Automatic resource matching method and system based on credibility dynamic grading

Abstract

The invention provides an automatic resource matching method and system based on trust degree dynamic grading, and relates to the technical field of data analysis and decision making, and the method comprises the steps: building a feature vector matrix through obtaining user historical behavior data, generating a trust degree grading model through the training of a deep learning model, and carrying out the trust degree grading of a target user; determining an accessible system resource pool based on the credibility level, collecting real-time service demand information, and generating a target resource configuration scheme through a resource demand prediction model; and the resource use efficiency is monitored in real time, the credibility level is dynamically updated, and resource redistribution is carried out. Accurate matching and dynamic adjustment of system resources are achieved, the resource utilization efficiency is improved, resource waste is reduced, and the safety and stability of system operation are guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of data analysis and decision-making technology, and mainly to a resource automatic matching method and system based on dynamic trust grading. Background Art

[0002] With the rapid development of cloud computing technology, enterprise users' demands for computing resources, storage resources and network bandwidth have become diversified and dynamic. How to establish an accurate resource allocation mechanism based on the user's credit level and business demand characteristics to achieve efficient use of system resources has become an important research direction in the current cloud computing field. At the same time, the user's resource usage behavior and credit status often change dynamically over time, which requires the resource allocation system to have real-time monitoring and dynamic adjustment capabilities.

[0003] Existing resource allocation systems generally adopt static user classification standards, ignoring the analytical value of user historical behavior data, and are unable to establish an accurate user trust evaluation model, resulting in a lack of differentiation and pertinence in resource allocation; and traditional resource allocation solutions mainly rely on simple rule matching or manual experience, lack the ability to intelligently predict users' real-time business needs, and are difficult to achieve accurate matching of resources, resulting in low resource utilization efficiency; there is a lack of effective monitoring and feedback mechanism after resource allocation, and it is impossible to dynamically adjust the trust level and resource allocation strategy according to the user's actual resource utilization efficiency, affecting the fairness and efficiency of system resource utilization. Therefore, it is urgent to construct an intelligent resource allocation method and system based on user trust to solve the technical problems of inaccurate user trust evaluation, inaccurate resource matching and lack of dynamic adjustment capabilities in the existing technology. Summary of the invention

[0004] In order to solve the above-mentioned problems existing in the prior art, the present invention provides an automatic resource matching method and system based on dynamic trust grading, which analyzes user historical behavior through a deep learning model to achieve accurate assessment of trust; and combines real-time business demand prediction to establish a dynamic resource matching mechanism to achieve automatic adjustment of resource allocation.

[0005] The technical solution of the present invention is as follows:

[0006] The present invention provides a resource automatic matching method based on dynamic trust classification, comprising:

[0007] Acquire user historical behavior data, construct a user feature vector matrix based on the historical behavior data, construct a deep neural network model based on the user feature vector matrix, and generate a user trust scoring model through optimization training; apply the user trust scoring model to a target user to obtain an initial trust score of the target user, and classify the target user into a preset trust level according to the initial trust score;

[0008] Based on the trust level of the target user, determine the system resource pool accessible to the target user, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; collect the real-time business demand information of the target user, wherein the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; construct a resource demand prediction model, input the real-time business demand information into the resource demand prediction model, and output a target resource configuration plan;

[0009] Allocate system resources to the target user from the system resource pool according to the target resource configuration plan; monitor the target user's use of the allocated system resources in real time to obtain resource usage efficiency data; dynamically update the target user's trust level based on the resource usage efficiency data and in combination with preset trust dynamic adjustment rules; and automatically re-match and allocate system resources when the target user's trust level changes.

[0010] As a preferred implementation of the present invention, user historical behavior data is obtained, a user feature vector matrix is ​​constructed based on the historical behavior data, a deep neural network model is constructed based on the user feature vector matrix, and a user trust rating model is generated through optimization training as follows:

[0011] The historical behavior data includes user transaction records, user contract performance, user credit score and user resource usage records, wherein: the user transaction records include transaction time, transaction amount, transaction type and transaction status, and the user transaction records are constructed into a transaction behavior sequence; the user contract performance includes contract performance, payment timeliness and service completion; the user credit score includes credit report score, third-party credit score, platform credit record score and complaint handling score; the user resource usage record includes resource call frequency record, resource occupation time record, resource usage efficiency record and resource over-limit usage record;

[0012] Constructing a user feature vector matrix based on the historical behavior data includes: setting a performance weight matrix for the user's performance and quantifying it to obtain a performance index; using a hierarchical analysis method to determine the score weight for the user's credit score to obtain a comprehensive credit score; standardizing the transaction behavior sequence, the performance index, the comprehensive credit score and the user's resource usage record to obtain standardized data; using a sliding time window method to extract time series features from the standardized data to obtain time series feature data; correlating the transaction amount data in the time series feature data with the performance index to obtain transaction performance features, correlating the resource utilization efficiency record with the comprehensive credit score to obtain credit resource features; combining the transaction performance features, the credit resource features and the time series feature data to construct the user feature vector matrix;

[0013] A deep neural network model is constructed based on the user feature vector matrix and a user trust scoring model is generated through optimization training, including: constructing a deep neural network model, the input layer dimension of the deep neural network model is the same as the dimension of the user feature vector matrix, and the deep neural network model includes three hidden layers using ReLU activation function and one output layer using Sigmoid function; constructing a composite loss function including a mean square error loss term and a ranking loss term, and introducing a regularization term into the composite loss function; using a batch gradient descent method combined with an Adam optimizer to train the deep neural network model, and using a dynamic learning rate strategy during the training process, wherein the dynamic learning rate strategy includes using a first learning rate in the initial stage of training and a second learning rate in the later stage of training, so as to obtain a basic trust scoring model; performing weighted fusion on the output result of the basic trust scoring model and the output result of the preset rule model to generate the user trust scoring model.

[0014] As a preferred implementation of the present invention, applying the user trust rating model to a target user to obtain an initial trust score of the target user, and classifying the target user into a preset trust level according to the initial trust score includes:

[0015] Applying the user trust rating model to the target user specifically includes: collecting the target user's behavior data within ninety days before the current time point, the behavior data including transaction data, performance data, credit data and resource usage data; preprocessing the behavior data, including standard deviation normalization of numerical features, one-hot encoding of categorical features, time window feature extraction of time series data, and model mean filling of missing data items; using an exponential decay function to calculate the time weight coefficient of the preprocessed behavior data, the decay coefficient in the exponential decay function is inversely proportional to the time interval; weighting the preprocessed behavior data based on the time weight coefficient to obtain weighted behavior data;

[0016] Constructing a target user feature vector based on the weighted behavior data specifically includes: performing feature cross construction on the weighted behavior data to obtain cross features; normalizing the cross features to obtain the target user feature vector; inputting the target user feature vector into the user trust scoring model, and obtaining an initial trust score through forward propagation of the user trust scoring model; weighted fusion of the initial trust score and the evaluation result of the preset rule model to obtain the initial trust score of the target user;

[0017] The target user is divided into a preset trust level according to the initial trust score, specifically including: performing cluster analysis on the historical user trust scores based on the K-means clustering method to obtain a dynamic grading threshold; combining the dynamic grading threshold with a time decay factor to construct a dynamic grading standard; and grading the initial trust score of the target user based on the dynamic grading standard to obtain the initial trust level of the target user.

[0018] As a preferred implementation mode of the present invention, based on the trust level of the target user, a system resource pool accessible to the target user is determined, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; real-time business demand information of the target user is collected, wherein the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; a resource demand prediction model is constructed, and the real-time business demand information is input into the resource demand prediction model, and the output target resource configuration scheme includes:

[0019] Determine the system resource pool accessible to the target user according to the trust level of the target user, set the usage quota of the computing resources based on the number of CPU cores and GPU memory capacity, set the usage quota of the storage resources based on the storage capacity and IOPS performance index, and set the usage quota of the bandwidth resources based on the peak bandwidth and the number of concurrent connections;

[0020] Collecting the real-time business demand information of the target user, the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; using a deep learning model to classify and identify the business type, evaluating the business scale based on data processing volume, number of concurrent requests and storage space requirements, and determining the business priority based on business importance and service level agreement;

[0021] Constructing a resource demand prediction model, the resource demand prediction model includes: a long short-term memory network module for capturing the temporal dependency of business demand, an attention mechanism module for focusing on resource usage features at key time points, and a fully connected layer module for feature fusion; the resource demand prediction model uses a sliding time window method to extract the periodic features of business load, mines the correlation between different types of resource demands, identifies resource usage patterns in specific business scenarios, and inputs the features extracted by the long short-term memory network module and the attention mechanism module into the fully connected layer module of the resource demand prediction model for feature fusion;

[0022] The real-time business demand information is input into the resource demand prediction model, and an initial resource allocation plan is obtained through the collaborative processing of the long short-term memory network module and the attention mechanism module; a multi-objective optimization model is constructed, and the initial resource allocation plan is input into the multi-objective optimization model, and optimization is performed based on total resource constraints, priority constraints, performance constraints and cost constraints to obtain an optimized resource allocation plan, namely, a target resource allocation plan.

[0023] As a preferred implementation mode of the present invention, system resources are allocated to target users from a system resource pool according to a target resource configuration scheme, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; CPU share allocation and memory usage restrictions are performed on the computing resources through a resource management control component, storage capacity quotas and input and output performance restrictions are performed on the storage resources through a service quality control component, and concurrent connection numbers and link establishment rate restrictions are performed on the bandwidth resources through a flow control component;

[0024] Establish a real-time monitoring and collection mechanism, which includes a differentiated sampling period setting module and a monitoring data collection module; set a sampling period for the multi-dimensional monitoring indicators through the differentiated sampling period setting module, and collect multi-dimensional monitoring indicator data of the target user using the system resources through the monitoring data collection module; wherein the multi-dimensional monitoring indicator data includes system-level indicator data and business-level indicator data, wherein the system-level indicator data includes CPU usage, memory usage, disk input and output throughput, and network bandwidth utilization, and the business-level indicator data includes request response time, business processing success rate, and concurrent processing quantity;

[0025] Construct a real-time data processing pipeline, which includes a data compression module and an anomaly detection module; incrementally compress the multi-dimensional monitoring indicator data through the data compression module, identify data collection anomalies through the anomaly detection module and correct them, so as to obtain compressed and corrected monitoring data; pre-process the compressed and corrected monitoring data, use the Z-score method to filter outliers, use the sliding average method to smooth, and use the time series prediction method to fill missing values, so as to obtain pre-processed monitoring data;

[0026] Extract monitoring features from the preprocessed monitoring data, including: extracting time series features to obtain trend features and period features, calculating statistical features to obtain mean features and variance features, analyzing correlation features to obtain indicator association features; using hierarchical analysis method to determine the weights of the monitoring features, and weighted fusion of the monitoring features to obtain resource utilization efficiency data.

[0027] As a preferred implementation mode of the present invention, according to the resource utilization efficiency data, combined with the preset trust dynamic adjustment rule, the trust level of the target user is dynamically updated; when the trust level of the target user changes, the automatic matching and allocation of system resources is re-performed, including:

[0028] Obtaining a preset trust dynamic adjustment rule, wherein the preset trust dynamic adjustment rule includes a trust level division rule and a level adjustment trigger rule; the trust level division rule defines a trust level boundary value and a transition buffer zone, and the level adjustment trigger rule defines a level adjustment condition and a cooling period;

[0029] Dynamically updating the trust level of the target user according to the resource utilization efficiency data and the preset trust dynamic adjustment rule, including: when the resource utilization efficiency data is above the transition buffer zone in a continuous preset evaluation period, raising the trust level; when the resource utilization efficiency data is below the transition buffer zone or a violation occurs, lowering the trust level;

[0030] When the trust level of the target user changes, the resource quota is recalculated based on the new trust level to generate a target resource allocation scheme; the target resource allocation scheme includes a new computing resource quota, a storage resource quota and a network resource quota;

[0031] Execute automatic matching and allocation of system resources, including: evaluating the existing resource occupancy, locking the available resource pool, and generating a resource migration plan; the resource migration plan adopts a step-by-step migration strategy to complete resource reallocation while ensuring service continuity; monitor the service status in real time during the resource migration process, and automatically initiate a rollback mechanism when an abnormality occurs.

[0032] The present invention also provides a resource automatic matching system based on dynamic trust grading, the system comprising a user behavior and feature analysis module, a trust model and rating module, a resource pool management and allocation module, and a resource monitoring and dynamic adjustment module, wherein:

[0033] The user behavior and feature analysis module is used to obtain user historical behavior data, which includes user transaction records, user contract fulfillment, user credit score and user resource usage records;

[0034] The trust model and rating module constructs a user feature vector matrix based on the historical behavior data, and trains the user feature vector matrix through a deep learning model to generate a user trust rating model; applies the user trust rating model to a target user to obtain an initial trust score for the target user, and classifies the target user into a preset trust level according to the initial trust score;

[0035] The resource pool management and allocation module determines the system resource pool accessible to the target user based on the trust level of the target user, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; collects the real-time business demand information of the target user, wherein the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; constructs a resource demand prediction model, inputs the real-time business demand information into the resource demand prediction model, and outputs a target resource allocation plan; allocates system resources to the target user from the system resource pool according to the target resource allocation plan;

[0036] The resource monitoring and dynamic adjustment module monitors the target user's use of the allocated system resources in real time and obtains resource usage efficiency data; based on the resource usage efficiency data and in combination with preset trust dynamic adjustment rules, the trust level of the target user is dynamically updated; when the trust level of the target user changes, the system resources are automatically matched and allocated again.

[0037] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for automatically matching resources based on dynamic grading of trust as described in any one of the embodiments is implemented.

[0038] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the automatic resource matching method based on dynamic trust grading as described in any one of the embodiments.

[0039] The beneficial effects of this application are as follows:

[0040] 1. Improve resource utilization and system efficiency: Through a trust-based grading mechanism and automated resource matching, the system can accurately allocate resources based on the user's historical behavior and real-time needs. High-trust users can access a richer resource pool, while low-trust users will be restricted to a certain range, avoiding waste and abuse of resources. Dynamic adjustment and automated allocation of resources also reduce the cost of manual intervention and improve the overall operating efficiency of the system.

[0041] 2. Enhance system security and stability: The trust rating mechanism can effectively identify potential malicious users or high-risk users and restrict their access to system resources, thereby reducing the risk of system attacks or abuse. At the same time, dynamically adjusting trust and resource allocation based on user resource usage efficiency data can better ensure the stability and availability of system resources and prevent resources from being over-occupied or exhausted.

[0042] 3. Optimize user experience and personalized services: The system provides differentiated resource services based on the user's trust level and real-time needs to meet the personalized needs of different users. High-trust users can enjoy better and richer resources, thereby improving user experience and satisfaction. At the same time, the automated matching and allocation of resources also simplifies the process of users obtaining resources, making it more convenient and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram of a process of an automatic resource matching method based on dynamic trust classification according to an embodiment of the present invention;

[0044] Figure 2 The diagram is a structural diagram of an automatic resource matching system based on dynamic trust classification according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0047] Embodiment 1:

[0048] Figure 1 FIG. 1 is a flow chart of a resource automatic matching method based on dynamic trust classification according to Embodiment 1 of the present invention. Figure 1 As shown, the method includes:

[0049] S1. Obtain user historical behavior data, construct a user feature vector matrix based on the historical behavior data, construct a deep neural network model based on the user feature vector matrix, and generate a user trust scoring model through optimization training; apply the user trust scoring model to the target user to obtain the initial trust score of the target user, and classify the target user into a preset trust level according to the initial trust score;

[0050] S11. Obtain user historical behavior data:

[0051] The user's historical behavior data includes the user's transaction records, contract performance, credit score and resource usage records. The user's transaction records include transaction time, transaction amount, transaction type and transaction status, and these records will be constructed into a transaction behavior sequence. The user's contract performance includes contract performance, payment timeliness and service completion. The user's credit score includes the credit report score, third-party credit score, platform credit record score and complaint handling score. The user's resource usage record includes resource call frequency record, resource occupation time record, resource usage efficiency record and resource over-limit usage record;

[0052] For example, assuming that on an e-commerce platform, the user's transaction record can be: on January 1, 2024, purchased books for 100 yuan, the transaction was successful; on January 5, 2024, purchased electronic products for 500 yuan, the transaction was successful; on January 10, 2024, purchased clothing for 200 yuan, the transaction was cancelled. The user's performance can be: good contract performance, timely payment, and 100% service completion. The user's credit score can be: credit report score 85 points, third-party credit score 90 points, platform credit record score 92 points, and complaint handling score 95 points. The user's resource usage record can be: calling the API interface 100 times a day, the average occupancy time is 1 second, the resource utilization efficiency is 90%, and there is no record of excessive resource usage.

[0053] S12. Constructing a user feature vector matrix based on the historical behavior data:

[0054] S121. Set a performance weight matrix for the user's performance to quantify and obtain a performance index. For example, the weight of contract performance is 0.4, the weight of payment timeliness is 0.3, and the weight of service completion is 0.3. Based on the above example, the user's performance index is expressed as:

[0055] 0.4*1+0.3*1+0.3*1=1;

[0056] The user's credit score is determined by the hierarchical analysis method to obtain a comprehensive credit score. For example, the credit report score weight is 0.25, the third-party credit score weight is 0.25, the platform credit record score weight is 0.3, and the complaint handling score weight is 0.2. According to the above example, the user's comprehensive credit score is 0.25*85+0.25*90+0.3*92+0.2*95=89.95;

[0057] S122. Standardize the transaction behavior sequence, performance index, comprehensive credit score and user resource usage records to obtain standardized data. For example, perform minimum-maximum standardization on the transaction amounts of all users, and perform Z-score standardization on the performance index, comprehensive credit score and resource usage records. Next, use the sliding time window method to extract time series features from the standardized data to obtain time series feature data. For example, set the time window to 7 days, and extract features such as the total transaction amount, number of transactions, and average transaction amount in the past 7 days. Then perform correlation analysis on the standardized transaction amount data and the performance index to obtain transaction performance features. For example, calculate the Pearson correlation coefficient between the transaction amount and the performance index. And perform correlation analysis on the resource utilization efficiency records and the comprehensive credit score to obtain credit resource features. For example, calculate the Spearman correlation coefficient between resource utilization efficiency and the comprehensive credit score. Finally, combine the transaction performance features, credit resource features and time series feature data to construct a user feature vector matrix;

[0058] S13. Construct a deep neural network model based on the user feature vector matrix and generate a user trust rating model through optimization training:

[0059] S131.Building a deep neural network model:

[0060] The input layer dimension of the deep neural network model is the same as the dimension of the user feature vector matrix, and includes three hidden layers using the ReLU activation function and one output layer using the Sigmoid function; a composite loss function including a mean square error loss term and a ranking loss term is constructed, and a regularization term is introduced into the composite loss function;

[0061] S132. The deep neural network model is trained by using a batch gradient descent method combined with an Adam optimizer. A dynamic learning rate strategy is used during the training process. The dynamic learning rate strategy includes using a first learning rate in the early stage of training and a second learning rate that is smaller than the first learning rate in the late stage of training. The first learning rate and the second learning rate are both preset values. The first learning rate is set to a value between 0.01 and 0.1, and the specific value depends on the task complexity and the model scale. The second learning rate is set to 1 / 10 of the first learning rate. A basic trust scoring model is obtained. The output result of the basic trust scoring model is weightedly fused with the output result of the preset rule model to generate the user trust scoring model. The preset rules are customized according to user needs. For example, the preset rules may be: if the user has malicious order-brushing behavior, the trust score is directly reduced by 50%;

[0062] In steps S11-S13 of this embodiment, by combining the multi-dimensional historical behavior data of users and the deep learning model, the user's trustworthiness is evaluated more comprehensively and accurately, high-risk users are effectively identified, and the accuracy of trustworthiness evaluation is improved; a composite loss function and a dynamic learning rate strategy are used to improve the generalization ability and robustness of the model, so that it can still maintain a high accuracy rate when facing new data; and the automated user trust rating model can help the platform quickly identify and manage risks, improve risk control efficiency, reduce manual review costs, and improve risk control efficiency;

[0063] S14. Apply the user trust rating model to the target user to obtain an initial trust score for the target user, and classify the target user into a preset trust level according to the initial trust score:

[0064] S141. Apply the user trust rating model to target users:

[0065] S1411. Collect the behavior data of the target user within 90 days before the current time point, the behavior data includes transaction data, performance data, credit data and resource usage data. Furthermore, the performance data includes, for example, whether the performance is on time, the number of performances, and historical defaults; the credit data includes, for example, the credit score, the length of credit history, and negative credit records; the resource usage data includes, for example, the platform resource occupancy, resource usage time, and resource usage efficiency; an example of collecting the behavior data of the target user is to obtain the following information: User A has made 5 transactions in the past 90 days, with a total amount of 1,000 yuan, and has performed on time every time, with a credit score of 700 points, and a platform resource utilization rate of 20%;

[0066] S1412. Preprocess the behavior data, including standardizing the standard deviation of numerical features (such as transaction amount, credit score), converting the data into a distribution with a mean of zero and a standard deviation of one; performing one-hot encoding on categorical features (such as transaction objects), converting categorical variables into numerical variables; extracting time window features from time series data (such as transaction time), such as counting the number of transactions and transaction amounts in the past 7 days, 30 days and 90 days; filling missing data items with model mean, filling with the average value of the feature in the existing data, for example, if the resource usage data of user A is missing, it is filled with the average value of the resource usage data of all users;

[0067] S1413. Use an exponential decay function to calculate the time weight coefficient of the preprocessed behavior data. The more recent the data, the higher the weight; the older the data, the lower the weight; the decay coefficient in the exponential decay function is inversely proportional to the time interval, which means that the longer the time interval, the smaller the decay coefficient and the faster the weight decays. For example, the weight coefficient of user A's transaction data 90 days ago is 0.1, the weight coefficient of transaction data 30 days ago is 0.5, and the weight coefficient of transaction data 7 days ago is 0.9;

[0068] The pre-processed behavior data is weighted based on the time weight coefficient to obtain weighted behavior data. Specifically, each pre-processed behavior data is multiplied by its corresponding time weight coefficient to obtain weighted behavior data. For example, the transaction amount of user A 7 days ago was 200 yuan, and its time weight coefficient was 0.9, then the weighted transaction amount was 180 yuan;

[0069] S142. Constructing a target user feature vector based on the weighted behavior data:

[0070] Perform feature cross construction on the weighted behavior data to obtain cross features, for example, multiply the transaction amount and the transaction frequency to obtain the total transaction amount; perform normalization on the cross features, scale the feature values ​​to between 0 and 1, and obtain the target user feature vector; for example, the feature vector of user A is [0.8, 0.5, 0.7, 0.2], which respectively represent the weighted total transaction amount, the weighted number of fulfillment times, the weighted credit score, and the weighted resource utilization rate;

[0071] S143. Input the target user feature vector into the user trust scoring model, and obtain an initial trust score through forward propagation of the user trust scoring model. The user trust scoring model is constructed based on any machine learning model such as logistic regression, support vector machine, neural network, etc., which will not be described in detail here;

[0072] S144. The initial trust score is weighted and integrated with the evaluation result of the preset rule model to obtain the initial trust score of the target user. For example, the initial trust score of user A is 0.85, and the evaluation result of the rule model is 0.9. Then, the initial trust score after weighted integration is 0.87. The preset rule model sets rules according to business experience and user needs. For example, users with more than 10 transactions have a higher trust score.

[0073] S145. Classify the target user into a preset trust level according to the initial trust score:

[0074] Based on the K-means clustering method, cluster analysis is performed on the historical user trust scores to obtain a dynamic grading threshold; the dynamic grading threshold is combined with the time decay factor to construct a dynamic grade division standard. For example, based on the historical data clustering analysis, the thresholds of the three grades are 0.6, 0.8, and 0.9 respectively. Combined with the time decay factor, the threshold of the most recent month can be appropriately reduced, for example, by 0.05; the initial trust score of the target user is graded based on the dynamic grade division standard to obtain the initial trust grade of the target user. For example, the initial trust score of user A is 0.87, and according to the dynamic grade division standard, it is divided into the second grade;

[0075] Step S14 in this embodiment uses technical means such as multi-dimensional data collection, refined preprocessing, feature cross-construction and model fusion to more comprehensively and accurately evaluate the user's trust and reduce the misjudgment rate; uses an exponential decay function and a dynamic grade classification standard to dynamically adjust the trust evaluation result according to the timeliness of user behavior data and changes in historical data distribution, so as to more promptly and accurately reflect the user's credit status; and the design of automated data processing and model calculation reduces manual intervention, improves the efficiency of trust evaluation, and reduces operating costs.

[0076] S2. Based on the trust level of the target user, determine the system resource pool accessible to the target user, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; collect the real-time business demand information of the target user, wherein the real-time business demand information includes business type, business scale, business timeliness requirements and business priority; construct a resource demand prediction model, input the real-time business demand information into the resource demand prediction model, and output a target resource configuration plan;

[0077] S21. Establish a user trust level system, which divides users into different levels, such as level one, level two, level three, etc., and each level corresponds to different resource access rights. Level one users enjoy the highest resource access rights and can access all resource pools in the system; level two users are second and can access some resource pools; level three users have the lowest rights and can only access limited resource pools. Each resource pool has pre-set quotas for computing resources, storage resources, and bandwidth resources, wherein the computing resources set usage quotas based on the number of CPU cores and GPU memory capacity, the storage resources set usage quotas based on storage capacity and IOPS performance indicators, and the bandwidth resources set usage quotas based on peak bandwidth and number of concurrent connections; an example of determining the system resource pool accessible to the target user based on the trust level of the target user is: the resource pool A accessible to the level one user has a computing resource quota of 1000 CPU cores and 10TB GPU memory, a storage resource quota of 100TB storage capacity and 10,000 IOPS, and a bandwidth resource quota of 10Gbps peak bandwidth and 10,000 concurrent connections. Resource pool B, which is accessible to secondary users, has a computing resource quota of 500 CPU cores and 5TB GPU memory, a storage resource quota of 50TB storage capacity and 5,000 IOPS, and a bandwidth resource quota of 5Gbps peak bandwidth and 5,000 concurrent connections.

[0078] S22. Collect the target user's business demand information in real time, where the real-time business demand information includes business type, business scale, business timeliness requirements and business priority. Suppose a user submits an image recognition task, which is a typical machine learning application. Classify and identify the business type through a deep learning model to determine that it belongs to the image recognition category. Evaluate the business scale based on the number of pictures submitted by the user, the estimated number of concurrent requests and the required storage space. For example, a user submitted 10,000 pictures, the estimated number of concurrent requests is 100, and the required storage space is 1TB. Determine the business priority as high based on the service level agreement and the importance of the business;

[0079] S23. Construct a resource demand prediction model, which includes three main modules: a long short-term memory network module for capturing the temporal dependency of business demand, an attention mechanism module for focusing on resource usage features at key time points, and a fully connected layer module for feature fusion;

[0080] S231. The resource demand prediction model uses a sliding time window method to extract the periodic characteristics of the business load. For example, by counting the resource usage at the same time every day in the past week, it can be found that the resource demand of the image recognition task reaches a peak between 9 am and 12 pm every day;

[0081] S232. Mining the correlation between different types of resource requirements. For example, there is a positive correlation between the computing resource requirements and storage resource requirements of image recognition tasks, that is, the higher the computing resource requirements, the higher the storage resource requirements;

[0082] S233. Identify resource usage patterns in specific business scenarios. For example, image recognition tasks have higher requirements for GPU resources in the model training phase, and higher requirements for CPU resources in the inference phase.

[0083] S234. In this embodiment, through steps S231 to S233, the long short-term memory network module and the attention mechanism module extract time series features, resource demand correlation features, business scenario features and other features, wherein:

[0084] The time series features include periodic features extracted by the sliding time window method (e.g., resource usage at the same time period every day in the past week), which are used to help the model capture daily, weekly or monthly periodic patterns of resource demand; the time series features also include peak time point features obtained by identifying the peak of resource demand in a specific time period (e.g., 9 a.m. to 12 p.m.), and the time points in a specific time period usually correspond to the peak period of the business;

[0085] The resource demand correlation characteristics include computing resource demand (such as the usage of computing resources such as CPU and GPU), storage resource demand (such as the usage of storage resources such as memory and hard disk), network resource demand (such as network bandwidth) and correlation characteristics (such as the positive correlation between computing resource demand and storage resource demand in image recognition tasks);

[0086] The business scenario characteristics include task type characteristics and task stage characteristics obtained by different types of tasks (such as image recognition, natural language processing, etc.) having different resource demand patterns (such as image recognition tasks have higher demand for GPU resources in the model training stage, and higher demand for CPU resources in the inference stage):

[0087] Input these extracted features into the fully connected layer module for feature fusion;

[0088] S24. Input the real-time business demand information into the resource demand prediction model, and obtain the initial resource allocation plan through the collaborative processing of the long short-term memory network module and the attention mechanism module. The long short-term memory network module can capture the temporal dependencies of business needs, for example, predict future resource needs based on resource usage over the past period of time. The attention mechanism module focuses on resource usage characteristics at key time points, for example, focusing on resource usage during business peak periods. Through the collaborative processing of these two modules, an initial resource allocation plan is obtained, for example, it is preliminarily estimated that 200 CPU cores, 2TB GPU video memory, 2TB storage capacity and 2Gbps bandwidth need to be allocated;

[0089] S25. Input the initial resource allocation plan into the multi-objective optimization model, and optimize it by considering the total resource constraints, priority constraints, performance constraints and cost constraints, and output the target resource allocation plan. The total resource constraint means that the total amount of resources available to the system is limited; the priority constraint means that high-priority services should be given priority in resource allocation; the performance constraint means that the resource allocation plan needs to meet the performance requirements of the service; the cost constraint means minimizing the cost of resource use. Through the processing of the multi-objective optimization model, the final optimized resource allocation plan is obtained, for example, 150 CPU cores, 1.5TB GPU video memory, 1.5TB storage capacity and 1.5Gbps bandwidth are finally allocated;

[0090] In step S2 of this embodiment, through accurate resource demand prediction and multi-objective optimization, it is possible to avoid over-allocation and waste of resources, thereby improving resource utilization. For example, by analyzing and predicting historical data, resource utilization can be increased from 70% to 90%; by considering the priority and performance constraints of the business, it is possible to ensure that key businesses have sufficient resource guarantees to meet the performance requirements of the business. For example, the response time of high-priority businesses can be reduced from 1 second to 0.5 seconds; by optimizing the resource allocation plan, the cost of resource use can be reduced, for example, the monthly cloud service fee can be reduced by 10%;

[0091] S3. Allocate system resources from the system resource pool according to the target resource configuration scheme for the target user; monitor the target user's use of the allocated system resources in real time to obtain resource utilization efficiency data;

[0092] S31. Allocate system resources to the target user according to the target resource configuration scheme:

[0093] The resource management control component is used to allocate CPU shares and limit memory usage for the computing resources, the quality of service control component is used to limit storage capacity quotas and input and output performance for the storage resources, and the flow control component is used to limit the number of concurrent connections and link establishment rate for the bandwidth resources;

[0094] S32. Establish a real-time monitoring and collection mechanism to monitor the target user's use of the allocated system resources in real time, the real-time monitoring and collection mechanism includes a differentiated sampling period setting module and a monitoring data collection module; set a sampling period for the multi-dimensional monitoring indicators through the differentiated sampling period setting module, and collect multi-dimensional monitoring indicator data of the target user's use of the system resources through the monitoring data collection module; wherein the multi-dimensional monitoring indicator data includes system-level indicator data and business-level indicator data, the system-level indicator data includes CPU usage, memory usage, disk input and output throughput, and network bandwidth utilization, and the business-level indicator data includes request response time, business processing success rate, and concurrent processing quantity;

[0095] S33. Construct a real-time data processing pipeline, which includes a data compression module and an anomaly detection module; perform incremental compression on the multi-dimensional monitoring indicator data through the data compression module, identify data collection anomalies through the anomaly detection module and perform corrections to obtain compressed and corrected monitoring data; preprocess the compressed and corrected monitoring data, use the Z-score method to filter outliers, use the sliding average method to perform smoothing, and use the time series prediction method to complete missing values ​​to obtain preprocessed monitoring data;

[0096] S34. Extract monitoring features from the preprocessed monitoring data, including: extracting time series features to obtain trend features and cycle features, calculating statistical features to obtain mean features and variance features, analyzing correlation features to obtain indicator association features; using hierarchical analysis method to determine the weight of the monitoring features, and weighted fusion of the monitoring features to obtain resource utilization efficiency data;

[0097] Step S3 is further described by example:

[0098] Prepare a system resource pool. The resource pool contains a variety of system resources, such as computing resources (CPU, memory), storage resources (storage capacity, IOPS), and bandwidth resources (bandwidth limit, number of concurrent connections). Assume that the resource pool has 100 CPU cores, 1TB memory, 500TB storage space, and 1Gbps bandwidth.

[0099] Then, formulate a target resource allocation plan. According to the needs of the target users and the service level agreement (SLA), determine the quantity and performance indicators of various resources required by the target users. For example, allocate 20 CPU cores, 200GB memory, 10TB storage space, and 100Mbps bandwidth to a certain user.

[0100] Next, allocate system resources to the target user. Allocate the corresponding resources to the target user from the resource pool according to the target resource configuration scheme. Use control components, such as cgroups or containerization technologies such as Docker, to allocate CPU shares and limit memory usage for computing resources. Use quality of service control components, such as QoS policies or storage array management software, to set storage capacity quotas and input / output performance (IOPS) limits for storage resources. Use traffic control components, such as traffic shapers or firewall rules, to limit the number of concurrent connections and link establishment rates for bandwidth resources. For example, use cgroups to limit the user's CPU usage to no more than 80% and set the memory usage cap to 200GB. Use storage array management software to allocate 10TB of storage space to the user and set the IOPS cap to 10,000. Use traffic shapers to limit the user's bandwidth usage to no more than 100Mbps.

[0101] After that, a real-time monitoring collection mechanism is established. This mechanism includes a differentiated sampling period setting module and a monitoring data collection module. The differentiated sampling period setting module sets different sampling periods for different monitoring indicators according to their importance. For example, the sampling period of important indicators such as CPU usage and memory usage is set to 1 second, while the sampling period of secondary indicators such as disk input and output throughput and network bandwidth utilization is set to 5 seconds. The monitoring data collection module collects multi-dimensional monitoring indicator data of target users using system resources according to the set sampling period. These data include system-level indicator data, such as CPU usage, memory usage, disk I / O throughput and network bandwidth utilization, as well as business-level indicator data, such as request response time, business processing success rate and concurrent processing number. For example, the data collected at a certain point in time may be: CPU usage 70%, memory usage 150GB, disk I / O throughput 8000IOPS, network bandwidth utilization 80Mbps, request response time 100ms, business processing success rate 99%, and concurrent processing number 1000.

[0102] Build a real-time data processing pipeline. The pipeline includes a data compression module and an anomaly detection module. The data compression module incrementally compresses the collected multi-dimensional monitoring indicator data to reduce the data storage space and the amount of calculation for subsequent processing. For example, compression algorithms such as differential encoding or run-length encoding are used. The anomaly detection module identifies data collection anomalies, such as data loss, data duplication, or data outside a reasonable range, and corrects them. For example, if the CPU usage at a certain point in time suddenly becomes 0, it can be judged as a data collection anomaly and filled with the CPU usage at the previous point in time. Preprocess the compressed and corrected monitoring data. Use the Z-score method to filter outliers, such as treating data points with an absolute value of Z-score greater than 3 as outliers and removing them. Use the sliding average method for smoothing, such as using a sliding window of 5 data points to calculate the average. Use the time series prediction method to fill in missing values, such as using the ARIMA model to predict missing data points.

[0103] Extract features from the preprocessed monitoring data. Extract time series features, such as growth rate, decline rate, etc., to obtain trend features; extract periodic patterns, such as daily peak time, weekly fluctuation pattern, etc., to obtain period features. Calculate statistical features, such as mean, variance, maximum, minimum, etc., to obtain mean features and variance features. Analyze correlation features, such as the relationship between CPU usage and memory usage, the relationship between request response time and the number of concurrent processing, etc., to obtain indicator correlation features. For example, the average CPU usage of a user is 60%, the variance is 10%, the average request response time is 150ms, the variance is 20ms, and there is a positive correlation between CPU usage and memory usage.

[0104] The analytic hierarchy process is used to determine the weight of each feature, and the features are weighted and fused to obtain resource utilization efficiency data. For example, the weight of the trend feature is 0.3, the weight of the cycle feature is 0.2, the weight of the mean feature is 0.2, the weight of the variance feature is 0.1, and the weight of the indicator association feature is 0.2. The value of each feature is multiplied by the corresponding weight, and then summed to obtain the final resource utilization efficiency data. For example, the resource utilization efficiency data of a user is 0.85.

[0105] In step S3 of this embodiment, through real-time monitoring and dynamic adjustment of resource allocation, resources are allocated to users who need them most according to their actual needs, avoiding idleness and waste of resources, thereby improving overall resource utilization; through service quality control components and real-time monitoring, the user's service level agreement (SLA) can be guaranteed, and performance bottlenecks can be discovered and resolved in a timely manner, thereby ensuring the user's service quality; through resource utilization efficiency data, the resource utilization of different users can be evaluated, and data support can be provided for resource optimization strategies, thereby improving operational efficiency.

[0106] S4. According to the resource utilization efficiency data, combined with the preset trust dynamic adjustment rules, the trust level of the target user is dynamically updated; when the trust level of the target user changes, the system resources are automatically matched and allocated again;

[0107] S41. Obtaining preset trust dynamic adjustment rules, the preset trust dynamic adjustment rules include trust level division rules and level adjustment trigger rules; the trust level division rules define trust level boundary values ​​and transition buffers, the level adjustment trigger rules define level adjustment conditions and cooling periods;

[0108] S42. Dynamically update the trust level of the target user according to the resource utilization efficiency data and the preset trust dynamic adjustment rules, including: when the resource utilization efficiency data is above the transition buffer zone in a continuous preset evaluation period, increase the trust level; when the resource utilization efficiency data is below the transition buffer zone or a violation occurs, reduce the trust level;

[0109] S43. When the trust level of the target user changes, the resource quota is recalculated based on the new trust level to generate a target resource allocation scheme; the target resource allocation scheme includes a new computing resource quota, storage resource quota and network resource quota;

[0110] S44. Execute automatic matching and allocation of system resources:

[0111] Assess the existing resource usage, lock the available resource pool, and generate a resource migration plan; the resource migration plan adopts a step-by-step migration strategy to complete resource reallocation while ensuring service continuity; monitor the service status in real time during the resource migration process, and automatically initiate a rollback mechanism when an abnormality occurs;

[0112] Step S4 is further described by example:

[0113] Get the preset trust dynamic adjustment rules. These rules include trust level division rules and level adjustment trigger rules. Trust level division rules define different trust levels and transition buffers between levels. For example, the trust level is divided into five levels: low, lower, medium, higher, and high, with corresponding boundary values ​​of 0, 0.2, 0.4, 0.6, and 0.8, and the transition buffer is the boundary value ±0.05. The level adjustment trigger rule defines the conditions for level adjustment, such as resource utilization efficiency data being above or below the transition buffer for multiple consecutive evaluation cycles, and the cooling period, that is, the time interval between two level adjustments.

[0114] The trust level of the target user is dynamically updated based on the resource efficiency data and the preset trust dynamic adjustment rules. For example, if a user's initial trust level is medium, if his resource efficiency data is higher than 0.45 for three consecutive evaluation cycles, his trust level will be raised to high. If his resource efficiency data is lower than 0.35 or there is a violation, his trust level will be lowered to low.

[0115] When the target user's trust level changes, the resource quota is recalculated based on the new trust level to generate a target resource allocation plan. For example, when the user's trust level is increased from medium to high, his computing resource quota, storage resource quota, and network resource quota will increase by 10%, 20%, and 15%, respectively.

[0116] Finally, perform automated matching and allocation of system resources. This process includes evaluating the existing resource usage, locking the available resource pool, and generating a resource migration plan. The resource migration plan adopts a step-by-step migration strategy to complete resource reallocation while ensuring service continuity. During the resource migration process, the service status is monitored in real time, and the rollback mechanism is automatically initiated when an abnormality occurs. For example, to migrate a user's virtual machine from server A to server B, first migrate some services to server B, and after confirming that the services are running normally, migrate the remaining services to server B, and finally release the resources on server A;

[0117] Step S4 of this embodiment can allocate resources preferentially to more trustworthy users by dynamically adjusting resource quotas according to the user's trust level, thereby avoiding resource waste and improving overall resource utilization; by reducing the resource quota of untrustworthy users, their access to system resources can be restricted, thereby reducing security risks; by providing more resources for highly trusted users, the stability and performance of their services can be guaranteed, thereby improving user experience.

[0118] Embodiment 2:

[0119] Figure 2FIG. 2 is a schematic diagram of a resource automatic matching system based on dynamic trust level classification according to Embodiment 2 of the present invention. Figure 2 As shown, the system comprises:

[0120] User behavior and feature analysis module, trust model and rating module, resource pool management and allocation module, and resource monitoring and dynamic adjustment module, among which:

[0121] The user behavior and feature analysis module is used to obtain user historical behavior data, which includes user transaction records, user contract fulfillment, user credit score and user resource usage records;

[0122] The trust model and rating module constructs a user feature vector matrix based on the historical behavior data, and trains the user feature vector matrix through a deep learning model to generate a user trust rating model; applies the user trust rating model to a target user to obtain an initial trust score for the target user, and classifies the target user into a preset trust level according to the initial trust score;

[0123] The resource pool management and allocation module determines the system resource pool accessible to the target user based on the trust level of the target user, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; collects the real-time business demand information of the target user, wherein the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; constructs a resource demand prediction model, inputs the real-time business demand information into the resource demand prediction model, and outputs a target resource allocation plan; allocates system resources to the target user from the system resource pool according to the target resource allocation plan;

[0124] The resource monitoring and dynamic adjustment module monitors the target user's use of the allocated system resources in real time and obtains resource usage efficiency data; based on the resource usage efficiency data and in combination with preset trust dynamic adjustment rules, the trust level of the target user is dynamically updated; when the trust level of the target user changes, the system resources are automatically matched and allocated again.

[0125] Embodiment 3:

[0126] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the automatic resource matching method based on dynamic grading of trust as described in any one of Embodiment 1 is implemented.

[0127] Embodiment 4:

[0128] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for automatically matching resources based on dynamic trust grading as described in any one of the embodiments 1 is implemented.

[0129] It is worth noting that the system, electronic device and computer-readable storage medium described in the present invention are all based on the same inventive concept as the method described in Example 1, and will not be described in detail here.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A resource automatic matching method based on dynamic trust classification, characterized in that: include: Acquire user historical behavior data, construct a user feature vector matrix based on the historical behavior data, construct a deep neural network model based on the user feature vector matrix, and generate a user trust scoring model through optimization training; apply the user trust scoring model to a target user to obtain an initial trust score of the target user, and classify the target user into a preset trust level according to the initial trust score; Based on the trust level of the target user, determine the system resource pool accessible to the target user, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; collect the real-time business demand information of the target user, wherein the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; construct a resource demand prediction model, input the real-time business demand information into the resource demand prediction model, and output a target resource configuration plan; Allocate system resources to the target user from the system resource pool according to the target resource configuration plan; monitor the target user's use of the allocated system resources in real time to obtain resource usage efficiency data; dynamically update the target user's trust level based on the resource usage efficiency data and in combination with preset trust dynamic adjustment rules; and automatically re-match and allocate system resources when the target user's trust level changes.

2. The automatic resource matching method based on dynamic trust classification according to claim 1 is characterized in that: Acquire user historical behavior data, construct a user feature vector matrix based on the historical behavior data, construct a deep neural network model based on the user feature vector matrix, and generate a user trust rating model through optimization training as follows: The historical behavior data includes user transaction records, user contract performance, user credit score and user resource usage records, wherein: the user transaction records include transaction time, transaction amount, transaction type and transaction status, and the user transaction records are constructed into a transaction behavior sequence; the user contract performance includes contract performance, payment timeliness and service completion; the user credit score includes credit report score, third-party credit score, platform credit record score and complaint handling score; the user resource usage record includes resource call frequency record, resource occupation time record, resource usage efficiency record and resource over-limit usage record; Constructing a user feature vector matrix based on the historical behavior data includes: setting a performance weight matrix for the user's performance and quantifying it to obtain a performance index; using a hierarchical analysis method to determine the score weight for the user's credit score to obtain a comprehensive credit score; standardizing the transaction behavior sequence, the performance index, the comprehensive credit score and the user's resource usage record to obtain standardized data; using a sliding time window method to extract time series features from the standardized data to obtain time series feature data; correlating the transaction amount data in the time series feature data with the performance index to obtain transaction performance features, correlating the resource utilization efficiency record with the comprehensive credit score to obtain credit resource features; combining the transaction performance features, the credit resource features and the time series feature data to construct the user feature vector matrix; A deep neural network model is constructed based on the user feature vector matrix and a user trust scoring model is generated through optimization training, including: constructing a deep neural network model, the input layer dimension of the deep neural network model is the same as the dimension of the user feature vector matrix, and the deep neural network model includes three hidden layers using ReLU activation function and one output layer using Sigmoid function; constructing a composite loss function including a mean square error loss term and a ranking loss term, and introducing a regularization term into the composite loss function; using a batch gradient descent method combined with an Adam optimizer to train the deep neural network model, and using a dynamic learning rate strategy during the training process, wherein the dynamic learning rate strategy includes using a first learning rate in the initial stage of training and a second learning rate in the later stage of training, so as to obtain a basic trust scoring model; performing weighted fusion on the output result of the basic trust scoring model and the output result of the preset rule model to generate the user trust scoring model.

3. The resource automatic matching method based on dynamic trust classification according to claim 2 is characterized in that: Applying the user trust rating model to a target user to obtain an initial trust score for the target user, and classifying the target user into a preset trust level according to the initial trust score includes: Applying the user trust rating model to the target user specifically includes: collecting the target user's behavior data within ninety days before the current time point, the behavior data including transaction data, performance data, credit data and resource usage data; preprocessing the behavior data, including standard deviation normalization of numerical features, one-hot encoding of categorical features, time window feature extraction of time series data, and model mean filling of missing data items; using an exponential decay function to calculate the time weight coefficient of the preprocessed behavior data, the decay coefficient in the exponential decay function is inversely proportional to the time interval; weighting the preprocessed behavior data based on the time weight coefficient to obtain weighted behavior data; Constructing a target user feature vector based on the weighted behavior data specifically includes: performing feature cross construction on the weighted behavior data to obtain cross features; normalizing the cross features to obtain the target user feature vector; inputting the target user feature vector into the user trust scoring model, and obtaining an initial trust score through forward propagation of the user trust scoring model; weighted fusion of the initial trust score and the evaluation result of the preset rule model to obtain the initial trust score of the target user; The target user is divided into a preset trust level according to the initial trust score, specifically including: performing cluster analysis on the historical user trust scores based on the K-means clustering method to obtain a dynamic grading threshold; combining the dynamic grading threshold with a time decay factor to construct a dynamic grading standard; and grading the initial trust score of the target user based on the dynamic grading standard to obtain the initial trust level of the target user.

4. The resource automatic matching method based on dynamic trust level classification according to claim 1 is characterized in that: Based on the trust level of the target user, determine the system resource pool accessible to the target user, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; collect the real-time business demand information of the target user, wherein the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; construct a resource demand prediction model, input the real-time business demand information into the resource demand prediction model, and output the target resource configuration plan including: Determine the system resource pool accessible to the target user according to the trust level of the target user, set the usage quota of the computing resources based on the number of CPU cores and GPU memory capacity, set the usage quota of the storage resources based on the storage capacity and IOPS performance index, and set the usage quota of the bandwidth resources based on the peak bandwidth and the number of concurrent connections; Collecting the real-time business demand information of the target user, the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; using a deep learning model to classify and identify the business type, evaluating the business scale based on data processing volume, number of concurrent requests and storage space requirements, and determining the business priority based on business importance and service level agreement; Constructing a resource demand prediction model, the resource demand prediction model includes: a long short-term memory network module for capturing the temporal dependency of business demand, an attention mechanism module for focusing on resource usage features at key time points, and a fully connected layer module for feature fusion; the resource demand prediction model uses a sliding time window method to extract the periodic features of business load, mines the correlation between different types of resource demands, identifies resource usage patterns in specific business scenarios, and inputs the features extracted by the long short-term memory network module and the attention mechanism module into the fully connected layer module of the resource demand prediction model for feature fusion; The real-time business demand information is input into the resource demand prediction model, and an initial resource allocation plan is obtained through the collaborative processing of the long short-term memory network module and the attention mechanism module; a multi-objective optimization model is constructed, and the initial resource allocation plan is input into the multi-objective optimization model, and optimization is performed based on total resource constraints, priority constraints, performance constraints and cost constraints to obtain an optimized resource allocation plan, namely, a target resource allocation plan.

5. The resource automatic matching method based on dynamic trust level classification according to claim 1 is characterized in that: Allocating system resources to the target user from the system resource pool according to the target resource configuration scheme; Real-time monitoring of the target user's use of the allocated system resources to obtain resource usage efficiency data includes: Allocate system resources to target users from a system resource pool according to a target resource configuration plan, wherein the system resource pool includes computing resources, storage resources, and bandwidth resources; allocate CPU shares and limit memory usage on the computing resources through a resource management control component, impose storage capacity quotas and input and output performance restrictions on the storage resources through a service quality control component, and limit the number of concurrent connections and link establishment rate on the bandwidth resources through a flow control component; Establish a real-time monitoring and collection mechanism, which includes a differentiated sampling period setting module and a monitoring data collection module; set a sampling period for the multi-dimensional monitoring indicators through the differentiated sampling period setting module, and collect multi-dimensional monitoring indicator data of the target user using the system resources through the monitoring data collection module; wherein the multi-dimensional monitoring indicator data includes system-level indicator data and business-level indicator data, wherein the system-level indicator data includes CPU usage, memory usage, disk input and output throughput, and network bandwidth utilization, and the business-level indicator data includes request response time, business processing success rate, and concurrent processing quantity; Construct a real-time data processing pipeline, which includes a data compression module and an anomaly detection module; incrementally compress the multi-dimensional monitoring indicator data through the data compression module, identify data collection anomalies through the anomaly detection module and correct them, so as to obtain compressed and corrected monitoring data; pre-process the compressed and corrected monitoring data, use the Z-score method to filter outliers, use the sliding average method to smooth, and use the time series prediction method to fill missing values, so as to obtain pre-processed monitoring data; Extract monitoring features from the preprocessed monitoring data, including: extracting time series features to obtain trend features and period features, calculating statistical features to obtain mean features and variance features, analyzing correlation features to obtain indicator association features; using hierarchical analysis method to determine the weights of the monitoring features, and weighted fusion of the monitoring features to obtain resource utilization efficiency data.

6. The automatic resource matching method based on dynamic trust classification according to claim 1 is characterized in that: Dynamically updating the trust level of the target user according to the resource utilization efficiency data and in combination with a preset trust dynamic adjustment rule; When the trust level of the target user changes, re-automatic matching and allocation of system resources includes: Obtaining a preset trust dynamic adjustment rule, wherein the preset trust dynamic adjustment rule includes a trust level division rule and a level adjustment trigger rule; the trust level division rule defines a trust level boundary value and a transition buffer zone, and the level adjustment trigger rule defines a level adjustment condition and a cooling period; Dynamically updating the trust level of the target user according to the resource utilization efficiency data and the preset trust dynamic adjustment rule, including: when the resource utilization efficiency data is above the transition buffer zone in a continuous preset evaluation period, raising the trust level; when the resource utilization efficiency data is below the transition buffer zone or a violation occurs, lowering the trust level; When the trust level of the target user changes, the resource quota is recalculated based on the new trust level to generate a target resource allocation scheme; the target resource allocation scheme includes a new computing resource quota, a storage resource quota and a network resource quota; Execute automatic matching and allocation of system resources, including: evaluating the existing resource occupancy, locking the available resource pool, and generating a resource migration plan; the resource migration plan adopts a step-by-step migration strategy to complete resource reallocation while ensuring service continuity; monitor the service status in real time during the resource migration process, and automatically initiate a rollback mechanism when an abnormality occurs.

7. A resource automatic matching system based on dynamic trust grading, used to implement the resource automatic matching method based on dynamic trust grading as described in any one of claims 1 to 6, characterized in that: It includes user behavior and feature analysis module, trust model and rating module, resource pool management and allocation module and resource monitoring and dynamic adjustment module, among which: The user behavior and feature analysis module is used to obtain user historical behavior data, which includes user transaction records, user contract fulfillment, user credit score and user resource usage records; The trust model and rating module constructs a user feature vector matrix based on the historical behavior data, and trains the user feature vector matrix through a deep learning model to generate a user trust rating model; applies the user trust rating model to a target user to obtain an initial trust score for the target user, and classifies the target user into a preset trust level according to the initial trust score; The resource pool management and allocation module determines the system resource pool accessible to the target user based on the trust level of the target user, wherein the system resource pool includes computing resources, storage resources and bandwidth resources; collects the real-time business demand information of the target user, wherein the real-time business demand information includes business type, business scale, business timeliness requirement and business priority; constructs a resource demand prediction model, inputs the real-time business demand information into the resource demand prediction model, and outputs a target resource allocation plan; allocates system resources to the target user from the system resource pool according to the target resource allocation plan; The resource monitoring and dynamic adjustment module monitors the target user's use of the allocated system resources in real time and obtains resource usage efficiency data; based on the resource usage efficiency data and in combination with preset trust dynamic adjustment rules, the trust level of the target user is dynamically updated; when the trust level of the target user changes, the system resources are automatically matched and allocated again.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the automatic resource matching method based on dynamic grading of trust as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the automatic resource matching method based on dynamic trust grading as described in any one of claims 1 to 6 is implemented.

Images