A high-efficiency safe remote interaction transformer area intelligent fusion terminal and electric energy meter

Abstract

The application discloses a kind of high-efficiency safe remote interaction's transformer area intelligent fusion terminal and electric energy meter, it is related to power distribution system technical field, first acquisition each node real-time operation information, and combine node topological structure to build multidimensional time sequence characteristic model, determine the power-down detection trigger condition of each node, generate each trigger power-down detection node, reconstruct and verify power-down detection result to its state, simultaneously to the task priority of each trigger power-down detection node is determined, dynamically adjust message package sending strategy, then in the reporting process, the parameter of time sequence prediction model is adjusted in real time to optimize virtual prediction accuracy;Further automatically update message sending strategy, realize the continuous optimization of model prediction and scheduling strategy, can maintain the continuous perception and reliable reporting of node state under the condition of power-down or network fluctuation, dynamically optimize message scheduling, improve the reliability, response speed and data integrity of transformer area intelligent fusion terminal in high-efficiency remote interaction.

Description

Technical Field

[0001] This invention relates to the field of power distribution system technology, and in particular to a highly efficient and secure remote interactive smart converged terminal for distribution transformers and an energy meter. Background Technology

[0002] The development of smart distribution networks towards digitalization and self-healing is a practical necessity. With the widespread adoption of low-voltage distribution area monitoring and distributed power source integration, the dynamics and complexity of power grid operation have significantly increased. Traditional monitoring systems heavily rely on physical power supply and communication pathways; once a power outage occurs, the monitoring chain is interrupted, preventing power grid maintenance departments from immediately understanding the propagation characteristics and impact range of outage events. In an environment with the widespread integration of new energy sources, distributed energy storage, and intelligent dispatch, the continuity of real-time data directly affects the safe operation and recovery efficiency of the system. Therefore, to ensure "uninterrupted sensing during power outages and reconfigurable chain breaks," it is essential to achieve continuous sensing and remote interaction under power outage conditions through software-level virtual modeling, data fusion, intelligent prediction, and closed-loop correction. This will enable the smart distribution area fusion terminal to provide reliable data and intelligent decision support even during power outages.

[0003] For example, CN119726722B discloses an edge collaboration system for intelligent integrated terminals in a power distribution area. Its features include several intelligent integrated terminals and a cloud platform. Specifically, the intelligent integrated terminals introduce a data hierarchical processing mechanism, classifying sensor data into high and low priorities for real-time processing; and perform noise reduction processing on the sensor data based on a time-series anomaly detection noise reduction algorithm; establish a dynamic collaboration network among the intelligent integrated terminals in the distribution area according to the priority distribution of computational load, and adopt a hierarchical task offloading collaboration mechanism to dynamically allocate computational tasks to other intelligent integrated terminals; the cloud platform uses an LSTM model to predict distribution area load changes in real time based on the processed sensor data; the cloud platform introduces a distribution diagnosis model based on graph neural networks; and achieves fault self-healing through the dynamic collaboration network of the intelligent integrated terminals.

[0004] For example, announcement number CN112398222B discloses an AC data acquisition system for a distribution transformer intelligent fusion terminal, including an AC data acquisition board circuit, a main control board circuit, an AC data acquisition board debugging module, and a debugging software system; the AC data acquisition board circuit is connected to the AC data acquisition board debugging module, the AC data acquisition board debugging module is connected to the main control computer, and the main control computer is equipped with a debugging software system; the AC data acquisition board circuit is used to acquire the voltage and current signals of the low-voltage side of the 10kV distribution transformer and calculate the voltage, current, and power signals that meet the accuracy requirements of the fusion terminal.

[0005] The above-mentioned technology has at least the following technical problems:

[0006] When one or more low-voltage monitoring nodes within a distribution area lose power due to a power outage, the entire monitoring system can no longer collect and upload the electrical status data of that node in real time, creating a "blind spot" or "discontinuity" in status perception. This interruption not only affects the location and analysis of power outage events but may also lead to discontinuous recording of the power grid's operating status, thereby weakening the accuracy of remote decision-making and the timeliness of response. More seriously, if the system cannot detect changes in nodes during a power outage, it cannot determine whether the problem is a local line anomaly, equipment failure, or a problem with the upstream power supply, thus affecting the formulation of fault recovery strategies. Summary of the Invention

[0007] In order to solve the technical problems existing in the prior art, the present invention provides an efficient and secure remote interactive smart converged terminal for distribution transformers and an energy meter.

[0008] The first aspect of this invention provides a highly efficient and secure remote interactive intelligent converged terminal for distribution stations, the terminal comprising:

[0009] The virtual sampling generation module is used to collect the running information of each node, generate a multi-dimensional time series feature model based on the topology of each node, determine the triggering conditions for power failure detection of each node to obtain each triggered power failure detection node, reconstruct the state of each triggered power failure detection node, and verify the power failure detection.

[0010] The message priority queue module is used to determine the task priority of each triggered power failure detection node, and to obtain the reachability score of each triggered power failure detection node by combining the historical reporting success rate and the current link quality. The message packet sending strategy is dynamically adjusted according to the reachability score of each triggered power failure detection node, and the parameters of the time series prediction model are dynamically adjusted during the message reporting process. The closed-loop feedback correction module is used for the terminal to resample the real electrical data and compare it with the virtual prediction data, and automatically adjust and update the message sending strategy according to the comparison results.

[0011] Furthermore, the specific process for determining the triggering conditions for power failure detection at each node is as follows:

[0012] Each sampling period is preset, and the voltage signal of each node in each sampling period is collected in a sliding sampling window manner on the time series. The number of times the measured voltage value of each node in each sampling period is continuously lower than a set threshold is counted. The number of sampling points where the measured voltage value of each node in each sampling period is continuously lower than the voltage threshold in the database is also counted. If this number exceeds the allowed number threshold stored in the database, the window is marked as the first power-down window. Thus, each first power-down window is obtained. In each first power-down window of each node, the rate of change of voltage and the rate of change of voltage change of adjacent sampling points are subtracted from and coupled with the average rate of change of voltage and the average rate of change of voltage of adjacent sampling points in the preset historical steady state period, respectively, to obtain the voltage signal rate of change exceeding the limit value of each node. The duration of each first power-down window is counted. If the voltage signal rate of change exceeding the limit value of a node is higher than or equal to the voltage signal rate of change exceeding the limit value stored in the database, and the duration of the first power-down window is higher than the preset duration, then it is preliminarily determined that the node triggers power-down detection, and each triggered power-down detection node is obtained. Otherwise, it is recorded as not triggered.

[0013] Furthermore, the specific process of reconstructing the state of each triggered power-down detection node is as follows:

[0014] The system retrieves continuous sampled data sequences of each triggered power failure detection node before power failure, including voltage, current, power factor, and harmonic content. It also obtains the recording timestamp, topological node location, and power coupling coefficient of adjacent nodes. A set of multi-dimensional state vectors is generated and imported into the time series prediction model to output the power failure warning coefficient of each triggered power failure detection node. The system also retrieves the recovery curves of similar historical events to simulate and reconstruct the missing data of each triggered power failure detection node.

[0015] Furthermore, the verification process for the power failure detection is as follows:

[0016] The trend gradient in the virtual state is fitted and compared with the dynamic fluctuation characteristics of the window before the power outage, and the power outage warning coefficients of each triggered power outage detection node generated by the time series prediction model are obtained, including the power outage signal change rate and power outage signal acceleration. At the same time, the power outage signal change rate, power outage signal acceleration, and signal fluctuation variance in the sampling window before the power outage of each triggered power outage detection node are extracted and fused to obtain the power outage confidence factor of each triggered power outage detection node. The power outage confidence factor of each triggered power outage detection node reflects the dynamic trend and short-term fluctuation characteristics of the node before the power outage.

[0017] Furthermore, the specific process for determining the task priority of each triggered power failure detection node is as follows:

[0018] The scheduling priority of each triggered power failure detection node is obtained by matching the power failure confidence factor of each node with the scheduling priority corresponding to each interval of the power failure confidence factor of the node stored in the database. The task priority of each triggered power failure detection node is then determined based on the scheduling priority of each triggered power failure detection node.

[0019] Determining the task priority of each triggered power failure detection node also includes extracting nodes with high confidence scores whose power failure confidence factors are higher than the confidence factor threshold stored in the database. These nodes are designated as judgment nodes. If the variance of virtual signal fluctuation is higher than the upper limit of the historical steady-state segment stored in the database and the trend gradient continues to decrease, the power failure detection condition is confirmed, and the node is further marked as a power failure pending judgment node. At the same time, a power failure confirmation signal is output to trigger the subsequent virtual continuous sampling, message scheduling, and remote reporting process.

[0020] Furthermore, the specific process for obtaining the reachability score of each triggered power-down detection node is as follows:

[0021] The status information of each triggered power failure detection node is encapsulated into a complete power failure event data packet, and the historical reporting success rate and current link quality information of each triggered power failure detection node are obtained, including signal-to-noise ratio, received packet loss rate, duplicate packet rate, and current latency.

[0022] The reachability score of each triggered power outage detection node is obtained by combining the historical reporting success rate of each triggered power outage detection node with the current historical reporting success rate of each triggered power outage detection node and the current link quality information of the system through normalization. The reachability score of each triggered power outage detection node is used to evaluate the overall reliability of the triggered power outage detection node in sending messages to the cloud under the current network and historical performance conditions.

[0023] Furthermore, the specific process of dynamically adjusting the message packet sending strategy is as follows:

[0024] Extract the reachability score of each triggering power outage detection node and compare it with the reachability score threshold stored in the database. If the reachability score of a triggering power outage detection node is lower than or equal to the reachability score threshold, adjust the message packet sending strategy of that triggering power outage detection node; otherwise, there is no need to adjust the message packet sending strategy of that triggering power outage detection node.

[0025] Furthermore, the process of dynamically adjusting the parameters of the time series prediction model during message reporting is as follows:

[0026] During the message reporting process, the reporting status of virtual sampling data is monitored, including the number of times data packets of each triggered power failure detection node are frequently retransmitted and the acknowledgment delay. When it is found that the data packets of each triggered power failure detection node are frequently retransmitted or the acknowledgment delay exceeds the corresponding frequent retransmission or acknowledgment delay threshold stored in the preset database, the time series prediction model parameter dynamic adjustment program is triggered.

[0027] Furthermore, the method for resampling real electrical data at the terminal and comparing it with virtual predicted data, and automatically adjusting the update message sending strategy based on the comparison result, includes:

[0028] After being triggered, a resampling mechanism is initiated to continuously sample the key operating parameters of each triggered power outage detection node during the power outage period. The sampled data is then compared one by one with the virtual prediction data at the corresponding time point, and the message sending strategy is automatically adjusted and updated based on the comparison results.

[0029] The second aspect of this invention also provides a highly efficient and secure remote interactive smart integrated energy meter for distribution areas, including a data acquisition unit, which is responsible for real-time acquisition of operating parameters of each node, including voltage, current, power factor and harmonics, and generating a multi-dimensional state vector from the acquired data according to the sampling period;

[0030] The communication unit is used for remote data interaction with the host terminal, cloud platform and nearby smart meters, and supports dynamic scheduling of message queues and redundant channel transmission.

[0031] The processing unit integrates a time series prediction model and a virtual sampling module. It can virtually reconstruct the collected data, generate power outage warning coefficients and power outage confidence factors in the event of power failure or abnormality, and compare the prediction results with the real sampled data to achieve power outage detection verification and continuous sensing.

[0032] The storage unit is used to store historical sampling data, virtual prediction data, trend model parameters, and historical event databases, providing basic data for reconstruction, confidence calculation, and scheduling priority matching;

[0033] The power management unit ensures continuous operation in the event of power failure or abnormal voltage, providing temporary power support for virtual sampling, message queuing, and predictive computation.

[0034] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0035] (1) This invention proposes an efficient and secure remote interactive intelligent converged terminal for distribution areas. First, by collecting real-time operating information of each node and constructing a multi-dimensional time-series feature model in combination with the node topology, the power failure detection trigger conditions of each node are determined, each triggering power failure detection node is generated, its state is reconstructed and the power failure detection result is verified, and the task priority of each triggering power failure detection node is determined. The message packet sending strategy is dynamically adjusted. Then, during the reporting process, the parameters of the time series prediction model are adjusted in real time to optimize the virtual prediction accuracy. Furthermore, the message sending strategy is automatically updated to achieve continuous optimization of the model prediction and scheduling strategy. It can maintain continuous perception and reliable reporting of node status under power failure or network fluctuation conditions, dynamically optimize message scheduling, and improve the reliability, response speed and data integrity of the intelligent converged terminal for distribution areas in efficient remote interaction.

[0036] (2) This invention obtains the detection nodes that trigger power outages; subsequently, the system calls the continuous sampling data of these nodes before the power outage, including voltage, current, power factor, and harmonic content, and generates a multi-dimensional state vector by combining the timestamp, topological location, and power coupling coefficient of adjacent nodes. This vector is then imported into a time series prediction model to output the power outage warning coefficient. Simultaneously, by combining the recovery curves of similar historical events, the missing data is simulated and reconstructed, realizing the continuous estimation of node states during power outages. Through the above steps, the intelligent fusion terminal of the distribution area can identify key nodes and predict their operating status in real time when voltage anomalies or power outage events occur, ensuring data continuity and accuracy, providing reliable support for efficient and safe remote monitoring, control, and intelligent interaction, and significantly improving the system's response speed and operational stability to sudden power grid fluctuations.

[0037] (3) This invention fits and compares the trend gradient in the virtual state with the dynamic fluctuation characteristics of the sampling window before the power outage, and simultaneously obtains the power outage warning coefficients of each triggered power outage detection node generated by the time series prediction model, including the power outage signal change rate and acceleration. It then fuses these coefficients with the signal change rate, acceleration, and fluctuation variance of the nodes in the window before the power outage to calculate the power outage confidence factor for each triggered power outage detection node. This confidence factor comprehensively reflects the dynamic trend and short-term fluctuation characteristics of the node before the power outage. Subsequently, the system matches the power outage confidence factor of each node with the scheduling priority corresponding to the confidence factor interval stored in the database to determine the scheduling priority and task priority of each triggered power outage detection node, which guides subsequent message reporting and data processing. Through the above steps, the intelligent fusion terminal of the distribution area can prioritize nodes based on the uncertainty and criticality of their states, achieving rapid response and efficient scheduling of potential power outage events, thereby ensuring timely transmission and processing of key node information, improving the reliability of remote monitoring and the overall security of system operation.

[0038] (4) This invention helps dynamically adjust the message sending strategy by obtaining the reachability scores of each node that triggers power outage detection. This includes adjusting the sending frequency, retransmission count, and channel selection. During message reporting, the system monitors the reporting status of virtual sampled data in real time. When frequent retransmission of data packets or excessive acknowledgment delay is detected, the system triggers a dynamic adjustment program for the time series prediction model parameters and starts a resampling mechanism to continuously sample key operating parameters during the power outage. The message sending strategy is then automatically updated based on the comparison results. This enables reliable collection, intelligent scheduling, and high-priority reporting of the status information of each node, ensuring that power outage event information is transmitted to the cloud in a timely and complete manner, thereby improving the system's response speed to emergencies and the overall security and stability of remote interaction. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the system modules provided in an embodiment of the present invention;

[0041] Figure 2 This is a flowchart of a highly efficient and secure remote interactive intelligent converged terminal for distribution areas provided in an embodiment of the present invention. Detailed Implementation

[0042] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0043] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0044] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0045] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0046] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0047] like Figure 1 As shown, this embodiment of the invention provides an efficient and secure remote interactive intelligent converged terminal for distribution areas, including: a virtual sampling generation module, used to collect the operating information of each node, generate a multi-dimensional time-series feature model based on the topology of each node, determine the triggering conditions for power failure detection of each node to obtain each triggered power failure detection node, reconstruct the state of each triggered power failure detection node, and verify the power failure detection.

[0048] It should be noted that the multidimensional time series feature model is actually a three-dimensional tensor (or multidimensional matrix) in the form of a certain node at a certain point in time. Combined with the state characteristics of the network topology, the multidimensional time series feature model not only records the basic operating state of the node, but also calculates the short-term change rate and abnormal fluctuation threshold of the node, forming a trend curve model that can be dynamically updated.

[0049] The message priority queue module is used to determine the task priority of each triggered power failure detection node, and to obtain the reachability score of each triggered power failure detection node by combining the historical reporting success rate and the current link quality. The message packet sending strategy is dynamically adjusted according to the reachability score of each triggered power failure detection node, and the parameters of the time series prediction model are dynamically adjusted during the message reporting process.

[0050] The closed-loop feedback correction module is used to resample the actual electrical data at the terminal and compare it with the virtual prediction data, and automatically adjust and update the message sending strategy based on the comparison results.

[0051] It should be noted that by comparing the deviation between the actual electrical data resampled by the terminal and the previously virtual predicted data, the system's message sending strategy is dynamically adjusted, achieving self-constraint and correction of information flow and control flow. When the terminal detects a significant deviation between the actual voltage, current, or power factor and the predicted value, the module does not directly modify the data itself. Instead, it automatically adjusts the message reporting frequency, priority, or retransmission strategy according to the degree of deviation, making the communication process responsive to actual changes on site. If the deviation is large, the system will automatically shorten the message cycle and increase the sending density to ensure that abnormal states can be uploaded quickly; if the deviation tends to stabilize, the sending frequency will be reduced to save bandwidth and energy consumption. Through this strategy adjustment based on real feedback, the terminal forms a self-consistent loop of prediction-verification-scheduling-reverification, enabling the message sending process to have self-correction and dynamic adaptation capabilities. Thus, without changing the underlying model, a closed-loop feedback mechanism with prediction deviation as the core driving force is still realized.

[0052] like Figure 2 As shown, Figure 2 This invention provides a flowchart of a highly efficient and secure remote interactive intelligent converged terminal for distribution areas. When a voltage signal characteristic meets the power outage judgment condition, a power outage detection process is triggered, and continuous sampling data is immediately extracted for analysis after triggering. The system calls a time series prediction model to virtually extrapolate the power outage trend, generates an early warning coefficient, and reconstructs missing segments by combining historical event data to improve judgment accuracy. Subsequently, the system integrates the prediction coefficient and dynamic features to calculate a power outage confidence factor, thereby assessing the reliability of the event; scheduling priorities are allocated based on the confidence factor to determine the processing order and resource allocation. Afterward, the terminal encapsulates the analysis results into an event data packet and calculates a reachability score to evaluate the transmission quality of the communication link. If the reachability is low, the system automatically adjusts the sending strategy, such as increasing the number of retransmissions, shortening the sending interval, or using a parallel channel to ensure that information can be stably reported to the cloud. During the reporting process, the system simultaneously monitors the number of retransmissions and the acknowledgment delay to determine the stability of the communication link; if frequent retransmissions of low-confidence data or abnormal delays are detected, the prediction model parameters are dynamically adjusted to achieve adaptive updating and optimization of the model. The detection voltage signal characteristics satisfying the power failure determination condition include determining the trigger conditions for power failure detection at each node. The specific process is as follows:

[0053] Each sampling period is preset, and the voltage signal of each node in each sampling period is collected in a sliding sampling window manner on the time series. The number of times the measured voltage value of each node in each sampling period is continuously lower than a set threshold is counted. The number of sampling points where the measured voltage value of each node in each sampling period is continuously lower than the voltage threshold in the database is also counted. If this number exceeds the allowed number threshold stored in the database, the window is marked as the first power-down window. Thus, each first power-down window is obtained. In each first power-down window of each node, the rate of change of voltage and the rate of change of voltage change of adjacent sampling points are subtracted from and coupled with the average rate of change of voltage and the average rate of change of voltage of adjacent sampling points in the preset historical steady state period, respectively, to obtain the voltage signal rate of change exceeding the limit value of each node. The duration of each first power-down window is counted. If the voltage signal rate of change exceeding the limit value of a node is higher than or equal to the voltage signal rate of change exceeding the limit value stored in the database, and the duration of the first power-down window is higher than the preset duration, then it is preliminarily determined that the node triggers power-down detection, and each triggered power-down detection node is obtained. Otherwise, it is recorded as not triggered.

[0054] It should be noted that the allowable number of occurrences threshold is determined by statistically analyzing the distribution of sampling points where voltage dips briefly within a steady-state cycle without causing a power outage. The system calculates the distribution of the number of times each node's voltage momentarily drops below the threshold but recovers quickly over several past operating cycles, and takes its upper confidence boundary (e.g., the mean plus twice the standard deviation) as the allowable number of occurrences threshold to ensure that small fluctuations do not trigger false detections. The preset value for setting the threshold is derived from the voltage distribution of historical steady-state operating segments in the database, using the mean and standard deviation of the voltage across all steady-state segments as the preset threshold values. The average rate of change and average rate of change of historical steady-state segments are calculated by sliding window differencing of the steady-state data. The system calculates the rate of change of voltage difference between adjacent sampling points divided by the sampling interval for the steady-state voltage sequence, statistically analyzes the mean and standard deviation of this rate of change distribution for all steady-state segments, and takes the mean as the benchmark for the average rate of change. The voltage signal rate of change exceeding the limit is derived from the distribution characteristics of the voltage rate of change deviating from the steady-state template during power outage events. The system calculates the deviation of the rate of change (i.e., the difference between the rate of change and the steady-state average rate of change) within each power outage window. After obtaining the deviation distribution, the mean value is taken as the preset value of the voltage signal rate of change exceeding the limit threshold. The duration threshold is determined based on the distribution of the duration of "voltage continuously below the threshold" in historical power outage events. The duration samples of all power outage events are statistically analyzed, and the mean value is calculated and used as the preset value of the duration threshold as the window duration threshold for physical power outage characteristics.

[0055] The rate of change of voltage signals at each node exceeds the limit. The specific analysis conditions are as follows:

[0056] ;

[0057] In the formula, N iV1 represents the rate of change of the voltage signal at the i-th node exceeding the limit. ij V2 represents the rate of change of the voltage at the j-th adjacent sampling point of the i-th node. ij V1 represents the rate of change of the voltage of the i-th node at the j-th adjacent sampling point, V2 represents the average rate of change of the voltage of the adjacent sampling points in the historical steady state period, j represents the number of the adjacent sampling point, j=1,2,3,...,n, n represents the total number of adjacent sampling points, i represents the number of each node, i=1,2,3,...,m, m represents the total number of nodes.

[0058] It should be noted that i always represents the number or index of the sampling point in the sequence, and its meaning changes with the order of difference: in the first-order difference, it corresponds to the starting point number of the adjacent point pair; in the second-order difference, it corresponds to the starting point number of the first-order difference sequence, which is the first point of the three-point window in the original sequence. Therefore, dividing by n... 1 is because the first-order difference has n One value, divided by n 2 is because the second-order difference has n With two values, this numbering method allows you to clearly know which sampling points each difference value corresponds to, thus enabling you to calculate the deviation or out-of-limit value.

[0059] It should be noted that the rate of change of voltage at each adjacent sampling point specifically refers to the rate of change of voltage at adjacent sampling times, i.e., the time derivative or numerical differential of the voltage signal. Since the voltage signal is a function that changes continuously with time, and the sampling system can only acquire signal values ​​at discrete time intervals (sampling period), in analysis, the difference between adjacent sampling points is usually used to approximate the rate of change of the signal to reflect the trend of voltage change within that small time interval. The rate of change of voltage is obtained by taking the difference again from the former, which is equivalent to the "acceleration" or second derivative of the signal. It is used to characterize the drasticness of voltage change. By comparing the deviation between the current dynamic characteristics of the signal and the historical steady-state statistical characteristics, the strength and persistence of the signal anomaly are quantified. The first power-down window is not the detection result at a single moment, but is dynamically divided in the time series using a sliding sampling window method. Each window covers a certain number of sampling points. By statistically analyzing the continuous low value characteristics and rate of change characteristics of the voltage signal within the window, the early identification of the power-down process can be achieved. When a window is determined to be a power-down window, it means that the signal has experienced a continuous drop accompanied by a large rate of change during that time period, and the system can further statistically analyze its duration.

[0060] It should be noted that the duration of the window can be obtained by multiplying the number of sampling points within the window by the sampling interval.

[0061] It should be noted that the system samples the voltage value once in each sampling period and continuously monitors these consecutive sampling points. If multiple consecutive sampling points (e.g., 5 or 10 consecutive points) are below the threshold, it is considered that there may be a power-down trend. At this time, the system will combine these sampling points and several sampling periods before and after them into a sampling window, namely the first power-down window (e.g., 50 sampling points within 1 second). Then, within this first power-down window, the system calculates the voltage change rate sequence (i.e., the rate of change between every two adjacent sampling points) and the voltage acceleration sequence (the rate of change of the change rate) to determine whether the voltage drop shows a continuous and rapid trend.

[0062] Specifically, the state of each triggered power failure detection node is reconstructed, and the specific process is as follows:

[0063] The system retrieves continuous sampled data sequences of each triggered power failure detection node before power failure, including voltage, current, power factor, and harmonic content. It also obtains the recording timestamp, topological node location, and power coupling coefficient of adjacent nodes. A set of multi-dimensional state vectors is generated and imported into the time series prediction model to output the power failure warning coefficient of each triggered power failure detection node. The system also retrieves the recovery curves of similar historical events to simulate and reconstruct the missing data of each triggered power failure detection node.

[0064] It should be noted that the time series prediction model is used to predict and warn of the electrical state of each triggered power outage detection node before and after a power outage. Essentially, it captures the temporal variation patterns of electrical parameters such as voltage, current, power factor, and harmonic content through historical and current sampled data, thereby predicting the possible development trend of power outage events. The model's input is a set of multi-dimensional state vectors, including the continuous sampled data sequence of each node, the corresponding timestamp, the topological node location, and the power coupling coefficient with adjacent nodes. The output is the power outage warning coefficient for each triggered power outage detection node, used to quantify the risk of a power outage occurring at the current moment and in the short term. The model can employ pre-trained deep learning-based sequence prediction methods, such as Long Short-Term Memory (LSTM), Temporal Convolution Network (TCN), or lightweight recurrent neural networks. It captures the long-term dependence and short-term fluctuations of node electrical parameters through recursive or convolutional operations in the time dimension. At the same time, it enhances the correlation modeling capability between multiple nodes by combining node topology information and power coupling coefficients. In the prediction phase, the model receives the continuous sampling data sequence of the currently triggered node, generates a power outage warning coefficient through forward propagation, and simulates and reconstructs the missing data by combining the recovery curves of similar historical events. This provides a reliable basis for subsequent power outage confirmation, virtual continuous sampling, and message scheduling.

[0065] It should be noted that when a node is confirmed to have triggered a power outage detection and is preparing to reconstruct missing data, during the missing data simulation and reconstruction phase, the system uses the last valid sampled data before the power outage as the initial boundary, and utilizes its changing trend and the real-time or predicted signals of adjacent nodes to construct a multi-dimensional input space. Each dimension in this space represents a dynamic constraint relationship, such as the delay relationship between voltage changes and current response, the conduction coefficient of power fluctuations in adjacent nodes, and the power conservation constraint under the overall topology. The system first constructs a time series prediction curve based on the rate of change and acceleration of the signal before the power outage, and then plots the potential change trajectory of the node. Subsequently, through the power coupling matrix of adjacent nodes, it extracts the dynamic influence components of the target node in the topology network, that is, it uses the voltage and current changing trends of adjacent nodes to estimate the possible response offset of the target node. At the same time, the system retrieves event curves with similar topological locations, power outage durations, and recovery characteristics from the historical similar event database, and uses the Dynamic Time Warping (DTW) algorithm to register the historical curves with the current power outage segment on the time axis and amplitude axis, and calculates their similarity matrix at key change points. The system then fuses the trend prediction results of the current node, the topological projection results of adjacent nodes, and the weighted historical event template, and solves for the state estimates of the missing time points through multidimensional interpolation (time-dimensional interpolation combined with spatial topological interpolation). Finally, the system uses a time-series regression model (such as LSTM or Kalman smoother) to process the interpolated discrete points into a continuous form, generating virtual operating state values ​​for each time point while ensuring the smoothness of voltage and current change trends.

[0066] It should be noted that the data sequence includes the voltage, current, power factor and harmonic content of the nodes in each sampling period, along with timestamp information, the topological location of the nodes and the power coupling coefficient of adjacent nodes.

[0067] Specifically, the power failure detection is verified, and the specific process is as follows:

[0068] The trend gradient in the virtual state is fitted and compared with the dynamic fluctuation characteristics of the window before the power outage, and the power outage warning coefficients of each triggered power outage detection node generated by the time series prediction model are obtained, including the power outage signal change rate and power outage signal acceleration. At the same time, the power outage signal change rate, power outage signal acceleration, and signal fluctuation variance in the sampling window before the power outage of each triggered power outage detection node are extracted and fused to obtain the power outage confidence factor of each triggered power outage detection node. The power outage confidence factor of each triggered power outage detection node reflects the dynamic trend and short-term fluctuation characteristics of the node before the power outage.

[0069] It should be noted that the stability of virtual reconstruction is judged by the difference between residual variance and rate of change. If the reconstructed virtual sequence maintains temporal continuity with the trend before the power outage, spatially aligns with the power coupling direction of adjacent nodes, and the fluctuation variance of the virtual signal is higher than the upper limit of the historical steady-state segment stored in the database, then the virtual state of that node can be considered to have significant abnormal characteristics. After reconstructing the virtual state of the power-out node, the system first compares the virtual voltage, current, and power values ​​generated at each time point with the trend gradient and dynamic fluctuation characteristics of the sampling window before the power outage to determine the stability of the reconstruction. Specifically, the system compares the virtual rate of change at each time point with the historical rate of change in the window before the power outage. If the difference exceeds a set threshold, it is determined that the prediction at that time point has a deviation. At the same time, the residual variance of the virtual state is calculated. If the residual variance is higher than a preset threshold, it indicates that the virtual reconstruction may be affected by abnormal fluctuations and is unreliable. Meanwhile, the second-order change direction is used as a trend consistency check. When the consistency between the second-order change direction of the virtual state and the historical steady-state segment is lower than a threshold, the system considers that the reconstruction result may deviate from the actual trend and needs further correction.

[0070] It should be noted that the power failure confidence factors obtained by fusing the power failure detection nodes are formed through the natural fusion relationship of the normalized parameter inter-function coupling. That is, the mean of the power failure signal change rate and acceleration is calculated on the same scale as the power failure trend strength, and then multiplied by the reciprocal of the window variance to obtain the fused power failure confidence factors.

[0071] It should be noted that detecting a voltage drop below a preset threshold at a certain triggering power outage detection node for several consecutive sampling periods means that the voltage value of that node remains below a pre-set safety threshold across multiple consecutive sampling time points (i.e., the interval between periods during which the system periodically collects data). This indicates that the voltage drop is not an accidental instantaneous fluctuation, but rather a potentially persistent abnormal state. This condition of continuously falling below the threshold triggers the system's time series prediction model, thereby activating the early warning mechanism. By analyzing the persistence of voltage changes, it identifies potential power outage risks, ensuring that the model can respond promptly to slowly decaying or sudden voltage drop events.

[0072] It should be noted that the power outage warning coefficients for each node are dynamically and jointly modeled using a multi-dimensional time series fusion mechanism, based on the node's rate of change parameters, load inertia coefficient, and short-time stability index. Specifically, the system first constructs a multi-dimensional state sequence of nodes within a continuous sampling period, indexed by timestamps. Each frame of this sequence simultaneously contains basic data such as voltage, current, power factor, and harmonic content, as well as the instantaneous rate of change, load inertia response curve, and short-time stability evaluation value calculated from these data. To ensure temporal continuity, the system employs a sliding window mechanism, ensuring that historical samples in each window overlap with real-time data, thereby preserving the inertial characteristics of electrical parameter changes. Covariance updates are performed for each sampling window. Through this time-weighted fusion, the model can not only identify the difference between sudden voltage drops and slow voltage decays but also identify the energy transfer direction between nodes through parameter gradient changes between adjacent windows, further enhancing spatial coupling. Ultimately, the system generates an evolution curve at the output of the time series fusion, incorporating joint characteristics of rate of change, inertia, and stability. The rate of change and acceleration of this fused curve are updated within each sampling period, forming the power outage warning coefficient. This curve reflects the dynamic transition of the node's electrical state from normal operation to potential power outage. The fusion result retains the sensitive response characteristics of the rate of change while incorporating the smoothness of load inertia and the constraints of short-term stability. The output time series not only reflects transient fluctuations but also reveals potential persistent risks.

[0073] Specifically, the task priority of each node that triggers power failure detection is determined, and the specific process is as follows:

[0074] The scheduling priority of each triggered power failure detection node is obtained by matching the power failure confidence factor of each node with the scheduling priority corresponding to each interval of the power failure confidence factor of the node stored in the database. The task priority of each triggered power failure detection node is then determined based on the scheduling priority of each triggered power failure detection node.

[0075] It should be noted that the matching process, which involves matching the power failure confidence factor of each triggered power failure detection node with the scheduling priority corresponding to each interval of the node's power failure confidence factor stored in the database, is based on a pre-defined power failure confidence factor-scheduling priority mapping interval table in the database. This table divides the confidence factor into several intervals and assigns a corresponding priority to each interval. During matching, the system compares the confidence factor value calculated by the currently triggered power failure detection node with the interval table in the database, finds the confidence factor interval it falls into, and directly reads the corresponding scheduling priority.

[0076] Determining the task priority of each triggered power failure detection node also includes extracting nodes with high confidence scores whose power failure confidence factors are higher than the confidence factor threshold stored in the database. These nodes are designated as judgment nodes. If the variance of virtual signal fluctuation is higher than the upper limit of the historical steady-state segment stored in the database and the trend gradient continues to decrease, the power failure detection condition is confirmed, and the node is further marked as a power failure pending judgment node. At the same time, a power failure confirmation signal is output to trigger the subsequent virtual continuous sampling, message scheduling, and remote reporting process.

[0077] It should be noted that if the variance of the virtual signal fluctuation is higher than the upper limit of the historical steady-state segment stored in the database and the trend gradient continues to decrease, then for a power failure detection node that has been identified as a judgment node, if its virtual signal fluctuation variance is higher than the upper limit of the historical steady-state segment stored in the database and the trend gradient continues to decrease, then the power failure detection condition is confirmed, the node is marked as a power failure pending judgment node, and a power failure confirmation signal is output to trigger the subsequent virtual continuous sampling, message scheduling and remote reporting process.

[0078] It should be noted that a virtual signal fluctuation variance exceeding the upper limit of the historical steady-state period stored in the database, with a continuously decreasing trend gradient, means that in the voltage or current sequence reconstructed using algorithms for missing time periods, the short-term fluctuation amplitude significantly exceeds the acceptable fluctuation range during the node's historical normal operation (i.e., the variance exceeds the historical steady-state upper limit). Simultaneously, the overall trend of the sequence exhibits a continuous downward slope rather than a brief rebound. These two factors combined indicate that the current signal is both unstable and trending towards continuous decay. In simpler terms, this is not a one-off or transient drop caused by noise, but a persistent and severe downward trend, which typically increases the likelihood of a power outage. Therefore, this situation should be considered a high-priority anomaly in the system, triggering more stringent verification measures, prioritizing reporting, and initiating redundant transmission and subsequent fault assessment procedures.

[0079] It's important to note that continuity is primarily determined from a time series perspective. This can be achieved by ensuring the signal's trend gradient (i.e., the difference between adjacent points or the first-order difference) in the reconstructed virtual signal maintains the same direction, meaning the gradient sign is consistent, and the number of sampling points or the duration of this consistent change exceeds a preset threshold. For example, if the gradients of n consecutive sampling points are all negative (downward), and n or the time length exceeds the set minimum continuous window length, it can be determined as "continuous decline," avoiding misinterpreting occasional transient fluctuations as continuous decline. The downward direction actually refers to a negative gradient, indicating an overall decreasing trend of the signal over time; only the sign and continuity of the gradient are considered. If the gradient sign remains negative and the fluctuation variance exceeds the upper limit of the historical steady-state range, it indicates that the signal is both unstable and continuously declining, thus triggering high-priority power-down processing.

[0080] It should be noted that the task priority of each triggering power failure detection node is determined based on its scheduling priority as follows: the power failure confidence factor of each triggering power failure detection node is used as the direct input for scheduling priority. The higher the confidence factor, the higher the priority of the node. The system immediately increases its message sending frequency, shortens the retransmission interval, and increases the use of parallel redundant channels (e.g., using the main channel and backup channel simultaneously) to ensure that important information arrives as soon as possible. For nodes with medium confidence, an appropriate frequency is used, and incremental reporting or compressed reporting is triggered when necessary to save bandwidth. Nodes with low confidence factors enter the batch aggregation and delayed reporting mode, and are prioritized to send data in packets when the link is relatively idle, thereby reducing the risk of network congestion and conserving node energy.

[0081] It's important to note that determining high, medium, and low confidence factors primarily relies on a pre-defined "Power Failure Confidence Factor - Scheduling Priority Mapping Interval Table" in the system database. This table divides the entire confidence factor value range into several continuous intervals, each corresponding to a specific scheduling priority; for example, it can be divided into high-priority, medium-priority, and low-priority intervals. During system runtime, the system first calculates the power failure confidence factor for each node that triggers power failure detection. Then, it compares this confidence factor value with each interval in the mapping interval table to determine which interval it falls into. Nodes falling into the high-priority interval are marked as having a high confidence factor, triggering an enhanced message sending strategy; nodes falling into the medium-priority interval are marked as having a medium confidence factor, employing a moderate frequency or incremental reporting strategy; and nodes falling into the low-priority interval are marked as having a low confidence factor, employing a delayed reporting or batch aggregation strategy. The entire process is directly mapped and completed based on interval lookup. It does not rely on additional calculations. Instead, it uses the pre-set mapping relationship in the database to determine the task priority and subsequent sending strategy of each node, thereby ensuring that high-confidence nodes report first and quickly reach the cloud, while medium and low-confidence nodes are flexibly scheduled according to network load and energy consumption.

[0082] Specifically, the reachability score of each triggered power-down detection node is obtained, and the specific process is as follows:

[0083] The status information of each triggered power outage detection node is encapsulated into a complete power outage event data packet. The historical reporting success rate and current system link quality information of each triggered power outage detection node are obtained, including signal-to-noise ratio, received packet loss rate, duplicate packet rate, and current latency. The historical reporting success rate of each triggered power outage detection node is combined with the current link quality information using a normalization method to obtain the reachability score of each triggered power outage detection node. This reachability score is used to evaluate the overall reliability of the triggered power outage detection node in delivering messages to the cloud under current network and historical performance conditions.

[0084] It should be noted that the reachability score of each triggered power failure detection node obtained by normalization is obtained by first normalizing the historical reporting success rate of each triggered power failure detection node and various indicators in the current link quality information of the system (including signal-to-noise ratio, received packet loss rate, duplicate packet rate and current delay), so that all values ​​are mapped to a unified range of 0 to 1. Then, these normalized indicators are combined by averaging.

[0085] It should be noted that the complete power outage event data packet includes the last measured value (before the power outage), the virtual predicted value, and the confidence factor. Each data packet also records the node ID, timestamp, and event type, forming a standardized event recording format.

[0086] It should be noted that the reachability scores for each triggered power outage detection node are obtained by mapping the node's historical reporting success rate, historical average acknowledgment latency and packet loss rate, as well as the current channel's signal-to-noise ratio, real-time packet loss rate, and instantaneous latency, onto a common scale. The core of mapping these parameters to the same scale lies in using a normalization method based on historical data. For example, the current value of each indicator (such as latency and packet loss rate) is compared with its own historical baseline, such as the average value over the past 30 days. For indicators with different dimensions, the dimensions are eliminated by calculating the relative ratio of the current value to the historical baseline value, and then a scoring function, such as a sigmoid function, is used to map the result to the 0-1 range. The mapping employs a normalization method relative to a historical baseline or quantile, transforming each raw metric into a comparable score between the lowest acceptable value and the highest ideal value. For metrics with opposite meanings (such as latency and packet loss), a reverse mapping is performed to ensure that higher scores are always better. (In the algorithm design, each metric is predefined with a positive meaning, such as higher scores being better; for metrics like latency and packet loss rate, where lower scores are better, the scoring function is pre-programmed to be calculated inversely, for example, by subtracting the normalized value from 1). These normalized scores are then arithmetically averaged and a time smoother is applied to reduce short-term jitter, ultimately outputting an reachability score between zero and one.

[0087] Specifically, the message packet sending strategy is dynamically adjusted. The process is as follows: extract the reachability score of each node that triggers the power outage detection and compare it with the reachability score threshold stored in the database. If the reachability score of a node that triggers the power outage detection is lower than or equal to the reachability score threshold, then the message packet sending strategy of that node is adjusted. Otherwise, the message packet sending strategy of that node does not need to be adjusted.

[0088] It should be noted that the specific process for adjusting the message packet sending strategy of each node that triggers power failure detection is as follows:

[0089] When the reachability score of a node that triggers a power outage detection is lower than or equal to a preset threshold, the system will immediately make a series of dynamic adjustments to the message packet sending strategy of that node at the message queue level: First, the priority label of the node in the queue will be upgraded to "high attention" and a low reachability mark will be added to the packet header for downstream processing. At the same time, the messages of that node will be switched from the regular batch processing channel to the real-time channel to ensure timely scheduling. In terms of transmission parameters, the initial acknowledgment timeout will be shortened and the preset allowed number of retransmissions will be increased. At the same time, a backup channel or backup path will be opened in parallel for redundant transmission (e.g., sending messages simultaneously through the main channel and the backup channel or switching to a more robust encoding / framing method). To reduce the failure cost of each transmission, the system will adjust the message packet sending strategy of the node. The transmitted packets undergo reversible compression or incremental differential transmission, sending only the most critical fields along with confidence level labels. Low-priority, non-critical data is deferred or merged into subsequent batch packets when necessary. The transmission strategy also includes exponential backoff and rapid switching logic: in the event of multiple failures within a short period, channel switching is immediately triggered and link anomalies are reported to the cloud. Simultaneously, the number of transmission attempts and real-time link metrics are recorded locally to update the historical success rate database. The entire adjustment process ends upon receiving a cloud receipt or reaching the retransmission limit. Based on this, the system updates the node's reachability assessment and uses the transmission result as input for model correction and subsequent priority adjustments. This ensures timely reporting of critical power failure information while minimizing network resources and terminal power consumption.

[0090] Specifically, the parameters of the time series prediction model are dynamically adjusted during the message reporting process. The specific process is as follows: During the message reporting process, the reporting status of virtual sampling data is monitored, including the number of times data packets of each triggered power failure detection node are frequently retransmitted and the acknowledgment delay. When it is found that the data packets of each triggered power failure detection node are frequently retransmitted or the acknowledgment delay exceeds the corresponding frequent retransmission or acknowledgment delay threshold stored in the preset database, the time series prediction model parameter dynamic adjustment program is triggered.

[0091] It should be noted that the program that triggers the dynamic adjustment of time series prediction model parameters first uses the prediction residual generated by the current virtual sampling as the core reference information. It analyzes the magnitude, fluctuation characteristics, and differences from the historical steady-state period of the residual, maps the residual information to the model parameter space, and performs weighted correction on the noise variance term, neighborhood weight distribution, and prediction coefficients in the model to optimize the model's prediction capability under short-term continuous sampling. The corrected parameters are immediately written to the model storage area to ensure that subsequent virtual continuous sampling of power outage events can generate more reliable virtual state values ​​based on the updated model. At the same time, the system marks the transmission status of the reported data packets as valid or partially valid for subsequent fusion, state reconstruction, and scheduling priority calculation, thereby forming a closed-loop feedback mechanism. This enables the time series prediction model to adaptively optimize during the message reporting process, achieving efficient and secure remote perception and continuous simulation of the state of the intelligent fusion terminal in the distribution area during power outages.

[0092] Specifically, it is used for resampling real electrical data at the terminal and comparing it with virtual predicted data, and automatically adjusting the update message sending strategy based on the comparison results, including:

[0093] After being triggered, a resampling mechanism is initiated to continuously sample the key operating parameters of each triggered power outage detection node during the power outage period. The sampled data is then compared one by one with the virtual prediction data at the corresponding time point, and the message sending strategy is automatically adjusted and updated based on the comparison results.

[0094] It should be noted that when a power outage interrupts real-time sampling, the system does not directly replace the virtual sequence. Instead, it performs a strict one-to-one alignment and traceable comparison between the virtual data and the subsequently transmitted real data, using time and events as the axis: First, the last real sampling record before the power outage is used as the alignment anchor point. All subsequently generated virtual points are marked with an absolute timestamp (or sequence number) based on this anchor point and labeled as "predicted / virtual." At the same time, strict time interval and sampling frequency information are maintained in the local virtual cache. When the edge terminal resamples and reports the real data after power is restored... During data processing, the cloud or edge first verifies the time base (using a synchronized clock or time offset within the message header). Based on the expected sampling rhythm, it performs nearest neighbor matching between real and virtual points according to their timestamps. Specifically, dynamic time warping is used during matching to handle minor clock drift or network latency. If a precise correspondence exists between real and virtual points, they are paired one-to-one, and point-by-point residuals are calculated. Simultaneously, for periods when the real sequence occurs, if the real samples arrive at a higher frequency, the virtual sequence can be interpolated / upsampled to achieve point-by-point comparison; otherwise, the real sequence is downsampled or aggregated to align with the virtual time grid. After comparison, the system updates the confidence factor matrix for each time point (based on residual size, adjacency consistency, and historical template matching degree), and uses these residuals and alignment statistics as the basis for model self-correction and message status labeling. If, after recovery, a large number of virtual points are found to have small deviations, they can be batch-"written back" as temporary observations while retaining their source labels. If the deviation is significant, the relevant virtual segments are downgraded to low confidence and reconstruction or manual verification is triggered. If the true value is never returned (long-term disconnection), the system will use the virtual sequence as a temporary substitute and clearly indicate the prediction source and confidence level when reporting. When any subsequent true value receipt arrives, the system will then perform compensation and correction according to the above alignment process.

[0095] It should be noted that comparing the sampled data with the virtual prediction data at the corresponding time points involves sorting the operating parameters of each node, such as voltage, current, power factor, and harmonics, obtained from the terminal's resampling according to timestamps, matching them one by one with the predicted values ​​at the corresponding time points in the virtual prediction data, and then calculating the deviation value for each time point, including voltage deviation, current deviation, power factor deviation, and harmonic deviation. Simultaneously, the confidence index attached to the virtual prediction data is compared, and the deviation value is combined with the confidence level to assess the reliability of the prediction results. The deviation is then compared to see if it exceeds a preset threshold or historical tolerance range. Time points where the deviation exceeds the limit are marked, and the overall deviation distribution, average deviation, and maximum deviation are statistically analyzed to quantify the consistency between virtual prediction and actual sampling. This provides a basis for the dynamic adjustment of message sending strategies, including whether to increase the message sending frequency of that node, add redundant channels, or delay batch sending. At the same time, it provides feedback data to the trend model or virtual sampling module to optimize the accuracy of the next round of prediction and reconstruction, enabling the terminal to continuously verify the virtual prediction data and adaptively adjust the strategy.

[0096] It should be noted that the virtual prediction data includes key electrical operating parameters of each triggered power outage detection node during the power outage or at the time of missing sampling, such as voltage, current, power factor and harmonic content. At the same time, each prediction data point is accompanied by a corresponding confidence index to reflect the credibility of the prediction value. The entire dataset forms a continuous time series, which can be used to complete the node status information during the power outage, verify the power outage detection results, assist in message reporting decisions, and support the state reconstruction and anomaly analysis at the edge or cloud.

[0097] It should be noted that the virtual prediction data is generated by the terminal by taking the real time series of several sampling periods before power-off and combining the values ​​of voltage, current, power factor, harmonics, etc. with timestamps and node topological locations into a multi-dimensional state vector. Then, the temporal features such as trend gradient, instantaneous rate of change, and short-term acceleration are extracted from this vector as boundary conditions. Based on these boundary conditions, the module calls a time series predictor (such as a lightweight autoregressive model, LSTM / TCN variant, or hybrid attention network) trained offline / fine-tuned online to perform short-term regression extrapolation on several future sampling points. At the same time, the topological coupling matrix is ​​used to map the real-time or predicted dynamics of neighboring nodes into spatial constraints on the target node to ensure the consistency of the predicted values ​​in the spatial dimension. In parallel, the system retrieves the recovery curves of similar historical events and aligns the matched templates in terms of scale and time axis. These historical templates are used to compensate for possible systematic biases in the regression prediction through multi-dimensional interpolation. Subsequently, the regression prior, neighborhood projection, and historical templates are fused according to similarity and consistency rules, and the optimal estimate at each time step is solved using constrained optimization (such as least squares or Bayesian fusion framework). The confidence information of each generated estimation point is calculated simultaneously. The confidence is derived by combining the variance of the model prediction residual, the stability measure of the input window, the neighborhood consistency index and the historical template matching degree. Finally, the boundary effect is eliminated by time smoothing (such as Kalman smoothing or forward and backward filtering) and the continuous virtual voltage, current and power time series and corresponding confidence matrix are output for reporting, analysis and subsequent self-learning modules.

[0098] The second aspect of the present invention also provides a highly efficient and safe remote interactive smart integrated energy meter for distribution areas, including a data acquisition unit, which is responsible for real-time acquisition of operating parameters such as voltage, current, power factor and harmonics of each node, and generating a multi-dimensional state vector from the acquired data according to the sampling period.

[0099] The communication unit is used for remote data interaction with the host terminal, cloud platform and nearby smart meters, and supports dynamic scheduling of message queues and redundant channel transmission.

[0100] The processing unit integrates a time series prediction model and a virtual sampling module. It can virtually reconstruct the collected data, generate power outage warning coefficients and power outage confidence factors in the event of power failure or abnormality, and compare the prediction results with the real sampled data to achieve power outage detection verification and continuous sensing.

[0101] The storage unit is used to store historical sampling data, virtual prediction data, trend model parameters, and historical event databases, providing basic data for reconstruction, confidence calculation, and scheduling priority matching.

[0102] The power management unit ensures continuous operation in the event of power failure or abnormal voltage, providing temporary power support for virtual sampling, message queuing, and predictive computation.

[0103] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0104] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0105] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0106] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0107] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0108] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0109] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0110] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0111] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0112] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0113] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0114] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A high-efficiency safe remote interaction area intelligent fusion terminal, characterized in that, The terminal comprises: A virtual sampling generation module is configured to collect running information of each node, generate a multi-dimensional time sequence characteristic model according to a topology structure of each node, determine a triggering condition of a power-off detection of each node to obtain each triggered power-off detection node, reconstruct a state of each triggered power-off detection node, and verify the power-off detection; A message priority queue module is configured to determine a task priority of each triggered power-off detection node, obtain a reachability score of each triggered power-off detection node by combining a historical reporting success rate of each triggered power-off detection node and a current link quality, dynamically adjust a message packet sending strategy according to the reachability score of each triggered power-off detection node, and dynamically adjust a parameter of a time sequence prediction model in a message reporting process; A closed-loop feedback correction module is configured to re-sample real electrical data of the terminal and compare the real electrical data with virtual prediction data, and automatically adjust and update a message sending strategy according to a comparison result; The determination of the triggering condition of the power-off detection of each node includes the following specific process: A preset sampling period is set, voltage signals of each node in each sampling period are collected in a sliding sampling window manner on a time sequence, a number of voltage measured values of each node that are continuously lower than a set threshold in each sampling period is counted, a number of sampling points of the voltage measured values of each node that are continuously lower than voltage thresholds in a database in each sampling period is counted, if the number exceeds a threshold of an allowed number stored in the database, the window is marked as a first power-off window, thereby obtaining each first power-off window, a change speed of each adjacent sampling point of the voltage and a change speed of a voltage change rate in each first power-off window of each node are respectively subtracted from and coupled with a preset average change speed of adjacent sampling points of a historical steady state segment and a voltage average change rate to obtain a voltage signal change rate overrun value of each node, and a duration of each first power-off window is counted, if the voltage signal change rate overrun value of a certain node is higher than or equal to a voltage signal change rate overrun value stored in the database, and the duration of the first power-off window is higher than a preset duration, it is preliminarily determined that the node triggers the power-off detection to obtain each triggered power-off detection node, otherwise, it is recorded as not triggering.

2. The high-efficiency and safe remote interaction area intelligent fusion terminal according to claim 1, characterized in that, The reconstruction of the state of each triggered power-off detection node includes the following specific process: A continuous sampling data sequence before power-off of each triggered power-off detection node is called, including voltage, current, power factor and harmonic content, a record time stamp, a topology node position and a power coupling coefficient of an adjacent node are obtained at the same time, a group of multi-dimensional state vectors are generated and imported into a time sequence prediction model to output a power-off early warning coefficient of each triggered power-off detection node, and a recovery curve of a historical similar event is called to simulate and reconstruct missing data of each triggered power-off detection node.

3. The high-efficiency and safe remote interaction area intelligent fusion terminal according to claim 2, characterized in that, The verification of the power-off detection includes the following specific process: The trend gradient in the virtual state is fitted and compared with the dynamic fluctuation characteristics of the power-off window before power-off, and the power-off early warning coefficients of each power-off detection node generated by the time series prediction model are obtained, including the power-off signal change rate and the power-off signal acceleration. Meanwhile, the power-off signal change rate, the power-off signal acceleration, and the fluctuation variance of the signal in the sampling window before power-off of each power-off detection node are extracted, and the power-off confidence factor of each power-off detection node is obtained by fusion. The power-off confidence factor of each power-off detection node reflects the dynamic trend and short-time fluctuation characteristics of the node before power-off.

4. The high-efficiency and safe remote interaction area intelligent fusion terminal according to claim 3, characterized in that, The specific process of determining the task priority of each power-off detection node is as follows: The scheduling priority of each power-off detection node is obtained by matching the power-off confidence factor of each power-off detection node with the scheduling priority corresponding to each interval of the node's power-off confidence factor stored in the database. The task priority of each power-off detection node is determined according to the scheduling priority of each power-off detection node. The task priority of each power-off detection node also includes extracting nodes with high confidence scores whose power-off confidence factors are higher than the confidence factor threshold stored in the database, denoted as decision nodes. If the virtual signal fluctuation variance is higher than the upper limit of the historical steady-state segment stored in the database and the trend gradient is continuously decreasing, it is confirmed that the power-off detection condition is met, and the node is further marked as a power-off pending node. At the same time, a power-off confirmation signal is output to trigger subsequent virtual continuous sampling, message scheduling, and remote reporting processes.

5. The high-efficiency and safe remote interaction area intelligent fusion terminal according to claim 1, characterized in that, The specific process of obtaining the reachability score of each power-off detection node is as follows: The state information of each power-off detection node is encapsulated into a complete power outage event data packet, and the historical reporting success rate of each power-off detection node and the current link quality information of the system are obtained, including the signal-to-noise ratio, the packet loss rate, the repeated message rate, and the current delay. The historical reporting success rate of each power-off detection node is combined with the current link quality information of the system through normalization to obtain the reachability score of each power-off detection node. The reachability score of each power-off detection node is used to evaluate the comprehensive reliability of the triggered power-off detection node in sending messages to the cloud under the current network and historical performance conditions.

6. The intelligent fusion terminal of the high-efficient and safe remote interaction according to claim 1, characterized in that, The specific process of dynamically adjusting the message packet sending strategy is as follows: The reachability score of each power-off detection node is extracted and compared with the reachability score threshold stored in the database. If the reachability score of a certain power-off detection node is lower than or equal to the reachability score threshold, the message packet sending strategy of the power-off detection node is adjusted. Otherwise, the message packet sending strategy of the power-off detection node does not need to be adjusted.

7. The intelligent fusion terminal of the high-efficient and safe remote interaction according to claim 1, characterized in that, The specific process of dynamically adjusting the parameters of the time series prediction model in the message reporting process is as follows: In the message reporting process, the reporting state of virtual sampling data is monitored, including the number of times of frequent retransmission of data packets of each trigger power failure detection node and the delay of acknowledgement, and when it is found that the data packets of each trigger power failure detection node are frequently retransmitted or the delay of acknowledgement exceeds the corresponding frequent retransmission or acknowledgement delay threshold value stored in the preset database, a time series prediction model parameter dynamic adjustment program is triggered.

8. The intelligent fusion terminal of the high-efficient and safe remote interaction according to claim 1, characterized in that, The terminal re-samples real electrical data and compares it with virtual prediction data, and automatically adjusts and updates the message sending strategy according to the comparison result, including: After triggering, the re-sampling mechanism is started, the key operating parameters of each trigger power failure detection node during power failure are continuously sampled, and the sampling data is compared with the virtual prediction data at the corresponding time point, and the message sending strategy is automatically adjusted and updated according to the comparison result.

9. A high-efficiency and safe remote interaction transformer area intelligent fusion electric energy meter, applied to the high-efficiency and safe remote interaction transformer area intelligent fusion terminal of any one of claims 1-8, characterized in that, Including: The data acquisition unit is responsible for real-time acquisition of operating parameters of each node, including voltage, current, power factor and harmonic, and generates a multi-dimensional state vector according to the sampling period; The communication unit is used for remote data interaction with the upper terminal, cloud platform and adjacent smart meters, supports dynamic scheduling of message queue and redundant channel transmission; The processing unit is used to integrate time series prediction model and virtual sampling module, which can virtually reconstruct the collected data, generate power failure warning coefficient and power failure confidence factor under power failure or abnormal conditions, and compare the prediction results with the real sampling data to realize power failure detection verification and continuous perception; The storage unit is used to store historical sampling data, virtual prediction data, trend model parameters and historical event library, which provides basic data for reconstruction, confidence calculation and scheduling priority matching; The power management unit is used to ensure the continuous operation ability under the condition of power failure or voltage abnormality, and provides temporary power support for virtual sampling, message queue and prediction calculation.

Images