A computing resource optimization method and system based on channel performance intelligent analysis

Through nonlinear dimensionality reduction algorithm and LSTM model to identify key performance indicators and dynamically adjust cloud resources, the problems of low data analysis efficiency and inflexible resource allocation in traditional methods are solved, and efficient and accurate performance forecasting and resource optimization are achieved.

CN117971490BActive Publication Date: 2024-11-01北京嘉华铭品牌策划有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410195780.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-22
Publication Date
2024-11-01
Estimated Expiration
2044-02-22

AI Technical Summary

Technical Problem

Traditional data analysis methods are difficult to quickly and accurately extract performance influencing factors and predict performance. Cloud resource allocation lacks the ability to respond quickly to dynamic changes in business demand, resulting in low resource usage efficiency and poor performance stability.

Method used

Using intelligent analysis methods based on channel performance, through nonlinear dimensionality reduction algorithms and LSTM models, we identify key performance indicators, build high-accuracy prediction models, and dynamically adjust cloud computing resources to meet business needs.

Benefits of technology

It improves data processing and analysis efficiency, enhances the accuracy of the prediction model, optimizes resource use, reduces costs, and improves the stability of business performance and resource use efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117971490B_ABST
    Figure CN117971490B_ABST
Patent Text Reader

Abstract

The present invention discloses a computing resource optimization method and system based on intelligent analysis of channel performance, including a training phase and an online phase, wherein the training phase includes the following steps: step S1, collecting data of L channel indicators from multiple channels; step S2, for the collected data of each channel, reducing the high-dimensional data of the L channel indicators to a low-dimensional space based on a nonlinear dimensionality reduction algorithm, and determining N channel indicators as key indicators of the channel based on the coordinates of each time point in the low-dimensional space; step S3, establishing a prediction model for the N key indicators; step S4, determining a relationship model between the N key indicators and the flow of people. The present invention improves the efficiency and accuracy of data analysis, optimizes resource allocation, improves the performance stability of real-time acquisition and analysis of channel performance data, and enhances the accuracy of channel performance data analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a computing resource optimization method and system based on channel performance intelligent analysis. Background Art

[0002] With the rapid development of information technology and the rise of the Internet economy, the marketing channels of enterprises have gradually diversified, generating a large amount of marketing data. These data contain rich commercial value and are of great significance for enterprises to grasp market dynamics, optimize marketing strategies, and improve performance. However, due to the multidimensionality and complexity of data, traditional data analysis methods have been unable to meet the needs of quickly and accurately extracting performance influencing factors and predicting performance.

[0003] At the same time, the business systems deployed by enterprises on cloud platforms require reasonable computing resources to support their operation. How to dynamically adjust resources according to actual business needs is the key to improving economic benefits and service quality. Currently, most cloud resource allocations are based on experience or static rules, lacking the ability to quickly respond to dynamic changes in business needs.

[0004] In response to the above problems, there is an urgent need for a technology that can intelligently analyze channel performance and optimize the allocation of computing resources, so that in an Internet environment with huge data volumes and high real-time requirements, useful information can be quickly extracted from massive data and channel performance can be accurately predicted. It has the ability to process large-scale high-dimensional data, can identify key performance indicators through intelligent algorithms, build highly accurate prediction models, and can dynamically adjust computing resources according to performance prediction results to adapt to changes in business needs, improve resource utilization efficiency and analyze the performance stability of business applications. Summary of the invention

[0005] In view of the defects existing in the above-mentioned prior art, the present invention provides a computing resource optimization method based on channel performance intelligent analysis, including a training stage and an online stage, wherein the training stage includes the following steps:

[0006] Step S1, collecting data of L channel indicators from multiple channels, including collecting data of L channel indicators at multiple time points;

[0007] Step S2, for each channel's collected data, reducing the high-dimensional data of the L channel indicators to a low-dimensional space based on a nonlinear dimensionality reduction algorithm, and determining N channel indicators as key indicators of the channel based on the coordinates of each time point in the low-dimensional space; wherein N is less than L; Step S3, for each channel, establishing a prediction model for the N key indicators;

[0008] Step S4, for each channel, determining a relationship model between the N key indicators and the flow of people;

[0009] And, the online stage includes the following steps:

[0010] Step P1, predicting the flow of people in a future time period based on the relationship model between the N key indicators and the flow of people;

[0011] Step P2, allocating elastic cloud computing resources based on the predicted flow of people in the future time period;

[0012] And, when the error between the predicted pedestrian flow and the actual pedestrian flow exceeds a threshold, all steps of the training phase are repeated, and the relationship model between the N key indicators and the pedestrian flow in the cloud server is updated.

[0013] Wherein, in the step S1, collecting data of L channel indicators from multiple channels includes collecting data of L channel indicators at multiple time points and storing them in a data matrix, including creating a data matrix X;

[0014] Each row represents data at a time point, and each column represents a performance indicator. The 8 channel indicator data are collected at M time points, and the data matrix X is an M×8 matrix:

[0015]

[0016] Among them, x ij represents the value of the jth indicator at the i-th time point;

[0017] The corresponding relationship between channel indicators is as follows:

[0018] x ij Where j=1 represents sales revenue, j=2 represents sales growth rate, j=3 represents customer retention rate, j=4 represents conversion rate, j=5 represents market share, j=6 represents average promotion discount rate, j=7 represents average order value, and j=8 represents inventory turnover rate;

[0019] Wherein, each item in the data matrix is ​​a normalized value in the channel indicator. Wherein, in the step S2, for the collected data of each channel, the high-dimensional data of the M channel indicators are reduced to a lower dimension based on a nonlinear dimensionality reduction algorithm, including using a t-SNE algorithm to reduce the high-dimensional data of the L channel indicators to a low dimension.

[0020] Wherein, in the step S2, by using the coordinates of each time point in the low-dimensional space, based on the coordinates of each time point in the low-dimensional space, N channel indicators are determined as the key indicators of the channel; specifically, the following steps are included:

[0021] Step K1, use the K-means algorithm to cluster the M data points in the low-dimensional space Y:

[0022] That is, for Y, execute the K-means algorithm to divide the points into C clusters:

[0023] Among them, S c is the set of points in the cth cluster, μ c is the center of the cth cluster;

[0024] Step K2, determine the center point in the cluster: find the data point in each cluster that has the shortest average distance to other nodes in the cluster, i.e., the center point; i.e., select cluster S c Point y in center , so that:

[0025] In three-dimensional space, point y i and point y j The Euclidean distance between ||y i -y j || can be calculated by the following formula:

[0026]

[0027] Among them, (y i1 ,y i2 ,y i3 ) and (y j1 ,y j2 ,y j3 ) are the coordinates of time points i and j in three-dimensional space respectively; Step K3, select neighboring data points: for each cluster, determine the first K data points with the shortest distance from the center point, that is, for the center point y center , select the first K data points Y with the shortest distance K ;

[0028] Step K4, mapping back to high-dimensional space: find the corresponding points of these K data points in the original high-dimensional space X; that is, find Y K The corresponding point set X in the high-dimensional space X K ;

[0029] Step K5, calculate the distance between high-dimensional indicators: calculate the distance between the channel indicators of these points in the high-dimensional space; that is, for X K For every pair of points x i and x j , calculate their distance d on each channel indicator n ijn :d ijn =|x in -x jn |;

[0030] Among them, XK Represents the set of points in the high-dimensional space X corresponding to the K data points selected from the cluster in the low-dimensional space Y;

[0031] Among them, these points are selected according to the nearest neighbor relationship in the low-dimensional space, and the K points are those K points closest to the cluster center point in the low-dimensional representation;

[0032] x i and x j is any pair of points in a high-dimensional space X, which are K There is a corresponding expression in x i and x j is the feature vector of two data points in the original data set;

[0033] n represents a specific channel indicator in high-dimensional space;

[0034] In a high-dimensional dataset X, each point x i It is a vector composed of multiple channel indicators;

[0035] x in and x jn Represents the data point x i and x j The value of the indicator at the nth channel;

[0036] d ijn Represents data point x i and x j The distance on the nth channel indicator is a non-negative real number representing the difference between two values;

[0037] i.e. ijn =|x in -x jn |Calculates the value of two data points x on a specific channel indicator n i and x j The absolute value of the difference;

[0038] Step K6, determine the key indicators: determine the N indicators with the smallest cumulative distance sum of the original indicators as the key indicators; that is, for each channel indicator n, calculate its distance in X K The cumulative distance D between all point pairs in n , the calculation formula is:

[0039]

[0040] Then select the cumulative distance D n The smallest N indicators.

[0041] In step S3, for each channel, a prediction model for the N key indicators is established, including:

[0042] For each key indicator, deploy the trained LSTM model to the production environment and implement a regular retraining mechanism to adapt to new data;

[0043] N LSTM models are obtained and used to predict the N key indicators corresponding to each channel.

[0044] Among them, in step S4, for each channel, the relationship model between the N key indicators and the flow of people is determined, including assigning different weights to different key indicators in the prediction model for predicting the flow of people in step S4 based on the cumulative distance Dn of different key indicators determined in step S2, and considering the regularization term based on the cumulative distance Dn of different key indicators in the loss function.

[0045] Among them, based on the cumulative distance D n Different weights are assigned to different key indicators in the prediction model for predicting passenger flow, including the weight w assigned to the nth key indicator. n It can be defined by a transformation function f, so that the key indicators with smaller distances get larger weights: in, is equal to 1;

[0046] Based on the weight w n Construct a weighted linear relationship model and transform each key indicator X n Associated with the flow of people P:

[0047] in,

[0048] P is the flow of people;

[0049] β 0 is the intercept term;

[0050] β n Is related to key indicator X n The coefficient of correlation;

[0051] w n Is the key indicator X n The weight of

[0052] ∈ is the error term.

[0053] In step P2 of the online phase, elastic cloud computing resources are allocated based on the predicted flow of people in the future time period, including a resource demand function that converts the flow of people into the predicted required computing resources R t+1 , the predicted required computing resources R t+1 The calculation formula is:

[0054] R t+1 =α·P t+1 +R base ;

[0055] in,

[0056] R t+1 : The amount of computing resources required to be allocated in time period t+1;

[0057] α: proportionality coefficient, indicating the amount of resources required per unit of traffic;

[0058] P t+1 : The flow of people in time period t+1 predicted according to step P1;

[0059] R base : Basic resource requirements.

[0060] The present invention also discloses a computing resource optimization system based on intelligent analysis of channel performance. The system includes a memory and a processor. The processor is used to execute computer code stored in the memory to execute the aforementioned computing resource optimization method based on intelligent analysis of channel performance.

[0061] The present invention reduces the complexity of data and improves the efficiency of data processing and analysis through a nonlinear dimensionality reduction algorithm. By focusing on key indicators, channel performance can be more accurately reflected, and the accuracy of the prediction model is enhanced. Computing resources are dynamically allocated based on the results of traffic flow prediction, which optimizes the use of resources, reduces costs, improves resource utilization efficiency, and enables the response to potential service demand increases, thereby improving the stability of business performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0063] Figure 1 is a flow chart showing a training phase in a computing resource optimization method based on channel performance intelligent analysis according to an embodiment of the present invention;

[0064] Figure 2 It is a flowchart showing the online phase in a computing resource optimization method based on channel performance intelligent analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0067] It should be understood that although the terms first, second, third, etc. may be used to describe ... in the embodiments of the present invention, these ... should not be limited to these terms. These terms are only used to distinguish .... For example, without departing from the scope of the embodiments of the present invention, the first ... may also be referred to as the second ..., and similarly, the second ... may also be referred to as the first ....

[0068] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0069] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0070] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0071] Directly processing the high-dimensional channel indicator data will encounter the "dimensionality curse", high computational complexity, and difficulty in finding the potential relationship between indicators. The present invention uses a nonlinear dimensionality reduction algorithm to screen the data and then determines the key indicators from the high-dimensional space, and pre-allocates cloud computing resources based on the traffic volume predicted by the key indicators, making the real-time acquisition and analysis performance of channel performance data more stable.

[0072] A computing resource optimization method based on intelligent analysis of channel performance includes: in a training phase: step S1, collecting data of L channel indicators from multiple channels, including collecting data of L channel indicators at multiple time points;

[0073] Step S2, for each channel's collected data, reduce the high-dimensional data of the L channel indicators to a low-dimensional space based on a nonlinear dimensionality reduction algorithm, and determine N channel indicators as the key indicators of the channel based on the coordinates of each time point in the low-dimensional space; where N is less than L; Step S3, for each channel, establish a prediction model for the N key indicators. The goal of the prediction model is to determine the N key indicators for the next time period.

[0074] Step S4: for each channel, determine the relationship model between the N key indicators and the flow of people.

[0075] And, a computing resource optimization method based on intelligent analysis of channel performance, including the online stage:

[0076] Step P1, based on the relationship model between the N key indicators and the flow of people, predict the flow of people in the future time period.

[0077] Step P2, based on the predicted flow of people in the future time period, elastic cloud computing resources are allocated. And, when the error between the predicted flow of people and the actual flow of people exceeds a threshold, all steps of the training phase are repeated, and the relationship model between the N key indicators and the flow of people in the cloud server is updated.

[0078] The present invention reduces the complexity of data and improves the efficiency of data processing and analysis through a nonlinear dimensionality reduction algorithm. By focusing on key indicators, channel performance can be more accurately reflected, and the accuracy of the prediction model is enhanced. Computing resources are dynamically allocated based on the results of traffic flow prediction, which optimizes the use of resources, reduces costs, improves resource utilization efficiency, and enables the response to potential service demand increases, thereby improving the stability of business performance.

[0079] In one embodiment, in step S1, collecting data of L channel indicators from multiple channels includes collecting data of L channel indicators at multiple time points and storing them in a data matrix, including:

[0080] Create a data matrix X. Each row represents the data at a time point, and each column represents a performance indicator. Therefore, if there are M time points, the data matrix X is an M×8 matrix.

[0081]

[0082] Among them, x ij Represents the value of the jth indicator at the i-th time point.

[0083] The corresponding relationship between channel indicators is as follows:

[0084] x ij Here, j=1 represents sales revenue, j=2 represents sales growth rate, j=3 represents customer retention rate, j=4 represents conversion rate, j=5 represents market share, j=6 represents average promotion discount rate, j=7 represents average order value, and j=8 represents inventory turnover rate.

[0085] Each item in the data matrix may be a normalized value in the channel indicator.

[0086] In one embodiment, in step S2, for the collected data of each channel, the high-dimensional data of the L channel indicators are reduced to a lower dimension based on a nonlinear dimensionality reduction algorithm. This includes using a t-SNE algorithm to reduce the high-dimensional data of the eight indicators to a lower dimension for visualization. For example, from 8 dimensions to 3 dimensions, and displaying the performance at each time point through a scatter plot.

[0087] Among them, the key steps of the t-SNE algorithm include:

[0088] Calculate the similarity between point pairs in high-dimensional space, usually using Gaussian distribution to convert Euclidean distance to conditional probability p j|i .

[0089]

[0090] Among them, σ i It is a perplexity-related parameter, and a suitable value is found through binary search.

[0091] The symmetric joint probability p ij is defined as:

[0092]

[0093] Where M is the total number of data points.

[0094] Mapping in low-dimensional space, including initializing each point y i Then, a similar method is used to calculate the point y in the low-dimensional space i andj The similarity q between ij In t-SNE, the low-dimensional space usually uses the t distribution to calculate the similarity:

[0095]

[0096] Then minimize the objective function: The purpose of t-SNE is to find a low-dimensional representation so that the similarity P in the high-dimensional space ij The similarity q with the low-dimensional space ij To do this, it minimizes the Kullback-Leibler (KL) divergence between similarities:

[0097]

[0098] Iteratively adjust y through optimization algorithms such as gradient descent i , thereby minimizing the KL divergence. Finally, the distance between data points in the low-dimensional space represents their similarity in the original high-dimensional space. The Y matrix will be an M*3 matrix. Among them:

[0099] M is the number of time points.

[0100] 3 represents the three coordinates (x, y, z) of each sample in three-dimensional space.

[0101] In this reduced-dimensional Y matrix, each row represents the indicator data of a certain channel at a certain point in time, and each column represents a dimension.

[0102] The goal of t-SNE is to maintain the relative distance between sample points in the high-dimensional space, especially the local structure, in the low-dimensional space, so that similar points in the original high-dimensional space are also close in the low-dimensional space.

[0103] In one embodiment, in step S2, N channel indicators are determined as key indicators of the channel based on the coordinates of each time point in the low-dimensional space. Specifically, the following steps are included:

[0104] Step K1, use the K-means algorithm to cluster the M data points in the low-dimensional space Y:

[0105] That is, for Y, execute the K-means algorithm to divide the points into C clusters:

[0106] Among them, S c is the set of points in the cth cluster, μ c is the center of the cth cluster.

[0107] Step K2, determine the center point in the cluster: find the data point with the shortest average distance to other nodes in each cluster, that is, the center point. That is, select cluster S c Point y in center , so that:

[0108] In three-dimensional space, point y i and point y j The Euclidean distance between ||y i -y j || can be calculated by the following formula:

[0109]

[0110] Among them, (y i1 ,y i2 ,y i3 ) and (y j1 ,y j2 ,y j3 ) are the coordinates of time points i and j in three-dimensional space.

[0111] Step K3, select neighboring data points: For each cluster, determine the first K data points with the shortest distance from the center point. That is, for the center point y center , select the first K data points Y with the shortest distance K .

[0112] Step K4, map back to high-dimensional space: find the corresponding points of these K data points in the original high-dimensional space X. That is, find Y K The corresponding point set X in the high-dimensional space X K .

[0113] Step K5, calculate the distance between high-dimensional indicators: calculate the distance between the channel indicators of these points in the high-dimensional space. K For every pair of points x i and x j , calculate their distance d on each channel indicator n ijn :d ijn =|x in -x jn |.

[0114] Among them, X K Represents the set of points in the high-dimensional space X corresponding to the K data points selected from the cluster in the low-dimensional space Y. These points are selected based on the nearest neighbor relationship in the low-dimensional space, assuming that the K points are those closest to the cluster center in the low-dimensional representation.

[0115] x i and x jis any pair of points in a high-dimensional space X, which are K In other words, x i and x j is the feature vector of two data points in the original data set.

[0116] n represents the index of a specific channel indicator or feature in the high-dimensional space. In the high-dimensional data set X, each point x i is a vector consisting of multiple such indicators.

[0117] x in and x jn Represents the data point x i and x j The value of the indicator at the nth channel.

[0118] d ijn Represents data point x i and x j The distance on the nth channel indicator is a non-negative real number representing the difference between two values.

[0119] i.e. ijn =|x in -x jn |Calculates the value of two data points x on a specific channel indicator n i and x j The absolute value of the difference.

[0120] Step K6, determine the key indicators: determine the N indicators with the smallest cumulative distance sum of the original indicators as the key indicators. That is, for each channel indicator n, calculate its K The cumulative distance D between all point pairs in n , the calculation formula is:

[0121]

[0122] Then select the cumulative distance D n The smallest N indicators. Optionally, N=3.

[0123] In one embodiment, in step S3, for each channel, a prediction model for the N key indicators is established. The goal of the prediction model is to determine the N key indicators for the next time period, including:

[0124] For each key metric, choose the appropriate number of LSTM units, number of layers, add a Dropout layer after the LSTM layer, use mean square error (MSE) as the loss function; use cross entropy loss.

[0125] Determine the appropriate batch size and number of iterations based on the size of your dataset and your computing resources. Monitor the validation loss and stop training when you see that the validation performance is no longer improving to avoid overfitting.

[0126] During the training process, the model with the lowest validation loss is saved as the final model.

[0127] Deploy the trained LSTM model to the production environment and implement a regular retraining mechanism to adapt to new data.

[0128] And, monitor the real-time performance of the model to ensure the accuracy of the predictions. Regularly update the model with new data over time.

[0129] Finally, N LSTM models are obtained for predicting the N key indicators of the channel. In one embodiment, in step S4, for each channel, a relationship model between the N key indicators and the flow of people is determined, including assigning different weights to different key indicators in the prediction model for predicting the flow of people in step S4 based on the cumulative distance Dn of different key indicators determined in step S2, and considering the regularization term based on the cumulative distance Dn of different key indicators in the loss function to prevent overfitting when fitting the coefficients in the prediction model.

[0130] Among them, based on the cumulative distance D n Different weights are assigned to different key indicators in the prediction model for predicting passenger flow, including the weight w assigned to the nth key indicator. n It can be defined by a transformation function f, so that the key indicators with smaller distances get larger weights: Where = is a small constant (eg, 1).

[0131] Based on the weight w n Construct a weighted linear relationship model. This model attempts to transform each key indicator X n Associated with the flow of people P:

[0132] in,

[0133] P is the flow of people.

[0134] β 0 is the intercept term.

[0135] β n Is related to key indicator X n The coefficient of correlation.

[0136] w n Is the key indicator X n The weight of .

[0137] ∈ is the error term.

[0138] To fit the model parameters β 0 , β 1 , ..., β N To minimize the weighted loss function, consider the loss function L of the regularization term based on the cumulative distance Dn of different key indicators to determine the fitting model parameter β 0 , β 1 , ..., β N :

[0139]

[0140] in,

[0141] L(β 0 , β 1 , ..., β N ): The value of the loss function, which is a measure of the difference between the model's predictions and the true values.

[0142] P i+1 : The actual flow of people observed at the i+1th time point.

[0143] β 0 : Intercept term, baseline output of the model.

[0144] β n : The coefficient of the nth key indicator.

[0145] X ni : The value of the nth key indicator in the observation at the ith time point.

[0146] w n : The weight of the nth key indicator is adjusted according to the importance of the indicator.

[0147] λ: Regularization parameter, used to balance the weight between the least squares term and the regularization term.

[0148] ∈: Error term, which represents the error between the model prediction and the actual observation.

[0149] In order to enhance the importance of key indicators in crowd flow prediction, a weighted regularization term is added to the loss function, specifically a weighted L2 regularization term, so that the higher the importance of the indicator, the higher its sensitivity.

[0150] In one embodiment, in an online phase of a computing resource optimization method based on intelligent analysis of channel performance, the flow of people in a future time period is predicted.

[0151] In the online stage, before the prediction begins, it is necessary to obtain the key indicators and traffic prediction model corresponding to the channel. Specifically, it includes:

[0152] In the offline stage, historical data is used to determine the key indicators of different channels, and to train traffic prediction models for different channels. The accuracy of the models is determined through verification and testing.

[0153] Store the key indicators and trained traffic prediction models for all channels in an accessible server or cloud environment to ensure that the models can be loaded quickly during the online phase. During the online phase, key indicators and trained traffic prediction models for different channels are accessed through authenticated API endpoints or internal networks.

[0154] In the online phase, for each channel, key indicator data is collected from the data source regularly or in real time. The data of these key indicators are automatically collected through API calls, database queries, IoT device interfaces, etc. ETL tools are used to process and store data.

[0155] Before the prediction begins, load the latest crowd flow prediction model from the model management system. Input the key indicator data collected in real time into the model for prediction. Output the prediction results and optimize the computing resources accordingly.

[0156] In one embodiment, storing key indicators corresponding to the same channel and a trained traffic flow prediction model includes:

[0157] The data is divided into two types: static data (key indicator types and model parameters) and dynamic data (real-time values ​​of key indicators).

[0158] Static data is stored in the object storage service, while dynamic data can be stored in a time series database to support fast read and write operations.

[0159] In order to quickly access the data and models of a specific channel, create indexes for the data and models, and establish relationship mappings between channels and data and models.

[0160] In one embodiment, before the task of real-time prediction of pedestrian flow begins, it is necessary to obtain static data from the storage service, obtain the key indicator type from the static data, and then obtain the real-time value of the key indicator from the time series database.

[0161] In one embodiment, in step P1, based on the relationship model between the N key indicators and the flow of people, the target variable P (flow of people) is predicted according to the N key indicators in the relationship model. n There is a corresponding weight w n and coefficient β n , which determines the degree of influence of this indicator on the prediction of passenger flow. ∈ in the model represents the prediction error, which reflects the influence of random factors other than the key indicators considered by the model.

[0162] The online phase uses the relational model to make predictions, including:

[0163] In the online stage, the predicted time period is t, and the key indicator data X corresponding to the time period t is collected. n,t Then, based on the trained model parameters β 0 , β 1 , ..., β N and indicator weight w 1 , ..., w N To calculate the predicted value P of the flow of people t+1 , P t+1 is the prediction of the flow of people at the current time t, that is, the predicted value can be understood as the model estimated value of the flow of people in the next time period t.

[0164] The predicted flow of people at the next time point P t+1 It can be expressed as:

[0165]

[0166] This forecasting process does not include an error term ∈ because the error term represents random factors not captured in the model.

[0167] In a certain embodiment, elastic cloud computing resources are allocated based on predicted traffic volume in a future time period.

[0168] Based on the predicted passenger flow in the future time period in step P1, analyze and predict the demand for CPU computing resources by passenger flow. Determine the required cloud resource specifications based on the demand analysis results.

[0169] Formulate elastic scaling rules based on historical data analysis and set automatic expansion and contraction thresholds for resources.

[0170] Automatic and elastic allocation of resources is achieved through elastic computing services provided by cloud service providers, such as AWS Auto Scaling.

[0171] In one embodiment, in step P2 of the online phase, elastic cloud computing resources are allocated based on the predicted flow of people in the future time period, including:

[0172] Based on the resource demand function, which converts the flow of people into the required amount of resources R t+1 : R t+1 =α·P t+1 +R base ;

[0173] in,

[0174] R t+1 : The amount of resources that need to be allocated in time period t+1.

[0175] α: Proportional coefficient, indicating the amount of resources required per unit of traffic.

[0176] P t+1 : The passenger flow in time period t+1 predicted according to step P1.

[0177] R base : Basic resource requirements, that is, the minimum computing resources that need to be guaranteed regardless of the size of the traffic.

[0178] Optionally, to handle possible bursts of traffic, add a buffer to the resource requirements:

[0179] in,

[0180] R buffer : Resource buffer, determined based on experience R t+1 a fixed percentage of .

[0181] Optionally, set elastic scaling rules that dynamically adjust resource allocation based on actual monitoring data, including:

[0182] If the CPU util >80%:R new =R current +γ·R current ;

[0183] If the CPU util <20%:R new =R current -δ·R current .

[0184] in,

[0185] R new : The adjusted amount of resources.

[0186] R current : The current amount of resources allocated.

[0187] CPU util : CPU utilization.

[0188] γ: The expansion factor when resource demand increases, which is a fixed value, such as 20%.

[0189] δ: reduction factor when resource demand decreases, a fixed value, such as 10%.

[0190] In one embodiment, the process of determining the proportional coefficient α, i.e., the amount of resources required per unit of passenger flow, includes:

[0191] Collect historical data, including cloud computing resource usage at different traffic levels, where the cloud resource usage is collected when the traffic analysis performance meets the performance requirements at the traffic level, including computing delay and error rate:

[0192] P 1 , R 1

[0193] P 2 , R 2

[0194] …

[0195] P n , R n

[0196] Among them, P i represents the flow of people at the i-th observation point, and R i Indicates the corresponding cloud computing resource usage.

[0197] Select two traffic points and the corresponding resource usage. For example, select (P 1 , R 1 ) and (P 2 , R 2 ), where P 2 >P 1 . Calculate the ratio of the increase in resource usage to the increase in traffic between these two points:

[0198] Among them, α represents the 1 Increase to P 2 The increment of resource usage corresponding to the unit increase of passenger flow.

[0199] Optionally, an averaging process is performed, including that in order to obtain a more accurate α value, multiple α values ​​can be calculated using all data point pairs, and then the average of these α values ​​is taken. That is: Ensure P i+1 ≠P i to avoid division by zero.

[0200] Using alpha avg Predict all data points and calculate the error between the predicted value and the actual value. avg value to predict the new flow rate P t+1 The computing resource requirement R t+1 ,include:

[0201] R t+1 =α·P t+1 +R base , or R t+1 =αavg ·P t+1 +R base .

[0202] Where R base It is the basic resource demand, which can be estimated by the resource usage at the minimum traffic flow in historical data.

[0203] In practical applications, the demand for cloud computing resources is related to various hardware capacity limitations. Based on actual usage data modeling,

[0204] In one embodiment, when the error between the predicted pedestrian flow and the actual pedestrian flow exceeds a threshold, all steps in the training phase are repeated, and the relationship model between the N key indicators and the pedestrian flow in the cloud server is updated, including:

[0205] Set a threshold, when the error is greater than this threshold, trigger the retraining of the model. The error can be calculated using the following formula:

[0206] Error = |predicted traffic flow - actual traffic flow|. When the error is greater than the threshold, all steps S1 to S3 in the training phase are repeated, and the static data (key indicator types and model parameters) in the object storage service in the cloud storage are updated, as well as the dynamic data (real-time values ​​of key indicators) in the time series database.

[0207] Therefore, the present invention provides a method for intelligent analysis of channel performance and optimization of computing resources, which not only improves the efficiency and accuracy of data analysis, but also optimizes resource allocation, improves the performance stability of real-time acquisition and analysis of channel performance data, and enhances the accuracy of channel performance data analysis.

[0208] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0209] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0210] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0211] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0212] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0213] The above introduces the preferred embodiments of the present invention, which is intended to make the spirit of the present invention clearer and easier to understand, but is not intended to limit the present invention. All modifications, substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection outlined by the claims attached to the present invention.

Claims

1. A computing resource optimization method based on intelligent analysis of channel performance, including a training phase and an online phase, wherein: The training phase includes the following steps: Step S1, collecting data of L channel indicators from multiple channels, including collecting data of L channel indicators at multiple time points; Step S2, for each channel's collected data, reducing the high-dimensional data of the L channel indicators to a low-dimensional space based on a nonlinear dimensionality reduction algorithm, and determining N channel indicators as N key indicators of the channel based on the coordinates of each time point in the low-dimensional space; wherein N is less than L; Step S3, for each channel, establishing a prediction model for the N key indicators; Step S4, for each channel, determining a relationship model between the N key indicators and the flow of people; And, the online stage includes the following steps: Step P1, predicting the flow of people in a future time period based on the relationship model between the N key indicators and the flow of people; Step P2, allocating elastic cloud computing resources based on the predicted passenger flow in the future time period; and, when the error between the predicted passenger flow and the actual passenger flow exceeds a threshold, repeating all steps of the training phase and updating the relationship model between N key indicators and passenger flow in the cloud server.

2. A computing resource optimization method based on channel performance intelligent analysis as claimed in claim 1, characterized in that: In the step S1, data of L channel indicators are collected from multiple channels, including collecting L channel indicator data at multiple time points and storing them in a data matrix, including creating a data matrix X; Each row represents data at a point in time, and each column represents a performance indicator.

3. A computing resource optimization method based on channel performance intelligent analysis as claimed in claim 2, characterized in that: When the L channel indicator data are 8L channel indicator data: Collect 8 channel indicator data at M time points. The data matrix X is an M×8 matrix: Among them, the element x in the data matrix X ij Represents the value of the jth indicator at the i-th time point; the value range of i is 1 to M, and the value range of j is 1 to 8; The corresponding relationship between channel indicators is as follows: x ij Where j=1 represents sales revenue, j=2 represents sales growth rate, j=3 represents customer retention rate, j=4 represents conversion rate, j=5 represents market share, j=6 represents average promotion discount rate, j=7 represents average order value, and j=8 represents inventory turnover rate; Each item in the data matrix is ​​a normalized value in the channel indicator.

4. A computing resource optimization method based on channel performance intelligent analysis as claimed in claim 1 or 3, characterized in that: In the step S2, for the collected data of each channel, the high-dimensional data of the L channel indicators are reduced to a low-dimensional space based on a nonlinear dimensionality reduction algorithm, including using a t-SNE algorithm to reduce the high-dimensional data of the L channel indicators to a low dimension.

5. A computing resource optimization method based on channel performance intelligent analysis as claimed in claim 1 or 3, characterized in that: In the step S2, N channel indicators are determined as key indicators of the channel based on the coordinates of each time point in the low-dimensional space, which specifically includes the following steps: Step K1, use the K-means algorithm to cluster the M data points in the low-dimensional space Y: That is, for Y, execute the K-means algorithm to divide the points into C clusters: Among them, S c is the set of points in the cth cluster, μ c is the center of the cth cluster; Step K2, determine the center point in the cluster: find the data point in each cluster with the shortest average distance to other nodes in the cluster, that is, the center point; that is, select cluster S c Point y in center , so that: In three-dimensional space, point y i and point y j The Euclidean distance between i -y j ∣∣Calculated by the following formula: Among them, (y i1 ,y i2 ,y i3 ) and (y j1 ,y j2 ,y j3 ) are the coordinates of time points i and j in three-dimensional space respectively; Step K3, select neighboring data points: For each cluster, determine the first K data points with the shortest distance to the center point, that is, for the center point y center , select the first K data points Y with the shortest distance K ; Step K4, mapping back to high-dimensional space: find the corresponding points of these K data points in the original high-dimensional space X; that is, find Y K The corresponding point set X in the high-dimensional space X K ; Step K5, calculate the distance between high-dimensional indicators: calculate the distance between the channel indicators of these points in the high-dimensional space; that is, for X K For every pair of points x i and x j , calculate their distance d on each channel indicator n ijn :d ijn =|x in -x jn |; Among them, X K Represents the set of points in the high-dimensional space X corresponding to the K data points selected from the cluster in the low-dimensional space Y; Wherein, these points are selected based on the nearest neighbor relationship in the low-dimensional space, and the K data points are those K data points closest to the cluster center point in the low-dimensional representation; x i and x j is any pair of points in a high-dimensional space X, which are K There is a corresponding expression in x i and x j is the feature vector of two data points in the original data set; n represents a channel indicator in the high-dimensional space, where 1≤n≤L; In a high-dimensional dataset X, each point x i It is a vector composed of multiple channel indicators; x in and x jn Represents the data point x i and x j The value of the indicator at the nth channel; d ijn Represents data point x i and x j The distance on the nth channel indicator is a non-negative real number representing the difference between two values; i.e. ijn =|x in -x jn |Calculated two data points x on channel indicator n i and x j The absolute value of the difference; Step K6, determine the key indicators: determine the N indicators with the smallest cumulative sum of distances of the original indicators as the key indicators; that is, for each channel indicator n, calculate its K The cumulative distance D between all point pairs in n , the calculation formula is: Then select the cumulative distance D n The smallest N indicators.

6. The computing resource optimization method based on channel performance intelligent analysis according to claim 1, characterized in that: In step S3, for each channel, a prediction model for the N key indicators is established, including: For each key indicator, deploy the trained LSTM model to the production environment and implement a regular retraining mechanism to adapt to new data; N LSTM models are obtained and used to predict the N key indicators corresponding to each channel.

7. The computing resource optimization method based on channel performance intelligent analysis as claimed in claim 5, characterized in that: In the step S4, for each channel, a relationship model between the N key indicators and the flow of people is determined, including assigning different weights to different key indicators in the prediction model for predicting the flow of people in step S4 based on the cumulative distance Dn of different key indicators determined in step S2, and considering the regularization term based on the cumulative distance Dn of different key indicators in the loss function.

8. A computing resource optimization method based on channel performance intelligent analysis as claimed in claim 7, characterized in that: Based on the cumulative distance D n Different weights are assigned to different key indicators in the prediction model for predicting passenger flow, including the weight w assigned to the nth key indicator. n Defined by a transformation function f: in, is equal to 1; Based on the weight w n Construct a weighted linear relationship model and transform each key indicator X n Associated with the flow of people P: in, P is the flow of people; β0 is the intercept term; β n Is related to key indicator X n The coefficient of correlation; w n Is the key indicator X n The weight of ∈ is the error term.

9. The computing resource optimization method based on channel performance intelligent analysis according to claim 1, characterized in that: In step P2 of the online phase, elastic cloud computing resources are allocated based on the predicted flow of people in the future time period, including based on the resource demand function, which converts the flow of people in the t+1th time period into the predicted required computing resource amount R t+1 , the predicted computing resource required for the t+1th time period R t+1 The calculation formula is: R t+1 =α·P t+1 +R base ; in, R t+1 : The amount of computing resources required to be allocated in the t+1th time period; α: proportionality coefficient, indicating the amount of resources required per unit of traffic; P t+1 : The passenger flow in the t+1th time period predicted according to step P1; R base : Basic resource requirements.

10. A computing resource optimization system based on intelligent analysis of channel performance, the system comprising a memory and a processor, the processor being used to execute computer code stored in the memory to execute a computing resource optimization method based on intelligent analysis of channel performance as described in any one of claims 1-9.