Power distribution box remote monitoring method and system based on internet of things

Abstract

The present application relates to the technical field of power distribution monitoring of Internet of Things, and particularly relates to a power distribution box remote monitoring method and system based on Internet of Things. The method comprises: obtaining running state data through a sensor deployed in the power distribution box, transmitting the running state data to a remote server through an Internet of Things communication interface after time alignment and space mapping of the running state data, constructing a multi-dimensional correlation matrix in the remote server and determining the abnormal deviation relationship between the data in each dimension based on the matrix to obtain an abnormal state recognition result, classifying the abnormal state and generating corresponding control instructions and alarm information, issuing target control instructions to an executing mechanism to execute adjustment or protection actions, collecting feedback data and dynamically adjusting the multi-dimensional correlation matrix according to the feedback data and the abnormal recognition result. The present application can improve the real-time performance, accuracy and self-adaptive ability of power distribution box abnormal monitoring, and realize remote intelligent operation and maintenance management.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) power distribution monitoring technology, and in particular to a remote monitoring method and system for distribution boxes based on IoT. Background Technology

[0002] Distribution boxes, as core equipment for power distribution and protection, are widely deployed in industrial, commercial, and civil buildings. With the increasing automation of power systems, the demand for real-time monitoring of distribution box operating status is growing. Currently, the industry commonly adopts a monitoring architecture based on fieldbus or industrial Ethernet, installing various sensors and smart meters within the distribution boxes to collect electrical and non-electrical parameters such as voltage, current, temperature, and humidity. These data are then aggregated to a regional monitoring center or back-end management system via fixed communication links. This model achieves basic data visualization and simple over-limit alarm functions, providing maintenance personnel with a certain degree of operational reference.

[0003] Conventional methods typically rely on manually set static thresholds for anomaly detection, treating each monitored parameter independently. An alarm is triggered whenever a single value exceeds a preset limit. This logic is intuitive and easy to implement, and it can be effective in simple scenarios where fault characteristics are obvious and parameter coupling is weak. However, in actual power distribution system operation, there are complex relationships between different state variables. Local anomalies often manifest as coordinated deviations of multiple parameters, rather than isolated value exceeding limits. Static threshold methods struggle to capture such potential problems, easily leading to delayed or missed alarms, while also resulting in a high false alarm rate, increasing the burden on maintenance personnel to identify genuine faults.

[0004] Existing solutions generally lack unified alignment processing for multi-source data across both temporal and spatial dimensions. Inconsistent sensor sampling frequencies and random communication delays result in inconsistent data reaching the server along the timeline, directly impacting the accuracy of subsequent analysis and judgment. The lack of spatial alignment is even more pronounced; the system struggles to effectively map data from different electrically connected measurement points, making it impossible to assess the scope and propagation path of a fault from an overall topology perspective.

[0005] Furthermore, the monitoring and execution control processes are disconnected. When an anomaly is identified, the system typically only generates alarm information for manual handling, failing to automatically generate tiered control strategies for anomalies of varying severity. There is a lack of a closed-loop feedback mechanism between alarm information and actual equipment response, and the monitoring model cannot self-correct and optimize based on historical handling results during subsequent operation. Consequently, the adaptability and intelligence level of the monitoring system are significantly insufficient. Summary of the Invention

[0006] This invention provides a method and system for remote monitoring of distribution boxes based on the Internet of Things, which can solve the problems in the prior art.

[0007] A first aspect of the present invention provides a method for remote monitoring of distribution boxes based on the Internet of Things, comprising: The operating status data of the distribution box is obtained by sensors deployed inside the distribution box; The operation status data is time-aligned and spatially mapped to obtain spatiotemporally synchronized operation status data, which is then transmitted to a remote server via an IoT communication interface. A multidimensional correlation matrix is ​​constructed in the remote server to identify the different dimensions of the spatiotemporally synchronized running status data. Based on the multidimensional correlation matrix, the abnormal deviation relationship between the data of each dimension is determined to obtain the abnormal state identification result. Based on the abnormal state identification results, the abnormal state is divided into different levels, and corresponding control instructions and alarm information are generated for each level to obtain the target control instructions. The target control command is sent to the actuator of the distribution box, which performs the corresponding adjustment or protection action and collects the status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted.

[0008] The operational status data is time-aligned and spatially mapped to obtain spatiotemporally synchronized operational status data. This spatiotemporally synchronized operational status data is then transmitted to a remote server via an IoT communication interface, including: Based on the timestamp information of each data in the operational status data, a time series deviation matrix is ​​constructed, which represents the time offset between data collected by different sensors. Based on the time series deviation matrix, the operation status data is reconstructed in time, and each data is projected onto a unified time reference plane to obtain time-aligned operation status data. Extract sensor identification information from each data in the operating status data, obtain the three-dimensional spatial position coordinate information of each sensor in the power distribution box based on the sensor identification information, calculate the spatial distance relationship between each sensor, and construct a spatial topology diagram; The data in the time-aligned running status data are mapped and associated with the corresponding nodes in the spatial topology graph to obtain the spatiotemporally synchronized running status data. The spatiotemporally synchronized operating status data is subjected to data integrity verification. When the verification passes, the spatiotemporally synchronized operating status data is transmitted to a remote server through the Internet of Things communication interface.

[0009] Extract sensor identification information from the operational status data, obtain the three-dimensional spatial coordinates of each sensor within the distribution box based on the sensor identification information, calculate the spatial distance relationships between the sensors, and construct a spatial topology diagram, including: Extract sensor identification information from each data point in the operating status data, and query the three-dimensional spatial coordinate information of each sensor within the distribution box from the distribution box structure database based on the sensor identification information; Based on the three-dimensional spatial position coordinate information of each sensor, the spatial distance value between each sensor is determined, and sensor pairs whose spatial distance values ​​meet the preset proximity conditions are identified as spatial adjacency relationships, thus obtaining a set of spatial adjacency relationships; Based on the set of spatial adjacency relationships, a spatial topology graph is constructed with sensors as nodes and spatial adjacency relationships as edges. In the spatial topology graph, the three-dimensional spatial position coordinate information of the corresponding sensor is determined for each node, and the spatial distance weight value between the corresponding sensor pairs is determined for each edge. Perform connectivity detection on the spatial topology graph to identify isolated nodes in the spatial topology graph; For the identified isolated nodes, the spatial distance between the isolated nodes and other nodes is recalculated, and a spatial adjacency relationship is established between the isolated nodes and the nodes with the smallest spatial distance values ​​to obtain a connected spatial topology graph.

[0010] A multidimensional correlation matrix is ​​constructed in the remote server to identify the different dimensions of the spatiotemporally synchronized operational status data. Based on the multidimensional correlation matrix, the abnormal deviation relationships between the data in each dimension are determined to obtain the abnormal state identification results, including: The spatiotemporally synchronized running status data is divided into multiple dimensions. Extract the statistical feature values ​​of each dimension of data within a preset time window, calculate the correlation coefficient between the statistical feature values ​​of any two dimensions of data, and construct an initial multidimensional association matrix based on the correlation coefficient. A time-series evolution analysis is performed on the initial multidimensional correlation matrix to extract the change trajectory of matrix elements in multiple consecutive time windows, and the time-series change rate of each matrix element is calculated. The correlation coefficients in the initial multidimensional correlation matrix are dynamically weighted and adjusted according to the time-series change rate to obtain the time-series weighted multidimensional correlation matrix. Extract the baseline correlation matrix between data of each dimension from historical normal operation data, and calculate the matrix deviation between the time-weighted multidimensional correlation matrix and the baseline correlation matrix; Identify the dimensional data pairs that have deviated abnormally in the matrix deviation, extract the dimensional identifier information and deviation degree information corresponding to the dimensional data pairs, and obtain the abnormal state identification result.

[0011] A time-series evolution analysis is performed on the initial multidimensional correlation matrix to extract the change trajectory of matrix elements within multiple consecutive time windows, and the time-series change rate of each matrix element is calculated, including: Calculate the initial multidimensional correlation matrix corresponding to each time window within multiple consecutive time windows to obtain multiple initial multidimensional correlation matrices corresponding to multiple consecutive time windows. Then, serialize and arrange these multiple initial multidimensional correlation matrices along the time dimension to obtain a time-series matrix sequence. Extract values ​​from matrix elements at the same position in the time-series matrix sequence in chronological order to determine the time-series value sequence corresponding to each matrix element position; Adaptive sliding window smoothing is applied to each time series numerical sequence, and the smoothing window length is dynamically adjusted according to the fluctuation amplitude of the time series numerical sequence to obtain the smoothed time series numerical sequence; Perform time-series difference operation on the smoothed time-series numerical sequence to calculate the difference between values ​​at adjacent time points in the smoothed time-series numerical sequence; Based on the difference between adjacent time points in the smoothed time series numerical sequence, the time series change rate of each time series numerical sequence is calculated, and the time series change rate characterizes the rate of change of matrix element values ​​in the time dimension.

[0012] Based on the abnormal state identification results, the abnormal states are divided into different levels, and corresponding control commands and alarm information are generated for each level to obtain the target control commands, including: The severity quantification value of the abnormal state is determined based on the deviation degree information in the abnormal state identification result, and the dimensional range of the abnormal state's impact is determined based on the dimensional identification information in the abnormal state identification result. An abnormal state comprehensive evaluation index is constructed based on the severity quantification value and the dimension range. The abnormal state is then divided into multiple different levels according to the abnormal state comprehensive evaluation index and the preset level classification rules. For each abnormal state level, the control strategy corresponding to the abnormal state level is matched from the control strategy library; The control instructions containing execution priorities are generated according to the control strategy, and alarm information is generated according to the abnormal state level. The control instructions and the alarm information are then associated and bound to obtain the target control instructions. Multiple target control instructions are sorted according to their execution priority, and then sent to the distribution box control unit in sequence according to their priority.

[0013] The target control command is sent to the actuator of the distribution box, which performs corresponding adjustment or protection actions and collects status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted, including: The target control command is sent to the actuator of the distribution box. The actuator locates the object to be adjusted according to the object identifier in the target control command, and performs the corresponding adjustment or protection action according to the action type in the target control command. After execution, collect status feedback data including the operating parameters of the executed object and the operating parameters of surrounding related devices; Extract the post-execution values ​​of each dimension of the status feedback data, and extract the dimension data pairs that deviate abnormally in the abnormal status identification results and their degree of deviation information; Based on the changes in the values ​​of the corresponding dimension data before and after execution, calculate the influence coefficient of the control command execution on the correlation between the data pairs of each dimension. Based on the deviation information, the correction weight is determined, and the correlation coefficients of corresponding dimension data pairs in the multidimensional correlation matrix are weighted and corrected based on the influence coefficient and the correction weight to obtain the dynamically adjusted multidimensional correlation matrix.

[0014] A second aspect of this invention provides an Internet of Things (IoT)-based remote monitoring system for distribution boxes, comprising: A sensing unit is used to acquire the operating status data of the distribution box through sensors deployed inside the distribution box; The synchronization unit is used to perform time alignment processing and spatial location mapping on the running status data to obtain time-space synchronized running status data, and transmit the time-space synchronized running status data to a remote server through the Internet of Things communication interface. The analysis unit is used to construct a multidimensional correlation matrix between different dimensions of the spatiotemporally synchronized running status data in the remote server, determine the abnormal deviation relationship between each dimension of data based on the multidimensional correlation matrix, and obtain the abnormal state identification result. The control unit is used to classify the abnormal state into different levels according to the abnormal state identification result, generate corresponding control instructions and alarm information for different levels, and obtain the target control instruction. The adjustment unit is used to send the target control command to the actuator of the distribution box, so that the actuator can perform the corresponding adjustment action or protection action and collect the status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted.

[0015] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0017] This invention can eliminate interference caused by differences in data acquisition timing and spatial inconsistencies, enabling accurate alignment of operational status data in both time and space dimensions. This improves the remote server's perception accuracy of the overall situation of the distribution box, provides a complete and consistent data foundation for subsequent anomaly identification, and allows subsequent analysis to be based on a unified spatiotemporal reference. It also reduces redundant data transmission and saves IoT communication resources.

[0018] This invention utilizes a multidimensional correlation matrix to quantify the interdependence and coupling relationships between data in various dimensions. Abnormal deviation relationships are no longer limited to a single threshold judgment, but are based on multivariate joint analysis, effectively identifying early weak anomalies and complex faults, reducing false alarm rate and false negative rate, and improving the reliability and sensitivity of abnormal state identification results.

[0019] This invention generates corresponding control commands and alarm information according to the anomaly level, ensuring that on-site handling measures are precisely matched to the severity of the anomaly, avoiding over-operation or insufficient response, shortening fault response time, and improving the stability and safety of the distribution box operation. Feedback data is collected after the actuator's action and used for dynamic matrix adjustment, enabling the correlation model to continuously optimize with changes in operating conditions, forming a closed-loop adaptive monitoring mechanism, and enhancing the system's long-term adaptability and robustness.

[0020] This invention enables intelligent, precise, and automated remote monitoring of distribution boxes, reducing the frequency of manual inspections and maintenance costs, improving the safety of the power supply system, ensuring stable power supply for production and daily life, enhancing equipment operating efficiency, and extending service life. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the IoT-based remote monitoring method for distribution boxes according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the process of determining the spatial topology diagram according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the correlation coefficient matrix table according to an embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram of the temporal evolution trajectory of matrix elements according to an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0025] Figure 1 This is a flowchart illustrating the IoT-based remote monitoring method for distribution boxes according to an embodiment of the present invention. Figure 1 As shown, the method includes: The operating status data of the distribution box is obtained by sensors deployed inside the distribution box; The operation status data is time-aligned and spatially mapped to obtain spatiotemporally synchronized operation status data, which is then transmitted to a remote server via an IoT communication interface. A multidimensional correlation matrix is ​​constructed in the remote server to identify the different dimensions of the spatiotemporally synchronized running status data. Based on the multidimensional correlation matrix, the abnormal deviation relationship between the data of each dimension is determined to obtain the abnormal state identification result. Based on the abnormal state identification results, the abnormal state is divided into different levels, and corresponding control instructions and alarm information are generated for each level to obtain the target control instructions. The target control command is sent to the actuator of the distribution box, which performs the corresponding adjustment or protection action and collects the status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted.

[0026] The operational status data is time-aligned and spatially mapped to obtain spatiotemporally synchronized operational status data. This spatiotemporally synchronized operational status data is then transmitted to a remote server via an IoT communication interface, including: Based on the timestamp information of each data in the operational status data, a time series deviation matrix is ​​constructed, which represents the time offset between data collected by different sensors. Based on the time series deviation matrix, the operation status data is reconstructed in time, and each data is projected onto a unified time reference plane to obtain time-aligned operation status data. Extract sensor identification information from each data in the operating status data, obtain the three-dimensional spatial position coordinate information of each sensor in the power distribution box based on the sensor identification information, calculate the spatial distance relationship between each sensor, and construct a spatial topology diagram; The data in the time-aligned running status data are mapped and associated with the corresponding nodes in the spatial topology graph to obtain the spatiotemporally synchronized running status data. The spatiotemporally synchronized operating status data is subjected to data integrity verification. When the verification passes, the spatiotemporally synchronized operating status data is transmitted to a remote server through the Internet of Things communication interface.

[0027] The distribution box contains various types of sensors, including current sensors, voltage sensors, temperature sensors, and leakage current sensors. The sampling frequencies of these sensors differ, causing varying degrees of time-series offset in the operational status data collected by different sensors. To eliminate this time inconsistency, it is necessary to perform time alignment processing on the multi-channel sensor data based on the timestamp information carried by each data point.

[0028] A time series deviation matrix is ​​constructed based on the timestamp information of data from each sensor. , where matrix elements Indicates the first Road sensor and the first The time offset of the road sensors within the same acquisition period, in milliseconds. The time offset is calculated as follows: using the timestamp sequence corresponding to the sensor with the highest sampling frequency as the reference time axis, the timestamp sequences of other sensors are compared point-by-point with the reference time axis, and the difference between each corresponding moment is taken as the offset and filled into the corresponding position in the matrix. When it is positive, it indicates the first... The data acquisition time of the road sensor lags behind the first Road sensor; when When it is negative, it indicates that the first... The road sensor acquisition time is ahead of the first Circuit sensors. The time series deviation matrix can comprehensively depict the time synchronization error distribution among various sensors within the distribution box.

[0029] Based on the constructed time series deviation matrix, the operational status data is reconstructed in time. The core of time reconstruction is to project the discrete data sequences collected by each sensor onto a unified time reference plane using interpolation methods. Specifically, for sensor data with time offsets, linear interpolation or cubic spline interpolation methods are used to estimate the corresponding values ​​of the sensor at each sampling time on the reference time axis, thus ensuring that the data from all sensors are strictly aligned in the time dimension. During the interpolation process, if a sensor has a reference time point between two adjacent actual sampling times, the measurement values ​​of the sensor at these two actual sampling times are used for interpolation calculation to obtain the estimated value at the reference time point. After time reconstruction, the data from each sensor are arranged with a unified time step, forming time-aligned operational status data, eliminating the timing errors introduced by asynchronous sensor sampling.

[0030] After time alignment, the spatial location information of each sensor is further processed to achieve precise mapping between the operating status data and the physical space within the distribution box. Sensor identification information is extracted from each record in the operating status data. This identification information was bound to the sensor's three-dimensional spatial coordinates within the distribution box and stored in a local configuration file during the sensor installation and deployment phase. Based on the sensor identification information, the corresponding three-dimensional spatial coordinates of each sensor are retrieved from the configuration file. The coordinate system is established with the lower left corner of the distribution box as the origin, along the length, width, and height of the box. axis, shaft and The axes represent the spatial positions of each sensor in coordinates. It indicates that the subscript Number the sensor.

[0031] After obtaining the three-dimensional spatial coordinates of each sensor, the spatial distance relationship between any two sensors is calculated. (Sensor) With sensors Euclidean distance between The calculation method is as follows ,in , , Sensors The three-dimensional coordinate components, , , Sensors The three-dimensional coordinate components are determined. Based on the spatial distance relationships between the sensors, a spatial topology graph is constructed, where each node corresponds to one sensor, and the edge weights between nodes are taken from the corresponding spatial distance values. The spatial topology diagram not only records the physical distances between sensors but also reflects their relative layout within the distribution box through adjacency relationships, providing a spatial basis for subsequent multidimensional data correlation analysis. For example, current sensors and temperature sensors installed near the same busbar are closer in the spatial topology diagram, indicating a stronger physical correlation between their collected data, and should be assigned a higher correlation weight during anomaly detection.

[0032] Each record in the time-aligned operational status data is mapped and associated with its corresponding node in the spatial topology graph, giving each time-series data not only a temporal dimension but also a clear spatial location attribute. The mapping and association method involves using sensor identification information as the index key to bind the time-aligned data sequence with the node corresponding to the same sensor in the spatial topology graph, forming a two-dimensional "time-space" labeled data structure. After this mapping and association process, each operational status data record carries a precise timestamp and three-dimensional spatial coordinate information, resulting in spatiotemporally synchronized operational status data. This spatiotemporally synchronized data structure provides a unified and complete data foundation for subsequent multidimensional correlation analysis on a remote server.

[0033] Before uploading the spatiotemporally synchronized operational status data to the remote server via the IoT communication interface, data integrity verification is required to ensure the reliability of the transmitted data. Data integrity verification includes two levels: first, data format verification, checking whether the data type, value range, and required fields of each field conform to predefined data protocol specifications, such as voltage values ​​not being negative and timestamp formats conforming to ISO 8601 standards; second, data consistency verification, checking whether the time series of each sensor in the spatiotemporally synchronized data is continuous, whether there are any missing data or abnormal jumps, and verifying the integrity of the data packets using a cyclic redundancy check (CRC) code to ensure that the data has not been lost or tampered with during local processing. When both levels of verification pass, the data integrity verification is considered successful, and the spatiotemporally synchronized operational status data is then uploaded to the remote server via the IoT communication interface. If the verification fails, a local re-acquisition or data completion process is triggered, and verification is repeated after completion until it passes, thus ensuring that the data quality received by the remote server meets the requirements for subsequent analysis and processing.

[0034] The transmission protocol for the IoT communication interface can be flexibly selected based on on-site network conditions. In scenarios with sufficient network bandwidth, the MQTT protocol is prioritized for real-time push, while in scenarios with limited network conditions, the NB-IoT narrowband IoT protocol can be switched for low-power transmission. Data packets are encrypted during transmission, and a TLS encrypted channel is used to ensure data transmission security, preventing interception or tampering of operational status data during transmission. Upon receiving the data packet, the remote server stores it in the corresponding time-series database based on the sensor identification information and timestamp information within the packet, providing a continuously updated data source for subsequent construction of multi-dimensional correlation matrices and anomaly state identification.

[0035] Figure 2 This is a schematic diagram illustrating the process of determining the spatial topology graph according to an embodiment of the present invention. Figure 2 As shown, sensor identification information of each data point in the operating status data is extracted. Based on this sensor identification information, the three-dimensional spatial coordinates of each sensor within the distribution box are obtained. The spatial distance relationships between the sensors are calculated, and a spatial topology diagram is constructed, including: Extract the sensor identification information of each data in the operation status data, and query the three-dimensional spatial coordinate information of each sensor in the distribution box from the distribution box structure database based on the sensor identification information; Based on the three-dimensional spatial position coordinate information of each sensor, the spatial distance value between each sensor is determined, and sensor pairs whose spatial distance values ​​meet the preset proximity conditions are identified as spatial adjacency relationships, thus obtaining a set of spatial adjacency relationships; Based on the set of spatial adjacency relationships, a spatial topology graph is constructed with sensors as nodes and spatial adjacency relationships as edges. In the spatial topology graph, the three-dimensional spatial position coordinate information of the corresponding sensor is determined for each node, and the spatial distance weight value between the corresponding sensor pairs is determined for each edge. Perform connectivity detection on the spatial topology graph to identify isolated nodes in the spatial topology graph; For the identified isolated nodes, the spatial distance between the isolated nodes and other nodes is recalculated, and a spatial adjacency relationship is established between the isolated nodes and the nodes with the smallest spatial distance values ​​to obtain a connected spatial topology graph.

[0036] After acquiring the operating status data of the distribution box, each data record carries the identification information of the corresponding sensor. This identification information is usually embedded in the data frame header in the form of a unique code, used to mark which specific sensor inside the distribution box the data originated from. After extracting the sensor identification information of each data record, a query request is initiated to the distribution box structure database using this identification information as an index to obtain the three-dimensional spatial coordinate information of each sensor inside the distribution box. The distribution box structure database pre-stores the internal structural model of the distribution box, including the installation position of each sensor, and records the three-dimensional coordinate values ​​of each sensor in the local coordinate system of the distribution box with millimeter-level accuracy. This query process establishes a clear correspondence between the logical sensor identifiers in the operating status data and their physical spatial locations, providing basic data support for subsequent spatial topology modeling.

[0037] After obtaining the three-dimensional spatial coordinates of each sensor, it is necessary to calculate the spatial distance between any two sensors. For the sensor numbered in the distribution box... The sensor and the number The sensor, whose three-dimensional spatial coordinates are denoted as follows: and Euclidean distance between the two Calculate using the following formula: ; in , , Sensors The three coordinate components in the local coordinate system of the distribution box , , Sensors The three coordinate components in the local coordinate system of the distribution box Indicates sensor With sensors The Euclidean spatial distance between them. All sensors in the distribution box are paired up, and the spatial distance between each pair of sensors is calculated sequentially to form a complete distance matrix.

[0038] After obtaining the spatial distance values ​​of each sensor pair, the sensor pairs are filtered according to preset proximity conditions, which are usually based on a distance threshold. Given the form, this threshold is determined based on the actual size and internal layout characteristics of the distribution box, generally taken as a certain proportion of the maximum diagonal length of the effective space inside the distribution box. When the spatial distance value between a pair of sensors... satisfy When a sensor pair is identified as spatially adjacent, it is included in the spatial adjacency set. The spatial adjacency set records all sensor pairs within the distribution box that meet the proximity condition and their corresponding spatial distance values. It reflects the degree of physical proximity between the sensors and serves as the core basis for subsequently constructing the spatial topology diagram.

[0039] Based on the set of spatial adjacency relationships, a spatial topology graph is constructed, with each sensor as a node and spatial adjacency relationships as edges. The node attributes store the 3D spatial coordinates of the corresponding sensor for each node. This allows nodes to represent not only the logical identifier of a sensor but also its physical location information. In the graph's edge properties, each edge is assigned a spatial distance value between the corresponding sensor pairs. As a weight, the spatial distance weight value is used. The smaller the spatial distance weight value, the closer the two sensors are in physical space, and the stronger the spatial correlation between their collected data is usually. This weighted spatial topology diagram can fully express the physical distribution characteristics and mutual proximity relationships of each sensor in the distribution box, providing spatial structure information for subsequent multidimensional correlation analysis.

[0040] After constructing the spatial topology graph, connectivity testing is required to identify any isolated nodes. An isolated node is a sensor node that has no edge connections with any other node in the current spatial adjacency set. This is typically caused by the spatial distance between the sensor and all other sensors exceeding a preset proximity threshold. This results in the node being isolated in the spatial topology graph. Connectivity detection can be implemented using graph traversal algorithms, such as performing a depth-first or breadth-first traversal starting from any node. Nodes not visited after the traversal are considered isolated nodes. Once isolated nodes are identified, they are recorded in an isolated node set for subsequent repair processing.

[0041] For the identified isolated nodes, it is necessary to recalculate the spatial distance values ​​between the isolated node and all other nodes in the spatial topology graph. Other nodes Based on their respective three-dimensional spatial coordinates, calculate the Euclidean spatial distance between them. ,in , , isolated node The three-dimensional coordinate components, , , For nodes The three-dimensional coordinate components. Among all the calculated distance values, find the one that makes... The node that obtains the minimum value That is, with isolated nodes The node with the closest spatial distance. Isolated nodes. With nodes Establish spatial adjacency between them, and As the spatial distance weight of this edge, it is added to the spatial topology graph. Through this forced edge connection operation, isolated nodes are incorporated into the connected structure of the graph, ensuring that each sensor node is connected to at least one other sensor node in the graph by an edge.

[0042] After performing the forced edge connection operation on all identified isolated nodes, the connectivity of the spatial topology graph is re-verified to confirm that no more isolated nodes exist, resulting in a connected spatial topology graph. This connected graph ensures that all sensors within the distribution box are included in a unified spatial topology structure, with reachable paths between any two sensor nodes. This provides complete spatial structural constraints for constructing a multidimensional association matrix on a remote server. During the multidimensional association matrix construction phase, the spatial distance weights of each edge in the spatial topology graph serve as prior constraints for spatial association strength, allowing for a more reasonable modeling of the association relationships between physically adjacent sensor data, effectively improving the accuracy and reliability of abnormal state identification.

[0043] A multidimensional correlation matrix is ​​constructed in the remote server to identify the different dimensions of the spatiotemporally synchronized operational status data. Based on the multidimensional correlation matrix, the abnormal deviation relationships between the data in each dimension are determined to obtain the abnormal state identification results, including: The spatiotemporally synchronized running status data is divided into multiple dimensions. Extract the statistical feature values ​​of each dimension of data within a preset time window, calculate the correlation coefficient between the statistical feature values ​​of any two dimensions of data, and construct an initial multidimensional association matrix based on the correlation coefficient. A time-series evolution analysis is performed on the initial multidimensional correlation matrix to extract the change trajectory of matrix elements in multiple consecutive time windows, and the time-series change rate of each matrix element is calculated. The correlation coefficients in the initial multidimensional correlation matrix are dynamically weighted and adjusted according to the time-series change rate to obtain the time-series weighted multidimensional correlation matrix. Extract the baseline correlation matrix between data of each dimension from historical normal operation data, and calculate the matrix deviation between the time-weighted multidimensional correlation matrix and the baseline correlation matrix; Identify the dimensional data pairs that have deviated abnormally in the matrix deviation, extract the dimensional identifier information and deviation degree information corresponding to the dimensional data pairs, and obtain the abnormal state identification result.

[0044] When partitioning the spatiotemporally synchronized operational status data, the data is divided into several independent dimensions based on the types of physical quantities collected during the actual operation of the distribution box. These dimensions include voltage, current, temperature, power factor, and harmonic distortion. Each dimension corresponds to a time series of a type of physical measurement within the distribution box. The granularity of the dimension partitioning is determined based on the sensor deployment density and business analysis requirements. After the dimension partitioning is completed, the data from each dimension form a set of data that are independent of each other but coupled in terms of physical mechanisms, providing the basic input for the subsequent construction of the correlation matrix.

[0045] For each dimension of data, statistical feature values ​​are extracted within a preset time window. These feature values ​​typically include quantities describing the data distribution characteristics, such as mean, standard deviation, range, kurtosis, and skewness. These feature values ​​reflect the overall behavioral pattern of the data in that dimension within the current time window. The length of the preset time window is set based on a combination of the distribution box's operating cycle and the timeliness requirements for anomaly response. A time window that is too short will lead to large fluctuations in statistical feature values, while a time window that is too long will mask short-term anomalies. After obtaining the statistical feature values ​​for each dimension, the correlation coefficient between any two dimensions of data is calculated. Let the dimensions be... With dimension The statistical feature vectors within the current time window are respectively and The correlation coefficient between the two Calculated using the Pearson correlation coefficient method, i.e. ,in The number of sampling points within the time window. and Dimensions With dimension In the Feature components of each sampling point and These are the corresponding means. The correlation coefficients between all dimension pairs are also given. Fill in the corresponding positions in the matrix to construct the initial multidimensional correlation matrix. The rows and columns of the matrix correspond to the numbers of each dimension, and the matrix elements... That is, dimension With dimension The correlation coefficient between them.

[0046] Figure 3 This is a schematic diagram of the correlation coefficient matrix table according to an embodiment of the present invention. Figure 3As shown, the Pearson correlation coefficients between each pair of the five dimensions (voltage, current, temperature, power factor, and harmonic distortion) are listed in numerical form, including the following fields: dimension pair (e.g., "D2_current × D3_temperature", indicating the two types of physical quantities involved in the calculation), correlation coefficient r (i.e., the above). This represents the Pearson correlation coefficient calculated within the latest time window, with the value highlighted: strong correlation (|r|>0.6) is marked in red, and moderate correlation (|r|>0.35) is marked in orange. The correlation strength (the absolute value of the correlation coefficient is divided into three levels: "strong," "moderate," and "weak" for quick assessment of the degree of correlation), direction (indicated as positive or negative correlation, reflecting the same or opposite direction of the physical quantity change), and physical meaning (in conjunction with the operating mechanism of the distribution box, providing an engineering explanation of the coupling relationship for this dimension, such as current-temperature being labeled as "Joule heating effect (key)" to help business personnel understand the physical causes behind the anomalies).

[0047] For the initial multidimensional correlation matrix When performing time series evolution analysis, it is necessary to calculate the corresponding correlation matrix in multiple consecutive time windows to form a matrix time series. ,in This represents the total number of time windows participating in the temporal evolution analysis. For any element position in the matrix... Extract its in The trajectory of change within a time window And calculate the time-series rate of change of the element. The rate of change over time reflects the speed at which the correlation evolves in this dimension, and can be obtained by linear fitting or differencing the trajectory of change. Specifically, It can be defined as the mean of the differences between matrix elements between adjacent time windows, i.e. ,in This is an index for the time window sequence. Dimension pairs with larger absolute values ​​of time-series change rates indicate that their correlation has shifted significantly in the recent period and should be given higher weight in subsequent weighting.

[0048] Figure 4 This is a schematic diagram of the temporal evolution trajectory of matrix elements according to an embodiment of the present invention. Figure 4As shown, this diagram illustrates the trajectory of the correlation coefficients of key dimension pairs in the initial multidimensional correlation matrix over 10 consecutive time windows (W1 to W10). Each curve represents the evolution path of the correlation coefficient between a pair of physical quantities as the time window progresses. The horizontal axis represents the time window number (W1~W10), with a window length typically of 60 seconds. The vertical axis represents the Pearson correlation coefficient, ranging from -1 to 1. Taking current-temperature (red curve) as an example, W1~W7 are under normal operating conditions, with the correlation coefficient remaining stable at around 0.6; starting from W8, the correlation suddenly spikes to around 0.9, indicating a significant shift in the correlation between this dimension pair, corresponding to abnormal heat dissipation or an early sign of overload. Each curve is also marked with its corresponding historical baseline correlation coefficient using dashed lines for a visual comparison of the deviation between the current window and the normal pattern.

[0049] Based on the rate of change of time The correlation coefficients in the initial multidimensional correlation matrix are dynamically weighted and adjusted to obtain the time-weighted multidimensional correlation matrix. The core idea of ​​weighted adjustment is: for dimension pairs with more drastic changes in correlation, their current correlation coefficient should have a larger weight in the matrix, so that abnormal behavior of that dimension pair can be more sensitively captured in subsequent bias calculations. Weighting coefficients According to The normalized absolute value is determined, and the time-weighted matrix elements are... ,in The correlation coefficient within the latest time window. The dynamic weights are derived from the time-series rate of change. In this way, the time-weighted multidimensional correlation matrix... It retains the correlation information between the dimensions at the current moment and incorporates the influence of the recent correlation evolution trend, thereby enhancing the matrix's ability to represent abnormal states.

[0050] Extract the baseline correlation matrix between data from each dimension from historical normal operation data. At that time, historical data segments of the distribution box operating under known normal conditions were selected. Following the same method as constructing the initial multidimensional correlation matrix, the correlation coefficients between each dimension pair were calculated and filled into the baseline matrix. Baseline Correlation Matrix This represents the typical correlation patterns between various dimensions of data in a distribution box under normal conditions, serving as a reference benchmark for subsequent anomaly deviation judgment. The time-weighted multidimensional correlation matrix is ​​obtained after this process. Correlation matrix with benchmark Then, the matrix deviation between the two is calculated. Each element Dimensions reflecting the current operating status With dimension The degree of deviation of the correlation between them relative to the normal baseline.

[0051] Identification matrix deviation When a dimension data pair exhibits abnormal deviation, a deviation threshold is set. ,when At that time, the determination dimension With dimension An abnormal deviation occurred. For all dimensional data pairs that meet the abnormal deviation criteria, extract their corresponding dimension identifier information (i.e., dimension number or physical quantity name) and deviation degree information (i.e.,...). The specific numerical values ​​and their degree of exceedance relative to the threshold. Deviation information can be further quantified as a deviation ratio. This provides a basis for subsequent anomaly level classification. The dimensional identifiers and deviation degree information of all anomaly deviation dimensional data pairs are summarized to form an anomaly state identification result. This result clearly indicates which dimensions have abnormal relationships and the severity of the anomaly, providing accurate input for the generation of subsequent control commands and alarm information. In actual distribution box operation scenarios, for example, when the correlation coefficient between the temperature and current dimensions deviates significantly, it indicates that a feeder is overloaded or has abnormal heat dissipation. Through the above matrix deviation identification mechanism, the physical quantity dimension combination where the anomaly is located can be accurately located in the early stage, thereby supporting subsequent refined handling decisions.

[0052] A time-series evolution analysis is performed on the initial multidimensional correlation matrix to extract the change trajectory of matrix elements within multiple consecutive time windows, and the time-series change rate of each matrix element is calculated, including: Calculate the initial multidimensional correlation matrix corresponding to each time window within multiple consecutive time windows to obtain multiple initial multidimensional correlation matrices corresponding to multiple consecutive time windows. Then, serialize and arrange these multiple initial multidimensional correlation matrices along the time dimension to obtain a time-series matrix sequence. Extract values ​​from matrix elements at the same position in the time-series matrix sequence in chronological order to determine the time-series value sequence corresponding to each matrix element position; Adaptive sliding window smoothing is applied to each time series numerical sequence, and the smoothing window length is dynamically adjusted according to the fluctuation amplitude of the time series numerical sequence to obtain the smoothed time series numerical sequence; Perform time-series difference operation on the smoothed time-series numerical sequence to calculate the difference between values ​​at adjacent time points in the smoothed time-series numerical sequence; Based on the difference between adjacent time points in the smoothed time series numerical sequence, the time series change rate of each time series numerical sequence is calculated, and the time series change rate characterizes the rate of change of matrix element values ​​in the time dimension.

[0053] When performing time-series evolution analysis on the initial multidimensional correlation matrix, it is necessary to independently calculate the initial multidimensional correlation matrix corresponding to each window within multiple consecutive time windows. Each time window covers a fixed length of historical sampling data. Statistical features are extracted from the multidimensional operational status data within the window, and the correlation between the data of each dimension within the window is calculated according to the construction method of the correlation matrix, thus obtaining an initial multidimensional correlation matrix corresponding to that time window. As time progresses, the time windows are slid sequentially, and the same calculation process is performed for each time window, ultimately obtaining multiple initial multidimensional correlation matrices corresponding to multiple consecutive time windows. These matrices are then serialized along the time dimension according to the chronological order of their respective time windows, forming a time-series matrix sequence. This sequence completely records the historical trajectory of the changes in the correlation between each dimension over time.

[0054] Perform element-wise time axis extraction on the time series matrix sequence, that is, extract the time axis for each fixed position in the matrix. The values ​​of this position in the corresponding matrices across all time windows are arranged chronologically to form a time-series numerical sequence. Let the total number of time windows be... The time window number is ( ), then position The corresponding time series numerical sequence is ,in Indicates the first Dimensions within a time window With dimension The numerical values ​​of the elements in the correlation matrix are extracted position by position. This position-by-position extraction method transforms the time evolution of the multidimensional correlation matrix into several independent time-series numerical sequences, providing structured input for subsequent smoothing and rate of change calculations.

[0055] Because short-term pulse interference or acquisition noise exists in the operational status data, directly performing differential operations on the original time-series numerical sequences can easily introduce spurious abrupt changes. Therefore, adaptive sliding window smoothing is required for each time-series numerical sequence. The core of adaptive sliding window smoothing lies in dynamically adjusting the smoothing window length based on the fluctuation amplitude of the time-series numerical sequence itself: when the sequence fluctuation amplitude is large, the smoothing window length is appropriately increased to suppress the interference of noise on the judgment of the change trend; when the sequence fluctuation amplitude is small and the numerical change is gradual, the smoothing window length is appropriately shortened to retain the true fine-grained change information. The fluctuation amplitude can be measured by using the standard deviation or range of the sequence within a local interval as an evaluation index, and mapped to the corresponding smoothing window length level according to the preset fluctuation amplitude classification threshold. Let the position... The local standard deviation of the corresponding time series numerical sequence is The smooth window length is ,but Follow The value increases with the value of , and the specific mapping relationship is determined by a pre-calibrated piecewise function. Smoothing is performed using a moving average method, applying the mean value at each time step. The value is taken from the first and last two digits. The mean of each value is used as the smoothed result, and the endpoint effect is handled by mirror filling at the boundaries, thus obtaining the smoothed time series numerical sequence. .

[0056] Perform time-series differencing on the smoothed time series to calculate the difference between values ​​at adjacent time points, for example, for position... Adjacent time points and The difference between them is defined as ,in Difference sequence This reflects the increase or decrease of elements in the correlation matrix between each adjacent time step; positive values ​​indicate a stronger correlation, while negative values ​​indicate a weaker correlation. Temporal differencing transforms the original absolute numerical sequence into an incremental sequence, helping to highlight the dynamic trends of the correlation and eliminating the impact of numerical baseline offset on the assessment of the rate of change.

[0057] Calculating the time series change rate of each time series numerical sequence based on the difference sequence. This indicator characterizes the average rate of change of matrix element values ​​over time. The calculation of the time-series rate of change comprehensively considers the overall distribution characteristics of the difference sequence: the mean absolute value of the difference sequence is used as the basic measure of the rate of change, while the sign consistency coefficient of the difference sequence is introduced to correct for directional stability. The sign consistency coefficient is defined as the proportion of differences with the same sign to the total number of differences. When the sign consistency is high, it indicates that the correlation continues to evolve in the same direction, and the time-series rate of change corresponds to a strong monotonic trend; when the sign consistency is low, it indicates that the correlation is in an oscillating state, and the time-series rate of change mainly reflects the intensity of the oscillation. The final time-series rate of change is obtained by weighting these two factors together. This is used for the derivation of subsequent dynamic weighting coefficients and the construction of the time-series weighted correlation matrix.

[0058] In practical engineering deployments, the correlations between multi-dimensional data such as current, voltage, temperature, and humidity collected by different sensors within the distribution box can slowly drift or abruptly change with factors such as load variations, ambient temperature fluctuations, and equipment aging. Through the aforementioned time-series evolution analysis process, the dynamic evolution trajectory of the correlations across various dimensions can be continuously tracked from historical data, providing a temporal reference benchmark for subsequent identification of abnormal deviations. Adaptive sliding window smoothing effectively filters out spurious fluctuations introduced by short-term load shocks or communication jitter, ensuring that the calculated time-series change rate truly reflects the long-term evolution of the distribution box's operating status, rather than the impact of occasional disturbances. The combination of time-series difference operations and change rate extraction allows elements in the correlation matrix that experience slow drift and those that experience abrupt changes to exhibit significant differences in their change rate values, thus providing a more refined quantitative basis for subsequent abnormal deviation judgments.

[0059] Based on the abnormal state identification results, the abnormal states are divided into different levels, and corresponding control commands and alarm information are generated for each level to obtain the target control commands, including: The severity quantification value of the abnormal state is determined based on the deviation degree information in the abnormal state identification result, and the dimensional range of the abnormal state's impact is determined based on the dimensional identification information in the abnormal state identification result. An abnormal state comprehensive evaluation index is constructed based on the severity quantification value and the dimension range. The abnormal state is then divided into multiple different levels according to the abnormal state comprehensive evaluation index and the preset level classification rules. For each abnormal state level, the control strategy corresponding to the abnormal state level is matched from the control strategy library; The control instructions containing execution priorities are generated according to the control strategy, and alarm information is generated according to the abnormal state level. The control instructions and the alarm information are then associated and bound to obtain the target control instructions. Multiple target control instructions are sorted according to their execution priority, and then sent to the distribution box control unit in sequence according to their priority.

[0060] After obtaining the anomaly identification results, two key types of information need to be extracted from them: first, deviation degree information, which is the quantitative degree of deviation of each dimension's data from the baseline correlation; and second, dimension identification information, which is the specific dimension number where the anomaly deviation occurred and its combination relationship. Deviation degree information directly reflects the "depth" of the anomaly, while dimension identification information characterizes the "breadth" of the anomaly. Converting these two types of information into severity quantification values ​​and dimension ranges, respectively, is the basis for subsequent level classification.

[0061] The severity quantification value is determined based on the deviation ratio of each anomalous dimension pair in the deviation information. For all dimension pairs identified as anomalous in the identification results, the largest deviation ratio is taken as the main contributor to the severity quantification value. Simultaneously, a weighted correction based on the number of anomalous dimension pairs is introduced, so that the severity quantification value increases accordingly when multiple dimensions are anomalous simultaneously. Let the severity quantification value be... The maximum deviation ratio is The number of abnormal dimensions is The weighted correction factor is ,but .in This is a preset normal number used to control the contribution weight of the dimension to the quantity and severity, and can be calibrated according to the actual size of the distribution box.

[0062] Determining the dimensional range relies on dimensional identification information. All dimension numbers marked as abnormal are extracted from the abnormal state identification results. The total number of independent dimensions involved is counted, and the functional categories to which these dimensions belong are analyzed, such as voltage, current, temperature, and power. If the abnormal dimensions are concentrated in a single functional category, the dimensional range is marked as a local range; if the abnormal dimensions span multiple functional categories, it is marked as a cross-domain range. This classification method helps in the accurate matching of subsequent control strategies, avoiding excessive or insufficient control commands due to inaccurate abnormal range judgment.

[0063] Based on severity quantification Based on the dimensional range, construct a comprehensive evaluation index for abnormal states. The comprehensive evaluation index integrates the severity quantification value with the dimensional range encoding value. Specifically, it maps the dimensional range to discrete range weights. Local range corresponds to a smaller Values, with larger cross-domain ranges. Value. The comprehensive evaluation index is expressed as... The product-like approach allows severity and scope of impact to jointly determine the final assessment result. The pre-defined level classification rules will... The value space of is divided into several intervals, each corresponding to a different anomaly level. For example, when Below the threshold When, it is classified as a minor abnormality; when Between and When it is between, it is classified as a moderately abnormal level; when Exceed At that point, it is classified as a severe abnormality. The above threshold... , All parameters are pre-calibrated based on historical operating data of the distribution box and expert experience. They can be written into the configuration file during the system initialization phase and can be dynamically modified online.

[0064] For each identified abnormal state level, a strategy matching process is performed from the control strategy library. This library is indexed in two dimensions: abnormal level and abnormal type, with each index entry corresponding to a set of predefined control strategies. During system deployment, the control strategy library is populated by maintenance personnel based on the specific configuration of the distribution box and protection procedures, covering overload protection strategies, over-temperature protection strategies, voltage over-limit protection strategies, and multi-dimensional linkage protection strategies. The matching process first retrieves the candidate strategy set using the abnormal level as the primary key, then performs a secondary filtering using the functional category of the abnormal dimension as an auxiliary condition, ultimately determining the optimal matching strategy. If multiple dimensions are simultaneously abnormal and each corresponds to a different strategy, they are merged according to a preset strategy priority rule to avoid conflicts between control commands.

[0065] Based on the matched control strategy, control instructions with execution priorities are generated. The structure of each control instruction includes an instruction identifier, target actuator number, execution action type, execution parameter value, and execution priority field. Execution priorities are directly mapped to the anomaly level: severe anomalies correspond to the highest priority, moderate anomalies to medium priority, and minor anomalies to lower priority. For multiple control instructions under the same anomaly level, the priority ranking needs to be further refined based on the physical dependencies of the actuators to ensure that actions with logical sequential relationships are executed in the correct order. For example, before executing a circuit breaker trip instruction, a load transfer instruction must be completed first to prevent the power outage area from expanding.

[0066] Alarm information generation and control command generation occur in parallel. Alarm information includes fields such as alarm timestamp, anomaly level identifier, a list of involved dimensions, a description of the deviation degree, and suggested handling measures. For minor anomalies, alarm information is pushed to the monitoring terminal of maintenance personnel as a notification, requiring no immediate response. For moderate anomalies, alarm information is pushed as a warning and triggers audible and visual alerts. For severe anomalies, alarm information is pushed as an emergency alarm, and relevant responsible persons are notified via SMS or voice call. Alarm information and control commands are associated and bound after generation. The binding relationship is maintained through a unified event identifier, allowing maintenance personnel to directly trace the execution status of the corresponding control commands when viewing alarm information, achieving closed-loop management of alarms and controls.

[0067] The set of associated and bound target control commands is globally sorted according to execution priority. When multiple target control commands from abnormal events exist simultaneously, priority sorting ensures that the most urgent control action is issued first. After sorting, the target control commands are issued to the distribution box control unit in descending order of priority. Upon receiving the command, the control unit parses the target actuator number and execution action type in the command and drives the corresponding actuator to complete the corresponding adjustment or protection action. For lower priority commands, if the abnormal state has been mitigated by higher priority commands during the waiting period, the necessity of the command is reassessed to avoid executing redundant or contradictory control actions, thereby improving the accuracy and safety of overall control.

[0068] The target control command is sent to the actuator of the distribution box, which performs corresponding adjustment or protection actions and collects status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted, including: The target control command is sent to the actuator of the distribution box. The actuator locates the object to be adjusted according to the object identifier in the target control command, and performs the corresponding adjustment or protection action according to the action type in the target control command. After execution, collect status feedback data including the operating parameters of the executed object and the operating parameters of surrounding related devices; Extract the post-execution values ​​of each dimension of the status feedback data, and extract the dimension data pairs that deviate abnormally in the abnormal status identification results and their degree of deviation information; Based on the changes in the values ​​of the corresponding dimension data before and after execution, calculate the influence coefficient of the control command execution on the correlation between the data pairs of each dimension. Based on the deviation information, the correction weight is determined, and the correlation coefficients of corresponding dimension data pairs in the multidimensional correlation matrix are weighted and corrected based on the influence coefficient and the correction weight to obtain the dynamically adjusted multidimensional correlation matrix.

[0069] The process of sending target control commands to the distribution box actuator first relies on the execution object identification field carried in the command. This identification field is composed of three parts: equipment number, circuit number, and function type. After receiving the target control command, the actuator parses the execution object identification to accurately locate the physical device to be adjusted, such as a circuit breaker, contactor, voltage regulator module, or cooling fan. The execution action type field further indicates whether the operation is a regulation action or a protection action. Regulation actions include voltage level switching, load balancing, and temperature control, while protection actions include overcurrent tripping, short-circuit isolation, and grounding protection. The actuator completes the precise action execution based on the above two types of information, avoiding misoperation caused by ambiguity in command parsing.

[0070] Upon completion of execution, status feedback data is immediately collected, including operating parameters of the executed object and surrounding related equipment. The operating parameters of the executed object cover directly measurable quantities such as current, voltage, temperature, and power factor after the action. The operating parameters of surrounding related equipment include the operating status quantities of other equipment electrically connected to or physically adjacent to the executed object at the same time. The data acquisition time window is set to the stable response interval after the execution action is completed to ensure that the collected values ​​have moved beyond the transient process and reflect the true steady-state operating state after execution. The collected status feedback data is categorized and organized by dimension, forming a multi-dimensional vector consistent with the operating status data structure before execution, facilitating subsequent dimensional numerical comparison and analysis.

[0071] In the extraction phase, the required information is extracted from both the state feedback data and the anomaly identification results. From the state feedback data, the steady-state values ​​of each dimension after execution are extracted and recorded as the post-execution values. From the anomaly identification results, the set of dimension data pairs exhibiting abnormal deviations, along with the deviation degree information for each pair, is extracted. The deviation degree information originates from the deviation ratio calculated in the previous steps, reflecting the magnitude of the deviation of that dimension pair from the baseline correlation during anomaly identification. These two types of information are aligned to a unified dimension numbering system, forming the input basis for subsequent influence coefficient calculations and weight correction determination.

[0072] The impact coefficient is calculated based on the amount and direction of change in the data values ​​of each dimension before and after execution. For a data pair of a certain dimension in the multidimensional correlation matrix, let the dimension before execution be... The value is Dimensions after execution The value is Dimensions before execution The value is Dimensions after execution The value is Then the dimension The numerical change is Dimension The numerical change is Influence coefficient Defined as the normalized ratio of the product of the numerical changes in the two dimensions to the product of their respective original numerical values, specifically expressed as: ; in To prevent extremely small positive numbers with a denominator of zero. The sign of the value reflects whether the control instruction strengthens or weakens the correlation of that dimension, while the absolute value reflects the magnitude of the impact. For dimensions whose values ​​are close to zero before execution, the normalized relative change is used to replace the original value to ensure the numerical stability of the influence coefficient calculation.

[0073] The determination of correction weights is based on the degree of deviation information. Dimension pairs with a greater degree of deviation indicate that the correlation between them had already shifted significantly before the anomaly occurred, and their correlation matrix elements require more substantial correction based on the feedback from this execution. Dimension pairs with a smaller degree of deviation only require minor adjustments. Let the dimension pairs... The corresponding deviation information is (This parameter is independent of the deviation ratio parameter in the previous step and is specifically used for calculating the feedback correction weight in this step.) Correction weight Through the Normalization mapping yields: ; in This represents the set of all dimension pairs identified as abnormal deviations in the anomaly identification results. Normalization ensures that the sum of the correction weights for all abnormal dimension pairs is 1, keeping the overall correction amount within a controllable range and preventing the overall correlation matrix from becoming unstable due to excessively high weights for a single dimension pair.

[0074] Based on the influence coefficient With corrected weights The correlation coefficients of corresponding dimension data pairs in the multidimensional correlation matrix are weighted and corrected. Let the dimensions in the current multidimensional correlation matrix be... With dimension The corresponding correlation coefficient is The step size control parameter is corrected as follows: The dynamically adjusted correlation coefficient Calculate using the following formula: ; Correct step size control parameters Adaptively set according to the level of abnormality: Severe abnormalities correspond to larger values. The value allows the correlation matrix to be quickly updated to a new steady state after feedback is performed; minor anomalies correspond to smaller values. This value preserves the smooth evolution of the correlation matrix, preventing frequent and large fluctuations due to occasional minor anomalies. For sets of pairs not appearing in the abnormal dimension... The correlation coefficients of the dimension pairs in the data remain unchanged during this round of dynamic adjustment, and only evolve naturally in the regular update process of subsequent time windows.

[0075] After weighting and correcting all outlier dimension pairs, the validity of the dynamically adjusted multidimensional association matrix is ​​validated. Validation includes: matrix symmetry verification (ensuring...). The system includes correlation coefficient range constraints (ensuring that the absolute value of all elements does not exceed 1) and matrix positive definiteness checks. If an element exceeds the legal range after correction, it is truncated to the boundary value, and elements at symmetrical positions are adjusted simultaneously. After the verification passes, the dynamically adjusted multidimensional correlation matrix is ​​stored as the new benchmark correlation matrix for use in the next round of abnormal deviation judgment, thereby realizing the continuous adaptive evolution of the correlation matrix according to the actual operating state of the distribution box.

[0076] A second aspect of this invention provides an Internet of Things (IoT)-based remote monitoring system for distribution boxes, comprising: A sensing unit is used to acquire the operating status data of the distribution box through sensors deployed inside the distribution box; The synchronization unit is used to perform time alignment processing and spatial location mapping on the running status data to obtain time-space synchronized running status data, and transmit the time-space synchronized running status data to a remote server through the Internet of Things communication interface. The analysis unit is used to construct a multidimensional correlation matrix between different dimensions of data in the spatiotemporally synchronized running status data in the remote server, determine the abnormal deviation relationship between each dimension of data based on the multidimensional correlation matrix, and obtain the abnormal state identification result. The control unit is used to classify the abnormal state into different levels according to the abnormal state identification result, generate corresponding control instructions and alarm information for different levels, and obtain the target control instruction. The adjustment unit is used to send the target control command to the actuator of the distribution box, so that the actuator can perform the corresponding adjustment action or protection action and collect the status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted.

[0077] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0078] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0079] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote monitoring method for distribution boxes based on the Internet of Things, characterized in that, include: The operating status data of the distribution box is obtained by sensors deployed inside the distribution box; The operation status data is time-aligned and spatially mapped to obtain spatiotemporally synchronized operation status data, which is then transmitted to a remote server via an IoT communication interface. A multidimensional correlation matrix is ​​constructed in the remote server to identify the different dimensions of the spatiotemporally synchronized running status data. Based on the multidimensional correlation matrix, the abnormal deviation relationship between the data of each dimension is determined to obtain the abnormal state identification result. Based on the abnormal state identification results, the abnormal state is divided into different levels, and corresponding control instructions and alarm information are generated for each level to obtain the target control instructions. The target control command is sent to the actuator of the distribution box, which performs the corresponding adjustment or protection action and collects the status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted.

2. The method according to claim 1, characterized in that, The operational status data is time-aligned and spatially mapped to obtain spatiotemporally synchronized operational status data. This spatiotemporally synchronized operational status data is then transmitted to a remote server via an IoT communication interface, including: Based on the timestamp information of each data in the operational status data, a time series deviation matrix is ​​constructed, which represents the time offset between data collected by different sensors. Based on the time series deviation matrix, the operation status data is reconstructed in time, and each data is projected onto a unified time reference plane to obtain time-aligned operation status data. Extract sensor identification information from each data in the operating status data, obtain the three-dimensional spatial position coordinate information of each sensor in the power distribution box based on the sensor identification information, calculate the spatial distance relationship between each sensor, and construct a spatial topology diagram; The data in the time-aligned running status data are mapped and associated with the corresponding nodes in the spatial topology graph to obtain the spatiotemporally synchronized running status data. The spatiotemporally synchronized operating status data is subjected to data integrity verification. When the verification passes, the spatiotemporally synchronized operating status data is transmitted to a remote server through the Internet of Things communication interface.

3. The method according to claim 2, characterized in that, Extract sensor identification information from the operational status data, obtain the three-dimensional spatial coordinates of each sensor within the distribution box based on the sensor identification information, calculate the spatial distance relationships between the sensors, and construct a spatial topology diagram, including: Extract the sensor identification information of each data in the operation status data, and query the three-dimensional spatial coordinate information of each sensor in the distribution box from the distribution box structure database based on the sensor identification information; Based on the three-dimensional spatial position coordinate information of each sensor, the spatial distance value between each sensor is determined, and sensor pairs whose spatial distance values ​​meet the preset proximity conditions are identified as spatial adjacency relationships, thus obtaining a set of spatial adjacency relationships; Based on the set of spatial adjacency relationships, a spatial topology graph is constructed with sensors as nodes and spatial adjacency relationships as edges. In the spatial topology graph, the three-dimensional spatial position coordinate information of the corresponding sensor is determined for each node, and the spatial distance weight value between the corresponding sensor pairs is determined for each edge. Perform connectivity detection on the spatial topology graph to identify isolated nodes in the spatial topology graph; For the identified isolated nodes, the spatial distance between the isolated nodes and other nodes is recalculated, and a spatial adjacency relationship is established between the isolated nodes and the nodes with the smallest spatial distance values ​​to obtain a connected spatial topology graph.

4. The method according to claim 1, characterized in that, A multidimensional correlation matrix is ​​constructed in the remote server to identify the different dimensions of the spatiotemporally synchronized operational status data. Based on the multidimensional correlation matrix, the abnormal deviation relationships between the data in each dimension are determined to obtain the abnormal state identification results, including: The spatiotemporally synchronized running status data is divided into multiple dimensions. Extract the statistical feature values ​​of each dimension of data within a preset time window, calculate the correlation coefficient between the statistical feature values ​​of any two dimensions of data, and construct an initial multidimensional association matrix based on the correlation coefficient. A time-series evolution analysis is performed on the initial multidimensional correlation matrix to extract the change trajectory of matrix elements in multiple consecutive time windows, and the time-series change rate of each matrix element is calculated. The correlation coefficients in the initial multidimensional correlation matrix are dynamically weighted and adjusted according to the time-series change rate to obtain the time-series weighted multidimensional correlation matrix. Extract the baseline correlation matrix between data of each dimension from historical normal operation data, and calculate the matrix deviation between the time-weighted multidimensional correlation matrix and the baseline correlation matrix; Identify the dimensional data pairs that have deviated abnormally in the matrix deviation, extract the dimensional identifier information and deviation degree information corresponding to the dimensional data pairs, and obtain the abnormal state identification result.

5. The method according to claim 4, characterized in that, A time-series evolution analysis is performed on the initial multidimensional correlation matrix to extract the change trajectory of matrix elements within multiple consecutive time windows, and the time-series change rate of each matrix element is calculated, including: Calculate the initial multidimensional correlation matrix corresponding to each time window within multiple consecutive time windows to obtain multiple initial multidimensional correlation matrices corresponding to multiple consecutive time windows. Then, serialize and arrange these multiple initial multidimensional correlation matrices along the time dimension to obtain a time-series matrix sequence. Extract values ​​from matrix elements at the same position in the time-series matrix sequence in chronological order to determine the time-series value sequence corresponding to each matrix element position; Adaptive sliding window smoothing is applied to each time series numerical sequence, and the smoothing window length is dynamically adjusted according to the fluctuation amplitude of the time series numerical sequence to obtain the smoothed time series numerical sequence; Perform time-series difference operation on the smoothed time-series numerical sequence to calculate the difference between values ​​at adjacent time points in the smoothed time-series numerical sequence; Based on the difference between adjacent time points in the smoothed time series numerical sequence, the time series change rate of each time series numerical sequence is calculated, and the time series change rate characterizes the rate of change of matrix element values ​​in the time dimension.

6. The method according to claim 1, characterized in that, Based on the abnormal state identification results, the abnormal states are divided into different levels, and corresponding control commands and alarm information are generated for each level to obtain the target control commands, including: The severity quantification value of the abnormal state is determined based on the deviation degree information in the abnormal state identification result, and the dimensional range of the abnormal state's impact is determined based on the dimensional identification information in the abnormal state identification result. An abnormal state comprehensive evaluation index is constructed based on the severity quantification value and the dimension range. The abnormal state is then divided into multiple different levels according to the abnormal state comprehensive evaluation index and the preset level classification rules. For each abnormal state level, the control strategy corresponding to the abnormal state level is matched from the control strategy library; The control instructions containing execution priorities are generated according to the control strategy, and alarm information is generated according to the abnormal state level. The control instructions and the alarm information are then associated and bound to obtain the target control instructions. Multiple target control instructions are sorted according to their execution priority, and then sent to the distribution box control unit in sequence according to their priority.

7. The method according to claim 1, characterized in that, The target control command is sent to the actuator of the distribution box, which performs corresponding adjustment or protection actions and collects status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted, including: The target control command is sent to the actuator of the distribution box. The actuator locates the object to be adjusted according to the object identifier in the target control command, and performs the corresponding adjustment or protection action according to the action type in the target control command. After execution, collect status feedback data including the operating parameters of the executed object and the operating parameters of surrounding related devices; Extract the post-execution values ​​of each dimension of the status feedback data, and extract the dimension data pairs that deviate abnormally in the abnormal status identification results and their degree of deviation information; Based on the changes in the values ​​of the corresponding dimension data before and after execution, calculate the influence coefficient of the control command execution on the correlation between the data pairs of each dimension. Based on the deviation information, the correction weight is determined, and the correlation coefficients of corresponding dimension data pairs in the multidimensional correlation matrix are weighted and corrected based on the influence coefficient and the correction weight to obtain the dynamically adjusted multidimensional correlation matrix.

8. A remote monitoring system for distribution boxes based on the Internet of Things, used to implement the method as described in any one of claims 1-7, characterized in that, include: A sensing unit is used to acquire the operating status data of the distribution box through sensors deployed inside the distribution box; The synchronization unit is used to perform time alignment processing and spatial location mapping on the running status data to obtain time-space synchronized running status data, and transmit the time-space synchronized running status data to a remote server through the Internet of Things communication interface. The analysis unit is used to construct a multidimensional correlation matrix between different dimensions of data in the spatiotemporally synchronized running status data in the remote server, determine the abnormal deviation relationship between each dimension of data based on the multidimensional correlation matrix, and obtain the abnormal state identification result. The control unit is used to classify the abnormal state into different levels according to the abnormal state identification result, generate corresponding control instructions and alarm information for different levels, and obtain the target control instruction. The adjustment unit is used to send the target control command to the actuator of the distribution box, so that the actuator can perform the corresponding adjustment action or protection action and collect the status feedback data after execution. Based on the status feedback data and the abnormal status identification result, the multidimensional correlation matrix is ​​dynamically adjusted.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images