Server load balancing prediction method and system based on deep learning

Abstract

The invention relates to the technical field of server load prediction, in particular to a server load balancing prediction method and system based on deep learning, and the method comprises the steps: obtaining the real-time load data and historical load data of a server; based on the real-time load data and the historical load data, topology embedding transformation is executed; executing group theory transformation according to a topological embedding transformation result; performing spectral decomposition based on a result of the group theory transformation; executing nonlinear transformation according to a spectral decomposition result; based on a nonlinear transformation result, executing time sequence prediction by using a long short-term memory network; and outputting a future load prediction result of the server, and effectively capturing the geometric structure and symmetry of the load data by introducing topology embedding and group theory transformation, thereby not only improving the quality of feature representation, but also enhancing the invariance of the model to data rotation and translation, thereby improving the robustness and generalization ability of prediction, and improving the prediction efficiency. When a load mode with periodicity and symmetry is processed, obvious advantages are shown.

Description

Technical Field

[0001] The present invention relates to the technical field of server load prediction, and more specifically, to a server load balancing prediction method and system based on deep learning. Background Art

[0002] With the rapid development of cloud computing and big data technology, the scale of data centers continues to expand, and server load balancing prediction has become a key technology to ensure the stable operation of the system and improve resource utilization efficiency. Traditional load balancing prediction methods mainly rely on statistical models and simple machine learning algorithms, such as moving average method, exponential smoothing method and autoregressive model. These methods perform well when dealing with linear and short-term trends, but when faced with the complex and changeable load patterns of modern data centers, it is often difficult to capture long-term dependencies and nonlinear characteristics, resulting in insufficient prediction accuracy.

[0003] In recent years, with the rise of deep learning technology, some researchers have tried to apply recurrent neural networks (RNN) and long short-term memory networks (LSTM) to server load prediction. These methods have improved the prediction accuracy to a certain extent, especially showing obvious advantages in processing time series data. However, these methods still have some significant technical problems: first, they usually directly use the original load data as input, ignoring the intrinsic geometric structure and symmetry of the data, which makes it difficult for the model to fully utilize the implicit information in the data; second, these methods have low computational efficiency when processing high-dimensional data, and it is difficult to meet the needs of real-time prediction in large-scale data centers; third, existing methods generally lack effective modeling of multi-scale time dependencies, and it is difficult to capture short-term fluctuations and long-term trends at the same time.

[0004] In addition, existing load prediction methods often treat the prediction problem as a single regression task, ignoring the multidimensional characteristics of load data and the interaction between different features. This makes it difficult for the prediction results to fully reflect the actual operating status of the server, affecting the formulation and implementation of load balancing strategies. At the same time, most existing methods lack the interpretability of the prediction results, making it difficult for system administrators to understand and verify the decision-making basis of the prediction model, reducing the credibility of the prediction results in practical applications.

[0005] In view of the above problems, there is an urgent need for a load balancing prediction method that can fully consider the characteristics of server load data and has high accuracy, high efficiency and good interpretability. The present invention is an innovative solution proposed to meet this demand. Summary of the invention

[0006] The technical problem to be solved by the present invention is how to improve prediction efficiency, enhance the interpretability and adaptability of the model while ensuring prediction accuracy, so as to achieve more intelligent and efficient server load balancing. Specifically, the present invention is committed to solving the following key problems: how to effectively capture the geometric structure and symmetry of load data; how to reduce computational complexity while maintaining high-dimensional information; how to simultaneously model multi-scale time dependencies; and how to improve the interpretability and credibility of prediction results.

[0007] The present invention provides a server load balancing prediction method based on deep learning, comprising:

[0008] The acquisition steps include:

[0009] Get the real-time and historical load data of the server;

[0010] Processing steps include:

[0011] Based on the real-time load data and the historical load data, performing topology embedding transformation;

[0012] Perform group theory transformations based on the results of the topological embedding transformation;

[0013] Based on the results of group theory transformations, spectral decomposition is performed;

[0014] Based on the results of the spectral decomposition, a nonlinear transformation is performed;

[0015] Based on the results of nonlinear transformation, time series prediction is performed using long short-term memory networks;

[0016] Output steps include:

[0017] Output the server's future load prediction results.

[0018] Preferably, the topology embedding transformation specifically includes:

[0019] Map server load data onto a Riemann manifold;

[0020] Topological embedding of data is achieved using Riemann index mapping.

[0021] Preferably, the group theory transformation specifically includes:

[0022] Choose the special orthogonal group SO(n) as the transformation group;

[0023] The rotational invariance of the data is achieved through group action.

[0024] Preferably, the spectral decomposition specifically includes:

[0025] Perform eigenvalue decomposition on the data after group theory transformation;

[0026] Get the eigenvector matrix and the eigenvalue diagonal matrix.

[0027] Preferably, the nonlinear transformation specifically includes:

[0028] Use activation function to perform nonlinear mapping on the spectral decomposition results;

[0029] Compute the trace of the transformed matrix as output.

[0030] Preferably, the long short-term memory network performs time series prediction specifically including:

[0031] Use LSTM network to process data sequences after nonlinear transformation;

[0032] The final prediction result is output through the softmax function.

[0033] Preferably, the data preprocessing step is also included:

[0034] Normalizing the acquired load data;

[0035] Remove outliers and noisy data.

[0036] As an advantage, the model training step is also included:

[0037] Training the LSTM network using historical load data;

[0038] The network parameters are optimized by back-propagation algorithm.

[0039] As a preferred embodiment, the load balancing strategy formulation step is also included:

[0040] Based on the future load prediction result, generating a server resource allocation plan;

[0041] According to the resource allocation scheme, the load distribution of the server cluster is adjusted.

[0042] A server load balancing prediction system based on deep learning for executing the method comprises:

[0043] A data acquisition module is used to obtain real-time load data and historical load data of the server;

[0044] A topology embedding module, used for performing topology embedding transformation based on the real-time load data and the historical load data;

[0045] A group theory transformation module, used to perform group theory transformation according to the result of topological embedding transformation;

[0046] Spectral decomposition module, used to perform spectral decomposition based on the results of group theory transformation;

[0047] A nonlinear transformation module, used for performing nonlinear transformation according to the result of spectral decomposition;

[0048] A time series prediction module is used to perform time series prediction using a long short-term memory network based on the results of nonlinear transformation;

[0049] The output module is used to output the future load prediction results of the server.

[0050] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0051] This paper proposes a new server load balancing prediction method by innovatively combining topology, group theory and deep learning technology. This method is not only significantly better than the existing technology in prediction accuracy, but also has made important breakthroughs in computing efficiency, model adaptability and result interpretability.

[0052] First, the present invention effectively captures the geometric structure and symmetry of load data by introducing topological embedding and group theory transformation. This preprocessing not only improves the quality of feature representation, but also enhances the invariance of the model to data rotation and translation, thereby improving the robustness and generalization ability of prediction. In particular, the method of the present invention shows obvious advantages when dealing with periodic and symmetrical load patterns.

[0053] Secondly, the present invention adopts a combination of spectral decomposition and nonlinear transformation to effectively reduce the computational complexity while retaining the high-dimensional information of the data. This dimensionality reduction method is different from the traditional principal component analysis (PCA). It can retain the nonlinear characteristics of the data and provide richer and more meaningful input for the subsequent deep learning model. Experimental results show that this processing method not only improves the prediction accuracy, but also significantly reduces the calculation time, so that the method of the present invention can meet the needs of real-time prediction in large-scale data centers.

[0054] Furthermore, the present invention successfully models the multi-scale time dependency of load data through a multi-layer LSTM network structure. This structure can capture both short-term fluctuations and long-term trends, making the prediction results more comprehensive and accurate. In particular, when predicting load changes over a long time span (such as 24 hours or longer), the method of the present invention shows obvious advantages, which is of great significance for formulating long-term resource scheduling strategies.

[0055] In addition, the method of the present invention has good interpretability. By analyzing the results of topological embedding and group theory transformation, system administrators can intuitively understand the structural characteristics of the data; by observing the results of spectral decomposition, they can identify the key factors affecting load changes; by analyzing the attention mechanism of the LSTM network, they can understand the key time periods that the model focuses on when making predictions. This interpretability not only enhances the credibility of the prediction results, but also provides valuable insights for system optimization and fault diagnosis.

[0056] Finally, the method of the present invention has strong adaptability and scalability. Through the continuous learning mechanism, the model can automatically adapt to the dynamic changes of the load pattern; through modular design, the system can flexibly adjust or replace certain components according to specific needs to adapt to data centers of different sizes and types.

[0057] In summary, the server load balancing prediction method and system based on deep learning proposed in the present invention have made significant breakthroughs in prediction accuracy, computational efficiency, model adaptability, and result interpretability. These innovations not only solve the key problems faced by existing technologies, but also open up new possibilities for the intelligent management of data centers, and are expected to play an important role in improving resource utilization efficiency, reducing energy consumption, and ensuring service quality, providing strong technical support for server management in the era of cloud computing and big data. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a system flow chart of the present invention.

[0059] Figure 2 It is a logic block diagram of the data acquisition module of the present invention.

[0060] Figure 3 It is a detailed flow chart of the processing steps of the present invention. DETAILED DESCRIPTION

[0061] Please refer to Figure 1-3 The present invention relates to a server load balancing prediction method and system based on deep learning. The method realizes high-precision prediction of server load by innovatively combining topology, group theory and deep learning technology, and provides a reliable basis for the optimal allocation of server resources.

[0062] First, the method of the present invention includes an acquisition step, a processing step, and an output step. In the acquisition step, the method acquires real-time load data and historical load data of the server. These data generally include key indicators (KPIs) such as CPU usage, memory usage, and network traffic. Preferably, the present invention adopts a high-frequency sampling method, such as collecting data once per second, to ensure that subtle trends in load changes are captured. The accuracy of subsequent analysis is ensured. By comprehensively analyzing multiple KPIs, a more comprehensive view of server operation can be obtained, which helps to discover hidden problem patterns or abnormal situations.

[0063] In the processing step, the method first performs a topological embedding transformation based on the acquired real-time load data and historical load data. This innovative step maps the multi-dimensional load data into a geometrically meaningful topological space. Specifically, the present invention uses the following topological embedding function:

[0064] ,

[0065] in, for dimensional Euclidean space, is the number of server load characteristics, is a Riemann manifold. The specific implementation of this function is as follows:

[0066] ,

[0067] here, is the Riemann index map, is a fixed point on the manifold, is an orthogonal basis in the tangent space, The first Through this mapping, high-dimensional data can be converted into a more manageable low-dimensional representation while maintaining the topological structure of the data.

[0068] Here we assume Dimensional Euclidean space Vector in Represents a set of load characteristics of a server at a certain point in time (such as CPU usage, memory usage, etc.). is a space with a specific geometric structure. Mapping these features through the Riemann index Transform to Manifold On, among which is a fixed point on the manifold, and is an orthogonal basis in the tangent space.

[0069] For example, in a data center, if each server has 20 different performance indicators, the original data dimension is very high. Through topological embedding transformation, these 20 features can be mapped to a manifold with low dimensionality but retaining the intrinsic structure of the data, thereby reducing the computational complexity. Consider two servers A and B with similar load characteristics at different time points. After topological embedding transformation, they are on the manifold The representations on are still close, ensuring that the model can capture similar working modes or abnormal situations.

[0070] Next, the method performs group theory transformation based on the result of topological embedding transformation. The purpose of this step is to capture the symmetry in the data and improve the generalization ability of the model. The present invention selects the special orthogonal group SO(n) as the transformation group, where is the dimension of the topological space. The group action is defined as follows:

[0071] ,

[0072] in, is a group element, which can be learned through the network. This transformation can achieve the rotation invariance of the data, making the model robust to the rotation changes of the data.

[0073] For group elements ,in is the dimension of the topological space, which acts on the data after the topological embedding transformation This operation actually changes the perspective by rotating the image, but keeps the internal relationship of the data unchanged.

[0074] Imagine a data center with periodic load variation patterns (such as differences between weekdays and weekends). With group theory transformation, the model can recognize the same pattern even when the data is rotated (i.e., viewed from different angles), improving the stability and reliability of the prediction. For example, when the load distribution of some machines in a server cluster changes due to hardware failure, group theory transformation can help extract the symmetric characteristics behind this change, allowing the model to adapt to the new load pattern more effectively.

[0075] After the group-theoretic transformation, the method performs spectral decomposition based on the transformation results. Spectral decomposition is a powerful technique that can reveal the intrinsic structure of data.

[0076] After completing the spectral decomposition, this method performs a nonlinear transformation based on the decomposition results. This step introduces nonlinearity and enhances the expressiveness of the model. This nonlinear transformation can capture complex patterns in the data and improve the model's fit. Finally, based on the results of the nonlinear transformation, this method uses a long short-term memory network (LSTM) to perform time series forecasting. The LSTM network is a special type of recurrent neural network that is particularly suitable for processing time series data.

[0077] In the output step, the method outputs the future load prediction results of the server. These results can be the load trend in the future period of time (such as the next 1 hour, 24 hours, etc.), providing a basis for load balancing decisions.

[0078] The method of the present invention achieves high-precision prediction of server load by innovatively combining topology, group theory and deep learning technology. Compared with traditional methods, this method has the following advantages: first, the topological embedding transformation retains the geometric structure of the data, which helps to capture the inherent laws of load changes; second, the group theory transformation enhances the rotational invariance of the model and improves the stability of the prediction; third, spectral decomposition and nonlinear transformation effectively extract the key features of the data; finally, the LSTM network handles the temporal dependency of load data well. The synergistic effect of these innovative points enables this method to more accurately predict server load, providing strong support for achieving efficient load balancing.

[0079] In a preferred embodiment of the present invention, the spectral decomposition specifically includes performing eigenvalue decomposition on the data after the group theory transformation, and obtaining an eigenvector matrix and an eigenvalue diagonal matrix. This step is a further processing of the aforementioned group theory transformation results, aiming to reveal the intrinsic structure and main components of the data.

[0080] Specifically, the spectral decomposition operation can be expressed as:

[0081] ,

[0082] in, represents the spectral decomposition operation, is the data after group theory transformation, is the eigenvector matrix, is a diagonal matrix of eigenvalues. Through this decomposition, the method of the present invention can decompose a complex data structure into a series of orthogonal basis vectors (eigenvectors) and their corresponding importance weights (eigenvalues).

[0083] here, is the eigenvector matrix, is the eigenvalue diagonal matrix. Through spectral decomposition, the main components and change patterns of the data can be obtained, providing important information for subsequent prediction tasks.

[0084] Spectral decomposition is the process of performing eigenvalue decomposition on the data matrix after group theory transformation. is the data matrix after group theory transformation, then it can be decomposed into ,in is an orthogonal matrix consisting of eigenvectors, is a diagonal matrix containing the eigenvalues ​​sorted by size.

[0085] In a large data center, there may be a data set with tens of thousands of records. By selecting the eigenvectors corresponding to the largest eigenvalues, the dimensionality of the data can be significantly reduced while almost no loss of any useful information. For example, retaining only the first 50 principal components may be sufficient to describe more than 95% of the variance. In a practical environment, server load data will inevitably contain some random fluctuations or measurement errors. Spectral decomposition helps to remove the parts corresponding to these small eigenvalues, reduce noise interference, and improve prediction accuracy.

[0086] Consider a data center with 100 servers, each with 20 performance indicators, forming a 100×20 data matrix. After spectral decomposition, it is found that the first 10 eigenvalues ​​account for 98% of the total variance, so the corresponding 10 eigenvectors can be selected as the basis of the new feature space, forming a 100×10 reduced-dimensional data representation. This not only simplifies the problem, but also improves computational efficiency.

[0087] There may be a lot of redundant information or noise in the load data of the data center. Helps to separate out the main components, allowing the model to focus on truly meaningful patterns without being distracted by irrelevant details.

[0088] For example, load changes on certain servers may be caused by the same type of user activity, such as visiting a specific web page or running similar applications. The eigenvector matrix can identify these common behavior patterns and classify them into a few important eigenvectors, thereby reducing the model's sensitivity to noisy data.

[0089] Eigenvalue diagonal matrix It provides a quantitative way to evaluate the importance of each eigenvector. By selecting the eigenvectors corresponding to the largest eigenvalues, the dimension of the data can be significantly reduced without losing any useful information.

[0090] Continuing with the data center example mentioned earlier, if you want to further compress the data, you can set a threshold (such as 95% of the total variance) and then select all eigenvalues ​​and their corresponding eigenvectors that are greater than this threshold. This will not only significantly reduce the amount of data, but also ensure that the most critical information is retained.

[0091] Small eigenvalues ​​usually correspond to random fluctuations or measurement errors in the data. By ignoring these eigenvalues ​​and their corresponding eigenvectors, noise can be effectively removed and prediction accuracy can be improved.

[0092] In actual operation, data centers may encounter hardware failures, network jitters and other problems, resulting in abnormal fluctuations in the load data of some servers. The eigenvalue diagonal matrix helps identify and filter out these atypical data points, making the model more robust to capture normal working patterns.

[0093] On a large cloud computing platform, administrators want to adjust the resource allocation plan of the server cluster through a multi-objective optimization algorithm. In this process, spectral decomposition plays a vital role:

[0094] Initial stage: Collect server load data over a period of time and build a high-dimensional data matrix .

[0095] Spectral decomposition: Perform spectral decomposition on the data matrix to obtain the eigenvector matrix and the eigenvalue diagonal matrix .

[0096] Dimensionality reduction and information concentration: Select the most important principal components based on the size of the eigenvalues ​​and construct a new low-dimensional data representation.

[0097] Multi-objective optimization: Using the reduced-dimensional data as input, multiple non-dominated solutions (Pareto optimal solutions) are generated through multi-objective optimization algorithms such as NSGA-II, providing decision makers with a variety of feasible options. Through the above steps, not only the complexity of the problem is simplified and the computational efficiency is improved, but also the model's ability to understand the real-world load pattern is enhanced, ultimately achieving more efficient and stable resource management and load balancing.

[0098] Preferably, the present invention uses a singular value decomposition (SVD) algorithm to achieve spectral decomposition. The SVD algorithm has the characteristics of good numerical stability and high computational efficiency, and is particularly suitable for processing large-scale server load data. In practical applications, full SVD or truncated SVD can be selected according to specific needs to achieve a balance between computational efficiency and information retention.

[0099] When performing spectral decomposition, the system of the present invention usually retains principal components that account for more than 95% of the total variance. This threshold is an empirical value obtained through a large number of experiments, which can effectively reduce the dimension of the data while retaining key information and improve the efficiency of subsequent processing.

[0100] Next, the method of the present invention performs a nonlinear transformation based on the result of the spectral decomposition. This step introduces nonlinearity and significantly enhances the expressive power of the model. The specific form of the nonlinear transformation is as follows:

[0101] ,

[0102] In this formula, represents the nonlinear transformation function, is the activation function, tr represents the trace operation of the matrix, is the weight matrix, The present invention preferably uses ReLU (Rectified Linear Unit) as the activation function, because ReLU has the advantages of simple calculation and stable gradient, and is particularly suitable for the training of deep networks.

[0103] Nonlinear transformation introduces activation function (such as ReLU), weight matrix , bias , and the trace operation tr. The trace operation is used to compress a high-dimensional matrix into a scalar output while retaining important features of the matrix. In this example, is the matrix after spectral decomposition.

[0104] Take a complex server load scenario as an example, which has not only regular periodic fluctuations but also sudden traffic peaks. Nonlinear transformation allows the model to fit this complex functional relationship and improves the ability to capture this type of load pattern. By adjusting the weight matrix , the model can dynamically control the importance of each feature based on the knowledge learned during training. For example, if a feature (such as network latency) is found to be particularly important for predicting future load trends, the prediction performance can be optimized by increasing the weight corresponding to the feature.

[0105] It is worth noting that the present invention uses matrix trace operations in nonlinear transformations. This is an innovation that can effectively compress high-dimensional matrices into scalars while retaining important features of the matrix. , the method of the present invention can flexibly control the importance of each feature.

[0106] In one embodiment of the present invention, the initial value of the weight matrix W is set using the Xavier initialization method. This initialization method can make the output variance of each layer as equal as possible, which helps to prevent the gradient disappearance or explosion problem and accelerate the convergence of the network.

[0107] After completing the nonlinear transformation, the method of the present invention uses a long short-term memory network (LSTM) to perform time series prediction. The LSTM network is a special recurrent neural network, the core of which is the introduction of a gate mechanism that can effectively handle long-term dependency problems. The prediction process can be expressed as:

[0108] ,

[0109] ,

[0110] here, is the hidden state, is the unit state, To predict the output. Through the LSTM network, this method can effectively capture the long-term dependencies of the load data and improve the accuracy of the prediction. The internal structure of the LSTM unit includes an input gate, a forget gate, and an output gate. These gated units enable the LSTM to dynamically adjust the retention and forgetting of information according to the importance of the current input and historical information.

[0111] Long Short-Term Memory (LSTM) is a recurrent neural network designed specifically to process sequence data. The LSTM unit contains an input gate, a forget gate, and an output gate, which work together to decide which information should be remembered or forgotten. The hidden state at the current time step Calculated by the softmax layer.

[0112] Considering that the load of a data center usually has certain periodicity and trends (such as the traffic surge during the morning rush hour every day), LSTM is particularly good at handling dependencies over long time spans. This makes it possible to accurately predict load trends in the next few hours or even days.

[0113] Preferably, the LSTM network of the present invention adopts a bidirectional structure. The bidirectional LSTM structure can consider both past and future context information, which is particularly effective in capturing the periodic patterns and long-term trends of server load. For example, it can better identify the load difference between weekdays and weekends, or capture the load changes during special periods such as holidays.

[0114] In the network training process, the present invention adopts the Adam optimizer, the learning rate is initially set to 0.001, and the learning rate decay strategy is used. Specifically, after every 10 epochs, the learning rate decays to 0.9 times the original. This strategy can quickly approach the optimal solution in the early stage of training, and carefully adjust the parameters in the later stage of training to improve the performance of the model.

[0115] In addition, in order to prevent overfitting, the present invention introduces the Dropout regularization technology in the LSTM network. The Dropout rate is set to 0.5, which is an effective value widely verified in practice, which can significantly improve the generalization ability of the model while maintaining its expressive ability.

[0116] Through the above series of innovative processing steps, the method of the present invention can comprehensively and deeply analyze the characteristics of server load data, capture the complex patterns and long-term dependencies therein, and thus achieve high-precision load prediction. This provides reliable data support for the optimal allocation of server resources and the formulation of load balancing strategies, and helps to improve the operating efficiency and service quality of data centers. In another preferred embodiment of the present invention, the method also includes a model training step. This step is crucial to ensuring the performance of the prediction model. Specifically, the present invention uses historical load data to train the long short-term memory network and optimizes the network parameters through a back propagation algorithm.

[0117] During the model training process, the present invention adopts batch gradient descent method, and the number of samples in each batch is set to 64. This batch size is an empirical value obtained through multiple experiments, which can achieve a good balance between training stability and computational efficiency. The training data is divided into training set, validation set and test set in a ratio of 8:1:1. This division method can make full use of the data while ensuring that the generalization ability of the model is effectively evaluated.

[0118] Preferably, the present invention introduces an early stopping strategy during the training process. Specifically, if the loss on the validation set does not decrease for 10 consecutive epochs, the training is stopped. This strategy can effectively prevent overfitting while saving computing resources. In addition, the present invention also adopts a learning rate adaptive adjustment technology. The initial learning rate is set to 0.001. When the validation set loss does not decrease for 3 consecutive epochs, the learning rate is automatically reduced to half of the original value. This dynamic adjustment strategy can help the model converge to the optimal solution faster.

[0119] After training is completed, the present invention evaluates the model performance on the test set. The evaluation indicators include root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Generally, the method of the present invention can control the MAPE within 5%, which means that the prediction results have high practical value.

[0120] An important feature of the present invention is that model training is not a one-time process. In practical applications, the system of the present invention will retrain the model regularly (e.g., weekly) using newly collected data. This continuous learning mechanism enables the model to adapt to changes in server load patterns in a timely manner and maintain the accuracy of predictions.

[0121] After completing the load prediction, the method of the present invention further includes a load balancing strategy formulation step. This step converts the prediction results into actual resource scheduling decisions, which is a key link in the practical application of the present invention.

[0122] Specifically, the present invention generates a server resource allocation scheme based on the future load prediction results. This process involves the weighing of multiple factors, including but not limited to: predicted load level, current state of the server, energy efficiency, service quality requirements, etc. The present invention uses a multi-objective optimization algorithm to generate a resource allocation scheme, and its objective function can be expressed as:

[0123] ,

[0124] in, represents the resource allocation scheme, Indicates The present invention preferably uses the NSGA-II (Non-dominated Sorting Genetic Algorithm II) algorithm to solve this multi-objective optimization problem. The NSGA-11 algorithm can obtain multiple non-dominated solutions in one run, providing decision makers with multiple options.

[0125] Each optimization goal Each is a function of a specific performance metric that together determine the quality of the final resource allocation solution. These goals may include, but are not limited to:

[0126] Load balancing: Ensure that the workload on each server is distributed as evenly as possible.

[0127] Energy efficiency: Minimize power consumption throughout the data center.

[0128] Response time: Minimize the average processing time of user requests.

[0129] Quality of Service (QoS): Ensure that mission-critical and service priorities are met.

[0130] For the above four goals, the following specific mathematical expressions can be defined:

[0131] Load balancing: Assumption is the number of servers, For the The load balancing degree can be calculated by the standard deviation To measure:

[0132] ,

[0133] in, is the average load of all servers.

[0134] Energy efficiency: For the Servers in The power consumption at the moment, the total energy consumption This can be given by integration or summation:

[0135] or , response time: For the The average response time is for:

[0136] ,

[0137] in The number of requests.

[0138] Service quality: Considering the importance and urgency of different services, weighting factors can be introduced To evaluate the overall service quality

[0139] ,

[0140] in For the Quality rating of the service.

[0141] The present invention preferably uses NSGA-II (Non-dominated Sorting Genetic Algorithm II) to solve the above multi-objective optimization problem. NSGA-II is an evolutionary algorithm that can find a set of non-dominated solutions (i.e., Pareto optimal solutions) in one run, thereby providing decision makers with multiple feasible options.

[0142] The working principle of NSGA-II is as follows:

[0143] Initialize the population: Randomly generate a set of initial resource allocation plans as the first generation population.

[0144] Evaluate individual fitness: Calculate the value of each solution for each objective function and classify the individuals according to fast non-dominated sorting.

[0145] Selection operation: Select excellent individuals from the current population to participate in crossover and mutation to produce the next generation population.

[0146] Crossover and mutation: Gene exchange (crossover) and random changes (mutation) of selected individuals to create new solutions.

[0147] Environmental selection: merge the parent and offspring populations, and then select the next generation population members by crowding distance sorting.

[0148] Termination condition: When the preset maximum number of iterations is reached or other stopping criteria are met, the optimal solution set is output.

[0149] On a large cloud computing platform, administrators want to optimize load balancing, energy efficiency, response time and service quality at the same time. By using the NSGA-II algorithm, the system can generate a series of different resource allocation schemes, each of which corresponds to a specific set of optimization results. For example:

[0150] Solution A: Focuses on load balancing and energy efficiency, and is suitable for use during daily operations to maintain low costs and stable performance.

[0151] Solution B: Emphasizes response time and service quality. It is suitable for peak hours or important business periods to ensure that user experience is not affected.

[0152] Plan C: Balance various indicators and provide fully optimized resource allocation under normal circumstances.

[0153] Using multi-objective optimization algorithms to generate resource allocation plans has the following significant advantages:

[0154] Flexibility: Based on the changes in demand in different time periods, you can choose the optimization strategy that best suits the current situation.

[0155] Robustness: By considering multiple optimization objectives, the risks that may be caused by a single objective are reduced, and the stability and reliability of the system are improved.

[0156] Adaptive adjustment: As new data continues to flow in, the model can continue to learn and optimize its decision-making process, gradually improving the effect of load balancing.

[0157] Transparency and control: Provides managers with intuitive decision support tools that enable them to flexibly adjust resource allocation based on business priorities.

[0158] Through carefully designed objective functions and advanced multi-objective optimization algorithms, the present invention not only achieves high-precision server load prediction, but also provides a scientific basis for dynamically adjusting resource allocation, which helps to improve the overall operational efficiency and service quality of the data center.

[0159] According to the generated resource allocation scheme, the method of the present invention adjusts the load distribution of the server cluster. This may involve specific operations such as virtual machine migration, container rescheduling, load balancer configuration adjustment, etc. Preferably, the present invention adopts a progressive adjustment strategy, that is, only small adjustments are made within each time window (such as 5 minutes). This strategy can reduce the overhead caused by frequent large-scale migrations while maintaining the stability of the system.

[0160] It is worth noting that the load balancing strategy of the present invention not only considers the current and predicted load conditions, but also considers the effects of historical scheduling decisions. By introducing a reinforcement learning mechanism, the system of the present invention can continuously optimize its decision-making strategy and gradually improve the effect of load balancing.

[0161] Finally, the present invention also proposes a server load balancing prediction system based on deep learning. The system includes the following modules:

[0162] The data acquisition module 1 is used to obtain the real-time load data and historical load data of the server. This module can collect data through server monitoring agents, log collectors, etc., and perform preliminary data cleaning and formatting.

[0163] The topology embedding module 2 is used to perform topology embedding transformation based on the real-time load data and the historical load data. This module implements the topology embedding function described above and maps the high-dimensional load data into a topological space with geometric meaning.

[0164] The group theory transformation module 3 is used to perform group theory transformation according to the result of topological embedding transformation. This module realizes the rotation invariance processing of data through the group action of the special orthogonal group SO(n).

[0165] The spectral decomposition module 4 is used to perform spectral decomposition based on the result of group theory transformation. This module performs eigenvalue decomposition on the data to reveal the intrinsic structure of the data.

[0166] The nonlinear transformation module 5 is used to perform nonlinear transformation according to the result of spectral decomposition. This module enhances the expression ability of the model by introducing a nonlinear activation function.

[0167] The time series prediction module 6 is used to perform time series prediction based on the result of nonlinear transformation using a long short-term memory network. This module is the core of the entire system and is responsible for generating the final load prediction result.

[0168] The output module 7 is used to output the prediction results of the future server load. This module is not only responsible for the output of the results, but also includes the visualization and interpretation functions of the results, so that the administrator can understand and use the prediction results.

[0169] These modules work closely together to form a complete load prediction and balancing system. Through this modular design, the system of the present invention has good scalability and maintainability. For example, a specific module can be replaced or upgraded as needed without affecting the operation of the overall system.

[0170] In general, the server load balancing prediction method and system based on deep learning proposed in the present invention realizes high-precision load prediction and intelligent load balancing by innovatively combining topology, group theory, spectral decomposition and deep learning technology. This can not only improve the resource utilization efficiency of data centers, but also significantly improve service quality, providing strong technical support for server management in the era of cloud computing and big data.

[0171] In order to verify the superiority of the server load balancing prediction method and system based on deep learning of the present invention, a set of comparative experiments was designed. The experimental environment is a medium-sized data center, including 100 servers, the load data collection cycle is 5 minutes, and the prediction time span is 24 hours. The following is a detailed description of the embodiment and comparative example and the experimental results.

[0172] Example 1: Complete method of the present invention

[0173] In this embodiment, the method proposed by the present invention is fully implemented, including all steps such as topological embedding, group theory transformation, spectral decomposition, nonlinear transformation and LSTM network prediction. The system parameters are set as follows: topological embedding uses Riemann manifold, group theory transformation uses SO (3) group, the number of LSTM network layers is 3, and the number of hidden units is 128.

[0174] Comparative Example 1: Traditional Time Series Forecasting Method

[0175] This comparison uses the ARIMA (Autoregressive Integrated Moving Average) model for load forecasting. ARIMA is a classic time series forecasting method that is widely used in various forecasting tasks.

[0176] Comparative Example 2: Simple Deep Learning Method

[0177] This comparative example uses a single-layer LSTM network to directly predict the original load data, and does not include the preprocessing steps such as topological embedding and group theory transformation proposed in the present invention. The parameters of the LSTM network are consistent with those in Example 1.

[0178] The following three metrics were selected to evaluate the prediction performance:

[0179] 1.MAPE (mean absolute percentage error): reflects the average deviation between the predicted value and the actual value.

[0180] 2. RMSE (root mean square error): measures the average deviation between the predicted value and the actual value.

[0181] 3. Forecast time: The computation time required to complete a 24-hour load forecast.

[0182] The experimental results are shown in Table 1 below:

[0183] Table 1 Comparison of experimental results of Example 1, Comparative Example 1 and Comparative Example 2: method MAPE(%) RMSE Prediction time (s) Example 1 3.2 0.045 2.5 Comparative Example 1 8.7 0.112 1.8 Comparative Example 2 5.9 0.078 3.2

[0184] It can be seen from the experimental results in Table 1 that the method of the present invention (Example 1) is significantly better than the other two methods in terms of prediction accuracy. The specific analysis is as follows:

[0185] 1.MAPE: The method of the present invention controls MAPE at 3.2%, which is much lower than 8.7% of the traditional ARIMA method and 5.9% of the simple LSTM method. This means that the prediction results of the present invention are closer to the actual load situation and can provide a more reliable basis for load balancing decisions.

[0186] 2. RMSE: The RMSE of the method of the present invention is 0.045, which is also significantly ahead of the other two methods. The lower RMSE indicates that the prediction results of the present invention have less fluctuation and higher stability.

[0187] 3. Prediction time: Although the method of the present invention is more computationally complex than the traditional ARIMA method, the prediction time is only 0.7 seconds longer than ARIMA and 0.7 seconds faster than the simple LSTM method. Considering the significant improvement in prediction accuracy, this slight increase in time is completely acceptable.

[0188] These results fully demonstrate the effectiveness of the innovations proposed in this invention:

[0189] 1. Topological embedding and group theory transformation: These two steps effectively capture the geometric structure and symmetry of the load data, providing more meaningful feature representation for subsequent predictions. This explains why the method of the present invention can significantly improve the prediction accuracy.

[0190] 2. Spectral decomposition and nonlinear transformation: These steps further extract the key features of the data and introduce nonlinear expression capabilities, which enables the method of the present invention to handle complex load patterns such as periodic fluctuations, sudden peaks, etc.

[0191] 3. Multi-layer LSTM structure: Compared with the simple single-layer LSTM (Comparative Example 2), the multi-layer structure of the present invention can learn more complex time dependencies, thereby improving the accuracy of long-term predictions.

[0192] It is noteworthy that the method of the present invention maintains a high prediction accuracy while maintaining a good computational efficiency, thanks to a series of optimization strategies adopted in the algorithm design, such as using SVD for spectral decomposition and matrix trace operation for dimensionality reduction.

[0193] In general, this set of experimental results strongly proves the superiority of the present invention in the server load prediction task. Compared with traditional methods and simple deep learning methods, the present invention not only significantly improves the prediction accuracy, but also maintains acceptable computational efficiency. This means that in practical applications, the method of the present invention can more accurately predict server loads, thereby formulating better load balancing strategies, and ultimately improving the resource utilization efficiency and service quality of the data center.

[0194] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A server load balancing prediction method based on deep learning, characterized in that: include: The acquisition steps include: Get the real-time and historical load data of the server; Processing steps include: Based on the real-time load data and the historical load data, performing topology embedding transformation; Perform group theory transformations based on the results of the topological embedding transformation; Based on the results of group theory transformations, spectral decomposition is performed; Based on the results of the spectral decomposition, a nonlinear transformation is performed; Based on the results of nonlinear transformation, time series prediction is performed using long short-term memory networks; Output steps include: Output the server's future load prediction results.

2. The method according to claim 1, characterized in that The topology embedding transformation specifically includes: Map server load data onto a Riemann manifold; Topological embedding of data is achieved using Riemann index mapping.

3. The method according to claim 1, characterized in that The group theory transformation specifically includes: Choose the special orthogonal group SO(n) as the transformation group; The rotational invariance of the data is achieved through group action.

4. The method according to claim 1, characterized in that: The spectrum decomposition specifically includes: Perform eigenvalue decomposition on the data after group theory transformation; Get the eigenvector matrix and the eigenvalue diagonal matrix.

5. The method according to claim 1, characterized in that The nonlinear transformation specifically includes: Use activation function to perform nonlinear mapping on the spectral decomposition results; Compute the trace of the transformed matrix as output.

6. The method according to claim 1, characterized in that The long short-term memory network performs time series prediction specifically including: Use LSTM network to process data sequences after nonlinear transformation; The final prediction result is output through the softmax function.

7. The method according to claim 1, characterized in that It also includes data preprocessing steps: Normalizing the acquired load data; Remove outliers and noisy data.

8. The method according to claim 1, characterized in that Also includes the model training steps: Training the LSTM network using historical load data; The network parameters are optimized by back-propagation algorithm.

9. The method according to claim 1, characterized in that: It also includes the steps for developing a load balancing strategy: Based on the future load prediction result, generating a server resource allocation plan; According to the resource allocation scheme, the load distribution of the server cluster is adjusted.

10. A server load balancing prediction system based on deep learning that executes the method according to any one of claims 1 to 9, characterized in that: include: A data acquisition module is used to obtain real-time load data and historical load data of the server; A topology embedding module, used for performing topology embedding transformation based on the real-time load data and the historical load data; A group theory transformation module, used to perform group theory transformation according to the result of topological embedding transformation; Spectral decomposition module, used to perform spectral decomposition based on the results of group theory transformation; A nonlinear transformation module, used for performing nonlinear transformation according to the result of spectral decomposition; A time series prediction module is used to perform time series prediction using a long short-term memory network based on the results of nonlinear transformation; The output module is used to output the future load prediction results of the server.

Images