Emergency power supply automatic monitoring and power distribution optimization method based on Internet of Things technology

Abstract

The invention discloses an emergency power supply automatic monitoring and power distribution optimization method based on the Internet of Things technology. The method comprises the following steps: S1, obtaining and preprocessing operation data of an emergency power supply; s2, constructing an equivalent simulation model of actual operation of the emergency power supply, and generating a corresponding power topological graph according to a simulation result; s3, applying a Laplacian operator to the power topological graph, and extracting a low-order frequency domain signal change trend; s4, returning the target node to the power distribution measurement and control center, constructing an emergency power supply knowledge graph in combination with historical load data, and endowing scheduling priorities to high-risk nodes; s5, power distribution path simulation reconstruction is executed in the RTDS simulator, and the dynamic response behavior of each path is evaluated and recorded; and S6, screening an optimal path according to an evaluation result, and inputting the optimal path into the RTDS simulator for verification. According to the invention, intelligent closed loop of the emergency power supply from monitoring, identification to regulation is realized, and the real-time performance and stability of a power distribution system are improved.

Description

Technical Field

[0001] The present invention relates to the field of power distribution automation and monitoring technology, and in particular to an emergency power supply automation monitoring and power distribution optimization method based on Internet of Things technology. Background Art

[0002] In modern power infrastructure, emergency power systems, as a key means of ensuring continuous power supply to critical loads and responding to sudden power outages or grid anomalies, have been widely used in hospitals, data centers, rail transit, military facilities, and intelligent manufacturing. Traditional emergency power systems often employ fixed scheduling strategies or semi-automatic control methods, relying on manual intervention and empirical judgment for power switching and load management. These systems suffer from low response efficiency and delayed fault handling, which can lead to power outages to critical loads, economic losses, and even personal safety risks in severe cases. Particularly in complex load environments, traditional emergency power distribution methods struggle to rapidly perceive, judge, and dynamically optimize and control sudden anomalies, leading to increasingly prominent technical bottlenecks.

[0003] With the rapid development of IoT technology, power systems are rapidly evolving from traditional centralized management to a more sensor-based, networked, and intelligent system. A large number of distributed sensor devices can collect key operating parameters such as voltage, current, power, and frequency in seconds or even milliseconds, significantly improving data transmission capabilities and real-time performance, providing new opportunities for automated monitoring and dispatch optimization of emergency power supplies. However, despite existing research achieving certain results in IoT perception, IoT sensor data has yet to be deeply integrated with intelligent control mechanisms in emergency power distribution scenarios, resulting in weak real-time response and optimization capabilities and a fragmented monitoring strategy.

[0004] On the other hand, most current power scheduling optimization methods are based on static rules or heuristic algorithms, making it difficult to adjust power distribution plans in real time in dynamic environments. This is especially true in scenarios where node topology frequently changes and electrical characteristics are transient and complex. Traditional methods cannot accurately characterize the complex coupling relationships between nodes, nor can they accurately identify high-risk nodes and optimal distribution paths. In emergency power supply systems, there is significant heterogeneity in load characteristics, power structure, and operating status at different nodes. Simple linear models are no longer sufficient to address the nonlinear mapping relationship between multidimensional operating data and complex power topology structures.

[0005] Furthermore, existing simulation and emulation methods suffer from timeliness and real-time performance issues. Most studies use offline simulation tools to build dispatch models, which cannot support the coordinated closed-loop dispatch mechanism of "monitoring, simulating, and controlling" in emergency scenarios. RTDS, as a high-precision real-time simulation platform, possesses powerful dynamic modeling and parallel simulation capabilities. However, it has not yet achieved the real-time input of IoT sensor data and the coordinated optimization of distribution models in the field of emergency power supply automation scheduling. This decoupling of the physical grid model and the information model limits its real-time control capabilities.

[0006] In terms of topological analysis, existing research is mostly based on static analysis of graph models, failing to effectively utilize the frequency domain characteristics and time-varying trends of graph structures, especially in identifying short-term power disturbances such as voltage drops, current mutations, and frequency anomalies between nodes. As a core tool in graph analysis, the Laplacian operator is still in its infancy in the frequency domain analysis of power topology signals. In particular, the depth of algorithms used to construct power disturbance propagation paths and identify nodes with potential power outage risks needs to be improved. Existing methods fail to systematically identify high-risk nodes from the perspective of signal frequency disturbances, and lack low-order frequency domain anomaly detection methods based on graph Fourier transforms, resulting in the inability of scheduling strategies to proactively intervene.

[0007] In terms of knowledge representation and strategy generation, most current emergency dispatch systems lack the semantic fusion modeling capabilities for multi-dimensional relationships such as nodes, electrical properties, and spatial information, making it impossible to optimize dispatch decisions based on operational knowledge. In particular, there is a lack of a unified knowledge graph support structure for node priority judgments under multiple objectives and conditions. Traditional data-driven models rely solely on historical loads and operating curves, lacking structured expression capabilities and unable to perform complex logical reasoning and dynamic priority updates. This results in the inability to comprehensively evaluate high-risk nodes under multi-factor correlations. Furthermore, the lack of the ability to model cross-temporal and spatial power supply relationships makes it difficult for control strategies to adapt to the dynamic reconstruction needs of the power network.

[0008] Therefore, how to provide an emergency power supply automation monitoring and distribution optimization method based on Internet of Things technology is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0009] One purpose of the present invention is to propose an emergency power supply automation monitoring and distribution optimization method based on Internet of Things technology. The present invention makes full use of various technical means such as Internet of Things real-time perception, RTDS real-time simulation, power topology map Laplacian operator analysis and knowledge graph intelligent reasoning, and describes in detail the closed-loop process from emergency power supply data acquisition, topology construction, frequency domain anomaly identification, priority evaluation to optimal path control. It has the advantages of fast response speed, accurate node identification, intelligent control strategy and high system reliability.

[0010] According to an embodiment of the present invention, a method for automatic monitoring and distribution optimization of emergency power supply based on Internet of Things technology includes the following steps:

[0011] S1. Obtain the operating data of the emergency power supply through the power distribution measurement and control center and perform pre-processing;

[0012] S2. Establish a data exchange channel between the power distribution measurement and control center and the RTDS simulator, input the pre-processed operation data into the RTDS simulator, build an equivalent simulation model of the actual operation of the emergency power supply, and generate the corresponding power topology map based on the simulation results;

[0013] S3. Apply the Laplacian operator to the power topology graph to calculate the topological association weights between nodes, forming a weighted graph structure. It also extracts the trend of low-order frequency domain signals and identifies target nodes with risks of voltage sag, current surge, frequency anomaly, or instantaneous power outage.

[0014] S4. Return the target node to the power distribution measurement and control center, combine it with historical load data, build an emergency power supply knowledge graph, and assign scheduling priority to high-risk nodes;

[0015] S5. Based on the weighted graph structure and node priority, perform power distribution path simulation reconstruction in the RTDS simulator to generate a multi-path power distribution plan, and evaluate and record the dynamic response behavior of each path;

[0016] S6. Filter the optimal path based on the evaluation results, input the optimal path into the RTDS simulator for verification, and feed back the verification results to the distribution measurement and control center to form a dynamic closed-loop control process.

[0017] Optionally, the operating data includes voltage, current, power, frequency, power quality and load switch status.

[0018] Optionally, the preprocessing includes normalization, missing value filling and timestamp alignment.

[0019] Optionally, the S2 specifically includes:

[0020] S21, establish a data exchange channel between the power distribution measurement and control center and the RTDS simulator, and input the pre-processed operation data set D from the power distribution measurement and control center to the RTDS simulator, wherein, V i Represents the voltage value of the i-th node, I i represents the current value of the i-th node, P i represents the active power value of the i-th node, f i represents the frequency value of the i-th node, Q i Represents the reactive power value of the i-th node, S iIndicates the load switch state of the i-th node, with a value of 0 or 1, indicating the off or on state, and n represents the number of all monitored nodes;

[0021] S22. Construct a topology matrix B of the emergency power supply based on the operation data set D. The topology matrix is ​​composed of matrix elements:

[0022]

[0023] Among them, b ij Represents the matrix elements in the topological structure matrix, ∈ is a positive real constant to avoid the denominator being zero, V j Represents the voltage value of the jth node, I j represents the current value of the jth node, P j Indicates the active power value of the jth node, Q j represents the reactive power value of the jth node;

[0024] S23, input the topology matrix B into the RTDS simulator, perform network construction and steady-state calculation simulation, and output the node connection graph G = (N, E), where N = {n1, n2, ..., n n} represents a node set, E={e ij |a ij >0} represents the edge set, which constitutes the power topology graph of the actual operation of the emergency power supply.

[0025] Optionally, the power topology map is constructed by a residual graph convolutional network, and the edge set E is reconstructed according to the node embedding similarity to form an enhanced power topology map. The residual graph convolutional network adopts a residual skip connection mechanism to optimize the feature propagation path between nodes. The topology structure matrix B and the node feature matrix X = [V, I, P, Q, f] are used as inputs, and the constructed graph convolution propagation function is:

[0026]

[0027] Where V represents the voltage value set of all nodes, I represents the current value set of all nodes, P represents the active power value set of all nodes, Q represents the reactive power value set of all nodes, and f represents the frequency value set of all nodes. is the adjacency matrix after adding the self-loop, satisfying for The degree matrix, H (l) represents the graph convolution input feature of the lth layer, H (l+1) Represents the output features of the graph convolution at layer l+1, initially H (0) =X,W (l) represents the trainable weight matrix of the lth layer, and σ represents the nonlinear activation function.

[0028] Optionally, the S3 specifically includes:

[0029] S31. Apply the Laplacian operator to the power topology graph to calculate the topological association weights between nodes and form a weighted graph structure A = [a ij ] and the corresponding degree matrix D=diag(d1,d2,…,d n ), and calculate the Laplacian matrix L, satisfying L = DA, where diag(·) is a diagonal matrix, Represents node n i The total connection weight value, n represents the number of all monitoring nodes;

[0030] S32, perform eigendecomposition on the Laplacian matrix L to obtain the eigenvalue set Λ={λ1,λ2,…,λ n} and the corresponding eigenvector set U=[u1,u2,…,u n ], where λ k represents the kth eigenvalue, satisfying Lu k =λ k u k , and u k represents the kth eigenvector;

[0031] S33. Define the operating status matrix where x i =[V i ,I i ,f i ] represents the three-dimensional state vector of voltage, current, and frequency of the i-th node. Fourier transform is performed based on the feature transformation, and the change trend of the low-order frequency domain signal is extracted:

[0032]

[0033] in, represents the state component of node i in the jth running variable dimension, for u k The transpose of m is the set low-order frequency threshold index, e is the base of the natural logarithm, θ i Represents node n i The low-order frequency domain disturbance index value;

[0034] S34, set the abnormality detection threshold δ, when θ i >δ, the corresponding node n i Marked as target node, the target node set is recorded as N * ={n i |θ i>δ}, the abnormality type of the target node includes voltage sag, current surge, frequency abnormality or instantaneous power failure risk.

[0035] Optionally, the topology association weight is calculated based on the voltage, current and frequency state differences between nodes in the power topology graph to generate a weighted graph structure, and a degree matrix corresponding to the node connection strength is constructed accordingly.

[0036] Optionally, the S4 specifically includes:

[0037] S41, set the target node N * Return to the distribution control center and obtain the historical load data set H = {h it |i∈N * ,t=1,2,…,T}, where h it represents the active load value of the i-th node in the t-th cycle;

[0038] S42, construct triple set Φ={(n i ,r k ,o j )}, where n i ∈N * represents the target node, Indicates the spatial region number where the node is located. Represents the semantic relationship between nodes and regions, and constructs an emergency power supply knowledge graph based on the triple set in is a collection of entities, is the relationship set, Φ is the edge set in the graph;

[0039] S43. Define a scheduling priority function and assign scheduling priority to high-risk nodes:

[0040]

[0041] Among them, θ i Represents node n i The low-order frequency domain disturbance index value, θ j Represents node n j The low-order frequency domain disturbance index value, d i Represents node n i The total connection weight value, d j Represents node n j The total connection weight value, α∈[0,1] represents the frequency domain weight coefficient, β∈[0,1] represents the structural weight coefficient, π(n i ) represents node n i The scheduling priority of , max represents the maximum value function.

[0042] Optionally, the S42 specifically includes:

[0043] S421, collect emergency power supply operation data and historical load information, and construct a triple set Φ={(n i ,r k ,o j )}, and get the initial map where n i ∈N * represents the target node, Indicates the spatial region number where the node is located. Represents the semantic relationship between nodes and regions, is a collection of entities, is the relationship set, Φ is the edge set in the graph;

[0044] S422. Use graph neural network structure to build embedding function and introduce attention mechanism to optimize the target

[0045]

[0046] Among them, f(n i ) represents node n i The embedding vector, f(n j ) represents node n j The embedding vector of For node n i The historical load feature representation vector, w ij is the structural similarity weight of node i, j, cos(·) is the cosine similarity function, ‖·‖2 represents the Euclidean distance;

[0047] S423. Perform hierarchical clustering on the target nodes in the knowledge graph based on the embedding results, divide the high-risk emergency area set, and generate upstream and downstream regulatory relationships for the cluster center nodes, ultimately forming a multi-level structure of the emergency power supply knowledge graph.

[0048] Optionally, the S5 specifically includes:

[0049] S51, according to the weighted graph structure and node scheduling priority function, perform path search in the RTDS simulator and build a path set where p k Indicates that from the starting node To the end node An effective power distribution path, and all paths meet the node connection conditions

[0050] S52, according to the path collection Each path p in k , calculate the dynamic response evaluation value:

[0051]

[0052] Among them, V i Represents node n i The voltage value, V i+1 Represents node n i+1 The voltage value, I i Represents node n i The current value, I i+1 Represents node n i+1 Current value, f i Represents node n i The frequency value, π(n i ) represents node n i The scheduling priority, a i,i+1 Represents the connection weight between adjacent nodes, ε represents a constant to prevent division by zero, and γ represents the frequency fluctuation sensitivity coefficient. represents the first-order derivative of the node frequency, Ψ(p k ) represents the path p k Dynamic response evaluation value of

[0053] S53, record path set Ψ(p k ) values ​​and sort them as the basis for screening the optimal path.

[0054] The beneficial effects of the present invention are:

[0055] First, by establishing a real-time data exchange channel between the distribution measurement and control center and the RTDS simulator, this invention achieves high-precision modeling and dynamic simulation of the emergency power supply's operating status. This effectively addresses the issues of delayed information perception and inaccurate simulation modeling in traditional emergency systems, providing reliable basic data support for subsequent scheduling strategies. By collecting and preprocessing multidimensional operating parameters such as voltage, current, and frequency at high frequencies, the system's response speed to emergencies and the accuracy of its modeling are significantly improved.

[0056] Secondly, this invention introduces the Laplacian operator to perform frequency-domain analysis of power topology maps, combined with Fourier transforms to extract low-order disturbance characteristic signals. This enables forward-looking identification of high-risk nodes such as voltage sags, current fluctuations, and frequency anomalies, overcoming the technical bottleneck of traditional rule-driven detection methods, which suffer from poor robustness and high false positive rates in nonlinear power grid environments. Furthermore, by integrating historical load information with node structural characteristics, a knowledge graph for emergency dispatch is constructed, and mathematical functions are introduced to calculate dispatch priorities, providing a quantitative basis and interpretable model for node management and resource scheduling in complex power supply environments.

[0057] Finally, this paper constructs a multi-path power distribution reconstruction model based on the RTDS simulation platform. Through simulation verification and a dynamic feedback mechanism, it achieves closed-loop optimization of the scheduling strategy. In complex scenarios, it can automatically select the optimal path and complete scheduling deployment, avoiding scheduling errors and energy waste caused by human intervention delays. The entire method connects the three major links of data perception, model reasoning, and control execution, truly realizing the intelligent, efficient, and stable operation of the emergency power supply system, significantly improving the system's survivability and power supply guarantee capabilities in extreme events. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0059] Figure 1 This is a flow chart of the method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology proposed by the present invention;

[0060] Figure 2 This is a flowchart of RTDS simulation and topology generation for the method of emergency power supply automation monitoring and power distribution optimization based on Internet of Things technology proposed in the present invention;

[0061] Figure 3 This is a flowchart of Laplacian topology processing and frequency domain identification for the emergency power supply automation monitoring and distribution optimization method based on Internet of Things technology proposed in the present invention. DETAILED DESCRIPTION

[0062] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0063] refer to Figure 1-3 ,The emergency power supply automatic monitoring and power distribution optimization method based on the Internet of Things technology includes the following steps:

[0064] S1. Obtain the operating data of the emergency power supply through the power distribution measurement and control center and perform pre-processing;

[0065] S2. Establish a data exchange channel between the power distribution measurement and control center and the RTDS simulator, input the pre-processed operation data into the RTDS simulator, build an equivalent simulation model of the actual operation of the emergency power supply, and generate the corresponding power topology map based on the simulation results;

[0066] S3. Apply the Laplacian operator to the power topology graph to calculate the topological association weights between nodes, forming a weighted graph structure. It also extracts the trend of low-order frequency domain signals and identifies target nodes with risks of voltage sag, current surge, frequency anomaly, or instantaneous power outage.

[0067] S4. Return the target node to the power distribution measurement and control center, combine it with historical load data, build an emergency power supply knowledge graph, and assign scheduling priority to high-risk nodes;

[0068] S5. Based on the weighted graph structure and node priority, perform power distribution path simulation reconstruction in the RTDS simulator to generate a multi-path power distribution plan, and evaluate and record the dynamic response behavior of each path;

[0069] S6. Filter the optimal path based on the evaluation results, input the optimal path into the RTDS simulator for verification, and feed back the verification results to the distribution measurement and control center to form a dynamic closed-loop control process.

[0070] By constructing a complete process based on IoT perception, RTDS simulation, graph structure analysis and power distribution strategy optimization, the present invention realizes real-time monitoring of the operating status of the emergency power supply, fault identification and generation of multi-path power distribution solutions, forming an efficient, stable and closed-loop automated control system, which significantly improves the intelligence level and response speed of the emergency power supply system.

[0071] In this embodiment, the operating data includes voltage, current, power, frequency, power quality and load switch status.

[0072] The present invention clarifies the types of emergency power supply operation data, covering multi-dimensional information such as voltage, current, power, frequency, power quality and load status, thereby ensuring the comprehensiveness of system input and the accuracy of analysis basis, and providing precise data support for building a reliable scheduling optimization model.

[0073] In this embodiment, the preprocessing includes normalization, missing value filling and timestamp alignment.

[0074] The present invention adopts normalization, missing value filling and timestamp alignment operations in the data preprocessing stage, which improves the consistency and effectiveness of multi-source asynchronous perception data and significantly enhances the stability and accuracy of subsequent topology modeling and frequency domain analysis processes.

[0075] In this embodiment, S2 specifically includes:

[0076] S21, establish a data exchange channel between the power distribution measurement and control center and the RTDS simulator, and input the pre-processed operation data set D from the power distribution measurement and control center to the RTDS simulator, wherein, Vi Represents the voltage value of the i-th node, I i represents the current value of the i-th node, P i represents the active power value of the i-th node, f i represents the frequency value of the i-th node, Q i Represents the reactive power value of the i-th node, S i Indicates the load switch state of the i-th node, with a value of 0 or 1, indicating the off or on state, and n represents the number of all monitored nodes;

[0077] S22. Construct a topology matrix B of the emergency power supply based on the operation data set D. The topology matrix is ​​composed of matrix elements:

[0078]

[0079] Among them, b ij Represents the matrix elements in the topological structure matrix, ∈ is a positive real constant to avoid the denominator being zero, V j Represents the voltage value of the jth node, I j represents the current value of the jth node, P j Indicates the active power value of the jth node, Q j represents the reactive power value of the jth node;

[0080] S23, input the topology matrix B into the RTDS simulator, perform network construction and steady-state calculation simulation, and output the node connection graph G = (N, E), where N = {n1, n2, ..., n n} represents a node set, E={e ij |a ij >0} represents the edge set, which constitutes the power topology graph of the actual operation of the emergency power supply.

[0081] The present invention constructs a topology structure matrix based on operating data and implements real-time modeling in the RTDS simulator to generate a power topology map that accurately reflects the actual physical power grid state, significantly improving the authenticity of the model and the practicality of scheduling simulation.

[0082] In this embodiment, the power topology graph is constructed by a residual graph convolutional network, and the edge set E is reconstructed according to the node embedding similarity to form an enhanced power topology graph. The residual graph convolutional network adopts a residual skip connection mechanism to optimize the feature propagation path between nodes. The topology matrix B and the node feature matrix X = [V, I, P, Q, f] are used as inputs, and the constructed graph convolution propagation function is:

[0083]

[0084] Where V represents the voltage value set of all nodes, I represents the current value set of all nodes, P represents the active power value set of all nodes, Q represents the reactive power value set of all nodes, and f represents the frequency value set of all nodes. is the adjacency matrix after adding the self-loop, satisfying for The degree matrix, H (l) represents the graph convolution input feature of the lth layer, H (l+1) Represents the output features of the graph convolution at layer l+1, initially H (0) =X,W (l) represents the trainable weight matrix of the lth layer, and σ represents the nonlinear activation function.

[0085] The present invention introduces a residual graph convolutional network in topology construction, enhances the depth and effectiveness of feature transfer between nodes through the skip connection mechanism, and constructs a structure-enhanced power map based on the node state characteristics, thereby improving the robustness and perception accuracy of the overall system in complex scenarios.

[0086] In this embodiment, S3 specifically includes:

[0087] S31. Apply the Laplacian operator to the power topology graph to calculate the topological association weights between nodes and form a weighted graph structure A = [a ij ] and the corresponding degree matrix D=diag(d1,d2,…,d n ), and calculate the Laplacian matrix L, satisfying L = DA, where diag(·) is a diagonal matrix, Represents node n i The total connection weight value, n represents the number of all monitoring nodes;

[0088] S32, perform eigendecomposition on the Laplacian matrix L to obtain the eigenvalue set Λ={λ1,λ2,…,λ n} and the corresponding eigenvector set U=[u1,u2,…,u n ], where λ k represents the kth eigenvalue, satisfying Lu k =λ k u k , and u k represents the kth eigenvector;

[0089] S33. Define the operating status matrix where x i =[V i ,I i ,f i] represents the three-dimensional state vector of voltage, current, and frequency of the i-th node. Fourier transform is performed based on the feature transformation, and the change trend of the low-order frequency domain signal is extracted:

[0090]

[0091] in, represents the state component of node i in the jth running variable dimension, for u k The transpose of m is the set low-order frequency threshold index, e is the base of the natural logarithm, θ i Represents node n i The low-order frequency domain disturbance index value;

[0092] S34, set the abnormality detection threshold δ, when θ i >δ, the corresponding node n i Marked as target node, the target node set is recorded as N * ={n i |θ i >δ}, the abnormality type of the target node includes voltage sag, current surge, frequency abnormality or instantaneous power failure risk.

[0093] Based on the Laplacian operator and Fourier transform technology, the present invention accurately extracts abnormal disturbance trends from low-order frequency domain signals, and can promptly identify key risk nodes such as voltage drops, current fluctuations, and frequency anomalies, thereby enhancing the system's early warning capability for potential power failures.

[0094] In this embodiment, the topology association weight is calculated based on the voltage, current and frequency state differences between nodes in the power topology graph to generate a weighted graph structure, and a degree matrix corresponding to the node connection strength is constructed accordingly.

[0095] The present invention constructs topological association weights and forms a weighted graph structure based on the state differences between node voltages, currents and frequencies, thereby improving the degree of fit of the graph model to the physical characteristics of the power grid and providing a more physically meaningful structural foundation for scheduling path optimization.

[0096] In this embodiment, the S4 specifically includes:

[0097] S41, set the target node N * Return to the distribution control center and obtain the historical load data set H = {h it |i∈N * ,t=1,2,…,T}, where h it represents the active load value of the i-th node in the t-th cycle;

[0098] S42, construct triple set Φ={(n i ,r k ,o j )}, where n i ∈N * represents the target node, Indicates the spatial region number where the node is located. Represents the semantic relationship between nodes and regions, and constructs an emergency power supply knowledge graph based on the triple set in is a collection of entities, is the relationship set, Φ is the edge set in the graph;

[0099] S43. Define a scheduling priority function and assign scheduling priority to high-risk nodes:

[0100]

[0101] Among them, θ i Represents node n i The low-order frequency domain disturbance index value, θ j Represents node n j The low-order frequency domain disturbance index value, d i Represents node n i The total connection weight value, d j Represents node n j The total connection weight value, α∈[0,1] represents the frequency domain weight coefficient, β∈[0,1] represents the structural weight coefficient, π(n i ) represents node n i The scheduling priority of , max represents the maximum value function.

[0102] The present invention constructs an emergency power supply knowledge graph and introduces a quantitative scheduling priority function, integrating multi-factor evaluations such as historical load, frequency domain disturbance, and node connectivity, thereby achieving dynamic sorting and precise regulation of high-risk nodes and optimizing resource allocation efficiency.

[0103] In this embodiment, the S42 specifically includes:

[0104] S421, collect emergency power supply operation data and historical load information, and construct a triple set Φ={(n i ,r k ,o j )}, and get the initial map where n i ∈N * represents the target node, Indicates the spatial region number where the node is located. Represents the semantic relationship between nodes and regions, is a collection of entities, is the relationship set, Φ is the edge set in the graph;

[0105] S422. Use graph neural network structure to build embedding function and introduce attention mechanism to optimize the target

[0106]

[0107] Among them, f(n i ) represents node n i The embedding vector, f(n j ) represents node n j The embedding vector of For node n i The historical load feature representation vector, w ij is the structural similarity weight of node i, j, cos(·) is the cosine similarity function, ‖·‖2 represents the Euclidean distance;

[0108] S423. Perform hierarchical clustering on the target nodes in the knowledge graph based on the embedding results, divide the high-risk emergency area set, and generate upstream and downstream regulatory relationships for the cluster center nodes, ultimately forming a multi-level structure of the emergency power supply knowledge graph.

[0109] This paper combines graph neural networks with attention mechanisms to improve the semantic integrity and structural consistency of embedded representations during knowledge graph construction, while introducing hierarchical clustering to generate regulatory relationships, thereby enhancing the system's knowledge expression and reasoning capabilities.

[0110] In this embodiment, the S5 specifically includes:

[0111] S51, according to the weighted graph structure and node scheduling priority function, perform path search in the RTDS simulator and build a path set where p k Indicates that from the starting node To the end node An effective power distribution path, and all paths meet the node connection conditions

[0112] S52, according to the path collection Each path p in k , calculate the dynamic response evaluation value:

[0113]

[0114] Among them, V i Represents node n i The voltage value, Vi+1 Represents node n i+1 The voltage value, I i Represents node n i The current value, I i+1 Represents node n i+1 Current value, f i Represents node n i The frequency value, π(n i ) represents node n i The scheduling priority, a i,i+1 Represents the connection weight between adjacent nodes, ε represents a constant to prevent division by zero, and γ represents the frequency fluctuation sensitivity coefficient. represents the first-order derivative of the node frequency, Ψ(p k ) represents the path p k Dynamic response evaluation value of

[0115] S53, record path set Ψ(p k ) values ​​and sort them as the basis for screening the optimal path.

[0116] Based on the weighted graph structure and priority function, the present invention performs multi-path search and response evaluation in the RTDS simulator, establishes a quantitative dynamic scheduling mechanism, and can quickly screen the optimal distribution path according to the dynamic behavior of the path voltage, current, and frequency, significantly improving the real-time and scientific nature of the scheduling strategy.

[0117] Example 1:

[0118] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a certain guaranteed smart power distribution scenario. The scenario is a multi-load linkage power supply site with high power supply reliability requirements, which is equipped with 8 high-precision load devices, 2 independent UPS emergency power supply systems, 1 set of mobile diesel generator sets, and a standard power distribution measurement and control center. The site has a daily operating time of no less than 20 hours, and has the characteristics of large load fluctuations, low power outage tolerance, and complex operating status. The traditional emergency power supply system adopts manual duty + static power distribution strategy, which has problems such as slow response, inflexible path, and lack of intelligent judgment in scheduling.

[0119] During this implementation, a multi-dimensional data acquisition terminal supporting the present invention was first deployed at the power distribution site to collect key indicator data such as voltage, current, frequency, active power, reactive power, and load switch status of each power node in real time. The sensor sampling frequency was set to 20 times per second, and the data was uploaded to the power distribution measurement and control center via the edge gateway. Subsequently, a preprocessing algorithm was used to normalize, timestamp align, and complete missing data on the raw data to ensure data quality. The complete state matrix was transmitted once per round in 5-minute increments, with an average single-round data packet size of 2.7MB.

[0120] The preprocessed data is fed into the RTDS simulator in real time, which automatically builds a power topology that reflects the current distribution state. In this test scenario, the topology has 14 nodes and 18 edges. Using the topology construction matrix formula proposed in this paper, the generated topology achieved 97.2% accuracy in the simulation environment, with minimal deviation from the actual distribution structure. Errors are primarily concentrated in determining the critical state of circuit breakers.

[0121] On this basis, the Laplacian operator was applied to construct a weighted graph structure and extract low-order frequency-domain disturbance signals from the graph Fourier transform. During operation, at 12:18 on the fourth day, the frequency-domain disturbance index θ of a certain voltage node was detected to have risen to 0.847 (far exceeding the threshold δ = 0.420). The system then determined that it was a target node with a voltage sag risk. The system further inferred from the knowledge graph that the node was located in area B, had an average load of 1760W and a maximum load of 2150W over the past 72 hours, and frequently experienced start-stop switching, making it a high-risk node. The system automatically assigned the node the highest scheduling priority and generated a ranking result based on the scheduling function π(n).

[0122] The RTDS simulator then reconstructed multiple paths based on the weighted graph structure and scheduling priorities, and output a set of paths. A total of 12 paths were generated, three of which exhibited frequency sensitivity anomalies. The path with the lowest dynamic response evaluation index, R (P3), was P3, at 0.163. The system selected this path as the optimal scheduling path and dynamically verified it using the RTDS simulator. After verification, the simulation results were returned to the distribution control center. The system completed the path switching within 3.6 seconds, successfully ensuring uninterrupted power supply to the load.

[0123] During a total of 67 dispatches over 14 consecutive days, the system's average recognition response time was 1.72 seconds, the average path reconstruction time was 2.45 seconds, and the average closed-loop control process time was 4.21 seconds. Compared to the original manual operation response time of over 40 seconds and the average path dispatch time of 11 seconds, the overall efficiency increased by 85.6%. In addition, the system's recognition accuracy for abnormal conditions such as instantaneous power outages, voltage drops, and current mutations was 98.1%, and the false alarm rate was controlled within 2.7%, effectively avoiding a large number of false operation problems. The specific implementation data statistical results are shown below:

[0124] Table 1 Statistical table of implementation effects of the emergency power supply automation monitoring and power distribution optimization method based on the present invention

[0125]

[0126]

[0127] As shown in Table 1, in actual operation, this invention not only achieves rapid recognition and responsive path reconstruction, but also maintains high stability and accuracy throughout the entire scheduling closed loop, ensuring continuous power supply to critical loads during emergencies. This implementation case demonstrates the feasibility, stability, and intelligence of this invention in an IoT environment, demonstrating its potential for widespread adoption and engineering success.

[0128] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The method for automatic monitoring and distribution optimization of emergency power supply based on Internet of Things technology is characterized by: The steps include: S1. Obtain the operating data of the emergency power supply through the power distribution measurement and control center and perform pre-processing; S2. Establish a data exchange channel between the power distribution measurement and control center and the RTDS simulator, input the pre-processed operation data into the RTDS simulator, build an equivalent simulation model of the actual operation of the emergency power supply, and generate the corresponding power topology map based on the simulation results; S3. Apply the Laplacian operator to the power topology graph to calculate the topological association weights between nodes, forming a weighted graph structure. It also extracts the trend of low-order frequency domain signals and identifies target nodes with risks of voltage sag, current surge, frequency anomaly, or instantaneous power outage. S4. Return the target node to the power distribution measurement and control center, combine it with historical load data, build an emergency power supply knowledge graph, and assign scheduling priority to high-risk nodes; S5. Based on the weighted graph structure and node priority, perform power distribution path simulation reconstruction in the RTDS simulator to generate a multi-path power distribution plan, and evaluate and record the dynamic response behavior of each path; S6. Filter the optimal path based on the evaluation results, input the optimal path into the RTDS simulator for verification, and feed back the verification results to the distribution measurement and control center to form a dynamic closed-loop control process.

2. The method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology according to claim 1 is characterized in that: The operating data includes voltage, current, power, frequency, power quality and load switch status.

3. The method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology according to claim 1 is characterized in that: The preprocessing includes normalization, missing value filling and timestamp alignment.

4. The method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology according to claim 1 is characterized in that: The S2 specifically includes: S21, establish a data exchange channel between the power distribution measurement and control center and the RTDS simulator, and input the pre-processed operation data set D from the power distribution measurement and control center to the RTDS simulator, wherein, V i Represents the voltage value of the i-th node, I i represents the current value of the i-th node, P i represents the active power value of the i-th node, f i represents the frequency value of the i-th node, Q i Represents the reactive power value of the i-th node, S i Indicates the load switch state of the i-th node, with a value of 0 or 1, indicating the off or on state, and n represents the number of all monitored nodes; S22. Construct a topology matrix B of the emergency power supply based on the operation data set D. The topology matrix is ​​composed of matrix elements: Among them, b ij Represents the matrix elements in the topological structure matrix, ∈ is a positive real constant to avoid the denominator being zero, V j Represents the voltage value of the jth node, I j represents the current value of the jth node, P j Indicates the active power value of the jth node, Q j represents the reactive power value of the jth node; S23, input the topology matrix B into the RTDS simulator, perform network construction and steady-state calculation simulation, and output the node connection graph G = (N, E), where N = {n1, n2, ..., n n } represents a node set, E={e ij |a ij >0} represents the edge set, which constitutes the power topology graph of the actual operation of the emergency power supply.

5. The method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology according to claim 4 is characterized in that: The power topology map is constructed by the residual graph convolutional network, and the edge set E is reconstructed according to the node embedding similarity to form an enhanced power topology map. The residual graph convolutional network adopts the residual skip connection mechanism to optimize the feature propagation path between nodes. The topology structure matrix B and the node feature matrix X = [V, I, P, Q, f] are used as inputs. The constructed graph convolution propagation function is: Where V represents the voltage value set of all nodes, I represents the current value set of all nodes, P represents the active power value set of all nodes, Q represents the reactive power value set of all nodes, and f represents the frequency value set of all nodes. is the adjacency matrix after adding the self-loop, satisfying for The degree matrix, H (l) represents the graph convolution input feature of the lth layer, H (l +1) Represents the output features of the graph convolution at layer l+1, initially H (0) =X,W (l) represents the trainable weight matrix of the lth layer, and σ represents the nonlinear activation function.

6. The method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology according to claim 1 is characterized in that: The S3 specifically includes: S31. Apply the Laplacian operator to the power topology graph to calculate the topological association weights between nodes and form a weighted graph structure A = [a ij ] and the corresponding degree matrix D=diag(d1,d2,…,d n ), and calculate the Laplacian matrix L, satisfying L = DA, where diag(·) is a diagonal matrix, Represents node n i The total connection weight value, n represents the number of all monitoring nodes; S32, perform eigendecomposition on the Laplacian matrix L to obtain the eigenvalue set Λ={λ1,λ2,…,λ n } and the corresponding eigenvector set U=[u1,u2,…,u n ], where λ k represents the kth eigenvalue, satisfying Lu k =λ k u k , and u k represents the kth eigenvector; S33, define the running state matrix X=[x1,x2,…,x n ] T , where x i =[V i ,I i ,f i ] represents the three-dimensional state vector of voltage, current, and frequency of the i-th node. Fourier transform is performed based on the feature transformation, and the change trend of the low-order frequency domain signal is extracted: in, represents the state component of node i in the jth running variable dimension, for u k The transpose of m is the set low-order frequency threshold index, e is the base of the natural logarithm, θ i Represents node n i The low-order frequency domain disturbance index value; S34, set the abnormality detection threshold δ, when θ i >δ, the corresponding node n i Marked as target node, the target node set is recorded as N * ={n i |θ i >δ}, the abnormality type of the target node includes voltage sag, current surge, frequency abnormality or instantaneous power failure risk.

7. The method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology according to claim 6 is characterized in that: The topology association weight is calculated based on the voltage, current and frequency state differences between nodes in the power topology graph to generate a weighted graph structure, and a degree matrix corresponding to the node connection strength is constructed based on this.

8. The method for automatic monitoring and distribution optimization of emergency power supply based on Internet of Things technology according to claim 1 is characterized in that: The S4 specifically includes: S41, set the target node N * Return to the distribution control center and obtain the historical load data set H = {h it |i∈N * ,t=1,2,…,T}, where h it represents the active load value of the i-th node in the t-th cycle; S42, construct triple set Φ={(n i ,r k ,o j )}, where n i ∈N * represents the target node, Indicates the spatial region number where the node is located. Represents the semantic relationship between nodes and regions, and constructs an emergency power supply knowledge graph based on the triple set in is a collection of entities, is the relationship set, Φ is the edge set in the graph; S43. Define a scheduling priority function and assign scheduling priority to high-risk nodes: Among them, θ i Represents node n i The low-order frequency domain disturbance index value, θ j Represents node n j The low-order frequency domain disturbance index value, d i Represents node n i The total connection weight value, d j Represents node n j The total connection weight value, α∈[0,1] represents the frequency domain weight coefficient, β∈[0,1] represents the structural weight coefficient, π(n i ) represents node n i The scheduling priority of , max represents the maximum value function.

9. The method for automatic monitoring and power distribution optimization of emergency power supply based on Internet of Things technology according to claim 8 is characterized in that: The S42 specifically includes: S421, collect emergency power supply operation data and historical load information, and construct a triple set Φ={(n i ,r k ,o j )}, and get the initial map where n i ∈N * represents the target node, Indicates the spatial region number where the node is located. Represents the semantic relationship between nodes and regions, is a collection of entities, is the relationship set, Φ is the edge set in the graph; S422. Use graph neural network structure to build embedding function and introduce attention mechanism to optimize the target Among them, f(n i ) represents node n i The embedding vector, f(n j ) represents node n j The embedding vector of For node n i The historical load feature representation vector, w ij is the structural similarity weight of node i, j, cos(·) is the cosine similarity function, ‖·‖2 represents the Euclidean distance; S423. Perform hierarchical clustering on the target nodes in the knowledge graph based on the embedding results, divide the high-risk emergency area set, and generate upstream and downstream regulatory relationships for the cluster center nodes, ultimately forming a multi-level structure of the emergency power supply knowledge graph.

10. The method for automatic monitoring and distribution optimization of emergency power supply based on Internet of Things technology according to claim 1, characterized in that: The S5 specifically includes: S51, according to the weighted graph structure and node scheduling priority function, perform path search in the RTDS simulator and build a path set where p k Indicates that from the starting node To the end node An effective power distribution path, and all paths meet the node connection conditions S52, according to the path collection Each path p in k , calculate the dynamic response evaluation value: Among them, V i Represents node n i The voltage value, V i+1 Represents node n i+1 The voltage value, I i Represents node n i The current value, I i+1 Represents node n i+1 Current value, f i Represents node n i The frequency value, π(n i ) represents node n i The scheduling priority, a i,i+1 Represents the connection weight between adjacent nodes, ε represents a constant to prevent division by zero, and γ represents the frequency fluctuation sensitivity coefficient. represents the first-order derivative of the node frequency, Ψ(p k ) represents the path p k Dynamic response evaluation value of S53, record path set Ψ(p k ) values ​​and sort them as the basis for screening the optimal path.

Images