Network performance baseline adjustment method, electronic device and storage medium

Abstract

The present invention relates to the field of network monitoring technology, and in particular to a network performance baseline adjustment method, electronic device and storage medium. The method includes: taking a target server as a federal center, processing each target client as a pseudo-client, constructing a network performance baseline corresponding to each preset session indicator based on the session traffic data between any pseudo-client and the target server, combining with a preset reference baseline to obtain a first fitting baseline and a first buoy baseline, inputting the preset reference baseline, the first fitting baseline and the first buoy baseline into a DQN neural network for training to update the network performance baseline, the target server aggregates the updated baseline of each client and updates the corresponding preset reference baseline, and when a preset number of iterations is reached, the final network performance baseline and the preset reference baseline are obtained; the present invention can realize real-time dynamic adjustment of the global baseline, maintain the personalized characteristics of the client baseline, and make network monitoring more reliable.

Description

Technical Field

[0001] The present invention relates to the technical field of network monitoring, and in particular to a network performance baseline adjustment method, electronic equipment and storage medium. Background Art

[0002] Network performance monitoring is a crucial component of network operations and maintenance. It helps assess system health, ensuring smooth network operation, preventing network failures, and improving network return on investment. A network performance baseline is a metric used in network performance monitoring to define the normal operating conditions of an enterprise's network infrastructure. Engineers can compare network performance baselines with network monitoring to identify traffic fluctuations that could lead to problems. The quality of a network's performance directly impacts user experience and network standards.

[0003] Currently, network performance monitoring and control typically involves collecting network performance data, building an adaptive mathematical model to determine a baseline, and then comparing the system's dynamic performance data against the baseline to determine whether current system performance is normal. The key technology lies in determining the network performance baseline. Current production practices often employ monitoring methods based on static baselines. These alarm baselines use fixed thresholds, resulting in low alarm efficiency, high false positive and false negative rates, and inability to provide maintenance personnel with effective network traffic information. In contrast, dynamic baselines describe the normal fluctuation range of a metric, which changes over time and has different values at different points in time. Common dynamic baseline generation methods include linear exponential smoothing, quadratic exponential smoothing, and normalized exponential smoothing. These methods utilize historical data for prediction and are only suitable for stable, linear data. However, they struggle to obtain reliable, real-time baseline data in diverse network environments.

[0004] Therefore, the present invention provides a reliable network performance baseline adjustment method that can dynamically adjust the baseline in real time according to the network environment. Summary of the Invention

[0005] In response to the above technical problems, the present invention provides a network performance baseline adjustment method, electronic device and storage medium, which can realize real-time dynamic adjustment of the global baseline while maintaining the personalized characteristics of each client baseline, making network monitoring more reliable.

[0006] According to a first aspect of the present invention, a method for adjusting a network performance baseline is provided, comprising the following steps:

[0007] S1, based on federated learning, takes the target server as the federation center, and mimics several target clients communicating with the target server into mimicked clients.

[0008] S2, obtaining session traffic data between any mimicking client and the target server within a historical time period, and constructing a network performance baseline corresponding to each preset session indicator based on the session traffic data.

[0009] S3, for any preset session indicator, based on the network performance baseline corresponding to the preset session indicator and the preset reference baseline corresponding to the target server, calculate a first fitting baseline and a first buoy baseline corresponding to the network performance baseline.

[0010] S4, input the preset reference baseline, the first fitting baseline and the first buoy baseline as training samples into a given DQN neural network for training, and output the Q value matrix after reaching the preset training rounds, so that the mimetic client updates the network performance baseline according to the maximum Q value in the Q value matrix, and obtains a new network performance baseline corresponding to the mimetic client.

[0011] S5, sending the new network performance baseline corresponding to each mimicking client to the target server, so that the target server aggregates the new network performance baselines and updates the preset reference baseline corresponding to the target server according to the aggregation result.

[0012] S6, based on the new network performance baseline and the updated preset reference baseline, return to execute steps S3-S6 until the preset number of iterations is reached, and obtain the final network performance baseline and the preset reference baseline to achieve the adjustment of the global baseline.

[0013] According to a second aspect of the present invention, a non-transitory computer-readable storage medium is provided, in which at least one instruction or at least one program is stored. The at least one instruction or the at least one program is loaded and executed by a processor to implement the above-mentioned network performance baseline adjustment method.

[0014] According to a third aspect of the present invention, there is provided an electronic device comprising a processor and the above-mentioned non-transitory computer-readable storage medium.

[0015] The present invention has at least the following beneficial effects:

[0016] The present invention provides a network performance baseline adjustment method. First, based on a federated learning model, a target server is used as a federation center, and each target client is simulated and processed as a simulated client. A network performance baseline corresponding to each preset session indicator is constructed based on the session traffic data between any simulated client and the target server. Combined with the preset reference baseline corresponding to the target server, a first fitting baseline and a first buoy baseline are obtained. The preset reference baseline, the first fitting baseline and the first buoy baseline, that is, the action sequence buoy, are input into a given DQN neural network as training samples for training. The Q value matrix is continuously updated through the first fitting baseline and the preset reference baseline, and the baseline action sequence is adjusted to achieve the update of the network performance baseline. The target server aggregates the updated network performance baseline of each simulated client to update its own corresponding preset reference baseline. The above steps are iterated again based on the updated network performance baseline and the preset reference baseline. When the number of iterative updates reaches the preset number of iterations, the final network performance baseline and the preset reference baseline are obtained. The above steps can realize real-time dynamic adjustment of the global baseline, maintain the personalized characteristics of each client baseline, and combine the results of multiple iterations to make network monitoring more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flowchart of a network performance baseline adjustment method provided by an embodiment of the present invention;

[0019] Figure 2 A schematic diagram of network performance baselines corresponding to several preset session indicators provided in an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of the training process based on a given DQN neural network provided in an embodiment of the present invention;

[0021] Figure 4 A schematic diagram of the initial data volume distribution of each mimic client provided in an embodiment of the present invention;

[0022] Figure 5 A schematic diagram of data distribution in a training set provided by an embodiment of the present invention;

[0023] Figure 6 A schematic diagram of data distribution in a test set provided by an embodiment of the present invention;

[0024] Figure 7A schematic diagram of the change in loss of a federated center at different training iterations provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0026] The embodiment of the present invention provides a method for adjusting network performance baseline, such as Figure 1 As shown, the method includes the following steps:

[0027] S1, based on federated learning, takes the target server as the federated center, and mimics several target clients communicating with the target server into mimicked clients; it can be understood as: taking the target server as the local center, and mimicking each target client locally.

[0028] Furthermore, when the number of initial clients communicating with the target server is greater than the preset training rule threshold N, N initial clients are randomly selected as target clients, otherwise all initial clients are used as target clients to prevent the training scale from being too large and affecting other business operations.

[0029] As mentioned above, by using the federated learning method, multiple participants, namely several mimicry clients and target servers, can jointly train a shared machine learning model while maintaining the privacy and localization of their respective data, so that each mimicry client can approach the baseline of the target server while retaining its own baseline characteristics.

[0030] S2, obtaining session traffic data between any mimicking client and the target server within a historical time period, and constructing a network performance baseline corresponding to each preset session indicator based on the session traffic data.

[0031] Specifically, the preset session indicators include but are not limited to an average communication delay indicator, a session communication duration indicator, a maximum communication delay indicator, and a minimum communication delay.

[0032] Furthermore, step S2 includes the following steps:

[0033] S201 : Divide the historical time period into a plurality of first historical time slices, and extract a session traffic segment of each first historical time slice from session traffic data.

[0034] Furthermore, the duration of the first historical time slice is 1 minute; in a specific implementation, it can also be adjusted according to the accuracy required by the user.

[0035] S202: For any preset session metric, data corresponding to the preset session metric is obtained from each session traffic segment, and a session metric baseline is generated. For example, for the average communication delay metric, the average communication delay is calculated for each session traffic segment, and the session metric baseline is obtained based on the average communication delays corresponding to multiple session traffic segments.

[0036] S203: When the duration of the historical time period is less than the preset duration, the session indicator baseline data is supplemented based on the curve characteristics of the Gaussian distribution to construct a network performance baseline corresponding to the preset session indicator. This can be understood as: aligning the obtained session indicator baseline with time and supplementing the timestamps with missing data values based on the Gaussian distribution.

[0037] Specifically, when supplementing any missing data in the session metric baseline, the data value of the missing data meets the following conditions:

[0038] , where D t is the data value corresponding to the timestamp of the t-th missing data, D0 is the average value of the timestamps of several existing data, and σ is the standard deviation of several existing data; since the first historical time slice used above is in the form of a time segment, the t-th timestamp can be considered as the middle moment in the t-th time segment.

[0039] Furthermore, the duration corresponding to the network performance baseline corresponding to the preset session indicator is the preset duration. In a specific implementation, the preset duration can be 24 hours. For example, the completed network performance baseline is as follows: Figure 2 As shown in the figure, the bottom-up network performance baselines are the baselines corresponding to the average communication delay indicator, the session communication duration indicator, the maximum communication delay indicator, and the minimum communication delay indicator. The gray mask area is the known data, and the two sides of the mask are the completed data.

[0040] As described above, by performing baseline separation and data completion processing on session traffic data, the baseline corresponding to each preset session indicator can be individually analyzed and processed. In addition, based on the characteristics of traffic data in the communication system, for example, the latency indicator often exhibits Gaussian distribution characteristics, the Gaussian distribution is used to complete the baseline, which can provide reliable and comprehensive samples for subsequent training.

[0041] S3, for any preset session indicator, according to the network performance baseline corresponding to the preset session indicator and the preset reference baseline corresponding to the target server, calculate the first fitting baseline and the first buoy baseline corresponding to the network performance baseline; it can be understood that: the preset reference baseline serves as the observation baseline of the client, and in subsequent training, the network performance baseline of the client and the observation baseline continuously interact to update the Q value matrix.

[0042] Specifically, obtain the preset reference baseline corresponding to the target server through the following steps:

[0043] S10 , based on a target time period corresponding to a start point and an end point of the network performance baseline, dividing the target time period into a plurality of second historical time slices.

[0044] Specifically, the duration of the second historical time slice is the same as the duration of the first historical time slice. During implementation, the second historical time slice is aligned with the first historical time slice.

[0045] S20, based on the historical traffic data of the target server within the target time period, extract the historical traffic segments of each second historical time slice; it can be understood that: each target client has its own corresponding session traffic data, and the target server also has its own corresponding communication traffic data. For example, with respect to the average communication delay indicator, different target clients have different average communication delay indicators when communicating with the target server due to factors such as their own load and properties, and the target server also calculates its own average communication delay indicator when sending traffic.

[0046] S30, for any preset session indicator, obtain the data corresponding to the preset session indicator from each historical traffic segment, and generate a preset reference baseline corresponding to the target server; it can be understood that the preset reference baseline corresponding to the target server in step S3 refers to the preset reference baseline corresponding to the same preset session indicator to which the network performance baseline belongs.

[0047] Furthermore, the first fitting baseline is obtained by the following steps:

[0048] S301, extracting data values corresponding to the network performance baseline and the preset reference baseline at each moment.

[0049] S302 : For any moment, calculate the average value of the data value corresponding to the network performance baseline and the data value corresponding to the preset reference baseline, and generate a first fitting baseline according to the average value corresponding to each moment.

[0050] Furthermore, the first buoy baseline is obtained by the following steps:

[0051] S310: extracting data values corresponding to the network performance baseline and the preset reference baseline at each moment.

[0052] S320, for any moment, calculate the difference between the data value corresponding to the network performance baseline and the data value corresponding to the preset reference baseline, and generate a first buoy baseline according to the difference corresponding to each moment.

[0053] As mentioned above, by obtaining the first fitting baseline and the first buoy baseline, the difference between the network performance baseline and the preset reference baseline can be known. The obtained first fitting baseline preliminarily eliminates the error between the network performance baseline and the preset reference baseline. Subsequent training is then carried out based on the obtained first buoy baseline, which is conducive to making the network performance baseline closer to the preset reference baseline, reducing the prediction error of the baseline, and thus obtaining a more reliable dynamic baseline.

[0054] S4, the preset reference baseline, the first fitting baseline and the first buoy baseline are input as training samples into a given DQN neural network for training, and when the preset training round is reached, the Q value matrix is output so that the mimicking client updates the network performance baseline according to the maximum Q value in the Q value matrix, and obtains a new network performance baseline corresponding to the mimicking client; it can be understood that: the Q value matrix is the action-state matrix in the training process, each Q value corresponds to an action sequence, and the maximum Q value corresponds to the best action sequence, and the best action sequence is updated to the current network performance baseline to obtain a new network performance baseline. The training process diagram based on the given DQN neural network is as shown below Figure 3 shown.

[0055] Specifically, those skilled in the art set the preset training rounds according to actual needs, which will not be described in detail here.

[0056] To facilitate understanding of the solution, the training process is explained as follows: The DQN neural network is a combination of the Q-Learning algorithm and the neural network. Based on the Q-Learning algorithm, each mimicking client needs to maintain a Q-value table, that is, a Q-value matrix. The rows represent different states and the columns represent different actions. Initially, the Q value is set to a small random number. During the training process, the intelligent agent in the neural network updates the Q value in the Q-value table by continuously interacting with the preset reference baseline, that is, the observation baseline, and adjusts the action sequence of the first fitting baseline based on the current action sequence. The update formula follows the Bellman equation. By querying the Q-value table, the intelligent agent can quickly select the optimal action in the current state, thereby realizing goal-oriented behavioral decision-making. As the training progresses, the Q-value table will gradually converge to obtain a stable network performance baseline.

[0057] As described above, during the training process, the preset reference baseline is used as the observation baseline of the mimicking client. The Q-value matrix is updated by continuously interacting between its own baseline and the observation baseline, and the baseline action sequence is adjusted in real time to make the network performance baseline automatically approach the preset reference baseline, so that the baseline state adapts to changes in the network environment. Moreover, since the training is performed independently on the client side, its own baseline characteristics are also retained, which improves the applicability of the baseline in its own environment.

[0058] S5, sending the new network performance baseline corresponding to each mimicking client to the target server, so that the target server aggregates the new network performance baselines and updates the preset reference baseline corresponding to the target server according to the aggregation result.

[0059] In one embodiment, in step S5, the preset reference baseline corresponding to the target server is updated through the following steps:

[0060] S501: extracting data values corresponding to each new network performance baseline at each time point.

[0061] S502, for any time point, calculate the average value of the data values corresponding to several new network performance baselines, and generate a new preset reference baseline based on the average value of the data values corresponding to several new network performance baselines at each time point, so as to update the preset reference baseline corresponding to the target server.

[0062] As mentioned above, the target server aggregates the new network performance baseline corresponding to each mimicked client, and the network performance baseline is extracted based on the session traffic data between the mimicked client itself and the target server, so it can reflect part of the network communication situation of the target server. By taking the average of the network performance baselines of all mimicked clients, the dynamic adjustment of the preset reference baseline can be made more reliable.

[0063] In another embodiment, in step S5, the preset reference baseline corresponding to the target server is updated through the following steps:

[0064] S510: Extracting data values corresponding to each new network performance baseline at each time point.

[0065] S520, for any time point, calculating an average value of data values corresponding to several new network performance baselines, and generating an initial reference baseline according to the average value of data values corresponding to the several new network performance baselines at each time point.

[0066] S530, based on the preset reference baseline and the initial reference baseline, calculate the second fitting baseline and the second buoy baseline corresponding to the preset reference baseline; it can be understood as: taking the preset reference baseline as the network performance baseline in step S3, taking the initial reference baseline as the preset reference baseline in step S3, executing step S3 to obtain the second fitting baseline and the second buoy baseline, that is, the method of obtaining the second fitting baseline is consistent with the method of obtaining the first fitting baseline, and the method of obtaining the second buoy baseline is consistent with the method of obtaining the first buoy baseline, which will not be repeated here.

[0067] S540, the initial reference baseline, the second fitting baseline and the second buoy baseline are input as training samples into a given DQN neural network for training, and when the preset training rounds are reached, the Q value matrix is output so that the target server updates the preset reference baseline according to the maximum Q value in the Q value matrix, and obtains a new preset reference baseline corresponding to the target server; it can be understood that: the training process of this step is consistent with the training process of step S4, and will not be repeated here.

[0068] As mentioned above, since the network performance baseline can reflect part of the network communication status of the target server, the target server obtains the initial reference baseline by aggregating the new network performance baseline corresponding to each mimicking client, which serves as the observation baseline on the target server side, and adjusts the baseline action sequence in real time through continuous interaction between the preset reference baseline and the observation baseline, making the dynamic adjustment of the preset reference baseline more real and reliable.

[0069] S6, based on the new network performance baseline and the updated preset reference baseline, return to execute steps S3-S6 until the preset number of iterations is reached, and obtain the final network performance baseline and the preset reference baseline to achieve the adjustment of the global baseline.

[0070] Furthermore, the preset number of iterations is 5. In the embodiment of the present invention, the preset number of iterations is obtained through creative experiments, and the specific experimental steps are as follows:

[0071] (1) Data preparation

[0072] The data, totaling approximately 3.25GB and containing 11,628 files, contains 64KB of packet headers for traffic between an airport (network segment 10.48.0.0 / 16) and a headquarters data center (network segments 10.5.0.0 / 16 and 10.6.0.0 / 16), stored in .pcap format. The packet data represents actual traffic flow from June 25 to July 4, 2024, for several time periods, such as 5:30 AM to 8:30 AM on June 25 and 9:40 AM to 2:40 PM on July 4. While traffic flow is discontinuous between time periods, traffic within the time period is continuously captured, with the time value within the packet recording the specific timestamp.

[0073] (2) Experimental environment

[0074] The experimental equipment includes two Intel(R) Xeon(R) Gold 5218R CPUs, each running at 2.10GHz, for a total of two processors. The system memory is 128GB, all of which is available. The operating system is Windows 10 Pro. The storage capacity is 3.6TB. Furthermore, the equipment is equipped with two NVIDIA GeForce RTX 3090 graphics cards, capable of supporting large-scale data training.

[0075] (3) Network performance baseline generation

[0076] The baseline completion method based on Gaussian distribution is used to preliminarily supplement the missing data. The network performance baseline after completion is as follows: Figure 2 shown.

[0077] Because the amount of data varies from one mimetic client to another, for mimetic clients with smaller initial data volumes, the probability-based baseline accounts for a larger proportion, while the actual business data accounts for a smaller proportion. Directly using the padded baseline will affect monitoring performance. Therefore, when a mimetic client has less initial data, it will rely more heavily on the observed preset reference baseline. Figure 4 is the initial data volume distribution of each mimic client, Figure 4 It can be seen that the initial data volume of some mimicking clients is more than 700, while the initial data volume of a few clients is relatively small, which means that network activities are not frequent and the business access demand is not high.

[0078] According to statistics, among the 346 mimetic clients, the number of communication segment sessions in 279 clients exceeded 100. The remaining mimetic clients were not included in the mimetic training scope due to the small number of sessions.

[0079] (4) Training process

[0080] We selected 80% of the time period of the existing data as the training set, simulated a client to generate a network performance baseline locally using the training data, and then performed local training. We compared the deviation between the time period and the 20% test period in the source data and observed the changes in various training indicators. We selected the baseline between 5:04 and 15:00 in the source data as the training set, and the data between 15:01 and 16:52 as the test set. Figure 5 and Figure 6 The distribution of two sets of data in the training set and test set respectively, where, from bottom to top, they are the training set data and test set data corresponding to the average communication delay indicator, session communication duration indicator, maximum communication delay indicator and minimum communication delay indicator.

[0081] 50 clients are randomly selected to participate in the mimic training, and the global training iteration rounds are set to 5. The changing trend of the federation center loss under different iteration numbers is observed. Figure 7 The figure below is a diagram showing the change in loss of the federated center at different training iterations. The experimental results show that the average loss value of the target server converges during the fourth and fifth rounds of training. The loss convergence effect is obvious after five rounds of training, so the preset number of iterations is set to 5.

[0082] As described above, by performing multiple rounds of training and iterative adjustment on the network performance baseline on the client side and the preset reference baseline on the target server side, real-time dynamic adjustment of the global baseline involving multiple nodes is achieved, making each baseline state more adaptable to changes in the network environment. Even in a network environment with diverse situations, a reliable baseline state can be obtained, which is conducive to accurate monitoring of network performance.

[0083] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to implementing a method in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the network performance baseline adjustment method provided in the above embodiment.

[0084] An embodiment of the present invention further provides an electronic device including a processor and the aforementioned non-transitory computer-readable storage medium.

[0085] Although some specific embodiments of the present invention have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A network performance baseline adjustment method, characterized in that: The method comprises the following steps: S1, based on federated learning, uses the target server as the federation center and mimics several target clients communicating with the target server into mimic clients; S2, obtaining session traffic data between any mimicking client and the target server within a historical time period, and constructing a network performance baseline corresponding to each preset session indicator based on the session traffic data; The network performance baseline corresponding to the preset session indicator is constructed based on the session traffic data corresponding to each first historical time slice in the historical time period; S3, for any preset session indicator, calculating a first fitting baseline and a first buoy baseline corresponding to the network performance baseline based on the network performance baseline corresponding to the preset session indicator and the preset reference baseline corresponding to the target server; The preset reference baseline corresponding to the target server is constructed based on the traffic data of the target server corresponding to each second historical time slice in the target time period; the duration of the second historical time slice is the same as the duration of the first historical time slice; The first fitting baseline is obtained through the following steps: S301, extracting data values corresponding to the network performance baseline and the preset reference baseline at each moment; S302, calculating, at any moment, an average of a data value corresponding to the network performance baseline and a data value corresponding to a preset reference baseline, and generating a first fitting baseline based on the average values corresponding to each moment; Obtain the first buoy baseline by following these steps: S310, extracting data values corresponding to the network performance baseline and the preset reference baseline at each moment; S320, calculating the difference between the data value corresponding to the network performance baseline and the data value corresponding to the preset reference baseline at any moment, and generating a first buoy baseline based on the difference corresponding to each moment; S4, inputting the preset reference baseline, the first fitting baseline, and the first buoy baseline as training samples into a given DQN neural network for training, and outputting a Q value matrix after reaching a preset training round, so that the mimetic client updates the network performance baseline according to the maximum Q value in the Q value matrix, thereby obtaining a new network performance baseline corresponding to the mimetic client; S5, sending the new network performance baseline corresponding to each mimic client to the target server, so that the target server aggregates the new network performance baselines and updates the preset reference baseline corresponding to the target server according to the aggregation result; S6, based on the new network performance baseline and the updated preset reference baseline, return to execute steps S3-S6 until the preset number of iterations is reached, and obtain the final network performance baseline and the preset reference baseline to achieve the adjustment of the global baseline.

2. The network performance baseline adjustment method according to claim 1, characterized in that: Step S2 includes the following steps: S201, dividing the historical time period into a plurality of first historical time slices, and extracting a session traffic segment of each first historical time slice from the session traffic data; S202, for any preset session indicator, obtaining data corresponding to the preset session indicator from each session traffic segment, and generating a session indicator baseline; S203: When the duration of the historical time period is less than the preset duration, the data of the session indicator baseline is supplemented based on the curve characteristics of the Gaussian distribution to construct a network performance baseline corresponding to the preset session indicator; the duration corresponding to the network performance baseline corresponding to the preset session indicator is the preset duration.

3. The network performance baseline adjustment method according to claim 1, characterized in that: The preset session indicator is any one of an average communication delay indicator, a session communication duration indicator, a maximum communication delay indicator, and a minimum communication delay indicator.

4. The network performance baseline adjustment method according to claim 1, characterized in that: Obtain the preset reference baseline corresponding to the target server by following the steps below: S10, dividing the target time period into a plurality of second historical time slices based on a target time period corresponding to a start point and an end point of the network performance baseline; S20, extracting historical traffic segments of each second historical time slice based on historical traffic data of the target server within the target time period; S30 , for any preset session indicator, obtaining data corresponding to the preset session indicator from each historical traffic segment, and generating a preset reference baseline corresponding to the target server.

5. The network performance baseline adjustment method according to claim 1, characterized in that: In step S5, the preset reference baseline corresponding to the target server is updated through the following steps: S501, extracting the data values corresponding to each new network performance baseline at each time point; S502, for any time point, calculate the average value of the data values corresponding to several new network performance baselines, and generate a new preset reference baseline based on the average value of the data values corresponding to several new network performance baselines at each time point, so as to update the preset reference baseline corresponding to the target server.

6. The network performance baseline adjustment method according to claim 1, characterized in that: In step S5, the preset reference baseline corresponding to the target server is also updated through the following steps: S510, extracting data values corresponding to each new network performance baseline at each time point; S520, calculating an average of data values corresponding to a number of new network performance baselines at any time point, and generating an initial reference baseline based on the average of the data values corresponding to the number of new network performance baselines at each time point; S530, calculating a second fitting baseline and a second buoy baseline corresponding to the preset reference baseline based on the preset reference baseline and the initial reference baseline; S540, the initial reference baseline, the second fitting baseline and the second buoy baseline are input as training samples into a given DQN neural network for training, and when the preset training rounds are reached, the Q value matrix is output so that the target server updates the preset reference baseline according to the maximum Q value in the Q value matrix, and obtains a new preset reference baseline corresponding to the target server.

7. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the storage medium, characterized in that: The at least one instruction or the at least one program segment is loaded and executed by the processor to implement the network performance baseline adjustment method according to any one of claims 1 to 6.

8. An electronic device, characterized in that: The device comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 7.

Images