A power distribution network power supply guaranteeing method, system, device and medium

Abstract

This invention discloses a method, system, equipment, and medium for power supply protection in distribution networks, belonging to the field of distribution network automation technology. The method includes: acquiring multi-source parameters to establish a normalized state matrix; outputting anomaly judgment results based on sensitivity weighting; solving for the minimum comprehensive cost path by combining constraints and issuing control commands to restore power supply. The system includes a data acquisition module, a time alignment module, an intelligent judgment module, a cost evaluation module, an optimization and reconstruction module, and an execution recovery module. This invention breaks the strong binding between software and hardware, supports plug-and-play sensing devices and time-aligned fusion of multi-source heterogeneous data; it decentralizes decision-making power to the edge, relying on sensitivity weighted judgment and comprehensive cost optimization to complete local fault autonomous isolation and power supply restoration in millisecond-level closed loop, eliminating dependence on long-link central scheduling; and it uses execution feedback data to adaptively adjust weights to achieve iterative evolution of system strategies.

Description

Technical Field

[0001] This invention relates to the field of power distribution network automation technology, specifically to a power supply guarantee method, system, equipment, and medium for power distribution networks. Background Technology

[0002] The power distribution network system plays a crucial role in electrification construction, and the requirements for power supply reliability, fault response speed, and intelligence level are increasing. In order to enhance the operational resilience of the power distribution network, related technologies widely adopt feeder automation technology, distributed power source access control systems, and intelligent terminal deployment methods such as feeder terminal units and distribution transformer terminal units to ensure power supply.

[0003] However, existing power supply protection systems are structurally simple and closed, heavily reliant on remote centralized control and pre-set logic. When a fault occurs in the distribution network, the underlying data must traverse long communication links to be transmitted back to the central dispatch center, where it is processed uniformly before action commands are relayed downwards. This long-link centralized decision-making model clashes severely with the millisecond-level fault self-rescue requirements of the distribution network. Once the upper-level communication link becomes congested or disconnected, the field equipment completely loses its ability to respond. Furthermore, traditional systems are heavily tied to hardware and software, lacking modular expansion and flexible deployment capabilities, and cannot achieve plug-and-play functionality for new sensing devices. At the sensing level, current state perception granularity is too coarse. Most systems rely solely on traditional single current and voltage sensors, lacking a deep fusion mechanism to simultaneously capture and analyze multi-source heterogeneous data such as field environmental information, electrical transient characteristics, and equipment health status. This makes the system highly susceptible to missed detections when facing complex, latent faults such as high-resistance grounding. Furthermore, the information of each node is isolated from each other, lacking rapid closed-loop linkage from front-end perception, edge decision-making to terminal execution. When faced with power supply topology reconstruction or dynamic load changes, the terminal equipment cannot autonomously find the optimal power transfer path based on the actual constraints on site, which seriously restricts the local autonomy and rapid power restoration capability of the distribution network. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a power supply guarantee method and system for distribution networks to solve the problem that the existing centralized systems have long communication links and closed architectures, which makes it impossible to achieve rapid autonomous response and power supply restoration at the edge side when a fault occurs.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for ensuring power supply in a distribution network, comprising: Obtain multi-dimensional operating state parameters of distribution network nodes and construct a multi-dimensional state parameter dataset; Cross-scale features are extracted from the multi-dimensional state parameter dataset and a time-aligned mapping relationship is established to obtain a time-aligned normalized state matrix. In response to the completion of the normalized state matrix construction, weight mapping coefficients are configured for the normalized state matrix based on the sensitivity differences of the multidimensional operating state parameters, and a fault determination result based on weighted features is output for the first anomaly determination of the distribution network section. In response to the first anomaly determination being triggered, the topology and load constraint characteristics of the backup power supply path are analyzed, and a cost evaluation function based on the path electrical parameters is generated for the second optimal prediction of the backup power supply path. Combining the second preferred prediction, the coupling characteristics between the distance of the backup power supply path and the scale of switching actions are analyzed, and a path optimization method based on the minimum comprehensive cost is constructed. The path optimization method outputs action commands, and the weight mapping coefficients are adaptively adjusted based on the execution feedback data to ensure power supply to the distribution network nodes.

[0007] As a preferred embodiment of the power supply guarantee method for a distribution network according to the present invention, the step of obtaining multi-dimensional operating state parameters of distribution network nodes and constructing a multi-dimensional state parameter dataset includes: In response to the physical access of the sensing terminal to the power distribution network, the global device identifier code of the sensing terminal is parsed through the network layer protocol; Based on the global device identifier code, match the corresponding scene configuration template to establish an edge collaborative communication connection; Read the standardized object model of the sensing terminal and map it to establish the underlying acquisition channel of the multi-dimensional operating state parameters; The underlying acquisition channel is activated to continuously receive heterogeneous measurement point data, which is then aggregated to generate the multidimensional state parameter dataset.

[0008] As a preferred embodiment of the power supply guarantee method for a distribution network according to the present invention, the step of extracting cross-scale features from the multi-dimensional state parameter dataset and establishing a time-aligned mapping relationship to obtain a time-aligned normalized state matrix includes: In response to the completion of the aggregation of the multidimensional state parameter dataset, a multi-scale analysis window is divided based on a unified timestamp; Within the multi-scale analysis window, the transient effective value features of the high-frequency signal are extracted to complete the downsampling process; Numerical mapping rules are used to complete the steady-state interpolation characteristics of low-frequency signals; The time-aligned mapping relationship is established by integrating the transient effective value features and the steady-state interpolation features, and the normalized state matrix with dimensional differences is output.

[0009] The beneficial effects of this preferred technical solution are: it avoids timing misalignment caused by inconsistent sampling frequencies of multi-source heterogeneous parameters. By extracting high-frequency transient effective values ​​and interpolating to complete low-frequency steady-state characteristics, a cross-scale time alignment mapping relationship is established. This eliminates dimensional barriers between physical quantities, ensures accurate fusion of multi-dimensional data on the same time profile, and provides high-temporal-consistency underlying data support for fault diagnosis.

[0010] As a preferred embodiment of the power supply guarantee method for a distribution network according to the present invention, the step of configuring weight mapping coefficients for the normalized state matrix based on the sensitivity differences of the multi-dimensional operating state parameters and outputting a fault judgment result based on weighted features includes: In response to the normalized state matrix outputting feature components, the feature components are input into a preset decision tree cluster; The information gain index of the feature components when dividing the running status labels is statistically analyzed to determine the sensitivity difference; The weight mapping coefficients are assigned to the feature components based on the sensitivity differences. The anomaly detection threshold is set using the statistical distribution boundary of the baseline operating data, and the fault detection result is generated by combining the weight mapping coefficient.

[0011] As a preferred embodiment of the power supply guarantee method for a distribution network according to the present invention, the step of analyzing the topology and load constraint characteristics of the backup power supply path and generating a cost evaluation function based on the path electrical parameters includes: In response to the first anomaly determination being triggered, candidate paths are searched by traversing the distribution network topology. Verify the node connectivity status of the candidate paths and eliminate paths with isolation and locking attributes; Compare the rated current carrying capacity of the candidate paths with the power of the load to be supplied, and eliminate paths whose capacity exceeds the limit; Calculate the voltage drop magnitude at the end of the candidate path, eliminate paths with voltage drops exceeding the limit, and output the cost evaluation function that conforms to the topology and load constraint characteristics.

[0012] As a preferred embodiment of the power supply guarantee method for a distribution network according to the present invention, the step of analyzing the coupling characteristics between the distance of the backup power supply path and the scale of switch actions, and constructing a path optimization method based on minimizing the overall cost, includes: In response to the available path set output by the cost evaluation function, the electrical distance parameters, voltage deviation parameters, and number of switching actions parameters of each path in the available path set are extracted; Based on the safety margin requirements of the real-time operating conditions of the distribution network, the optimal proportions of the electrical distance parameter, the voltage deviation parameter, and the number of switch actions parameter are dynamically adjusted. The comprehensive cost of each path is calculated based on the aforementioned preferred ratio and overlap. The path with the lowest overall cost is selected, and the path optimization method is established.

[0013] As a preferred embodiment of the power supply guarantee method for a distribution network according to the present invention, the step of adaptively correcting the weight mapping coefficient by combining execution feedback data includes: In response to the completion of the action command, steady-state electrical parameters after the switching of distribution network nodes are collected; The steady-state electrical parameters are compared with the standard power supply threshold to verify the power supply recovery status. Based on the power supply recovery status, a structured sample set is formed by encapsulating the fault event record, the action command, and the steady-state electrical parameters; The weight mapping coefficients are corrected using the structured sample set.

[0014] In a second aspect, the present invention provides a power distribution network power supply protection system, comprising: The time alignment module is used to extract cross-scale features from the multi-dimensional state parameter dataset and establish a time alignment mapping relationship to obtain a time-aligned normalized state matrix. The intelligent judgment module is used to respond to the completion of the normalized state matrix construction, configure weight mapping coefficients for the normalized state matrix according to the sensitivity differences of the multi-dimensional operating state parameters, and output the fault judgment result based on weighted features for the first anomaly judgment of the distribution network section. The cost evaluation module is used to analyze the topology and load constraint characteristics of the backup power supply path in response to the first anomaly determination being triggered, and generate a cost evaluation function based on the path electrical parameters for the second optimal prediction of the backup power supply path. The optimization and reconstruction module is used to combine the second optimal prediction to analyze the coupling characteristics between the distance of the backup power supply path and the scale of switching action, and to construct a path optimization method based on the minimum comprehensive cost. The execution recovery module is used to output action commands through the path optimization method and adaptively correct the weight mapping coefficients based on the execution feedback data to ensure power supply to the distribution network nodes.

[0015] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of a power distribution network power supply guarantee method.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the power supply protection method for a power distribution network.

[0017] The beneficial effects of this invention are as follows: This solution breaks the strong binding between software and hardware, supports plug-and-play sensing devices and time-aligned fusion of multi-source heterogeneous data; it decentralizes decision-making power to the edge, relies on sensitivity-weighted judgment and comprehensive cost optimization to complete local fault autonomous isolation and power restoration in millisecond-level closed loop, and gets rid of the dependence on long links of central scheduling; and it uses execution feedback data to adaptively correct weights to realize the iterative evolution of system strategies. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a power distribution network power supply guarantee method provided in one embodiment of the present invention.

[0020] Figure 2 This is a general framework diagram of a power distribution network power supply guarantee system provided in one embodiment of the present invention.

[0021] Figure 3 This is a flowchart illustrating the multi-source data fusion and time alignment process of a power supply guarantee method for a power distribution network, as provided in one embodiment of the present invention.

[0022] Figure 4 This is a structural diagram of a fault determination model for a power distribution network power supply system, provided as an embodiment of the present invention.

[0023] Figure 5 This is a schematic diagram of the topology and power supply path of a power distribution network power supply system provided in one embodiment of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0025] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for ensuring power supply in a distribution network is provided, comprising: S100: Obtain multi-dimensional operating status parameters of distribution network nodes and construct a multi-dimensional status parameter dataset; S200: Extract cross-scale features from the multi-dimensional state parameter dataset and establish a time-aligned mapping relationship to obtain a time-aligned normalized state matrix; S300: In response to the completion of the normalized state matrix construction, the weight mapping coefficients are configured for the normalized state matrix according to the sensitivity differences of the multidimensional operating state parameters, and the fault judgment result based on the weighted features is output for the first anomaly judgment of the distribution network section. S400: In response to the first anomaly determination being triggered, analyze the topology and load constraint characteristics of the backup power supply path, generate a cost evaluation function based on the path electrical parameters, and use it for the second optimal prediction of the backup power supply path; S500: Combining the second preferred prediction, analyze the coupling characteristics between the distance of the backup power supply path and the scale of switching actions, and construct a path optimization method based on the minimum comprehensive cost; S600: The path optimization method outputs action commands, and the weight mapping coefficients are adaptively corrected in combination with the execution feedback data to ensure the power supply of the distribution network nodes.

[0026] It should be noted that current distribution network fault handling heavily relies on remote centralized dispatch centers. When an anomaly occurs on-site, data must traverse long communication links to be transmitted back to headquarters for calculation before instructions are issued, resulting in a significant delay in response. Furthermore, once upper-level communication is lost, on-site equipment loses its ability to handle the situation. Simultaneously, traditional sensing methods often rely solely on triggering events by exceeding limits for single electrical parameters, lacking multi-source heterogeneous data fusion and cross-scale time alignment, leading to a high rate of missed detections for latent faults such as high-resistance grounding. In addition, conventional systems lack closed-loop verification mechanisms after on-site actions, and their isolation and power transfer strategies are rigid, unable to adaptively adjust according to real-time topology and load constraints, easily triggering cascading trips or large-scale power outages.

[0027] Therefore, to address the aforementioned main issues, the S100-S600 architecture is designed to construct multi-dimensional perception and second-level autonomous closed-loop control steps at the edge. By delegating decision-making authority to the field edge control unit, a time-aligned normalized state matrix combined with sensitivity weighting is used to achieve high-precision local fault location. Furthermore, a multi-dimensional cost evaluation function is used to quickly select the power transfer path with the lowest overall cost. Finally, an execution feedback linkage adaptive correction judgment weight is introduced to completely break the constraints of long-link communication, achieve millisecond-level fault self-rescue, and provide highly reliable power supply for critical and sensitive loads.

[0028] Example 2, refer to Figures 1 to 2 This is one embodiment of the present invention. Based on the previous embodiment, a method for ensuring power supply in a distribution network is provided.

[0029] Before outlining the plan, refer to Figure 2 First, let's introduce the edge computing and multi-source data fusion architecture of the power distribution network. For example... Figure 2 As shown, the overall system architecture, from top to bottom, comprises a central coordination layer, a perception layer, an edge control layer, and an execution terminal layer. The perception layer includes electrical parameter sensing modules and environmental / equipment sensing modules, responsible for collecting underlying source data and outputting it to the edge control layer. The central coordination layer includes modules for model training and parameter distribution, global topology management, and global optimization decision-making, responsible for distributing top-level model parameters to the edge control layer and managing global information. The edge control layer, as the core decision-making hub, includes modules for data fusion and time alignment, fault identification, and local policy generation, responsible for receiving data from the perception layer and parameters from the central coordination layer, and outputting control policies locally. The execution terminal layer includes smart switches, energy storage / distributed power supplies, and adjustable loads, responsible for receiving instructions from the edge control layer and executing actions. Its technical system can be divided into three main categories based on the decision-making location: centralized control, distributed collaboration, and edge autonomy.

[0030] Centralized control, based on global data training and scheduling, aims to coordinate the mapping relationships of resources across the entire network through a remote master station. Typical applications include traditional distribution automation systems (DMS), which are widely used in scenarios such as macro-grid planning and global load balancing. Distributed collaboration relies on peer-to-peer communication between adjacent nodes to exchange state patterns, such as intelligent distributed feeder automation based on the GOOSE protocol, which is suitable for rapid fault isolation in simple chain topologies. Edge autonomy, on the other hand, pushes sensing, computing, and decision-making capabilities down to the core nodes of the distribution network. By optimizing local closed-loop strategies through built-in fusion models, it has unique advantages in dynamic decision-making scenarios such as communication constraints, master station disconnection, or millisecond-level sudden fault self-rescue.

[0031] Key technologies supporting autonomous edge applications include heterogeneous data fusion, topology reconstruction optimization, and adaptive closed-loop mechanisms. Data fusion is a core component of front-end perception, encompassing data cleaning operations such as unified timestamp alignment and dimensional elimination, as well as key steps like multi-scale feature extraction (e.g., high-frequency transient RMS values ​​and low-frequency steady-state interpolation). In the power distribution field, features need to be designed by incorporating knowledge of physical equipment operating rules, such as line current distortion rate and transformer temperature rise gradient. Model optimization techniques cover multi-objective cost function construction, topology constraint verification, and weight mapping allocation. For example, edge control units can construct a comprehensive cost value by superimposing electrical distance and the number of switching actions, reducing voltage fluctuation variance during power transfer and improving the system's robustness to dynamic topology noise in the distribution network.

[0032] In terms of control framework, the traditional open-loop framework is suitable for unidirectional command issuance, while the closed-loop collaborative framework based on execution feedback can use steady-state verification data after terminal action to correct judgment weights and efficiently process massive multi-dimensional feature streams at the edge of the distribution network, such as the hundreds of GB of high-frequency waveforms and real-time operation data of environmental sensors generated daily by a large industrial park distribution network.

[0033] In this embodiment of the application, step S100 obtains the multi-dimensional operating state parameters of the distribution network nodes and constructs a multi-dimensional state parameter dataset; Specifically, the multi-dimensional operating status parameter types can cover high-frequency electrical parameters such as current and voltage in the physical field of the power distribution network, as well as low-frequency non-electrical parameters such as temperature, humidity, partial discharge vibration, and gas concentration; generate heterogeneous data streams that conform to the multi-scale characteristics of the power distribution field, and provide multi-dimensional source inputs for subsequent processing.

[0034] In one optional implementation, S100 can dynamically acquire multi-dimensional operating status parameters based on a device self-discovery protocol. Starting from the physical connection of the sensing terminal to the power distribution network, it parses the global device identifier code through a network layer protocol, matches the corresponding scenario configuration template based on the identifier code, and establishes an edge collaborative communication connection. Subsequently, it reads the standardized object model of the sensing terminal, maps and establishes the underlying acquisition channel, and continuously receives heterogeneous measurement point data. This approach retains the plug-and-play characteristics of the device, facilitates multi-source fusion in subsequent steps, breaks down system closure, and enhances the scalability of the sensing layer in actual field operations.

[0035] In another optional implementation, S100 can also directly collect various types of data sequences already existing in the substation or switching station via the aggregation module, including the operating status of intelligent protection devices read via the IEC61850MMS protocol, and historical records of existing temperature and humidity sensors accessed via the MQTT protocol. This method can also serve as an input source for generating multidimensional status parameter datasets, covering static ledger features and multi-source monitoring features.

[0036] S200: Extract cross-scale features from the multi-dimensional state parameter dataset and establish a time-aligned mapping relationship to obtain a time-aligned normalized state matrix; In this embodiment of the application, the cross-scale feature extraction in step S200 includes: multi-scale analysis window division, high-frequency transient effective value feature extraction, and low-frequency steady-state interpolation feature completion; such as Figure 3 As shown, multi-source input data such as current, voltage, temperature, and vibration are distributed according to processing logic. For high-frequency signals, their RMS value, peak value, or frequency domain characteristics are extracted; for low-frequency signals, methods such as hold, interpolation, or moving average are used for processing. Subsequently, these two signals are aggregated into a unified time window alignment stage, and then handed over to a normalization processing stage to eliminate dimensional differences, finally constructing and outputting a normalized sensing matrix.

[0037] Specifically, multi-scale analysis window partitioning, or time-dimension slicing and rasterization, is a process of distributing converged asynchronous multi-source data streams to a unified time scale.

[0038] For example, in response to the completion of the aggregation of the multidimensional state parameter dataset, the recorded current with a sampling rate of 10kHz and the ambient temperature with a sampling rate of 1Hz are allocated to a fundamental frequency period analysis window of 20ms, based on the master control clock at the edge of the distribution network.

[0039] It should be noted that its core is to map physical quantities with different sampling frequencies onto the same time analysis profile, so that the intelligent fault judgment model can capture the multi-dimensional feature distortion law of the same fault moment.

[0040] Specifically, high-frequency transient effective value feature extraction refers to the process of extracting key physical state parameters and reducing data dimensionality by means of root mean square operation for high-frequency sampling sequences such as current and voltage. It should be noted that its core is to filter out redundant high-frequency sampling points and retain only the key feature values ​​that reflect the transient characteristics of distribution network faults.

[0041] Preferably, this embodiment extracts transient features using a half-wave RMS algorithm. Spatially, the data stream is divided into primary equipment main circuit electrical quantities and secondary equipment operating condition quantities according to the equipment monitoring level. The main circuit electrical quantities include line phase current, zero-sequence current, and bus voltage, which meet the millisecond-level protection action requirements and have a direct impact on fault isolation; the operating condition quantities include the displacement of the switch mechanism, etc. Temporally, the data is divided into high-frequency and low-frequency categories according to the sampling rate level. Feature compression of high-frequency data is achieved by extracting transient RMS values.

[0042] For example, a sliding time window is used to extract the root mean square value of the current sampling point set within 20 milliseconds, output a single effective value parameter, and save this mapping relationship for subsequent superposition calculations.

[0043] Specifically, low-frequency steady-state interpolation feature completion refers to the process of generating intermediate time estimates for low-frequency, slowly changing feature quantities such as ambient temperature, humidity, or equipment partial discharge using numerical approximation methods.

[0044] For example, steady-state feature completion: Low-frequency sampling data such as temperature are filled into a 20-millisecond analysis window using linear interpolation or spline interpolation to ensure that each high-frequency data analysis window has a corresponding temperature feature.

[0045] It should be noted that its core objective is to fill the data gaps of low-frequency sensors during rapid transient processes, enabling the edge control unit to utilize continuous steady-state environmental characteristics to assist in the determination of complex faults such as high-resistance grounding.

[0046] In an optional implementation, based on the above feature extraction implementation method, it may also include zero drift suppression and dead zone filtering. Based on the physical characteristics of the power distribution network, a dead zone threshold rule is designed to eliminate invalid fluctuations such as sensor zero drift or extremely weak environmental white noise, so as to improve the data quality of the normalized state matrix, reduce the invalid occupation of noise on the computing resources of the edge computing module, and at the same time retain the real over-limit signals, which facilitates the accurate mapping of weights in S300.

[0047] In another optional implementation, based on the above feature extraction implementation, it may also include phase sequence decoupling and sequence component stripping. The three-phase asymmetrical current and voltage signals are decomposed into positive sequence, negative sequence and zero sequence components by the symmetrical component method, eliminating the three-phase coupling correlation differences, which facilitates the subsequent grounding fault assessment.

[0048] In this embodiment of the application, the step S200 of establishing the time alignment mapping relationship and obtaining the normalized state matrix includes: A1: The timestamp anchoring method is used to concatenate and combine the extracted transient effective value features and steady-state interpolation features in time sequence; Specifically, timestamp anchoring is a method used in heterogeneous data fusion to first generate a reference time axis using a high-precision edge clock source as the reference, and then strictly align the feature values ​​of each physical quantity to the same time slice according to the timestamp label.

[0049] For example, the matching can be shifted in steps of every millisecond to ensure that the voltage sag characteristics and the partial discharge peak characteristics correspond strictly in the time profile, thus avoiding asynchronous misalignment.

[0050] It should be noted that its core objective is to ensure absolute synchronization of the multidimensional data input to the model at the moment of physical occurrence, avoiding false correlation judgments caused by network transmission delays or different sensor sampling rates. In multi-source fusion analysis of distribution networks, timestamp anchoring can effectively solve the problem of fault judgment failure caused by heterogeneous network delays by forcibly synchronizing the data frames of each sensor channel, and is a core means to improve the accuracy of edge computing.

[0051] A2: Apply numerical mapping transformation to the time-aligned feature set. The numerical mapping transformation includes eliminating differences in different physical dimensions and mapping to a fixed numerical range to adapt to model fusion requirements.

[0052] Specifically, when extracting cross-scale features and obtaining the normalized state matrix, the specific numerical mapping rules are as follows: The normalized calculation method for current is as follows: in Mapped to interval, To collect the effective value of the current, This is the rated current. The voltage normalization calculation method is as follows: ,in Mapped to interval, To collect the effective value of the voltage, This is the rated voltage of the power distribution network.

[0053] The temperature normalization calculation method is as follows: The reference value is the room temperature. This represents the temperature anomaly threshold range.

[0054] The normalization calculation method for vibration frequency is as follows: The power frequency reference vibration frequency, This represents the abnormal threshold frequency.

[0055] It should be noted that feature dimension transformation refers to the process of converting numerical values ​​with different physical units, such as kilovolt-level voltage, ampere-level current, and Celsius-level temperature, into a dimensionless standard range based on benchmark physical constraints. Its core lies in breaking the magnitude limitations of the original physical quantities and, through methods such as extreme value normalization and standard deviation scaling, enabling the new feature matrix to more evenly reflect the severity of anomalies in each parameter. In the field of power distribution networks, the original feature magnitudes differ greatly; direct calculation would cause the sensitivity contribution of smaller numerical features to be completely overwhelmed by larger numerical features. Dimension transformation, by smoothing out unit differences, transforms physical quantities into feature units with equivalent mathematical benchmarks, and is a crucial step in the transformation of heterogeneous data from the physical acquisition space to the algorithmic judgment space.

[0056] S300: In response to the completion of the normalized state matrix construction, it configures weight mapping coefficients for the normalized state matrix based on the sensitivity differences of multi-dimensional operating state parameters, and outputs fault judgment results based on weighted features for the first anomaly judgment of the distribution network section. In this embodiment of the application, step S300 configures weights based on the sensitivity differences of multidimensional operating state parameters and outputs the judgment result, including steps B1 to B3: like Figure 4 As shown, the normalized values ​​of the input parameters I(t) current, V(t) voltage, T(t) temperature, and F(t) vibration are calculated respectively. Then, each normalized value is input to a weighted node, and the values ​​are weighted according to the configured weights wi to calculate and output the fusion state index Φ(t). The system then determines whether the fusion state index Φ(t) is greater than a preset threshold δ. If it is, the branch outputs a result indicating an anomaly; otherwise, the branch outputs a result indicating normality.

[0057] B1: In response to the output feature components of the normalized state matrix, the information gain index of the feature components when dividing the running state labels is statistically determined to determine the sensitivity difference; Specifically, in the sensitivity assessment phase, based on the multi-source heterogeneous characteristics of distribution network fault features, the feature components are input into a pre-defined decision tree cluster. The decision tree cluster quantifies the contribution of each feature to fault classification by calculating the change in information entropy before and after node splitting. The sensitivity assessment selects an information gain algorithm, which is suitable for the rapid location of key features in multi-dimensional mixed data, and is particularly suitable for the identification task of complex latent faults in this multi-source sensing scenario of the distribution network. The formula for calculating its information gain is as follows: Where Gain represents the information gain obtained by partitioning dataset D with feature a, Ent represents the information entropy of the dataset, V represents the number of possible values ​​of feature a, and the absolute value symbol represents the number of samples in the corresponding set.

[0058] B2: Assign weight mapping coefficients to feature components based on sensitivity differences; In one alternative implementation, B2 can be implemented using a normalized exponential function method to map the information gain values ​​of each feature to weight constants between zero and one, with a total sum of one. This emphasizes amplifying the proportion of high-gain features such as zero-sequence voltage abrupt changes in the calculation, while suppressing the interference of low-gain features such as fluctuations in ambient humidity. This ensures the system can accurately focus on the core fault characterization signals and output the corresponding weight information for each feature for subsequent superposition calculations.

[0059] B3: The statistical distribution boundary of the benchmark operating data is used to set the anomaly judgment threshold, and the fault judgment result is generated by combining the weight mapping coefficient.

[0060] Specifically, the mean and standard deviation are calculated based on the characteristic distribution of the historical normal operation cycle of the distribution network, and the mean plus three times the standard deviation is set as the anomaly judgment threshold.

[0061] In another alternative implementation, the normalized state matrix with configured weight mapping coefficients is linearly weighted and summed to obtain a comprehensive fusion state index.

[0062] Specifically, calculate the fusion state index. Mathematical model: In the formula, To perceive the total number of data types, These are the normalized values ​​for each parameter. These are the weighting coefficients, and all weighting coefficients satisfy the constraint that their sum is 1. After determining the weights using the information gain index mentioned in the final draft, the fusion state index is calculated using this formula, and then compared with historical data from normal operation. The baseline threshold set by the principle In comparison, when An exception is triggered at that time.

[0063] Subsequently, the fusion status index is compared with the anomaly judgment threshold in real time on the edge monitoring subset of the distribution network segment. If the fusion status index exceeds the preset judgment threshold, it indicates a suspected fault in the current segment, triggering the first anomaly judgment and saving the over-limit record; otherwise, it is judged as a normal state and returns to B1 to continue cyclic monitoring. This step mainly ensures the reliability of multi-source fusion analysis in edge micro-scenes and supports seamless linkage with subsequent topology path optimization.

[0064] It should be noted that feature sensitivity assessment and dynamic weight mapping are key steps in improving fault identification accuracy during the model construction process. By calculating information gain and setting exponential mapping, the optimal combination of attention points can be found in the preset feature space, which can effectively solve the problem that weak fault features such as high-resistance grounding in the distribution network are easily overwhelmed by massive normal operation fluctuations.

[0065] It should also be noted that decision tree clusters are a collection of supervised learning algorithms based on a hierarchical tree structure. Their core objective is to accurately classify data from different operating states through a series of decision nodes composed of feature attributes. In a multidimensional feature space, if multiple fault modes exist, the decision tree attempts to find the feature dimension with the greatest improvement in information purity as the splitting benchmark; this benchmark is the internal node of the tree. For complex nonlinear fault feature associations, decision tree clusters introduce an ensemble learning mechanism to comprehensively vote on the results of multiple independently trained subtrees, making the model more robust when facing local feature loss or sudden electrical noise interference. In solving for the optimal splitting path, the algorithm transforms the problem into a recursive process of finding locally optimal features, using information theory criteria to solve for the maximum classification gain of each split. Decision tree clusters perform exceptionally well in classifying high-dimensional distribution network physical data that mixes discrete and continuous data. Even with limited computing power at edge terminals, this model can be used for lightweight and rapid inference on multi-source heterogeneous data, accurately identifying the normal and abnormal states of equipment, effectively avoiding false alarms from single-point sensors, and comprehensively improving the accuracy and robustness of local autonomous judgments.

[0066] S400: In response to the first anomaly determination being triggered, analyze the topology and load constraint characteristics of the backup power supply path, generate a cost evaluation function based on the path electrical parameters, and use it for the second optimal prediction of the backup power supply path; In this embodiment of the application, step S400 analyzes the topology and load constraint characteristics of the backup power supply path and generates a cost evaluation function, including steps D1 to D4: like Figure 5 As shown, the distribution network topology mainly includes line A, line B, and line C. Line A is equipped with intelligent circuit breakers ICB-A1 and ICB-A2, and is connected to load A; line B is equipped with ICB-B1 and ICB-B2, and is connected to load B; line C is equipped with ICB-C1, and is connected to load C. The lines are interconnected via tie switches. ICB-A2 and ICB-B2 are connected via tie switch LBS-AB, and ICB-B2 and ICB-C1 are connected via tie switch LBS-BC. When a fault occurs at the downstream end of ICB-A2 on line A, the system, based on the above topology calculations, plans the power reconfiguration path P1 as shown in the diagram. That is, through tie switch LBS-AB, backup power is supplied from ICB-B2 on the line B side to load A on line A.

[0067] D1: In response to the first anomaly detection being triggered, traverse and search for candidate paths in the distribution network topology; Specifically, the traversal of the distribution network topology, or the path search process in graph theory, is a process that starts from the current power-out node and searches for all interconnecting switches and backup power sources that can provide power through the distribution network node connection matrix.

[0068] For example, a depth-first search algorithm or a breadth-first search algorithm can be used to scan adjacent closed and open nodes along the feeder path to find three potential power transfer corridors.

[0069] It should be noted that its core is to quickly define the boundaries of physically reachable connected graphs and eliminate absolutely unusable paths caused by maintenance or physical disconnection.

[0070] D2: Verify the node connectivity of candidate paths and eliminate paths with isolation and locking attributes; Specifically, the isolation and interlocking attribute refers to the electrical characteristic of certain switch nodes that are forcibly prohibited from being closed due to safety anti-misoperation logic or being in a maintenance grounding state.

[0071] For example, when reading the switch failure blocking signal reported by the terminal, the candidate path containing the faulty switch is directly removed from the available set.

[0072] D3: Compare the rated current carrying capacity of the candidate paths with the power of the load to be supplied, and eliminate paths with excessive capacity. In one optional implementation, D3 involves obtaining the sum of real-time active and reactive power demands for the area to be restored, and extracting the rated current-carrying capacity of the thinnest cable or the smallest transformer in the candidate paths as the bottleneck capacity of that path. If the total power to be restored exceeds the bottleneck capacity multiplied by a safety margin factor, the capacity is deemed to be out of limit and the path is eliminated. The power supply capacity constraint must meet the current of the load to be supplied. Less than or equal to the maximum allowed capacity of the candidate path Right now .

[0073] This step is mainly to ensure that the transfer of power will not cause overload tripping in non-faulty areas, thus preventing the scope of the accident from expanding.

[0074] D4: Calculate the voltage drop magnitude at the end of the candidate path, eliminate paths with voltage drops exceeding the limit, and output a cost evaluation function that meets the topology and load constraint characteristics.

[0075] In another alternative implementation, power flow calculations are used to assess voltage losses during power transmission along long-distance feeders. The calculation can be expressed as a node voltage deviation equation: in, This represents the voltage drop magnitude. Represents the active power transmitted. Represents reactive power. Represents the equivalent resistance of the circuit. Represents the equivalent reactance of the line. This represents the rated voltage. If the calculated voltage drop at the end exceeds the allowable deviation range specified in the standard, the path is discarded. The final output is the set of valid paths that pass through all hard physical constraints, with each path assigned an initial cost value characterizing its electrical distance and health status, forming a cost evaluation function.

[0076] In addition, voltage stability constraints must satisfy the path-end voltage. Within the allowable range of voltage deviation, i.e. .

[0077] It should be noted that during the distribution network reconfiguration process, multi-dimensional constraint verification is the physical red line for ensuring the safe operation of the power grid. By searching the topology and then verifying capacity and voltage, legitimate and compliant power supply corridors can be found in the dynamically changing network graph, which can effectively solve the problem of secondary system impact and cascading failures caused by blind switching.

[0078] It should also be noted that graph theory and the push-back power flow algorithm are the underlying core theories of edge autonomous computing. Graph theory abstracts the complex distribution network into a mathematical model composed of vertices and edges, where vertices represent buses or load nodes and edges represent transmission lines or transformers. Based on this model, the push-back algorithm utilizes the typical radial operation characteristics of the distribution network, calculating branch currents through a back-substitution process from end nodes to power source nodes, and then calculating node voltages through a push-back process from power source nodes to end nodes, performing iterative solutions. This algorithm exhibits extremely high convergence speed and computational stability in handling load flow distribution tasks in tree-like distribution networks, making it particularly suitable for deployment in edge control units with limited computing power. Using this algorithm, real-time electrical parameter simulations of multiple backup paths can accurately identify the safety carrying capacity boundaries of each line, effectively avoiding the risk of thermal collapse of heavily loaded lines, and comprehensively enhancing the safety barrier of power supply guarantee strategies.

[0079] S500: Combining the second preferred prediction, analyze the coupling characteristics between the distance of the backup power supply path and the scale of switching actions, and construct a path optimization method based on the minimum comprehensive cost; In the embodiments of this application, reference is made to Figure 5 Step S500 analyzes the coupling characteristics between the distance of the backup power supply path and the scale of the switching action and constructs a path optimization method, including steps E1 to E4: E1: In response to the available path set output by the cost evaluation function, extract the electrical distance parameters, voltage deviation parameters, and number of switching actions parameters for each path in the available path set; Specifically, the electrical distance parameter is the physical representation of the line's equivalent impedance; the voltage deviation parameter is the ratio of the difference between the actual node voltage and the system's rated voltage; and the switch operation quantity parameter refers to the total number of circuit breakers or load switches required to complete the power transfer operation.

[0080] For example, from the set of available paths that have passed security verification, the equivalent impedance value of backup path one, the percentage of voltage drop at the end, and the total number of switches that need to be remotely closed and opened are read.

[0081] E2: Based on the safety margin requirements of the real-time operating conditions of the distribution network, dynamically adjust the optimal proportions of electrical distance parameters, voltage deviation parameters, and the number of switch actions parameters; Specifically, the preferred weighting is the allocation of weighting coefficients in multi-objective optimization calculations.

[0082] In one alternative implementation, E2 assesses whether the current distribution network system is operating under heavy or light load conditions. If it is under heavy load, the system has very little safety margin to withstand voltage fluctuations, so the weight of the voltage deviation parameter is increased. If the system faces a large-scale power outage requiring rapid repair and restoration, the weight of the switch action quantity parameter is increased to reduce mechanical action delay.

[0083] E3: Calculate the comprehensive cost of each path based on the overlap of the optimization ratio; Specifically, comprehensive value is a process of reducing and integrating multiple evaluation indicators into a single scalar value.

[0084] In another alternative implementation, E3 is specifically calculated using a linear weighted summation model. The calculation formula can be expressed as a comprehensive cost function: in, This is the normalized value of the path electrical distance. This is the normalized value of the terminal voltage deviation rate. This is a normalized value for the number of switching operations. The weight parameter must be greater than or equal to 0 and satisfy the following conditions: The weight ratios of these three factors are dynamically adjusted based on the safety margin requirements of real-time operating conditions.

[0085] E4: Select the path with the lowest overall cost and establish the path optimization method.

[0086] For example, by comparing the comprehensive cost of all candidate paths horizontally, the optimal connection corridor with the smallest value is selected as the final power transfer execution channel.

[0087] The expression for the optimal power supply path selection is: It should be noted that in path optimization, the dynamic weighting and comprehensive calculation of multi-objective parameters are key steps in achieving flexible reconfiguration of the distribution network. By extracting physical impedance and operational characteristics and then calculating the cost value through weighted averages, the optimal solution with the highest overall benefit can be found among multiple safe and compliant corridors. This effectively solves the problem of local voltage exceeding limits or frequent switching operations leading to reduced equipment mechanical life caused by traditional local power supply transfer strategies.

[0088] It should also be noted that multi-objective optimization theory and heuristic optimization algorithms are the underlying logical support for intelligent decision-making. Multi-objective optimization aims to find a Pareto optimal solution set among multiple conflicting objectives. In distribution network reconfiguration calculations, it is often necessary to trade off the economic objective of reducing the number of switching operations with the safety objective of reducing line voltage deviation. The comprehensive cost function method transforms the complex multi-objective evaluation into an intuitive single-objective optimization by introducing preference weight parameters. In order to find the minimum value from the huge switch combination state space in a very short time, this scheme implicitly integrates the local optimum search idea of ​​the greedy algorithm, which converges quickly within the constrained decision-making region. This mechanism shows extremely high decision-making efficiency in handling combinatorial optimization tasks with multiple constraints, and is particularly suitable for the rapid autonomous response of distribution network edge computing nodes in communication interruption scenarios. It effectively avoids the computational delay caused by the time consumption of complex algorithms at the remote master station, and comprehensively improves the agility of local microgrid autonomous power restoration.

[0089] S600: The path optimization method outputs action commands, and the weight mapping coefficients are adaptively corrected in combination with the execution feedback data to ensure the power supply of the distribution network nodes.

[0090] In this embodiment of the application, step S600, which combines execution feedback closed-loop linkage to obtain the autonomous recovery model, includes steps F1 to F4: F1: In response to the completion of the action command, collect the steady-state electrical parameters after the switching of the distribution network node.

[0091] Specifically, issuing an action command refers to the operation of the edge control unit sending a remote closing or opening message to the terminal actuator. Steady-state electrical parameters refer to the effective values ​​of voltage and current when the system reaches a stable equilibrium state again after the switch action and a brief electromagnetic transient decay.

[0092] For example, after the system sends a closing command to the tie switch and waits for 300 milliseconds for the mechanical action and system oscillation to subside, the edge terminal begins to continuously collect the bus voltage and branch current of the node as steady-state electrical parameters.

[0093] F2: Compare steady-state electrical parameters with standard power supply thresholds to verify the power supply recovery status.

[0094] Specifically, the standard power supply threshold refers to the allowable voltage deviation range and the upper limit of equipment current carrying capacity in the power grid safety operation specifications.

[0095] In one optional implementation, F2 calculates the ratio of the difference between the actual voltage of the node after power transfer and the rated voltage of the system. If this ratio is within the standard range of ±7%, and the actual current of each branch does not exceed the rated current carrying capacity of the corresponding equipment, then the power transfer restoration status is verified as successful; otherwise, it is determined as a restoration failure or a sub-healthy state.

[0096] F3: Based on the power supply recovery status, encapsulate fault event records, action commands, and steady-state electrical parameters to form a structured sample set.

[0097] Specifically, structured sample sets are standard data formats used for later algorithm iterations and updates. The encapsulation process is a step that links and binds the early multi-dimensional perception of anomaly features, the mid-term optimization decision-making actions, and the later physical execution results.

[0098] For example, the heterogeneous sensing data stream captured before the occurrence of this local fault, the path selection control command with the lowest overall cost, and the steady-state electrical parameters of the final success or failure of power transfer are packaged and a globally unified timestamp label is attached to generate a complete structured sample.

[0099] F4: Adaptively adjusts the weight mapping coefficients using a structured sample set.

[0100] In another optional implementation, F4 utilizes the reward and penalty logic in feedback control. If the power supply recovery is successful, the weight mapping coefficient of the feature component that played a major role in this fault determination is finely adjusted and increased in the multidimensional heterogeneous sensing data stream; if the recovery fails or triggers a new over-limit alarm, the weight mapping coefficient of the corresponding feature that caused the misjudgment is decreased, and the corrected coefficient is updated in the intelligent fault judgment model to achieve a closed-loop strategy.

[0101] It should be noted that during the autonomous recovery control phase, the execution of feedback linkage and adaptive correction are the core elements that break the rigid strategy of traditional open-loop control. By issuing action commands and then verifying the power supply recovery status, a complete data closed-loop link is established in the complex physical power grid, from end-side decision-making to hardware execution and then to status feedback. This can effectively solve the pain points of blindly closing circuits and the inability to self-correct in traditional distribution network fault handling.

[0102] It should also be noted that adaptive control theory and online learning mechanisms are the underlying foundation for the continuous evolution of edge systems. Adaptive control aims to handle internal parameter drift and external physical environment disturbances. In the long-term operation of distribution networks, aging of equipment materials or seasonal load changes often lead to changes in fault representation distribution. The closed-loop feedback mechanism introduces physical execution results as objective evaluation indicators, transforming the real steady-state electrical parameters into a post-hoc score for the pre-judgment model. To ensure that edge nodes become increasingly accurate over their long lifespan, this scheme implicitly integrates data-driven online incremental learning, using a continuously updated set of structured samples to dynamically adjust the sensitivity weights of multi-dimensional features. This mechanism demonstrates strong vitality in handling fault identification tasks in dynamic, non-stationary distribution network environments, and is particularly suitable for the maintenance-free self-evolution of distribution terminals. It effectively avoids the risk of rigid threshold rules becoming ineffective over time, comprehensively improving the long-term intelligence level of local microgrid power supply systems.

[0103] To more intuitively demonstrate the application effect of the power supply guarantee method for distribution networks of the present invention, this embodiment takes the distribution network of three feeder lines from a 10kV urban substation as the implementation object, and presents the full-process numerical simulation and time consumption statistics.

[0104] Basic parameters and initial state of the distribution network: Three feeders, A, B, and C, are branched off from the 10kV substation. Under normal operating conditions, feeder A has a total active load of 80kW, a stable operating current of 150A, and a terminal voltage of 10.05kV. Feeder B has a total active load of 50kW, a stable operating current of 95A, and a remaining current carrying capacity of 205A.

[0105] Fault scenario simulation and multi-dimensional state parameter acquisition: At t=0ms, a single-phase metallic ground fault occurs between the intelligent circuit breakers ICB-A1 and ICB-A2 on line A. At t=5ms, the sensing unit at ICB-A1 synchronously collects all raw fault data at a sampling rate of 1kHz: the line current suddenly increases from the normal 150A to 900A, the phase voltage drops from the normal 10kV to 2kV, the equipment temperature is 27℃, and the equipment vibration frequency is 185Hz.

[0106] Cross-scale feature extraction and state matrix normalization: The sensing unit maps the above data to the same time analysis profile and eliminates dimensional differences to generate a normalized state matrix. Based on the sensor range, rated parameters, and abnormal threshold, the normalized values ​​are calculated as follows: current 0.90, voltage 0.80, temperature 0.10, and vibration 0.30. At t=10ms, this normalized data is synchronously uploaded to the edge control unit.

[0107] Sensitivity weighting and first anomaly determination: The edge control unit, trained based on historical short-circuit fault data, calculates sensitivity differences and retrieves weighted mapping coefficients, specifically: current weight 0.5, voltage weight 0.3, temperature weight 0.1, and vibration weight 0.1. The fusion state index is then calculated. The normal operation threshold set for this scenario The value is 0.50. Since 0.73 is greater than the set threshold of 0.50, the edge control unit triggers the first anomaly determination at t=30ms, locating the faulty section as between ICB-A1 and ICB-A2.

[0108] Backup power supply path topology analysis and second optimal prediction: The system traverses the topology to generate two candidate power supply paths. Path P1 involves the downstream load of line A being transferred to line B via a tie switch, and path P2 involves the downstream load of line A being transferred to line C via a tie switch and line B. The electrical parameters of the paths are calculated and normalized: the normalized value for electrical distance of path P1 is 0.40, the normalized value for voltage deviation is 0.10, and the normalized value for the number of switch actions is 0.333; the corresponding values ​​for path P2 are 1.00, 0.25, and 1.000, respectively. The load constraint of P1 is verified: the remaining capacity of line B (205A) is greater than the pending load of line A (150A), thus satisfying the power supply capacity constraint.

[0109] Comprehensive cost calculation and action command execution: This scenario involves emergency power supply for a primary load. The core objectives are to prioritize voltage quality, reduce the complexity of switching operations, and consider line losses. Therefore, optimal weighting parameters are set. Calculate the comprehensive cost of each path: the cost of path P1 is... The cost of path P2 is... The system selects path P1, which has the lowest overall cost, as the optimal reconfiguration method. At t=50ms, the edge control unit sends a tripping control command to the corresponding smart switch.

[0110] Feedback verification and time consumption statistics: At t=310ms, the intelligent switch completed its operation, and the sensing unit collected data showing that the downstream voltage of line A had recovered to 10.02kV and the current had recovered to 148A, verifying the successful restoration of power supply. Closed-loop feedback data was synchronously packaged into a structured sample set for adaptively correcting the weight mapping coefficients. The total time from fault occurrence to power restoration was 310ms, verifying that the system solution meets the technical indicator of second-level fault recovery.

[0111] Example 3 illustrates a schematic scheme for a power supply guarantee method for a distribution network. It should be noted that the technical solution of this power supply guarantee system for a distribution network belongs to the same concept as the technical solution of the power supply guarantee method for a distribution network described above. Details not described in detail in this example can be found in the description of the technical solution of the power supply guarantee system for a distribution network described above.

[0112] This embodiment also provides a power supply protection system for a distribution network, including: The data acquisition module is used to acquire multi-dimensional operating status parameters of distribution network nodes and construct a multi-dimensional status parameter dataset; The time alignment module is used to extract cross-scale features from the multi-dimensional state parameter dataset and establish a time alignment mapping relationship to obtain a time-aligned normalized state matrix. The intelligent judgment module is used to respond to the completion of the normalized state matrix construction, configure weight mapping coefficients for the normalized state matrix according to the sensitivity differences of the multi-dimensional operating state parameters, and output the fault judgment result based on weighted features for the first anomaly judgment of the distribution network section. The cost evaluation module is used to analyze the topology and load constraint characteristics of the backup power supply path in response to the first anomaly determination being triggered, and generate a cost evaluation function based on the path electrical parameters for the second optimal prediction of the backup power supply path. The optimization and reconstruction module is used to combine the second optimal prediction to analyze the coupling characteristics between the distance of the backup power supply path and the scale of switching action, and to construct a path optimization method based on the minimum comprehensive cost. The execution recovery module is used to output action commands through the path optimization method and adaptively correct the weight mapping coefficients based on the execution feedback data to ensure power supply to the distribution network nodes.

[0113] In the modular sensing layer, the electrical parameter sensing module uses a combination of high-precision Rogowski coils and electromagnetic transformers to collect data in dimensions such as phase current, zero-sequence current, phase voltage, and line voltage. The sampling frequency is adjustable from 10kHz to 1MHz. After second-order filtering, linear amplification, and zero-point calibration, the digital signal is output by the built-in 16-bit or higher high-precision analog-to-digital converter.

[0114] The non-electrical parameter sensing module is divided into two parts: environmental sensing and equipment health sensing. The environmental sensing submodule integrates sensors for temperature and humidity, air pollution level, and SF6 gas concentration. The equipment health sensing submodule integrates infrared temperature measurement, vibration, and ultra-high frequency partial discharge sensors.

[0115] The core processing module of the sensing unit uses a low-power industrial-grade microcontroller or microprocessor with a built-in embedded real-time operating system. The power module is equipped with a supercapacitor backup power supply, which can maintain data buffering and emergency communication for at least 4 hours after the main power supply fails.

[0116] At the execution terminal layer, intelligent switching equipment includes intelligent vacuum circuit breakers and intelligent load switches. Intelligent vacuum circuit breakers have a rated short-circuit breaking capacity of 31.5kA, built-in electric operating mechanisms and measurement and control protection units, and are deployed at line outgoing points and tie nodes. Intelligent load switches have a rated load current breaking capacity of 630A and are deployed at branch nodes.

[0117] The system adopts a layered standardized communication architecture. The process layer adopts the IEC61850 general communication standard for power system automation, the field layer adopts the TCP / IP protocol suite, the management layer adopts the MQTT lightweight publish-subscribe communication protocol, and is equipped with the IEC62351 power system communication security protection mechanism.

[0118] When a modular sensing unit is connected to the power distribution network, the edge control unit first checks its online status via a TCP / IP link. Then, the sensing unit sends a registration request containing device identity information. The edge control unit uses the IEC 62351 security mechanism to complete identity and access verification.

[0119] After successful authentication, the edge control unit uses the standardized information model of the IEC61850MMS service reading module to parse its functional capabilities and parameter attributes, establishes a mapping relationship based on the globally unique device identifier ID, and automatically writes it into the partition device registry to complete the automatic registration and filing of the device.

[0120] In the control command issuance phase, the edge control unit and the smart switch comply with the IEC61850-8-1 standard. The smart switch's built-in measurement and control unit achieves standardized functional encapsulation through XCBR or XSWI logic nodes. The edge control unit issues opening and closing commands through the GOOSE service, with an end-to-end transmission delay of no more than 3ms. After verifying the built-in signature, the smart switch drives the operating mechanism and transmits position signals and execution results back in real time via GOOSE messages.

[0121] This embodiment also provides a computer device applicable to a power distribution network power supply guarantee situation, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement a power distribution network power supply guarantee method as proposed in the above embodiment.

[0122] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements a method for ensuring power supply in a power distribution network as proposed in the above embodiments.

[0123] The storage medium proposed in this embodiment belongs to the same inventive concept as the power supply guarantee method for power distribution networks proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0124] From the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0125] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for ensuring power supply in a distribution network, characterized in that, include: Obtain multi-dimensional operating state parameters of distribution network nodes and construct a multi-dimensional state parameter dataset; Cross-scale features are extracted from the multi-dimensional state parameter dataset and a time-aligned mapping relationship is established to obtain a time-aligned normalized state matrix. In response to the completion of the normalized state matrix construction, weight mapping coefficients are configured for the normalized state matrix based on the sensitivity differences of the multidimensional operating state parameters, and a fault determination result based on weighted features is output for the first anomaly determination of the distribution network section. In response to the first anomaly determination being triggered, the topology and load constraint characteristics of the backup power supply path are analyzed, and a cost evaluation function based on the path electrical parameters is generated for the second optimal prediction of the backup power supply path. Combining the second preferred prediction, the coupling characteristics between the distance of the backup power supply path and the scale of switching actions are analyzed, and a path optimization method based on the minimum comprehensive cost is constructed. The path optimization method outputs action commands, and the weight mapping coefficients are adaptively adjusted based on the execution feedback data to ensure power supply to the distribution network nodes.

2. The power supply guarantee method for a distribution network as described in claim 1, characterized in that: The process of obtaining multi-dimensional operating state parameters of distribution network nodes and constructing a multi-dimensional state parameter dataset includes: In response to the physical access of the sensing terminal to the power distribution network, the global device identifier code of the sensing terminal is parsed through the network layer protocol; Based on the global device identifier code, match the corresponding scene configuration template to establish an edge collaborative communication connection; Read the standardized object model of the sensing terminal and map it to establish the underlying acquisition channel of the multi-dimensional operating state parameters; The underlying acquisition channel is activated to continuously receive heterogeneous measurement point data, which is then aggregated to generate the multidimensional state parameter dataset.

3. The power supply guarantee method for a distribution network as described in claim 1, characterized in that: The step of extracting cross-scale features from the multi-dimensional state parameter dataset and establishing a time-aligned mapping relationship to obtain a time-aligned normalized state matrix includes: In response to the completion of the aggregation of the multidimensional state parameter dataset, a multi-scale analysis window is divided based on a unified timestamp; Within the multi-scale analysis window, the transient effective value features of the high-frequency signal are extracted to complete the downsampling process; Numerical mapping rules are used to complete the steady-state interpolation characteristics of low-frequency signals; The time-aligned mapping relationship is established by integrating the transient effective value features and the steady-state interpolation features, and the normalized state matrix with dimensional differences is output.

4. The power supply guarantee method for a distribution network as described in claim 1, characterized in that: The step of configuring weight mapping coefficients for the normalized state matrix based on the sensitivity differences of the multidimensional operating state parameters, and outputting a fault determination result based on weighted features, includes: In response to the normalized state matrix outputting feature components, the feature components are input into a preset decision tree cluster; The information gain index of the feature components when dividing the running status labels is statistically analyzed to determine the sensitivity difference; The weight mapping coefficients are assigned to the feature components based on the sensitivity differences. The anomaly detection threshold is set using the statistical distribution boundary of the baseline operating data, and the fault detection result is generated by combining the weight mapping coefficient.

5. A power supply guarantee method for a distribution network as described in claim 1, characterized in that: The analysis of the topology and load constraint characteristics of the backup power supply path generates a cost evaluation function based on the path's electrical parameters, including: In response to the first anomaly determination being triggered, candidate paths are searched by traversing the distribution network topology. Verify the node connectivity status of the candidate paths and eliminate paths with isolation and locking attributes; Compare the rated current carrying capacity of the candidate paths with the power of the load to be supplied, and eliminate paths whose capacity exceeds the limit; Calculate the voltage drop magnitude at the end of the candidate path, eliminate paths with voltage drops exceeding the limit, and output the cost evaluation function that conforms to the topology and load constraint characteristics.

6. A power supply guarantee method for a distribution network as described in claim 1, characterized in that: The method analyzes the coupling characteristics between the distance of the backup power supply path and the scale of switching actions, and constructs a path optimization method based on minimizing the overall cost, including: In response to the available path set output by the cost evaluation function, the electrical distance parameters, voltage deviation parameters, and number of switching actions parameters of each path in the available path set are extracted; Based on the safety margin requirements of the real-time operating conditions of the distribution network, the optimal proportions of the electrical distance parameter, the voltage deviation parameter, and the number of switch actions parameter are dynamically adjusted. The comprehensive cost of each path is calculated based on the aforementioned preferred ratio and overlap. The path with the lowest overall cost is selected, and the path optimization method is established.

7. A power supply guarantee method for a distribution network as described in claim 1, characterized in that: The adaptive correction of the weight mapping coefficients based on execution feedback data includes: In response to the completion of the action command, steady-state electrical parameters after the switching of distribution network nodes are collected; The steady-state electrical parameters are compared with the standard power supply threshold to verify the power supply recovery status. Based on the power supply recovery status, a structured sample set is formed by encapsulating the fault event record, the action command, and the steady-state electrical parameters; The weight mapping coefficients are corrected using the structured sample set.

8. A power distribution network power supply protection system, using the method as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire multi-dimensional operating status parameters of distribution network nodes and construct a multi-dimensional status parameter dataset; The time alignment module is used to extract cross-scale features from the multi-dimensional state parameter dataset and establish a time alignment mapping relationship to obtain a time-aligned normalized state matrix. The intelligent judgment module is used to respond to the completion of the normalized state matrix construction, configure weight mapping coefficients for the normalized state matrix according to the sensitivity differences of the multi-dimensional operating state parameters, and output the fault judgment result based on weighted features for the first anomaly judgment of the distribution network section. The cost evaluation module is used to analyze the topology and load constraint characteristics of the backup power supply path in response to the first anomaly determination being triggered, and generate a cost evaluation function based on the path electrical parameters for the second optimal prediction of the backup power supply path. The optimization and reconstruction module is used to combine the second optimal prediction to analyze the coupling characteristics between the distance of the backup power supply path and the scale of switching action, and to construct a path optimization method based on the minimum comprehensive cost. The execution recovery module is used to output action commands through the path optimization method and adaptively correct the weight mapping coefficients based on the execution feedback data to ensure power supply to the distribution network nodes.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the power supply guarantee method for a power distribution network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, which, when executed by a processor, implement the steps of the power supply protection method for a power distribution network as described in any one of claims 1 to 7.

Images