Research on Transparent Grid Dynamic Surveillance and Fault Early Warning System for High-Density Distributed Power Supply Access Areas Incorporating Artificial Intelligence Technology
Publié en ligne: 26 mars 2025
Reçu: 02 nov. 2024
Accepté: 23 févr. 2025
DOI: https://doi.org/10.2478/amns-2025-0810
Mots clés
© 2025 Zhongqiang Zhou et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
With the massive construction of infrastructure, the energy pattern dominated by fossil fuels is facing an unprecedented crisis. On the one hand, the rapid development of economy depends on energy, which leads to the increase of energy consumption year by year, especially the increasing depletion of non-renewable resources such as coal and oil [1-2]. On the other hand, the environmental pollution problem brought by the large use of fossil energy is also becoming more and more prominent [3-4]. Compared with centralized power generation, distributed power generation technology has the advantages of less environmental pollution, high reliability, low investment, and high comprehensive energy utilization rate, and thus has a better development prospect [5-7].
Under the incentive of China’s “double carbon” goal, the new power system based on renewable energy scale access will become the development trend, and the structure of the distribution network will also change. Under such a background, it directly leads to the utilization of resources more in favor of renewable resources, and the degree of utilization of renewable resources in the future will continue to increase [8-11]. And the power industry will be accessed to the distribution grid in the form of multi-point and high-density through distributed power generation, which requires that the development speed of the distributed power grid should be improved, so as to better adapt to the trend of the times [12-14]. The stochastic and intermittent characteristics of distributed power supply make the power output of this power source has an unstable situation, and the high penetration of distributed photovoltaic power supply will have an impact on the safe operation of the distribution grid [15-17]. If the distributed power supply fails, it is difficult to guarantee the power supply reliability of the distribution network, and it will also affect the operation effect and economic benefits [18-20]. In order to minimize the accidents of distributed power supply and protect the safe and reliable operation of the distribution network, it is of great significance to carry out dynamic monitoring and fault warning for distributed PV power supply [21-23].
The multi-source heterogeneous data in the distribution network contains a wealth of information, and the use of artificial intelligence methods for feature extraction and analysis can realize the identification and localization of faults in the distribution network system. Liao, Y. et al. showed that line loss is an important cause of energy loss in distribution networks, and proposed an intelligent diagnostic algorithm (ISSA-SVM) combining the sparrow search algorithm and support vector machine to analyze the line loss as well as diagnose the anomalies of the distribution system, which significantly improves the power supply efficiency as well as the security of the distribution system [24]. Stock, S. et al. proposed a combined method of convolutional neural network and segmentation function for power system fault monitoring, which realizes the detection, classification and localization of faults in the distribution network by fitting the voltage waveforms and establishing a correlation vector function for analysis [25]. Baghaee, H. R. et al. used the support vector machine algorithm to construct a classification model for the islanding detection problem in distribution networks, in addition to this model can effectively deal with grid fault detection in distribution network systems, which achieves high classification efficiency while also improving classification accuracy and effectiveness [26]. Liwen, Q. et al. used neural networks to learn and train the distribution network data after preprocessing, so as to obtain the deterministic diagnostic rules and optimal generalized decision rules for distribution network faults, which not only reduces the neural sample data and fault diagnostic time, but also improves the diagnostic efficiency in fault diagnosis [27]. Perez, R. et al. constructed fuzzy rule sets in terms of magnitude of phase and neutral currents and designed a fault identification model for distribution network systems based on artificial intelligence techniques such as fuzzy logic, which is reliable in fault identification and localization [28]. Hosseini, M. M. et al. explored the application of Monte Carlo tree search and -greedy search algorithms for solving uncertain and high-dimensional spatial problems in distribution network systems, constructing targeted supervised and unsupervised deep learning models for the related problems, which are able to enhance the operation monitoring capability and anomaly detection of the distribution network system, and improve the resilience of the distribution network system [29]. The above studies discussed the fault diagnosis methods in distribution network systems, but due to the limited fault recovery capability of distribution networks, the power supply service cannot be restored in time when subjected to unexpected conditions, therefore, effective early warning of fault risks during distribution network operation is a key way to improve the reliability of power supply in distribution networks.
Based on this background, a large number of scholars have carried out research on distribution network fault early warning.Ghaemi, A. used a stacked integrated learning approach to identify the state of the distribution network system and the external condition factors affecting the system, and established a defect early warning system through situational awareness of the distribution network system [30]. Ma, X. et al. designed a fuzzy clustering-based adjustment method for fault alarm thresholds in distribution network information monitoring system, which improved the adaptive adjustment ability of the thresholds, effectively avoided the redundancy interference generated by them, and ensured real-time and reliable distribution network monitoring and warning results [31]. Yang, Y. et al. established a multi-load early warning system for distribution networks based on big data, which forms a visual display of the operating status of the distribution network by processing and analyzing the monitoring data and provides an effective early warning of impending distribution network faults [32]. Further, for the distribution network accessed by distributed power sources, some scholars have also researched on its monitoring and early warning system.Gu, C. et al. introduced a load state sensing early warning method for high penetration distributed distribution networks, which dynamically monitors the tidal current fluctuation and voltage overruns existing in the power supply system, and can provide timely early warning information on distribution network faults [33]. Shen, Z. et al. proposed a dynamic tidal current model based on complex network theory to cope with the distributed distribution network fault warning problem, and realized timely fault warning by describing the relationship between the model and the distribution network line load loss [34]. Xiang, M. et al. constructed a fault early warning model for distribution networks based on fault gene table, which establishes a fault gene table based on the historical fault data of each distribution network, and matches the acquired real-time distribution network state data with the fault genes through an algorithm in order to realize fault early warning for distributed distribution networks [35]. However, the proposed algorithm not only reduces the management efficiency of distribution networks, but also affects the accuracy, and it is crucial to further study the distribution network monitoring and warning system with high accuracy and efficiency.
The article examines how artificial intelligence technology can be used to create a transparent grid dynamics monitoring and fault warning system for high-density distributed power access areas. The technical route explored in the article includes the combination of “data collection” and “computational prediction” to realize the transparency of the distribution network and to solve the existing problems and new challenges. By analyzing the system structure and formulating the functional design process, a four-layer grid dynamic monitoring and fault warning system is designed. The system’s multi-functional modules (implementation of monitoring data, intelligent communication, alarm messages, etc.) Work together to assist in monitoring and early warning of grid faults using artificial intelligence calculations.
Reinforcement learning [36] belongs to an important research field in machine learning. Compared with supervised and unsupervised learning, reinforcement learning is a kind of active learning, i.e., it is essentially a process of learning strategies in the environment by “taking into account the situation”, emphasizing how to adaptively adjust the actions based on the state in order to maximize the expected benefits. Reinforcement learning Reinforcement learning is mainly applied to power grid security control, automatic power generation control and power grid dispatch optimization.
In power grid security control, reinforcement learning methods only need to react to the evaluation information of the current control effect, which has higher control real-time and robustness, and thus has been applied in power system security and stability control [37]. Automatic power generation control is a dynamic multilevel decision-making problem, and its control process can be regarded as a Markov decision-making process in order to ultimately realize the matching of generation output and load power within the whole system. For this reason, based on the Markov chain control process, offline learning control algorithms for load frequency control of automatic power system control have been studied by many scholars with a view to achieving fast and automatic control system optimization. In power scheduling optimization, reinforcement learning converts the constraints, actions and objectives in the scheduling optimization problem into states, actions and rewards in the algorithm, and dynamically searches for the optimal actions through continuous trial and error, backtracking, iteration and other operations, which achieves good results in dealing with multi-objective scheduling optimization problems, and has a strong feasibility and validity [38].
According to the current high-density distributed power supply access area transparent power grid monitoring technology and electronic equipment parameter correction technology deficiencies, this paper designs the power grid intelligent dynamic monitoring and fault early warning system based on artificial intelligence technology, and puts forward the following requirements:
Change the traditional manual regular field inspection of the grid operation status, and adopt information technology means to realize the real-time collection of line operation status on the branch lines of the low-voltage distribution network. Change the mode of after-the-fact processing to avoid electricity safety accidents as much as possible. Realize the Internet of Things, upload the relevant line operation parameter information collected by the monitoring equipment to the cloud platform, and display the relevant data on the front-end page to realize the real-time monitoring of the low-voltage distribution network. Development of supporting visualization software monitoring system, grid operation and inspection staff through the front-end page of the data display, overview of the entire low-voltage distribution network operation status, to achieve the overall operation of the low-voltage distribution network for real-time control. Fault alarm function, when the monitoring equipment detects that the line is in abnormal operation state, it can upload the operation state information to the cloud platform in time, display the abnormal state in the visualization monitoring page, and inform the operation and inspection staff through SMS to ensure that the staff can receive the alarm information in time in order to troubleshoot as soon as possible and improve the quality of power supply service. Harmonic analysis function, through the algorithm to analyze the harmonics in the power grid, and the harmonic parameters obtained from the analysis will be reported to the cloud platform, which can help to analyze the faults when the line is running abnormally. Parameter automatic correction function, monitoring equipment in the factory needs to go through parameter correction, improve the collection accuracy, reduce the error, change the traditional manual parameter correction manually, the design of supporting parameter automatic correction system, simplify the correction process, improve the efficiency of correction, save time and labor costs, to ensure that the accuracy, stability and uniformity of the equipment collection. Integrate intelligent algorithms, multivariate information and data-driven methods to realize efficient monitoring and accurate fault location of the grid system through the steps of grid historical data collection, big data analysis preprocessing and real-time data input and output.
The overall structure of the intelligent high-density distributed power distribution network dynamic monitoring and fault detection system designed in this paper is shown in Fig. 1. The system is divided into two parts:
Online far-infrared monitoring unit, mainly including: far-infrared monitoring unit probe, nylon support fixtures, scanning control platform, control circuit, column, solar charging board and so on. Receiving unit, mainly including: desktop computer (pre-installed DXJ online high-voltage electrical equipment monitoring and warning system software), receiving unit RTU relay, and related connection lines.
The overall structure of the system
The system design requires 6 temperature monitoring points per switchgear, i.e. 6 temperature sensors connected to each temperature acquisition module. The selection and function of each module are as follows:
Temperature sensor: The temperature sensor is the most important component of the bottom layer of the monitoring network, which can largely ensure the long-term and stable operation of the sensor. Temperature acquisition module A temperature acquisition module is set for each switchgear cabinet, which can largely improve the real-time and reliability of the monitoring data acquisition, and at the same time, it is necessary to transmit the collected data to the monitoring center. RS-485 bus The RS-485 cable is used to realize the link between the temperature acquisition module and the monitoring center, so as to ensure the safety and reliability of data transmission to a large extent, and the RS-485 communication range should be set within 1200m, so as to provide reliable communication quality. Monitoring Center The monitoring center’s primary function is to process the corresponding signals and then display them using visualization technology to achieve the goal of early warning.
The functional modules of the system are shown in Figure 2. The system is divided into two parts of the design of the front and backend, the frontend mainly realizes the reality of voting, commenting and system projects as well as the information of user registration, login and posting comments. The background mainly realizes the function of information query, comments and other information management and so on.
The function flow chart of the system
The system presented in this paper is divided into four layers, and each layer is interconnected with each other to achieve dynamic monitoring of the transparent power grid and fault warning. The hierarchical structure of the system is depicted in Figure 3.
System hierarchy diagram Data layer Running on the database server of the cluster, with real-time data as the core. This includes the real-time database, spatial database, attribute database, and media database, which periodically store the data required for platform calculations. Application component layer Functional parts for data processing, deployed in component mode. Interface layer Running on the interface gateway server, it defines the standard data exchange model and handles the data exchange function of the neighboring systems. Application layer Functional performance and human-computer interaction based on the WEB mode, running on the WEB server.
The detailed functional module structure of the system is shown in Figure 4.
System function module structure
After entering the main interface of the monitoring platform, if you need to operate, you need to log in first, click on the “User Registration” menu, the pop-up menu shown in the left figure, click on the “User Login”, the user login dialog box appears, enter the user name and password, the user will be able to log in successfully. The user login dialog box will appear, enter your user name and password, and the user will be logged in successfully. If you need to change the user during operation, click “Change User” menu item, the Change User dialog box will appear, enter the user name and password of other users to change the user.
In the system designed in this paper, a real-time display of monitored data items is required, and the types of monitoring include telemetry, telecommunication, telepulse, faults, and harmonic quantities for substation equipment.
In the real-time data dialog box, in order to make the operation of the substation user convenient, the layout of the function interface adopts the board type of the tree structure on the left side and the function interface on the right side. The left side is designed to display the names of specific substation devices whose detected data quantities include telemetry, telecommunication, telepulse, fault, and harmonic quantities.
Intelligent communication program is at the bottom of the system, completing all data acquisition tasks of the monitoring station and having the function of dual-machine standby, which can realize the fast switching of the main and standby machines. Intelligent communication program communicates and exchanges data with monitoring and protection devices through RS232/RS485 or Ethernet. Intelligent communication program to complete a communication task consists of two parts, the main program and the statute dynamic library program (statute name.DLL). The main program completes some general functions, and the driver is responsible for interpreting the statute, interpreting the byte stream read out from the channel by the main program into usable data, and transforming the commands issued by the main program into byte streams according to the requirements of the statute. The main program controls multiple drivers to complete communication with devices from different manufacturers.
The main function of the alarm information module is to monitor the operation of generating units and various equipment within the power grid, and there are many types of alarms. The most commonly used types of alarms include exceeding the threshold, equipment unresponsiveness, link failure, and abnormal equipment status. Monitoring the response to the alarm information, can be stored in real time to the database response data table, and in the front section can also be a real-time display, and draw the trend of the operating state of the equipment, in order to be able to summarize the change rule of the operating state of the equipment in a timely manner.
The dispatch data management system is used to create, modify and manage dispatch data. The main function of the dispatching data management system is to set the data contained in each data frame of each dispatching channel and its comparison with the data in the system for the communication program, so that the communication system can write the received data to the real-time database according to the corresponding relationship, or take out the data from the real-time database and send it out according to the corresponding relationship. It is the foundation for the operation of the entire dispatching system.
With the continuous development of power grid automation technology, the current distribution network can collect the grid operation data more and more present multi-source, a large amount of data can provide extremely effective information for the stable operation of the distribution network and fault location, improve the stability of the entire power grid operation.
The multi-source information used in the fault early warning process includes distribution network protection circuit breaker action information, distribution network event sequence information, fault indicator alarm information, and fault electrical quantity information. The sources of these four types of information and their roles in fault warning are analyzed below:
Distribution network protection action information. The protection of the distribution network is usually configured on the circuit breaker at the outlet of its feeder line, and when a fault occurs on a distribution line, the fault current will change. When a fault current flows through the relevant line, if a protection device is installed on that line, the protection device will disconnect the line through the circuit breaker to remove the fault. The information about distribution network protection actions is extremely important signal data when the distribution network is faulty, and can provide great help in localizing distribution network faults. Therefore, automated collection devices are installed on distribution feeders to realize the detection and collection of protection device action information, which is then provided to the fault early warning analysis system through dynamic monitoring to improve the accuracy of fault early warning of distribution networks. Distribution network event sequence information. The so-called event sequence is the sequence of the actions of each device after the occurrence of faults in the distribution network, for example, the circuit breaker will act only after receiving the protection signal, the switch action after the operation of the knife gate, etc., and then deposited into the grid dynamic monitoring and fault warning system. The sequence information of these events can provide a reference for the fault analysis of the active distribution network and facilitate the determination of the exact location of the fault. Therefore, when a fault occurs in the distribution network, the report information in the system database can be called up to assist in the location judgment of the fault based on the combination information of switch action, protection signal, alarm signal, site information, and institutional signal action of different lines. Fault indicator alarm information. The distribution network fault indicator can sense the current of the line fault and provide alarm information, and line repair personnel can locate the fault according to the alarm information provided by the fault indicator. When a fault occurs in the distribution feeder, the alarm indicator on the fault feeder will send an alarm signal and store it in the grid dynamic monitoring and fault warning system. When carrying out fault localization of the distribution network, the fault indicator alarm can be called to determine the faulty distribution feeder. Fault electrical quantity. Distribution network fault voltage and fault current changes are the most significant features of distribution network faults, and distribution network feeder faults can lead to increased fault voltage and current and phase changes. The traditional distribution network feeder acquisition device only collects the voltage and current amplitude of the line for fault localization. With the progress of artificial intelligence technology, the traditional feeder acquisition device can now acquire distribution network voltage and current phase angle after improvement. The volume measurement collected by the phase volume measurement device (PMU) contains a wealth of fault electrical information, such as frequency, phase angle, and amplitude. It can be collected and stored very accurately, effectively improving the accuracy of the grid fault warning.
In view of the current operation status of the power grid, full consideration of high-density distributed power grid and collection devices is becoming an increasingly intelligent situation. In this paper, the fault early warning is specifically expressed as follows: firstly, the distribution network is optimized for partitioning based on multi-source information, and on this basis, artificial intelligence technology is used to locate the fault section. Then, the distance measurement of the fault in the region, mainly using multi-port electrical distance measurement method. In active distribution network fault segment localization, if the zoning area contains protection devices, the comprehensive judgment of faults can be carried out based on event sequence information and protection action information.
In order to verify the feasibility of this paper’s grid dynamic monitoring and fault early warning system combined with artificial intelligence techniques, this section collects 63 sets of data generated from a feeder before the occurrence of a cable burnout, which is first screened by taking the amplitude threshold of 10, when the amplitude of the zero sequence current is less than 10 the data sets are removed, while the ones greater than 10 are retained as the input-output data for the training and prediction of the artificial intelligence techniques. Here the threshold value of 10 is taken due to the fact that in the measurement signals are sometimes disturbed or it takes more than 1 year before the insulation is completely damaged. By filtering out 8 such sets of data 55 sets of data can be obtained, out of which 50 sets of data are used as input samples for training and these 50 sets of data are used as evaluating the feasibility of the system.
The comparison and percentage of error between the predicted output of the system and the original data are shown in Fig. 5 and Fig. 6 respectively. From the two figures, it can be seen that the system in this paper applies artificial intelligence techniques for fault monitoring and warning with high fitting ability, and is able to recognize various different fault states in the process of fault occurrence, which proves the feasibility of the system in this paper in predicting the occurrence of faults in high-density distributed power supply access areas. Among them, the predicted output values are basically consistent with the true values, with a high degree of fit, and the percentage of error is between ±0.15, and the overall error is more evenly distributed among the groups, and will not be concentrated in certain groups with relatively large or small fault values. In light of the errors, the system’s prediction accuracy can be improved by increasing the number of samples significantly.
The comparison between the system’s forecast output and the original data
Percentage of system prediction error
After completing the fault information acquisition, the high-density distributed power supply access area fault localization is analyzed, based on the above fault data, the system fault localization results in accurate statistical analysis as shown in Table 1. As can be seen from the table, the system determines that there are five kinds of faults in the distribution network in the high-density distributed power supply access area, and the success rate of fault information acquisition is 100%, and the fault localization accuracy rate is above 95%, which basically meets the requirements of real-time and accuracy of fault localization in the power grid in the high-density distributed power supply access area.
Statistical error warning accuracy statistics
| Fault type | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|
| Actual number of failures | 33 | 17 | 40 | 29 | 24 |
| The number of failures of the system | 32 | 17 | 39 | 29 | 24 |
| Location is not sure | 1 | 0 | 1 | 0 | 0 |
| Failure acquisition success rate | 100% | 100% | 100% | 100% | 100% |
| Positioning accuracy | 96.97% | 100% | 97.44% | 100% | 100% |
In this study, a transparent grid dynamic monitoring and fault early warning system based on artificial intelligence technology for high-density distributed power supply access areas is developed and applied to short-circuit current monitoring in distribution grids containing distributed power supplies. The field monitoring is carried out in the context of an urban low voltage grid under the jurisdiction of a power supply bureau of a power grid company. Since June 2022, the monitoring system has been in operation and the equipment is functioning normally, the data transmission is efficient, the overall operation results are satisfactory, and the monitoring results are analyzed as follows.
After the system was commissioned, a loss of power fault was monitored on this grid system. The current and voltage waveforms captured during one of the grid power loss faults are given in Fig. 7. From the figure, it can be seen that the moment of the fault is in the 3rd IF cycle and the fault occurs on the power supply side or on an adjacent branch. When a short-circuit occurs on the power supply side of the grid system or on an adjacent branch, the fault current is supplied by both the system power supply and the PV power supply. Due to the high short-circuit current provided by the system, the system-side protective switch action trips, after which the fault current is provided only by the PV power supply, and the current monitored by the system is also the fault current provided by the PV power supply. The PV inverter works in the maximum power point operation mode, its output power is not enough to maintain the load demand after the fault, the output current gradually increases and the output voltage gradually decreases, in the 9th weekly wave after the fault occurs, the inverter output current exceeds the protection limit or the islanding effect is monitored, the PV inverter automatically shuts down, and the whole grid loses all power supply.
Current voltage waveform captured during grid failure
Fig. 8 shows the current and voltage waveforms during a current inrush fault in a grid system. The moment of fault occurrence is in the 3rd and 9th industrial frequency cycles, and a large inrush current occurs in the distribution grid system containing distributed power supply, which is estimated to be due to the fact that there is a large motor-type load startup in the distribution branch where the protection switch is located with a large startup current, and the PV and power supply both present a large current inrush for the startup of the load, but since the PV inverter works in the maximum power point operation mode, under a certain output power condition, the inrush current is too large to pull down the voltage, and since motor starting is an instantaneous process, the system quickly returns to normal.
Current voltage waveform of current shock failure of the grid system
From the analysis of the results in the figure, it can be seen that this fault condition is detected by the sudden change of current instantaneous value or voltage RMS criterion in the composite criterion, which proves that the system in this paper is sensitive to the judgment of fault signals. The remote reception, storage, and visualization of fault waveform data are normal, which proves that the dynamic monitoring and fault warning system based on artificial intelligence technology operates reliably.
Figure 9 shows a set of voltage transient drop waveforms captured by the system, and it can be seen that this fault was detected by a sudden change in voltage instantaneous value or voltage RMS criterion. The system experienced a typical voltage transient drop phenomenon that lasted for approximately 5 IF cycles and had a depth of drop of more than 45%.
The waveform diagram of the branch of the network
The voltage drop waveform is detected by the voltage instantaneous value mutation or voltage RMS criterion in the composite criterion, and the composite criterion is reasonable and effective, and the judgment of the fault signal is sensitive and highly accurate. In the process of voltage transient drop, the distributed power supply reacts to the grid abnormality with a current impact process, after which the change of current is relatively smooth. In the instant of grid voltage transient and recovery, there are inrush currents, but the size of the inrush current does not exceed the maximum output of the PV.
In this paper, a transparent grid dynamic monitoring and fault early warning system for high-density distributed power supply access areas is designed using artificial intelligence technology. The system is divided into a four-layer structure: data layer, application formation layer, interface layer, and application layer. By collecting information such as distribution network protection circuit breaker action, distribution network event sequence, fault indicator alarm and fault electrical quantity, the dynamic monitoring and fault warning of power grid is realized under this system.
The system in this paper provides an excellent fit between the predicted output value and the real value for grid fault monitoring and warning, with the error percentage between -0.1244 and 0.121, and the fault localization accuracy rate exceeds 95%. The system has a good effect in the power grid monitoring of a power supply bureau of a power grid company, and can make timely fault warning.
In a power loss fault, the system monitors that the fault occurs in the third cycle, and the entire power grid loses power in the 9th cycle. The real-time data shows that the system monitors the duration of a voltage-dropout fault to be approximately 5 cycles.
Through the visualization of grid data, the system enables real-time monitoring and accurate positioning of grid faults, and provides effective early warning. Looking ahead, the application of new technologies will have an impact on the traditional distribution network management mode, and it is necessary to study the new requirements for planning and monitoring, operation and maintenance management, dispatching and operation, and customer service.