Real-time monitoring technology for power internet of things data in converter station scene

Through edge computing technology and generalized S transformation, combined with topological potential value evaluation, the problems of large-scale data processing and rapid response of power grid faults in the Internet of Things are solved, real-time monitoring of power grid nodes and rapid fault positioning are realized, and the grid operation is improved stability and responsiveness.

CN118941073APending Publication Date: 2024-11-12GUIYANG BUREAU OF CHINA SOUTHERN POWER GRID CO LTD EHV TRANSMISSION CO
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410952038.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-12

Smart Images

  • Figure CN118941073A_ABST
    Figure CN118941073A_ABST
Patent Text Reader

Abstract

The invention provides an electric power Internet of Things data real-time monitoring technology in a converter station scene. The invention relates to the field of electric power Internet of Things, and aims to solve the problem that a cloud computing mode in the electric power Internet of Things cannot effectively meet the requirements of mass data transmission, computing and storage acquired by a large-scale intelligent terminal. The technology provided by the invention comprises the following steps: regarding various devices or regions in a power grid as nodes, extracting time-frequency characteristics of signals through generalized S transformation, and realizing rapid harmonic identification and fault analysis; analyzing and calculating the security risk of each node through a potential function to obtain a node needing to be focused; a synchronization mechanism is established by the cloud service platform and the edge computing node, risk analysis is performed on the power grid when the circuit does not fail, rapid fault positioning and fault analysis are realized from top to bottom when the circuit fails, and the method can be applied to an actual power grid system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of electric power Internet of Things, to edge computing technology, and specifically to a real-time monitoring technology for electric power Internet of Things data in a converter station scenario. Background Art

[0002] The power Internet of Things is the product of the integration of smart grid and Internet of Things. The power Internet of Things provides effective support for the operation of basic grid services. Sensors, charging piles, smart substations, smart meters and other terminals in all links of the power system have achieved interconnection and interoperability, promoting the intelligent and integrated construction of the power grid. In order to improve the intelligence level of the power grid and the operating efficiency of the power system, strict requirements are put forward in terms of positioning, identification, detection, monitoring, and inspection.

[0003] Edge computing provides real-time computing and communication services to terminals by deploying edge servers near the terminals. Edge computing and cloud computing are complementary. Edge computing deploys computing and storage resources on data source servers close to the edge. Edge computing extends the cloud computing resources of the power communication network center to the edge network, enabling it to better support real-time data processing of the power information system to meet the different needs of the power system business and complete various grid distribution businesses.

[0004] With the continuous development of smart grid systems, intelligent equipment and wireless communication methods are constantly updated, and power automation and intelligence are gradually entering a new stage. The power system relies on distribution equipment for power distribution and power consumption activities. Due to the limitation of technical level, power grid failure is inevitable, and the "power quality" problem is also the focus of people's attention. Qualified power quality means that the power provided to sensitive equipment and the grounding system set are suitable for the normal operation of the equipment. However, in actual applications, there are a large number of high-power, nonlinear, and unbalanced loads, which cause a certain degree of pollution to the power grid, causing a series of power quality problems such as harmonics, interharmonics, three-phase imbalance, voltage fluctuations and flicker, voltage drops and short-term interruptions. How to effectively monitor the operation of the distribution network or the power outage and restoration has become the focus of improving the level of power supply services. At present, the data monitoring of the power Internet of Things is mainly achieved by converging all data into the cloud for calculation. With the expansion of the scale of the power grid and the increase in the number of terminals, the cloud computing model cannot effectively meet the needs of massive data transmission, calculation and storage collected by large-scale intelligent terminals, nor can it make decisions in the first time when the power grid encounters an emergency to reduce losses and restore the operation of the power grid. Summary of the invention

[0005] The present invention provides a real-time monitoring technology for power Internet of Things data in a converter station scenario, which can realize real-time monitoring of the working conditions of each node or device in the power Internet of Things, can quickly discover problems existing in the power grid, and evaluate the risks of each current node, so as to guide workers to make judgments to reduce potential risks and losses. First, several key nodes in the power grid are selected for monitoring. The node can be a device or a working point in the power grid. Each node constitutes an area, and each area can constitute a wider area as a node in the form of a whole. The three-phase voltage and current signals of the node are collected by synchronous sampling and frequency tracking technology, the analog quantity is converted into a digital signal, and the time-frequency characteristics of the signal are extracted based on the generalized S transform, so as to monitor the short circuit, open circuit and impact fault of the power grid, and can realize rapid identification of harmonics, analyze the disturbance existing in the power grid in real time, and actively transmit node working information to edge computing. An edge computing node is responsible for collecting the working information of all nodes in an area. The edge computing node first calculates the node quality of each node or transmits its node quality by the node itself. For multi-source heterogeneous data, a mathematical method is used to convert the working conditions of each node to the same scale, and then calculate the node quality. When a region is used as a node, the edge nodes in the region calculate the expectation and variance of heterogeneous data in the region and transmit them to the next edge node for calculation. A graph of interactions between nodes is established, and the node influence factor is determined by minimizing the topological potential entropy. The topological potential value of each node is calculated in the next step, and the topological potential value is used as the risk value of the node. A synchronization mechanism is established between the cloud service platform and the edge computing node. The cloud server is informed of the overall working status and completes fault location, fault monitoring, and real-time topological analysis from top to bottom. Adjust and improve nodes with higher topological potential values, that is, nodes with higher risks, and quickly locate and analyze faults for failed nodes.

[0006] The technical solution of the present invention is: a real-time monitoring technology of power Internet of Things data in a converter station scenario, mainly including:

[0007] 1. Use sensors to collect node working data, pre-process it, and upload it to the edge node.

[0008] 2. Use generalized S transform to perform harmonic analysis on the signal to achieve power grid fault analysis and monitoring.

[0009] 3. The edge nodes calculate the node quality of each node, establish the interaction relationship diagram between nodes, determine the node influence factor by minimizing the topological potential entropy, and calculate the topological potential value of each node.

[0010] 4. The edge node calculates the working conditions in the area and transmits them to the next node to calculate the topological potential value of the next scale area. For nodes with higher risk values, manual intervention is used to ensure the stability of the power grid operation.

[0011] 5. The cloud service platform and edge nodes establish a synchronization mechanism to achieve fault location and risk analysis from top to bottom. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A technical flow chart of real-time monitoring of power Internet of Things data in a converter station scenario;

[0013] Figure 2 Signal harmonic analysis flow chart;

[0014] Figure 3 Node risk assessment flow chart. DETAILED DESCRIPTION

[0015] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation. According to an embodiment of the present invention, a real-time monitoring technology for power Internet of Things data in a converter station scenario is provided. Figure 1 The proposed technology flow chart of real-time monitoring of power Internet of Things data based on edge computing for this method includes:

[0016] Step 101, collect node working data and upload it to the edge computing node.

[0017] Specifically, several nodes in the area are selected for monitoring. The nodes can be a certain working point or a certain device in the power grid. For the power equipment or a certain circuit in the circuit system that needs to be monitored, the sensor is connected, and the three-phase voltage and current signals of the node are collected through synchronous sampling and frequency tracking technology, and the analog quantity is converted into a digital signal. After the collection is completed, the data is transmitted to the edge computing node for storage and the next step of calculation.

[0018] Step 102: The edge node performs a generalized S transform on each node signal to achieve harmonic identification and fault analysis.

[0019] Specifically, Figure 2 This is a flow chart for harmonic identification and fault analysis.

[0020] Step 201: The edge computing node performs a generalized S transform on the signal.

[0021] Specifically, this step only operates on the faulty device or other basic nodes for fault analysis and can also be used for real-time observation. The signal is represented as a discrete sequence a(p) (p = 0, T, 2T, ..., (N-1)T), where T is the sampling period, which is involved in synchronous sampling. Directly calculate the S transform S of the discrete a(p) T [m,n] is:

[0022]

[0023] Use the fast FT algorithm to directly calculate the S transform S F [m,n] is:

[0024]

[0025] Where: m,n=0,1,...,N-1; A(l) is the discrete Fourier transform of a(m). Therefore, the steps of discrete fast S transform are:

[0026] 1) Calculate the first column S[jT,0] of the S matrix when n=0.

[0027] 2) Calculate the Fourier transform A[n / (NT)] of a[kT] when n=1, 2, ..., N-1.

[0028] 3) Expand A[n / (NT)] in the frequency domain to A[(n+m) / (NT)].

[0029] 4) Calculate the frequency domain Fourier transform of the Gaussian window function.

[0030] 5) Calculate the product of the frequency domain expanded dimension array A[(n+m) / (NT)] with m=0,1,...,N-1 and the Fourier transform of the frequency domain Gaussian window function.

[0031] 6) Perform a discrete inverse Fourier transform on the product.

[0032] 7) Repeat steps 5) and 6) until all frequency points are calculated.

[0033] After the signal has been transformed by S, it is easy to determine the type of fault in the power grid, including power grid short circuit, short circuit and impact fault. It can accurately detect faulty equipment in the power grid and actively report the fault to the edge computing node. If there is a disturbance signal in the power grid, the S transform can accurately extract the frequency of the disturbance signal. In the process of extracting the frequency characteristics of the disturbance signal, when the frequencies of the two disturbance signals are similar, frequency overlap will occur when the frequency characteristics are directly identified using the S transform two-dimensional contour map, which is not conducive to the extraction of frequency characteristics. In order to improve the time-frequency resolution of the signal, adjustable factors α and β are introduced in the S transform. The generalized S transform is defined as:

[0034]

[0035] α is the Gaussian window amplitude stretch factor; β is the frequency scale stretch factor. When and only when α=1, β=1, the generalized S transform is the S transform; when α>1 or β>1, the signal time resolution is improved while the frequency resolution is reduced; when α<1 or β<1, the signal frequency resolution is improved while the time resolution is reduced. By adjusting the values ​​of the two, the signal harmonic distribution can be more comprehensively reflected. In the subsequent result visualization process, the results of different factors can be presented in the same spectrum diagram.

[0036] Step 202: input into the fuzzy SOM neural network to analyze the disturbance.

[0037] Specifically, any trained neural network or any other means can be used to analyze the disturbance type. In the present invention, the fuzzy SOM neural network is taken as an example. The square, mean, and characteristic frequency points of the transformed time amplitude are taken as input samples of the neural network, and the neural network is trained in the cloud server. The edge computing node performs data preprocessing, and when a fault is found in the system or the risk of the current node is high in the risk analysis, the trained model sent from the cloud server is called when the disturbance analysis is performed on the current node. After being put into use, the fuzzy SOM neural network can qualitatively and quantitatively classify the superposition of multiple voltage disturbances and accurately reflect the actual situation of the circuit disturbance. The fuzzy SOM neural network uses the following rules to determine the competition layer structure and fuzzy clustering rules respectively.

[0038] The SOM neural network consists of an input layer, a competitive layer, a fuzzy recognition layer, and an output layer. The determination of the competitive layer is very important for the recognition of the perturbation features of the input. Even if the number of times the network is trained is increased, the classification performance of the network cannot be completely changed.

[0039] Specifically, there are two empirical formulas for determining the number of competitive layer structures:

[0040] 1) p1×p2>k, where k is the total number of sample categories, p1 is the number of row units in the competition layer, and p2 is the number of column units in the competition layer.

[0041] 2) or Where m is the number of output neurons, n is the number of input units, and z is a constant between [1,10].

[0042] Optionally, an exhaustive number of structures is constructed, the perturbation sample data of the test array is performed on the competing structures in all cases, and the structure with the best performance is selected.

[0043] For fuzzy clustering rules, the fuzzy SOM neural network needs to perform perturbation classification on the perturbation data. Assume that the ideal output of the SOM network is G, and the actual output of the test data is g. When the distance between the two outputs is the smallest, the perturbation classification result can be output, that is:

[0044]

[0045] Among them, g i is the clustering result closest to the ideal output G. N is the total number of clusters.

[0046] For the superposition of two adjacent disturbances, the primary and secondary disturbances can be determined by the minimum distance, that is,

[0047]

[0048] For more interference situations, clustering rules can be determined similarly.

[0049] The edge service node collects data, cleans, filters, and labels the data, and uploads it to the cloud server for model training. After the training is completed, the model parameters are sent to the edge computing node. During the application process, the cloud server still trains and updates the model parameters, and regularly sends new parameters for use.

[0050] Step 203: visualize the analysis results and control the circuit according to the results.

[0051] Specifically, after the S transform, a set of two-dimensional vectors is obtained, with the dimensions of time and frequency. For the time-frequency diagram, different colors are used to represent the amplitude. Therefore, if a certain frequency harmonic exceeds the preset value, its color is set to a warning color for warning; for the spectrum diagram, in the generalized S transform, by adjusting the values ​​of α and β, the transformation results under different time resolutions and frequency resolutions can be displayed, thereby avoiding the phenomenon of poor feature extraction caused by frequency overlap to a large extent. Help decision makers locate abnormal harmonic frequencies. According to the actual scenario, the fuzzy SOM neural network can determine the type of disturbance existing in the circuit and the percentage of disturbance they occupy, helping decision makers locate the source of the problem more quickly.

[0052] The time-frequency diagram and spectrum diagram help locate abnormal time-frequency points, and the disturbance analysis results help locate the source of the problem. Combining the two, real-time analysis and monitoring of circuit harmonics can be achieved, and decision makers can control the power grid, eliminate the source of the problem, and ensure the safety of the power grid.

[0053] Step 103, establish a node interaction graph in each area, calculate the node topological potential value, and integrate the data to send to the next level and cloud server.

[0054] Specifically, the edge node is responsible for the data integration and calculation of several nodes in a region. Several regions serve as new nodes to form the next level of regions, until the next level of the current region is the cloud server. For any region, if the data collected by the node is multi-source heterogeneous, the data needs to be integrated and normalized before it can be sent to the next level for the next step of calculation.

[0055] When testing multi-source heterogeneous data, multiple sensors need to be set between each parallel device group. Sensor data collection and management are very difficult. It is necessary to integrate the sensor differential data of each parallel device group to improve the efficiency of synchronous testing. Assume that the number of sensors is n and the sensor data set is {d1, d2, …, d n}, the sensor acquisition time is t, based on the probability data fusion theory, {λ1(t),λ2(t),…,λ n (t)} is the collection data set. The sensor collection accuracy is improved through weighted calculation. The weighted calculation result is:

[0056]

[0057] Among them, ω i For sensor weighting, it is important to mention that due to the rich types of nodes, the scale of data collected by sensors may be inconsistent, and mathematical methods are needed to determine the weighting values ​​to scale the data to the same scale for subsequent calculations.

[0058] Based on the theory of multivariate probability, {λ1(t),λ2(t),…,λ n (t)} should satisfy the n-dimensional normal distribution, and its distribution function is:

[0059]

[0060] The data set collected by the combined sensor {λ1(t),λ2(t),…,λ n (t)} can be expressed by the distribution density function of the weighted result λ:

[0061]

[0062] After completing sensor data fusion, the mathematical expectation value of multi-source heterogeneous data is:

[0063]

[0064] That is, the weighted sum of the mathematical expectations of each sensor.

[0065] The standard deviation is:

[0066]

[0067] Figure 3 The node risk assessment flow chart proposed by the present invention includes:

[0068] Step 301, create a risk map.

[0069] Specifically, according to the location and interaction relationship of each node in the actual power grid, the relationship between nodes is abstracted into a topological graph to obtain a graph G, that is, G = {V, E} represents the power system node risk graph, E is the set of edges, V = {v1, v2, ..., v n} is a graph node.

[0070] Step 302: Minimize the topological potential entropy to determine the impact factor.

[0071] Specifically, the interaction between topological potential field nodes is expressed as:

[0072]

[0073] in, is the inter-node effect; m i is the node quality, which can be represented by the working status of each node. The healthier the working status of the node, the higher the quality. The description process of the present invention adopts the difference between the actual mathematical expectation and the normal expectation of the node working voltage or other parameters, and the variance to represent it together; is the path length between nodes, which can be the actual distance between nodes, or the distance between edge nodes in each area, and can also be represented by the number of hops between nodes; σ is the node influence factor.

[0074] The topological potential of different nodes can be expressed as follows:

[0075]

[0076] The topological potential entropy H of the graph is:

[0077]

[0078] The more concentrated the number of nodes, the greater the influence of the node; the smaller the distance between nodes, the stronger their correlation and the greater the mutual influence between nodes.

[0079] To calculate the topological potential value of each node, it is necessary to first determine the node influence factor σ and obtain the influence factor by minimizing the potential entropy. Specifically, by traversal or other search algorithms, find the influence factor value corresponding to the minimum topological potential entropy as the optimal σ value.

[0080] Step 303, calculate the topological potential value, and maintain the circuit according to the result.

[0081] Specifically, after obtaining the optimal σ value, the topological potential value of each node is calculated according to formula (12). The larger the topological potential value, the greater the risk of the node. It is necessary to focus on nodes with higher risks or rectify them to avoid failure. On the other hand, if the circuit has failed, the node with the largest risk value can be searched to quickly locate the fault.

[0082] Step 104: Implement fault location and risk analysis from top to bottom.

[0083] Specifically, the topological potential values ​​of all nodes in the power Internet of Things are sent up by the edge nodes one level at a time and summarized to the cloud server. The cloud server proposes rectification suggestions for the higher-risk areas below, and finds the higher-risk areas or nodes under the corresponding areas, so that the risk nodes can be quickly located. If a power grid fails, the located risk node is the node with the highest probability of failure. The generalized S transform result and disturbance analysis result of the edge computing node corresponding to the node are called to analyze its fault type. If the node does not have a fault, the node with the second highest fault probability is found to perform the same operation. This method can quickly find high-risk nodes, thereby avoiding power grid failures, and can achieve rapid fault location and fault analysis after the failure occurs.

Claims

1. A real-time monitoring technology for power Internet of Things data in a converter station scenario, characterized in that: The content includes: collecting node working data and uploading it to the edge computing node; the edge node performs generalized S transform on the signal of each node to realize harmonic identification and fault analysis; the edge computing node performs generalized S transform on the signal; inputs it into the fuzzy SOM neural network to analyze the disturbance; visualizes the analysis results, and controls the circuit according to the results; establishes a node interaction diagram in each area, calculates the node topological potential value, integrates the data and sends it to the next level and cloud server; establishes a risk diagram; minimizes the topological potential entropy to determine the influencing factor; calculates the topological potential value, and maintains the circuit according to the result; realizes fault location and risk analysis from top to bottom.

Citation Information

Cited By

  • Power distribution network intelligent switch control method and system considering communication fault

    CN121036357A

  • A power distribution network intelligent switch control method and system considering communication failure

    CN121036357B