Intelligent dispatching and coordinated operation method for medium and low voltage direct current distribution system

Abstract

The present application relates to the technical field of medium and low voltage direct current power distribution, and specifically discloses a method for intelligent scheduling and coordinated operation of a medium and low voltage direct current power distribution system, which comprises the following steps: deploying and dividing intelligent terminal groups at the starting ends of each T-connection branch along the direct current power distribution trunk line of a long and narrow infrastructure, collecting branch current waveforms, extracting features to obtain power distribution operation data, constructing a baseline feature template and determining whether a disturbance has occurred, if a disturbance has occurred, sorting disturbance power distribution operation data records, dividing edge autonomous areas within the intelligent terminal groups, forming a power distribution topology structure, executing edge collaborative fault discrimination, if an edge collaborative fault has occurred, identifying a fault section and executing a protection action, generating operation and maintenance feedback data, maintaining a dynamically evolved interval probability graph, executing a double-layer monitoring mechanism including a trend analysis layer and an instant analysis layer, completing post-sorting of full-process record data, and realizing closed-loop management from passive protection to active risk prevention.

Description

Technical Field

[0001] This invention relates to the field of medium and low voltage DC power distribution technology, and more specifically to a method for intelligent scheduling and coordinated operation of medium and low voltage DC power distribution systems. Background Technology

[0002] With the large-scale construction of narrow infrastructure such as urban underground utility tunnels and traffic tunnels, the medium- and low-voltage DC power distribution trunk lines laid longitudinally along these lines have become the core circuits for providing power to distributed loads along the route. Such scenarios are characterized by long power supply distances, numerous T-junction branches, enclosed spaces, and difficulties in inspection.

[0003] Existing technologies have the following prominent problems in such scenarios: First, the centralized protection architecture gathers disturbance data from all intelligent terminals on the entire line to the scheduling layer for global decision-making. Any single-point disturbance will trigger data broadcasting across the entire line, resulting in high communication load and large end-to-end latency, making it difficult to meet the millisecond-level protection requirements of DC systems in terms of real-time fault isolation.

[0004] Secondly, fixed threshold criteria are the mainstream configuration of existing protection devices, but they cannot adapt to the gradually changing operating conditions in the pipe gallery due to seasonal changes and load peak and valley fluctuations. They are prone to frequent false trips during peak summer loads or humid transitional seasons, and may fail to trip during low winter loads, resulting in insufficient long-term operational reliability.

[0005] Furthermore, existing fault diagnosis methods struggle to accurately distinguish between faults within T-junction branches and faults in main line sections. Once a disturbance occurs, the outage area is often expanded for rudimentary isolation, negating the flexibility of continuous power supply from DC distribution trunk lines. Simultaneously, existing operation and maintenance models are limited to passive responses to existing faults, lacking quantitative assessment methods for potential risks such as gradual insulation degradation and continuous load growth trends, thus failing to enable preventative topology reconfiguration or planned maintenance before faults occur.

[0006] To address the above problems, this invention proposes an intelligent scheduling and coordinated operation method for medium and low voltage DC power distribution systems. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent scheduling and coordinated operation method for medium and low voltage DC power distribution systems, in order to solve the problems in existing narrow infrastructure scenarios where centralized protection schemes cause data broadcasting and global decision-making across the entire line due to single-point disturbances, resulting in poor real-time fault isolation, and the inability of fixed threshold criteria to adapt to gradual changes in operating conditions, leading to false tripping or failure to trip.

[0008] The objective of this invention can be achieved through the following technical solution: a method for intelligent dispatching and coordinated operation of medium and low voltage DC power distribution systems, comprising the following steps: For DC distribution trunk lines of narrow infrastructure, deploy intelligent terminals at the beginning of each T-junction branch, divide the intelligent terminal groups, and collect branch current waveforms according to typical operating conditions. During the operation period, feature extraction is performed on the branch current waveform to obtain power distribution operation data, a baseline feature template is constructed, and it is determined whether a disturbance occurs. If a disturbance occurs, the disturbance power distribution operation data record is compiled. Within the intelligent terminal group, edge autonomous regions are divided to form a power distribution topology. Edge collaborative fault identification is performed based on the recorded data of the power distribution operation that causes disturbances. If an edge collaborative fault occurs, the faulty section is identified and protection actions are performed to generate operation and maintenance feedback data. By combining operation and maintenance feedback data, a dynamically evolving interval probability map is maintained, and a two-layer monitoring mechanism including a trend analysis layer and a real-time analysis layer is implemented. The trend analysis layer extracts key parameters of waveform features and uses a time-series anomaly detection algorithm to perform long-term drift trend analysis to determine whether to initiate preventive topology reconfiguration. If initiated, a reconfiguration record is generated. The real-time analysis layer determines whether to initiate emergency takeover. If initiated, a takeover record is generated. After completion, the entire process record data is organized.

[0009] Furthermore, the method for acquiring the branch current waveform is as follows: The dispatch layer pre-defines all adjacent intelligent terminals in each fire compartment along the longitudinal direction of the DC power distribution trunk line as the same intelligent terminal group, using the physical boundary of the fire compartment as the dividing boundary. The dispatch layer issues data collection instructions to each group. The data collection instructions include typical operating conditions and the length of the data collection time window. Typical operating conditions include combinations of different seasonal indicators and load level indicators. When any typical operating condition is met simultaneously, the intelligent terminal records the branch current waveform at the beginning of the T-connection branch and adds the corresponding typical operating condition identifier.

[0010] Furthermore, the method for processing power distribution operation data is as follows: Feature extraction is performed on the branch current waveform. The extracted feature dimensions include current amplitude, current change rate, current second derivative, high-frequency energy ratio, and partial discharge pulse count. The corresponding feature components are extracted for each feature dimension, and all feature components are concatenated into a multi-dimensional feature vector as power distribution operation data. Among them, the high-frequency energy ratio is the ratio of the total energy of the frequency band above the high-frequency cutoff frequency to the total energy of the entire frequency band after the branch current waveform is transformed by fast Fourier transform. The partial discharge pulse count is the number of pulses in the branch current waveform whose current amplitude exceeds the preset partial discharge pulse detection threshold and whose pulse width is less than the preset upper limit of pulse width.

[0011] Furthermore, the method for constructing baseline feature templates and determining whether perturbations have occurred is as follows: The runtime phase includes the cold start initial build phase and the online runtime detection phase; In the initial construction phase of cold start, the power distribution operation data collected during the initialization period are grouped according to typical operating conditions. The mean of each feature component of the multidimensional feature vector under the same typical operating condition is used to construct the baseline feature template, and the baseline threshold of each feature component is calculated. During the online operation detection phase, the corresponding baseline feature template is retrieved based on the typical operating conditions of the power distribution operation data. The current amplitude and current change rate are compared with the corresponding baseline thresholds. If either exceeds the threshold, a disturbance is determined, and the data is marked as disturbed power distribution operation data and is not included in the updated sample set. Otherwise, the data is marked as steady-state power distribution operation data and is included in the updated sample set. The baseline feature template is updated on a rolling basis with the update cycle as the interval.

[0012] Furthermore, the method for dividing edge autonomous regions within a smart terminal group is as follows: The scheduling layer obtains the total number of smart terminals in the smart terminal group and divides them based on the preset upper limit and lower limit of the number of terminals; If the number of T-connection branches within a smart terminal group does not exceed the preset upper limit of the number of terminals, it is directly divided into an edge autonomous region. If the number of T-connection branches does not exceed the preset lower limit of the number of terminals, it is recorded as an isolated autonomous region. If the number of terminals exceeds the preset upper limit, the terminals are sorted by trunk line mileage and then divided into initial groups based on the preset lower limit of terminal number. If there are not enough terminals remaining, they are assigned to the last group. Starting from the first initial group, the terminals are merged with adjacent initial groups in turn. The merging condition is that the number of terminals after merging does not exceed the preset upper limit of terminal number and the distance spanned does not exceed the preset upper limit of area span. Otherwise, it is retained as an edge autonomous area.

[0013] Furthermore, the execution method for edge collaboration fault detection is as follows: Edge collaboration fault detection includes a single-terminal autonomous determination stage and a multi-terminal collaborative determination stage; In the single-terminal autonomous judgment stage, if only one smart terminal in the edge autonomous region has a disturbance distribution operation data record within the preset waiting time window, it enters the single-terminal judgment branch. The current change rate and high-frequency energy ratio of the disturbance distribution operation data record are compared with the baseline values ​​of the corresponding baseline feature template. If both exceed the preset fault judgment multiple, the fault type is judged as an internal fault of the T-connection branch. In the multi-terminal collaborative judgment stage, if two or more intelligent terminals have disturbance distribution operation data records, each intelligent terminal extracts multi-dimensional feature vectors from the disturbance distribution operation data records, and uses a dynamic time warping algorithm to calculate the similarity distance between each pair of multi-dimensional feature vectors. If any similarity distance exceeds the preset similarity threshold, the fault type is determined to be a main line section fault between adjacent terminals.

[0014] Furthermore, the similarity distance is calculated as follows: The two multidimensional feature vectors to be paired are normalized by the maximum and minimum values ​​to obtain normalized multidimensional feature vectors. The normalized multidimensional feature vectors are expanded into time series, and a cumulative distortion matrix is ​​constructed. The matrix elements are the Euclidean distances between the normalized multidimensional feature vectors at corresponding time points. By finding the regular path with the minimum cumulative distortion value, the cumulative distortion value at the end of the path is used as the similarity distance.

[0015] Furthermore, the interval probability graph is maintained as follows: The scheduling layer constructs an interval probability map for each edge autonomous region. Using the trunk line mileage station number as the horizontal axis, the DC distribution trunk line segment is discretized into several interval units according to the preset mileage step size. Each interval unit corresponds to a probability storage location with an initial value of zero. After receiving the operation and maintenance feedback data, the fault section information is extracted and mapped to the corresponding interval unit. The cumulative probability estimate of the probability storage location is updated using the exponential sliding window accumulation algorithm. The update method is to weight the current value with a preset probability decay factor and then sum it with the probability increment corresponding to the current edge collaboration fault. The probability increment takes the preset weight coefficient of the fault type. The preset weight coefficient corresponding to the trunk section fault is greater than the preset weight coefficient corresponding to the internal fault of the T-connection branch.

[0016] Furthermore, the method for determining whether to initiate preventative topology reconfiguration is as follows: When the cumulative probability estimate of any interval unit exceeds the preset preventive maintenance probability threshold, the trend analysis layer is triggered, and the high-frequency energy ratio of steady-state power distribution operation data within the trend analysis interval window is extracted as the partial discharge trend indicator and the current amplitude is extracted as the load trend indicator. A time-series anomaly detection algorithm is used to perform long-term drift trend analysis on the time series of partial discharge trend indicators and load trend indicators. If a comprehensive degradation risk is determined, the power distribution topology is combined to determine whether there is a backup power supply path that meets the capacity conditions. If so, a preventive topology reconfiguration is initiated, generating a topology reconfiguration instruction sequence arranged in the order of first closing and then disconnecting.

[0017] Furthermore, the method for determining whether to initiate emergency takeover is as follows: When the cumulative probability estimate of any interval unit exceeds the preset emergency takeover probability threshold, the real-time analysis layer is triggered. The preset emergency takeover probability threshold is higher than the preset preventive maintenance probability threshold. The scheduling layer extracts the protection action result of the most recent edge collaborative fault judgment of the interval unit. If the fault section has been isolated, it is recorded as an isolation event and the probability storage position is reset to zero. If it has not been isolated, emergency takeover is initiated, and an emergency disconnection command is directly issued to the intelligent terminal at the beginning of the corresponding T-connection branch to perform the disconnection operation and update the power distribution topology.

[0018] The beneficial effects of this invention are as follows: 1. This invention decouples the DC power distribution trunk line into an edge autonomous region. By comparing the similarity of disturbance waveforms of adjacent terminals through a dynamic time warping algorithm, it accurately distinguishes between internal faults in T-connection branches and faults in the main trunk line section and isolates them locally. This avoids a single-point disturbance causing a power outage across the entire line and significantly improves the real-time performance and selectivity of fault handling.

[0019] 2. This invention constructs an adaptive baseline feature template for operating conditions and updates it on a rolling basis, so that the disturbance criterion continuously follows the gradual change of the environment and reduces the probability of false activation or failure to activate; at the same time, it uses a dynamic evolution interval probability map to drive a dual-layer monitoring mechanism of trend analysis and real-time analysis, so as to realize closed-loop management from passive protection to active risk prevention. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart illustrating the steps of the intelligent scheduling and coordinated operation method for medium and low voltage DC power distribution systems according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the specific steps involved in generating operation and maintenance feedback data in the intelligent scheduling and coordinated operation method for medium and low voltage DC power distribution systems described in this embodiment of the invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Example 1

[0024] Please see Figure 1As shown in the embodiments of the present invention, the intelligent scheduling and coordinated operation method for medium and low voltage DC distribution systems aims to solve the problems of poor real-time fault isolation and false tripping caused by the inability of fixed threshold criteria to adapt to gradual changes in operating conditions in existing narrow infrastructure scenarios due to the centralized protection scheme causing data broadcasting and global decision-making across the entire line due to single-point disturbances. By decoupling the entire DC distribution trunk line longitudinally into several edge autonomous regions, and comparing the similarity of disturbance waveforms of adjacent intelligent terminals in each edge autonomous region based on a dynamic time warping algorithm to perform edge collaborative fault discrimination, the fault is accurately located as a fault within a T-junction branch or a fault in a section of the main trunk line and isolated locally. An interval probability map is constructed and dynamically evolved to drive a two-layer monitoring mechanism that includes trend analysis and real-time analysis, realizing closed-loop management from passive protection to active risk prevention. Specifically, it includes the following steps: S1: For DC power distribution trunk lines of narrow infrastructure, deploy intelligent terminals at the beginning of each T-connection branch, divide the intelligent terminal groups, and collect branch current waveforms according to typical operating conditions. Specifically, at the beginning of each T-connection branch along the DC power distribution trunk line of the narrow infrastructure, a smart terminal with high-speed sampling and edge computing capabilities is deployed. The smart terminal continuously collects the current signal at the beginning of the T-connection branch at a preset sampling frequency through its built-in high-speed analog-to-digital conversion unit. It should be noted that the preset sampling frequency is pre-calibrated based on the characteristic frequency range of transient disturbances in the DC power distribution system, combined with the Nyquist sampling theorem and engineering margin. It should be noted that elongated infrastructure refers to enclosed or semi-enclosed structures that extend in a single direction and whose length is much greater than their width and height, including urban underground utility tunnels and traffic tunnels. DC power distribution trunk line refers to a medium- or low-voltage DC power supply circuit that runs longitudinally along the elongated infrastructure to provide power to distributed loads along the line. T-connection branch refers to a branch circuit that is drawn from the DC power distribution trunk line to supply power to a single or group of load points, and its starting point is the electrical connection point between the branch line and the DC power distribution trunk line. The back-end main station system that establishes communication connections with each smart terminal is referred to as the scheduling layer. The scheduling layer pre-determines the spatial location of each T-connection branch within the narrow infrastructure, using the physical boundary of the fire protection zone of the pipe gallery as the dividing boundary, and divides all adjacent smart terminals in each fire protection zone along the longitudinal direction of the DC power distribution trunk line into the same smart terminal group. For each group of smart terminals, the scheduling layer issues a collection instruction to each group. The collection instruction includes typical operating conditions and the length of the collection time window. The typical operating conditions include a combination of different seasonal identifiers and load level identifiers. Based on the collection instruction, the smart terminal records the branch current waveform at the beginning of the T-connection branch with the length of the collection time window when any typical operating condition is met at the same time, and adds the corresponding typical operating condition identifier to the branch current waveform. Among them, the seasonal indicators include summer, winter and transition seasons, and the load level indicators include peak load, off-peak load and off-peak load; The scheduling layer calibrates the load classification threshold based on the historical load data of the section where the smart terminal group is located, and takes a sampling period of a preset duration forward from the sampling time as the end point at each sampling time. It obtains the active power time series of the section where the smart terminal group is located within the sampling period, calculates the average power of the active power time series, compares it with the load classification threshold, and obtains the corresponding load level identifier. It should be noted that historical load data refers to the active power time sequence recorded by each T-connection branch in the past operating cycle according to a preset fixed time interval, stored in the scheduling layer database. It should be noted that the typical operating conditions included in the acquisition instructions cover all operating conditions. The continuous acquisition of the current signal at the beginning of the T-connection branch by the intelligent terminal is executed as a continuous task throughout the entire life cycle of the system. The acquired current signals are collectively referred to as branch current waveforms. The scheduling layer adds corresponding typical operating condition identifiers to the branch current waveforms through the typical operating conditions in the acquisition instructions in order to achieve classified management according to operating conditions. S2: During the operation period, feature extraction is performed on the branch current waveform to obtain power distribution operation data, a baseline feature template is constructed, and it is determined whether a disturbance occurs. If a disturbance occurs, the disturbance power distribution operation data record is organized. Specifically, the intelligent terminal performs feature extraction on the branch current waveform. The extracted feature dimensions include current amplitude, current change rate, current second derivative, high-frequency energy ratio, and partial discharge pulse count. For each feature dimension, the corresponding feature components are extracted. Among them, the current amplitude is taken as the maximum absolute value of all sampled values ​​in the corresponding branch current waveform. The absolute value of the slope of the branch current waveform at each sampling time is calculated and organized into a change rate time series. The maximum value of the change rate time series is taken as the extracted current change rate. The second derivative of the current is taken as the maximum absolute value of the slope of the change rate time series at each sampling time. The calculation method for the high-frequency energy ratio is as follows: after performing a fast Fourier transform on the branch current waveform, a preset multiple of the DC power distribution system switching frequency is used as the high-frequency cutoff frequency. The ratio of the total energy in the frequency band above the high-frequency cutoff frequency to the total energy in the entire frequency band is calculated as the high-frequency energy ratio of the branch current waveform. The partial discharge pulse count is calculated as follows: the number of pulses in the branch current waveform whose current amplitude exceeds the preset partial discharge pulse detection threshold and whose pulse width is less than the preset upper limit of pulse width is counted to obtain the partial discharge pulse count. All feature components are concatenated into a multi-dimensional feature vector to obtain the corresponding power distribution operation data; Based on power distribution operation data, a baseline feature template is constructed during the operation period and updated at a preset baseline update cycle. The operation period includes a cold start initial construction phase and an online operation detection phase. It should be noted that the cold start initial construction phase refers to the period after the system is first powered on, during which no baseline feature templates are stored in the databases of each smart terminal, while the online operation detection phase refers to the period during which baseline feature templates are stored in the databases of each smart terminal. Specifically, during the initial cold start construction phase, the intelligent terminal enters the initialization acquisition mode. Within the preset initialization time, the system is in a disturbance-free steady state by default. All power distribution operation data collected within the initialization time are grouped according to typical operating conditions. For the multi-dimensional feature vector under the same typical operating condition, the mean of each feature component is taken as the baseline value of the corresponding feature component. The baseline feature template under the corresponding typical operating condition is constructed, and the sum of the mean and the corresponding three times the standard deviation is calculated as the corresponding baseline threshold. During the initial cold start construction phase, no judgment on whether a disturbance has occurred is performed. It should be noted that after the initial cold start construction phase is completed, the smart terminal stores the baseline feature template in the local non-volatile storage unit and uploads the copy to the scheduling layer for archiving via the uplink communication link; During the online operation detection phase, the intelligent terminal uses the latest baseline feature template as a benchmark to perform disturbance judgment on each newly generated power distribution operation data; Specifically, the intelligent terminal retrieves the baseline feature template under the same typical operating conditions from the locally stored baseline feature template based on the typical operating condition identifier corresponding to the power distribution operation data, and compares the current amplitude and current change rate in the power distribution operation data with the corresponding baseline thresholds respectively. If the current amplitude exceeds the corresponding baseline threshold or the current change rate exceeds the corresponding baseline threshold, a disturbance is determined to have occurred, and the power distribution operation data is marked as disturbed power distribution operation data. Disturbed power distribution operation data is not included in the updated sample set of the baseline feature template; otherwise, no disturbance is determined to have occurred. If no disturbance occurs, the power distribution operation data is marked as steady-state power distribution operation data. The intelligent terminal incorporates the steady-state power distribution operation data into the updated sample set under the corresponding typical operating conditions. At the end of each update cycle, the mean of each feature dimension under the typical operating conditions is recalculated, and the baseline feature template is updated on a rolling basis so that the baseline feature template continuously follows the gradual drift of equipment status and environmental conditions. If a disturbance occurs, the smart terminal reads the pre-configured spatial location identifier of the T-connection branch from the local storage. The spatial location identifier includes the fire compartment number of the utility tunnel where the smart terminal is located and the trunk line mileage station number. The trunk line mileage station number indicates the longitudinal mileage position of the beginning of the T-connection branch on the DC power distribution trunk line. The spatial location identifier is encapsulated into the disturbance power distribution operation data according to the preset data frame format to obtain the disturbance power distribution operation data record, which is stored in the local non-volatile storage unit. It should be noted that, in response to the problem that existing fixed threshold criteria cannot adapt to the gradual drift of equipment status and environmental conditions under typical operating conditions such as seasonal changes and load peak-valley fluctuations, which can easily lead to false tripping and failure to trip, this step aims to construct baseline feature templates for different typical operating conditions and use only steady-state power distribution operation data to continuously update the templates. This ensures that the baseline threshold for disturbance judgment continuously follows the actual operating conditions and system evolution, significantly reducing the probability of false tripping and failure to trip, and improving the long-term reliability of medium and low voltage DC power distribution systems in narrow infrastructure. S3: Divide edge autonomous regions within the intelligent terminal group to form a power distribution topology. Perform edge collaborative fault identification based on the disturbance power distribution operation data records. If an edge collaborative fault occurs, identify the fault section and perform protection actions to generate operation and maintenance feedback data. like Figure 2 As shown, the specific method for generating operation and maintenance feedback data is as follows; Specifically, the scheduling layer obtains the total number of smart terminals in each smart terminal group, and, in combination with the preset upper limit and lower limit of the number of terminals, directly calculates the division scheme of the edge autonomous region. If the number of T-connection branches within a single smart terminal group does not exceed the preset upper limit of the number of terminals, the smart terminal group will be directly divided into an edge autonomous region. If the number of T-connection branches within the edge autonomous region does not exceed the preset lower limit of the number of terminals, the edge autonomous region will be recorded as an isolated autonomous region. If the number of T-connection branches in a single smart terminal group exceeds the preset terminal number limit, the scheduling layer obtains the total number of smart terminals in each smart terminal group, arranges them in ascending order of trunk line mileage station number, and divides the smart terminal sequence into several initial groups starting from the beginning, using the preset terminal number limit as the basic unit. Each initial group contains the preset terminal number limit number of adjacent terminals. If the number of remaining smart terminals at the end is less than the preset terminal number limit, then all remaining terminals are assigned to the last initial group. For initial groups that are not classified as edge autonomous regions, starting from the first initial group, merge with adjacent initial groups. If the number of terminals after merging does not exceed the preset terminal number limit and the distance spanned does not exceed the preset region span limit, then merge into one initial group; otherwise, retain the first initial group as an edge autonomous region. After traversing the initial groups and completing the process, if there are still unmerged initial groups, then each initial group is itself a marginal autonomous region. It should be noted that the purpose of dividing the edge autonomous region is to decouple the entire DC distribution trunk line longitudinally into several local sections that can perform edge collaborative fault identification. This limits the sharing scope of disturbance distribution operation data records, the scope of participating terminals in edge collaborative fault identification, and the impact scope of protection actions to local sections, avoiding single-point disturbances from triggering full-line data broadcasting and global decision-making. This enables rapid, local, and accurate fault isolation in the distributed architecture of narrow infrastructure. It should be noted that the purpose of setting a preset lower limit and a preset upper limit for the number of terminals is to achieve an engineering balance between the feasibility and real-time performance of edge collaborative fault detection. The preset lower limit of 2 terminals ensures that there are at least two adjacent smart terminals in the edge autonomous region, so that edge collaborative fault detection can be physically executed. The preset upper limit of terminals prevents the number of terminals in the edge autonomous region from being too large or the vertical span from being too large, which would cause the end-to-end delay of hop forwarding to exceed the preset waiting time window, or the edge computing unit to process too many smart terminals, which would burden the computing power and affect real-time performance. It should be noted that an isolated autonomous region refers to an edge autonomous region where the number of T-connection branches within a smart terminal group is too small to form adjacent terminal pairings within the region. Therefore, it can only rely on its own single-terminal criteria to independently complete fault diagnosis and does not have the ability to make collaborative judgments with multiple terminals. The division results of each peripheral autonomous region and isolated autonomous region are generated by the scheduling layer during the system initialization phase by generating a regional configuration table. The regional configuration table records the spatial location identifiers of all intelligent terminals under the jurisdiction of the peripheral autonomous regions and isolated autonomous regions. The regional configuration table is uniformly distributed to each intelligent terminal by the scheduling layer through the communication network. The connection relationship of the main line segments between T-connection branches and the division boundary of the peripheral autonomous regions and isolated autonomous regions together form the power distribution topology. Within each edge autonomous region, each intelligent terminal establishes a communication connection through a point-to-point high-speed edge communication link laid along the pipe gallery or tunnel wall. The logical topology of the communication connection is a linear topology. Each intelligent terminal establishes a direct communication relationship only with the upstream and downstream intelligent terminals adjacent to each other along the longitudinal direction of the DC power distribution trunk line within its edge autonomous region. The intelligent terminals located at both ends of the edge autonomous region have only one adjacent communication node. The physical medium of the point-to-point high-speed edge communication link is shielded twisted pair cable, and the communication protocol adopts a preset lightweight real-time data exchange protocol. If any smart terminal generates a disturbance distribution operation data record, it will forward the disturbance distribution operation data record to all other smart terminals in the same edge autonomous region hop by hop according to the communication relationship of the linear topology through the point-to-point high-speed edge communication link. Each smart terminal maintains a regional shared data cache queue locally to temporarily store the received disturbance distribution operation data records. After each intelligent terminal has completed the sharing of data recording of the disturbance power distribution operation within the area, it independently runs the same edge decision algorithm to perform edge collaborative fault identification. Specifically, the edge decision-making algorithm is deployed in the edge computing unit on each smart terminal. The execution process of edge collaborative fault judgment includes two stages: the single terminal autonomous judgment stage and the multi-terminal collaborative judgment stage. In the single-terminal autonomous determination stage, the number of smart terminals with disturbance distribution operation data records in the statistical area shared data cache queue is counted. If only one smart terminal has disturbance distribution operation data records and does not receive disturbance distribution operation data records sent by any other smart terminal in the edge autonomous area within the preset waiting time window, then the single-terminal determination branch is entered. It should be noted that the preset waiting time window is determined by multiplying the sum of the maximum end-to-end transmission delay of the point-to-point high-speed edge communication link and the processing delay inside the smart terminal with a preset tolerance coefficient. In the single-terminal judgment branch, the smart terminal reads two feature components, the current change rate and the proportion of high-frequency energy, from the disturbance power distribution operation data record. It also retrieves the baseline feature template corresponding to the current typical operating conditions from local storage and compares the two feature components with the corresponding baseline values. If both feature components exceed the preset fault judgment multiple of the corresponding baseline values, it is determined that an edge coordination fault has occurred, and the fault type is the internal fault of the T-connection branch corresponding to the smart terminal. For internal faults in T-connection branches, the intelligent terminal immediately generates a local disconnection command and controls the solid-state circuit breaker installed at the beginning of the T-connection branch to perform a disconnection operation through the solid-state circuit breaker drive circuit, cutting the faulty branch off from the DC power distribution trunk line. If two or more smart terminals have records of disturbance power distribution operation data, the multi-terminal collaborative judgment stage will be entered. In the multi-terminal collaborative judgment stage, a multi-terminal collaborative judgment branch is adopted. Each intelligent terminal extracts all the disturbance distribution operation data records sent by intelligent terminals that have disturbance distribution operation data records from the regional shared data cache queue, and extracts the multi-dimensional feature vector in each disturbance distribution operation data record. For any smart terminal, the smart terminal will pair its own multidimensional feature vector with the multidimensional feature vectors of other smart terminals in the regional shared data cache queue one by one, and use the dynamic time warping algorithm to calculate the similarity distance between each pair of multidimensional feature vectors. Specifically, the calculation method of the dynamic normalization time algorithm is as follows: for two multidimensional feature vectors to be paired, each feature component is normalized based on the pre-calibrated corresponding standard threshold interval to obtain a normalized multidimensional feature vector including the normalized feature components. The normalized multidimensional feature vectors of each time series are expanded into two time series according to the time sampling order. A cumulative distortion matrix is ​​constructed with the lengths of the two time series being the number of rows and columns, respectively. Each element in the cumulative distortion matrix represents the Euclidean distance between the normalized multidimensional feature vectors of the two time series at the corresponding time point. Starting from the bottom left corner of the cumulative distortion matrix, the cumulative distortion is gradually advanced in the three directions of right, up, and upper right, selecting the direction with the smallest cumulative distortion value at each step, until the top right corner of the cumulative distortion matrix is ​​reached. The cumulative distortion value on the final path is the normalized path distance between the two time series, and the normalized path distance is output as the similarity distance. Wherein, the cumulative distortion value represents the sum obtained by gradually accumulating the local distance between two selected time points along each step of the regular path from the starting point of the regular path to the current step. Within the autonomous region at the edge, each smart terminal is sorted in ascending order of the trunk line mileage markers, forming a one-dimensional terminal sequence along the trunk line spatial direction. The smart terminals traverse from one end of the one-dimensional terminal sequence to the other, and for each pair of adjacent smart terminals along the trunk line, the similarity distance between the smart terminals is read. If any similarity distance exceeds the preset similarity threshold, an edge coordination failure is determined to have occurred. The failure type is a backbone segment failure between adjacent terminals. The fault segment is located as the backbone segment between adjacent terminals, and adjacent fault segments are merged into one fault segment. For the intelligent terminals at both ends of the fault section, a coordinated disconnection command is generated for each intelligent terminal. After receiving the coordinated disconnection command, each intelligent terminal immediately controls the solid-state circuit breaker installed at the beginning of the T-connection branch to perform a disconnection operation, thus isolating the fault section from the DC power distribution trunk line. Among them, trunk line section faults between adjacent terminals and internal faults in T-connection branches are two types of edge coordination faults. If the similarity distance between all adjacent terminals does not exceed the preset similarity threshold, it is determined that no edge coordination failure has occurred and is recorded as an external disturbance penetration event. Each smart terminal maintains its corresponding solid-state circuit breaker in the closed conduction state to ensure uninterrupted normal power supply. The multi-dimensional feature vector of the external disturbance penetration event, the judgment timestamp, and the number of smart terminals participating in the judgment are organized into an external disturbance penetration record. This record is then aggregated to a preset reporting terminal within the edge autonomous region through a point-to-point high-speed edge communication link. The reporting terminal then uploads the external disturbance penetration record to the scheduling layer through the uplink communication link between the edge autonomous region and the scheduling layer. It should be noted that the reporting terminal is designated by the scheduling layer during the system initialization phase based on a comprehensive evaluation of the communication quality and processing capabilities of the smart terminals in each edge autonomous region, and the reporting terminal identifier is marked in the regional configuration table. It should be noted that the isolated autonomous region contains only one smart terminal, does not establish a point-to-point high-speed edge communication link, and does not execute multi-terminal collaborative judgment branch. After generating disturbance power distribution operation data records, the smart terminal in the isolated autonomous region directly executes the single-terminal judgment branch to complete edge collaborative fault identification and protection actions. After completing the edge collaborative fault identification and protection actions, each smart terminal participating in the identification will integrate the complete process data of edge collaborative fault identification, including the number of smart terminals with disturbance power distribution operation data records, the similarity distance between each adjacent terminal, fault section information, the final protection action and action timestamp, into operation and maintenance feedback data. Operation and maintenance feedback data is stored in the local non-volatile storage unit of each smart terminal and uploaded to the scheduling layer through the uplink communication link; It should be noted that, in response to the problem that centralized protection cannot quickly and accurately identify fault sections in the event of local faults due to the numerous T-connection branches along the DC distribution trunk line in narrow infrastructure, the edge collaborative fault identification based on the dynamic time warping algorithm calculates the similarity distance between the normalized multidimensional feature vectors of the disturbance distribution operation data collected by adjacent intelligent terminals within the edge autonomous region. This enables accurate differentiation and location of faults inside the T-connection branch and faults in the trunk line section, limiting the scope of protection decisions and actions to the local area, and achieving rapid, local, and accurate fault isolation. S4: Combine operation and maintenance feedback data to maintain a dynamically evolving interval probability map, and execute a two-layer monitoring mechanism including a trend analysis layer and a real-time analysis layer. The trend analysis layer extracts key parameters of waveform features and uses a time-series anomaly detection algorithm to perform long-term drift trend analysis to determine whether to initiate preventive topology reconstruction. If initiated, a reconstruction record is generated. The real-time analysis layer determines whether to initiate emergency takeover. If initiated, a takeover record is generated. After completion, the entire process record data is organized. Specifically, during the system initialization phase, the scheduling layer constructs and maintains an interval probability map for each edge autonomous region. The interval probability map uses the trunk line mileage station number along the longitudinal direction of the DC distribution trunk line within the edge autonomous region as the horizontal axis and the preset mileage step size as the minimum granularity. It discretizes the DC distribution trunk line segment under the jurisdiction of the edge autonomous region into several continuous interval units. Each interval unit corresponds to a probability storage location. The initial value of each probability storage location is set to zero. The vertical axis of the interval probability map represents the cumulative probability estimate of the occurrence of edge collaborative failure in the corresponding interval unit. It should be noted that the preset mileage step size is determined comprehensively based on the average spacing of T-junction branches within the narrow infrastructure and the fault location accuracy requirements. The mileage step size is no greater than the minimum value of the spacing between T-junction branches, which is the lower limit of the preset number of terminals, to ensure that at least each T-junction branch corresponds to an independent interval unit. When the scheduling layer receives maintenance feedback data uploaded from any smart terminal or reporting terminal via the uplink communication link, the scheduling layer extracts fault section information from the maintenance feedback data. The fault section information is represented in the form of a continuous interval of trunk mileage station number range within the edge autonomous area participating in the judgment. The scheduling layer maps the fault section to the corresponding interval unit in the interval probability map and performs probability accumulation update on the probability storage location corresponding to the interval unit covered by the fault section. The calculation method for probability accumulation update is as follows: For each interval unit covered by the faulty section, the scheduling layer reads the currently stored cumulative probability estimate of the interval unit, adopts the exponential sliding window accumulation algorithm, takes the current time as the benchmark, and performs weighted attenuation on the currently stored cumulative probability estimate according to the preset probability attenuation factor, and sums it with the probability increment corresponding to the current edge coordination fault. The calculated data is used to update the cumulative probability estimate. The probability increment takes the preset weight coefficient of the fault type of the edge coordination fault. Among them, the preset weight coefficient corresponding to the trunk section fault is greater than the preset weight coefficient corresponding to the internal fault of the T-connection branch. It should be noted that the preset weighting coefficient is determined according to the degree of impact of the fault type on the continuous power supply capacity of the DC distribution trunk line. The fault in the main trunk line section requires the isolation of the loads at both ends of the section, and the impact range is greater than that of the internal fault of the T-connection branch. Therefore, a higher weighting coefficient is assigned. For the interval probability map after each probability accumulation update, the scheduling layer determines whether to trigger the long-term drift trend analysis of the trend analysis layer: if the cumulative probability estimate of any interval unit in the interval probability map exceeds the preset preventive maintenance probability threshold, the trend analysis layer is triggered. The scheduling layer starts from the segment corresponding to the interval unit, takes the segment as the starting point, and extends the preset correlation analysis mileage range upstream and downstream in the longitudinal direction of the DC distribution trunk line to form a trend analysis interval window. The dispatch layer retrieves the multi-dimensional feature vectors corresponding to the steady-state power distribution operation data uploaded to the dispatch layer archive by all intelligent terminals within the trend analysis interval window within the preset trend backtracking time, and extracts key waveform feature parameters from the multi-dimensional feature vectors. Key waveform characteristic parameters include the proportion of high-frequency energy and the current amplitude in the multi-dimensional feature vector corresponding to steady-state power distribution operation data. The dispatching layer uses the extracted high-frequency energy proportion as a partial discharge trend indicator and the current amplitude as a load trend indicator. The scheduling layer uses a preset time-series anomaly detection algorithm to perform long-term drift trend analysis on the time series of partial discharge trend indicators and load trend indicators within the trend backtracking time period; Specifically, the preset time-series anomaly detection algorithm adopts a combination of Mann-Kendall trend test and Sen's slope estimator. For the partial discharge trend index time series, the scheduling layer takes the start time of the preset trend backtracking duration as the starting point, arranges the sampling timestamps of the steady-state power distribution operation data into a sequence, performs the Mann-Kendall trend test, calculates the trend test statistic and the corresponding significance level. If the significance level is less than the preset trend significance threshold, and the slope value calculated by Sen's slope estimator is positive, it is determined that the partial discharge trend index has a significant monotonically increasing trend, indicating that the insulation state of the corresponding section has a continuous deterioration drift, and a partial discharge trend alarm record is generated. It should be noted that steady-state power distribution operation data refers to power distribution operation data that has been determined to be undisturbed during the online operation detection phase and has been included in the baseline feature template update sample set. This data is periodically uploaded by the smart terminal or after generation to the scheduling layer for archiving via the uplink communication link. The scheduling layer maintains a time-series database of steady-state power distribution operation data under corresponding typical operating conditions for each smart terminal. For the time series of load trend indicators, the scheduling layer uses the same combined algorithm of Mann-Kendall trend test and Sen's slope estimator. If the time series of load trend indicators has a significant monotonically increasing trend and the slope value calculated by Sen's slope estimator is positive, it indicates that the load level of the corresponding section has a continuous growth drift, and a load growth trend alarm record is generated. The dispatch layer aggregates partial discharge trend alarm records and load growth trend alarm records. If partial discharge trend alarm records and load growth trend alarm records exist simultaneously within the trend analysis interval window, and the interval units corresponding to the partial discharge trend alarm records and load growth trend alarm records overlap in spatial location, the dispatch layer determines that the corresponding segment of the trend analysis interval window has a comprehensive deterioration risk and generates a comprehensive trend analysis alarm record. The dispatch layer determines whether to initiate preventive topology reconfiguration based on trend analysis, comprehensive alarm records, and the current power distribution topology. Specifically, the scheduling layer maps the interval units corresponding to the trend analysis and comprehensive alarm records to the distribution topology. The distribution topology is a description of the topological relationship formed by the spatial location identifiers of each smart terminal, the connection relationship of the main line segments between T-connection branches, and the division boundaries of edge autonomous regions and isolated autonomous regions. The dispatch layer reads the power distribution topology, starts from the interval unit corresponding to the trend analysis comprehensive alarm record, traces upstream along the vertical direction of the DC power distribution trunk line to the beginning of the nearest T-connection branch with tie switch function, and traces downstream to the beginning of the nearest T-connection branch with tie switch function. It then determines whether there is a backup power supply path for load transfer through tie switch in the upstream and downstream sections, under the condition of avoiding the interval unit corresponding to the trend analysis comprehensive alarm record. It should be noted that the T-connection branch start point with tie switch function means that in addition to the solid circuit breaker, the T-connection branch start point is also equipped with a tie path to the adjacent power supply circuit. The dispatch layer database stores the spatial location identifier and tie relationship table of each T-connection branch start point with tie switch function. The tie relationship table records the backup power supply path that the tie switch can access and its available capacity. If a feasible backup power supply path exists, and the available capacity of the backup power supply path can meet the power demand of the load to be transferred, the dispatch layer determines that the conditions for preventive topology reconfiguration are met and initiates preventive topology reconfiguration. It should be noted that the power demand of the load to be transferred is calculated by the dispatching layer based on the current amplitude in the steady-state power distribution operation data uploaded by the intelligent terminal within the trend analysis interval window, combined with the rated voltage of the DC power distribution trunk line. The execution process of preventive topology reconfiguration is as follows: The dispatch layer generates a topology reconfiguration instruction sequence, which includes closing instructions for the tie switches corresponding to the backup power supply path and opening instructions for the solid-state circuit breakers at both ends of the section unit corresponding to the trend analysis comprehensive alarm record. The instructions are arranged in the order of closing first and then opening. The dispatch layer sends the topology reconfiguration instruction sequence to the corresponding smart terminal or tie switch controller for execution one by one through the uplink communication link. After the topology reconfiguration is completed, the dispatch layer updates the distribution topology structure and generates a reconfiguration record. The reconfiguration record includes the start time of the preventive topology reconfiguration, the interval unit corresponding to the trend analysis comprehensive alarm record, the spatial location identification of the tie switches and solid circuit breakers involved in the reconfiguration, and a snapshot comparing the distribution topology before and after the reconfiguration. If there is no feasible backup power supply path, or the available capacity of the backup power supply path does not meet the power demand of the load to be transferred, the dispatch layer determines that the conditions for preventive topology reconfiguration are not met, generates a preventive maintenance work order, which includes the interval unit corresponding to the trend analysis comprehensive alarm record, a summary of the trend analysis results, and a suggested maintenance window time. The work order is pushed to the operation and maintenance personnel through the human-machine interface of the dispatch layer, and topology reconfiguration is not performed. For the interval probability map after each probability accumulation update, if the cumulative probability estimate of any interval cell in the interval probability map exceeds the preset emergency takeover probability threshold, the dispatch layer triggers the real-time analysis layer. The preset emergency takeover probability threshold is higher than the preset preventive maintenance probability threshold. It should be noted that the preset emergency takeover probability threshold indicates that the probability of an edge coordination failure occurring in a short period of time has risen to the point where direct intervention and control by the dispatch layer is required. Its value is calibrated based on the reliability target of the DC power distribution system and the historical fault statistical distribution. After the real-time analysis layer is triggered, the scheduling layer reads the fault section information corresponding to the interval unit that exceeds the preset emergency takeover probability threshold, and extracts the complete process data of the most recent edge collaborative fault identification corresponding to the interval unit from the operation and maintenance feedback data. The complete process data includes the number of smart terminals with disturbance distribution operation data records, the similarity distance between each adjacent terminal, the fault section, and the final protection action executed. The scheduling layer determines whether the protection action finally executed by the interval unit in the most recent edge collaborative fault identification has isolated the faulty section; If the protection action has isolated the faulty section, it means that the edge autonomous region has completed the fault isolation. The scheduling layer will record this real-time analysis layer trigger as an executed isolation event, generate a takeover record, which includes the trigger time, the corresponding interval unit, a summary of the protection actions executed in the edge autonomous region, and reset the probability storage location of the corresponding interval unit to zero. This real-time analysis layer processing ends. If the protection action fails to isolate the faulty section, it indicates that the fault still exists in the DC power distribution trunk line, and the dispatching level determines that emergency takeover needs to be initiated. The emergency takeover execution process is as follows: The dispatch layer directly generates an emergency disconnection command for the solid-state circuit breaker at the beginning of the T-connection branch at both ends of the interval unit. The emergency disconnection command bypasses the edge decision algorithm of the edge autonomous region and is directly sent by the dispatch layer to the corresponding smart terminal through the uplink communication link. After receiving the emergency disconnection command, the smart terminal immediately controls the solid-state circuit breaker installed at the beginning of the T-connection branch to perform the disconnection operation. After confirming that the disconnection operation has been completed, the dispatch layer updates the power distribution topology and marks the corresponding interval unit as an emergency isolation state in the interval probability diagram, generating a takeover record. The takeover record includes the trigger time of the emergency takeover, the section unit corresponding to the takeover action, the spatial location identifier of the solid circuit breaker that performed the disconnection operation, the timestamp confirming the completion of the disconnection operation, and a description of the changes in the power distribution topology before and after the takeover action. Throughout the entire lifecycle of the system, the scheduling layer continuously summarizes reconstruction and takeover records, and merges them with operation and maintenance feedback data, trend analysis alarm records, partial discharge trend alarm records, and load growth trend alarm records into full-process record data. The full-process record data is stored in the scheduling layer database, and provides multi-dimensional retrieval and statistical analysis functions by edge autonomous region, by time range, and by fault type through the scheduling layer's data service interface. It should be noted that, in response to the problem that traditional protection schemes can only passively respond to faults that have occurred and cannot quantitatively assess and proactively warn of fault risks in DC power distribution trunk sections, the interval probability map uses an exponential sliding window accumulation algorithm to update and decay the segment probability of each edge collaborative fault, forming a dynamically evolving spatial risk quantification model, which provides a data foundation for operation and maintenance decisions to shift from passive response to proactive prevention. It should be noted that, in order to address the difficulty in timely detection of gradual insulation degradation and continuous load growth trends from massive steady-state data during routine operation and maintenance, and the potential ineffective isolation of faults by edge autonomous regions when the probability of failure increases sharply, this step employs a dual-layer monitoring mechanism triggered by preset preventive maintenance probability thresholds and emergency takeover probability thresholds. The trend analysis layer actively identifies long-term drift trends and initiates preventive topology reconstruction using a time-series anomaly detection algorithm, while the real-time analysis layer directly executes emergency takeover when necessary, bypassing edge autonomous regions. This achieves a seamless closed loop from passive protection to proactive risk prevention and hierarchical control. The technical solution of this invention is as follows: For the DC distribution trunk line of a narrow infrastructure, deploy intelligent terminals at the beginning of each T-junction branch, divide the intelligent terminal group, and collect the branch current waveform according to typical operating conditions. During the operation period, extract the features of the branch current waveform to obtain distribution operation data, construct a baseline feature template and determine whether a disturbance has occurred. If a disturbance has occurred, organize the disturbance distribution operation data record, divide the edge autonomous region within the intelligent terminal group to form a distribution topology, perform edge collaborative fault discrimination based on the disturbance distribution operation data record, identify the fault section and perform protection actions if an edge collaborative fault has occurred, generate operation and maintenance feedback data, combine the operation and maintenance feedback data to maintain a dynamically evolving interval probability map, and execute a dual-layer monitoring mechanism including a trend analysis layer and a real-time analysis layer. The trend analysis layer extracts key parameters of waveform features and uses a time-series anomaly detection algorithm to perform long-term drift trend analysis to determine whether to initiate preventive topology reconfiguration. If initiated, generate a reconfiguration record. The real-time analysis layer determines whether to initiate emergency takeover. If initiated, generate a takeover record. After completion, organize the entire process record data.

[0025] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for intelligent dispatching and coordinated operation of medium and low voltage DC power distribution systems, characterized in that: Includes the following steps: For DC distribution trunk lines of narrow infrastructure, deploy intelligent terminals at the beginning of each T-junction branch, divide the intelligent terminal groups, and collect branch current waveforms according to typical operating conditions. During the operation period, feature extraction is performed on the branch current waveform to obtain power distribution operation data, a baseline feature template is constructed, and it is determined whether a disturbance occurs. If a disturbance occurs, the disturbance power distribution operation data record is compiled. Within the intelligent terminal group, edge autonomous regions are divided to form a power distribution topology. Edge collaborative fault identification is performed based on the recorded data of the power distribution operation that causes disturbances. If an edge collaborative fault occurs, the faulty section is identified and protection actions are performed to generate operation and maintenance feedback data. By combining operation and maintenance feedback data, a dynamically evolving interval probability map is maintained, and a two-layer monitoring mechanism including a trend analysis layer and a real-time analysis layer is implemented. The trend analysis layer extracts key parameters of waveform features and uses a time-series anomaly detection algorithm to perform long-term drift trend analysis to determine whether to initiate preventive topology reconfiguration. If initiated, a reconfiguration record is generated. The real-time analysis layer determines whether to initiate emergency takeover. If initiated, a takeover record is generated. After completion, the entire process record data is organized.

2. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 1, characterized in that: The method for acquiring the branch current waveform is as follows: The dispatch layer pre-defines all adjacent intelligent terminals in each fire compartment along the longitudinal direction of the DC power distribution trunk line as the same intelligent terminal group, using the physical boundary of the fire compartment as the dividing boundary. The dispatch layer issues data collection instructions to each group. The data collection instructions include typical operating conditions and the length of the data collection time window. Typical operating conditions include combinations of different seasonal indicators and load level indicators. When any typical operating condition is met simultaneously, the intelligent terminal records the branch current waveform at the beginning of the T-connection branch and adds the corresponding typical operating condition identifier.

3. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 1, characterized in that: The method for processing power distribution operation data is as follows: Feature extraction is performed on the branch current waveform. The extracted feature dimensions include current amplitude, current change rate, current second derivative, high-frequency energy ratio, and partial discharge pulse count. The corresponding feature components are extracted for each feature dimension, and all feature components are concatenated into a multi-dimensional feature vector as power distribution operation data. Among them, the high-frequency energy ratio is the ratio of the total energy of the frequency band above the high-frequency cutoff frequency to the total energy of the entire frequency band after the branch current waveform is transformed by fast Fourier transform. The partial discharge pulse count is the number of pulses in the branch current waveform whose current amplitude exceeds the preset partial discharge pulse detection threshold and whose pulse width is less than the preset upper limit of pulse width.

4. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 3, characterized in that: The method for constructing baseline feature templates and determining whether perturbations have occurred is as follows: The runtime phase includes the cold start initial build phase and the online runtime detection phase; In the initial construction phase of cold start, the power distribution operation data collected during the initialization period are grouped according to typical operating conditions. The mean of each feature component of the multidimensional feature vector under the same typical operating condition is used to construct the baseline feature template, and the baseline threshold of each feature component is calculated. During the online operation detection phase, the corresponding baseline feature template is retrieved based on the typical operating conditions of the power distribution operation data. The current amplitude and current change rate are compared with the corresponding baseline thresholds. If either exceeds the threshold, a disturbance is determined, and the data is marked as disturbed power distribution operation data and is not included in the updated sample set. Otherwise, the data is marked as steady-state power distribution operation data and is included in the updated sample set. The baseline feature template is updated on a rolling basis with the update cycle as the interval.

5. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 1, characterized in that: The method for dividing edge autonomous regions within a smart terminal group is as follows: The scheduling layer obtains the total number of smart terminals in the smart terminal group and divides them based on the preset upper limit and lower limit of the number of terminals; If the number of T-connection branches within a smart terminal group does not exceed the preset upper limit of the number of terminals, it is directly divided into an edge autonomous region. If the number of T-connection branches does not exceed the preset lower limit of the number of terminals, it is recorded as an isolated autonomous region. If the number of terminals exceeds the preset upper limit, the terminals are sorted by trunk line mileage and then divided into initial groups based on the preset lower limit of terminal number. If there are not enough terminals remaining, they are assigned to the last group. Starting from the first initial group, the terminals are merged with adjacent initial groups in turn. The merging condition is that the number of terminals after merging does not exceed the preset upper limit of terminal number and the distance spanned does not exceed the preset upper limit of area span. Otherwise, it is retained as an edge autonomous area.

6. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 1, characterized in that: The execution method for edge collaboration fault detection is as follows: Edge collaboration fault detection includes a single-terminal autonomous determination stage and a multi-terminal collaborative determination stage; In the single-terminal autonomous judgment stage, if only one smart terminal in the edge autonomous region has a disturbance distribution operation data record within the preset waiting time window, it enters the single-terminal judgment branch. The current change rate and high-frequency energy ratio of the disturbance distribution operation data record are compared with the baseline values ​​of the corresponding baseline feature template. If both exceed the preset fault judgment multiple, the fault type is judged as an internal fault of the T-connection branch. In the multi-terminal collaborative judgment stage, if two or more intelligent terminals have disturbance distribution operation data records, each intelligent terminal extracts multi-dimensional feature vectors from the disturbance distribution operation data records, and uses a dynamic time warping algorithm to calculate the similarity distance between each pair of multi-dimensional feature vectors. If any similarity distance exceeds the preset similarity threshold, the fault type is determined to be a main line section fault between adjacent terminals.

7. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 6, characterized in that: The similarity distance is calculated as follows: The two multidimensional feature vectors to be paired are normalized by the maximum and minimum values ​​to obtain normalized multidimensional feature vectors. The normalized multidimensional feature vectors are expanded into time series, and a cumulative distortion matrix is ​​constructed. The matrix elements are the Euclidean distances between the normalized multidimensional feature vectors at corresponding time points. By finding the regular path with the minimum cumulative distortion value, the cumulative distortion value at the end of the path is used as the similarity distance.

8. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 1, characterized in that: The interval probability graph is maintained as follows: The scheduling layer constructs an interval probability map for each edge autonomous region. Using the trunk line mileage station number as the horizontal axis, the DC distribution trunk line segment is discretized into several interval units according to the preset mileage step size. Each interval unit corresponds to a probability storage location with an initial value of zero. After receiving the operation and maintenance feedback data, the fault section information is extracted and mapped to the corresponding interval unit. The cumulative probability estimate of the probability storage location is updated using the exponential sliding window accumulation algorithm. The update method is to weight the current value with a preset probability decay factor and then sum it with the probability increment corresponding to the current edge collaboration fault. The probability increment takes the preset weight coefficient of the fault type. The preset weight coefficient corresponding to the trunk section fault is greater than the preset weight coefficient corresponding to the internal fault of the T-connection branch.

9. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 8, characterized in that: The method for determining whether to initiate preventative topology reconfiguration is as follows: When the cumulative probability estimate of any interval unit exceeds the preset preventive maintenance probability threshold, the trend analysis layer is triggered, and the high-frequency energy ratio of steady-state power distribution operation data within the trend analysis interval window is extracted as the partial discharge trend indicator and the current amplitude is extracted as the load trend indicator. A time-series anomaly detection algorithm is used to perform long-term drift trend analysis on the time series of partial discharge trend indicators and load trend indicators. If a comprehensive degradation risk is determined, the power distribution topology is combined to determine whether there is a backup power supply path that meets the capacity conditions. If so, a preventive topology reconfiguration is initiated, generating a topology reconfiguration instruction sequence arranged in the order of first closing and then disconnecting.

10. The intelligent dispatching and coordinated operation method for medium and low voltage DC power distribution systems according to claim 8, characterized in that: The method for determining whether to initiate emergency takeover is as follows: When the cumulative probability estimate of any interval unit exceeds the preset emergency takeover probability threshold, the real-time analysis layer is triggered. The preset emergency takeover probability threshold is higher than the preset preventive maintenance probability threshold. The scheduling layer extracts the protection action result of the most recent edge collaborative fault judgment of the interval unit. If the fault section has been isolated, it is recorded as an isolation event and the probability storage position is reset to zero. If it has not been isolated, emergency takeover is initiated, and an emergency disconnection command is directly issued to the intelligent terminal at the beginning of the corresponding T-connection branch to perform the disconnection operation and update the power distribution topology.

Images