Edge computing node exception recovery method and system, electronic equipment and storage medium

Abstract

The invention provides an edge computing node exception recovery method and system, electronic equipment and a storage medium, and relates to the technical field of edge computing and intelligent control, and the method comprises the steps: obtaining a plurality of different operation parameter sequences of an edge computing node; performing smoothing processing on each operation parameter sequence to obtain a plurality of smoothing parameter sequences; inputting corresponding smooth operation parameter values in each smooth parameter sequence at the current moment into a preset classification model to obtain a current operation state; inputting each smoothing parameter sequence into a preset time sequence model to obtain a plurality of predicted operation parameter values, and inputting each predicted operation parameter value into a preset classification model to obtain a predicted operation state; if the current operation state or the predicted operation state is abnormal, acquiring an enhanced state vector; and determining a target recovery action instruction according to the enhanced state vector and sending the target recovery action instruction to the edge computing node. According to the invention, the abnormity identification precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of edge computing and intelligent control, in particular to an edge computing node abnormal recovery method and system, an electronic device and a storage medium. BACKGROUND

[0002] An edge computing node is a distributed computing device deployed near a data source, such as a power distribution terminal of a power distribution network, an Internet of Things gateway (IoU) of an industrial Internet of Things, and an intelligent monitoring terminal of a smart city. Edge computing nodes are generally operated in complex communication environments, frequent voltage and current fluctuations, and devices susceptible to high temperatures. Currently, edge computing nodes rely on traditional fixed thresholds or simple watchdog mechanisms for abnormal processing, which is not suitable for complex dynamic scenarios. Therefore, it is a technical problem to be solved to provide a feasible edge computing node abnormal recovery method in a complex dynamic environment. SUMMARY

[0003] Therefore, the purpose of the present application is to overcome the deficiencies in the prior art and provide an edge computing node abnormal recovery method, system, electronic device and storage medium. The present application provides the following technical solutions: In a first aspect, the present application provides an edge computing node abnormal recovery method, which comprises: obtaining a plurality of different running parameter sequences of an edge computing node within a preset time, wherein the running parameter sequence comprises a plurality of running parameter values collected within the preset time; performing smoothing processing on each running parameter sequence through a sliding window to obtain a plurality of smoothed parameter sequences, wherein the smoothed parameter sequence comprises a plurality of smoothed running parameter values; inputting the smoothed running parameter values corresponding to each smoothed parameter sequence at the current time into a preset classification model to obtain the current running state of the edge computing node; inputting each smoothed parameter sequence into a preset time sequence model to obtain a plurality of different predicted running parameter values, and inputting each predicted running parameter value into the preset classification model to obtain the predicted running state of the edge computing node; if the current running state is abnormal or the predicted running state is abnormal, then concatenating the smoothed running parameter values corresponding to each smoothed parameter sequence at the current time and each predicted running parameter value to obtain an enhanced state vector of the edge computing node; determining, according to the enhanced state vector, a target recovery action instruction from a plurality of candidate recovery action instructions by a reinforcement learning algorithm, and sending the target recovery action instruction to the edge computing node to control the edge computing node to perform a target recovery action.

[0004] In an embodiment, the method further comprises: respectively determining whether each of the predicted running parameter values belongs to a corresponding preset parameter value range; if any of the predicted running parameter values does not belong to the corresponding preset parameter value range, sending a light recovery instruction to the edge computing node to control the edge computing node to perform a light recovery action.

[0005] In an embodiment, after the target recovery action instruction is sent to the edge computing node, the method further comprises: reacquiring a plurality of running parameter values from the edge computing node, inputting each of the reacquired running parameter values into the preset classification model to determine a post-recovery running state, the post-recovery running state being a running state of the edge computing node after performing the target recovery action; determining a recovery reward value according to the post-recovery running state and a preset reward mechanism, the recovery reward value being used to measure a recovery effect of the edge computing node performing the target recovery action; determining a four-tuple data vector according to the current running state, the target recovery action, the recovery reward value, and the post-recovery running state; optimizing the reinforcement learning algorithm according to the four-tuple data vector.

[0006] In an embodiment, the preset reward mechanism comprises: if the post-recovery running state is normal, determining the recovery reward value as a first reward value; if the post-recovery running state is abnormal, determining at least one abnormal running parameter value from the reacquired plurality of running parameter values, and if the number of abnormal running parameter values is greater than or equal to a preset number threshold, determining the recovery reward value as a second reward value; if the number of abnormal running parameter values is less than the preset number threshold, determining the recovery reward value as a third reward value.

[0007] In an embodiment, the determining at least one abnormal running parameter value from the reacquired plurality of running parameter values comprises: respectively determining whether each of the running parameter values belongs to a corresponding preset running parameter value range, and determining the running parameter values that do not belong to the corresponding preset running parameter value range as the abnormal running parameter values.

[0008] In an embodiment, when the number of edge computing nodes is greater than 1, the method further comprises: obtaining an abnormality occurrence timestamp and a topological relationship of each edge computing node, determining an abnormality co-occurrence relationship between each edge computing node according to the abnormality occurrence timestamp and the topological relationship, and constructing an abnormality co-occurrence relationship graph between each edge computing node by taking a node as the edge computing node, an edge as the abnormality co-occurrence relationship, and an edge weight as a co-occurrence frequency; performing clustering division on each node in the abnormality co-occurrence relationship graph by using a community discovery algorithm; determining a key degree corresponding to each node respectively by using centrality calculation based on the abnormality co-occurrence relationship graph and the clustering division result; determining a recovery priority of each node according to each key degree.

[0009] In an embodiment, when the number of edge computing nodes is greater than 1, the method further comprises: for the i th edge computing node, splicing the smooth running parameter values respectively corresponding in each smooth parameter sequence at the current moment to obtain a current state vector of the i th edge computing node; determining a node similarity between the i th edge computing node and the j th edge computing node by using a similarity algorithm, i ≥ 1, j ≥ 1, and i ≠ j; if the node similarity is greater than a preset similarity threshold, issuing a risk warning prompt, the risk warning prompt being used to prompt abnormality checking on the i th edge computing node and the j th edge computing node.

[0010] In a second aspect, the application provides an edge computing node abnormality recovery system, characterized in that the system comprises: a data acquisition module configured to acquire a plurality of different running parameter sequences of an edge computing node within a preset time, wherein each running parameter sequence comprises a plurality of running parameter values collected within the preset time; a smoothing processing module configured to perform smoothing processing on each running parameter sequence by using a sliding window to obtain a plurality of smooth parameter sequences, wherein each smooth parameter sequence comprises a plurality of smooth running parameter values; a state determination module configured to input the smooth running parameter values respectively corresponding in each smooth parameter sequence at the current moment into a preset classification model to obtain a current running state of the edge computing node; The state prediction module is configured to input each of the smoothed parameter sequences into a preset time series model to obtain a plurality of different predicted running parameter values, and input each of the predicted running parameter values into the preset classification model to obtain a predicted running state of the edge computing node. The splicing module is configured to splice the smoothed running parameter values corresponding to each of the smoothed parameter sequences and the predicted running parameter values at the current time to obtain an enhanced state vector of the edge computing node if the current running state is abnormal or the predicted running state is abnormal. The recovery control module is configured to determine a target recovery action instruction from a plurality of candidate recovery action instructions according to the enhanced state vector through a reinforcement learning algorithm, and send the target recovery action instruction to the edge computing node to control the edge computing node to perform a target recovery action.

[0011] In a third aspect, the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program performs the edge computing node abnormal recovery method of the first aspect when running on the processor.

[0012] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program implements the edge computing node abnormal recovery method of the first aspect when executed by a processor.

[0013] The edge computing node abnormal recovery method, system, electronic device and storage medium provided by the application, by acquiring a plurality of different running parameter sequences of the edge computing node within a preset time, the running parameter sequence includes: a plurality of running parameter values collected within the preset time; by sliding window, each running parameter sequence is respectively smoothed to obtain a plurality of smoothed parameter sequences, the smoothed parameter sequence includes: a plurality of smoothed running parameter values; the smoothed running parameter values respectively corresponding to each smoothed parameter sequence at the current time are input into a preset classification model to obtain the current running state of the edge computing node; each smoothed parameter sequence is input into a preset time sequence model to obtain a plurality of different predicted running parameter values, and each predicted running parameter value is input into the preset classification model to obtain the predicted running state of the edge computing node; if the current running state is abnormal or the predicted running state is abnormal, the smoothed running parameter values respectively corresponding to each smoothed parameter sequence at the current time and each predicted running parameter value are spliced to obtain an enhanced state vector of the edge computing node; according to the enhanced state vector, a target recovery action instruction is determined from a plurality of candidate recovery action instructions through a reinforcement learning algorithm, and the target recovery action instruction is sent to the edge computing node to control the edge computing node to execute a target recovery action, which improves the abnormal identification precision, can prevent and control the abnormality in advance, quickly and efficiently recovers the node stability, reduces the business interference, and improves the reliability and intelligent management level of the edge computing node operation.

[0014] In order to make the above objectives, characteristics and advantages of the present application more apparent and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation to the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0016] Figure 1 A flowchart of an edge computing node abnormal recovery method provided by an embodiment of the application is shown; Figure 2 A structure diagram of an edge computing node abnormal recovery system provided by an embodiment of the application is shown; Figure 3 A structure diagram of an electronic device provided by an embodiment of the application is shown.

[0017] Main element symbol explanation: 200 - Edge computing node abnormal recovery system; 210 - Data acquisition module; 220 - Smoothing processing module; 230 - State determination module; 240 - State prediction module; 250 - Splicing module; 260 - Recovery control module; 300 - Electronic device; 301 - Transceiver; 302 - Processor; 303 - Memory. DETAILED DESCRIPTION

[0018] Embodiments of the present application are described in detail below with reference to examples illustrated in the accompanying drawings, in which the same or similar components have the same or similar designations throughout the various figures and identical or similar components have the same or similar reference characters. The embodiments described below with reference to the drawings are exemplary only, and are merely intended to explain the present application, and should not be understood as limiting the present application.

[0019] In addition, the terms "first", "second", etc. are used only for descriptive purposes and should not be construed as indicating or implying relative importance or an indicated number of technical features. Thus, features defined with "first", "second", etc. can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the template herein is only for the purpose of describing specific embodiments and is not intended to limit the present application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Embodiment 1 The embodiments of the present application provide a method for recovering an abnormal edge computing node, please refer to Figure 1 , the method comprises steps S110-S160.

[0022] Step S110, obtaining a plurality of different running parameter sequences of the edge computing node within a preset time, the running parameter sequence comprising a plurality of running parameter values collected within the preset time.

[0023] In this embodiment, the edge computing node is continuously obtained within a preset time, such as voltage, current, network delay, device temperature, log data, etc. All values collected in this time period are arranged into a sequence, for example, if the preset time is one minute, the voltage and current are collected every second, then 60 voltage values are formed into a voltage sequence, and 60 current values are formed into a current sequence. These running parameter sequences collectively reflect the multi-dimensional running state of the edge computing node within the preset time period.

[0024] Step S120: Smooth each of the running parameter sequences using a sliding window to obtain multiple smoothed parameter sequences, the smoothed parameter sequences including multiple smoothed running parameter values.

[0025] In this embodiment, for each type of operating parameter sequence collected by the edge computing node, such as voltage sequence and current sequence, a sliding window method is used to process the data to eliminate instantaneous fluctuations. That is, a fixed-size time window is set, such as a window containing five consecutive sampling points. The window is slid sequentially in time order, and the statistical results of the parameter values ​​in each window are calculated. Usually, the mean is used as the smoothed operating parameter value of the window. Finally, a smoothed parameter sequence consisting of multiple smoothed values ​​is generated for each original parameter sequence.

[0026] For example, given a voltage sequence of [220V, 222V, 219V, 225V, 221V, 223V], if the sliding window size is 3, then the first window calculates (220+222+219)÷3=220.3V, the second window calculates (222+219+225)÷3=222V, and so on, resulting in [220.3V, 222V, 221.7V], which is the smoothing parameter sequence corresponding to this voltage sequence, where each value is a smoothing operation parameter value within the window.

[0027] Step S130: Input the smoothing operation parameter values ​​corresponding to each smoothing parameter sequence at the current time into the preset classification model to obtain the current running state of the edge computing node.

[0028] In this embodiment, at the current time point, the smoothed running parameter values ​​corresponding to the current time are extracted from each parameter sequence after the sliding window smoothing process. These smoothed values ​​from different smoothed parameter sequences are integrated into a comprehensive feature vector, which is input into a preset classification model, namely a pre-trained support vector machine model (SVM). The preset classification model analyzes these multi-dimensional smoothed features and outputs the running status of the edge computing node at the current time, such as normal or abnormal.

[0029] For example, after smoothing, the smoothed voltage value at the current moment is 220.3V, the smoothed current value is 1.2A, and the smoothed network latency value is 85ms. These three values ​​are combined to form a feature vector [220.3V, 1.2A, 85ms], which is then input into a preset classification model. By combining the feature patterns of normal or abnormal states in historical training data, the current operating state of the edge computing node is ultimately determined to be normal. This process, by integrating multi-dimensional smoothed operating parameters, avoids the interference of fluctuations in a single operating parameter, making the state determination more accurate.

[0030] Step S140: Input each of the smoothing parameter sequences into a preset time series model to obtain multiple different predicted running parameter values. Input each of the predicted running parameter values ​​into the preset classification model to obtain the predicted running status of the edge computing node.

[0031] In this embodiment, for each smoothed sequence of operating parameters, such as voltage smoothing sequence and current smoothing sequence, a preset time series model is input, such as a pre-trained Autoregressive Integrated Moving Average (ARIMA), Exponentially Weighted Moving Average (EWMA), or Long Short-Term Memory (LSTM) model. By analyzing the historical trend of the parameters through the preset time series model, the predicted value of each parameter at a certain point in the future is obtained, i.e., multiple different predicted operating parameter values. Subsequently, these predicted operating parameter values ​​are integrated into a feature vector and input into a preset classification model to determine the operating status of the edge computing node at that point in the future.

[0032] Step S150: If the current running state is abnormal, or the predicted running state is abnormal, then the smoothing running parameter values ​​corresponding to each smoothing parameter sequence at the current time and each predicted running parameter value are concatenated to obtain the enhanced state vector of the edge computing node.

[0033] In this embodiment, when the current operating state of an edge computing node is determined to be abnormal, or when its future operating state is predicted to be abnormal, it is necessary to integrate two sets of parameter values ​​into an enhanced state vector. Specifically, this includes the smoothed operating parameter values ​​corresponding to each smoothed parameter sequence at the current time, and the future values ​​of each parameter predicted by the time series model. These two types of parameter values ​​are concatenated and combined in a certain order to form an enhanced state vector that can simultaneously reflect the current abnormal state and future abnormal trend of the edge computing node, providing a more comprehensive feature basis for the subsequent abnormal recovery strategy formulation.

[0034] Step S160: Based on the enhanced state vector, a target recovery action instruction is determined from multiple candidate recovery action instructions using a reinforcement learning algorithm, and the target recovery action instruction is sent to the edge computing node to control the edge computing node to execute the target recovery action.

[0035] In this embodiment, based on the constructed enhanced state vector, a reinforcement learning algorithm is used to filter and make decisions among a number of preset candidate recovery action instructions, such as link reset, cache clearing, and module restart, and finally determine the most effective target recovery action instruction. Then, the target recovery action instruction is sent to the edge computing node to control the edge computing node to execute the corresponding recovery action, such as executing link reset to reduce network latency, thereby resolving the current anomaly or preventing the occurrence of the predicted anomaly.

[0036] For example, if the augmented state vector indicates an abnormal current network latency of 120ms and predicts an abnormal latency of 130ms after 10 seconds, the reinforcement learning algorithm, combined with historical experience (such as the 90% success rate of link reset in similar states), selects link reset as the target recovery action from multiple candidate instructions. This instruction is then sent to the edge computing node, which executes the action to bring the network latency back to normal, achieving anomaly recovery. This process, through the dynamic decision-making of reinforcement learning, ensures the targeted and effective nature of the recovery action.

[0037] In one embodiment, after sending the target recovery action instruction to the edge computing node, the method further includes: re-acquiring multiple operating parameter values ​​from the edge computing node, inputting the re-acquiring operating parameter values ​​into the preset classification model to determine the post-recovery operating state, wherein the post-recovery operating state is the operating state of the edge computing node after performing the target recovery action; determining a recovery reward value based on the post-recovery operating state and a preset reward mechanism, wherein the recovery reward value is used to measure the recovery effect of the edge computing node performing the target recovery action; determining a four-tuple data vector based on the current operating state, the target recovery action, the recovery reward value, and the post-recovery operating state; and optimizing the reinforcement learning algorithm based on the four-tuple data vector.

[0038] In this embodiment, firstly, multiple operating parameter values ​​of the edge computing node are re-collected and input into a preset classification model to determine the post-recovery operating state of the edge computing node after performing a recovery action. Next, based on the post-recovery operating state and preset reward rules, such as high score for normal recovery and low score for no recovery, a recovery reward value is determined to measure the effectiveness of the recovery action. Then, the current operating state before the action, the target recovery action taken, the obtained recovery reward value, and the post-recovery operating state after the action are combined into a four-tuple data vector. Finally, this four-tuple data vector is used to update the parameters of the reinforcement learning algorithm, enabling it to learn from the experience gained during this recovery process, such as the effectiveness of choosing a certain action under a certain abnormal state, thereby improving the accuracy of decision-making when facing similar abnormalities in the future.

[0039] For example, if a node's original abnormal state is "high network latency", after executing a link reset, the re-collected network latency returns to normal, and it receives a reward of 10 points according to the reward mechanism, thus forming a quadruple [high network latency, link reset, 10, normal]. The reinforcement learning algorithm learns from this data that choosing link reset is effective when network latency is high, and will give priority to this action when encountering similar situations in the future.

[0040] The reinforcement learning algorithm is Q-learning, Deep Q-Network, or Actor-Critic. The reinforcement learning algorithm provided in this application learns online by accumulating a large number of quadruplet data vectors, so that the recovery strategy can continuously evolve to adapt to more scenarios and improve the overall self-recovery capability.

[0041] In one embodiment, the preset reward mechanism includes: if the restored operating state is normal, then the restoration reward value is determined to be a first reward value; if the restored operating state is abnormal, then at least one abnormal operating parameter value is determined from multiple reacquired operating parameter values; if the number of abnormal operating parameter values ​​is greater than or equal to a preset quantity threshold, then the restoration reward value is determined to be a second reward value; if the number of abnormal operating parameter values ​​is less than the preset quantity threshold, then the restoration reward value is determined to be a third reward value.

[0042] In this embodiment, the preset reward mechanism is a scoring rule used to measure the effect of the recovery action. Specifically, when the edge computing node performs the target recovery action, if the running status after recovery is determined to be normal, a first reward value is given; if it is still abnormal after recovery, the number of abnormal parameters in the re-collected running parameters is further counted. If the number of abnormal parameters reaches or exceeds a preset threshold, such as 3 or more, a second reward value is given; if the number of abnormal parameters is less than the preset threshold, such as 2, a third reward value is given.

[0043] For example: if the state is normal after recovery, the first reward value is 10 points; if it is still abnormal after recovery and there are 4 abnormal parameters, the second reward value is -5 points; if it is abnormal after recovery but only 1 parameter is abnormal, the third reward value is 3 points. This tiered reward can accurately reflect the effect of the action and help the reinforcement learning algorithm distinguish between three situations: effective recovery, partial improvement and complete ineffectiveness.

[0044] In one embodiment, determining at least one abnormal operating parameter value from a plurality of reacquired operating parameter values ​​includes: determining whether each of the operating parameter values ​​belongs to a corresponding preset operating parameter value range, and determining the operating parameter values ​​that do not belong to the corresponding preset operating parameter value range as the abnormal operating parameter values.

[0045] In this embodiment, when identifying abnormal parameters from multiple operating parameter values ​​re-acquired after the recovery action is performed, the following steps are required: a normal value range is pre-defined for each operating parameter, and each re-acquired parameter value is checked one by one to see if it falls within its corresponding preset normal range. Parameter values ​​that exceed the corresponding range are marked as abnormal operating parameter values.

[0046] For example, the reacquired operating parameter values ​​include a voltage of 250V, whose preset range is 198V-242V; a current of 1.2A, whose preset range is 0.8A-1.5A; and a network latency of 110ms, whose preset range is 0-100ms. It is determined that the voltage of 250V and the network latency of 110ms both exceed their respective preset ranges, so these two values ​​are identified as abnormal operating parameter values.

[0047] In one embodiment, the method further includes: determining whether each of the predicted operating parameter values ​​belongs to the corresponding preset parameter value range; if any of the predicted operating parameter values ​​does not belong to the corresponding preset parameter value range, then sending a gentle recovery command to the edge computing node to control the edge computing node to perform a gentle recovery action.

[0048] In this embodiment, after obtaining the predicted operating parameter values, each predicted value is checked one by one to see if it falls within its corresponding preset normal range, such as whether the predicted voltage is between 198V and 242V, or whether the predicted network latency is between 0 and 100ms. If any predicted parameter value exceeds the corresponding normal range, a gentle recovery command is sent to the edge computing node, such as link reset or cache clearing, which have less interference with the service. This controls the edge computing node to execute these gentle recovery actions in advance to avoid the predicted anomalies from actually occurring.

[0049] For example, if the network latency is predicted to be 110ms in 10 seconds, which is outside the preset range of 0-100ms, there is no need to wait for the actual occurrence of the anomaly. Instead, a gentle recovery command for link reset can be sent directly to the edge computing node. After the edge computing node executes the command, the network latency can be reduced in advance to prevent subsequent anomalies from occurring.

[0050] In one embodiment, when the number of edge computing nodes is greater than 1, the method further includes: Obtain the anomaly occurrence timestamp and topological relationship of each edge computing node; determine the anomaly co-occurrence relationship between each edge computing node based on the anomaly occurrence timestamp and the topological relationship; and construct an anomaly co-occurrence relationship graph between each edge computing node by using nodes to represent edge computing nodes, edges to represent the anomaly co-occurrence relationship, and edge weights to represent co-occurrence frequency. The community detection algorithm is used to cluster the nodes in the abnormal co-occurrence relationship graph. Based on the aforementioned abnormal co-occurrence relationship diagram and clustering results, the criticality of each node is determined by centrality calculation. The recovery priority of each node is determined based on its criticality.

[0051] In this embodiment, the timestamps of all node anomalies and network topology relationships are first collected. These two pieces of information are combined to determine the anomaly co-occurrence relationship between nodes, i.e., whether edge computing nodes with similar times and topological associations are synchronously abnormal. Then, an anomaly co-occurrence relationship graph is constructed, with each edge computing node as a node, the anomaly co-occurrence relationship as an edge, and the co-occurrence frequency as the edge weight. Next, a community detection algorithm is used to cluster the nodes in the graph, grouping nodes with close anomaly associations into the same group. Subsequently, based on the relationship graph and the clustering results, the criticality of each node is evaluated through centrality calculation, such as whether it is a core hub for anomaly propagation. Finally, based on the criticality of each node, the recovery priority when anomalies occur is determined. The higher the criticality, the higher the recovery priority, so that multi-node anomalies can be handled efficiently according to priority in the subsequent process.

[0052] In one embodiment, when the number of edge computing nodes is greater than 1, the method further includes: For the i-th edge computing node, the smoothing operation parameter values ​​corresponding to each smoothing parameter sequence at the current time are concatenated to obtain the current state vector of the i-th edge computing node; A similarity algorithm is used to determine the node similarity between the i-th edge computing node and the j-th edge computing node, where i≥1, j≥1, and i≠j; If the node similarity is greater than a preset similarity threshold, a risk warning is issued. The risk warning is used to prompt anomaly investigation for the i-th edge computing node and the j-th edge computing node.

[0053] In this embodiment, for each i-th edge computing node, the smoothing operation parameter values ​​corresponding to each smoothing parameter sequence at the current time are first concatenated to form a current state vector that reflects the current operating state of the node. Then, a similarity algorithm, such as cosine similarity or Euclidean distance, is used to calculate the node similarity between the i-th edge computing node and any j-th edge computing node, thereby measuring the similarity between the current operating states of the two nodes. If the calculated node similarity exceeds a preset similarity threshold, a risk warning will be issued to remind relevant personnel to promptly investigate the two highly similar nodes for anomalies, because edge computing nodes with similar states may have the same potential anomaly risks, and early investigation can prevent the spread or simultaneous occurrence of anomalies.

[0054] The edge computing node anomaly recovery method provided in this application involves acquiring multiple different operating parameter sequences of an edge computing node within a preset time period. These operating parameter sequences include multiple operating parameter values ​​collected within the preset time period. A sliding window is used to smooth each of the operating parameter sequences, resulting in multiple smoothed parameter sequences, each including multiple smoothed operating parameter values. The smoothed operating parameter values ​​corresponding to each of the smoothed parameter sequences at the current time are input into a preset classification model to obtain the current operating state of the edge computing node. Each smoothed parameter sequence is then input into a preset time series model to obtain multiple different predicted operating parameter values. Finally, these predicted operating parameter values ​​are input into the preset classification model. The predicted operating state of the edge computing node is obtained. If the current operating state is abnormal, or the predicted operating state is abnormal, the smoothed operating parameter values ​​corresponding to each smoothed parameter sequence at the current time, and each predicted operating parameter value are concatenated to obtain the enhanced state vector of the edge computing node. Based on the enhanced state vector, a target recovery action instruction is determined from multiple candidate recovery action instructions through a reinforcement learning algorithm, and the target recovery action instruction is sent to the edge computing node to control the edge computing node to execute the target recovery action. This improves the accuracy of anomaly identification, enables early prevention and control of anomalies, and quickly and efficiently restores node stability. It also reduces business interference and improves the reliability and intelligent management level of edge computing node operation.

[0055] Example 2 In addition, please see Figure 2 This application embodiment also provides an edge computing node anomaly recovery system 200, including: Data acquisition module 210 is used to acquire multiple different operating parameter sequences of edge computing nodes within a preset time period, and each of the operating parameter sequences includes: multiple operating parameter values ​​acquired within the preset time period; Smoothing module 220 is used to smooth each of the running parameter sequences through a sliding window to obtain multiple smoothed parameter sequences, wherein the smoothed parameter sequences include multiple smoothed running parameter values; The state determination module 230 is used to input the smoothing operation parameter values ​​corresponding to each smoothing parameter sequence at the current time into a preset classification model to obtain the current running state of the edge computing node; The state prediction module 240 is used to input each of the smoothing parameter sequences into a preset time series model to obtain multiple different predicted running parameter values, and input each of the predicted running parameter values ​​into the preset classification model to obtain the predicted running state of the edge computing node. The splicing module 250 is used to splice the smoothing operation parameter values ​​corresponding to each smoothing parameter sequence at the current time and the predicted operation parameter values ​​to obtain the enhanced state vector of the edge computing node if the current running state is abnormal or the predicted running state is abnormal. The recovery control module 260 is used to determine the target recovery action instruction from multiple candidate recovery action instructions based on the enhanced state vector using a reinforcement learning algorithm, and send the target recovery action instruction to the edge computing node to control the edge computing node to execute the target recovery action.

[0056] The edge computing node anomaly recovery system 200 provided in this application embodiment can execute the edge computing node anomaly recovery method provided in the above method embodiment 1. To avoid repetition, it will not be described again here.

[0057] Example 3 Furthermore, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the computer program executes the edge computing node anomaly recovery method provided in Embodiment 1 when running on the processor.

[0058] For details, please see Figure 3 The electronic device 300 includes a transceiver 301, a bus interface, and a processor 302. The processor 302 is configured to acquire multiple different operating parameter sequences of an edge computing node within a preset time period. Each operating parameter sequence includes multiple operating parameter values ​​collected within the preset time period. The processor 302 then smooths each operating parameter sequence using a sliding window to obtain multiple smoothed parameter sequences, each smoothed parameter sequence including multiple smoothed operating parameter values. Finally, the processor 302 inputs the smoothed operating parameter values ​​corresponding to each smoothed parameter sequence at the current time into a preset classification model to obtain the current operating state of the edge computing node. The processor 302 also inputs each smoothed parameter sequence into a preset time sequence. The model is used to obtain multiple different predicted operating parameter values. Each predicted operating parameter value is input into the preset classification model to obtain the predicted operating state of the edge computing node. If the current operating state is abnormal, or the predicted operating state is abnormal, the smoothed operating parameter values ​​corresponding to each smoothed parameter sequence at the current time and each predicted operating parameter value are concatenated to obtain the enhanced state vector of the edge computing node. Based on the enhanced state vector, a target recovery action instruction is determined from multiple candidate recovery action instructions through a reinforcement learning algorithm, and the target recovery action instruction is sent to the edge computing node to control the edge computing node to execute the target recovery action.

[0059] In this embodiment of the invention, the electronic device 300 further includes a memory 303.Figure 3 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 302) and memory (memory 303). The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 301 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. The processor 302 is responsible for managing the bus architecture and general processing, and the memory 303 can store data used by the processor 302 during operation.

[0060] The electronic device 300 provided in this application embodiment can execute the edge computing node anomaly recovery method provided in the above method embodiment 1. To avoid repetition, it will not be described again here.

[0061] Example 4 Furthermore, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the edge computing node anomaly recovery method provided in Embodiment 1.

[0062] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0063] The computer-readable storage medium provided in this embodiment can implement the edge computing node anomaly recovery method provided in Embodiment 1. To avoid repetition, it will not be described again here.

[0064] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0065] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0066] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for edge computing node anomaly recovery, characterized in that, The method comprises: acquiring a plurality of different running parameter sequences of an edge computing node within a preset time, the running parameter sequence comprising a plurality of running parameter values collected within the preset time; respectively smoothing each of the running parameter sequences through a sliding window to obtain a plurality of smoothed parameter sequences, the smoothed parameter sequence comprising a plurality of smoothed running parameter values; inputting the smoothed running parameter values respectively corresponding to each of the smoothed parameter sequences at a current time into a preset classification model to obtain a current running state of the edge computing node; inputting each of the smoothed parameter sequences into a preset time sequence model to obtain a plurality of different predicted running parameter values, and inputting each of the predicted running parameter values into the preset classification model to obtain a predicted running state of the edge computing node; if the current running state is abnormal or the predicted running state is abnormal, then splicing the smoothed running parameter values respectively corresponding to each of the smoothed parameter sequences at the current time and each of the predicted running parameter values to obtain an enhanced state vector of the edge computing node; determining a target recovery action instruction from a plurality of candidate recovery action instructions through a reinforcement learning algorithm according to the enhanced state vector, and sending the target recovery action instruction to the edge computing node to control the edge computing node to perform a target recovery action.

2. The edge computing node anomaly recovery method of claim 1, wherein, The method further comprises: respectively determining whether each of the predicted running parameter values belongs to a corresponding preset parameter value range; if any of the predicted running parameter values does not belong to the corresponding preset parameter value range, then sending a light recovery instruction to the edge computing node to control the edge computing node to perform a light recovery action.

3. The edge computing node anomaly recovery method of claim 1, wherein, After the target recovery action instruction is sent to the edge computing node, the method further comprises: reacquiring a plurality of running parameter values from the edge computing node, inputting each of the reacquired running parameter values into the preset classification model to determine a post-recovery running state, the post-recovery running state being a running state of the edge computing node after performing the target recovery action; determining a recovery reward value according to the post-recovery running state and a preset reward mechanism, the recovery reward value being used to measure a recovery effect of the edge computing node performing the target recovery action; determining a four-tuple data vector according to the current running state, the target recovery action, the recovery reward value, and the post-recovery running state; optimizing the reinforcement learning algorithm according to the four-tuple data vector.

4. The edge computing node anomaly recovery method of claim 3, wherein, The preset reward mechanism comprises: if the post-recovery running state is normal, then determining the recovery reward value as a first reward value; if the post-recovery running state is abnormal, then determining at least one abnormal running parameter value from the reacquired plurality of running parameter values, and if the number of abnormal running parameter values is greater than or equal to a preset number threshold, then determining the recovery reward value as a second reward value; if the number of abnormal running parameter values is less than the preset number threshold, then determining the recovery reward value as a third reward value.

5. The edge computing node anomaly recovery method of claim 4, wherein, The determination of at least one abnormal running parameter value from the reacquired plurality of running parameter values comprises: Determine the abnormal running parameter value as the running parameter value not belonging to the corresponding preset running parameter value range.

6. The edge computing node anomaly recovery method of claim 1, wherein, When the number of edge computing nodes is greater than 1, the method further comprises: Obtain the abnormal occurrence timestamp and the topological relationship of each edge computing node, determine the abnormal co-occurrence relationship between each edge computing node according to the abnormal occurrence timestamp and the topological relationship, and represent the edge computing nodes as nodes, the abnormal co-occurrence relationship as edges, and the co-occurrence frequency as edge weights, and construct an abnormal co-occurrence relationship graph between the edge computing nodes; Cluster and divide each node in the abnormal co-occurrence relationship graph by using a community discovery algorithm; Determine the key degree corresponding to each node based on the abnormal co-occurrence relationship graph and the clustering and division result by using centrality calculation; Determine the recovery priority of each node according to the key degree.

7. The edge computing node anomaly recovery method of claim 1, wherein, When the number of edge computing nodes is greater than 1, the method further comprises: For the i-th edge computing node, splice the smooth running parameter values respectively corresponding to each smooth parameter sequence at the current moment to obtain the current state vector of the i-th edge computing node; Determine the node similarity between the i-th edge computing node and the j-th edge computing node by using a similarity algorithm, i≥1, j≥1, and i≠j; If the node similarity is greater than a preset similarity threshold, a risk warning prompt is issued, and the risk warning prompt is used to prompt abnormal troubleshooting of the i-th edge computing node and the j-th edge computing node.

8. An edge computing node anomaly recovery system, comprising: The system comprises: A data acquisition module configured to acquire a plurality of different running parameter sequences of an edge computing node within a preset time, wherein each running parameter sequence comprises a plurality of running parameter values collected within the preset time; A smoothing processing module configured to perform smoothing processing on each running parameter sequence by using a sliding window to obtain a plurality of smooth parameter sequences, wherein each smooth parameter sequence comprises a plurality of smooth running parameter values; A state determination module configured to input the smooth running parameter values respectively corresponding to each smooth parameter sequence at the current moment into a preset classification model to obtain a current running state of the edge computing node; A state prediction module configured to input each smooth parameter sequence into a preset time series model to obtain a plurality of different predicted running parameter values, and input each predicted running parameter value into the preset classification model to obtain a predicted running state of the edge computing node; A splicing module configured to splice the smooth running parameter values respectively corresponding to each smooth parameter sequence at the current moment and each predicted running parameter value to obtain an enhanced state vector of the edge computing node if the current running state is abnormal or the predicted running state is abnormal. A recovery control module is configured to determine a target recovery action instruction from a plurality of candidate recovery action instructions according to the enhanced state vector by using a reinforcement learning algorithm, and send the target recovery action instruction to the edge computing node to control the edge computing node to perform a target recovery action.

9. An electronic device, comprising: An edge computing node comprises a memory and a processor, the memory stores a computer program, and the computer program performs the edge computing node abnormal recovery method in any one of claims 1-7 when running on the processor.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program performs the edge computing node abnormal recovery method in any one of claims 1-7 when executed by a processor.

Images