Cloud service distributed flow limiting method based on time sequence prediction algorithm
Through the distributed flow limiting method of cloud services based on time series prediction algorithm, the current limiting parameters are dynamically adjusted, which solves the problem of being unable to actively adapt to traffic changes in the existing technology, and dynamic learning and prediction of traffic patterns are realized, and the stability and resource utilization of the system are improved.
Patent Information
- Application Number
- CN202510663452.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-22
- Publication Date
- 2025-08-08
AI Technical Summary
The existing distributed current limiting method lacks the ability to predict traffic trends and is unable to actively adapt to traffic changes, resulting in the system's slow response when facing burst traffic, which may not be able to effectively prevent system overload.
The distributed flow limiting method of cloud services based on time series prediction algorithm is adopted, and the flow limiting parameters are dynamically adjusted to realize dynamic learning and prediction of traffic mode through data acquisition, preprocessing, feature engineering, construction of prediction models and flow limiting parameter management.
Adaptive adjustment of the current limiting strategy is realized, responding to potential traffic changes in advance, preventing system overload, improving resource utilization, and optimizing the synchronization of the current limiting strategy between distributed nodes to ensure the consistency and accuracy of the current limiting.
Smart Images
Figure CN120455376A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of cloud computing technology, and in particular to a cloud service distributed current limiting method based on a time series prediction algorithm. Background Art
[0002] In distributed systems, current limiting is a common flow control technique designed to prevent system overload and ensure service stability. When a system experiences sudden high traffic, current limiting can maintain normal system operation by temporarily rejecting some requests or delaying their processing.
[0003] In the prior art, distributed current limiting methods generally rely on static rules or preset thresholds to control traffic, such as using a token bucket algorithm or a leaky bucket algorithm to limit the rate of incoming requests.
[0004] Among these traditional methods, the token bucket algorithm generates tokens at a fixed rate. Requests must consume tokens to be processed. If insufficient tokens are available, requests are rejected or queued. The leaky bucket algorithm allows requests through at a fixed rate, while any excess requests are queued. These rate-limiting mechanisms typically require operations staff or developers to set fixed parameters, such as the token generation rate and bucket capacity, based on experience. Furthermore, traditional rate-limiting methods lack the ability to predict traffic trends and cannot proactively adapt to traffic changes, relying solely on reactive responses. This results in a system that reacts slowly to sudden traffic bursts and may be unable to effectively prevent system overloads at critical moments.
[0005] In order to solve the above problems, the present invention proposes a cloud service distributed current limiting method based on a time series prediction algorithm. Summary of the Invention
[0006] In order to overcome the shortcomings of the prior art, the present invention provides a simple and efficient cloud service distributed current limiting method based on a time series prediction algorithm.
[0007] The present invention is achieved through the following technical solutions:
[0008] A distributed throttling method for cloud services based on a time series prediction algorithm includes the following steps:
[0009] Step S1: Data collection
[0010] Deploy a data collection module in the distributed architecture of the cloud computing environment to regularly collect traffic data and related performance indicators of each node according to a custom cycle;
[0011] Step S2: Preprocessing and feature engineering
[0012] Perform noise data processing on the collected data in specific scenarios, such as abnormal traffic caused by DDoS attacks or system failures, including but not limited to filling missing values, removing outliers, and data normalization and standardization;
[0013] Customize and extract traffic prediction features based on preprocessed data, including but not limited to historical request rates and request response times;
[0014] Step S3: Build a prediction model to achieve traffic prediction
[0015] Identify patterns and trends in traffic data through time series analysis. Based on the analysis results, select a suitable prediction model (such as ARIMA, LSTM, etc.) for training to achieve short-term or long-term traffic forecasts.
[0016] Step S4: Current limiting parameter management
[0017] Based on the prediction results, dynamically adjust the flow control parameters (such as rate and capacity) and distribute the adjusted flow control parameters to all relevant nodes in the cloud computing environment to ensure that each node adopts consistent flow control measures;
[0018] Step S5: Monitoring and feedback
[0019] Monitor and collect feedback system performance and traffic data in real time, compare actual conditions with predicted results, and adjust the prediction model based on deviations to achieve continuous optimization and adjustment of the prediction model.
[0020] In step S3, the following steps are performed:
[0021] Step S3.1: Analyze the cloud service traffic characteristics and select a suitable time series model and corresponding model parameters based on the traffic characteristics;
[0022] In step S3.1, the historical traffic data of the cloud service is analyzed in detail to observe whether there are obvious periodic fluctuations (such as daily peak hours), long-term trends (such as traffic increases caused by user growth), or seasonal patterns (such as traffic surges during holidays);
[0023] If the traffic data is relatively stable and exhibits obvious periodicity and seasonality, the ARIMA model is used; this model is good at capturing these characteristics and effectively predicting future short-term traffic.
[0024] For the ARIMA model, the difference order d, the number of autoregressive terms p, and the number of moving average terms q are determined by the autocorrelation function ACF and partial autocorrelation function PACF graphs. The formula is as follows:
[0025] ACF(T)→q
[0026] PACF(T)→p
[0027] Where T is the lag order;
[0028] Seasonal parameters are determined using seasonal differences for periodic data, including the seasonal difference order D, the number of seasonal autoregressive terms P, the number of seasonal moving average terms Q, and the length of the seasonal cycle s. The seasonal difference order D is determined by the number of differences.
[0029] The formula is as follows:
[0030] S ACF (T s )→Q
[0031] S PACF (T s )→P
[0032] Among them, T s is the lag order after seasonal differencing;
[0033] If the traffic data shows complex nonlinear patterns, or long-term memory capabilities are required to make predictions based on long-term traffic trends, the LSTM model is used; LSTM can process long sequences of data and capture deep features in the time series.
[0034] For the LSTM model, select the network structure and training parameters;
[0035] The network structure includes the number of layers L and the number of neurons in each layer N;
[0036] The training parameters include the appropriate batch size BatchSize and the learning rate \eta.
[0037] Architecture(L,N)=LSTM(Layers=L,Neurons=N)
[0038] Hyperparams(BatchSize,n)=Tune(BatchSize,n)
[0039] Step S3.2: Divide the preprocessed historical data into a training set and a test set, and use the training set to train the selected time series model;
[0040] In step S3.2, when the ARIMA model is selected as the time series model, the appropriate difference order, number of autoregressive terms, and number of moving average terms are determined during the training process;
[0041] When the LSTM model is selected as the time series model, the network structure is designed and the weights are adjusted through the backpropagation algorithm during the training process.
[0042] Step S3.3: Use the test set to evaluate the prediction performance of the model and verify and optimize the trained time series model;
[0043] Evaluation indicators include mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE).
[0044] In step S3.3, the hyperparameters of the time series model are tuned according to the performance evaluation results;
[0045] When the ARIMA model is used as the time series model, the information criterion (such as Akaike Information Criterion AIC or Bayesian Information Criterion BIC) is used to select the optimal parameter combination and minimize the information criterion value. The formula is as follows:
[0046] When the LSTM model is used as the time series model, random dropout D or weight regularization λ is applied to reduce overfitting;
[0047] Regularization(D,λ)=LSTM(Dropout=D,Weight R egularization
[0048] Train the model using the Backpropagation Through Time (BPTT) algorithm and an appropriate loss function (such as mean squared error MSE):
[0049] Train(BPTT,MSE)=Optimize(BPTT,Loss=MSE)
[0050] In step S3.3, different training sets and test sets are divided, and a cross-validation method is used to further verify the stability and generalization ability of the model;
[0051] When the ARIMA model is selected as the time series model, the model fit is checked through residual analysis to ensure that the residual sequence is close to the white noise sequence;
[0052] (The latter is the fitted value of the model), the ACF and PACF plots of the residuals should show that the correlation of all lagged terms is close to zero; at the same time, use rolling forecasts or k-fold cross-validation to assess the generalization ability of the model and adjust the parameters based on the prediction error:
[0053] CV error =f(Model ARIMA ,T′ validation )
[0054] Ensure that the optimized parameters should minimize the cross-validation error;
[0055] When the LSTM model is selected as the time series model, use the validation set or cross-validation to evaluate the model performance and select the model configuration with the smallest error;
[0056] BestModel=SelectModel(ValidationSet,CrossValidation)
[0057] Use model ensemble techniques, such as bagging or boosting, to improve the stability and accuracy of predictions;
[0058] EnsembleModel=Ensemble(Bagging,Boosting)
[0059] After the model is deployed, the deviation between its predicted results and actual traffic is monitored in real time, and the model is fine-tuned based on the feedback to maintain or improve the prediction accuracy.
[0060] The cloud service distributed current limiting system based on the time series prediction algorithm is used to implement the above method, including a data acquisition module, a preprocessing and feature engineering module, a prediction model module, a current limiting parameter management module, and a monitoring and feedback module;
[0061] The data collection module is deployed in the distributed architecture of the cloud computing environment and is responsible for regularly collecting traffic data and related performance indicators of each node according to a custom cycle;
[0062] The preprocessing and feature engineering module is responsible for processing noise data collected in specific scenarios, such as abnormal traffic caused by DDoS attacks or system failures, including but not limited to filling missing values, removing outliers, and data normalization and standardization.
[0063] Customize and extract traffic prediction features based on preprocessed data, including but not limited to historical request rates and request response times;
[0064] The prediction model module is responsible for identifying patterns and trends in traffic data through time series analysis. Based on the analysis results of the time series, it customizes and selects appropriate prediction models (such as ARIMA and LSTM) for training to achieve short-term or long-term traffic forecasts.
[0065] The flow control parameter management module is responsible for dynamically adjusting flow control parameters (such as rate and capacity) based on the prediction results and distributing the adjusted flow control parameters to all relevant nodes in the cloud computing environment to ensure that each node adopts consistent flow control measures.
[0066] The monitoring and feedback module is responsible for real-time monitoring and collecting feedback system performance and traffic data, comparing the actual situation with the predicted results, and adjusting the prediction model according to the deviation to achieve continuous optimization and adjustment of the prediction model.
[0067] A cloud service distributed current limiting device based on a time series prediction algorithm is characterized by comprising a memory and a processor; the memory is used to store a computer program, and the processor is used to implement the above-mentioned method steps when executing the computer program.
[0068] A readable storage medium, characterized in that: a computer program is stored on the readable storage medium, and the computer program implements the above method steps when executed by a processor.
[0069] The beneficial effects of the present invention are as follows: the cloud service distributed current limiting method based on the time series prediction algorithm realizes the dynamic learning and prediction of the traffic pattern, so that the current limiting strategy can be adaptively adjusted, responding to potential traffic changes in advance, more effectively preventing system overload, and improving resource utilization; at the same time, it optimizes the current limiting strategy synchronization mechanism between distributed nodes, reduces synchronization delay, and ensures the consistency and accuracy of current limiting. BRIEF DESCRIPTION OF THE DRAWINGS
[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0071] Figure 1 Schematic diagram of the cloud service distributed current limiting method based on the time series prediction algorithm of the present invention. DETAILED DESCRIPTION
[0072] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.
[0073] A distributed throttling method for cloud services based on a time series prediction algorithm includes the following steps:
[0074] Step S1: Data collection
[0075] Deploy a data collection module in the distributed architecture of the cloud computing environment to regularly collect traffic data and related performance indicators of each node according to a custom cycle;
[0076] Step S2: Preprocessing and feature engineering
[0077] Perform noise data processing on the collected data in specific scenarios, such as abnormal traffic caused by DDoS attacks or system failures, including but not limited to filling missing values, removing outliers, and data normalization and standardization;
[0078] Customize and extract traffic prediction features based on preprocessed data, including but not limited to historical request rates and request response times;
[0079] Step S3: Build a prediction model to achieve traffic prediction
[0080] Identify patterns and trends in traffic data through time series analysis. Based on the analysis results, select a suitable prediction model (such as ARIMA, LSTM, etc.) for training to achieve short-term or long-term traffic forecasts.
[0081] Step S4: Current limiting parameter management
[0082] Based on the prediction results, dynamically adjust the flow control parameters (such as rate and capacity) and distribute the adjusted flow control parameters to all relevant nodes in the cloud computing environment to ensure that each node adopts consistent flow control measures;
[0083] Step S5: Monitoring and feedback
[0084] Monitor and collect feedback system performance and traffic data in real time, compare actual conditions with predicted results, and adjust the prediction model based on deviations to achieve continuous optimization and adjustment of the prediction model.
[0085] In step S3, the following steps are performed:
[0086] Step S3.1: Analyze the cloud service traffic characteristics and select a suitable time series model and corresponding model parameters based on the traffic characteristics;
[0087] In step S3.1, the historical traffic data of the cloud service is analyzed in detail to observe whether there are obvious periodic fluctuations (such as daily peak hours), long-term trends (such as traffic increases caused by user growth), or seasonal patterns (such as traffic surges during holidays);
[0088] If the traffic data is relatively stable and exhibits obvious periodicity and seasonality, the ARIMA model is used; this model is good at capturing these characteristics and effectively predicting future short-term traffic.
[0089] For the ARIMA model, the difference order d, the number of autoregressive terms p, and the number of moving average terms q are determined by the autocorrelation function ACF and partial autocorrelation function PACF graphs. The formula is as follows:
[0090] ACF(T)→q
[0091] PACF(T)→p
[0092] Where T is the lag order;
[0093] Seasonal parameters are determined using seasonal differences for periodic data, including the seasonal difference order D, the number of seasonal autoregressive terms P, the number of seasonal moving average terms Q, and the length of the seasonal cycle s. The seasonal difference order D is determined by the number of differences.
[0094] The formula is as follows:
[0095] S ACF (T s )→Q
[0096] S PACF (T s )→P
[0097] Among them, T s is the lag order after seasonal differencing;
[0098] If the traffic data shows complex nonlinear patterns, or long-term memory capabilities are required to make predictions based on long-term traffic trends, the LSTM model is used; LSTM can process long sequences of data and capture deep features in the time series.
[0099] For the LSTM model, select the network structure and training parameters;
[0100] The network structure includes the number of layers L and the number of neurons in each layer N;
[0101] The training parameters include the appropriate batch size BatchSize and the learning rate \eta.
[0102] Architecture(L,N)=LSTM(Layers=L,Neurons=N)
[0103] Hyperparams(BatchSize,n)=Tune(BatchSize,n)
[0104] Step S3.2: Divide the preprocessed historical data into a training set and a test set, and use the training set to train the selected time series model;
[0105] In step S3.2, when the ARIMA model is selected as the time series model, the appropriate difference order, number of autoregressive terms, and number of moving average terms are determined during the training process;
[0106] When the LSTM model is selected as the time series model, the network structure is designed and the weights are adjusted through the backpropagation algorithm during the training process.
[0107] Step S3.3: Use the test set to evaluate the prediction performance of the model and verify and optimize the trained time series model;
[0108] Evaluation indicators include mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE).
[0109] In step S3.3, the hyperparameters of the time series model are tuned according to the performance evaluation results;
[0110] When the ARIMA model is used as the time series model, the information criterion (such as Akaike Information Criterion AIC or Bayesian Information Criterion BIC) is used to select the optimal parameter combination and minimize the information criterion value. The formula is as follows:
[0111]
[0112] When the LSTM model is used as the time series model, random dropout D or weight regularization λ is applied to reduce overfitting;
[0113] Regularization(D,λ)=LSTM(Dropout=D,Weight R egularization = λ)
[0114] Train the model using the Backpropagation Through Time (BPTT) algorithm and an appropriate loss function (such as mean squared error MSE):
[0115] Train(BPTT,MSE)=Optimize(BPTT,Loss=MSE)
[0116] In step S3.3, different training sets and test sets are divided, and a cross-validation method is used to further verify the stability and generalization ability of the model;
[0117] When the ARIMA model is selected as the time series model, the model fit is checked through residual analysis to ensure that the residual sequence is close to the white noise sequence;
[0118] (The latter is the fitted value of the model), the ACF and PACF plots of the residuals should show that the correlation of all lagged terms is close to zero; at the same time, use rolling forecasts or k-fold cross-validation to assess the generalization ability of the model and adjust the parameters based on the prediction error:
[0119] CV error =f(Model ARIMA ,T′ validation )
[0120] Ensure that the optimized parameters should minimize the cross-validation error;
[0121] When the LSTM model is selected as the time series model, use the validation set or cross-validation to evaluate the model performance and select the model configuration with the smallest error;
[0122] BestModel=SelectModel(ValidationSet,CrossValidation)
[0123] Use model ensemble techniques, such as bagging or boosting, to improve the stability and accuracy of predictions;
[0124] EnsembleModel=Ensemble(Bagging,Boosting)
[0125] After the model is deployed, the deviation between its predicted results and actual traffic is monitored in real time, and the model is fine-tuned based on the feedback to maintain or improve the prediction accuracy.
[0126] The cloud service distributed current limiting system based on the time series prediction algorithm is used to implement the above method, including a data acquisition module, a preprocessing and feature engineering module, a prediction model module, a current limiting parameter management module, and a monitoring and feedback module;
[0127] The data collection module is deployed in the distributed architecture of the cloud computing environment and is responsible for regularly collecting traffic data and related performance indicators of each node according to a custom cycle;
[0128] The preprocessing and feature engineering module is responsible for processing noise data collected in specific scenarios, such as abnormal traffic caused by DDoS attacks or system failures, including but not limited to filling missing values, removing outliers, and data normalization and standardization.
[0129] Customize and extract traffic prediction features based on preprocessed data, including but not limited to historical request rates and request response times;
[0130] The prediction model module is responsible for identifying patterns and trends in traffic data through time series analysis. Based on the analysis results of the time series, it customizes and selects appropriate prediction models (such as ARIMA and LSTM) for training to achieve short-term or long-term traffic forecasts.
[0131] The flow control parameter management module is responsible for dynamically adjusting flow control parameters (such as rate and capacity) based on the prediction results and distributing the adjusted flow control parameters to all relevant nodes in the cloud computing environment to ensure that each node adopts consistent flow control measures.
[0132] The monitoring and feedback module is responsible for real-time monitoring and collecting feedback system performance and traffic data, comparing the actual situation with the predicted results, and adjusting the prediction model according to the deviation to achieve continuous optimization and adjustment of the prediction model.
[0133] The cloud service distributed current limiting device based on the time series prediction algorithm includes a memory and a processor; the memory is used to store a computer program, and the processor is used to implement the above method steps when executing the computer program.
[0134] The readable storage medium stores a computer program, which implements the above method steps when executed by a processor.
[0135] Compared with existing technologies, this cloud service distributed throttling method based on time series prediction algorithm has the following characteristics:
[0136] 1) Through real-time traffic prediction, the system can foresee performance bottlenecks and make advance adjustments to the throttling strategy, thereby reducing or avoiding the risk of service interruptions and performance degradation, directly improving system reliability and user experience;
[0137] 2) Dynamic flow control parameter adjustment enables the system to flexibly allocate resources according to actual traffic demand, providing sufficient service capacity during peak traffic and saving resources during low traffic, thereby achieving optimal resource allocation and maximizing cost-effectiveness;
[0138] 3) Distributed policy deployment ensures that consistent rate limiting policies are implemented across all nodes in the entire cloud computing environment. This consistency is crucial for maintaining the stable operation of the entire distributed system.
[0139] 4) Real-time monitoring and feedback mechanisms can continuously optimize the prediction model, so that the model's predictive ability continues to improve over time. This self-optimization capability ensures the long-term effectiveness and adaptability of the technical solution.
[0140] The embodiment described above is only one specific implementation of the present invention. Common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.
Claims
1. A cloud service distributed throttling method based on a time series prediction algorithm, characterized by: The following steps are involved: Step S1: Data collection Deploy a data collection module in the distributed architecture of the cloud computing environment to regularly collect traffic data and related performance indicators of each node according to a custom cycle; Step S2: Preprocessing and feature engineering Perform noise data processing on the abnormal traffic collected in specific scenarios, including but not limited to filling missing values, removing outliers, and data normalization and standardization; Customize and extract traffic prediction features based on preprocessed data, including but not limited to historical request rates and request response times; Step S3: Build a prediction model to achieve traffic prediction Identify patterns and trends in traffic data through time series analysis. Based on the analysis results of the time series, select a custom prediction model for training to achieve short-term or long-term traffic forecasts. Step S4: Current limiting parameter management Dynamically adjust the flow control parameters based on the prediction results and distribute the adjusted flow control parameters to all relevant nodes in the cloud computing environment to ensure that each node adopts consistent flow control measures; Step S5: Monitoring and feedback Monitor and collect feedback system performance and traffic data in real time, compare actual conditions with predicted results, and adjust the prediction model based on deviations to achieve continuous optimization and adjustment of the prediction model.
2. The cloud service distributed current limiting method based on the time series prediction algorithm according to claim 1 is characterized by: In step S3, the following steps are performed: Step S3.1: Analyze the cloud service traffic characteristics and customize the time series model and appropriate model parameters based on the traffic characteristics; Step S3.2: Divide the preprocessed historical data into a training set and a test set, and use the training set to train the selected time series model; Step S3.3: Use the test set to evaluate the prediction performance of the model and verify and optimize the trained time series model; Evaluation indicators include mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE).
3. The cloud service distributed current limiting method based on the time series prediction algorithm according to claim 2 is characterized by: In step S3.1, the historical traffic data of the cloud service is analyzed in detail to observe whether there are obvious cyclical fluctuations, long-term trends or seasonal patterns; If the traffic data is stable and shows periodicity and seasonality, the ARIMA model is used; For the ARIMA model, the difference order d, the number of autoregressive terms p, and the number of moving average terms q are determined by the autocorrelation function ACF and partial autocorrelation function PACF graphs; Seasonal parameters are determined using seasonal differences for periodic data, including the seasonal difference order D, the number of seasonal autoregressive terms P, the number of seasonal moving average terms Q, and the length of the seasonal cycle s. The seasonal difference order D is determined by the number of differences. If the traffic data shows complex nonlinear patterns, or if long-term memory is needed to predict long-term trends in traffic, the LSTM model is used. For the LSTM model, select the network structure and training parameters; The network structure includes the number of layers L and the number of neurons in each layer N; The training parameters include the appropriate batch size BatchSize and the learning rate \eta.
4. The cloud service distributed current limiting method based on the time series prediction algorithm according to claim 3 is characterized by: In step S3.2, when the ARIMA model is selected as the time series model, the appropriate difference order, number of autoregressive terms, and number of moving average terms are determined during the training process; When the LSTM model is selected as the time series model, the network structure is designed and the weights are adjusted through the backpropagation algorithm during the training process.
5. The cloud service distributed current limiting method based on the time series prediction algorithm according to claim 3 is characterized by: In step S3.3, the hyperparameters of the time series model are tuned according to the performance evaluation results; When the ARIMA model is selected as the time series model, the information criterion is used to select the optimal parameter combination and minimize the information criterion value; When the LSTM model is used as the time series model, random dropout D or weight regularization λ is applied to reduce overfitting; The model is trained using the backpropagation algorithm and loss function.
6. The cloud service distributed current limiting method based on the time series prediction algorithm according to claim 3 is characterized by: In step S3.3, different training sets and test sets are divided, and a cross-validation method is used to further verify the stability and generalization ability of the model; When the ARIMA model is selected as the time series model, the model fit should be checked through residual analysis to ensure that the residual series is close to the white noise series. The ACF and PACF plots of the residuals should show that the correlation of all lagged terms is close to zero. At the same time, rolling forecasts or k-fold cross-validation should be used to evaluate the generalization ability of the model, and the parameters should be adjusted based on the forecast error to ensure that the optimized parameters should minimize the cross-validation error. When the LSTM model is selected as the time series model, use the validation set or cross-validation to evaluate the model performance and select the model configuration with the smallest error; Use model ensemble techniques to improve forecast stability and accuracy; After the model is deployed, the deviation between its predicted results and actual traffic is monitored in real time, and the model is fine-tuned based on the feedback to maintain or improve the prediction accuracy.
7. A cloud service distributed current limiting system based on a time series prediction algorithm, characterized by: Used to implement the method according to any one of claims 1 to 6, comprising a data acquisition module, a preprocessing and feature engineering module, a prediction model module, a current limiting parameter management module, and a monitoring and feedback module; The data collection module is deployed in the distributed architecture of the cloud computing environment and is responsible for regularly collecting traffic data and related performance indicators of each node according to a custom cycle; The preprocessing and feature engineering module is responsible for processing the noise data of the abnormal traffic collected in specific scenarios, including but not limited to filling missing values, removing outliers, and data normalization and standardization. Customize and extract traffic prediction features based on preprocessed data, including but not limited to historical request rates and request response times; The prediction model module is responsible for identifying patterns and trends in traffic data through time series analysis. Based on the analysis results of the time series, it customizes the prediction model for training to achieve short-term or long-term traffic forecasts. The flow control parameter management module is responsible for dynamically adjusting the flow control parameters based on the prediction results and distributing the adjusted flow control parameters to all relevant nodes in the cloud computing environment to ensure that each node adopts consistent flow control measures; The monitoring and feedback module is responsible for real-time monitoring and collecting feedback system performance and traffic data, comparing the actual situation with the predicted results, and adjusting the prediction model according to the deviation to achieve continuous optimization and adjustment of the prediction model.
8. A cloud service distributed current limiting device based on a time series prediction algorithm, characterized by: The method comprises a memory and a processor; the memory is used to store a computer program, and the processor is used to implement the method according to any one of claims 1 to 6 when executing the computer program.
9. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.