Container cloud node load prediction method combining grey dominance analysis and neural network

Abstract

The invention provides a container cloud node load prediction method combining gray dominance analysis and a neural network, and the method comprises the steps: obtaining a prediction result through gray dominance analysis when the collection frequency does not reach a preset frequency, training a neural network through a gray dominance analysis result, and carrying out the prediction when the prediction result is greater than or equal to a preset value. Carrying out capacity expansion operation on the container cloud nodes, otherwise, not carrying out capacity expansion operation on the container cloud nodes, and considering small sample scene prediction; and after the preset number of times is reached, combining the grey dominance analysis and the neural network prediction result to obtain a comprehensive prediction result, when the comprehensive prediction result is greater than or equal to a preset value, carrying out capacity expansion operation on the container cloud node, otherwise, not carrying out capacity expansion operation on the container cloud node, and utilizing grey analysis to deal with small samples and quickly respond. Meanwhile, by means of neural network optimization, prediction precision and timeliness are improved, capacity expansion is executed according to different prediction results, and reasonable scheduling of container cloud node resources is effectively guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cloud computing resource scheduling, and more particularly to a container cloud node load prediction method combining gray advantage analysis and neural network. BACKGROUND

[0002] In the field of container cloud technology, node load prediction is crucial to ensure system stability and efficient operation. In the container cloud environment, it is necessary to accurately predict node load to reasonably schedule resources and avoid failures. Existing container cloud node load prediction mainly relies on a single model, such as Long Short-Term Memory (LSTM) or Autoregressive Integrated Moving Average (ARIMA), which has obvious defects: First, the prediction accuracy is low in small sample scenarios. Traditional neural networks (such as LSTM) rely on a large amount of data for training. In the cold start phase of the container cloud or when data collection is insufficient, the model cannot learn the rules due to the lack of sufficient samples, resulting in significant prediction errors and an inability to accurately reflect the node load state. Second, the dynamic adaptability is insufficient. Container clouds are prone to sudden traffic, such as business peaks and instantaneous high-concurrency requests. Single models are difficult to quickly adjust parameters and respond to load changes, leading to prediction lag and an inability to gain effective time for resource scheduling. Finally, there is a waste of computing resources. Node load prediction involves CPU usage, memory usage, and other multi-dimensional features. Directly inputting high-dimensional features into a single model significantly increases model complexity, lengthens inference time, and causes excessive consumption of computing resources, affecting the overall performance of the container cloud.

[0003] Therefore, there is an urgent need for a new container cloud node load prediction method to break through the limitations of single models, accurately predict in small sample scenarios, enhance dynamic adaptability, optimize computing resource utilization, and ensure stable and efficient operation of the container cloud. SUMMARY

[0004] To solve the problems of low prediction accuracy, insufficient dynamic adaptability, and waste of computing resources in existing small sample scenarios, the present application proposes a container cloud node load prediction method combining gray advantage analysis and neural network.

[0005] To achieve the above technical effects, the technical solutions of the present application are as follows: A container cloud node load prediction method combining gray advantage analysis and neural network, comprising the following steps: Step 1: Collect container cloud node parameters according to a predetermined time frequency; Step 2: Determine whether the number of node parameter collection times is less than the preset number. If yes, execute step 3, otherwise execute step 5; Step 3: Perform gray advantage analysis on the node parameters to obtain a prediction result L greyand according to the grey advantage analysis result, a neural network is trained and constructed; Step 4: judging the prediction result L grey whether greater than or equal to a preset value, if yes, the container cloud node is expanded, otherwise, the container cloud node is not expanded; Step 5: the node parameters are subjected to grey advantage analysis and prediction by using the trained neural network to obtain a comprehensive prediction result L final Step 6: judging the comprehensive prediction result L final whether greater than or equal to a preset value, if yes, the container cloud node is expanded, otherwise, the container cloud node is not expanded.

[0006] Compared with the prior art, the beneficial effects of the technical scheme of the present application are: The present application proposes a container cloud node load prediction method combining grey advantage analysis and neural network, which combines grey advantage analysis and neural network, obtains a prediction result by grey advantage analysis when the number of collection times does not reach a preset number, trains a neural network according to the grey advantage analysis result, expands the container cloud node when the prediction result is greater than or equal to a preset value, otherwise, does not expand the container cloud node, and takes into account small sample scene prediction; after the preset number is reached, a comprehensive prediction result is obtained by combining the grey advantage analysis and the neural network prediction result, the container cloud node is expanded when the comprehensive prediction result is greater than or equal to a preset value, otherwise, the container cloud node is not expanded, the grey analysis is used to cope with small samples and fast response, and the neural network is used for optimization to improve the prediction accuracy and timeliness, and the expansion is performed according to different prediction results, which effectively guarantees the reasonable scheduling of container cloud node resources, and solves the problems of single model in small sample, dynamic adaptation and resource utilization, etc. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 The grey advantage analysis and neural network combined container cloud node load prediction method flowchart shown in the embodiment of the present application.

[0008] Figure 2 The neural network structure diagram shown in the embodiment of the present application. DETAILED DESCRIPTION

[0009] ​The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description of the exemplary embodiments is intended to apply to any exemplary embodiment, unless specified otherwise. It is understood that the description of the exemplary embodiments is intended to be illustrative, and not in a limiting sense, as many other embodiments of the present application are possible.

[0010] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0011] It is to be understood that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0012] The present application will be described in detail herein with reference to the attached drawings and specific embodiments.

[0013] Embodiment 1 This embodiment proposes a container cloud node load prediction method combining gray advantage analysis and neural network, the flow chart thereof is as shown in Figure 1 The method comprises the following steps: Step 1: Collecting container cloud node parameters according to a predetermined time frequency; Step 2: Determining whether the number of node parameter collection is less than a preset number, if yes, executing step 3, otherwise executing step 5; Step 3: Performing gray advantage analysis on the node parameters to obtain a prediction result L grey , and training the constructed neural network according to the gray advantage analysis result; Step 4: Determining whether the prediction result L grey is greater than or equal to a preset value, if yes, performing expansion operation on the container cloud node, otherwise, not performing expansion operation on the container cloud node; Step 5: Performing gray advantage analysis on the node parameters and prediction using the trained neural network to obtain a comprehensive prediction result Lfinal ; Step 6: judging the comprehensive prediction result L final whether greater than or equal to a preset value, if yes, expanding the container cloud node, otherwise, not expanding the container cloud node.

[0014] In the embodiment, when the collection times do not reach the preset times, the prediction result is obtained through the grey advantage analysis, the neural network is trained according to the grey advantage analysis result, when the prediction result is greater than or equal to the preset value, the container cloud node is expanded, otherwise, the container cloud node is not expanded, the small sample scene prediction is considered; after the preset times are reached, the grey advantage analysis and the neural network prediction result are combined to obtain a comprehensive prediction result, when the comprehensive prediction result is greater than or equal to the preset value, the container cloud node is expanded, otherwise, the container cloud node is not expanded, the grey analysis is used to cope with the small sample and fast response, at the same time, the neural network optimization is used to improve the prediction accuracy and timeliness, and according to different prediction results, the expansion is performed, the container cloud node resource reasonable scheduling is effectively guaranteed, and the problems of the existing single model in small sample, dynamic adaptation and resource utilization are solved.

[0015] Embodiment 2 The embodiment explains the case when the collection times are less than the preset times in detail.

[0016] In an optional embodiment, the container cloud node parameters include CPU usage, memory usage, network traffic and active connection number.

[0017] In an optional embodiment, the CPU usage represents the ratio of the used CPU and the total CPU, the memory usage represents the ratio of the used memory and the total memory, the network traffic represents the total amount of data transmitted by the network interface of the node, and the active connection number represents the number of TCP / UDP connections currently established and maintained in the active state with the node.

[0018] Further, the node parameters are collected every minute, and when the number of collection times N is less than 1000, the grey advantage analysis is used for load prediction, and the last 5 collected data are taken as a sequence for grey advantage analysis to predict the node expansion probability, x0(k)=[x0(1),x0(2),x0(3),x0(4),x0(5)], k=1, 2, …, 5 is the CPU usage rate sequence, x0(1) represents the node CPU usage rate collected for the first time in the last 5 collected data, wherein x1(k)=[x1(1),x1(2),x1(3),x1(4),x1(5)] represents the memory usage rate sequence in the last 5 collected data, x2(k)=[x2(1),x2(2),x2(3),x2(4),x2(5)] represents the network traffic sequence in the last 5 collected data, and x3(k)=[x3(1),x3(2),x3(3),x3(4),x3(5)] is the active connection sequence.

[0019] In an optional embodiment, the node parameters are subjected to grey advantage analysis to obtain a prediction result L grey comprising the following steps: taking the last collected container cloud node parameters as a sequence, and normalizing each parameter sequence; taking the node CPU usage rate sequence after normalization as a reference sequence, and taking the sequence obtained by normalizing the node memory usage rate sequence, the sequence obtained by normalizing the network traffic sequence, and the sequence obtained by normalizing the active connection number sequence as comparison sequences; obtaining the absolute difference between the comparison sequence parameters and the reference sequence parameters according to the reference sequence and the comparison sequence; obtaining the grey correlation degree according to the minimum and maximum values of the absolute difference between the comparison sequence parameters and the reference sequence parameters and the absolute difference between the comparison sequence parameters and the reference sequence parameters; obtaining the prediction result according to the parameter with the largest correlation degree and through the accumulated sequence and the grey micro equation L grey .

[0020] In an optional embodiment, the expression of the normalization processing of each parameter sequence is:

[0021] wherein, x’ represents the normalized value of each parameter, and the value range is [0, 1], x represents the original data of each parameter.

[0022] Furthermore, min(x) is the minimum value in the original data sequence of each set of parameters, such as the minimum value in the node CPU utilization sequence, and max(x) is the maximum value in the original data sequence of each set of parameters, such as the maximum value in the node CPU utilization sequence.

[0023] In one optional embodiment, the grey relational degree is obtained based on the minimum and maximum values ​​of the absolute differences between the comparison sequence parameters and the reference sequence parameters; its expression is:

[0024] in, x’ 0 represents the reference sequence. x’ i Indicates a comparison sequence, Indicates the first k The comparison sequence parameters of the second acquisition and the first acquisition k The absolute difference of the reference sequence parameters from the next acquisition; The first k The comparison sequence parameters of the second acquisition and the first acquisition k The absolute difference of the reference sequence parameters acquired in each acquisition; its expression is:

[0025] in, x’ 0 represents the reference sequence. x’ i Indicates a comparison sequence, k Indicates the first k Second-rate.

[0026] Furthermore, the normalized node CPU utilization is used as a reference sequence, denoted as x'0(k) = [x'0(1), x'0(2), x'0(3), x'0(4), x'0(5)], k = 1, 2, ..., 5, where k represents the number of times. That is, x'0(1) represents the normalized value of the node CPU utilization collected for the first time in the last 5 data collections, x'0(2) represents the node CPU utilization collected for the second time in the last 5 data collections, and so on.

[0027] The comparison sequence, denoted as x', consists of normalized node memory usage, normalized network traffic, and normalized active connection count. i (k), i=1, 2, 3, k=1, 2, ..., 5, x'1(k) represents the normalized value of memory usage collected in the kth time, x'2(k) represents the normalized value of network traffic collected in the kth time, and x'3(k) represents the normalized value of active connection count collected in the kth time.

[0028] Let be the absolute difference between the comparison sequence parameters acquired in the k-th acquisition and the reference sequence parameters acquired in the k-th acquisition. for The minimum value in, for The maximum value in.

[0029] Furthermore, the grey relational degree of all comparison sequences and reference sequences is calculated. And sort them to obtain the correlation degree. Let i = 1, 2, 3. Select the sequence with the highest correlation to the reference sequence. Parameters (such as the number of active connections) are used as strong correlation terms for grey prediction.

[0030] The initial sequence is denoted as and the parameter-normalized sequence is denoted as and the sequence with the highest correlation to the reference sequence is denoted as . .

[0031] In one alternative embodiment, the prediction result is obtained based on the parameter with the highest correlation and through a cumulative sequence and a grey microequation. L grey ; k+1 The expression for time is:

[0032] in, L grey This indicates the load forecast result. Indicates the cumulative sequence at the th k+1 The value at time; The cumulative sequence expression is:

[0033] in, x’ (0) Indicates the initial sequence; The expression for the gray differential equation is:

[0034] in, x’ (1) Represents an accumulated sequence. a Represents the development coefficient. b Indicates the amount of gray action. t Represents a time variable.

[0035] in, , , .

[0036] Furthermore, the cumulative sequence is defined in the first... k+1 value of time The result of grey advantage analysis is the predicted initial sequence.k+1 The value of the time When , the node needs to be expanded.

[0037] When the number of node parameter collection N is less than 1000, the neural network is in the training stage, the input is the value of the node parameter, and the output is whether the node is expanded. The first four times of node parameter collection are defaulted to not expand the node, and after the fifth time of node parameter collection, the training is performed according to the gray advantage analysis result.

[0038] Further, the purpose of neural network training is to find the most suitable , and .

[0039] Embodiment 3 This embodiment further explains the present application based on Embodiment 1 and Embodiment 2.

[0040] The input sample of each group of the neural network is , the schematic diagram of the neural network structure is shown in Figure 2 , the neural network includes an input layer, a hidden layer and an output layer, x0(t) is the node CPU usage rate collected at t time, x1(t) is the node memory usage rate at t time, x2(t) is the node network traffic at t time, and x3(t) is the node active connection number at t time. The output sample is whether to expand, and expansion is to make , and non-expansion is to make .

[0041] In an optional embodiment, the prediction result of the trained neural network is the output of the output layer node, and its expression is:

[0042] wherein, L nn represents the prediction result of the neural network, represents the input of the output layer node, e represents the base number of the natural logarithm, represents the function mapping relationship.

[0043] In an optional embodiment, the input of the output layer node is obtained according to the output of the first j node of the hidden layer; and its expression is:

[0044] wherein, represents the input of the output layer node, represents the output of the first j node of the hidden layer; The first jThe output calculation expression of the node is:

[0045] wherein, represents the output of the i-th node of the hidden layer, j represents the weighted input of the i-th node of the hidden layer, represents the mother wavelet, j represents the scale stretching factor, represents the time translation factor. Further, ,

[0046] represents the weighted input of the i-th node of the hidden layer, , represents the mother wavelet, j represents the scale stretching factor, represents the time translation factor, represents the connection weight from the input layer to the hidden layer. In an optional embodiment, the node parameters are subjected to grey superiority analysis and prediction with the trained neural network to obtain a comprehensive prediction result final ; the expression is:

[0047] L

[0048] wherein, L final represents the comprehensive load prediction result.

[0049] Further, when the node parameter acquisition frequency N is less than 1000, the result of the grey superiority analysis is taken as the system prediction value, and when the node parameter acquisition frequency N is greater than or equal to 1000, a mixed prediction value is defined: wherein When L final > 0.6, it is judged that the node needs to be expanded.

[0050] ​​​The various embodiments in the present application are described in a progressive manner, and the same or similar parts among the various embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, the device embodiments are described more simply because they are basically similar to the method embodiments, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely exemplary, and the modules described as separate components can or can not be physically separated, and the functions of the modules can be implemented in one or more software and / or hardware in the implementation of the present application. Part or all of the modules can be selected to achieve the purpose of the present embodiment according to actual needs.

[0051] Obviously, the above-described embodiments of the present application are merely examples for clearly illustrating the present application, and are not intended to limit the implementation manner of the present application. Based on the above description, any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the claims of the present application.

Claims

1. A container cloud node load prediction method combining gray dominance analysis and neural networks, characterized in that, The method comprises the following steps: Step 1: collecting container cloud node parameters at a predetermined time frequency; Step 2: determining whether the number of node parameter collection times is less than a preset number, if yes, executing step 3, otherwise executing step 5; Step 3: Grey advantage analysis of node parameters to obtain prediction results L grey and training the constructed neural network according to the grey advantage analysis results; Step 4: judging the prediction result L grey whether the number is greater than or equal to a preset value, and if yes, performing the expansion operation on the container cloud node, otherwise, not performing the expansion operation on the container cloud node. Step 5: Grey dominance analysis of node parameters and prediction with trained neural network to obtain comprehensive prediction results L final ; Step 6: judging the comprehensive prediction result L final whether the value is greater than or equal to a preset value, and if yes, performing the expansion operation on the container cloud node, otherwise, not performing the expansion operation on the container cloud node.

2. The container cloud node load prediction method combining gray dominance analysis and neural network according to claim 1, characterized in that, The container cloud node parameters comprise CPU usage, memory usage, network traffic and active connection number.

3. The container cloud node load prediction method of claim 2, wherein, The CPU usage represents a ratio of used CPU to total CPU, the memory usage represents a ratio of used memory to total memory, the network traffic represents a total amount of data transmitted by a network interface of the node, and the active connection number represents a number of TCP / UDP connections currently established and maintained in an active state with the node.

4. The container cloud node load forecasting method combining grey superiority analysis and neural network according to claim 1, characterized in that, The node parameters are subjected to grey superiority analysis to obtain a prediction result L grey comprising the steps of: The collected container cloud node parameters are taken as sequences, and each parameter sequence is normalized; The normalized node CPU usage sequence is taken as a reference sequence, and the normalized node memory usage sequence, the normalized network traffic sequence and the normalized active connection number sequence are taken as comparison sequences; An absolute difference between the comparison sequence parameter and the reference sequence parameter is obtained according to the reference sequence and the comparison sequence; A grey correlation degree is obtained according to the absolute difference between the comparison sequence parameter and the reference sequence parameter and minimum and maximum values in the absolute difference between the comparison sequence parameter and the reference sequence parameter; The prediction result is obtained according to the parameter with the largest correlation degree and through accumulation sequence and grey micro equation L grey .

5. The container cloud node load forecasting method combining grey superiority analysis and neural network according to claim 4, characterized in that, The normalization of each parameter sequence; The expression is: wherein, x’ denotes the normalized value of each parameter, which ranges from [0, 1], x denotes the original data of each parameter.

6. The container cloud node load forecasting method combining grey superiority analysis and neural network according to claim 5, characterized in that, The grey correlation degree is obtained according to the absolute difference between the comparison sequence parameter and the reference sequence parameter and minimum and maximum values in the absolute difference between the comparison sequence parameter and the reference sequence parameter; The expression is: in, x’ 0 represents the reference sequence. x’ i Indicates a comparison sequence, Indicates the first k The comparison sequence parameters of the second acquisition and the first acquisition k The absolute difference of the reference sequence parameters from the next acquisition; the first k acquired comparison sequence parameter and the reference sequence parameter of the second k acquired absolute difference; the expression is: wherein, x’ 0 indicates a reference sequence, x’ i indicates a comparison sequence, k indicates the k th.

7. The container cloud node load forecasting method combining gray superiority analysis and neural network according to claim 6, characterized in that, The prediction result is obtained according to the parameter with the largest correlation degree and through accumulation sequence and grey micro equation L grey ; k+1 The expression thereof at the moment is wherein, L grey denotes the load prediction result, denotes the value of the cumulative sequence at the k+1 moment in time; The cumulative sequence expression is: wherein x’ (0) represents the initial sequence; The grey micro-equation expression is: wherein x’ (1) denotes an accumulation sequence, a denotes a development coefficient, b denotes a grey action amount, t denotes a time variable.

8. The container cloud node load forecasting method combining grey superiority analysis and neural network according to claim 1, characterized in that, The prediction result of the trained neural network is an output of an output layer node, and the expression is: wherein, L nn denotes a neural network prediction result, denotes an input of an output layer node, e denotes a base number of a natural logarithm, denotes a function mapping relationship.

9. The container cloud node load forecasting method combining grey superiority analysis and neural network according to claim 8, characterized in that, The input of the output layer node is obtained according to the output of the hidden layer node j . The expression is: wherein, represents an input of an output layer node, represents an output of an i-th node of a hidden layer; j represents an output of an i-th node of a hidden layer; The output calculation expression of the hidden layer first j node is: wherein, represents the output of the jth node of the hidden layer, represents the weighted input of the jth node of the hidden layer, represents the mother wavelet, represents the scale dilation factor, represents the time translation factor.

10. The container cloud node load forecasting method combining gray superiority analysis and neural network according to any one of claims 1 to 9, characterized in that, The node parameters are subjected to grey advantage analysis and prediction by using the trained neural network to obtain a comprehensive prediction result L final The expression is: wherein, L final represents the comprehensive load prediction result, represents the weight coefficient.

Images