Cloud mobile phone virtualization network function optimization method and related equipment

Abstract

The invention discloses a cloud mobile phone virtualized network function optimization method and related equipment, and relates to the technical field of computer application, and the method comprises the steps: obtaining network performance monitoring data and resource utilization parameters of a cloud mobile phone service platform; generating network demand trend data through a preset prediction model based on the network performance monitoring data; generating a dynamic resource allocation strategy through a preset optimization model based on the network demand trend data and the resource utilization parameters; and performing dynamic adjustment operation on the virtualized network function instance deployed on the cloud mobile phone service platform based on the dynamic resource allocation strategy. By introducing the preset prediction model and the preset optimization model, dynamic resource scheduling and elastic management of the virtualized network function are realized, and the network performance, the resource utilization rate and the service stability of the cloud mobile phone system can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer application, more particularly, the present application relates to a cloud phone virtualization network function optimization method and related equipment. BACKGROUND

[0002] With the rapid development of cloud computing and virtualization technology, cloud phones, as a new mobile terminal service form, are gradually widely used in cloud games, mobile office, remote control, intelligent testing and other scenarios. Cloud phone service platforms provide virtual terminal services on demand by deploying mobile operating systems and applications in cloud servers. In this process, the stability of network performance and the efficiency of resource utilization are key factors that determine service quality and user experience, especially in scenarios with high concurrency and frequent network fluctuations. How to ensure the continuity and efficiency of network connection becomes a core problem of system performance optimization.

[0003] In the prior art, for network resource allocation and scheduling in cloud phone services, static or semi-static configuration methods are usually used, such as presetting bandwidth thresholds or fixed allocation of virtual network instance numbers. Although this method is simple, it lacks the ability to perceive the dynamic changes of network load and is difficult to respond to fluctuations in actual business needs, often causing resource waste or network congestion, especially in high concurrency or burst traffic scenarios, which can easily cause service lag, response delay and data transmission bottlenecks, significantly affecting the availability of cloud phone systems and user experience. That is, the related art generally has the technical problems of resource scheduling response lag, insufficient network performance prediction, low resource utilization efficiency, and low intelligence level. SUMMARY

[0004] A series of simplified concepts are introduced in the summary part of the present application, which will be further described in detail in the specific embodiment part. The summary part of the present application does not mean to try to limit the key features and necessary technical features of the claimed technical solution, and even less means to determine the protection scope of the claimed technical solution.

[0005] The cloud phone virtualization network function optimization method and related equipment provided by the present application can realize dynamic resource scheduling and elastic management of virtualization network functions by introducing preset prediction models and preset optimization models, which can improve the network performance, resource utilization and service stability of the cloud phone system.

[0006] In a first aspect, the present application provides a cloud mobile virtualization network function optimization method, comprising: obtaining network performance monitoring data and resource utilization parameters of a cloud mobile service platform, wherein the network performance monitoring data comprises at least one of network delay, network bandwidth, data throughput and data packet loss rate, and the resource utilization parameters comprise at least one of processor utilization, memory occupancy, instance runtime, instance quantity and input / output rate; generating network demand trend data based on the network performance monitoring data through a preset prediction model; generating a dynamic resource allocation strategy based on the network demand trend data and the resource utilization parameters through a preset optimization model; and performing a dynamic adjustment operation on a virtualization network function instance deployed on the cloud mobile service platform based on the dynamic resource allocation strategy.

[0007] In some embodiments, the preset prediction model comprises a first prediction sub-model and a second prediction sub-model; the generating network demand trend data based on the network performance monitoring data through a preset prediction model comprises: performing sliding window sampling on the network performance monitoring data to generate a first time series feature sequence, wherein the size of the sliding window is adjusted according to the current system time; performing missing value filling and noise filtering processing on the first time series feature sequence to obtain a second time series feature sequence; inputting the second time series feature sequence into the first prediction sub-model to obtain a basic bandwidth demand prediction value, wherein the first prediction sub-model is a long short-term memory network model based on an attention mechanism; determining a processor utilization rate change rate based on the processor utilization rate within a preset period; inputting the processor utilization rate change rate into the second prediction sub-model to obtain a resource constraint compensation coefficient, wherein the second prediction sub-model is a gradient boosting decision tree model; generating an intermediate predicted bandwidth demand according to the product of the basic bandwidth demand prediction value and the resource constraint compensation coefficient; and performing time domain smoothing processing on the intermediate predicted bandwidth demand in combination with historical network delay fluctuation characteristics to output the network demand trend data with a confidence interval.

[0008] In some embodiments, the generating, by the preset optimization model, the dynamic resource allocation strategy based on the network demand trend data and the resource utilization parameter comprises: generating a multi-dimensional state vector based on the network demand trend data and the resource utilization parameter; inputting the multi-dimensional state vector into an action decision network of the preset optimization model to output an initial resource allocation action space, wherein the action decision network comprises a fully connected decision layer and a policy normalization layer; calculating an immediate reward value of each action in the initial resource allocation action space by a value evaluation network of the preset optimization model; generating an action value distribution matrix based on the immediate reward value; performing feasibility filtering on the action value distribution matrix based on a preset resource constraint condition to obtain an effective action subset; determining a target action with the highest value evaluation score in the effective action subset as a basic resource allocation strategy; and performing dynamic correction on the basic resource allocation strategy in combination with the processor utilization rate change to generate the dynamic resource allocation strategy comprising a bandwidth allocation weight, a routing priority coefficient, and an instance deployment density.

[0009] In some embodiments, the calculating, by the value evaluation network of the preset optimization model, the immediate reward value of each action in the initial resource allocation action space comprises: determining a first reward correction coefficient according to a confidence interval width in the network demand trend data; determining a second reward correction coefficient based on a deviation degree of a memory occupancy rate in the resource utilization parameter from a preset occupancy rate threshold; generating a basic reward value by fusing a network delay reduction gain, a throughput improvement gain, and a resource saving gain through a weighting function; and determining a product of the basic reward value and the first reward correction coefficient and the second reward correction coefficient as the immediate reward value.

[0010] In some embodiments, the preset resource constraint condition comprises at least one of a maximum instance deployment quantity constraint, a minimum available bandwidth threshold constraint, and a processor utilization rate safety boundary constraint; the maximum instance deployment quantity constraint is determined according to a total number of resource isolation units of the cloud phone service platform, the minimum available bandwidth threshold constraint is determined according to a service level agreement of the cloud phone service platform, and the processor utilization rate safety boundary constraint is determined according to a real-time monitored temperature alarm value.

[0011] In some embodiments, the dynamic adjustment operation performed on the virtualized network function instance deployed on the cloud phone service platform based on the dynamic resource allocation strategy comprises: adjusting the available bandwidth proportion of a data transmission channel of the virtualized network function instance according to the bandwidth allocation weight; updating the path selection priority in a routing forwarding table of the cloud phone service platform based on the routing priority coefficient; creating a new virtualized network function instance in a target area when the instance deployment density of the target area is less than or equal to a first preset density threshold; migrating a specified number of virtualized network function instances to other areas when the instance deployment density of the target area is greater than or equal to a second preset density threshold; and maintaining the current instance distribution state when the instance deployment density satisfies greater than the first preset density threshold and less than the second preset density threshold.

[0012] In some embodiments, the cloud phone virtualized network function optimization method further comprises: after the dynamic adjustment operation is completed, obtaining performance feedback data of the virtualized network function instance, wherein the performance feedback data comprises adjusted network delay, instance startup time consumption, and resource reuse rate; generating a strategy correction coefficient based on a deviation value of the performance feedback data and a preset expected performance threshold; inputting the strategy correction coefficient into a strategy iteration module of the preset optimization model to update the weight parameters of the action decision network and the reward calculation rule of the value evaluation network.

[0013] In a second aspect, the present application further provides a cloud phone virtualized network function optimization device, comprising: a data acquisition unit configured to acquire network performance monitoring data and resource utilization parameters of a cloud phone service platform, wherein the network performance monitoring data comprises at least one of network delay, network bandwidth, data throughput, and data packet loss rate, and the resource utilization parameters comprise at least one of processor utilization rate, memory occupancy rate, instance runtime length, instance number, and input / output rate; a trend prediction unit configured to generate network demand trend data based on the network performance monitoring data through a preset prediction model; a strategy generation unit configured to generate a dynamic resource allocation strategy based on the network demand trend data and the resource utilization parameters through a preset optimization model; and a strategy execution unit configured to perform a dynamic adjustment operation on a virtualized network function instance deployed on the cloud phone service platform based on the dynamic resource allocation strategy.

[0014] In a third aspect, the present application further provides an electronic device, comprising: a memory and a processor, wherein the processor is configured to implement the steps of the cloud phone virtualized network function optimization method of the first aspect when executing a computer program stored in the memory.

[0015] In a fourth aspect, the present application also provides a computer readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the cloud mobile phone virtualization network function optimization method of the first aspect.

[0016] In a fifth aspect, the present application also provides a computer program product comprising a computer program or computer executable instructions, which, when executed by a processor, implements the cloud mobile phone virtualization network function optimization method provided by the embodiments of the present application.

[0017] To sum up, the present application can dynamically perceive and predict network state changes by collecting network performance monitoring data in real time and generating network demand trend data based on a prediction model, so as to timely adjust resource allocation strategies, effectively reduce network delay and data transmission bottlenecks, and improve the stability and response speed of network services, thereby improving the quality of experience of cloud mobile phone users in high real-time scenarios such as cloud gaming and video streaming media. By introducing resource utilization parameters such as processor utilization, memory occupancy, and instance quantity as optimization basis, and combining the dynamically predicted network trend, intelligent and on-demand management of virtualization network function instances can be achieved, effectively avoiding the performance deficiency problems caused by resource waste and static allocation, thereby improving the overall resource utilization efficiency of the system and reducing the operating cost. Based on the preset optimization model, a dynamic resource allocation strategy is generated to construct an intelligent optimization mechanism with learning and adaptation capabilities, which can automatically adjust resource distribution according to the predicted demand changes, realize the transition from “passive response” to “active optimization”, and reduce manual intervention. By dynamically adjusting the state of the virtualization network function instance deployed in the cloud mobile phone service platform, resources can be quickly expanded or released according to changes in business load, and good elastic scheduling capability is achieved. To sum up, the cloud mobile phone virtualization network function optimization method provided by the present application realizes dynamic resource scheduling and elastic management of virtualization network functions by introducing a preset prediction model and a preset optimization model, which can improve the network performance, resource utilization, and service stability of the cloud mobile phone system. BRIEF DESCRIPTION OF DRAWINGS

[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments, and are not intended to limit the present application. Moreover, like reference numerals are used to designate like parts throughout the accompanying drawings. In the drawings:

[0019] Figure 1 A flowchart of a cloud mobile phone virtualization network function optimization method provided by an embodiment of the present application;

[0020] Figure 2A component structure schematic diagram of a cloud mobile phone virtualization network function optimization device provided by an embodiment of the present application is provided.

[0021] Figure 3 A component structure schematic diagram of an electronic device provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0022] The terms in the specification, claims and drawings of the present application, such as “first”, “second”, “third”, “fourth” and the like (if any), are used to distinguish similar objects, and are not intended to describe a specific order or sequence. Therefore, it is understood that these terms can be used interchangeably under appropriate circumstances, so that the described embodiments can adopt different orders, unless the drawings or descriptions specifically require otherwise. In addition, the terms “is” and “has” and any variants thereof in the present application are intended to cover non-exclusive inclusion of all possible constituent elements. For example, a process, method, system, product or device including several steps or units does not necessarily limit to only the explicitly listed steps or units, but can also include other steps or units not explicitly listed, or steps or units inherent to the process, method, product or device.

[0023] In the present application, “module” or “unit” refers to a computer program or a part of a computer program with a specific function, and cooperates with other related parts to achieve a predetermined target. These modules or units can be implemented by software, hardware (such as processing circuitry or memory) or a combination of both. One or more processors or memories can implement one or more modules or units. At the same time, each module or unit can also be part of a larger module or unit.

[0024] The technical solutions in the present application will be described in detail below in conjunction with the drawings in the embodiments. It should be noted that the described embodiments are only a part of the present application, not all embodiments. In the following description, “some embodiments” mentioned is only a subset of all possible embodiments, which can be the same or different subset, and different embodiments can be combined with each other without conflict.

[0025] Figure 1 A flowchart of a cloud mobile phone virtualization network function optimization method provided by an embodiment of the present application is provided. For example, see Figure 1 The cloud mobile phone virtualization network function optimization method provided by the embodiment of the present application can include the following steps 101 to 104:

[0026] In step 101, network performance monitoring data and resource utilization parameters of the cloud mobile service platform are acquired, wherein the network performance monitoring data can include at least one of network delay, network bandwidth, data throughput and data packet loss rate, and the resource utilization parameters can include at least one of processor utilization, memory occupancy, instance runtime, instance quantity and input / output rate.

[0027] In some examples, the network performance monitoring data is used to intuitively reflect the network running state of the cloud mobile service platform, wherein the network delay refers to the round-trip time consumption of data transmission between the cloud mobile and the server, which can be obtained by sending a specific probe packet to the server and calculating the response time, and directly affects the real-time feedback of user operation; the network bandwidth represents the maximum data amount that can be transmitted by the network link in unit time, which is usually allocated by the network service provider and is limited by the physical link and device performance; the data throughput is the data amount successfully transmitted in unit time in the actual transmission process, which can be calculated by monitoring the network interface traffic, and reflects the actual carrying capacity of the network; and the data packet loss rate is the proportion of data packets that do not successfully arrive at the destination, which is caused by network congestion, link failure and the like, and a too high packet loss rate will cause data transmission errors. The resource utilization parameters are used to evaluate the resource usage of the cloud mobile service platform, the processor utilization refers to the proportion of CPU resources occupied by the virtualized network function module when running, which is obtained through the operating system performance monitoring tool, and can judge whether the computing resource is tight; the memory occupancy reflects the ratio of the used memory of the virtualized network function instance to the total allocated memory, and a too high memory occupancy can cause system lag; the instance runtime records the time of each cloud mobile instance continuously working, which is used to analyze the instance life cycle and resource usage efficiency; the instance quantity is the total number of currently running cloud mobile instances, which helps to understand the service scale and load distribution; and the input / output rate refers to the data read / write speed of the storage device or network interface, which is obtained by monitoring the disk I / O or network port traffic, and can evaluate the efficiency of data storage and transmission. These parameters collectively provide data support for subsequent dynamic adjustment of network resources and optimization of virtualized network function modules.

[0028] Through the implementation of step 101, the network performance indicators and system resource utilization conditions are comprehensively collected, the running state of the cloud mobile service platform can be reflected in real time, and then the current load and resource bottleneck can be dynamically perceived, which helps to provide accurate input for subsequent prediction and optimization, and improves the accuracy and response efficiency of overall management.

[0029] In step 102, network demand trend data is generated through a preset prediction model based on the network performance monitoring data.

[0030] In some examples, network performance monitoring data, as input to the preset prediction model, covers historical collected network delay, bandwidth, throughput, packet loss rate, etc. time series data, which is collected by distributed probes at a frequency of seconds and stored in a time series database. The preset prediction model is an algorithm model based on machine learning or statistical theory, and the common types include: time series model (such as ARIMA, Prophet), which predicts future indicators by analyzing the periodicity, trend and seasonality characteristics of the data; deep learning model (such as LSTM, GRU neural network), which uses a gating mechanism to capture the nonlinear variation of network traffic; the hybrid model combines statistical methods and machine learning to improve the prediction accuracy in complex network scenarios. Model preset parameters are usually determined based on historical data training, such as the number of hidden layer nodes of LSTM, the size of the time window, etc. When generating network demand trend data, the model first performs normalization, denoising and other preprocessing on the input data, extracts features such as traffic peak period, burst traffic pattern, and then outputs network demand prediction values for the next 5 minutes to 1 hour through sliding window or rolling prediction, including bandwidth peak, delay threshold, throughput fluctuation interval, etc. Trend indicators. These trend data can be used to predict network congestion risks and plan resource expansion in advance, for example, when it is predicted that the bandwidth demand of cloud game users in a certain area will exceed the current link capacity by 20% in 30 minutes, a load balancing strategy or link bandwidth expansion can be triggered in advance.

[0031] Through the implementation of step 102, by inputting historical network performance data into the trained prediction model, the network resource demand trend in the future can be estimated in advance, which can more effectively plan resource allocation and reduce temporary congestion, resource mismatch and other problems, thereby ensuring the stable experience of high concurrency and high real-time business.

[0032] Step 103, based on network demand trend data and resource utilization parameters, a dynamic resource allocation strategy is generated through a preset optimization model;

[0033] In some examples, the network demand trend data is a quantitative indicator of network resource demand in the future period (such as 5-30 minutes) predicted by time series analysis or machine learning models (such as LSTM, ARIMA) based on historical network performance monitoring data (such as delay, bandwidth, throughput, packet loss rate), including expected bandwidth peak, delay tolerance threshold, traffic fluctuation range, etc.; the resource utilization parameters reflect the current resource usage status of the cloud phone service platform in real time, covering processor utilization (such as virtualized network function instance CPU core occupancy), memory occupancy (used memory to total allocated memory ratio), instance runtime (single virtualized network function instance continuous working time), instance number (total number of currently active virtualized network function instances), and input / output rate (such as disk I / O read / write speed, network interface data transfer rate), which are obtained through cloud platform monitoring API or container orchestration tools (such as Kubernetes). The preset optimization model is an algorithm framework based on multi-objective decision theory, common types include: reinforcement learning model, which defines resource allocation actions (such as bandwidth adjustment, instance scaling) and reward functions (such as delay reduction, resource cost minimization), iteratively learns the optimal strategy in a dynamic environment; heuristic algorithms (such as genetic algorithm, particle swarm optimization) simulate natural evolution or swarm intelligence processes to search for global or approximate optimal solutions under complex constraints; linear programming model converts resource allocation problems into objective functions (such as maximizing resource utilization, minimizing cost) under linear constraints. When generating dynamic resource allocation strategies, the optimization model first maps network demand trend data into resource demand targets (such as ensuring 200Mbps bandwidth, delay ≤50ms for the next 10 minutes), while combining resource utilization parameters to evaluate the availability of the current resource pool (such as remaining allocatable bandwidth, number of idle CPU cores), then iteratively calculates through the model internal algorithm, outputs specific resource allocation instructions, for example: add 50Mbps bandwidth quota to a regional cloud gaming virtualized network function instance, migrate instances with processor utilization exceeding 80% to physical servers with more abundant resources, start 2 standby virtualized network function instances in advance to achieve load balancing according to the predicted traffic peak, these strategies aim to balance network performance (such as low delay, high throughput) and resource cost (such as avoiding excessive allocation), ultimately automatically executed through cloud platform API or pushed to manual decision system for auxiliary operation.

[0034] Through the implementation of step 103, the predicted demand trend and the current resource status are fused, and real-time resource allocation decisions are generated through the optimization model, which can realize dynamic scheduling of resources, make optimal or suboptimal choices according to changes in network and computing resources, improve system resource utilization, reduce unnecessary resource redundancy and energy consumption, and further optimize the cost structure.

[0035] At step 104, a dynamic adjustment operation is performed on the virtualized network function instance deployed on the cloud phone service platform based on the dynamic resource allocation strategy.

[0036] In some examples, the dynamic resource allocation strategy is a specific operation instruction generated by an optimization model based on network demand trend data and resource utilization parameters, such as a bandwidth quota adjustment scheme, a virtualized network function instance scaling plan, or a node migration suggestion, etc. These strategies provide clear execution basis for the dynamic adjustment of virtualized network function instances. The virtualized network function instance deployed on the cloud phone service platform, i.e. the running instance of the virtualized network function module, can exist in the form of a container (such as a Docker container) or a lightweight virtual machine (such as KVM, Xen), covering the instantiation components of data forwarding, protocol parsing, traffic control, etc. The running state of the virtualized network function instance directly affects the network performance and resource efficiency. The dynamic adjustment operation includes: starting a new virtualized network function instance to cope with sudden traffic peaks, such as adding data forwarding instances when the concurrent number of cloud game users increases; stopping redundant instances to release resources, such as closing idle traffic control instances when network load decreases; migrating instances to physical nodes with more resources, such as migrating high-load protocol parsing instances from servers with excessive CPU utilization to idle servers to balance cluster load. The adjustment operation is executed through the cloud platform management system or container orchestration tools (such as Kubernetes), using API interfaces to manage the life cycle of instances, and combining service discovery and registration mechanisms to ensure the continuity of network connection during instance migration. The core role of this step is to convert the resource allocation scheme at the strategy level into actual instance operations, and through the elastic adjustment of the number, distribution and resource quota of virtualized network function instances, to achieve the dual goals of network performance optimization and resource utilization improvement, such as dynamically scaling content distribution instances during video streaming peak periods to reduce video loading delay and avoid excessive consumption of server resources.

[0037] Through the implementation of step 104, after the strategy is formulated, dynamic operations such as bandwidth adjustment, priority update, instance addition or reduction, or migration can be automatically performed on the virtualized network function instance deployed on the cloud phone service platform, which can improve the adaptability of the cloud phone service platform to sudden traffic, peak load or regional congestion, has good scalability and scheduling flexibility, and helps to maintain stable service quality and avoid performance degradation due to rigid resource configuration.

[0038] In summary, the embodiment of the present application can dynamically perceive and predict network state changes by collecting network performance monitoring data in real time and generating network demand trend data based on a prediction model, thereby timely adjusting resource allocation strategies, effectively reducing network latency and data transmission bottlenecks, and improving the stability and response speed of network services to improve the quality of experience of cloud phone users in high real-time scenarios such as cloud games and video streaming media. The processor utilization, memory occupancy, and instance number are introduced as optimization basis, and the dynamically predicted network trend can realize intelligent and on-demand management of virtualized network function instances, effectively avoiding the performance deficiency problem caused by resource waste and static allocation, thereby improving the overall resource utilization efficiency of the system and reducing the operating cost. Based on the preset optimization model, a dynamic resource allocation strategy is generated to construct an intelligent optimization mechanism with learning and adaptation ability, which can automatically adjust the resource distribution according to the predicted demand changes, realize the transformation from "passive response" to "active optimization", and reduce manual intervention. By dynamically adjusting the state of the virtualized network function instance deployed in the cloud phone service platform, the resource can be quickly expanded or released according to the business load changes, and the system has good elastic scheduling capability. In summary, the cloud phone virtualized network function optimization method provided by the embodiment of the present application realizes dynamic resource scheduling and elastic management of virtualized network functions by introducing a preset prediction model and a preset optimization model, which can improve the network performance, resource utilization, and service stability of the cloud phone system.

[0039] In some embodiments, the foregoing preset prediction model can include a first prediction sub-model and a second prediction sub-model; and the foregoing step 102 can include: performing sliding window sampling on the network performance monitoring data to generate a first time series feature sequence, wherein the size of the sliding window is adjusted according to the current system time; performing missing value filling and noise filtering processing on the first time series feature sequence to obtain a second time series feature sequence; inputting the second time series feature sequence into the first prediction sub-model to obtain a basic bandwidth demand prediction value, wherein the first prediction sub-model is a long short-term memory network model based on an attention mechanism; determining a processor utilization rate change rate based on the processor utilization rate within a preset period; inputting the processor utilization rate change rate into the second prediction sub-model to obtain a resource constraint compensation coefficient, wherein the second prediction sub-model is a gradient boosting decision tree model; generating an intermediate predicted bandwidth demand according to the product of the basic bandwidth demand prediction value and the resource constraint compensation coefficient; and performing time domain smoothing processing on the intermediate predicted bandwidth demand in combination with historical network latency fluctuation characteristics to output network demand trend data with a confidence interval.

[0040] In some examples, the preset prediction model is composed of a first prediction sub-model and a second prediction sub-model. The first prediction sub-model is an attention mechanism based long short-term memory network model (LSTM-Attention), which is responsible for capturing the time-dependent features of network bandwidth demand. The second prediction sub-model is a gradient boosting decision tree model (GBDT), which is used to quantify the impact of processor resource constraints on bandwidth demand. The combination of the two models achieves more accurate network demand prediction. Sliding window sampling divides continuous network performance monitoring data into fixed-length sequence segments according to the time dimension. The window size is dynamically adjusted according to the current system time. For example, a shorter window (e.g., 5 minutes) is used during network traffic peak periods (e.g., evening cloud gaming period) to capture high-frequency fluctuations, and a longer window (e.g., 15 minutes) is used during low peak periods to smooth noise. The generated first time series features include the timestamp and numerical information of the original sampling data. Missing value filling and noise filtering are preprocessing of the first time series features. Missing values are filled using linear interpolation or forward filling methods, and noise filtering uses moving average or wavelet transform algorithms to eliminate abnormal values caused by network transient fluctuations (such as burst packet loss), resulting in a smoother second time series features, providing a stable data basis for model input. The first prediction sub-model (LSTM-Attention) receives the second time series features, captures long-term dependencies (such as daily / weekly periodic traffic patterns) through LSTM units, and assigns weights to features at different time steps (such as focusing on the last 1 hour of data) through an attention mechanism. The output is the basic bandwidth demand prediction value, which reflects the theoretical bandwidth demand without considering resource constraints. The processor utilization rate of change is the derivative calculated based on the processor utilization rate within a preset period (e.g., the last 30 minutes), obtained through the difference method (current value - previous value) / time interval, reflecting the growth or decline trend of resource usage. This indicator is collected in real time by the system monitoring component and normalized to the [-1, 1] interval. The second prediction sub-model (GBDT) takes the processor utilization rate of change as input and outputs a resource constraint compensation coefficient (range 0-1.5) by integrating the prediction results of multiple decision trees. When the processor utilization rate rises rapidly (e.g., change rate > 0.3), the coefficient < 1 indicates that resources are tight and the bandwidth prediction value needs to be adjusted downward. When the utilization rate decreases (e.g., change rate <-0.2), the coefficient > 1 indicates that resources are abundant and the prediction value can be adjusted upward. The intermediate predicted bandwidth demand is the product of the basic bandwidth demand prediction value and the resource constraint compensation coefficient, which realizes the resource adaptability correction of the original prediction. Time domain smoothing processing combines historical network delay fluctuation features (such as standard deviation, peak frequency) to perform weighted averaging on the intermediate predicted bandwidth demand using exponential smoothing methods (such as the Holt-Winters model), outputting network demand trend data with a confidence interval (such as 95% confidence). The width of the confidence interval is determined by the fluctuation amplitude of the historical data (the larger the fluctuation, the wider the interval). This data not only contains the prediction mean, but also provides an uncertainty range, providing a more comprehensive decision basis for subsequent resource allocation strategies.

[0041] Through the implementation of the above embodiments, the dual prediction model composed of the long short-term memory network model with attention mechanism and the gradient boosting decision tree model is introduced, which not only improves the time series prediction accuracy of the bandwidth demand trend, but also combines the constraint influence of resource usage change on network load to generate a prediction result with better context awareness. In particular, through the processing means such as sliding window, adaptive processing of missing and noise information, and output with confidence interval, the robustness and generalization ability of the model can be enhanced, and the challenges such as large data fluctuation and difficult to fix prediction window in actual cloud platform can be effectively coped with, so as to ensure the accuracy and stability of the prediction result under various business load conditions.

[0042] In some embodiments, the foregoing step 103 can include: generating a multi-dimensional state vector based on the foregoing network demand trend data and resource utilization parameters; inputting the multi-dimensional state vector into an action decision network of a preset optimization model to output an initial resource allocation action space, wherein the action decision network can include a fully connected decision layer and a policy normalization layer; calculating the immediate reward value of each action in the initial resource allocation action space through a value evaluation network of the foregoing preset optimization model; generating an action value distribution matrix based on the immediate reward value; performing feasibility filtering on the action value distribution matrix based on a preset resource constraint condition to obtain an effective action subset; determining a target action with the highest value evaluation score in the effective action subset as a basic resource allocation strategy; and performing dynamic correction on the basic resource allocation strategy in combination with the processor utilization rate change to generate a dynamic resource allocation strategy which can include bandwidth allocation weight, routing priority coefficient and instance deployment density.

[0043] In some examples, the network demand trend data covers indicators such as bandwidth demand prediction values, delay tolerance thresholds, and traffic fluctuation ranges in a future period of time, which are generated by a prediction model based on historical network performance data, and the resource utilization parameters include real-time monitoring data such as processor utilization, memory occupancy, and instance runtime, which together constitute the basis information of the multi-dimensional state vector. The multi-dimensional state vector is a numerical sequence formed by arranging the network demand trend data and the resource utilization parameters in a specific order, such as [predicted bandwidth peak, delay tolerance threshold, processor utilization, memory occupancy, and instance quantity], which comprehensively reflects the current system's network demand and resource state as input data for the preset optimization model. The action decision network of the preset optimization model includes a fully connected decision layer and a policy normalization layer. The fully connected decision layer performs feature extraction and nonlinear transformation on the multi-dimensional state vector through multiple layers of neurons to mine potential relationships between data, and the policy normalization layer maps the action parameters output by the fully connected decision layer to a reasonable value range, such as mapping the bandwidth adjustment ratio to 0-1, thereby outputting an initial resource allocation action space that includes a series of possible resource allocation actions such as increasing bandwidth, adjusting routing, and starting or stopping instances. The value evaluation network is a component in the preset optimization model that quantifies the impact of each action on the system. It calculates the immediate reward value of each action in the initial resource allocation action space based on network demand and resource state, combined with a pre-set reward mechanism (such as giving positive rewards for delay reduction and negative rewards for resource waste), which reflects the short-term change in system performance after executing a certain action. The action value distribution matrix is a matrix representation of each action and its corresponding immediate reward value. Each row of the matrix represents an action, and the columns contain the specific parameters and reward scores of the action, which visually displays the value differences of different actions. The preset resource constraints include hardware resource upper limits (such as server maximum bandwidth and CPU core number limits), quality of service requirements (such as a delay that must be lower than 50 ms), etc. When performing feasibility filtering on the action value distribution matrix, the system automatically removes actions that violate the constraint conditions, such as bandwidth allocation schemes that exceed the server bandwidth capacity, thereby obtaining an effective action subset. The target action with the highest value evaluation score is selected from the effective action subset to determine the basic resource allocation strategy, which has the optimal performance in balancing network performance and resource utilization.The processor utilization rate change rate reflects the dynamic trend of processor resource use. When the dynamic correction is performed on the base resource allocation strategy in combination with the change rate, if the processor utilization rate rapidly rises, the system reduces the bandwidth allocation weight to avoid resource overload, increases the routing priority coefficient to optimize the data transmission path to reduce processing pressure, and reduces the instance deployment density to prevent excessive tasks from competing for resources. If the utilization rate decreases, the parameters are adjusted in the opposite direction. Finally, a dynamic resource allocation strategy including the bandwidth allocation weight, the routing priority coefficient, and the instance deployment density is generated. The strategy can flexibly adjust resource allocation according to the real-time state of the system, and realizes the dynamic balance of network performance and resource utilization.

[0044] Through the implementation of the above embodiments, an optimization model including state evaluation, action selection, and value evaluation is constructed, the traditional static strategy is replaced by a dynamic scheduling scheme with intelligent decision-making ability, and especially the optimal resource allocation strategy is generated by the cooperation of the action decision network and the value evaluation network. Not only is the fine response to network demand realized, but also the feasibility filtering mechanism and the dynamic correction means are introduced, which can optimize the key parameters such as bandwidth allocation, routing path, and instance density in real time according to the change of processor load, so as to realize the intelligent, fine, and flexible scheduling of the cloud phone network resources, and improve the resource utilization rate and service continuity.

[0045] In some embodiments, the aforementioned value evaluation network through the preset optimization model calculates the immediate reward value of each action in the initial resource allocation action space, which can include: determining a first reward correction coefficient according to the confidence interval width in the network demand trend data; determining a second reward correction coefficient based on the deviation degree of the memory occupancy rate in the resource utilization parameter and the preset occupancy rate threshold; generating a basic reward value by fusing the network delay reduction gain, the throughput improvement gain, and the resource saving gain through a weighting function; and determining the product of the basic reward value and the first reward correction coefficient and the second reward correction coefficient as the immediate reward value.

[0046] In some examples, in the process of calculating the immediate reward value of each action in the initial resource allocation action space by the value evaluation network of the preset optimization model, the confidence interval width in the network demand trend data is generated by the prediction model when outputting the network demand trend data, which reflects the uncertainty degree of the prediction result, the greater the width, the lower the credibility of the prediction, and the system maps the confidence interval width to a certain range to determine the first reward correction coefficient, if the confidence interval width is narrow, the prediction is more accurate, the first reward correction coefficient is close to 1, and the subsequent reward value is less affected, if the width is wide, the correction coefficient is less than 1, and is used to reduce the reward risk caused by prediction uncertainty. The memory occupancy rate in the resource utilization parameter is the ratio of the used memory of the virtualized network function instance to the total allocated memory obtained by real-time monitoring of the cloud platform, and the preset occupancy rate threshold is a reasonable upper limit of memory usage preset according to the stable operation demand of the system. The deviation degree of the memory occupancy rate and the preset occupancy rate threshold reflects the tension degree of the current memory resource, the greater the deviation degree, the more tense the memory resource, and the system determines the second reward correction coefficient according to the deviation degree, when the memory occupancy rate is close to or exceeds the threshold, the second reward correction coefficient is less than 1, to punish the action that may further aggravate the memory burden, otherwise, when the memory resource is abundant, the correction coefficient is greater than 1 to encourage the action of fully utilizing the resource. The generation of the basic reward value depends on the fusion of the network delay reduction gain, the throughput improvement gain and the resource saving gain by the weighting function, the network delay reduction gain refers to the performance improvement benefit brought by the reduction of network delay after executing a certain resource allocation action, the gain value is set by comparing the delay difference before and after the action and combining the delay sensitivity of the user; the throughput improvement gain is the benefit brought by the improvement of data transmission efficiency after the action, which is calculated according to the throughput change; the resource saving gain is the benefit generated by the saving of resources (such as CPU, memory, bandwidth) after the action, which is evaluated by quantifying the resource saving amount, the weighting function gives different weights to the three kinds of gains according to the current emphasis of the system on different indicators, adds them to obtain the basic reward value, which reflects the comprehensive improvement effect of the action on the system performance and resource utilization in the ideal case. Finally, the basic reward value is multiplied by the first reward correction coefficient and the second reward correction coefficient to obtain the immediate reward value, which considers the prediction uncertainty, the memory resource status and the improvement of the action itself on the system performance, and is used to measure the actual value of each resource allocation action in the current system state, providing a quantitative basis for subsequent selection of the optimal resource allocation strategy.

[0047] Through implementation of the foregoing embodiments, the reward value calculation mechanism of fused multi-dimensional indexes is adopted, and correction factors such as confidence interval width and resource deviation degree are introduced, so that the yield effect of different resource scheduling actions can be more reasonably evaluated; compared with the traditional single or static weight method, the embodiments of the present application have higher adaptability and differentiation ability in the multi-objective optimization scene, and are particularly suitable for the cloud phone platform environment in which the network load fluctuates greatly and the resource bottleneck frequently appears, so as to ensure that the scheduling strategy is not only effective at present, but also has stronger forward-looking and robustness.

[0048] In some embodiments, the foregoing preset resource constraint condition can include at least one of a maximum instance deployment quantity constraint, a minimum available bandwidth threshold constraint, and a processor utilization safety boundary constraint; the maximum instance deployment quantity constraint is determined according to the total number of resource isolation units of the cloud phone service platform, the minimum available bandwidth threshold constraint is determined according to the service level agreement of the cloud phone service platform, and the processor utilization safety boundary constraint is determined according to the real-time monitored temperature alarm value.

[0049] In some examples, the preset resource constraints are key rules to ensure the stable operation of the cloud phone service platform, including maximum instance deployment quantity constraint, minimum available bandwidth threshold constraint, and processor utilization safety boundary constraint, etc. Each constraint is set based on specific system resources and service requirements. The maximum instance deployment quantity constraint is determined by the total number of resource isolation units of the cloud phone service platform. The resource isolation unit is an independent running environment divided by containerization or virtual machine technology, which is used to ensure that virtualized network function modules do not interfere with each other. For example, if the platform uses Kubernetes for container orchestration, the upper limit of the number of containers that can be divided on each physical server is the total number of resource isolation units. The maximum instance deployment quantity needs to reserve 10%-20% of the redundant space on this basis to cope with sudden traffic or fault switching, so as to avoid system crash caused by resource contention due to excessive instance deployment. The minimum available bandwidth threshold constraint is determined according to the service level agreement (SLA) signed between the cloud phone service platform and the user. The SLA clearly specifies the minimum network quality requirements for services such as cloud gaming and video streaming media. For example, the cloud gaming SLA requires a delay of less than 50ms and a bandwidth of not less than 10Mbps. The minimum available bandwidth threshold is set according to these indicators and combined with a certain redundancy coefficient (such as 1.2 times) to ensure that the service quality can still be met when the network fluctuates, preventing the user experience from being reduced due to insufficient bandwidth. The processor utilization safety boundary constraint is based on the real-time monitoring of the temperature alarm value. When the server CPU is running at high load, it will generate a lot of heat. When the temperature reaches the preset alarm value (such as 80℃), the processor performance will decrease due to automatic frequency reduction, and even may cause hardware failure. Therefore, the system reverses the corresponding processor utilization threshold (such as 75%-80%) based on the temperature alarm value as the safety boundary constraint. When the processor utilization is close to the boundary, new resource allocation requests are limited or instance migration is triggered to avoid performance degradation due to overheating. These constraint conditions work together to filter potential solutions during dynamic resource allocation, ensuring that the allocation strategy meets business requirements while ensuring the stability and reliability of the system.

[0050] By implementing the above embodiments, the actual capacity of the platform and the service level agreement requirements are mapped into resource constraints. The optimization model can automatically exclude infeasible or high-risk operations during the strategy generation process, ensuring that the generated solution balances between physical resource limitations and service guarantees, which can improve the security and stability of the system, especially in multi-tenant or business priority difference obvious cloud phone deployment environment, which helps to achieve resource fair allocation and differentiated service quality guarantee.

[0051] In some embodiments, the foregoing step 104 can include: adjusting the proportion of available bandwidth of the data transmission channel of the virtualized network function instance according to the bandwidth allocation weight; updating the path selection priority in the routing forwarding table of the cloud phone service platform based on the routing priority coefficient; creating a new virtualized network function instance in the target area when the instance deployment density of the target area is less than or equal to a first preset density threshold; migrating a specified number of virtualized network function instances to other areas when the instance deployment density of the target area is greater than or equal to a second preset density threshold; and maintaining the current instance distribution state when the instance deployment density satisfies greater than the first preset density threshold and less than the second preset density threshold.

[0052] In some examples, the bandwidth allocation weight is a key parameter in the dynamic resource allocation strategy, calculated by a preset optimization model according to network demand trend data and resource utilization parameters, usually ranging from 0 to 1, used to accurately adjust the proportion of available bandwidth of the data transmission channel of the virtualized network function instance. Specifically, the system will redivide the link bandwidth occupied by each virtualized network function instance through the network management interface of the cloud platform. For example, when the bandwidth allocation weight of a certain cloud game service is 0.8, the system will increase the available bandwidth of the data transmission channel where the service is located to 80% of the total bandwidth, thereby ensuring low-latency and high-stability transmission of game data. The routing priority coefficient is also generated by the optimization model and is used to update the path selection priority in the routing forwarding table of the cloud phone service platform. The routing forwarding table is a core configuration file that guides the transmission path of data in the network. The system will sort different transmission paths according to the routing priority coefficient, and the path with a higher coefficient will be selected first during data forwarding, thereby achieving routing optimization and reducing data transmission delay and congestion. The instance deployment density refers to the number of virtualized network function instances per unit of computing resource, used to measure the intensity of resource usage. The first preset density threshold and the second preset density threshold are two critical values preset according to the hardware resource carrying capacity and service quality requirements of the cloud phone service platform. When the instance deployment density of the target area is less than or equal to the first preset density threshold, it means that the resource utilization of the area is low, and the system will create new virtualized network function instances in the target area through a container orchestration tool (such as Kubernetes) to improve resource utilization and cope with potential traffic growth. When the instance deployment density is greater than or equal to the second preset density threshold, it indicates that the resources of the area are approaching saturation, which may cause performance degradation. At this time, the system will migrate a specified number of virtualized network function instances to other areas with relatively abundant resources according to the preset migration strategy, achieving load balancing. When the instance deployment density is greater than the first preset density threshold and less than the second preset density threshold, it means that the current instance distribution state is reasonable, and the system will maintain the status quo without additional instance adjustment operations. Through these operations based on different thresholds, the cloud phone service platform can always maintain efficient and stable operation in a dynamically changing network environment.

[0053] Through the implementation of the above embodiments, the virtualized network function instance is dynamically adjusted from three dimensions of bandwidth, routing, and instance deployment, effectively realizing comprehensive scheduling and management of network transmission paths, load distribution, and computing resources; supporting automatic addition, deletion, or migration of instances according to deployment density, with good scalability and regional load balancing capability; ensuring that the cloud phone system can maintain an efficient operation state under different business density and network fluctuation conditions, reducing congestion and resource redundancy, and improving system operation efficiency.

[0054] In some embodiments, the aforementioned cloud mobile virtualization network function optimization method can further include: after completing the aforementioned dynamic adjustment operation, obtaining performance feedback data of the virtualization network function instance, wherein the performance feedback data can include the adjusted network delay, instance startup time consumption, and resource reuse rate; based on the deviation value between the performance feedback data and the preset expected performance threshold, generating a policy correction coefficient; inputting the policy correction coefficient into a policy iteration module of the preset optimization model to update the weight parameters of the action decision network and the reward calculation rule of the value evaluation network.

[0055] In some examples, in the cloud mobile virtualization network function optimization process, after completing the dynamic adjustment operation, the system needs to evaluate and feedback the optimization effect, so as to realize the continuous iteration of the strategy. The performance feedback data is a key indicator for measuring the effectiveness of the dynamic adjustment operation. The adjusted network delay is obtained by sending a probe data packet to the virtualization network function instance and calculating the round-trip time, which is used to intuitively reflect the network transmission efficiency. The instance startup time consumption records the time from triggering the instance startup instruction to the instance completely entering the available state, which reflects the response speed of resource scheduling. The resource reuse rate is the ratio of the reused resource amount to the total resource amount, which is used to evaluate the resource utilization efficiency. These data are collected in real time by the cloud platform monitoring component. The preset expected performance threshold is a standard set in advance according to the service level agreement (SLA) of the cloud mobile service and the user experience requirements, for example, the network delay needs to be less than 50 ms, the instance startup time consumption needs to be no more than 3 seconds, and the resource reuse rate needs to be higher than 70%. Based on the deviation value between the performance feedback data and the preset expected performance threshold, the system generates a policy correction coefficient through a specific algorithm. If the actual network delay is higher than the threshold, it means that the optimization effect is not as expected, and the policy correction coefficient will be less than 1, otherwise it will be greater than 1. The same applies to the instance startup time consumption and the resource reuse rate. The coefficient quantifies the gap between the current strategy and the ideal state. The policy iteration module of the preset optimization model is the core component for realizing the self-optimization of the model. After inputting the policy correction coefficient into the module, the action decision network will adjust the weight parameters of each layer of neurons according to the coefficient, for example, in the fully connected decision layer, the influence weights of different input features are redistributed, so that the network pays more attention to the factors that have a greater impact on performance in subsequent decision-making. The value evaluation network updates the reward calculation rule according to the policy correction coefficient, such as adjusting the weights of indicators such as network delay reduction gain and throughput improvement gain, so that the reward mechanism is more in line with the actual needs. In this way, the system can continuously optimize the resource allocation strategy, gradually reduce the gap between the actual performance and the expected performance, improve the network performance and resource utilization rate of the cloud mobile service platform, and realize the closed-loop management of dynamic optimization.

[0056] Through the implementation of the above embodiments, a performance feedback loop is introduced to obtain the actual operating effect of the network instance after each round of resource adjustment, and the policy correction factor is automatically generated by comparison with the expected target. The factor is fed back to the model for weight and rule update, and a continuous iterative optimization mechanism for the policy is constructed. This can give the optimization system continuous learning and self-adaptation capabilities, enhance the system's adaptability to changing business scenarios and long-term scheduling performance, and provide key support for building a closed-loop intelligent resource management system.

[0057] Furthermore, as an implementation of the aforementioned method embodiment, the present application also provides a cloud phone virtualized network function optimization device for implementing the aforementioned method embodiment. This device embodiment corresponds to the aforementioned method embodiment. For ease of reading, this cloud phone virtualized network function optimization device embodiment will no longer repeat the details of the aforementioned method embodiment one by one, but it should be clear that the device in the embodiment of the present application can correspond to all the contents of the aforementioned method embodiment. Figure 2 As shown, the cloud phone virtualized network function optimization device 20 includes: a data acquisition unit 201, a trend prediction unit 202, a policy generation unit 203 and a policy execution unit 204, wherein the data acquisition unit 201 is used to obtain network performance monitoring data and resource utilization parameters of the cloud phone service platform, wherein the aforementioned network performance monitoring data may include at least one of network delay, network bandwidth, data throughput and data packet loss rate, and the aforementioned resource utilization parameters may include at least one of processor utilization, memory occupancy, instance running time, number of instances, and input and output rates; the trend prediction unit 202 is used to generate network demand trend data through a preset prediction model based on the network performance monitoring data; the policy generation unit 203 is used to generate a dynamic resource allocation strategy through a preset optimization model based on the network demand trend data and resource utilization parameters; the policy execution unit 204 is used to perform dynamic adjustment operations on the virtualized network function instances deployed on the cloud phone service platform based on the dynamic resource allocation strategy.

[0058] The present application also provides a computer-readable storage medium, which stores computer-executable instructions or computer programs. When the computer-executable instructions or computer programs are executed by a processor, the processor will execute any step of the cloud phone virtualization network function optimization method provided in the present application.

[0059] In some embodiments, the computer-readable storage media can be a memory such as random access memory (RAM), read-only memory (ROM), flash memory, a magnetic storage, an optical disc, or a Compact Disc Read-Only Memory (CD-ROM), etc. The computer-readable storage media can also be various devices including one or any combination of the above-mentioned memory.

[0060] In some embodiments, the computer-executable instructions can take the form of programs, software, software modules, scripts, or code, written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0061] In some embodiments, the computer-executable instructions can, but need not, correspond to files in a file system, can be stored in a portion of a file that holds other programs or data, for example, one or more scripts stored in a Hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files, for example, files that store one or more modules, sub programs, or portions of code.

[0062] In some embodiments, the computer-executable instructions can be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed across multiple sites and interconnected by a communication network.

[0063] As shown in Figure 3 The present application also provides an electronic device 30, which includes a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, any step of the cloud phone virtualization network function optimization method described above is implemented.

[0064] The present application also provides a computer program product, which includes a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer program or computer-executable instructions from the computer-readable storage medium, and the processor executes the computer program or computer-executable instructions, so that the electronic device performs any step of the cloud phone virtualization network function optimization method described above.

[0065] The above examples are only used to illustrate the technical solutions of the present application, but not to limit the same; although the present application has been described in detail with reference to the foregoing examples, it should be understood by those of ordinary skill in the art that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A cloud phone virtualization network function optimization method, characterized in that, The method comprises: obtaining network performance monitoring data and resource utilization parameters of a cloud mobile service platform, wherein the network performance monitoring data comprises at least one of network delay, network bandwidth, data throughput and data packet loss rate, and the resource utilization parameters comprise at least one of processor utilization, memory occupancy, instance runtime, instance quantity and input / output rate; generating network demand trend data based on the network performance monitoring data through a preset prediction model; generating a dynamic resource allocation strategy based on the network demand trend data and the resource utilization parameters through a preset optimization model; performing a dynamic adjustment operation on a virtualized network function instance deployed on the cloud mobile service platform based on the dynamic resource allocation strategy.

2. The cloud-hands optimization method of virtualizing network functions according to claim 1, characterized in that, The preset prediction model comprises a first prediction sub-model and a second prediction sub-model; and the generation of the network demand trend data based on the network performance monitoring data through the preset prediction model comprises: performing sliding window sampling on the network performance monitoring data to generate a first time sequence feature sequence, wherein the size of the sliding window is adjusted according to the current system time; performing missing value filling and noise filtering processing on the first time sequence feature sequence to obtain a second time sequence feature sequence; inputting the second time sequence feature sequence into the first prediction sub-model to obtain a basic bandwidth demand prediction value, wherein the first prediction sub-model is a long short-term memory network model based on an attention mechanism; determining a processor utilization rate change rate based on the processor utilization rate within a preset period; inputting the processor utilization rate change rate into the second prediction sub-model to obtain a resource constraint compensation coefficient, wherein the second prediction sub-model is a gradient boosting decision tree model; generating an intermediate predicted bandwidth demand according to the product of the basic bandwidth demand prediction value and the resource constraint compensation coefficient; performing time domain smoothing processing on the intermediate predicted bandwidth demand in combination with historical network delay fluctuation characteristics to output the network demand trend data with a confidence interval.

3. The cloud phone virtualized network function optimization method according to claim 2, characterized in that: The generation of the dynamic resource allocation strategy based on the network demand trend data and the resource utilization parameters through the preset optimization model comprises: generating a multi-dimensional state vector based on the network demand trend data and the resource utilization parameters; inputting the multi-dimensional state vector into an action decision network of the preset optimization model to output an initial resource allocation action space, wherein the action decision network comprises a fully connected decision layer and a policy normalization layer; calculating an immediate reward value of each action in the initial resource allocation action space through a value evaluation network of the preset optimization model; generating an action value distribution matrix based on the immediate reward value; performing feasibility filtering on the action value distribution matrix based on a preset resource constraint condition to obtain an effective action subset; determining a target action with the highest value evaluation score in the effective action subset as a basic resource allocation strategy; performing dynamic correction on the basic resource allocation strategy in combination with the processor utilization rate change rate to generate the dynamic resource allocation strategy comprising a bandwidth allocation weight, a routing priority coefficient and an instance deployment density.

4. The cloud-hands optimization method of claim 3, wherein, The value evaluation network passing through the preset optimization model calculates an immediate reward value of each action in the initial resource allocation action space, and the value evaluation network comprises: According to the confidence interval width in the network demand trend data, a first reward correction coefficient is determined; Based on the deviation degree of the memory occupancy rate in the resource utilization parameter and the preset occupancy rate threshold, a second reward correction coefficient is determined; The basic reward value is generated by fusing the network delay reduction gain, the throughput improvement gain and the resource saving gain through a weighting function; The product of the basic reward value and the first reward correction coefficient and the second reward correction coefficient is determined as the immediate reward value.

5. The cloud-hands optimization method of virtualizing network functions according to claim 3, characterized in that, The preset resource constraint condition comprises at least one of a maximum instance deployment quantity constraint, a minimum available bandwidth threshold constraint and a processor utilization safety boundary constraint; the maximum instance deployment quantity constraint is determined according to the total number of resource isolation units of the cloud mobile service platform, the minimum available bandwidth threshold constraint is determined according to the service level agreement of the cloud mobile service platform, and the processor utilization safety boundary constraint is determined according to the real-time monitored temperature alarm value.

6. The cloud-hands optimization method of virtualizing network functions according to claim 3, characterized in that, The dynamic adjustment operation is performed on the virtualized network function instance deployed on the cloud mobile service platform based on the dynamic resource allocation strategy, and the dynamic adjustment operation comprises: According to the bandwidth allocation weight, the available bandwidth proportion of the data transmission channel of the virtualized network function instance is adjusted; Based on the routing priority coefficient, the path selection priority in the routing forwarding table of the cloud mobile service platform is updated; When the instance deployment density of the target area is less than or equal to a first preset density threshold, a new virtualized network function instance is created in the target area; When the instance deployment density of the target area is greater than or equal to a second preset density threshold, a specified number of virtualized network function instances are migrated to other areas; When the instance deployment density satisfies greater than the first preset density threshold and less than the second preset density threshold, the current instance distribution state is maintained.

7. The cloud-hands optimization method of virtualizing network functions according to claim 3, characterized in that, The cloud mobile virtualized network function optimization method further comprises: After the dynamic adjustment operation is completed, performance feedback data of the virtualized network function instance is obtained, wherein the performance feedback data comprises adjusted network delay, instance startup time consumption and resource reuse rate; Based on the deviation value of the performance feedback data and a preset expected performance threshold, a strategy correction coefficient is generated; The strategy correction coefficient is input into a strategy iteration module of the preset optimization model, and the weight parameter of the action decision network and the reward calculation rule of the value evaluation network are updated.

8. A cloud phone virtualization network function optimization apparatus, characterized in that, Comprise: The data acquisition unit is configured to acquire network performance monitoring data and resource utilization parameters of a cloud mobile service platform, wherein the network performance monitoring data comprises at least one of network delay, network bandwidth, data throughput and data packet loss rate, and the resource utilization parameters comprise at least one of processor utilization rate, memory occupancy rate, instance running time, instance quantity and input / output rate; The trend prediction unit is configured to generate network demand trend data by a preset prediction model based on the network performance monitoring data; A policy generation unit is configured to generate a dynamic resource allocation policy based on the network demand trend data and the resource utilization parameters through a preset optimization model. A policy execution unit is configured to perform a dynamic adjustment operation on a virtualized network function instance deployed on the cloud phone service platform based on the dynamic resource allocation policy.

9. An electronic device comprising: A memory and a processor, wherein the processor is configured to implement the steps of the cloud phone virtualized network function optimization method according to any one of claims 1 to 7 when executing a computer program stored in the memory.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is configured to implement the steps of the cloud phone virtualized network function optimization method according to any one of claims 1 to 7 when executed by the processor.

Images