratio. The Stability Bound (SB) is represented by a randomly generated number in

[0, 1]

γ

and

α

decay is likely to transpire if an atom’s constancy stage results in Control and Optimization surpass SB. In EVO, this procedure is represented mathematically as a position-updating method.

In $[1, c]$ , the secretion of energy in the resolution nominee is guided by two random numbers, $α$ Index I and II. Energy from the atom takes the place of these released rays. The outline of the mathematical representation is given in equation (17).

Y_{j}^{N e w 1} = Y_{j} (Y_{A T} (Y_{j}^{i})), {\begin{cases} j = 1, 2, \dots ., m \\ i = α I I . \end{cases}

(17)

where the highest stability level particle,

Y_{j}

the current vector particle, and the

i^{t h}

emitted ray, respectively.

Y_{j}^{N e w 1}

Y_{A T}

, and

Y_{j}^{i}

are the new vector particles. The

γ

decay and

γ

rays provide a situational updating system within the EVO that helps stabilize excited particles. This comprises the creation of two accidental numerals

γ

Index indicates the number of photons released and the photons to take into account. To simulate relations with previous elements, a Neighboring Particle (XNg) is used in place of these photons which are considered as choice variables. With coordinates

(X_{2} - X_{1})

and

(Y_{2} - Y_{1})

, the sum of space

C_{j}^{l}

among the specified

l^{t h}

and

j^{t h}

atoms is intended to find the closest particle available for substitution using equation (18).

C_{j}^{l} = \sqrt{{(Y_{2} - Y_{1})}^{2} + {(X_{2} - X_{1})}^{2},} {\begin{cases} j = 1, 2, \dots ., m \\ l = 1, 2, \dots, m - 1 \end{cases}

(18)

Mimicking a dynamic search for ideal solutions in complicated systems improves fault identification. EVO enhances the identification process by efficiently navigating the energy globe, improving the accuracy of operational issue detection, and ensuring speedier, more precise status evaluations for PI.

IntGen-EVO integrates the advantages of IntGen and EVO for fault identification and condition estimation in PI. By constantly modifying search parameters, IntGen-EVO effectively searches the system’s parameter space to find problems, ensuring precise real-time fault detection. By boosting the analysis of time-stamped sensor data, it also improves condition estimates, enhancing the system’s capacity to anticipate operating states and identify anomalies early. This approach improves decision-making and lowers feature extraction errors eventually resulting in increased system reliability, quicker problem-solving, and more efficient predictive maintenance. Algorithm 1 shows the IntGen-EVO method.

Algorithm 1: Intelligent Genetic Energy Valley Optimizer (IntGen-EVO)

Result

This research uses Python 3.10, to assess the effectiveness of the suggested IntGen-EVO approach for PI failure detection. The performance is estimated employing different metrics such as recall, precision, F1-score, accuracy interruption frequency, reliability, and ROC analysis served as the foundation for the assessment. The IntGen-EVO compared with conventional techniques like Decision Trees (DTs),²⁷ Deep Neural Networks (DNNs),³⁰ Deep Graph Neural Networks with Multi-Layer Perceptron (DGNN-MLP),²⁹ and Gated Recurrent Units (GRUs).²⁸

Accuracy

A statistical metric called accuracy is utilized to evaluate how a detection model performs. It shows the percentage of accurately predicted cases of True Positive (TP) and True Negative (TN) among all the instances in the data. It shows the frequency of accurate model predictions. Table 1 and Figure 3 show the evaluation of accuracy. Accuracy is calculated using equation (19).

a c c u r a c y = \frac{T N + T P}{T N + F N + T p + T N}

(19)

Table 1.

Estimation of accuracy values.

Methods	Accuracy (%)
DT²⁷	97
GRU²⁸	92.13
DGNN-MLP²⁹	82
IntGen-EVO [proposed]	98.7

Figure 3.

Evaluation of accuracy.

With an accuracy of 98.7%, the IntelGen-EVO offers better performance than the other techniques. The results show that it outperforms DGNN-MLP at 82%, GRU at 92.13%, and DT at 97%. As demonstrated by these outcomes, IntGen-EVO is a viable method for applications requiring accurate identification due to its ability to achieve high predicted accuracy.

Precision

It quantifies the proportion of positively anticipated cases that are positively predicted. A model with high precision is assumed to have a low False Positive (FP) rate, which makes it crucial in applications where the expense of FP is substantial. Figure 4 and Table 2 demonstrate the outcomes of precision. Precision is calculated using the equation (20).

p r e c i s i o n = \frac{T P}{T P + F P}

(20)

Figure 4.

Precision outcomes.

Table 2.

Assessment of precision values.

Methods	Precision (%)
DT²⁷	98
DNN³⁰	93
DGNN-MLP²⁹	87
IntGen-EVO [proposed]	98.3

The precise comparison of the various approaches reveals that the suggested IntGen-EVO strategy performs better than the DT (98%), DNN (93%), and DGNN-MLP (87%), with a precision of 98.3%. While the notable improvements over DNN and DGNN-MLP highlight its resilience in managing complicated data, the slight improvement over the DT shows how effective the suggested improvements are. These results show the potential of the IntGen-EVO technology for high-accuracy applications, making the most precise approach.

Recall

The model’s capacity to recognize every pertinent occurrence of the positive class is measured. Recall shows the degree to which the model identifies defects in the context of PI. Table 3 and Figure 5 display the evaluation of recall. Recall is calculated using equation (21).

r e c a l l = \frac{T P}{F N + T P}

(21)

Table 3.

Outcomes of recall.

Methods	Recall (%)
DT²⁷	97
DNN³⁰	90
DGNN-MLP²⁹	82
IntGen-EVO [proposed]	98

Figure 5.

Estimation of recall.

The results demonstrate that the suggested IntGen-EVO outperformed the DNN at 90%, the DGNN-MLP at 82%, and the DT at 97%, with a recall of 98%. These findings show how reliable and successful IntGen-EVO is at increasing recall, which is essential for reducing FN. This enhancement represents a step toward improving fault detection and ensuring precise forecasts in the relevant field.

F1-score

It is used to measure a recognition system’s performance. It measures the statistical significance that combines recall and precision. It is helpful in cases of imbalanced data when a class has a higher prevalence than others. Table 4 and Figure 6 demonstrate the estimation of the F1-score. The F1-score is calculated using equation (22).

F 1 S c o r e = 2 \times \frac{T P}{2 T P + F P + F N}

(22)

Table 4.

Determination of F1-score.

Methods	F1-score (%)
DT²⁷	98
DNN³⁰	91.5
DGNN-MLP²⁹	81
IntGen-EVO [proposed]	98.4

Figure 6.

Evaluation of F1-score.

With an F1-Score of 98.4%, the IntGen-EVO method performed better than the current methods, surpassing DT (98%), DNN (91.5%), and DGNN-MLP (81%). While the notable difference between DNN and DGNN-MLP emphasizes IntGen-EVO’s resilience in managing intricate patterns, the slight improvement over DT shows its improved optimization capacity. These findings confirm that the suggested approach is successful in producing accurate and consistent results in the assessed task.

Interruption frequency analysis

This finds patterns in system disruptions and detects irregularities or anomalies through the analysis of frequency data. Raw data is pre-processed by standardization and emphasizing fault patterns. This improved the IntGen-EVO algorithm’s detection precision and reliability, organized data inputs that ensured precise fault detection and condition estimates. Figure 7 shows the outcome of the frequency analysis.

Figure 7.

Analysis of interruption frequency.

The examination of interruption frequency shows that performance steadily improves with an increase in sample size. With increasing sample sizes, lower interruption frequencies and improved grid reliability, as evidenced by the reliability ratio’s progression from 73% to over 99%. This demonstrates its ability to monitor and minimize interruptions over time.

Reliability analysis

It assesses a system’s capacity to function consistently throughout time. It examines persistent problems and identifies weak points in the PI to provide a consistent and unbroken power supply. It comprises a comprehensive assessment of the PI dependability while taking various factors into account. The reliability of the PI’s ability to deliver electricity continuously and uninterruptedly must be carefully assessed. The model identifies weak spots in the grid and learns about recurring issues. Figure 8 demonstrates the PI reliability analysis.

Figure 8.

Reliability analysis of the PI.

The reliability analysis shows that the PI reliability ratio continuously improves as the number of samples rises. This pattern implies that the technique improves with bigger sample sizes, underscoring its potential to provide increased reliability in power grid analysis as more data is used for assessment.

ROC analysis

It is employed to evaluate a model’s capacity for identification and design the TP Rate about the FP Rate across thresholds. The ROC curve uses an AUC value to assess the IntGen-EVO model’s predicting performance for time-series forecasting system failures. In Figure 9, the anticipated AUC value is shown.

Figure 9.

Evaluation of ROC outcome.

The ROC curve shows how well the IntGen-EVO algorithm performs in fault identification by classifying flaws across various thresholds. The system has exceptional prediction accuracy, successfully reducing FP, with a curve in the top-left corner and an AUC of 0.90. This dramatic increase in decreased FP rates demonstrates the algorithm’s strong decision-making abilities and validates the accuracy and dependability of the suggested optimization framework in identifying system defects.

Discussion

Research aimed to improve real-time monitoring and fault detection in PI using TSM. The proposed method was compared with conventional techniques like DT,²⁷ DNN,³⁰ DGNN-MLP,²⁹ and GRU.²⁸ This addressed some limitations, like slow response times, overfitting, and inefficiency were problems that traditional approaches faced when working with complicated, real-time data or huge datasets. For instance, DTs²⁷ struggle with high-dimensional data, whereas DNNs³⁰ require large amounts of labeled data and processing power. GRUs²⁸ frequently failed to retain long-term memory across lengthy sequences, while DGNN-MLP²⁹ models were computationally costly, despite their effectiveness for graph-based data. The system addressed the drawbacks of conventional methods in large-scale PI management by leveraging Wi-Fi communication protocols and employing Fast FFT for feature extraction. This ensured quick data transmission and examination, improving operational effectiveness, reducing dependence on huge data, and offering accurate, sensible insights for defensive protection. Despite the promising results achieved through the IntGen-EVO approach, some limitations need to be acknowledged. The methodology’s reliance on specific sensor types could impact consistency across diverse power infrastructure environments. Additionally, while the method demonstrated strong performance metrics, its computational complexity may pose challenges when implemented in real-time systems with extensive data inputs. Directions for future research could include exploring alternative optimization techniques that may streamline processing times and enhance scalability. Investigating hybrid models that combine IntGen-EVO with other machine learning algorithms could also reveal synergies that improve accuracy and reliability in fault detection. These considerations will be essential for developing adaptive systems capable of addressing the evolving needs of power infrastructure management.

Conclusion

Monitoring and problem identification were necessary for PI to be reliable, efficient, and safe. Early fault detection made probable by concurrent monitoring lowers safeguarding costs and downtime. Advanced technologies like ML and IoT sensors were utilized to enlarge infrastructure flexibility, decreasing malfunctions and enhancing overall operational efficiency. The objective was to create an automated method that precisely monitors and detects PI faults using TSM and complicated approaches, like the IntGen and EVO. This research highlighted how significant it was to monitor and analyze defects in PI concurrent to reduce facility disruption and ensure operational consistency. Based on performance criteria like precision (98.3%), accuracy (98.7%), recall (98%), and F1-score (98.4%), the outcomes demonstrated that the IntGen-EVO approach performed superior to traditional methods. The examination of interruption frequency and consistency measures also established improved grid reliability and proficient fault identification. The research studies have certain drawbacks, such as its reliance on particular sensor types and communication practices that do not work in every PI system. Expanding the data for wider simplification and including more complicated defect detection systems also develops the method’s performance. The system’s scalability and flexibility to diverse PI settings should be optimized in future research, along with the combination of analytical protection techniques and the examinations of ML models’ possibility for more definite and immediate defect detection in PI organization.

Statements and declarations

Footnotes

Conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the science and technology project of State Grid Shandong Electric Construction Company of China: “Deepening Application Research of the whole Process of Planning and Construction based on 3D Design Results - Research on digital intelligence construction technology based on 3D design results” (No. 520632220001).

ORCID iD

Rui Li

References

Khazaali

Bocchini

. Component-based fragility analysis of transmission towers subjected to hurricane wind load. Eng Struct 2021; 242: 112586. DOI: 10.1016/j.engstruct.2021.112586.

Manninen

Ramlal

Singh

, et al. Multi-stage deep learning networks for automated assessment of electricity transmission infrastructure using fly-by images. Elec Power Syst Res 2022; 209: 107948. DOI: 10.1016/j.epsr.2022.107948.

Shihavuddin

Rashid

Maruf

, et al. Image-based surface damage detection of renewable energy installations using a unified deep learning approach. Energy Rep 2021; 7: 4566–4576. DOI: 10.1016/j.egyr.2021.07.045.

Odo

McKenna

Flynn

, et al. Aerial image analysis using deep learning for electrical overhead line network asset management. IEEE Access 2021; 9: 146281–146295. DOI: 10.1109/ACCESS.2021.3123158.

Siddiqui

Park

. A drone-based transmission line components inspection system with deep learning technique. Energies 2020; 13(13): 3348. DOI: 10.3390/en13133348.

Djenouri

Belhadi

Srivastava

, et al. Fast and accurate deep learning framework for secure fault diagnosis in the industrial internet of things. IEEE Internet Things J 2021; 10(4): 2802–2810. DOI: 10.1109/JIOT.2021.3092275.

Zhang

Wang

Song

, et al. An edge visual incremental perception framework based on deep semi-supervised learning for monitoring power transmission lines. CSEE J Power Energy Sys 2023; 9(2): 759–768. DOI: 10.17775/CSEEJPES.2022.03120.

Braik

Koliou

. A novel digital twin framework of electric power infrastructure systems subjected to hurricanes. Int J Disaster Risk Reduct 2023; 97: 104020. DOI: 10.1016/j.ijdrr.2023.104020.

Salvi

Spagnoletti

Noori

. Cyber-resilience of critical cyber infrastructures: integrating digital twins in the electric power ecosystem. Comput Secur 2022; 112: 102507. DOI: 10.1016/j.cose.2021.102507.

10.

Peng

. Internet infrastructure, digital development and urban energy efficiency. J Digital Econ 2024; 3: 62–74. DOI: 10.1016/j.jdec.2024.11.001.

11.

Broo

Bravo-Haro

Schooling

. Design and implementation of a smart infrastructure digital twin. Autom ConStruct 2022; 136: 104171. DOI: 10.1016/j.autcon.2022.104171.

12.

Wang

. Digital infrastructure and innovation: digital divide or digital dividend? J Innov Knowl 2024; 9(3): 100542. DOI: 10.1016/j.jik.2024.100542.

13.

Rivas

Abrao

. Faults in smart grid systems: monitoring, detection and classification. Elec Power Syst Res 2020; 189: 106602. DOI: 10.1016/j.epsr.2020.106602.

14.

Elsisi

Tran

Mahmoud

, et al. Effective IoT-based deep learning platform for online fault diagnosis of power transformers against cyberattacks and data uncertainties. Measurement 2022; 190: 110686. DOI: 10.1016/j.measurement.2021.110686.

15.

Dhiman

Deb

Muyeen

, et al. Wind turbine gearbox anomaly detection based on adaptive threshold and twin support vector machines. IEEE Trans Energy Convers 2021; 36(4): 3462–3469. DOI: 10.1109/TEC.2021.3075897.

16.

Darvishi

Ciuonzo

Eide

, et al. Sensor-fault detection, isolation and accommodation for digital twins via modular data-driven architecture. IEEE Sens J 2020; 21(4): 4827–4838. DOI: 10.1109/JSEN.2020.3029459.

17.

Andrusca

Adam

Dragomir

, et al. Innovative integrated solution for monitoring and protection of power supply system from railway infrastructure. Sensors 2021; 21(23): 7858. DOI: 10.3390/s21237858.

18.

Kandanaarachchi

Anantharama

Munoz

. Early detection of vegetation ignition due to powerline faults. IEEE Trans Power Deliv 2020; 36(3): 1324–1334. DOI: 10.1109/TPWRD.2020.3006553.

19.

Cepeda

Orozco-Henao

Percybrooks

, et al. Intelligent fault detection system for microgrids. Energies 2020; 13(5): 1223. DOI: 10.3390/en13051223.

20.

Manninen

Ramlal

Singh

, et al. Toward automatic condition assessment of high-voltage transmission infrastructure using deep learning techniques. Int J Electr Power Energy Syst 2021; 128: 106726. DOI: 10.1016/j.ijepes.2020.106726.

21.

Ulasyar

Khattak

, et al. Cloud based IoT solution for fault detection and localization in power distribution systems. Energies 2020; 13(11): 2686. DOI: 10.3390/en13112686.

22.

Odongo

Musabe

Hanyurwimfura

, et al. An efficient LoRa-enabled smart fault detection and monitoring platform for the power distribution system using self-powered IoT devices. IEEE Access 2022; 10: 73403–73420. DOI: 10.1109/ACCESS.2022.3189002.

23.

Yousaf

Singh

Khalid

, et al. Enhancing HVDC transmission line fault detection using disjoint bagging and Bayesian optimization with artificial neural networks and scientometric insights. Sci Rep 2024; 14(1): 23610. DOI: 10.1038/s41598-024-74300-z.

24.

Azizi

Seker

. Microgrid fault detection and classification based on the boosting ensemble method with the Hilbert-Huang transform. IEEE Trans Power Deliv 2021; 37(3): 2289–2300. DOI: 10.1109/TPWRD.2021.3109023.

25.

Muhs

Parvania

Shahidehpour

. Wildfire risk mitigation: a paradigm shift in power systems planning and operation. IEEE Open J Power Energy 2020; 7: 366–375. DOI: 10.1109/OAJPE.2020.3030023.

26.

Alhanaf

Farsadi

Balik

. Fault detection and classification in ring power system with DG penetration using hybrid CNN-LSTM. IEEE Access 2024; 12: 59953–59975. DOI: 10.1109/ACCESS.2024.3394166.

27.

Khan

Asad

Vaimann

, et al. Improved Fault classification and localization in power transmission networks using VAE-generated synthetic data and machine learning algorithms. Machines 2023; 11(10): 963. DOI: 10.3390/machines11100963.

28.

Almasoudi

. Grid distribution fault occurrence and remedial measures prediction/forecasting through different deep learning neural networks by using real time data from Tabuk city power grid. Energies 2023; 16(3): 1026. DOI: 10.3390/en16031026.

29.

Ukwuoma

Cai

Bamisile

, et al. Power transmission system’s fault location, detection, and classification: pay close attention to transmission nodes. Int J Electr Power Energy Syst 2024; 156: 109771. DOI: 10.1016/j.ijepes.2023.109771.

30.

Govea

Gaibor-Naranjo

Villegas-Ch

. Transforming cybersecurity into critical energy infrastructure: a study on the effectiveness of artificial intelligence. Systems 2024; 12(5): 165. DOI: 10.3390/systems12050165.