Email Alert
Abstract
The rapid development of the power system requires high reliability and real-time situational awareness of power equipment. The current agent-based condition-monitoring perception mode is not suitable for widely distributed power equipment due to the potential of single-point failure and high communication and data costs. Therefore, the technical development path of the power equipment perception mode is analyzed based on the development trend of the future power system. The concept of intelligent power equipment (IPE) is introduced, which combines online sensing, data mining, remote communication, and primary and secondary fusion technologies to develop an intelligent object that can realize autonomous situational awareness. IPE can actively interact with the control center and operation and maintenance (O&M) personnel according to its situation. This gives the power company an efficient and comprehensive perception of the equipment. Then, based on the actual situation of the power grid and emerging technology research directions, the challenges faced by each key technology supporting IPE and the corresponding technology enhancement solutions are presented. In addition, the O&M method applicable to IPE is discussed, which achieves proactive maintenance and prognosis management through autonomous equipment perception. Finally, the feasibility and effectiveness of IPE are verified by the performance of current IPE applications in an actual power grid.
POWER system operators are now promoting the development of intelligent, decentralized, and self-healing technologies, including the deep integration of the industrial Internet of Things (IOT) with the power system [
Subsequently, the periodic maintenance is used for power equipment. This mode enhances the ability to perceive the condition of power equipment. To date, this mode is still used by many power companies. However, the disadvantages of periodic maintenance are also prominent. First, it requires manual operation, testing, and recording, which lead to a large workload. Second, there are overhauls due to periodic dismantling, which add risks to the operation [
The agent-based perception mode is introduced for power equipment with the improvement of computer technologies [
The number and types of power equipment are surging currently. Taking transformers as an example, the global transformer market size was expected to have a large increase compared with 2017 [
1) For the periodic maintenance mode, the future volume of ponderous equipment will create great pressure on the operation and maintenance (O&M) personnel. The economic cost for expanding human resources is more than that power companies can afford.
2) For the agent-based power perception mode, introducing such a large volume of power equipment can bring a significant security risk of single-point fault [
To this end, [
To get rid of the agent-based power perception mode, the future development trend of power equipment is analyzed. Then, the concept of intelligent power equipment (IPE) is proposed and its meaning and technical content are clarified. IPE is a new power equipment management mode, which is a paradigm shift.
This paper is organized as follows. Section II analyzes the development of the power equipment perception mode. In Section III, we describe the concept of IPE includung its conceptual meaning, intelligent functions, and situational awareness system. The key technologies required for IPE and their solutions are proposed in Section IV. In Section V, we discuss the IPE O&M system. In Section VI, we introduce the field application of this concept in an actual power grid. A review of related work concludes this paper in Section VII.
The limiting factor in the development of the current power equipment sensing model is that it lacks intelligence. As a result, it cannot provide reference information for the O&M system in real time, and it cannot interact intelligently with other power equipment. The popularity of the IoT provides a new solution to the improvement of power equipment perception mode [
Autonomous computing meets its high requirements for real-time services, intelligent perception, and security and privacy by configuring computing power for the equipment itself [
Self-perception refers to the introduction of online monitoring technology and intelligent algorithms based on edge computing for power equipment to autonomously collect real-time information of itself and locally detect, analyze, and evaluate its condition [
Information interconnection is based on IoT technology to manage power equipment [
In summary, the power equipment perception mode in the future needs to configure autonomous computing capabilities for power equipment to realize its autonomous situational awareness. The information interconnection function is used to achieve comprehensive perception and real-time interaction of equipment information to meet the construction requirements of the Energy Internet. Based on the analysis of the future development trend, we propose the concept of IPE.
IPE mainly includes two parts of technical content, namely autonomous situational awareness and active O&M. Autonomous situational awareness refers to IPE using embedded sensors and artificial intelligence (AI) modules to achieve its sensing, status analysis, and health management. Active O&M refers to IPE actively interacting with the O&M system as an independent agent based on realizing autonomous situational awareness. The above two parts of technical content require IPE to have autonomous computing and information interconnection capabilities. Thus, it establishes information interoperability and early warning linkage with the logical environment of the power grid, peripheral equipment, and O&M platforms. Compared with ordinary power equipment, IPE can become an intelligent object for autonomous situational awareness in power system operation. In addition, IPE adopts primary and secondary fusion technology in hardware development, so that it also remains physically independent.
Fig. 1 Functional flow and related support algorithm for IPE to achieve autonomous situational awareness.
In the overall management, the backend server is only responsible for sending the updated parameters of the algorithm to the equipment. It reduces the amount of communication data, communication costs, and reliance on highly reliable communication while reducing the pressure on backend computing. Autonomous computing improves the timeliness of the information and avoids data congestion. IPE provides a paradigm shift for power equipment perception modes to meet the need for an efficient and comprehensive perception of power equipment in the future power grid.
To make this idea a reality, the key technologies for IPE will need to be developed and implemented with the following characteristics.
1) Self-sensing: sensors must be compact, sensitive, and cost-effective to ensure that they can be used in IPE for large-scale engineering applications.
2) Data pre-processing: the algorithm must have the ability to process multi-source heterogeneous data, including structured and unstructured data. The full exploitation of the IPE panoramic monitoring data by the algorithm realized at a later stage.
3) State analysis: the fault diagnosis and fault prediction must use lightweight algorithms and consider the correlation between multi-dimensional quantities to ensure the analysis accuracy of IPE condition.
4) State assessment: the degradation model of IPE with varying performance must be both accurate and generalizable. The health assessment system must achieve equipment health ratings while obtaining time-continuous equipment reliability.
5) Self-protection and control: self-protection and control of IPE can cause a change in the logical structure of the grid. Therefore, it must be confirmed that no chain of faults in the grid will be triggered when the IPE protects itself.
6) Remote communication: a communication method with high-security flexibility and outstanding economy must be selected to meet the communication needs of the widely distributed IPE to the remote servers and O&M personnel.
The implementation of IPE requires the support of several key technologies. This section examines the challenges faced by each key technology in the implementation of IPE and the corresponding improvement schemes.
Self-sensing is the data basis for IPE to realize its autonomous situational awareness.
| Category | Quantity | Problem representation |
|---|---|---|
| Electricity | Voltage Uk | Working condition of equipment k |
| Current Ik | ||
| Partial discharge uk | Minor internal defect of equipment k | |
| Machinery | Vibration signal VIk | Physical or electrical fault of equipment k |
| Environment | Temperature TEk | Internal short circuit of equipment k |
| Humidity HUk | Environmental humidity condition of equipment k | |
| Gas Gk,m | Fault of equipment k characterized by gas m | |
| Infrared image Pk | Localized overheating of equipment k |
In order to make IPE remain independent, the condition-monitoring sensors need to be embedded in the IPE body. It faces many challenges.
1) Hardware performance: the condition-monitoring data change frequently and unstably, e.g., partial discharge (PD) and vibration data, etc. This requires the signal sampling system to have a high data acquisition rate and data volume. Various performance requirements are imposed on the hardware system, such as sampling frequency, communication bandwidth, and energy consumption.
2) Economy: the cost of IPE data collection system needs to be low enough to facilitate large-scale promotion of IPE. On the one hand, the price of an analog-to-digital converter (ADC) increases exponentially as the sampling frequency increases. On the other hand, the data communication and storage costs are also the factors that cannot be ignored.
An effective measure to address these challenges of IPE sensing units is to develop auxiliary software algorithms. Among them, the compressed sensing (CS) can reconstruct the original signal from low-rank data [
The measurement matrix is used to realize the mixing of time-domain information. Subsequently, the measurement vector y is obtained by sampling with a low-frequency ADC. The sampling model can be developed as:
| (1) |
where is the measurement matrix, m is the number of measurement channels, and n is the original signal length; is the signal sparse basis, and l represents the sparse basis dimension; is the corresponding sparse coefficient; is the m-dimensional measurement vector; x is the original signal; and W is the CS operator.
needs to be set in advance. and need to have low relevance, so that the original information can be preserved to the greatest extent. Meanwhile, can achieve compressed collection and transmission of data.
Since the dimension of y is much smaller than that of x, it is necessary to use the sparsity of x in to solve (1). As s only contains a few non-zero elements, the reconstruction algorithm can get s from y, and then perform a sparse inverse transformation to obtain . The solution model can be described as:
| (2) |
Due to the sparsity of x and the low relevance between and , the l0-norm algorithm can be used to reconstruct the original signal from y.
We have achieved the sub-Nyquist data output rate of the PD signal, which mainly includes three technical methods. A joint sparse dictionary learning method based on multi-dimensional parameters of PD signals, a three-valued observation matrix design method based on incoherent continuation, and a sparse decomposition method of PD signals based on dual residual ratio thresholds are proposed [
IPE for autonomous situational awareness requires data pre-processing to obtain panoramic monitoring data. The challenges faced by IPE data pre-processing are as follows.
1) Heterogeneous data processing: IPE data sources include structured data such as electrical quantities and mechanical characteristics and unstructured data such as infrared images and recorded text. Normalizing the heterogeneous data and forming a unified data model are difficult problems to be solved by pre-processing.
2) Error data detection: to prevent false or missing data from monitoring resulting from sensing, communication, and other reasons, it is necessary to detect error data. The equipment monitoring data exhibit the 4V characteristics, i.e., volume, variety, value, velocity, of big data and vary in magnitude and dimension. The quick detection of such a large amount of data is a problem that must be solved by pre-processing.
We built a multi-dimensional soft-sensor model in the AI module to perform data pre-processing and integration. As a data middleware, a multi-dimensional soft-sensor can provide data sources for self-perception, and output real-time cross-sectional information of equipment condition. This model is developed as:
| (3) |
where X and are the pre-processed monitoring data and original data, respectively; F is the processing function; and the subscripts Str and Uns denote the structured and unstructured data, respectively.
Normalization can be performed directly on the structured data. For the unstructured data, an ontology dictionary can be constructed based on a hidden Markov model, and the k-nearest neighbor algorithm is used to classify the data. The corresponding vector space model is then used to vectorize the data.
For the error data detection, we conduct multi-dimensional data comparisons to make the algorithm both accurate and fast. The monitoring data are compared with 2-dimensional data both horizontally and vertically to detect error data. The error data detection model is expressed as:
| (4) |
where xref is the median of xi sequence; and is the average of N correlation covariates of xi,t at time t. That is, the data credibility is judged from the M data in the neighborhood of at time t and the changes of the other N correlation covariates xj,t at time t.
satisfying (4) is error. That is, there are abrupt changes in the time series and no corresponding changes in other relevant quantities. It requires data rejection or correction, and troubleshooting of the corresponding sensing unit.
If a sensor has long-term data anomalies due to the factors such as energy, the data will be defaulted. We use the tensor decomposition method for data filling [
State refers to the operating characteristics of power equipment, i.e., equipment fault or abnormality, health degree as characterized by data performance, and change trends. State analysis means that IPE analyzes its operating characteristics in real time based on access to monitoring data by AI modules, including anomaly detection, fault prediction, and fault diagnosis.
The framework of the state analysis is illustrated in
Fig. 2 Framework of state analysis.
Anormaly detection is applied to sudden functional faults. That is, characteristic quantities suddenly exceed the threshold range and there is no abnormal trend in the previous period. The mathematical model is expressed as:
| (5) |
where A is the monitoring data; ait is the value of quantity i at time t; and and are the thresholds.
This part focuses on setting threshold for each quantity. The IPE needs to set dynamic thresholds based on its monitoring data to achieve both accuracy and robustness.
For this, we embed the trend evolution algorithm with the adaptive thresholding function in the AI module of the IPE, where the threshold for the quantity i is expressed as:
| (6) |
where ai, , Ni, and pi are the monitoring data, data standard deviation, noise, and trend evolution of the quantity i, respectively.
When a threshold is exceeded, the IPE calls the fault diagnosis algorithm, which sends alarm information to the control center and the associated O&M personnel.
When each quantity is within the normal threshold range, the fault prediction should be used to determine whether there is a trend of data dispersion, i.e.,
| (7) |
where T is the length of the detection time interval; is the change rate; and is the predicted value.
The equipment fault prediction is divided into 2 methods: physical model based and data-driven methods [
The real-time cross-sectional information of IPE contains multi-dimensional quantities. The change of correlations among the quantities can better indicate the real operating condition of the equipment. Therefore, we make full use of high-order moment correlations among the quantities to design the fault prediction algorithm for the IPE.
First, the network is trained with the labeled dataset to learn the dynamic features of the equipment under each operating condition and extract the correlation relationships between the quantities. The stable correlations among the data are broken under the abnormal condition, and the abnormal dataset E deviates from e reconstructed using the normal denoising auto-encoder network.
Define the reconstruction error Re as an evaluation index for the detection of abnormal equipment:
| (8) |
The threshold JUL is adaptively set based on the confidence interval of the parameter to monitor the trend of Re as an alarm condition of equipment anomaly. Equipment anomalies will be reflected in the residual trends of the relevant variables. The use of changes in parameter residuals can enable a preliminary analysis of anomalies, and then isolate the relevant variables, analyze the cause of the anomaly, and proactively report the information. The residual of quantity i is related to its reconstructed value e and the actual value ai:
| (9) |
We have conducted research on this issue and the related characteristics of IPE. We have conducted in-depth research in this technical path and designed a fault diagnosis algorithm based on a deep autoencoding (DAE) network and XGBoost [
The role of fault diagnosis is to determine the fault location and type, and provide reference information for O&M personnel. The key part is given by the extraction and classification of fault features in monitoring data. The current fault diagnosis mainly has 2 kinds of algorithms: signal decomposition and deep learning.
In signal decomposition, the feature vector is extracted and clustered to discern its fault type. This algorithm is simple, but it is not suitable to handle high-dimensional mixed data.
The deep learning algorithm of the convolutional neural network (CNN) has a powerful feature extraction capability, which is suitable for fault diagnosis based on complex data [
| (10) |
| (11) |
where gs is the feature map learned by the convolutional kernel; is the weight of each feature data;bs is the bias term; is the output of CNN; is the probability of selecting the convolution kernel after determining the input data and model parameters; , , and are the input data dimensions; K is the number of convolutional kernels in the network; and denotes the model parameters.
However, the deep learning algorithms have problems of poor generalization and long calculation cycles, which cannot meet the real-time requirements of IPE.
We propose a power equipment fault identification algorithm based on multi-deep neural networks [
State assessment enables IPE to perform deep data mining on its condition information. Thus, it can evaluate the health degree in real time. It supports control center queries and own information upload.
Equipment degradation refers to the existence of failure behavior over time [
Given that degradation is initially stochastic, IPE first builds a degradation model based on a stochastic process. The data collection process considers the extreme behavior possibility and real average degradation behavior information. Subsequently, the degradation model of IPE is established based on the monitoring data. The nonlinear function of the degradation condition at time t is solved based on time variables and random variables, which is expressed as:
| (12) |
where is the random factor; and and are the drift and diffusion coefficients, respectively. Decomposing the stochastic differential equation according to (11), we can obtain:
| (13) |
where is the standard Brownian motion concerning t; and is the Brownian motion function. The probability of the transition of to at is counted by (13), and the degradation condition is derived as:
| (14) |
The initial values of the model parameters are estimated based on the expectation maximization (EM) algorithm. However, different IPEs have different degradation paths due to different operating environments and workloads.
To this end, a real-time degradation model of power equipment based on an improved Wiener stochastic process relying on a strong tracking filter (STF) is proposed [
The equipment health assessment refers to using condition monitoring data to determine the current health condition of the equipment. The current method is to manually assign weights without considering the degradation distribution of the equipment [
Thus, we propose a health assessment method for IPE. We integrate discrete and continuous model assessment methods in establishing the IPE health assessment system. Weights are determined autonomously based on contingency theory to enhance sensitivity to data changes. The reliability combined with the real-time degradation condition of the equipment is obtained based on the continuous model. A discrete model for health assessment is established, as shown in
Fig. 3 Discrete model for health assessment.
The equipment quantities are first graded. The weight vector W of each quantity is determined according to the contingency theory, as shown in (15). Subsequently, the HI model is established based on the data distribution distance. Finally, based on the object topological model, the equipment condition is rated using the closeness function.
| (15) |
where d(ai) is the weight of the data vector ai.
After obtaining the health rating, a continuous model for health assessment is also studied to incorporate the real-time degradation condition of the equipment, modeled as follows.
Suppose z(t) denotes the maintenance index of the equipment at time t, and . And the variables are independent of each other. The distribution of the random increment is denoted by FR. Subsequently, a continuous model of reliability Le as a function of A, FR, and t is developed as:
| (16) |
where is the reliability function. The Weibull proportional failure model is introduced to establish the function , which can consider both A and t and is expressed as:
| (17) |
where k0 is the initial risk function that follows the Weibull distribution. The algorithm uses the product of the covariate of A Coi(t) and its parameter to represent the relationship between A and Le(t).
IPE can integrate HI and Le(t) to characterize its health degree and the ability to maintain health using our method. Meanwhile, IPE supports independent reporting of health information and real-time queries from the control center.
In IPE, there is an electric connection between the AI module and sensors. The communication method becomes very important when IPE sends information to remote servers and O&M personnel. Table II compares the characteristics of communication methods.
| Method | Economy | Security | Reliability | Anti-interference | Anti-eavesdropping |
|---|---|---|---|---|---|
| Wireless | M | M | G | M | L |
| Optical fiber | P | G | M | G | E |
| Carrier | G | P | P | P | G |
Note: E means excellent, G means good, M means medium, and P means poor.
The communication method of the IPE must satisfy the following characteristics.
1) Economy: the communication cost is one of the limiting factors for IPE rollout. Therefore, the cost should be controlled to meet the engineering application requirements of IPE.
2) Security: the power system has the highest requirement for stability [
3) Flexibility: IPE, as an independent body, needs to realize the self-organizing characteristics in its communication method.
Therefore, it is necessary to combine local conditions when configuring communication methods for IPE. For switchgear stations or substations with centralized equipment and fiber-optic access, IPE in the station can access fiber optics for high communication security. Carrier communication can be used for IPE that is close to the backend, but the encoding and decoding need to use special keys to prevent information theft.
For the rest of the scattered IPE, it is not economical to build communication lines. The virtual private network can be established based on a cellular network to achieve secure wireless remote communication. When establishing VPN lines, a key management system should be established to ensure key security. Meanwhile, the identity authentication encryption algorithm is introduced in the network to identify the device ID and realize two-way authentication of the device and network. Cellular wireless networking enables the automatic access and exit of authorized devices by establishing wireless channels at the network layer. The resulting scalability of the network is outstanding.
In addition, the communication protocols are selected to interconnect equipment with different communication message formats. The commonly-used protocols include TCP/IP, NetBEUI, and IPX/SPX. Among them, the connection-oriented TCP/IP protocol is implemented with high reliability and profitability, which is suitable for IPE.
Based on the hardware principle of primary and secondary convergences, IPE protects itself and can autonomously execute external decision commands. The self-protection of IPE is achieved by breaking itself actively before it is damaged by an accident.
Owing to the limitation of edge-side computing capability, IPE cannot consider the global operation conditions of the power grid. The technical difficulty of self-protection lies in how to avoid chain faults caused by the self-protection of a single IPE.
To this end, IPE prevents chain faults by interfacing with the coordination system of the control center. IPE presets the alarm time delay before self-protection, sends the corresponding priority alarm message to the control center according to its importance, and sends a breaking warning to its neighboring IPEs. The neighboring IPEs need to be manually set according to the topological relationship of the power grid and equipment functions at the beginning of IPE operation. The control center adopts the appropriate coordination scheme. External decision commands refer to the instructions sent by the control center for switching or changing the operation mode, and the IPE executes the above decisions independently.
Given that IPE has the ability of autonomous situational awareness, the active maintenance based on process supervision can meet the O&M requirements of IPE. It transforms the maintenance from fault repair to equipment prognosis management.
Active maintenance requires interfacing the computerized maintenance management system (CMMS) processes of work order management, planning and scheduling, and maintenance metrics with an expert system based on web services. Experts operate remotely or develop maintenance strategies online based on the relationship of each piece of equipment. A schematic diagram of the IPE O&M system is shown in
Fig. 4 Schematic diagram of IPE O&M system.
In this O&M system, IPE obtains the equipment conditions in real time based on local data flow, and the internal algorithms improve the accuracy and robustness by synchronizing their information. There is a two-way connection between the control center and O&M personnel for assigning work orders and feedback. The control center can also send control commands to the equipment. The backend server can call the pre-processed data for a model update when the channel is free to make full use of its abundant computing resources.
We move out of the laboratory and deployed IPE in a real grid. A regional distribution system owned by the State Grid Corporation of China is selected for a pilot application to verify the effectiveness of IPE. The pilot region is located in a remote area with a harsh operating environment for power equipment.
The ideal IPE is to configure software and hardware during the manufacturing process. IPE can be directly connected to the network. However, it is unrealistic to replace the current power equipment operating in the grid at a large scale. The current application method is to utilize the physical assets of existing power equipment and upgrade them into IPEs by adding non-intrusive sensors, AI modules with corresponding algorithms, intelligent terminals, and communication capabilities, without affecting the security of the equipment. It can interact with other IPEs, control centers, and O&M personnel, as shown in Supplementary Material A Fig. S1.
In addition to the development of IPE-related technical content, we have developed a supporting mobile terminal application for O&M personnel, as shown in Supplementary Material A Fig. S2. We have also developed a real-time monitoring and display interface based on the Web, as shown in Supplementary Material A Fig. S3, for the control center. The complete IPE object and O&M prototype are established.
During the operation, the IPE triggers an abnormal PD alarm, as shown in
Fig. 5 IPE PD monitoring case. (a) Ultrasonic PD mapping of switchgear. (b) Abnormal temperature of transformer components.
The above results demonstrate that IPE actively participates in O&M management as an intelligent object. While significantly improving the reliability of the power grid, it makes the power equipment become an intelligent object that can perceive itself and interact with the O&M system.
In this paper, the concept of IPE and its intelligent functions are systematically described. Through the configuration of embedded sensors and AI modules to make power equipment with autonomous thinking, it can carry out its real-time detection and alarm, state analysis, and state assessment. IPE becomes an intelligent object that can realize autonomous situational awareness. In the O&M system, IPE serves as an independent intelligent entity that can analyze its condition in real time, enabling the power company to have an efficient and comprehensive awareness of equipment. Then, the corresponding technology enhancement solutions are given for the challenges faced by each key technology applied in IPE. This lays the theoretical foundation for IPE and shows that IPE can be applied in engineering at a large scale. The process of interaction between IPE and other entities or platforms in the O&M system is illustrated based on its perception mode. IPE meets the needs of managing the proliferation and variety of equipment in the complex power grid and is in line with the development trend of the Energy Internet.
We are now in close contact with equipment manufacturers to develop an IPE-independent object. We aim to achieve a paradigm shift in the perception mode of equipment in the future power grid.
References
Y. Kabalci, E. Kabalci, S. Padmanaban et al., “Internet of Things applications as energy Internet in smart grids and smart environments,” Electronics, vol. 8, no. 9, p. 972, Aug. 2019. [Baidu Scholar]
A. Marot, A. Kelly, M. Naglic et al., “Perspectives on future power system control centers for energy transition,” Journal of Modern Power Systems and Clean Energy, vol. 10, no. 2, pp. 328-344, Mar. 2022. [Baidu Scholar]
C. Choi, C. Esposito, H. Wang et al., “Intelligent power equipment management based on distributed context-aware inference in smart cities,” IEEE Communications Magazine, vol. 56, no. 7, pp. 212-217, Jul. 2018. [Baidu Scholar]
A. Sannino, H. Breder, and E. K. Nielsen. (2006, Jun.). Reliability of collection grids for large offshore wind parks. [Online]. Available: https://ieeexplore.ieee.org/document/4202378 [Baidu Scholar]
S. Raghavan and B. Chowdhury, “State diagram-based life cycle management plans for power plant components,” IEEE Transactions on Smart Grid, vol. 6, no. 2, pp. 965-972, Mar. 2015. [Baidu Scholar]
H. de Faria Jr, J. G. S. Costa, and J. L. M. Olivas, “A review of monitoring methods for predictive maintenance of electric power transformers based on dissolved gas analysis,” Renewable and Sustainable Energy Reviews, vol. 46, pp. 201-209, Jun. 2015. [Baidu Scholar]
M. Manbachi and M. Ordonez, “Intelligent agent-based energy management system for islanded AC-DC microgrids,” IEEE Transactions on Industrial Informatics, vol. 16, no. 7, pp. 4603-4614, Jul. 2020. [Baidu Scholar]
H. F. Habib, T. Youssef, M. H. Cintuglu et al., “Multi-agent-based technique for fault location, isolation, and service restoration,” IEEE Transactions on Industry Applications, vol. 53, no. 3, pp. 1841-1851, May 2017. [Baidu Scholar]
Z. Zhu, Y. Xiang, M. Li et al., “Preventive maintenance subject to equipment unavailability,” IEEE Transactions on Reliability, vol. 68, no. 3, pp. 1009-1020, Sept. 2019. [Baidu Scholar]
Puhua Policy Information Consulting Co. Ltd. (2020, Nov.). 2021-2026 industrial Internet industry supply and demand status quo and investment prospect forecast report, Beijing, China. [Online]. Available: https://wenku.baidu.com/view/50e7b54383c758f5f61fb7360b4c2e3f5727258d.html [Baidu Scholar]
K. Perdon, M. Scarpellini, S. Magoni et al., “Modular online monitoring system to allow condition-based maintenance for medium voltage switchgear,” CIRED-Open Access Proceedings Journal, vol. 2017, no. 1, pp. 346-349, Oct. 2017. [Baidu Scholar]
S. Xu, H. Liu, and T. Bi, “Field PMU test and calibration method – Part II: test signal identification methods and field test applications,” Journal of Modern Power Systems and Clean Energy, vol. 10, no. 1, pp. 243-253, Jan. 2023. [Baidu Scholar]
G. Bedi, G. K. Venayagamoorthy, R. Singh et al., “Review of Internet of Things (IoT) in electric power and energy systems,” IEEE Internet of Things Journal, vol. 5, no. 2, pp. 847-870, Apr. 2018. [Baidu Scholar]
S. S. Reka and T. Dragicevic, “Future effectual role of energy delivery: a comprehensive review of Internet of Things and smart grid,” Renewable and Sustainable Energy Reviews, vol. 91, pp. 90-108, Aug. 2018. [Baidu Scholar]
L. Long, “Research on status information monitoring of power equipment based on Internet of Things,” Energy Reports, vol. 8, pp. 281-286, May 2022. [Baidu Scholar]
M. Elsisi, M. Q. Tran, K. Mahmoud et al., “Towards secured online monitoring for digitalized GIS against cyber-attacks based on IoT and machine learning,” IEEE Access, vol. 9, pp. 78415-78427, May 2021. [Baidu Scholar]
R. Solozabal, A. Sanchoyerto, E. Atxutegi et al., “Exploitation of mobile edge computing in 5G distributed mission-critical push-to-talk service deployment,” IEEE Access, vol. 6, pp. 37665-37675, Jun. 2018. [Baidu Scholar]
S. O. Muhanji, W. C. H. Schoonenberg, and A. M. Farid, “Transforming the grid’s architecture: enterprise control, the energy Internet of Things, and heterofunctional graph theory,” IEEE Power and Energy Magazine, vol. 17, no. 5, pp. 71-81, Sept. 2019. [Baidu Scholar]
P. Deane and M. Brinkerink, “Connecting the continents – a global power grid,” IEEE Power and Energy Magazine, vol. 18, no. 2, pp. 121-127, Mar. 2020. [Baidu Scholar]
S. N. Shirazi, A. Gouglidis, A. Farshad et al., “The extended cloud: review and analysis of mobile edge computing and fog from a security and resilience perspective,” IEEE Journal on Selected Areas in Communications, vol. 35, no. 11, pp. 2586-2595, Nov. 2017. [Baidu Scholar]
R. Khelif, B. Chebel-Morello, S. Malinowski et al., “Direct remaining useful life estimation based on support vector regression,” IEEE Transactions on Industrial Electronics, vol. 64, no. 3, pp. 2276-2285, Mar. 2017. [Baidu Scholar]
S. Yang, Y. Su, Y. Chang et al., “Short-term traffic prediction for edge computing-enhanced autonomous and connected cars,” IEEE Transactions on Vehicular Technology, vol. 68, no. 4, pp. 3140-3153, Apr. 2019. [Baidu Scholar]
Navigant Consulting, Inc. (2019, Dec.). Edge computing and distributed intelligence for grid operations – AMI, distribution automation, volt/var optimization, smart inverters, and DER integration: global market analysis and forecasts, Chicago, Illinois, USA. [Online]. Available: https://www.navigantresearch.com/reports/edge-computing-and-distributed-intelligence-for-grid-operations [Baidu Scholar]
A. Muller, A. C. Marquez, and B. Iung, “On the concept of e-maintenance: review and current research,” Reliability Engineering & System Safety, vol. 93, no. 8, pp. 1165-1187, Aug. 2008. [Baidu Scholar]
D. Zhang, F. Haider, M. St-Hilaire et al., “Model and algorithms for the planning of fog computing networks,” IEEE Internet of Things Journal, vol. 6, no. 2, pp. 3873-3884, Apr. 2019. [Baidu Scholar]
G. Fulli, M. Masera, A. Spisto et al., “A change is coming: how regulation and innovation are reshaping the European Union’s electricity markets,” IEEE Power and Energy Magazine, vol. 17, no. 1, pp. 53-66, Jan. 2019. [Baidu Scholar]
C. Caione, D. Brunelli, and L. Benini, “Compressive sensing optimization for signal ensembles in WSNs,” IEEE Transactions on Industrial Informatics, vol. 10, no. 1, pp. 382-392, Feb. 2014. [Baidu Scholar]
S. Zhao, H. Zhao, Y. Qu et al., “Compressed sampling of partial discharge signal based on virtual channel extension,” Transactions of China Electrotechnical Society, vol. 38, no. 12, pp. 3366-3378, Mar. 2022. [Baidu Scholar]
S. Zhao, H. Zhao, L. Ma et al., “Partial discharge signal compression reconstruction method based on transfer sparse representation and dual residual ratio threshold,” IET Science, Measurement & Technology, vol. 17, no. 6, pp. 230-242, Aug. 2023. [Baidu Scholar]
H. Zhao, P. Shou, and L. Ma, “A tensor completion method of missing data in transformer district,” Proceedings of the CSEE, vol. 40, no. 22, pp. 7328-7336, Nov. 2020. [Baidu Scholar]
W. Jiang, Z. Zhang, F. Li et al., “Joint label consistent dictionary learning and adaptive label prediction for semisupervised machine fault classification,” IEEE Transactions on Industrial Informatics, vol. 12, no. 1, pp. 248-256, Feb. 2016. [Baidu Scholar]
Y. Tan, Y. Chen, Y. Li et al., “Linearizing power flow model: a hybrid physical model-driven and data-driven approach,” IEEE Transactions on Power Systems, vol. 35, no. 3, pp. 2475-2478, May 2020. [Baidu Scholar]
R. Rai and C. K. Sahu, “Driven by data or derived through physics? A review of hybrid physics guided machine learning techniques with cyber-physical system (CPS) focus,” IEEE Access, vol. 8, pp. 71050-71073, Apr. 2020. [Baidu Scholar]
H. Zhao, X. Yan, G. Wang et al., “Fault diagnosis of wind turbine generator based on deep autoencoder network and XGBoost,” Automation of Electric Power Systems, vol. 43, no. 1, pp. 81-86, Nov. 2019. [Baidu Scholar]
K. Zhang, B. Tang, Y. Qin et al., “Fault diagnosis of planetary gearbox using a novel semi-supervised method of multiple association layers networks,” Mechanical Systems and Signal Processing, vol. 131, pp. 243-260, Sept. 2019. [Baidu Scholar]
Y. Qu, H. Zhao, L. Ma et al., “Multi-depth neural network synthesis method for power transformer fault identification,” Proceedings of the CSEE, vol. 41, no. 23, pp. 8223-8231, Aug. 2021. [Baidu Scholar]
Y. Shirasaka, H. Murase, S. Okabe et al., “Cross-sectional comparison of insulation degradation mechanisms and lifetime evaluation of power transmission equipment,” IEEE Transactions on Dielectrics and Electrical Insulation, vol. 16, no. 2, pp. 560-573, Apr. 2009. [Baidu Scholar]
W. Huang, C. Shao, M. Dong et al., “Modeling the aging-dependent reliability of transformers considering the individualized aging threshold and lifetime,” IEEE Transactions on Power Delivery, vol. 37, no. 6, pp. 4631-4645, Dec. 2022. [Baidu Scholar]
H. Zhao, J. Chang, and Y. Qu, “Prediction method of residual life of transformer oil-paper insulation based on Wiener random process improved by strong tracking filter,” IET Generation, Transmission & Distribution, vol. 16, no. 19, pp. 4007-4016, Oct. 2022. [Baidu Scholar]
J. H. Jürgensen, A. Godin, and P. Hilber. (2007, Jun.). Health index as condition estimator for power system equipment: a critical discussion and case study. Presented at 2017 24th CIRED. [Online]. Available: https://www.diva-portal.org/smash/get/diva2:1120997/FULLTEXT03 [Baidu Scholar]
N. Zhou, Y. Xu, S. Cho et al., “A systematic review for switchgear asset management in power grids: condition monitoring, health assessment, and maintenance strategy,” IEEE Transactions on Power Delivery, vol. 38, no. 5, pp. 3296-3311, Oct. 2023. [Baidu Scholar]
D. Du, M. Zhu, X. Li et al., “A review on cybersecurity analysis, attack detection, and attack defense methods in cyber-physical power systems,” Journal of Modern Power Systems and Clean Energy, vol. 11, no. 3, pp. 727-743, May 2023. [Baidu Scholar]
