Transformer station equipment full life cycle management platform based on edge calculation

Abstract

The invention discloses a transformer substation equipment full life cycle management platform based on edge calculation, and relates to the technical field of transformer substation equipment full life cycle management, and the transformer substation equipment full life cycle management platform comprises a network flow monitoring module, a flow data modeling module, a surge feature extraction module, a surge risk prediction module and a dynamic frequency limiting control module. The network traffic monitoring module is used for monitoring traffic information of all incoming and outgoing data packets of the edge nodes through a lightweight network monitoring unit deployed on each edge node and establishing a comprehensive network traffic visual channel; and the traffic data modeling module is used for constructing the data packet traffic information acquired in real time into structured data by a short-time sliding window mechanism. According to the invention, by deploying a network monitoring and traffic prediction mechanism at the edge node, real-time pre-judgment and active frequency limiting of traffic surge are realized, burst data pressure is effectively relieved, key data are ensured not to be lost, the accuracy of fault early warning and the system stability are improved, and intelligent management of substation equipment in a full life cycle is assisted.

Description

Technical Field

[0001] The present invention relates to the technical field of full life cycle management of substation equipment, and specifically to a full life cycle management platform for substation equipment based on edge computing. Background Art

[0002] Edge computing-based substation equipment lifecycle management refers to the deployment of edge nodes with computing and analytical capabilities at the substation site to perform on-site real-time processing and intelligent analysis of data (such as condition monitoring data, fault records, and environmental information) generated by the equipment during various lifecycle stages, including operation, maintenance, overhaul, and decommissioning. Edge computing enables low-latency, high-reliability local decision-making and feedback, effectively reducing the pressure on data backhaul centers. It also supports key links such as accurate diagnosis of equipment operating status, risk prediction, maintenance optimization, and lifespan assessment, thereby improving equipment operational safety, extending service life, reducing operation and maintenance costs, and building a more intelligent, autonomous, and efficient equipment lifecycle management system.

[0003] The existing technology has the following deficiencies:

[0004] In existing edge computing-based substation equipment lifecycle management solutions, when faced with emergencies (such as large-scale equipment startup, short-term electrical disturbances, or lightning-induced partial discharge anomalies) that cause multiple sensing devices to simultaneously upload large amounts of raw data in a short period of time, a single edge node is prone to rapidly filling up its cache space due to limited local cache resources (such as RAM or temporary storage areas). Due to lagging processing capabilities or imperfect queuing mechanisms, data exceeding the cache capacity will be directly discarded, resulting in the early warning data generated by critical equipment at the beginning of the fault not being received and processed in a timely manner, making it impossible to complete status identification or trend prediction. This seriously affects the accuracy and timeliness of fault warnings, and there are technical risks such as permanent loss of abnormal information and missed warning windows, which restricts the practical value and stability of edge intelligence in ensuring the safe operation of power equipment.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a full life cycle management platform for substation equipment based on edge computing. By deploying a lightweight network monitoring unit at each edge node, combined with short-term sliding window structured modeling and machine learning-driven traffic surge prediction algorithm, a traffic control system with autonomous analysis, real-time prediction and dynamic intervention capabilities is constructed. When it is detected that the data traffic is about to surge, the frequency of data collection is actively reduced to suppress the explosive growth of unnecessary data volume from the source, significantly reducing the cache pressure and packet loss risk of edge nodes under sudden situations, thereby ensuring the complete retention of key symptom data and the normal operation of the fault warning model, further improving the edge intelligent system's recognition accuracy, response timeliness and operational stability of abnormal states of power equipment, and providing a strong support means for the intelligent management of substation equipment throughout its life cycle to solve the problems in the above-mentioned background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solutions: a full lifecycle management platform for substation equipment based on edge computing, including a network traffic monitoring module, a traffic data modeling module, a surge feature extraction module, a surge risk prediction module, and a dynamic frequency limiting control module;

[0008] The network traffic monitoring module deploys a lightweight network monitoring unit on each edge node, binds the network monitoring unit to all data communication interfaces of the edge node, monitors all incoming and outgoing data packet traffic information of the edge node, establishes a comprehensive network traffic visualization channel, and provides an accurate raw data source for traffic fluctuation analysis and dynamic prediction;

[0009] The traffic data modeling module, during the monitoring process, converts the real-time data packet traffic information (such as packet size, packet interval, transmission rate, source / destination port) into structured data using a short-term sliding window mechanism, forming a data traffic map and recording the multi-dimensional traffic behavior of the edge node;

[0010] The surge feature extraction module uses feature engineering technology to extract packet traffic surge features from each data traffic map. After in-depth analysis of the extracted features within a set short sliding window (such as 5 seconds or 10 seconds), the module quantifies the surge potential of short-term packet traffic bursts based on the analyzed features.

[0011] The surge risk prediction module inputs the analyzed features into a machine learning model that has been pre-trained with a large number of labeled samples. The machine learning model then predicts the risk of packet traffic surges at edge nodes in the short term.

[0012] The dynamic frequency limiting control module automatically reduces the sampling period of the edge node's data sampling terminal based on the prediction results when the machine learning model predicts that there is a risk of a surge in packet traffic at the edge node in the future, so as to dynamically suppress the sudden packet traffic and reduce traffic pressure at the source.

[0013] Preferably, the data packet flow information obtained in real time is constructed into structured data using a short-term sliding window mechanism to form a data flow map. The specific steps are as follows:

[0014] The data packet traffic is divided into time windows using a short-time sliding window mechanism, and the data packet traffic information within the same time window is classified as one analysis unit;

[0015] Preprocess and structure the data in each analysis unit, calculate and extract the statistical characteristics of data packet traffic information, and form structured traffic feature data with time series attributes;

[0016] The structured feature vectors of multiple consecutive short-term sliding windows are spliced ​​in chronological order to generate a complete data traffic map to describe the multi-dimensional network traffic behavior trajectory of edge nodes in continuous time periods.

[0017] Preferably, for each data traffic graph, a feature engineering technique is used to extract data packet traffic surge characteristics, wherein the extracted characteristics include the change in the time interval between adjacent data packets and the rate of change of the number of newly established TCP connections. Within a set short-term sliding window, after an in-depth analysis of the extracted characteristics, a packet interval shortening factor and a connection number growth factor are generated respectively, and the surge potential of a short-term burst of data packet traffic is quantified based on the packet interval shortening factor and the connection number growth factor.

[0018] Preferably, within the set short-time sliding window, the analyzed packet interval shortening factor and connection number growth factor are input as feature vectors into a machine learning model that has been pre-trained with a large number of labeled samples, and a traffic surge coefficient is generated based on the machine learning model. The traffic surge coefficient is used to predict the risk of data packet traffic surge at the edge node in a short period of time in the future.

[0019] Preferably, the traffic surge coefficient generated by the trained machine learning model when predicting the degree of data packet traffic surge of the edge node in a short period of time in the future is compared with a pre-set traffic surge coefficient reference threshold to predict the risk of a short-term burst of data packet traffic at the edge node. The specific steps are as follows:

[0020] If the traffic surge coefficient is greater than the traffic surge coefficient reference threshold, it is determined that the edge node data packet traffic has a risk of a short-term burst surge;

[0021] If the traffic surge coefficient is less than or equal to the traffic surge coefficient reference threshold, it is determined that there is no risk of a short-term burst surge in the data packet traffic of the edge node.

[0022] Preferably, when the machine learning model predicts that there is a risk of a surge in data packet traffic at the edge node in the future, the sampling period of the data sampling terminal of the edge node is automatically reduced according to the prediction result. The specific steps are as follows:

[0023] First, a dynamic sampling period adjustment coefficient is constructed based on the current predicted traffic surge coefficient to measure the degree to which the sampling frequency should be reduced. The calculation formula of the sampling period adjustment coefficient is as follows:

[0024]

[0025] , where: Traffic surge is the traffic surge coefficient currently output by the machine learning model, which represents the traffic surge intensity within the forecast period (short-term sliding window); T surge_ref is the reference threshold of the flow surge coefficient, which is used for normalization processing to make the results comparable to the model output; δ v is the nonlinear enhancement coefficient, which is used to control the response intensity and is usually in the range of [1, 5]. The tanh(·) function is used to smooth the nonlinear mapping between the flow surge degree and the adjustment response to prevent mutations. adj is the sampling adjustment coefficient, and its value range is (0, 1). The closer the value is to 1, the more obvious the reduction is.

[0026] In obtaining the sampling adjustment coefficient κ adj After that, the actual sampling period of the current edge node data sampling terminal is further updated, and the updated expression is:

[0027] Δt new =Δt base ·(1+λ m κ adj )

[0028] , where: Δt base The original basic sampling period; λ m To adjust the amplification factor, determine the extent to which the sampling period can be expanded under the maximum response condition. The recommended value is [0.5, 2.0]; Δt new The sampling frequency of the data sampling terminal is dynamically adjusted according to the new sampling period after adjustment.

[0029] Preferably, within a set short-time sliding window, the specific steps of generating a packet interval shortening factor after performing an in-depth analysis on the time interval changes between adjacent data packets of the edge node are as follows:

[0030] Set a short-term sliding window of fixed length, and record the timestamps of the consecutive data packets captured in the short-term sliding window as {τ1, τ2, ..., τ N}, calculate the time interval sequence between adjacent data packets as: Δ i =τ i+1 -τ i , i=1,2,…,N-1;

[0031] Next, construct the packet interval reduction trend sequence, and the constructed expression is:

[0032]

[0033] , where: ε is a small positive constant introduced to prevent division by zero or numerical oscillation (e.g. 10 -6 ), Λ i is the logarithmic ratio factor of the packet interval reduction, Λ i >0 means the packet interval is shortening, i <0 means the packet interval is getting longer;

[0034] The logarithmic ratio factors (i.e., Λ i >0) to perform cumulative growth mapping, introduce the surge trend enhancement function to construct the packet interval shortening factor, and the calculation expression of the packet interval shortening factor is:

[0035]

[0036] , where: Packet int represents the packet interval shortening factor; tanh(·) is the hyperbolic tangent function used for nonlinear enhancement, which responds smoothly to small shortening trends and significantly to large shortening trends; δ is the trend amplification coefficient, which is used to control the response sensitivity to the packet interval shortening amplitude, and δ>0; is the indicator function, when the packet interval is shortened (Λ i >0) takes 1, when the packet interval becomes longer (Λ i <0) takes 0.

[0037] Preferably, within a set short-term sliding window, the specific steps of generating a connection number growth factor after performing an in-depth analysis on the change rate of the number of new TCP connections of the edge node are as follows:

[0038] In a short-term sliding window of a set length, the number of new TCP connections in each time slice within the short-term sliding window is counted. The connection transition factor is defined as the relative increase in the number of new TCP connections in adjacent time slices. The specific calculation formula for the connection transition factor is:

[0039]

[0040] ,in: is the number of new TCP connections in the pth time slice; is the number of new TCP connections in the p-1th time slice; k is the growth sensitivity factor, which is used to enhance the mutation response capability, k>1; Λ (p) is the connection transition factor of the p-th time slice, which represents the increase in the number of new TCP connections from the p-1-th time slice to the p-th time slice;

[0041] The transition factor sequence {Λ (1) , Λ (2) ,…,Λ (n)} is constructed as a change trajectory, and the trend-enhanced logarithmic integral transformation function is applied to the transition factor sequence to generate the final connection number growth factor. The generation expression of the connection number growth factor is:

[0042]

[0043] , where n is the total number of time slices contained in the short-term sliding window; μ is the trend amplification coefficient, which is used to enhance the response to the cumulative change of the connection transition factor, μ>0; ln(·) is the natural logarithm function, which is used to compress the gradient fluctuation of extreme value growth; C cxi The connection growth factor is the core indicator for quantifying the potential for packet traffic surge.

[0044] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0045] The present invention deploys a lightweight network monitoring unit at each edge node, combines short-term sliding window structured modeling with machine learning-driven traffic surge prediction algorithms, and builds a traffic control system with autonomous analysis, real-time prediction, and dynamic intervention capabilities. When it detects that data traffic is about to surge, it actively reduces the frequency of data collection, suppresses the explosive growth of unnecessary data volume from the source, and significantly reduces the cache pressure and packet loss risk of edge nodes under sudden situations, thereby ensuring the complete retention of key symptom data and the normal operation of the fault warning model, further improving the edge intelligent system's recognition accuracy, response timeliness, and operational stability of abnormal states of power equipment, and providing a strong support means for the intelligent management of substation equipment throughout its life cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0047] Figure 1 This is a module diagram of the substation equipment full life cycle management platform based on edge computing in the present invention. DETAILED DESCRIPTION

[0048] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0049] The present invention provides Figure 1 The edge computing-based substation equipment full lifecycle management platform shown includes a network traffic monitoring module, a traffic data modeling module, a surge feature extraction module, a surge risk prediction module, and a dynamic frequency limiting control module;

[0050] The network traffic monitoring module deploys a lightweight network monitoring unit on each edge node, binds the network monitoring unit to all data communication interfaces of the edge node, monitors all incoming and outgoing data packet traffic information of the edge node, establishes a comprehensive network traffic visualization channel, and provides an accurate raw data source for traffic fluctuation analysis and dynamic prediction;

[0051] A lightweight network monitoring unit (NMU) is a type of tool or software module that uses minimal computing resources and has strong real-time performance. It is used to monitor and analyze data traffic. In the edge computing environment of a substation, common NMUs include:

[0052] Nethogs: A Linux-based network bandwidth monitoring tool that displays network traffic by process. It displays real-time bandwidth usage by process and is ideal for accurately tracking process-level data traffic.

[0053] iftop: Another Linux-based command-line tool that displays real-time statistics about network traffic. It shows the traffic and bandwidth consumption of currently active connections, making it easy to monitor network load in real time.

[0054] iptraf: A comprehensive network monitoring tool that provides traffic statistics and connection tracking. It displays information about IP packet traffic entering and leaving edge nodes, facilitating detailed traffic analysis.

[0055] Custom SNMP extension: Based on the standard SNMP (Simple Network Management Protocol) protocol, the customized monitoring module is used to obtain network interface information and traffic data of edge devices, providing flexible remote management and monitoring capabilities.

[0056] The lightweight nature of these monitoring units ensures that they can run stably on edge nodes without excessively occupying computing and storage resources. At the same time, they have the ability to efficiently and accurately obtain traffic data and support real-time network behavior analysis.

[0057] The data communication interface of an edge node refers to the data exchange channel between the edge computing device and the external network or other devices. In a substation environment, edge nodes generally involve the following main communication interfaces:

[0058] Ethernet interface: The most common wired communication method, typically used for high-bandwidth, high-stability communication between edge nodes and local area networks or core control systems. Ethernet interfaces support fast data transmission and high-volume data exchange, and are widely used in monitoring, control, and data transmission tasks.

[0059] Serial-to-TCP interface: Used to connect legacy devices or other devices whose interfaces do not support IP networks. This serial-to-TCP interface allows legacy serial devices to exchange data with edge computing devices through a conversion module. This is suitable for industrial devices that do not support IP protocols.

[0060] 5G communication modules: In scenarios requiring long-distance data transmission, such as substations, or data exchange between edge nodes and remote cloud platforms, 5G communication modules provide low-latency, wide-coverage, and high-speed wireless communication capabilities. The 5G interface enables edge nodes to reliably transmit real-time data even when they are far from fixed networks.

[0061] Wireless interfaces such as Wi-Fi or LoRa: In more complex application scenarios, edge nodes may also use wireless communication methods such as Wi-Fi or LoRa to transmit data to other devices or terminals. LoRa is generally used for low-power, long-distance transmission and is suitable for connecting low-data-rate devices in a large area.

[0062] The traffic data modeling module, during the monitoring process, converts the real-time data packet traffic information (such as packet size, packet interval, transmission rate, source / destination port) into structured data using a short-term sliding window mechanism, forming a data traffic map and recording the multi-dimensional traffic behavior of the edge node;

[0063] The real-time data packet traffic information is converted into structured data using a short-term sliding window mechanism to form a data traffic map. The specific steps are as follows:

[0064] The data packet traffic is divided into time windows using a short-time sliding window mechanism (e.g., 5 seconds or 10 seconds), and the data packets in the same time window are grouped as one analysis unit;

[0065] Through this short-term sliding window mechanism, the original data packet traffic data will be divided into subsets of multiple time periods, ensuring that the temporal and local characteristics of the data can be clearly displayed.

[0066] Preprocess and structure the data in each analysis unit, calculate and extract statistical features such as average transmission rate, maximum packet length, minimum packet interval, and source IP distribution density, and form structured traffic feature data with time series attributes;

[0067] This process generates traffic characteristic data with time series properties, providing accurate numerical support for subsequent behavioral analysis.

[0068] The structured feature vectors of multiple continuous sliding windows are spliced ​​in chronological order to generate a complete data traffic map to describe the multi-dimensional network traffic behavior trajectory of edge nodes in continuous time periods, providing a high-quality original data foundation for subsequent traffic surge identification and trend prediction.

[0069] The surge feature extraction module uses feature engineering technology to extract packet traffic surge features from each data traffic map. After in-depth analysis of the extracted features within a set short sliding window (such as 5 seconds or 10 seconds), the module quantifies the surge potential of short-term packet traffic bursts based on the analyzed features.

[0070] For each data traffic graph, feature engineering technology is used to extract the data packet traffic surge characteristics. The extracted features include the changes in the time interval between adjacent data packets and the change rate of the number of newly established TCP connections. After in-depth analysis of the extracted features within the set short-term sliding window, the packet interval shortening factor and connection number growth factor are generated respectively. Based on the packet interval shortening factor and connection number growth factor, the surge potential of short-term bursts of data packet traffic is quantified.

[0071] When the time intervals between adjacent data packets in the data traffic graph constructed for an edge node show an exponentially shortening trend, it usually indicates that the edge node has a significant potential for traffic surges. This is because the rapid compression of the time intervals between data packets reflects that the number of data packets transmitted per unit time is increasing dramatically, which means that the data flow is rapidly changing from a sparse state to a dense state. This exponential interval contraction is usually not caused by smooth or periodic traffic, but often occurs in scenarios such as batch startup of devices, outbreaks of abnormal event perception, or upstream network push storms. Therefore, this feature can be used as a precursor indicator of short-term traffic bursts, and is particularly suitable for identifying in advance the cache pressure or transmission bottlenecks that edge nodes will face, thereby triggering sampling period regulation or flow control strategies in a timely manner, and enhancing the system's response to sudden loads.

[0072] The specific steps for generating the packet interval reduction factor after in-depth analysis of the time interval changes between adjacent data packets at the edge node within a set short-time sliding window are as follows:

[0073] Set a short-term sliding window of fixed length, and record the timestamps of the consecutive data packets captured in the short-term sliding window as {τ1, τ2, ..., τ N}, calculate the time interval sequence between adjacent data packets as: Δ i =τ i+1 -τ i , i=1,2,…,N-1;

[0074] Next, construct the packet interval reduction trend sequence, and the constructed expression is:

[0075]

[0076] , where: ε is a small positive constant introduced to prevent division by zero or numerical oscillation (e.g. 10 -6 ), Λ i is the logarithmic ratio factor of the packet interval reduction, Λ i >0 means the packet interval is shortening, i <0 means the packet interval is getting longer;

[0077] The purpose of this step is to capture the continuous change trend of packet intervals in a logarithmic compression manner, remove the data scale dependency, and make the subsequent index construction more nonlinearly sensitive.

[0078] The logarithmic ratio factors (i.e., Λ i >0) to perform cumulative growth mapping, introduce the surge trend enhancement function to construct the packet interval shortening factor, and the calculation expression of the packet interval shortening factor is:

[0079]

[0080] , where: Packet int represents the packet interval shortening factor; tanh(·) is the hyperbolic tangent function used for nonlinear enhancement, which responds smoothly to small shortening trends and significantly to large shortening trends; δ is the trend amplification coefficient, which is used to control the response sensitivity to the packet interval shortening amplitude, and δ>0; is the indicator function, when the packet interval is shortened (Λ i >0) takes 1, when the packet interval becomes longer (Λ i <0) takes 0;

[0081] The larger the resulting packet interval reduction factor, the more significantly the time interval between consecutive data packets decreases, reflecting the potential for significant short-term surges in current data packet traffic. This packet interval reduction factor, as one of the input features of the prediction model, can be used to identify potential traffic storms at edge nodes at an early stage.

[0082] It can be seen from the packet interval shortening factor that within the set short-term sliding window, the packet interval shortening factor generated after in-depth analysis of the changes in the time interval between adjacent data packets of the edge node is larger, indicating that the edge node has a higher potential for a short-term surge in data packet traffic. Conversely, it indicates that the edge node has a lower potential for a short-term surge in data packet traffic. The reason is that the packet interval shortening factor amplifies and accumulates the logarithmic ratio of the time difference between consecutive packets and is only calculated when the packet interval continues to shorten. Therefore, the packet interval shortening factor can sensitively capture the growth trend of data flow density per unit time. This increase in density is usually a precursor to the outbreak of burst traffic (such as event triggering, concurrent reporting, and abnormal linkage), thereby effectively reflecting the risk level of traffic surge.

[0083] For each edge node, the higher the rate of change in the number of newly established TCP connections, the higher the potential for a short-term burst of data packet traffic faced by the edge node. The reason is that the establishment of a TCP connection usually means that new data will be transmitted. Especially in the edge node environment, each new connection often corresponds to a new device report, service request or data return task. When the number of newly established TCP connections per unit time increases rapidly, it reflects a sudden increase in the access pressure of the edge node and a surge in the number of communication sessions, which in turn leads to a significant increase in data packet concurrency and total traffic. Especially in the full life cycle management scenario of substation equipment, this surge in connections is often accompanied by a short-term upload of a large amount of raw data (such as equipment anomaly triggering, batch status collection, etc.). Therefore, the rate of change in the number of newly established connections can be used as an effective precursor indicator of sudden traffic surges, and can be used as a pre-trigger condition for real-time traffic risk prediction of edge nodes.

[0084] The specific steps for generating a connection growth factor after in-depth analysis of the change rate of the number of new TCP connections at the edge node within a set short sliding window are as follows:

[0085] In a short-term sliding window of a set length, the number of new TCP connections in each time slice within the short-term sliding window is counted. The connection transition factor is defined as the relative increase in the number of new TCP connections in adjacent time slices. The specific calculation formula for the connection transition factor is:

[0086]

[0087] ,in: is the number of new TCP connections in the pth time slice; is the number of new TCP connections in the p-1th time slice; κ is the growth sensitivity factor used to enhance the mutation response capability, κ>1; Λ (p) is the connection transition factor of the p-th time slice, which represents the increase in the number of new TCP connections from the p-1-th time slice to the p-th time slice;

[0088] This step uses a nonlinear power function to highlight the sudden change in the number of new TCP connections in the time series, which is particularly suitable for capturing local surge trends.

[0089] The transition factor sequence {Λ (1) , Λ (2) ,…,Λ (n)} is constructed as a change trajectory, and the trend-enhanced logarithmic integral transformation function is applied to the transition factor sequence to generate the final connection number growth factor. The generation expression of the connection number growth factor is:

[0090]

[0091] , where n is the total number of time slices contained in the short-term sliding window; μ is the trend amplification coefficient, which is used to enhance the response to the cumulative change of the connection transition factor, μ>0; ln(·) is the natural logarithm function, which is used to compress the gradient fluctuation of extreme value growth; C cxi The connection growth factor is the core indicator for quantifying the potential for packet traffic surges.

[0092] This step uses the square of the quadratic increment of each jump transition value for accumulation and controls the growth rate through logarithmic transformation, so that the connection number growth factor is stable when there is a small change, but rises rapidly when there is a local surge. It has a highly nonlinear response capability and is suitable as a strong response quantitative indicator of the surge trend in the early warning system.

[0093] The connection growth factor, generated by deeply analyzing the rate of change in the number of new TCP connections at an edge node within a specified short-term sliding window, shows that a larger value indicates a higher potential for a short-term surge in packet traffic at that edge node. Conversely, a smaller value indicates a lower potential for a short-term surge in packet traffic at that edge node. This is because the connection growth factor is constructed by nonlinearly amplifying and accumulating the relative amplitude of the jump in the number of new TCP connections within a short-term sliding window, making it sensitive to dramatic changes in the number of connections over a short period of time. In edge computing environments, the establishment of each TCP connection typically implies a subsequent data communication process. Therefore, the rapid increase in a large number of connections inevitably leads to a simultaneous increase in the number of packets, transmission rate, and network pressure. The connection growth factor, through trend reinforcement and logarithmic transformation mechanisms, effectively captures the cumulative intensity of this bursty connection behavior, making it a highly responsive indicator for measuring the potential for data traffic surges. Therefore, a higher connection growth factor indicates that the edge node has experienced more dramatic changes in connection growth within the current time window, reflecting a higher risk of traffic surges. Conversely, a lower connection growth factor indicates a relatively stable network.

[0094] The surge risk prediction module inputs the analyzed features into a machine learning model that has been pre-trained with a large number of labeled samples. The machine learning model then predicts the risk of packet traffic surges at edge nodes in the short term.

[0095] Within the set short-time sliding window, the analyzed packet interval shortening factor and connection number growth factor are input as feature vectors into the machine learning model that has been pre-trained with a large number of labeled samples. The traffic surge coefficient is generated based on the machine learning model, and the risk of data packet traffic surge at the edge node in the short term in the future is predicted through the traffic surge coefficient.

[0096] A trained machine learning model refers to an artificial intelligence model that has the ability to predict future trends in network traffic changes by learning, extracting features, and summarizing rules from actual traffic surge behaviors in historical network traffic data. This model is not a statically configured rule system, but rather actively builds a functional mapping relationship between traffic change characteristics and surge risks through an iterative training process on a large amount of labeled sample data, enabling it to have the ability to judge, classify, and quantify the output of new input data. In this technical solution, the machine learning model uses the packet interval shortening factor and the connection number growth factor as input feature variables, and through methods such as deep learning or ensemble learning, forms a prediction output mechanism for the traffic surge coefficient with the support of training on a large amount of labeled data to support real-time intelligent regulation of edge nodes.

[0097] Specifically, the training process involves the following key steps: First, a high-quality, structured training dataset must be constructed. This dataset consists of network behavior data collected from historical edge node operations. Each traffic behavior sample in this data is labeled (e.g., "normal traffic," "mild surge," "severe surge," etc.) using manual or automated algorithms. These labels represent the actual traffic status at the time. Once the data is prepared, feature engineering methods are used to extract key feature parameters, such as the packet interval reduction factor (describing the changing trend of the interval between packets) and the connection growth factor (characterizing the increase in the number of new connections per unit time), which serve as input feature vectors for the model. During the training phase, these feature vectors and their corresponding traffic surge labels are fed into a machine learning algorithm, such as an LSTM (Long Short-Term Memory) network, a GRU (Gated Recurrent Unit), XGBoost, LightGBM, or a support vector machine (SVM). The training algorithm continuously adjusts the model's internal weights, parameters, and structure to ensure that the model's output, given the input features, is as consistent as possible with the true labels, thereby minimizing the loss function and improving the model's generalization ability in prediction tasks.

[0098] After training is complete and cross-validation and test set evaluation are performed to ensure the model's accuracy, stability, and robustness, the model can be deployed in a live environment, becoming a "trained machine learning model." During the actual deployment phase, new data collected by edge nodes undergoes the same feature extraction process to generate real-time feature vectors, which are then fed into the model. Based on its learned decision boundaries and pattern recognition capabilities, the model outputs a numerical value representing the trend of traffic surges, known as the traffic surge coefficient. This traffic surge coefficient is typically a continuous variable that quantifies the degree of traffic surges that edge nodes may experience in the near future. Higher values ​​indicate a greater risk of future traffic surges.

[0099] The machine learning model is not specifically limited here, and can achieve the goal of shortening the packet interval factor Packet int and the connection growth factor C cxi Perform comprehensive analysis to generate traffic surge coefficient Traffic surge In order to realize the technical solution of the present invention, the present invention provides a specific implementation method; Traffic surge coefficient Traffic surge The generated expression is: Traffic surge =w f1 *Packet int +w f2 *C cxi , where w f1 、w f2Packet int and the connection growth factor C cxi The preset proportional coefficient, and w f1 、w f2 The preset proportional coefficient refers to the weighting factor w set artificially for the two core input indicators (packet interval shortening factor and connection number growth factor) when constructing the traffic surge coefficient. f1 With w f2 The role of these coefficients is to provide a basis for adjusting the weight distribution of different features in the calculation of comprehensive indicators. Since the contribution of the packet interval shortening factor and the connection number growth factor to the traffic surge may vary in different network environments, directly adding them may cause deviations. Therefore, the relative importance of them is controlled by setting proportional coefficients. For example, in some sudden situations, the impact of the increase in the number of connections on traffic changes is more significant, so the w f2 value, thereby increasing the number of connections by a factor of C cxi The preset scaling factor is typically obtained through historical data fitting, expert experience, or training set optimization. During actual deployment, it is either fixed or dynamically adjusted based on policy to improve the model's sensitivity to traffic surge trends and its accuracy.

[0100] It can be seen from the traffic surge coefficient that within the set short-term sliding window, the larger the performance value of the packet interval shortening factor generated after in-depth analysis of the changes in the time interval between adjacent data packets of the edge node, the larger the performance value of the connection number growth factor generated after in-depth analysis of the change rate of the number of newly established TCP connections of the edge node. In other words, the larger the performance value of the traffic surge coefficient generated when the trained machine learning model predicts the degree of data packet traffic surge of the edge node in a short period of time in the future, the higher the risk of a short-term burst of data packet traffic surge at the edge node, and vice versa.

[0101] The traffic surge coefficient generated by the trained machine learning model when predicting the degree of packet traffic surge at the edge node in the short term is compared with the pre-set traffic surge coefficient reference threshold to predict the risk of a short-term burst of packet traffic at the edge node. The specific steps are as follows:

[0102] If the traffic surge coefficient is greater than the traffic surge coefficient reference threshold, it is determined that the edge node data packet traffic has a risk of a short-term burst surge;

[0103] If the traffic surge coefficient is less than or equal to the traffic surge coefficient reference threshold, it is determined that there is no risk of a short-term burst surge in the data packet traffic of the edge node.

[0104] The dynamic frequency limiting control module automatically reduces the sampling period of the edge node's data sampling terminal based on the predicted result when the machine learning model predicts a risk of a surge in packet traffic in the future. This dynamically suppresses the sudden packet traffic and reduces traffic pressure at the source.

[0105] When the machine learning model predicts that there is a risk of a surge in packet traffic at an edge node in the future, the sampling period of the data sampling terminal at the edge node is automatically reduced based on the prediction result. The specific steps are as follows:

[0106] First, a dynamic sampling period adjustment coefficient is constructed based on the current predicted traffic surge coefficient to measure the degree to which the sampling frequency should be reduced. The calculation formula of the sampling period adjustment coefficient is as follows:

[0107]

[0108] , where: Traffic surge is the traffic surge coefficient currently output by the machine learning model, which represents the traffic surge intensity within the forecast period (short-term sliding window); T surge_ref is the reference threshold of the flow surge coefficient, which is used for normalization processing to make the results comparable to the model output; δ v is the nonlinear enhancement coefficient, which is used to control the response intensity and is usually in the range of [1, 5]. The tanh(·) function is used to smooth the nonlinear mapping between the flow surge degree and the adjustment response to prevent mutations. adj is the sampling adjustment coefficient, and its value range is (0, 1). The closer the value is to 1, the more obvious the reduction is.

[0109] This step constructs a nonlinear growth function, allowing the system to fine-tune the sampling period when the traffic surge is small, and significantly extend the sampling period when the traffic surge is significant, thereby achieving an adaptive response to different risk levels and avoiding the problem of linear adjustment strategies being too aggressive or slow.

[0110] In obtaining the sampling adjustment coefficient κ adj After that, the actual sampling period of the current edge node data sampling terminal is further updated, and the updated expression is:

[0111] Δt new =Δt base ·(1+λ m κ adj )

[0112] , where: Δt base The original basic sampling period; λ m To adjust the amplification factor, determine the extent to which the sampling period can be expanded under the maximum response condition. The recommended value is [0.5, 2.0]; Δt newThe new sampling period after adjustment is used to dynamically adjust the sampling frequency of the data sampling terminal according to the value;

[0113] This step implements the specific behavior of "adaptive sampling frequency limiting," mapping traffic surge predictions to direct control of the physical sampling rhythm. This suppresses the data generation rate at the source, alleviating edge node cache pressure and data reporting channel load. Because the regulation is based on exponential smoothing and nonlinear enhancement strategies, it is highly responsive, stable, and predictively driven, making it particularly suitable for intelligent traffic control in high-density sensing environments.

[0114] A prediction-driven proactive rate limiting mechanism proactively controls data traffic growth at the source before it fully materializes, effectively mitigating the risks of buffer overflows, processing delays, and packet loss faced by edge nodes under high load. Specifically, when a machine learning model detects a sharp increase in network traffic in the near future, the system doesn't passively wait for the data storm to arrive. Instead, based on the prediction, it immediately implements a "sampling cycle reduction" control measure on the data collection terminals. This involves extending the interval between each sampling, thereby reducing the amount of data generated and reported per unit time. This approach, equivalent to "rate limiting" at the source of information generation, fundamentally reduces computing and cache pressure on edge nodes, preventing data overflows, critical data loss, or system overload caused by sudden data surges. Furthermore, compared to traditional fixed-threshold triggering strategies, this mechanism offers greater foresight, adaptability, and precision. It intelligently optimizes system resource scheduling efficiency without sacrificing core monitoring capabilities, making it a key strategic support for building highly robust edge intelligence systems.

[0115] The aforementioned edge computing-based full-lifecycle management platform solution effectively addresses the issue of critical data loss at edge nodes due to limited cache resources during emergencies. Its primary benefit lies in shifting data traffic management from a "post-event response" approach to a "pre-event prediction" approach. By deploying a lightweight network monitoring unit at each edge node and combining short-term sliding window structured modeling with a machine learning-driven traffic surge prediction algorithm, the platform builds a traffic control system capable of autonomous analysis, real-time prediction, and dynamic intervention. When an impending data traffic surge is detected, the system proactively reduces data collection frequency, suppressing the explosive growth of unnecessary data at the source. This significantly reduces cache pressure and packet loss risks at edge nodes during emergencies, ensuring the complete retention of critical symptom data and the proper functioning of fault warning models. This further improves the edge intelligent system's accuracy in identifying abnormal conditions in power equipment, the timeliness of its response, and operational stability, providing strong support for intelligent management of substation equipment throughout its lifecycle.

[0116] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0117] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0118] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. The substation equipment full life cycle management platform based on edge computing is characterized by: It includes network traffic monitoring module, traffic data modeling module, surge feature extraction module, surge risk prediction module and dynamic frequency limiting control module; The network traffic monitoring module monitors all incoming and outgoing data packet traffic information of edge nodes through a lightweight network monitoring unit deployed on each edge node, establishing a comprehensive network traffic visual channel; The traffic data modeling module converts the real-time data packet traffic information into structured data using a short-term sliding window mechanism, forming a data traffic map and recording the multi-dimensional traffic behavior of edge nodes. The surge feature extraction module uses feature engineering techniques to extract packet traffic surge features from each data traffic graph. Within a set short-term sliding window, the extracted features are deeply analyzed and the surge potential of packet traffic bursts is quantified based on the analyzed features. The surge risk prediction module inputs the analyzed features into a machine learning model that has been pre-trained with a large number of labeled samples. The machine learning model then predicts the risk of packet traffic surges at edge nodes in the short term. The dynamic frequency limiting control module automatically reduces the sampling period of the edge node data sampling terminal according to the prediction results when the machine learning model predicts that there is a risk of a surge in packet traffic at the edge node in the future, dynamically suppressing the sudden packet traffic.

2. The substation equipment full life cycle management platform based on edge computing according to claim 1 is characterized in that: The real-time data packet traffic information is converted into structured data using a short-term sliding window mechanism to form a data traffic map. The specific steps are as follows: The data packet traffic is divided into time windows using a short-time sliding window mechanism, and the data packet traffic information within the same time window is classified as one analysis unit; Preprocess and structure the data in each analysis unit, calculate and extract the statistical characteristics of data packet traffic information, and form structured traffic feature data with time series attributes; The structured feature vectors of multiple consecutive short-term sliding windows are spliced ​​in chronological order to generate a complete data traffic map to describe the multi-dimensional network traffic behavior trajectory of edge nodes in continuous time periods.

3. The substation equipment full life cycle management platform based on edge computing according to claim 1 is characterized in that: For each data traffic graph, feature engineering technology is used to extract the data packet traffic surge characteristics. The extracted features include the changes in the time interval between adjacent data packets and the change rate of the number of newly established TCP connections. After in-depth analysis of the extracted features within the set short-term sliding window, the packet interval shortening factor and connection number growth factor are generated respectively. Based on the packet interval shortening factor and connection number growth factor, the surge potential of short-term bursts of data packet traffic is quantified.

4. The substation equipment full life cycle management platform based on edge computing according to claim 3 is characterized in that: Within the set short-time sliding window, the analyzed packet interval shortening factor and connection number growth factor are input as feature vectors into the machine learning model that has been pre-trained with a large number of labeled samples. The traffic surge coefficient is generated based on the machine learning model, and the risk of data packet traffic surge at the edge node in the short term in the future is predicted through the traffic surge coefficient.

5. The substation equipment full life cycle management platform based on edge computing according to claim 4 is characterized in that: The traffic surge coefficient generated by the trained machine learning model when predicting the degree of packet traffic surge at the edge node in the short term is compared with the pre-set traffic surge coefficient reference threshold to predict the risk of a short-term burst of packet traffic at the edge node. The specific steps are as follows: If the traffic surge coefficient is greater than the traffic surge coefficient reference threshold, it is determined that the edge node data packet traffic has a risk of a short-term burst surge; If the traffic surge coefficient is less than or equal to the traffic surge coefficient reference threshold, it is determined that there is no risk of a short-term burst surge in the data packet traffic of the edge node.

6. The substation equipment full life cycle management platform based on edge computing according to claim 5 is characterized in that: When the machine learning model predicts that there is a risk of a surge in packet traffic at an edge node in the future, the sampling period of the data sampling terminal at the edge node is automatically reduced based on the prediction result. The specific steps are as follows: A dynamic sampling period adjustment coefficient is constructed based on the current predicted traffic surge coefficient to measure the degree of reduction in the sampling frequency. The calculation formula for the sampling period adjustment coefficient is as follows: ， Among them: Traffic surge is the traffic surge coefficient currently output by the machine learning model, indicating the traffic surge intensity during the forecast period; T surg e _ref is the reference threshold of the flow surge coefficient, used for normalization processing; δ v is the nonlinear enhancement coefficient, which is used to control the response intensity; κ adj is the sampling adjustment coefficient, and its value range is (0, 1); After obtaining the sampling adjustment coefficient k adj After that, the actual sampling period of the current edge node data sampling terminal is updated, and the updated expression is: Δt new =Δt base ·(1+λ m ·k adj ), Where: Δt base The original basic sampling period; λ m To adjust the amplification factor, determine the extent to which the sampling period can be expanded under the maximum response condition; Δt new The new sampling period after adjustment is used to dynamically adjust the sampling frequency of the data sampling terminal.

7. The substation equipment full life cycle management platform based on edge computing according to claim 3 is characterized in that: The specific steps for generating the packet interval reduction factor after in-depth analysis of the time interval changes between adjacent data packets at the edge node within a set short-time sliding window are as follows: Set a short-term sliding window of fixed length, and record the timestamps of the consecutive data packets captured in the short-term sliding window as {τ1, τ2, ..., τ N }, calculate the time interval sequence between adjacent data packets as: Δ i =τ i+1 -τ i , i=1,2,…,N-1; Construct a data packet interval reduction trend sequence, the constructed expression is: ， Among them: ε is a very small positive constant, Λ i is the logarithmic ratio factor of the packet interval reduction, Λ i >0 means the packet interval is shortening, i <0 means the packet interval is getting longer; All logarithmic ratio factors showing a shortening trend in the packet interval reduction trend sequence are cumulatively mapped, and a surge trend enhancement function is introduced to construct the packet interval shortening factor. The calculation expression of the packet interval shortening factor is: ， Among them: Packet int Indicates the packet interval shortening factor; δ is the trend amplification coefficient, which is used to control the response sensitivity to the packet interval shortening amplitude, δ>0; is an indicator function that takes the value 1 when the packet interval is shortened and takes the value 0 when the packet interval is lengthened.

8. The substation equipment full life cycle management platform based on edge computing according to claim 3 is characterized in that: The specific steps for generating a connection growth factor after in-depth analysis of the change rate of the number of new TCP connections at the edge node within a set short sliding window are as follows: In a short-term sliding window of a set length, the number of new TCP connections in each time slice within the short-term sliding window is counted. The connection transition factor is defined as the relative increase in the number of new TCP connections in adjacent time slices. The specific calculation formula for the connection transition factor is: ， in: is the number of new TCP connections in the pth time slice; is the number of new TCP connections in the p-1th time slice; κ is the growth sensitivity factor used to enhance the mutation response capability, κ>1; Λ (p) is the connection transition factor of the p-th time slice, which represents the increase in the number of new TCP connections from the p-1-th time slice to the p-th time slice; The transition factor sequence {Λ (1) , Λ (2) ,…,Λ (n) } is constructed as a change trajectory, and the final connection number growth factor is generated based on the transition factor sequence. The generation expression of the connection number growth factor is: ， Where: n is the total number of time slices contained in the short-term sliding window; μ is the trend amplification coefficient, which is used to enhance the response to the cumulative change of the connection transition factor, μ>0; C cxi is the connection growth factor.

Images