Node stability assessment method and system for terminal computing power network

Abstract

The invention discloses a node stability assessment method and system for a terminal computing power network, and relates to the technical field of node stability assessment, and the method comprises the steps: collecting resource index data of nodes, and guaranteeing that the number of data points of each resource index is consistent and the time is synchronous; performing standardization processing on the resource index data; performing feature extraction on the standardized data by adopting a PCA method; constructing a GRU model, inputting the selected feature data into the GRU model, and outputting predicted values of related dimension indexes by the GRU model; and calculating a comprehensive stability score of the node according to the predicted value of the related dimension index. According to the method, the comprehensive stability of the node can be predicted and evaluated by utilizing the resource indexes collected by the end-side equipment, and a data basis is provided for the management of the end-side node and the scheduling of the application by a terminal computing network system.

Description

Technical Field

[0001] The present invention relates to the technical field of node stability evaluation, and specifically to a method and system for evaluating the node stability of a terminal computing power network. Background Art

[0002] In today's era, the Internet of Things and edge computing are booming at an unprecedented speed, making the status of end-side devices increasingly crucial. Take smartphones, which are indispensable tools in people's daily lives. They not only undertake conventional functions such as communication, social networking, and entertainment but also play the role of data collection and preliminary processing in the Internet of Things system. For example, they collect users' motion data, location information, etc. through various sensors, and then provide support for applications such as health monitoring and intelligent transportation. Looking at sensors again, they are widely distributed in various fields such as industrial production, environmental monitoring, and smart homes. From monitoring the operating status of equipment on the factory production line to real-time perception of urban air quality, and then to the collection of temperature, humidity, and security information in the home, sensors are indispensable. Embedded systems are deeply embedded in various devices such as smart home appliances, vehicle-mounted systems, and medical devices, endowing these devices with intelligent control and data processing capabilities.

[0003] However, these end-side devices face many thorny problems during actual operation. In terms of resources, whether it is computing resources, storage resources, or energy supply, they are very limited. Take some small sensors as an example. The performance of their computing chips is relatively weak and they cannot perform complex operations. Moreover, their storage capacity may only be a few KB to dozens of KB, making it difficult to store a large amount of data. At the same time, in order to achieve long-term autonomous operation, their energy supply often relies on small batteries, resulting in limited battery life. From the perspective of hardware configuration, the hardware configurations of end-side devices produced by different manufacturers vary greatly. From entry-level to flagship smartphones, the processor performance, memory capacity, storage type, etc. are all different; there are also significant differences in the accuracy, sensitivity, communication interfaces, etc. of sensors; embedded systems have great flexibility in hardware selection according to different application scenarios. This diversity of hardware configurations makes unified system adaptation and optimization extremely difficult. In addition, the operating environment of end-side devices is dynamically changing. In industrial sites, sensors may face harsh environments such as high temperature, high humidity, and strong electromagnetic interference; when used outdoors, smartphones are affected by different temperatures, humidities, and changes in network signal strength; vehicle-mounted embedded systems have to cope with vibrations, bumps, and different road conditions during vehicle driving.

[0004] Based on the above situation, in order to ensure the stable and efficient operation of the Internet of Things and edge computing systems, it is crucial to accurately predict and evaluate the comprehensive stability of end-side nodes. Only by accurately grasping the operating status and stability trend of end-side devices can measures be taken in advance to avoid service interruptions or data errors caused by equipment failures or performance degradation. However, most traditional stability assessment methods are established based on fixed models and static environmental assumptions, and they tend to be ineffective when faced with the high complexity and dynamic changes of the end-side environment. For example, traditional methods may not be able to adapt to the dynamic adjustment of hardware configurations in a timely manner, nor can they effectively handle the impact of various unexpected factors in the operating environment on device stability. Therefore, there is an urgent need for a new node stability assessment method and system that can adapt to the complex end-side environment to meet the needs of the rapid development of the current Internet of Things and edge computing. Summary of the Invention

[0005] In view of the requirements and deficiencies of the current technological development, the present invention provides a method and system for evaluating the node stability of a terminal computing power network, aiming to predict and evaluate the comprehensive stability of nodes by using resource metrics (such as power, network bandwidth, memory, and CPU usage rate, etc.) collected by end-side devices. The stability score is calculated based on the predicted values, providing a data basis for the terminal computing power network system to manage end-side nodes and schedule applications.

[0006] In the first aspect, the present invention provides a method for evaluating the node stability of a terminal computing power network. The technical solutions adopted to solve the above technical problems are as follows:

[0007] A method for evaluating the node stability of a terminal computing power network includes the following steps:

[0008] S1. Collect resource metric data of the node to ensure that the number of data points for each resource metric is the same and the time is synchronized;

[0009] S2. Standardize the resource metric data;

[0010] S3. Use the PCA method to extract features from the standardized data;

[0011] S4. Construct a GRU model, input the selected feature data into the GRU model, and the GRU model outputs the predicted values of relevant dimension metrics;

[0012] S5. Calculate the comprehensive stability score of the node based on the predicted values of relevant dimension metrics.

[0013] Optionally, when performing step S2, the Z-score method is used to standardize the resource metric data so that the mean of each feature is 0 and the variance is 1.

[0014] Optionally, step S3 specifically includes:

[0015] First, calculate the covariance matrix, its eigenvalues, and eigenvectors. Then, select the eigenvectors corresponding to the top k largest eigenvalues, project the data from the original high-dimensional space to the low-dimensional principal component space, thereby extracting the principal components, reducing the data dimension, and retaining the key information.

[0016] Optionally, step S4 specifically includes:

[0017] S4.1. Construct a GRU model, which has the ability to handle long-term and short-term dependencies in time series data;

[0018] S4.2. Use the pre-prepared training data to train the GRU model so that the GRU model can output the predicted values of relevant dimensional indicators at a future moment based on the input current moment and previous feature data. During the training process of the GRU model, continuously adjust the weights and biases of the GRU model through the backpropagation algorithm so that the GRU model can learn the time series features and patterns in the data;

[0019] S4.3. Evaluate the trained GRU model and output the GRU model that meets the evaluation indicators;

[0020] S4.4. Use the feature data extracted by the PCA method as the input, and input it into the GRU model that meets the evaluation indicators according to the set time step. The GRU model outputs the predicted values of relevant dimensional indicators at a future moment based on the input current moment and previous feature data.

[0021] Optionally, step S5 specifically includes:

[0022] S5.1. Assign weights to each indicator based on the predicted values of relevant dimensional indicators, node characteristics, historical data, or test feedback to reflect their different impacts on node stability;

[0023] S5.2. For each indicator, calculate its stability score at the current moment;

[0024] S5.3. Calculate the comprehensive stability score of the node according to the standardized score and weight of each indicator.

[0025] In a second aspect, the present invention provides a node stability evaluation system for a terminal computing power network. The technical solution adopted to solve the above technical problems is as follows:

[0026] A node stability evaluation system for a terminal computing power network, which includes:

[0027] A data collection module for collecting resource metric data of nodes, ensuring that the number of data points for each resource metric is consistent and the time is synchronized;

[0028] A data standardization module for standardizing resource metric data;

[0029] A feature extraction module for extracting features from the standardized data using the PCA method;

[0030] A model processing module for building a GRU model, inputting the selected feature data into the GRU model, and the GRU model outputting the predicted values of relevant dimension indicators;

[0031] A stability calculation module for calculating the comprehensive stability score of a node based on the predicted values of relevant dimension indicators.

[0032] Optionally, the involved data standardization module standardizes the resource metric data using the Z-score method, making the mean of each feature 0 and the variance 1.

[0033] Optionally, the involved feature extraction module extracts features from the standardized data using the PCA method, and this process specifically includes:

[0034] First, calculate the covariance matrix and its eigenvalues and eigenvectors, then select the eigenvectors corresponding to the top k largest eigenvalues, project the data from the original high-dimensional space to the low-dimensional principal component space, thereby extracting the principal components, reducing the data dimension, and retaining key information.

[0035] Optionally, the involved model processing module specifically includes:

[0036] A model construction unit for building a GRU model, which has the ability to handle long-term and short-term dependencies in time series data;

[0037] A model training unit for training the GRU model using pre-prepared training data, enabling the GRU model to output the predicted values of relevant dimension indicators at a future moment based on the current moment and previous feature data input; during the training process of the GRU model, continuously adjust the weights and biases of the GRU model through the backpropagation algorithm, enabling the GRU model to learn the time series features and patterns in the data;

[0038] A model evaluation unit for evaluating the trained GRU model and outputting a GRU model that meets the evaluation indicators;

[0039] A GRU model for receiving the feature data extracted by the PCA method at a set time step and outputting the predicted values of relevant dimension indicators at a future moment based on the current moment and previous feature data received.

[0040] Optionally, the involved stability calculation module specifically includes:

[0041] A weight setting unit, configured to assign weights to each indicator according to the predicted values of relevant dimension indicators, node characteristics, historical data, or test feedback, so as to reflect their different impacts on node stability;

[0042] An indicator calculation unit, configured to calculate the stability score of each indicator at the current moment;

[0043] A stability calculation unit, configured to calculate the comprehensive stability score of the node according to the standardized scores and weights of each indicator.

[0044] A method and system for evaluating the node stability of a terminal computing power network according to the present invention have the beneficial effects compared with the prior art as follows:

[0045] Based on the original resource indicators of the edge node, the present invention adopts a prediction method combining the PCA method and the GRU model to predict the future indicator data of the node, and further uniformly processes the predicted values, and calculates the comprehensive stability score of the node based on the predicted values of multi-dimensional indicators, realizing the comprehensive measurement of the edge device with strong dynamics. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Attached Figure 1 is the flowchart of the method according to Embodiment 1 of the present invention;

[0047] Attached Figure 2 is the block diagram of module connection according to Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] In order to make the technical solutions, technical problems solved, and technical effects of the present invention clearer and more understandable, the following combines specific embodiments to clearly and completely describe the technical solutions of the present invention.

[0049] Embodiment 1:

[0050] Combined with Attached Figure 1 , this embodiment proposes a method for evaluating the node stability of a terminal computing power network, which includes the following steps:

[0051] S1. Collect the resource indicator data of the node, and ensure that the number of data points of each resource indicator is the same and the time is synchronized.

[0052] When specifically executing this step, use the monitoring system to continuously collect multi-dimensional resource indicator data of the edge node at different time points, and these indicators at least cover key resource information such as CPU load, memory load, network bandwidth, and available power.

[0053] During the collection process, through technical means such as timestamp marking, ensure that the number of data points for each resource metric is consistent and the time is synchronized, so that the data has consistency and comparability during subsequent analysis. For example, data collection for all metrics can be set to occur at fixed time intervals (such as every 1 minute), ensuring that each metric has corresponding data records on the same time series. The collected data needs to be preliminarily verified to ensure data integrity and accuracy, such as removing obviously incorrect or abnormal data points and performing reasonable interpolation or filling for missing values.

[0054] S2. Since the units and ranges of each resource metric vary greatly. For example, CPU load and memory load are expressed as percentages, network bandwidth is in units of bps, and power is usually a value between 0 and 100. Therefore, the Z-score method is used to standardize the resource metric data, making the mean of each feature 0 and the variance 1.

[0055] The formula involved in the standardization process is as follows:

[0056] Z = (X - μ) / σ,

[0057] In the formula, X represents the original metric data, μ represents the mean of the metric data, and σ represents the standard deviation of the metric data; through this standardization process, the mean of each feature is 0 and the variance is 1, eliminating the influence of the dimension between different metrics and facilitating subsequent unified analysis and processing.

[0058] Before performing the standardization calculation, it is necessary to preliminarily analyze the data distribution to ensure that the Z-score method is applicable to this set of data. If the data has a severe skewed distribution, other standardization methods or data transformation methods may need to be considered.

[0059] S3. The PCA method is used to extract features from the standardized data, specifically including:

[0060] First, calculate the covariance matrix, its eigenvalues, and eigenvectors, and then select the eigenvectors corresponding to the top k largest eigenvalues, project the data from the original high-dimensional space to the low-dimensional principal component space, thereby extracting the principal components, reducing the data dimension, and retaining the key information.

[0061] When selecting the principal components, the determination of the k value can refer to the cumulative variance contribution rate. Usually, the cumulative variance contribution rate is set to reach 80% - 95% to ensure that the extracted principal components can explain the variance of the original data to the greatest extent. At the same time, it is necessary to verify the rationality of the extracted principal components, such as observing the relationship between the principal components and the characteristics of the original data through visualization means to ensure that the principal components have practical significance.

[0062] S4. Build a GRU (Gated Recurrent Unit) model, input the selected feature data into the GRU model, and the GRU model outputs the predicted values of relevant dimensional indicators, specifically including:

[0063] S4.1. Build a GRU model, which has the ability to handle long-term and short-term dependencies in time series data. It should be noted that the determination of the model structure needs to comprehensively consider factors such as the complexity of the data and computing resources. The performance of GRU models with different numbers of layers and different numbers of units can be compared through experiments to select the optimal structure.

[0064] S4.2. Use the pre-prepared training data to train the GRU model so that the GRU model can output the predicted values of relevant dimensional indicators at a future moment based on the input current moment and previous feature data. During the training process of the GRU model, the weights and biases of the GRU model are continuously adjusted through the backpropagation algorithm so that the GRU model can learn the time series features and patterns in the data. Reasonable training parameters, such as learning rate, number of training epochs, batch size, etc., need to be set during the training process. A strategy of dynamically adjusting the learning rate can be adopted to improve the training efficiency and the model convergence speed. At the same time, to prevent overfitting, regularization methods such as L1 or L2 regularization, and Dropout technology can be used.

[0065] S4.3. Evaluate the trained GRU model and output the GRU model that meets the evaluation indicators. The evaluation indicators can be selected as mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), etc. Determine a reasonable evaluation indicator threshold according to the actual application scenario. Only when the evaluation indicators of the model on the validation set meet the threshold requirements is the model considered to be successfully trained.

[0066] S4.4. Take the feature data extracted by the PCA method as the input, and input it into the GRU model that meets the evaluation indicators according to the set time step (for example, using the data of 10 consecutive time points as a time step). The GRU model outputs the predicted values of relevant dimensional indicators at a future moment based on the input current moment and previous feature data. Before inputting the data, the data needs to be preprocessed, such as normalizing it to the range required by the GRU model input, and at the same time, the data is processed in batches to adapt to the computing resources and memory limitations of the model.

[0067] S5. Calculate the comprehensive stability score of the node based on the predicted values of relevant dimensional indicators, specifically including:

[0068] S5.1. Assign weights to each indicator based on the predicted values of relevant dimensional indicators, node characteristics, historical data, or test feedback to reflect their different impacts on node stability. The determination of weights can adopt scientific methods such as the Analytic Hierarchy Process (AHP). By constructing a judgment matrix, calculate the relative importance weights of each indicator. At the same time, regularly update and optimize the weights according to new data and actual situations to ensure the timeliness and accuracy of the weights.

[0069] S5.2. Calculate the stability score of each indicator at the current moment.

[0070] S5.3. Calculate the comprehensive stability score of the node according to the standardized score and weight of each indicator.

[0071] For step S5, assume the following weights are set: the weight of CPU load is w 1 , the weight of memory load is w 2 , the weight of network bandwidth is w 3 , the weight of available power is w 4 , and the sum of the weights is 1:

[0072] w 1 + w 2 + w 3 + w 4 = 1.

[0073] ① For CPU load and memory load, generally, the higher the value, the higher the load and the more unstable the device may be. Therefore, the stability score can be calculated as:

[0074]

[0075] where CPU_load represents the CPU load, Mem_load represents the memory load, and the units of both variables are percentages;

[0076] ② For network bandwidth, if the bandwidth usage is close to the upper limit, it indicates a network bottleneck and poor device stability. The score is calculated as:

[0077]

[0078] where Bandwidth_usage represents the current bandwidth usage and Bandwidth represents the maximum bandwidth;

[0079] ③ For the device power, the lower the available power may indicate poor device stability. Therefore, the stability score can be calculated as:

[0080]

[0081] Among them, Charge_remaining represents the available power, and Charge_capacity represents the full charge capacity of the battery.

[0082] According to the standardized scores and weights of each indicator, calculate the comprehensive stability score of the node:

[0083] S node = w 1 ·S cpu + w 2 ·S memory + w 3 ·S network + w 4 ·S battery ,

[0084] Among them, w 1 represents the CPU load weight, w 2 represents the memory load weight, w 3 represents the network bandwidth weight, w 4 represents the available power weight, S CPU represents the stability score of the CPU load, S memory represents the stability score of the memory load, S network represents the stability score of the network bandwidth, S battery represents the stability score of the device power.

[0085] After calculating the comprehensive stability score, the node stability can be divided into different levels, such as high, medium, and low, according to the pre-set stability level threshold, so as to intuitively evaluate the stability status of the node.

[0086] Example 2:

[0087] Combined with the attached Figure 2 , this embodiment proposes a node stability evaluation system for the terminal computing power network, which includes:

[0088] A data collection module for collecting resource indicator data of the node to ensure that the number of data points of each resource indicator is the same and the time is synchronized;

[0089] A data standardization module for standardizing the resource indicator data using the Z-score method to make the mean of each feature 0 and the variance 1;

[0090] A feature extraction module for extracting features from the standardized data using the PCA method. This process specifically includes: first calculating the covariance matrix and its eigenvalues and eigenvectors, then selecting the eigenvectors corresponding to the first k largest eigenvalues, projecting the data from the original high-dimensional space to the low-dimensional principal component space, thereby extracting the principal components, reducing the data dimension, and retaining the key information;

[0091] A model processing module for constructing a GRU model, inputting the selected feature data into the GRU model, and the GRU model outputs the predicted values of relevant dimensional indicators.

[0092] A stability calculation module for calculating the comprehensive stability score of a node based on the predicted values of relevant dimensional indicators.

[0093] In this embodiment, the involved model processing module specifically includes:

[0094] A model construction unit for constructing a GRU model, which has the ability to handle long-term and short-term dependencies in time series data.

[0095] A model training unit for training the GRU model using pre-prepared training data, enabling the GRU model to output the predicted values of relevant dimensional indicators at a future moment based on the input current moment and previous feature data; during the training process of the GRU model, continuously adjust the weights and biases of the GRU model through the backpropagation algorithm, so that the GRU model can learn the time series features and patterns in the data.

[0096] A model evaluation unit for evaluating the trained GRU model and outputting a GRU model that meets the evaluation indicators.

[0097] The GRU model is used to receive the feature data extracted by the PCA method at a set time step, and based on the received current moment and previous feature data, output the predicted values of relevant dimensional indicators at a future moment.

[0098] In this embodiment, the involved stability calculation module specifically includes:

[0099] A weight setting unit for assigning weights to each indicator according to the predicted values of relevant dimensional indicators, node characteristics, historical data or test feedback, to reflect their different impacts on node stability.

[0100] An indicator calculation unit for calculating the stability score of each indicator at the current moment.

[0101] A stability calculation unit for calculating the comprehensive stability score of the node according to the standardized scores and weights of each indicator.

[0102] In summary, by using the node stability evaluation method and system of a terminal computing power network of the present invention, the comprehensive stability of nodes can be predicted and evaluated by using the resource indicators collected by the edge devices, providing a data basis for the management of edge nodes and the scheduling of applications in the terminal computing network system.

[0103] The above specific application examples have elaborated in detail the principle and implementation manner of the present invention. These examples are only used to help understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made by those skilled in the art of this technical field without departing from the principle of the present invention shall fall within the scope of patent protection of the present invention.

Claims

1. A method for evaluating node stability of a terminal computing network, characterized in that: The steps include: S1. Collect the resource indicator data of the node to ensure that the number of data points of each resource indicator is consistent and time synchronized; S2. Standardize resource indicator data; S3, using PCA method to extract features from the standardized data; S4, build a GRU model, input the selected feature data into the GRU model, and the GRU model outputs the predicted value of the relevant dimension index; S5. Calculate the comprehensive stability score of the node based on the predicted values ​​of the relevant dimension indicators.

2. A method for evaluating node stability of a terminal computing network according to claim 1, characterized in that: Execute step S2 and use the Z-score method to standardize the resource indicator data so that the mean of each feature is 0 and the variance is 1.

3. A method for evaluating node stability of a terminal computing network according to claim 1, characterized in that: The step S3 specifically includes: First, the covariance matrix and its eigenvalues ​​and eigenvectors are calculated, and then the eigenvectors corresponding to the first k largest eigenvalues ​​are selected to project the data from the original high-dimensional space to the low-dimensional principal component space, thereby extracting the principal components, reducing the data dimension, and retaining key information.

4. A method for evaluating node stability of a terminal computing network according to claim 1, characterized in that: The step S4 specifically includes: S4.

1. Construct a GRU model that has the ability to handle long-term and short-term dependencies in time series data; S4.

2. Use the pre-prepared training data to train the GRU model, so that the GRU model can output the predicted value of the relevant dimension index at a certain moment in the future based on the input feature data at the current moment and before; during the GRU model training process, the weights and biases of the GRU model are continuously adjusted through the back propagation algorithm, so that the GRU model can learn the time series characteristics and laws in the data; S4.3, evaluate the trained GRU model and output the GRU model that meets the evaluation indicators; S4.

4. The feature data extracted by the PCA method is used as input and input into the GRU model that meets the evaluation index according to the set time step. The GRU model outputs the predicted value of the relevant dimension index at a certain moment in the future based on the input current moment and previous feature data.

5. A method for evaluating node stability of a terminal computing network according to claim 1, characterized in that: The step S5 specifically includes: S5.

1. Assign weights to each indicator based on the predicted values ​​of the relevant dimensional indicators, node characteristics, historical data or test feedback to reflect their different impacts on node stability; S5.

2. For each indicator, calculate its stability score at the current moment; S5.

3. Calculate the comprehensive stability score of the node based on the standardized score and weight of each indicator.

6. A node stability evaluation system for a terminal computing network, characterized in that: It includes: The data collection module is used to collect the resource indicator data of the node to ensure that the number of data points for each resource indicator is consistent and time synchronized; Data standardization module, used to standardize resource indicator data; A feature extraction module is used to extract features from the standardized data using the PCA method; The model processing module is used to build a GRU model, input the selected feature data into the GRU model, and the GRU model outputs the predicted values ​​of the relevant dimension indicators; The stability calculation module is used to calculate the comprehensive stability score of the node based on the predicted values ​​of relevant dimensional indicators.

7. A node stability evaluation system for a terminal computing network according to claim 6, characterized in that: The data standardization module uses the Z-score method to standardize the resource indicator data so that the mean of each feature is 0 and the variance is 1.

8. A node stability evaluation system for a terminal computing network according to claim 6, characterized in that: The feature extraction module uses the PCA method to extract features from the standardized data. This process specifically includes: First, the covariance matrix and its eigenvalues ​​and eigenvectors are calculated, and then the eigenvectors corresponding to the first k largest eigenvalues ​​are selected to project the data from the original high-dimensional space to the low-dimensional principal component space, thereby extracting the principal components, reducing the data dimension, and retaining key information.

9. A node stability evaluation system for a terminal computing network according to claim 6, characterized in that: The model processing module specifically includes: The model building unit is used to build a GRU model that has the ability to handle long-term and short-term dependencies in time series data; The model training unit is used to train the GRU model using the pre-prepared training data, so that the GRU model can output the predicted value of the relevant dimension index at a certain moment in the future based on the input feature data at the current moment and before. During the GRU model training process, the weight and bias of the GRU model are continuously adjusted through the back propagation algorithm, so that the GRU model can learn the time series characteristics and laws in the data. The model evaluation unit is used to evaluate the trained GRU model and output a GRU model that meets the evaluation indicators; The GRU model is used to receive feature data extracted by the PCA method according to the set time step, and output the predicted value of the relevant dimension index at a certain moment in the future based on the received feature data at the current moment and before.

10. A node stability evaluation system for a terminal computing network according to claim 6, characterized in that: The stability calculation module specifically includes: The weight setting unit is used to assign weights to each indicator based on the predicted values ​​of the relevant dimensional indicators, node characteristics, historical data or test feedback to reflect their different impacts on node stability; An indicator calculation unit, used to calculate the stability score of each indicator at the current moment; The stability calculation unit is used to calculate the comprehensive stability score of the node based on the standardized score and weight of each indicator.

Images