A power distribution network fault early warning method and system based on online monitoring

Abstract

The application discloses a power distribution network fault early warning method and system based on online monitoring, and relates to the field of fault early warning. The method comprises the following steps: continuously collecting real-time electrical signals; predicting the signal waveform of the real-time electrical signals according to the prediction result of the real-time signals, and judging whether early fault signals exist in the target power distribution network; if early fault signals exist in the target power distribution network, obtaining power distribution network information, environmental prediction data and geographic location data; constructing a power distribution network topology structure, and extracting early fault information of the target power distribution network; constructing an icing evolution model of the target power distribution network; coupling the fault risk between the icing evolution model and the early fault information; performing risk accumulation analysis on the power distribution network topology structure, and outputting the fault early warning information of the target power distribution network according to the risk accumulation analysis result. The application can effectively improve the fault early warning precision and efficiency of the power distribution network.

Description

Technical Field

[0001] This application relates to the field of fault early warning, and in particular to a method and system for early warning of distribution network faults based on online monitoring. Background Technology

[0002] As a crucial link connecting the transmission network and users in the power system, the stability of the distribution network directly affects the reliability and security of power supply. However, the distribution network is widely distributed, has a complex topology, and is susceptible to multiple factors such as natural environment (e.g., icing, lightning strikes, typhoons), equipment aging, and load fluctuations. Especially under extreme natural conditions, such as blizzards and typhoons, the probability of distribution network failures increases significantly. Therefore, it is necessary to monitor the distribution network in real time and provide early warning of failures to prevent the scope of failures from expanding and causing large-scale power outages.

[0003] Existing power distribution network fault early warning technologies mainly rely on single or multiple types of sensor data. By analyzing the sensor data, faults in the power distribution network can be identified in a simple way. However, this method is difficult to identify minute fault signals and may lead to missed faults. In addition, performing single-signal fault analysis on the power distribution network without coordinating with other factors may lead to errors in fault detection results, resulting in delayed fault information and failing to meet the needs of refined and intelligent operation and maintenance of the power distribution network. Summary of the Invention

[0004] This application provides a method and system for early warning of distribution network faults based on online monitoring, which solves the problem that existing technologies are unable to accurately and timely identify distribution network faults.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0006] Firstly, a method for early warning of distribution network faults based on online monitoring is provided, the method comprising:

[0007] Real-time electrical signals from all monitoring points in the target power distribution network are continuously collected by multiple types of sensors pre-installed in the target power distribution network.

[0008] For any monitoring point, real-time signal prediction is performed based on the real-time electrical signal of the monitoring point and using the extended Kalman filter algorithm. The real-time signal prediction results and preset early fault characteristics are combined to determine whether there are early fault signals in the target distribution network.

[0009] If there are early fault signals in the target distribution network, obtain the distribution network information of the target distribution network, as well as the environmental prediction data and geographical location data of the target distribution network location;

[0010] The distribution network topology is constructed based on the distribution network information, and the early fault information of the target distribution network is extracted by combining the distribution network topology and all early fault signals.

[0011] An icing evolution model for the target distribution network was constructed by combining distribution network information, environmental prediction data, and geographic location data.

[0012] By completing the fault risk coupling between the icing evolution model and early fault information, a set of distribution network risks for the distribution network topology is obtained.

[0013] Repeat the above steps to continuously acquire the distribution network risk set of the distribution network topology, perform risk accumulation analysis on the distribution network topology based on the continuously acquired distribution network risk set, and output fault warning information of the target distribution network based on the risk accumulation analysis results.

[0014] Optionally, determining whether there are early fault signals in the target distribution network based on the real-time electrical signals from the monitoring points and using the extended Kalman filter algorithm for real-time signal prediction, and combining the real-time signal prediction results with preset early fault characteristics, includes the following steps:

[0015] A second-order generalized integral phase-locked loop is used to synchronously sample the real-time electrical signals at the monitoring points, thereby obtaining the electrical sampling signals;

[0016] Based on the electrical sampling signal and using the extended Kalman filter algorithm, the prior estimation of the target distribution network at the current moment is completed to obtain the signal prediction state, which includes the signal prediction instantaneous value, signal prediction frequency, signal prediction phase and signal prediction amplitude.

[0017] Extract the signal sampling state of the electrical sampling signal and calculate the signal difference index between the signal sampling state and the signal prediction state;

[0018] The fault similarity between the signal sampling state and the preset early fault characteristics is calculated using a similarity formula.

[0019] By combining signal difference indicators and fault similarity, early fault identification of real-time electrical signals can be achieved;

[0020] If the real-time electrical signal is a normal electrical signal, then continue to collect the real-time electrical signal of the monitoring point;

[0021] If the real-time electrical signal is a suspected fault signal, then the secondary fault identification of the suspected fault signal is completed by injecting an auxiliary detection signal into the monitoring point, and the real-time electrical signal is determined to be an early fault signal based on the secondary fault identification result.

[0022] If the real-time electrical signal is an early fault signal, it indicates that there is an early fault signal in the target distribution network.

[0023] Optionally, secondary fault detection of suspected fault signals is performed by injecting auxiliary detection signals into the monitoring points, and the determination of whether the real-time electrical signal is an early fault signal based on the secondary fault detection results includes the following steps:

[0024] By injecting auxiliary detection signals into the monitoring points to enhance suspected fault signals, an electrical enhancement signal is obtained;

[0025] Modal decomposition of the electrical enhancement signal yields multiple intrinsic modes of the signal;

[0026] The signal mode energy of all inherent modes of the signal is calculated using the energy operator;

[0027] Weights are assigned to all signal intrinsic modes based on the signal modal energy, and the signals of all signal intrinsic modes after weight assignment are reconstructed to obtain the electrical reconstructed signal;

[0028] Calculate the signal difference between the electrical reconfiguration signal and the pre-acquired historical reference signal;

[0029] If the signal difference is less than or equal to the preset difference threshold, the real-time electrical signal is determined to be a normal electrical signal.

[0030] If the signal difference is greater than the difference threshold, the real-time electrical signal is determined to be an early fault signal.

[0031] Optionally, the distribution network information includes distribution network structure data and line load data, and the environmental prediction data includes predicted temperature, predicted humidity, predicted wind speed, predicted wind direction, predicted precipitation, and predicted freezing rain time.

[0032] Optionally, constructing a distribution network topology based on distribution network information and extracting early fault information of the target distribution network by combining the distribution network topology and all early fault signals includes the following steps:

[0033] The early fault locations of all early fault signals are located using a signal localization algorithm.

[0034] Based on the distribution network structure data, the target distribution network is abstracted into a distribution network topology. The nodes of the distribution network topology include power supply nodes, branch nodes, monitoring nodes, equipment nodes, and fault nodes, and the edges of the distribution network topology are the distribution lines of the target distribution network.

[0035] For any early fault signal, the topological fault features of the early fault signal are extracted based on the distribution network topology. The topological fault features include the fault branch coefficient, the fault reflection coefficient, and the fault load coefficient.

[0036] Basic fault features of early fault signals were extracted based on wavelet transform algorithm;

[0037] A fault feature tensor is constructed by stacking topological fault features and basic fault features;

[0038] The fault feature tensor is input into the pre-built signal fault identification model, and the early fault type of the target distribution network is output through the fault signal identification model. The signal fault identification model is built based on the residual network.

[0039] The early fault location and early fault type of the early fault signal are integrated into the early fault information of the target distribution network.

[0040] Optionally, extracting the basic fault features of early fault signals based on the wavelet transform algorithm includes the following steps:

[0041] Preprocess early fault signals;

[0042] The preprocessed early fault signal is decomposed by stationary wavelet transform to obtain several early fault sub-bands.

[0043] Each early fault sub-band is divided into several early fault segments at equal intervals.

[0044] The basic fault features of all early fault segments are extracted. These basic fault features include fault mean, fault variance, fault energy value, fault energy entropy, fault frequency entropy value, and fault entropy weight.

[0045] Optionally, constructing an icing evolution model for the target distribution network by combining distribution network information, environmental prediction data, and geographic location data includes the following steps:

[0046] The windward characteristics of the distribution lines in the target distribution network are determined based on the route of the distribution network lines and the predicted wind direction. The windward characteristics include lines against the wind, lines with the wind, and lines across the wind.

[0047] The target distribution network is divided into several distribution network areas based on the line load data and windward characteristics of the distribution lines;

[0048] A preliminary icing evolution model for each power distribution network area was constructed based on environmental prediction data.

[0049] For any primary icing evolution model, geographical location data, windward characteristics, and line load data are used as icing correction factors. The primary icing evolution model is then corrected using the icing correction factors to obtain the regional icing evolution model.

[0050] By combining the icing evolution models of all regions, the icing evolution model of the target distribution network is obtained.

[0051] Optionally, to couple the fault risk between the icing evolution model and early fault information to obtain the distribution network risk set of the distribution network topology, the following steps are included:

[0052] For any fault node in the distribution network topology, the fault evolution level and fault risk level of the fault node are determined according to the early fault type in the early fault information corresponding to the fault node, and an early fault risk value is assigned to the fault node in combination with the fault evolution level and fault risk level.

[0053] The icing thickness at the location corresponding to the fault node was calculated based on the icing evolution model.

[0054] The icing type of the target distribution network is determined by combining predicted temperature, predicted humidity and predicted wind speed. The icing type includes stable icing and unstable icing.

[0055] Assign line icing risk values ​​to fault nodes based on line icing thickness and icing type;

[0056] The topological centrality of faulty nodes is determined based on the distribution network topology, and a faulty topological risk value is assigned to the faulty nodes based on the topological centrality.

[0057] The node coupling risk value of the fault node is obtained by weighted fusion of early fault risk value, line icing risk value and fault topology risk value;

[0058] By integrating the node coupling risk values ​​of all faulty nodes, a distribution network risk set for the distribution network topology is obtained.

[0059] Optionally, performing risk accumulation analysis on the distribution network topology based on the continuously acquired distribution network risk set, and outputting fault early warning information for the target distribution network based on the risk accumulation analysis results includes the following steps:

[0060] The cumulative risk value of all faulty nodes in the distribution network topology is obtained by performing time integration on all faulty nodes based on the continuously acquired distribution network risk set.

[0061] When the cumulative risk value of any faulty node is greater than or equal to the preset risk threshold, a fault warning message for the target distribution network is output.

[0062] Secondly, this application provides a power distribution network fault early warning system based on online monitoring, comprising:

[0063] The memory is configured to store instructions; and

[0064] The processor is configured to retrieve instructions from memory and, when executing the instructions, to implement the online monitoring-based power distribution network fault early warning method according to the first aspect.

[0065] The above technical solution enables real-time monitoring of the target distribution network by continuously collecting real-time electrical signals, thus allowing for timely detection of early fault signals. To ensure the accuracy of early fault signal identification, for easily identifiable electrical fault signals, the extended Kalman filter algorithm is used to predict the signal prediction state of the target distribution network at the next moment, along with preset early fault characteristics for fault identification. For suspected fault signals that are difficult to determine, auxiliary detection signals are injected into the monitoring points to amplify the fault characteristics of the suspected fault signals, followed by secondary fault identification. This method ensures that minor faults in the target distribution network are not missed, further improving the accuracy of distribution network fault monitoring. Furthermore, when extreme weather such as blizzards occurs in the target distribution network's location, the ice thickness increases the risk of faults. The ice thickness is also affected by weather conditions and is constantly changing, potentially increasing. Therefore, it is necessary to construct an ice evolution model based on environmental prediction data, geographical location data, and distribution network information of the target distribution network. By combining early fault signals identified through online monitoring with icing evolution models for fault early warning of the target distribution network, risk omissions caused by single-dimensional fault monitoring can be avoided, thus improving the accuracy of distribution network risk early warning. In summary, this application can effectively improve the accuracy of distribution network risk early warning, provide a data foundation for the safe operation of the distribution network, ensure the safety of electricity use in people's daily lives and industrial production, and promote social stability and development.

[0066] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0067] Figure 1 A flowchart illustrating a power distribution network fault early warning method based on online monitoring, provided as an embodiment of this application;

[0068] Figure 2 A schematic flowchart illustrating the identification of early fault signals provided in an embodiment of this application;

[0069] Figure 3 This is a schematic diagram of a process for constructing an icing evolution model provided in an embodiment of this application. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0071] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0072] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0073] Figure 1 The illustration shows a flowchart of a distribution network fault early warning method based on online monitoring according to an embodiment of this application. Figure 1 As shown in the figure, this application provides a method for early warning of distribution network faults based on online monitoring. The method may include the following steps:

[0074] S101. Real-time electrical signals of all monitoring points in the target distribution network are continuously collected by multiple types of sensors pre-set in the target distribution network.

[0075] In this embodiment, real-time electrical signals include parameters such as three-phase current (RMS and peak values), three-phase voltage (RMS and phase angle), active power, reactive power, and power factor. Various types of sensors are used, including current sensors and voltage sensors. The selection of monitoring points is determined based on factors such as the importance of equipment in the target distribution network and the probability of historical faults. For example, substations (high and low voltage sides of transformers), line branch points (such as T-junctions), ring main units, and switch stations are key points in the target distribution network, requiring monitoring points to be set at corresponding locations. In addition, there are sections prone to faults that require more monitoring points. Furthermore, after acquiring real-time electrical signals, various types of sensors do not immediately upload them. Instead, the raw data for the first 10 minutes is cached by the front-end edge gateway of the target distribution network (such as edge computing devices installed in ring main units), and then uploaded in batches to the cloud database (such as the time-series database InfluxDB) at the second level. This avoids data loss due to network fluctuations. In addition, the monitoring system of the target distribution network also has a built-in sensor status diagnosis module to detect sensor offline and data drift in real time. If the current value is 0 for a long time or exceeds the range, an alarm is triggered and the system automatically switches to the backup sensor. If the main sensor of a branch line fails, redundant data interpolation from adjacent monitoring points is used to supplement the data. Through the monitoring system of the target distribution network, full-node monitoring and all-time signal acquisition of the corresponding target distribution network can be achieved, providing spatiotemporally continuous raw data for subsequent fault identification and risk assessment.

[0076] S102. For any monitoring point, real-time signal prediction is performed based on the real-time electrical signal of the monitoring point and using the extended Kalman filter algorithm. The real-time signal prediction result and the preset early fault characteristics are combined to determine whether there is an early fault signal in the target distribution network.

[0077] In this embodiment, a nonlinear state-space model is established based on the electrical sampling signal (which includes fundamental frequency, fundamental phase, fundamental amplitude, instantaneous voltage value, and instantaneous current value) obtained by sampling through a second-order generalized integral phase-locked loop. The signal prediction state is predicted through prior estimation, and the signal sampling state of the real-time electrical signal is extracted. The signal sampling state includes fundamental frequency, fundamental phase, fundamental amplitude, instantaneous voltage value, and instantaneous current value. Then, the difference between the signal sampling state and the signal prediction state is calculated, and all differences are integrated as a signal difference index. Common early faults include insulation degradation, arcing grounding, and poor contact. Early fault features are extracted from these early faults, including fault harmonic distortion rate, fault current growth rate, fault phase shift, and fault current slope. These features are integrated into a fault feature set, which is then vectorized to obtain a fault feature vector. Similarly, features corresponding to the signal sampling state are extracted, including harmonic distortion rate, current growth rate, current amplitude, and voltage phase shift. These are also integrated into a sampling feature set, which is then vectorized to obtain a sampling feature vector. The fault similarity between the fault feature vector and the sampling feature vector is calculated using cosine similarity.

[0078] Next, by combining signal difference indicators and fault similarity, early fault identification of the real-time electrical signal is completed, that is, determining whether the real-time electrical signal is a normal electrical signal, a suspected fault signal, or an early fault signal. If the early fault identification result of the real-time electrical signal is a normal electrical signal, it means that there is no fault, and no further fault warning steps are needed; online monitoring of the monitoring point can continue. If the early fault identification result of the real-time electrical signal is an early fault signal, the subsequent fault warning steps continue. If the early fault identification result of the real-time electrical signal is a suspected fault signal, a secondary fault identification is required after amplifying the real-time electrical signal by injecting an auxiliary detection signal.

[0079] In summary, the above steps can accurately identify early fault signals in the target distribution network, providing a data foundation for subsequent fault early warning. Furthermore, compared to fault prediction schemes relying on single thresholds (such as current exceeding limits or voltage fluctuation exceeding limits), the above steps provide a more accurate solution, significantly reducing false alarms or missed faults. Missed faults can lead to untimely subsequent fault warnings, causing irreparable losses, such as large-scale power outages due to fault propagation. False alarms can also have serious consequences, especially during blizzards. Environmental factors significantly increase the false alarm rate when using conventional methods for fault prediction, leading to incorrect warnings and potentially flawed decisions. For example, a normally functioning distribution network might trip due to a false alarm, causing a large-scale power outage and disrupting residents' lives. This is especially true during blizzards, when some residents may rely on electrical appliances for heating; incorrect tripping can severely impact their lives. Additionally, if personnel are dispatched to perform distribution network repairs in blizzards when repairs are unnecessary due to false alarms, it results in significant resource waste and could even cause unnecessary casualties. Therefore, it is necessary to accurately identify early faults in the target distribution network.

[0080] S103. If there are early fault signals in the target distribution network, obtain the distribution network information of the target distribution network, as well as the environmental prediction data and geographical location data of the target distribution network location.

[0081] In this embodiment, the distribution network information of the target distribution network includes distribution network structure data and line load data. Environmental prediction data includes predicted temperature, predicted humidity, predicted wind speed, predicted wind direction, predicted precipitation, and predicted freezing rain time. Geographical location data refers to the micro-topographical data of the target distribution network's location, such as high-altitude mountainous areas or inland arid regions. The distribution network structure data includes the attributes and connection relationships of the core components of the target substation, such as the name, number, and geographical coordinates (latitude and longitude) of the substation / switching station, the number, type, and installation location (to which the line segment or node belongs), and the plan layout and main wiring diagram used to guide the planning and construction of the target substation. Line load data refers to the current load transmitted by the distribution lines in the target distribution network, for example, a main line of the target distribution network transmits a current of more than 200A, and a branch line transmits a current of less than 50A. Environmental prediction data includes predicted temperature, predicted humidity, predicted wind speed, predicted wind direction, predicted precipitation, and predicted freezing rain time, which are obtained from the meteorological station where the target distribution network is located and can be queried through the official website of the local meteorological station.

[0082] S104. Construct the distribution network topology based on the distribution network information, and extract the early fault information of the target distribution network by combining the distribution network topology and all early fault signals.

[0083] In this embodiment, the target distribution network is abstracted into a distribution network topology based on the distribution network structure data. Power sources, various devices (e.g., switching equipment), distribution network branches, monitoring points, and fault points are considered as nodes in the distribution network topology, and the power lines connecting these nodes are considered as edges. Next, based on the distribution network topology, topological fault features of early fault signals are extracted. These features include fault branch coefficients, fault reflection coefficients, and fault load coefficients. Then, wavelet decomposition is performed on the early fault signals to extract basic faults such as fault mean, fault variance, fault energy value, fault energy entropy, fault frequency entropy value, and fault entropy weight. A two-dimensional matrix is ​​constructed based on the topological and basic fault features, and then stacked into a three-dimensional tensor to obtain the fault feature tensor. The fault feature tensor is then normalized and input into a pre-constructed signal fault identification model, which outputs the early fault type of the target distribution network. The fault signal identification model is built upon a residual network (ResNet), a deep convolutional neural network that addresses the vanishing gradient and network degradation problems of deep networks by introducing skip connections. Early fault information includes the location and type of the early fault signal.

[0084] Identifying early fault types in target power distribution systems through early fault signals helps quantify the risk weights of different faults. For example, insulation degradation deteriorates slowly, potentially taking months to years, while arcing grounding develops extremely rapidly, spreading and worsening within hours. This can lead to the burning of distribution network equipment, causing large-scale power outages, and even fires. Therefore, if the early fault type is arcing grounding, a modified risk weight needs to be assigned to it. Furthermore, accurately identifying early fault types and locations allows repair personnel to quickly pinpoint the fault and carry appropriate tools and spare parts (e.g., insulated gloves for short circuits, welding equipment for broken wires), reducing on-site investigation time, thus shortening the fault repair cycle, reducing power outage duration for users, and improving power supply reliability.

[0085] S105. By combining distribution network information, environmental prediction data and geographical location data, an icing evolution model of the target distribution network is constructed.

[0086] In this embodiment, both windward characteristics and line load data affect the icing evolution rate. Windward characteristics can be determined based on line alignment and predicted risks, and include downwind, crosswind, and headwind lines. Based on the line load data and windward characteristics, the target distribution network is divided into several distribution network areas, including headwind-main line area, headwind-branch line area, downwind-main line area, downwind-branch line area, crosswind-main line area, and crosswind-branch line area. A preliminary icing evolution model for each distribution network area is initially constructed based on environmental prediction data. Then, geographical location data, windward characteristics, and line load data are used as icing correction factors to correct the preliminary icing evolution model. Specifically, windward correction coefficients and load correction coefficients are set for the preliminary icing evolution model of each distribution network area based on windward characteristics and line load data. Micro-topographic parameters are set based on geographical location data. The preliminary icing evolution model is corrected using the windward correction coefficient, load correction coefficient, and micro-topographic parameters to obtain the regional icing evolution model. Finally, all regional icing evolution models are combined to obtain the icing evolution model of the target distribution network.

[0087] Icing is a typical hidden danger in power distribution networks (especially transmission lines, towers, insulators, etc.) under low temperature and high humidity environments. Changes in the thickness, density, and duration of icing can directly lead to excessive mechanical loads on lines, decreased insulation performance, and even serious faults such as line breaks, tower collapses, and flashovers. By constructing an icing evolution model, the development trend of icing risks in the target power distribution network can be dynamically quantified. Furthermore, if insulation degradation occurs in the target power distribution network (such as aging of insulator surfaces or damage to skirts), under non-icing conditions, it only leads to a decrease in local insulation resistance. However, under icing conditions, icing forms a conductive water film on the insulator surface. The electrolyte during ice melting accelerates the increase in leakage current in the degraded area, causing a significant reduction in insulation breakdown voltage (potentially from the normal 100kV to below 50kV). In this case, the coupled risk of icing and insulation degradation is far greater than the sum of the risks of the two individually. Therefore, in blizzard weather, the impact of icing on the target power distribution network needs to be considered, making it necessary to construct an icing evolution model.

[0088] S106. Complete the fault risk coupling between the icing evolution model and early fault information to obtain the distribution network risk set of the distribution network topology.

[0089] In this embodiment, the fault evolution level and fault risk level of each fault node are determined based on the early fault type in the early fault information corresponding to each fault node. Then, an early fault risk value is assigned to the fault node based on the fault evolution level and fault risk level. After that, the line icing thickness at the location corresponding to the fault node is calculated based on the icing evolution model. The icing type of the target distribution network is determined based on the predicted temperature, predicted humidity and predicted wind speed. The icing type includes stable icing and unstable icing. Therefore, after assigning an initial line icing risk value to the fault node based on the line icing thickness, a corresponding risk correction coefficient is set according to the icing type. The product of the initial line icing risk value and the risk correction coefficient is calculated to obtain the line icing risk value.

[0090] Next, the topological centrality of the faulty node is determined based on the distribution network topology. Topological centrality can be the degree centrality or betweenness centrality of the faulty node, etc. Calculating topological centrality reflects the influence range of the faulty node; the higher the topological centrality, the larger the influence range of the corresponding faulty node and the greater its impact on the target distribution network. Therefore, a higher fault topological risk value needs to be assigned. After assigning corresponding weights to the early fault risk value, line icing risk value, and fault topological risk value, they are weighted and fused to obtain the node coupling risk value of the faulty node. The weights can be optimized using the analytic hierarchy process (AHP) or machine learning. For example, if the proportion of power outage losses caused by line icing is high in historical data, its weight is increased. Integrating the node coupling risk values ​​of all faulty nodes yields the distribution network risk set of the distribution network topology.

[0091] S107. Repeat the above steps to continuously obtain the distribution network risk set of the distribution network topology, perform risk accumulation analysis on the distribution network topology based on the continuously obtained distribution network risk set, and output the fault warning information of the target distribution network based on the risk accumulation analysis results.

[0092] In this embodiment, fault monitoring of the target distribution network continues, and the above steps are repeated to continuously calculate whether early fault signals continue to appear at each fault point. The cumulative sum of node coupling risk values ​​of each fault node is calculated to obtain the node cumulative risk value of each fault node at each monitoring time point. When the node cumulative risk value of any fault node is greater than or equal to the preset risk threshold, fault warning information is output to remind maintenance personnel to carry out maintenance or perform tripping operations. The fault warning information includes information such as the early fault type and early fault location, so that maintenance personnel can quickly reach the maintenance point. When performing tripping operations, the power outage range can also be planned according to the early fault location to minimize the power outage range and ensure people's normal power supply.

[0093] In one embodiment, reference is made to Figure 2The process of determining whether an early fault signal exists in the target distribution network based on real-time electrical signals from monitoring points and using an extended Kalman filter algorithm for real-time signal prediction, combined with the real-time signal prediction results and preset early fault characteristics, includes the following steps:

[0094] S201. A second-order generalized integral phase-locked loop is used to complete the synchronous sampling of the real-time electrical signals of the monitoring points to obtain electrical sampling signals;

[0095] S202. Based on the electrical sampling signal and using the extended Kalman filter algorithm, perform a priori estimation of the target distribution network at the current moment to obtain the signal prediction state, wherein the signal prediction state includes the signal prediction instantaneous value, signal prediction frequency, signal prediction phase and signal prediction amplitude.

[0096] S203. Extract the signal sampling state of the electrical sampling signal and calculate the signal difference index between the signal sampling state and the signal prediction state;

[0097] S204. Calculate the fault similarity between the signal sampling state and the preset early fault characteristics using the similarity formula;

[0098] S205. Combine signal difference indicators and fault similarity to complete early fault identification of real-time electrical signals;

[0099] S206. If the real-time electrical signal is a normal electrical signal, continue to collect the real-time electrical signal of the monitoring point;

[0100] S207. If the real-time electrical signal is a suspected fault signal, then the secondary fault identification of the suspected fault signal is completed by injecting an auxiliary detection signal into the monitoring point, and the real-time electrical signal is determined to be an early fault signal based on the secondary fault identification result.

[0101] S208. If the real-time electrical signal is an early fault signal, it indicates that there is an early fault signal in the target distribution network.

[0102] In this embodiment, the second-order generalized integral phase-locked loop (SOGI-PLL) is a phase-locking technique that uses a second-order generalized integrator to generate two orthogonal signals (with the same frequency and amplitude) to achieve a 90° phase shift, thereby locking the phase difference between the input signal and the reference signal. Through SOGI-PLL, the fundamental frequency, fundamental phase, fundamental amplitude, and instantaneous voltage and current values ​​at the synchronous sampling moment of the real-time electrical signal can be accurately extracted, providing a synchronous and stable electrical sampling signal for subsequent fault prediction and avoiding waveform distortion or feature loss caused by asynchronous sampling. Then, a nonlinear state-space model is established based on the electrical sampling signal obtained from the SOGI-PLL (which includes the fundamental frequency, fundamental phase, fundamental amplitude, instantaneous voltage value, and instantaneous current value) to predict the prior state at the current moment, i.e., the signal prediction state. The signal prediction state includes the signal prediction instantaneous value, signal prediction frequency, signal prediction phase, and signal prediction amplitude.

[0103] The Extended Kalman Filter (EKF) algorithm is a state estimation method for nonlinear dynamic systems, achieving recursive filtering through local linearization of the nonlinear function. Its core steps include a prediction step and an update step. In the prediction step, based on the state estimate from the previous time step and the system dynamic model, the prior state estimate and error covariance matrix for the current time step are predicted. Specifically, the predicted value of the electrical sampled signal is calculated using the nonlinear state-space model, and the state transition matrix (system Jacobian matrix) is calculated to predict the propagation of the error covariance matrix. In the update step, the prior estimate is corrected using the actual electrical sampled signal at the current time step. First, the measured predicted value and the measurement Jacobian matrix are calculated, then the Kalman gain is calculated, and finally, the state estimate and error covariance matrix are updated. The Kalman gain determines the contribution weight of the measurement information to the state update; it depends on the relative magnitudes of the prediction error covariance and the measurement noise covariance. Through this prediction-update loop, the EKF algorithm can provide optimal estimates of the state and parameters, taking into account system nonlinearity and measurement noise. Predicting the prior state of the target distribution network at the current moment using the Extended Kalman Filter (EKF) algorithm provides a quantitative benchmark for whether subsequent signal sampling states deviate significantly. The EKF algorithm is essentially a normal operation model built based on current and historical signals. It predicts the normal electrical signal at the current moment and compares it with the actual real-time electrical signal. If the difference between the two is large, indicating a significant signal difference index, it suggests that the real-time electrical signal deviates from the normal electrical signal, meaning that an early fault is highly likely to exist in the real-time electrical signal.

[0104] The real-time electrical signal sampling state is extracted, including the fundamental frequency, fundamental phase, fundamental amplitude, instantaneous voltage value, and instantaneous current value. Then, the difference between the sampled signal state and the predicted signal state is calculated, and all differences are integrated as a signal difference index. Common early faults include insulation degradation, arcing grounding, and poor contact. Specific early fault characteristics include: Insulation degradation: High-frequency harmonics appear in the voltage waveform (e.g., 3rd and 5th harmonic distortion rate > 5%), and the effective current value rises slowly (daily growth rate > 1%). Arcing grounding: Intermittent spikes appear in the current waveform (amplitude 2-3 times the rated value, duration 10-50ms), and voltage phase shift > 5°. Poor contact: "Glitches" appear in the current waveform (abrupt changes in rising / falling edge slope), and frequency fluctuations > 0.2Hz. Early fault features of the aforementioned early faults are extracted, namely, the fault harmonic distortion rate, fault current growth rate, fault phase shift, and fault current slope corresponding to each early fault. These features are integrated into a fault feature set, which is then vectorized to obtain a fault feature vector. Features corresponding to the signal sampling state are also extracted, namely, harmonic distortion rate, current growth rate, current amplitude, and voltage phase shift. These features are also integrated into a sampling feature set, which is then vectorized to obtain a sampling feature vector. The fault similarity between the fault feature vector and the sampling feature vector is calculated using cosine similarity.

[0105] Next, by combining signal difference indicators and fault similarity, early fault identification of real-time electrical signals is completed, that is, determining whether the real-time electrical signal is a normal electrical signal, a suspected fault signal, or an early fault signal. Specifically, if the signal difference indicator shows that the difference between the signal sampling state and the signal prediction state of the real-time electrical signal is less than the minimum value of the corresponding difference interval, for example, the voltage difference between the instantaneous voltage value and the instantaneous voltage value in the signal prediction is less than the minimum value of the preset voltage interval (the voltage interval is [U, 2U], and the minimum value of the voltage interval is U), and at the same time, the fault similarity is less than the minimum value of the corresponding similarity interval, that is, none of the characteristics of the signal sampling state meet the characteristics of an early fault. If both conditions are fully met, it indicates that the real-time electrical signal is a normal electrical signal and no early fault has occurred. The multi-type sensors at the corresponding monitoring point continue to perform online monitoring. If the difference between any one or more data points in the signal sampling state and the signal prediction state is greater than the maximum value of the corresponding difference interval, for example, the difference between the instantaneous current value and the instantaneous current value in the signal prediction is greater than the maximum value of the preset current interval, the current interval is [I, 2I], and the maximum value of the current interval is 2I, and at the same time, the fault similarity between the signal sampling state and any early fault feature is greater than the maximum value of the corresponding similarity interval, that is, the signal sampling state conforms to a certain early fault feature, then it is indicated that the real-time electrical signal is an early fault signal.

[0106] If the difference between any one or more data points in the signal sampling state and the signal prediction state is within the corresponding difference interval, and the difference between other data points in the signal sampling state and the signal prediction state is less than the minimum value of the corresponding difference interval, or if the fault similarity between the signal sampling state and any early fault feature is within the similarity interval, and the fault similarity between the signal sampling state and other early fault features is less than the minimum value of the corresponding similarity interval, then the real-time electrical signal is located between the normal electrical signal and the early fault signal. It is difficult to directly determine whether the real-time electrical signal has an early fault. This is because there are some minor faults in the distribution network. These minor faults may also be accompanied by arcing of the power grid, causing harmonic disturbances, which may lead to damage to the insulation of the distribution network lines or cause major fire accidents. Therefore, minor faults also need to be paid attention to and should not be ignored as normal electrical signals just because the fault characteristics are not obvious. Of course, suspected fault signals may not be caused by a fault, but by external interference. Especially in extreme snowstorms, the power distribution network may be affected by weather, and voltage and current may fluctuate briefly, which is not a fault. In order to avoid false faults and trigger unnecessary protection actions (such as circuit breaker tripping), which would affect the normal power supply of residents or businesses, secondary fault identification is required. If the secondary fault identification result shows that the real-time electrical signal is an early fault signal, the subsequent fault warning steps will continue. If the secondary fault identification result shows that the real-time electrical signal is a normal electrical signal, it means that there is no fault and there is no need to take the subsequent fault warning steps. The monitoring point can continue to be monitored online.

[0107] In summary, the above steps can accurately identify early fault signals in the target distribution network, providing a data foundation for subsequent fault early warning. Furthermore, compared to fault prediction schemes relying on single thresholds (such as current exceeding limits or voltage fluctuation exceeding limits), the above steps provide a more accurate solution, significantly reducing false alarms or missed faults. Missed faults can lead to untimely subsequent fault warnings, causing irreparable losses, such as large-scale power outages due to fault propagation. False alarms can also have serious consequences, especially during blizzards. Environmental factors significantly increase the false alarm rate when using conventional fault prediction methods, leading to incorrect warnings and potentially flawed decisions. For example, a normally functioning distribution network might trip due to a false alarm, causing a large-scale power outage and disrupting residents' lives. Especially during blizzards, some residents may rely on electrical appliances for heating, and a mistaken trip could severely impact their lives. Additionally, if personnel are dispatched to repair distribution network facilities in blizzards when repairs are unnecessary due to false alarms, it results in significant resource waste and could even cause unnecessary casualties. Therefore, it is necessary to accurately identify early faults in the target distribution network.

[0108] In one embodiment, secondary fault detection of suspected fault signals is completed by injecting auxiliary detection signals into the monitoring point, and the determination of whether the real-time electrical signal is an early fault signal based on the secondary fault detection results includes the following steps:

[0109] By injecting auxiliary detection signals into the monitoring points to enhance suspected fault signals, an electrical enhancement signal is obtained;

[0110] Modal decomposition of the electrical enhancement signal yields multiple intrinsic modes of the signal;

[0111] The signal mode energy of all inherent modes of the signal is calculated using the energy operator;

[0112] Weights are assigned to all signal intrinsic modes based on the signal modal energy, and the signals of all signal intrinsic modes after weight assignment are reconstructed to obtain the electrical reconstructed signal;

[0113] Calculate the signal difference between the electrical reconfiguration signal and the pre-acquired historical reference signal;

[0114] If the signal difference is less than or equal to the preset difference threshold, the real-time electrical signal is determined to be a normal electrical signal.

[0115] If the signal difference is greater than the difference threshold, the real-time electrical signal is determined to be an early fault signal.

[0116] In this embodiment, actively injecting auxiliary detection signals that do not affect the normal operation of the distribution network can amplify the electrical quantity differences of suspected fault signals, such as voltage amplitude and harmonic content, making subtle fault characteristics easier to detect. During normal operation of the distribution network, the voltage at the beginning and end of the line is symmetrically distributed. After the auxiliary detection signal is injected, due to the symmetry of the line parameters (impedance, admittance), the signal is canceled out and does not change the normal voltage amplitude. However, when a minor fault occurs, the fault point forms an asymmetrical impedance (such as high-resistance grounding or cable insulation defects), disrupting the line symmetry. At this time, the injected auxiliary detection signal interacts with the fault point, causing a significant increase in the voltage difference between the beginning and end of the faulty line, and amplifying fault characteristics (such as nonlinearity caused by arcing and harmonic components). Therefore, if the suspected fault signal is an early-stage fault signal, injecting an auxiliary detection signal can enhance the suspected fault signal, facilitating subsequent fault identification. Generally, the selection of the auxiliary detection signal needs to follow relevant regulations. Typically, a signal with a peak voltage of 3% to 7% of the rated voltage and a frequency of 50Hz is selected as the auxiliary detection signal to ensure that the fault characteristics of the suspected fault signal are amplified to the greatest extent possible without affecting power quality. The auxiliary detection signal can be generated using a multilevel converter pre-installed at the monitoring point.

[0117] After injecting auxiliary detection signals into the monitoring point to enhance the suspected fault signals, an electrical enhancement signal is obtained. The electrical enhancement signal is then subjected to mode decomposition, which can be variational mode decomposition (VMD), ensemble empirical mode decomposition (EEMD), complementary ensemble empirical mode decomposition (CEEMD), etc. The main purpose is to decompose the electrical enhancement signal into multiple intrinsic mode functions, i.e., signal intrinsic modes, and to separate the fault-related high-frequency modes (such as IMF2-IMF4) from the low-frequency interference modes (such as the fundamental frequency 50Hz) to separate the fault characteristics and interference signals in the electrical enhancement signal, providing interference-free data input for subsequent calculation of signal mode energy.

[0118] The signal modal energy (EME) of all inherent modes of a signal is calculated using energy operators. Commonly used energy operators include the Teager energy operator (TEO) and the Hilbert-Huang transform (HHT) energy operator. Taking the Teager energy operator as an example, for each inherent mode of a signal, each sampling point is traversed, i.e., the signal value at different times. The instantaneous energy of each sampling point is calculated by squaring the signal value corresponding to that sampling point and then subtracting the product of the signal values ​​between the previous and next sampling points. This gives the instantaneous energy of that sampling point. The instantaneous energy of all sampling points in each inherent mode of a signal is calculated in the same way. After normalizing all the instantaneous energies, they are integrated into an energy sequence, i.e., the signal modal energy. The signal modal energy represents the energy change of the inherent mode of the signal over the entire time range. By calculating the signal modal energy of all inherent modes of a signal, the energy characteristics of a fault can be quantified, and the difference between normal signals and fault signals can be more intuitively distinguished.

[0119] The distribution probability of instantaneous energy in each energy sequence (signal mode energy) is statistically analyzed. This involves dividing the instantaneous energy into multiple equidistant intervals, counting the frequency of all instantaneous energies in each energy sequence across different intervals, and calculating their probability distribution. This probability distribution is then input into the Shannon formula to calculate information entropy. Weights are assigned to all signal intrinsic modes based on the information entropy, with higher weights allocated to intrinsic modes with higher information entropy. Finally, all intrinsic modes are weighted and fused to obtain the reconstructed electrical signal. This is because normal signal intrinsic modes have energy stably concentrated in a few equidistant intervals, resulting in a concentrated probability distribution and low information entropy. To suppress their interference with subsequent fault identification steps, their proportion should be minimized during signal reconstruction, thus reducing their weight allocation. Conversely, fault-related signal intrinsic modes, due to sudden energy changes during faults, have a wider instantaneous energy distribution and a more dispersed probability distribution, resulting in higher calculated information entropy. Therefore, higher weights should be allocated to fault-related signal intrinsic modes with higher information entropy.

[0120] The pre-acquired historical reference signal refers to the historical electrical signal of the distribution network under normal operating conditions, collected at the same monitoring points beforehand. After performing the same signal decomposition and reconstruction processes on the historical reference signal, a reference reconstructed signal is obtained. Then, the signal difference between the two is calculated. The signal difference can be measured by the Wasserstein distance (WD), Jensen-Shannon divergence (JSD), or Bartholomew's distance, all of which can measure the difference between the two. Taking the Wasserstein distance (WD) as an example, it is a mathematical tool used to measure the difference between two probability distributions. Its core idea originates from optimal transport theory. It quantifies the difference by calculating the minimum cost required to "transport" one distribution to another. Euclidean distance can be chosen as the basic distance between sample points of two probabilities to quantify the distance cost of "transporting" a single sample from one distribution to another, finding the transport scheme with the minimum total distance cost. This minimum total distance cost is the Wasserstein distance between the reference reconstructed signal and the electrical reconstructed signal, which can also be called the signal difference between the two. If the signal difference is less than or equal to a preset difference threshold, the real-time electrical signal is determined to be a normal electrical signal; if the signal difference is greater than the difference threshold, the real-time electrical signal is determined to be an early fault signal. This scheme can accurately identify minute early fault signals, providing data support for subsequent fault early warning.

[0121] In one embodiment, the power distribution network information includes power distribution network structure data and line load data, and the environmental prediction data includes predicted temperature, predicted humidity, predicted wind speed, predicted wind direction, predicted precipitation, and predicted freezing rain time.

[0122] In this embodiment, the power distribution network structure data includes the attributes and connection relationships of the core components of the target power distribution station, such as the name, number, and geographical coordinates (latitude and longitude) of the substation / switching station, the number, type, and installation location (to which the line segment or node belongs), as well as the layout plan and main wiring diagram used to guide the planning and construction of the target power distribution station. Line load data refers to the current load transmitted by the distribution lines in the target power distribution network; for example, a main line in the target power distribution network transmits a current of over 200A, while a branch line transmits a current of less than 50A. Environmental prediction data includes predicted temperature, predicted humidity, predicted wind speed, predicted wind direction, predicted precipitation, and predicted freezing rain time, which are obtained from the meteorological station where the target power distribution network is located and can be queried through the official website of the local meteorological station.

[0123] In one embodiment, constructing a distribution network topology based on distribution network information and extracting early fault information of the target distribution network by combining the distribution network topology and all early fault signals includes the following steps:

[0124] The early fault locations of all early fault signals are located using a signal localization algorithm.

[0125] Based on the distribution network structure data, the target distribution network is abstracted into a distribution network topology. The nodes of the distribution network topology include power supply nodes, branch nodes, monitoring nodes, equipment nodes, and fault nodes, and the edges of the distribution network topology are the distribution lines of the target distribution network.

[0126] For any early fault signal, the topological fault features of the early fault signal are extracted based on the distribution network topology. The topological fault features include the fault branch coefficient, the fault reflection coefficient, and the fault load coefficient.

[0127] Basic fault features of early fault signals were extracted based on wavelet transform algorithm;

[0128] A fault feature tensor is constructed by stacking topological fault features and basic fault features;

[0129] The fault feature tensor is input into the pre-built signal fault identification model, and the early fault type of the target distribution network is output through the fault signal identification model. The signal fault identification model is built based on the residual network.

[0130] The early fault location and early fault type of the early fault signal are integrated into the early fault information of the target distribution network.

[0131] In this embodiment, the signal localization algorithm includes impedance method, traveling wave method, and injection signal method. For early fault signals that are directly identified without secondary fault identification, the impedance method can be used for fault localization. The principle of the impedance method is that after a fault occurs, the fault point forms a loop with the power supply and monitoring point, and the loop impedance is positively correlated with the distance to the fault point (the farther the distance, the greater the line impedance). By calculating the fault loop impedance, the fault distance can be deduced to obtain the early fault location. The fault loop impedance can be calculated from the voltage and current of the early fault signal. If the early fault signal is identified through secondary fault identification, the injection signal method can be used for fault localization. This is because an auxiliary signal of a specific frequency (such as a low-frequency pulse) is injected into the faulty line, and the signal will be reflected at the fault point (the impedance of the fault point is different from that of the normal line). The product of the emission time of the reflected signal and the signal propagation speed is calculated to obtain the fault distance between the monitoring point and the fault point, thereby determining the early fault location of the early fault signal.

[0132] Next, based on the distribution network structure data, the target distribution network is abstracted into a distribution network topology. The power sources, various equipment (e.g., switching equipment), distribution branches, monitoring points, and fault points of the target distribution network are considered as nodes in the distribution network topology, and the distribution lines connecting these nodes are considered as edges. Then, based on the topological fault characteristics of early fault signals, the fault branch coefficient refers to the proportion of early fault signals shunted at branch nodes, i.e., the ratio of the "signal amplitude flowing to the monitoring point" to the "total signal amplitude at the fault point." When the main line fault occurs, the branch coefficient is close to 1 (no shunting), while when the branch line fault occurs, the branch coefficient is <1 (part of the signal is shunted by other branches). The fault reflection coefficient refers to the proportion of early fault signals reflected at branch nodes (the reflection phenomenon when the signal encounters a sudden impedance change). The stronger the reflection, the more obvious the "oscillation characteristic" of the monitoring point signal. It can be calculated based on the impedances on both sides of the branch node closest to the fault point. Z1 represents the impedance on one side of the branch node, Z2 represents the impedance on the other side of the branch node, and the reflection coefficient r = |(Z1-Z2) / (Z1+Z2)|. The fault load factor refers to the proportion of change in load current before and after the fault point, used to distinguish between faults and normal load fluctuations. Then, wavelet decomposition is performed on the early fault signal to extract basic fault parameters such as fault mean, fault variance, fault energy value, fault energy entropy, fault frequency entropy value, and fault entropy weight.

[0133] After constructing a two-dimensional matrix based on topological fault features and basic fault features, the two-dimensional matrix is ​​stacked into a three-dimensional tensor to obtain the fault feature tensor. Then, the fault feature tensor is normalized and input into a pre-constructed signal fault identification model, which outputs the early fault type of the target distribution network.

[0134] The steps for constructing a signal fault identification model based on residual networks include: The signal fault identification model includes an input layer, a feature extraction layer, and a classification layer. The input layer adapts the fault feature tensor and is used to input the fault feature tensor into the feature extraction layer. The feature extraction layer is constructed based on a multi-channel fusion convolutional layer and multiple improved residual modules (IRBs) stacked together. The multi-channel fusion convolutional layer is used to fuse the initial features of multiple channels, i.e., the fault feature tensor. Multiple improved residual modules (IRBs) are used for feature depth mining. The classification layer includes a pooling layer, a channel attention module, a fully connected layer, and an output layer. The probability distribution of fault types is obtained through the Softmax function, and the fault type prediction result is output.

[0135] Residual Networks (ResNet) are deep convolutional neural networks that address the vanishing gradient and network degradation problems of deep networks by introducing skip connections. After constructing a signal fault identification model based on the ResNet, a large number of fault samples, such as 15,000, are generated using PSCAD simulation. These samples need to cover most early fault types, such as insulation degradation, arcing grounding, and poor contact, with the same number of samples for each type. The fault samples are then divided into training, validation, and test sets according to a certain ratio. After constructing feature tensors from these sets using the same steps, they are normalized, and each early fault type is converted to one-hot encoding to adapt to the cross-entropy loss function. Next, training parameters are set, such as setting the loss function to cross-entropy loss, the optimizer to the Adam optimizer, the batch size to 32, and the training epochs to 50. The network parameters of the signal fault identification model are initialized first. Then, the training set is input into the network in batches, passing through the feature extraction layer and classification layer, outputting the predicted probability distribution. Next, the cross-entropy loss between the predicted probability and the true label is calculated. Through gradient descent (Adam optimizer) and backpropagation, all trainable parameters (convolutional kernel weights, BN layer parameters, fully connected layer weights, etc.) are updated. After each training epoch, the model performance (accuracy, loss value) is evaluated using a validation set. If the validation accuracy does not improve for five consecutive epochs, training is stopped (early stopping strategy) to avoid overfitting. When the preset number of training epochs is reached or the test set accuracy reaches a preset threshold, model training is complete, resulting in a trained signal fault identification model. This trained model is then used for early fault type identification. Finally, the early fault location and early fault type of the early fault signal are integrated into the early fault information of the target distribution network.

[0136] In one embodiment, extracting the basic fault features of early fault signals based on the wavelet transform algorithm includes the following steps:

[0137] Preprocess early fault signals;

[0138] The preprocessed early fault signal is decomposed by stationary wavelet transform to obtain several early fault sub-bands.

[0139] Each early fault sub-band is divided into several early fault segments at equal intervals.

[0140] The basic fault features of all early fault segments are extracted. These basic fault features include fault mean, fault variance, fault energy value, fault energy entropy, fault frequency entropy value, and fault entropy weight.

[0141] In this embodiment, the early fault signal is first denoised and standardized to eliminate interference from residual injected auxiliary signals or power grid background noise. Standardization is performed to avoid the impact of dimensional differences on subsequent model recognition. Denoising can be done using wavelet thresholding, and standardization can be done using maximum-minimum standardization. Next, the preprocessed early fault signal is decomposed using a stationary wavelet transform (SWT). For example, the db4 wavelet can be used to decompose the signal into 5 levels, resulting in one level of low-frequency approximation coefficients (reflecting the fundamental frequency component) and five levels of high-frequency detail coefficients (reflecting transient change characteristics). The early fault signal is a redundant wavelet transform method. Unlike the traditional Discrete Wavelet Transform (DWT), it does not perform downsampling, thus exhibiting translation invariance (i.e., small temporal shifts in the signal do not lead to drastic changes in the decomposition results). This makes SWT more stable in tasks such as signal denoising, feature extraction, and change detection. The decomposition steps include selecting wavelet basis functions (such as Haar, Daubechies, Symlet, etc.), choosing an appropriate number of decomposition levels, and then using the wavelet basis functions to decompose the early fault signal layer by layer, obtaining several early fault sub-bands. Each early fault sub-band is equally spaced into several early fault segments, and the basic fault characteristics of each segment are extracted. These basic fault characteristics include fault mean, fault variance, fault energy value, fault energy entropy, fault frequency entropy value, and fault entropy weight. The fault mean measures the central tendency of all sampled values ​​(values ​​obtained by discretizing continuous voltage, current, and other signals of the early fault signal) in the early fault segment, reflecting the stability of the early fault signal. The fault variance measures the dispersion of the sampled values, reflecting the volatility of the early fault signal. The fault energy value characterizes the energy distribution of the early fault signal in that segment and can represent the signal strength. The fault energy value can be obtained by calculating the sum of the squares of all sampled values ​​within the segment. The mean and variance are existing technologies and will not be described in detail here. The fault energy entropy characterizes the degree of disorder in the energy distribution within the segment and can be used to distinguish between noise (high entropy) and fault (low entropy). The formula for calculating the fault energy entropy is as follows:

[0142]

[0143] in, Indicates the fault energy value. Indicates the first In the first early fault segment Each sample value, This represents the total number of sampled values.

[0144] The fault frequency entropy value reflects the randomness of the frequency distribution within the early fault segment, used to identify complex faults (such as high-frequency arc oscillations). The formula for calculating the fault frequency entropy value is as follows:

[0145]

[0146] in, The normalized power spectral density is calculated by performing a Fourier transform on the early fault segment. Power spectral density Next, the power spectral density was analyzed. After normalization, we can obtain , This indicates the frequency point of the early fault segment after Fourier transform.

[0147] The fault entropy weight represents the importance of the fault frequency entropy value within the early fault segment. The formula for calculating the fault entropy weight is as follows:

[0148]

[0149] in, Indicates the first Within the first early fault subband The fault energy entropy of an early fault segment.

[0150] In one embodiment, reference is made to Figure 3 The icing evolution model of the target distribution network, which combines distribution network information, environmental prediction data, and geographical location data, includes the following steps:

[0151] S301. Determine the windward characteristics of the power distribution lines in the target power distribution network based on the route of the power distribution network lines and the predicted wind direction. The windward characteristics include lines against the wind, lines with the wind, and lines across the wind.

[0152] S302. Divide the target distribution network into several distribution network areas based on the line load data and windward characteristics of the distribution lines.

[0153] S303. Based on environmental prediction data, a preliminary primary icing evolution model for each power distribution network area is constructed.

[0154] S304. For any primary icing evolution model, use geographical location data, windward characteristics and line load data as icing correction factors, and use the icing correction factors to correct the primary icing evolution model to obtain the regional icing evolution model.

[0155] S305. By combining the icing evolution models of all regions, the icing evolution model of the target distribution network is obtained.

[0156] In this embodiment, even within the same power distribution network, the direction of the power line can affect the rate of icing evolution, even if the geographical terrain is similar. When the power distribution network line is perpendicular to the predicted wind direction, its windward characteristic is a crosswind line. At this time, the airflow impacts the conductor vertically, and the supercooled water droplets almost "collide head-on" with the surface of the power distribution network conductor, resulting in the highest impact probability. The icing growth rate is faster than that of the downwind line. The downwind line refers to the power distribution network line being parallel to the predicted wind direction. The airflow flows along the axial direction of the power distribution conductor, and the water droplets mostly skim over the side of the conductor, with only the edge parts being captured. Therefore, the icing growth rate is slower. There are also some special cases. When the target power distribution network is located in special terrain, such as when the line is erected uphill along a hillside, against the downhill wind, the power distribution conductor is a upwind line. The airflow and the conductor form an "angle of attack," and the water droplet impact efficiency is between that of the crosswind and the downwind line. The icing evolution rate is also between that of the downwind and the crosswind line.

[0157] Line load data also affects the icing rate of distribution network conductors. For main lines with heavy loads (such as conductors with a load data greater than 200A), the conductor resistance heats up significantly, and its surface temperature is higher than the ambient temperature, resulting in a lower icing rate. Conversely, for branch lines with light loads (such as conductors with a load data less than 50A), less heat is generated, and the surface temperature is close to the ambient temperature, leading to a higher icing rate compared to the main lines. Based on the line load data and windward characteristics of the distribution lines, the target distribution network is divided into several distribution network areas, including the upwind-main line area, upwind-branch line area, downwind-main line area, downwind-branch line area, crosswind-main line area, and crosswind-branch line area.

[0158] Next, based on environmental prediction data, a preliminary primary icing evolution model was constructed for each power distribution network area. The primary icing evolution model is as follows:

[0159]

[0160] in, This refers to the icing density, which can be queried based on the icing type. For the density of water, For distribution network area The predicted precipitation To predict the timing of freezing rain, For distribution network area Predicted wind speed, For distribution network area The liquid water content in saturated air, .

[0161] Next, geographical location data, windward characteristics, and line load data are used as icing correction factors to modify the primary icing evolution model. Specifically, windward correction coefficients and load correction coefficients are set for the primary icing evolution model of each distribution network area based on windward characteristics and line load data. For example, the correction coefficient for crosswind lines is set to 1, the correction coefficient for downwind lines is set to 0.4, and the correction coefficient for upwind lines is set to 0.8. The windward correction coefficient can be set by measuring the water droplet impact efficiency at different angles through wind tunnel experiments. Then, the load correction coefficient is set based on the line load data. Since lightly loaded branch lines generate less heat and their surface temperature is close to the ambient temperature, the load correction coefficient for branch lines can be set to 1. Heavy-load main line conductors generate significant resistive heat, and their surface temperature is higher than the ambient temperature; therefore, their load correction coefficient needs to be appropriately reduced, for example, set to 0.8. In addition, geographical location data also affects icing speed. For example, special terrain such as wind gaps or windward slopes will increase wind speed and thicken icing, while leeward slopes or valleys will have the opposite effect. Therefore, for plain terrain, the micro-topographical parameter set based on geographical location data can be 1, with 1.0~1.3 for terrain that increases wind speed and 0.8~1.0 for terrain that decreases wind speed. The regional icing evolution model is as follows:

[0162]

[0163] in, For distribution network area The windward correction factor, For distribution network area The load correction factor, For distribution network area The micro-topographic parameters are obtained. By combining the icing evolution models of the entire region, the icing evolution model of the target power distribution network is obtained.

[0164] Icing is a typical hidden danger in power distribution networks (especially transmission lines, towers, insulators, etc.) under low temperature and high humidity environments. Changes in the thickness, density, and duration of icing can directly lead to excessive mechanical loads on lines, decreased insulation performance, and even serious faults such as line breaks, tower collapses, and flashovers. By constructing an icing evolution model, the development trend of icing risks in the target power distribution network can be dynamically quantified. Furthermore, if insulation degradation occurs in the target power distribution network (such as aging of insulator surfaces or damage to skirts), under non-icing conditions, it only leads to a decrease in local insulation resistance. However, under icing conditions, icing forms a conductive water film on the insulator surface. The electrolyte during ice melting accelerates the increase in leakage current in the degraded area, causing a significant reduction in insulation breakdown voltage (potentially from the normal 100kV to below 50kV). In this case, the coupled risk of icing and insulation degradation is far greater than the sum of the risks of the two individually. Therefore, in blizzard weather, the impact of icing on the target power distribution network needs to be considered, making it necessary to construct an icing evolution model.

[0165] In one embodiment, coupling the fault risk between the icing evolution model and early fault information to obtain the distribution network risk set of the distribution network topology includes the following steps:

[0166] For any fault node in the distribution network topology, the fault evolution level and fault risk level of the fault node are determined according to the early fault type in the early fault information corresponding to the fault node, and an early fault risk value is assigned to the fault node in combination with the fault evolution level and fault risk level.

[0167] The icing thickness at the location corresponding to the fault node was calculated based on the icing evolution model.

[0168] The icing type of the target distribution network is determined by combining predicted temperature, predicted humidity and predicted wind speed. The icing type includes stable icing and unstable icing.

[0169] Assign line icing risk values ​​to fault nodes based on line icing thickness and icing type;

[0170] The topological centrality of faulty nodes is determined based on the distribution network topology, and a faulty topological risk value is assigned to the faulty nodes based on the topological centrality.

[0171] The node coupling risk value of the fault node is obtained by weighted fusion of early fault risk value, line icing risk value and fault topology risk value;

[0172] By integrating the node coupling risk values ​​of all faulty nodes, a distribution network risk set for the distribution network topology is obtained.

[0173] In this embodiment, the fault evolution level and fault risk level of each fault node are determined based on the early fault type in the early fault information corresponding to each fault node. For example, insulation degradation evolves slowly, generally taking months to years from the onset of insulation degradation to circuit breakdown. Arc grounding evolves relatively quickly, generally taking only minutes to hours from its occurrence to triggering a phase-to-phase short circuit. Poor contact degradation also evolves slowly, taking weeks to months from its early stage to a severe fault (such as joint burnout). Therefore, fault evolution levels can be assigned to different types of early faults, with levels ranging from 1 to 5, where level 1 is the slowest and level 5 is instantaneous. Then, based on the consequences of different types of early faults, such as power outage range, maintenance costs, and safety risks, they are classified into low risk → medium risk → high risk → emergency risk. For example, early insulation degradation has minimal impact on the target distribution network and is therefore considered low risk. The fault risk level of different types of early faults can also be determined using methods such as analytic hierarchy process (AHP), Bayesian networks, or fuzzy logic. Next, early fault risk values ​​are assigned to the fault node based on the fault evolution level and fault risk level. For example, evolution risk values ​​and fault risk values ​​are assigned to the fault node according to the fault evolution level and fault risk level, respectively. For example, fault evolution levels 1-5 are scored from 1 to 5 points, and low risk to emergency risk are scored from 2 to 4 to 6 to 8 points, respectively. Then, the evolution risk values ​​and fault risk values ​​are weighted and summed to obtain the early fault risk value of the fault node. Next, the icing thickness of the line at the location of the fault node is calculated according to the icing evolution model. It is only necessary to determine which distribution network area the fault node is located in, and the corresponding line icing thickness can be calculated using the corresponding regional icing evolution model. Next, the icing type of the target power distribution network is determined by combining the predicted temperature, predicted humidity, and predicted wind speed. Icing types include stable icing and unstable icing. Stable icing refers to icing types with strong adhesion that are not easy to fall off, such as rime icing and snow icing. Unstable icing refers to icing types with weak adhesion that are easy to fall off, such as hoarfrost icing and mixed rime icing. The formation conditions for rime icing are predicted temperature: -5-0℃, predicted humidity: 90%, and predicted wind speed: 3-15m / s. The formation conditions for snow icing are predicted temperature: -10-0℃ and predicted wind speed: relatively low. The formation conditions for hoarfrost icing are predicted temperature: -10--3℃ and -13--8℃. The formation conditions for mixed rime icing are temperature: -8--2℃ and predicted wind speed: 2-5m / s. Because stable icing is not easy to detach and tends to accumulate, it has a greater impact on the target distribution network. Unstable icing is easy to detach, so it may detach when its thickness increases to a certain extent. Therefore, compared with stable icing, it has a smaller impact on the target distribution network.Therefore, after assigning an initial line icing risk value to the fault node based on the line icing thickness, a corresponding risk correction coefficient is set according to the icing type. For example, the risk correction coefficient for stable icing is 1.2, and the risk correction coefficient for unstable icing is 0.8. The product of the initial line icing risk value and the risk correction coefficient is calculated to obtain the line icing risk value.

[0174] Next, the topological centrality of the faulty node is determined based on the distribution network topology. Topological centrality can be the degree centrality or betweenness centrality of the faulty node. Degree centrality refers to the number of nodes directly connected to the faulty node, while betweenness centrality refers to the frequency with which the faulty node appears on the shortest path between other nodes. Calculating topological centrality reflects the influence range of the faulty node; the higher the topological centrality, the greater the influence range and the greater the impact on the target distribution network. Therefore, a higher fault topological risk value needs to be assigned. After assigning corresponding weights to the early fault risk value, line icing risk value, and fault topological risk value, they are weighted and fused to obtain the node coupling risk value of the faulty node. The weights can be optimized using the analytic hierarchy process (AHP) or machine learning. For example, if the proportion of power outage losses caused by line icing is high in historical data, its weight should be increased. Integrating the node coupling risk values ​​of all faulty nodes yields the distribution network risk set of the distribution network topology. Historical data refers to the data accumulated by the target distribution network before this point, including historical fault records and historical maintenance records.

[0175] In one embodiment, performing risk accumulation analysis on the distribution network topology based on the continuously acquired distribution network risk set, and outputting fault early warning information for the target distribution network based on the risk accumulation analysis results includes the following steps:

[0176] The cumulative risk value of all faulty nodes in the distribution network topology is obtained by performing time integration on all faulty nodes based on the continuously acquired distribution network risk set.

[0177] When the cumulative risk value of any faulty node is greater than or equal to the preset risk threshold, a fault warning message for the target distribution network is output.

[0178] In this embodiment, fault monitoring of the target distribution network continues, and the above steps are repeated to continuously calculate whether early fault signals continue to appear at each fault point. The cumulative sum of node coupling risk values ​​of each fault node is calculated to obtain the node cumulative risk value of each fault node at each monitoring time point. When the node cumulative risk value of any fault node is greater than or equal to the preset risk threshold, fault warning information is output to remind maintenance personnel to carry out maintenance or perform tripping operations. The fault warning information includes information such as the early fault type and early fault location, so that maintenance personnel can quickly reach the maintenance point. When performing tripping operations, the power outage range can also be planned according to the early fault location to minimize the power outage range and ensure people's normal power supply.

[0179] The above solution reduces unnecessary power outages. Many early fault signals may be transient disturbances (such as short-term icing vibrations or momentary poor equipment contact) or minor anomalies (such as weak signals from early aging), and are not actually risks that will develop into faults. Blindly tripping in these situations would directly cause power outages for users, affecting power continuity. For example, if a line experiences slight icing and vibrations due to a momentary wind speed, triggering a fault signal, direct tripping would cause several hours of power outage, even though the signal might not actually cause a fault. This solution can distinguish between "transient disturbances" and "continuously worsening risks," only issuing a warning when the risk accumulates and exceeds a threshold, avoiding unnecessary tripping due to misjudgment of a single signal and significantly improving power supply reliability. Furthermore, for some low-risk faults, such as equipment triggering an early fault signal due to a short-term increase in ambient humidity, immediate repair might reveal that the equipment is actually in good condition and only requires adjustment of the operating environment, without replacing components. In such cases, repair resources (manpower, equipment, and time) are wasted, and unplanned power outages may occur due to the repair work. In blizzard conditions, the cost of a single maintenance operation is high, making on-demand maintenance even more crucial. Furthermore, frequent passive tripping or emergency maintenance (such as live-line repairs or forced shutdowns) can cause additional wear and tear on equipment (e.g., frequent motor starts and stops, mechanical wear from repeated operation of switches), shortening equipment lifespan. Therefore, it is essential to accurately determine the timing of maintenance and adopt planned maintenance (such as during periods of low equipment load) to reduce forced operation and extend the equipment's lifespan.

[0180] This application also provides a distribution network fault early warning system based on online monitoring, including:

[0181] The memory is configured to store instructions; and

[0182] The processor is configured to retrieve instructions from memory and, when executing the instructions, to implement the online monitoring-based power distribution network fault early warning method described above.

[0183] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it.

[0184] The memory can be an internal storage unit of a computer device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) provided on the computer device. Furthermore, the memory can be a combination of internal storage units and external storage devices of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0185] This application also provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described method for early warning of distribution network faults based on online monitoring.

[0186] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0188] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0189] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0190] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0191] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0192] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0193] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0194] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for early warning of distribution network faults based on online monitoring, characterized in that, The method includes the following steps: Real-time electrical signals from all monitoring points in the target power distribution network are continuously collected by multiple types of sensors pre-installed in the target power distribution network. For any monitoring point, real-time signal prediction is performed based on the real-time electrical signal of the monitoring point and using the extended Kalman filter algorithm. The real-time signal prediction results and preset early fault characteristics are combined to determine whether there are early fault signals in the target distribution network. If there are early fault signals in the target distribution network, the distribution network information of the target distribution network, as well as the environmental prediction data and geographical location data of the target distribution network are obtained. The distribution network information includes line load data, and the environmental prediction data includes predicted wind direction. The distribution network topology is constructed based on the distribution network information, and the early fault information of the target distribution network is extracted by combining the distribution network topology and all early fault signals. The windward characteristics of the distribution lines in the target distribution network are determined based on the route of the distribution network lines and the predicted wind direction. The windward characteristics include lines against the wind, lines with the wind, and lines across the wind. The target distribution network is divided into several distribution network areas based on the line load data and windward characteristics of the distribution lines; A preliminary icing evolution model for each power distribution network area was constructed based on environmental prediction data. For any primary icing evolution model, geographical location data, windward characteristics, and line load data are used as icing correction factors. The primary icing evolution model is then corrected using the icing correction factors to obtain the regional icing evolution model. By combining the icing evolution models of all regions, the icing evolution model of the target distribution network is obtained; By completing the fault risk coupling between the icing evolution model and early fault information, a set of distribution network risks for the distribution network topology is obtained. Repeat the above steps to continuously acquire the distribution network risk set of the distribution network topology, perform risk accumulation analysis on the distribution network topology based on the continuously acquired distribution network risk set, and output fault warning information of the target distribution network based on the risk accumulation analysis results.

2. The method according to claim 1, characterized in that, The process of determining whether there are early fault signals in the target distribution network based on real-time electrical signals from monitoring points and using the extended Kalman filter algorithm for real-time signal prediction, and combining the real-time signal prediction results with preset early fault characteristics, includes the following steps: A second-order generalized integral phase-locked loop is used to synchronously sample the real-time electrical signals at the monitoring points, thereby obtaining the electrical sampling signals; Based on the electrical sampling signal and using the extended Kalman filter algorithm, the prior estimation of the target distribution network at the current moment is completed to obtain the signal prediction state, which includes the signal prediction instantaneous value, signal prediction frequency, signal prediction phase and signal prediction amplitude. Extract the signal sampling state of the electrical sampling signal and calculate the signal difference index between the signal sampling state and the signal prediction state; The fault similarity between the signal sampling state and the preset early fault characteristics is calculated using a similarity formula. By combining signal difference indicators and fault similarity, early fault identification of real-time electrical signals can be achieved; If the real-time electrical signal is a normal electrical signal, then continue to collect the real-time electrical signal of the monitoring point; If the real-time electrical signal is a suspected fault signal, then the secondary fault identification of the suspected fault signal is completed by injecting an auxiliary detection signal into the monitoring point, and the real-time electrical signal is determined to be an early fault signal based on the secondary fault identification result. If the real-time electrical signal is an early fault signal, it indicates that there is an early fault signal in the target distribution network.

3. The method according to claim 2, characterized in that, The process of injecting auxiliary detection signals into monitoring points to perform secondary fault identification of suspected fault signals, and determining whether the real-time electrical signal is an early fault signal based on the secondary fault identification results, includes the following steps: By injecting auxiliary detection signals into the monitoring points to enhance suspected fault signals, an electrical enhancement signal is obtained; Modal decomposition of the electrical enhancement signal yields multiple intrinsic modes of the signal; The signal mode energy of all inherent modes of the signal is calculated using the energy operator; Weights are assigned to all signal intrinsic modes based on the signal modal energy, and the signals of all signal intrinsic modes after weight assignment are reconstructed to obtain the electrical reconstructed signal; Calculate the signal difference between the electrical reconfiguration signal and the pre-acquired historical reference signal; If the signal difference is less than or equal to the preset difference threshold, the real-time electrical signal is determined to be a normal electrical signal. If the signal difference is greater than the difference threshold, the real-time electrical signal is determined to be an early fault signal.

4. The method according to claim 1, characterized in that, The power distribution network information also includes power distribution network structure data, and the environmental prediction data also includes predicted temperature, predicted humidity, predicted wind speed, predicted precipitation, and predicted freezing rain time.

5. The method according to claim 4, characterized in that, The process of constructing a distribution network topology based on distribution network information and extracting early fault information of the target distribution network by combining the distribution network topology and all early fault signals includes the following steps: The early fault locations of all early fault signals are located using a signal localization algorithm. Based on the distribution network structure data, the target distribution network is abstracted into a distribution network topology. The nodes of the distribution network topology include power supply nodes, branch nodes, monitoring nodes, equipment nodes, and fault nodes, and the edges of the distribution network topology are the distribution lines of the target distribution network. For any early fault signal, the topological fault features of the early fault signal are extracted based on the distribution network topology. The topological fault features include the fault branch coefficient, the fault reflection coefficient, and the fault load coefficient. Basic fault features of early fault signals were extracted based on wavelet transform algorithm; A fault feature tensor is constructed by stacking topological fault features and basic fault features; The fault feature tensor is input into the pre-built signal fault identification model, and the early fault type of the target distribution network is output through the fault signal identification model. The signal fault identification model is built based on the residual network. The early fault location and early fault type of the early fault signal are integrated into the early fault information of the target distribution network.

6. The method according to claim 5, characterized in that, The basic fault features extracted from early fault signals based on the wavelet transform algorithm include the following steps: Preprocess early fault signals; The preprocessed early fault signal is decomposed by stationary wavelet transform to obtain several early fault sub-bands. Each early fault sub-band is divided into several early fault segments at equal intervals. The basic fault features of all early fault segments are extracted. These basic fault features include fault mean, fault variance, fault energy value, fault energy entropy, fault frequency entropy value, and fault entropy weight.

7. The method according to claim 4, characterized in that, The process of coupling fault risk between the icing evolution model and early fault information to obtain the distribution network risk set of the distribution network topology includes the following steps: For any fault node in the distribution network topology, the fault evolution level and fault risk level of the fault node are determined according to the early fault type in the early fault information corresponding to the fault node, and an early fault risk value is assigned to the fault node in combination with the fault evolution level and fault risk level. The icing thickness at the location corresponding to the fault node was calculated based on the icing evolution model. The icing type of the target distribution network is determined by combining predicted temperature, predicted humidity and predicted wind speed. The icing type includes stable icing and unstable icing. Assign line icing risk values ​​to fault nodes based on line icing thickness and icing type; The topological centrality of faulty nodes is determined based on the distribution network topology, and a faulty topological risk value is assigned to the faulty nodes based on the topological centrality. The node coupling risk value of the fault node is obtained by weighted fusion of early fault risk value, line icing risk value and fault topology risk value; By integrating the node coupling risk values ​​of all faulty nodes, a distribution network risk set for the distribution network topology is obtained.

8. The method according to claim 1, characterized in that, The step of performing risk accumulation analysis on the distribution network topology based on the continuously acquired distribution network risk set, and outputting fault early warning information of the target distribution network based on the risk accumulation analysis results, includes the following steps: The cumulative risk value of all faulty nodes in the distribution network topology is obtained by performing time integration on all faulty nodes based on the continuously acquired distribution network risk set. When the cumulative risk value of any faulty node is greater than or equal to the preset risk threshold, a fault warning message for the target distribution network is output.

9. A distribution network fault early warning system based on online monitoring, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve instructions from memory and, when executing the instructions, to implement the online monitoring-based power distribution network fault early warning method according to any one of claims 1 to 8.

Images