Resource urgency assessment and dynamic scheduling method and system based on behavior characteristics

Abstract

The invention provides a resource urgency degree evaluation and dynamic scheduling method and system based on behavior characteristics, and the method comprises the steps: collecting equipment protocol characteristic data, equipment resource state data and user historical behavior data, and carrying out the preprocessing, and obtaining a standardized data set; by fusing a convolutional neural network, a bidirectional gating circulation unit and a network model of an attention mechanism, extracting spatial-temporal characteristics and a long and short term dependency relationship of the standardized data set, and generating depth behavior characteristics; dynamically calculating a task urgency degree weight from three dimensions of business priority, an equipment performance baseline and environment health degree by adopting a fuzzy analytic hierarchy process based on the depth behavior characteristics, and generating a comprehensive urgency degree score; according to the comprehensive urgency score, generating a dynamic scheduling strategy and executing resource allocation; and monitoring the resource utilization rate and the service level agreement compliance rate, and feeding back for optimizing a subsequent scheduling strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of resource scheduling, and in particular to a method and system for resource urgency assessment and dynamic scheduling based on behavioral characteristics. Background Art

[0002] In cutting-edge fields like computer networking, the Internet of Things, and intelligent device management, efficient allocation of device resources, such as network bandwidth, computing power, and storage, is key to ensuring efficient business operations. Proper resource allocation not only impacts system stability but also directly impacts business response speed and user experience.

[0003] Resource allocation refers to the rational distribution of limited resources to various links based on the needs of different services and equipment. QoS (Quality of Service) mechanisms are an important means of ensuring that different services receive corresponding service levels.

[0004] Among conventional technologies, mainstream resource allocation technologies mostly rely on static rules or protocol characteristics, which have many limitations: First, the extensive allocation dominated by static policies often uses fixed bandwidth quotas or protocol priority queues to schedule resources, which can only cope with known scenarios. When faced with sudden business or equipment anomalies, it will cause critical business to be stuck and non-critical business resources to be redundant. Second, the accuracy of business scheduling is insufficient. Traditional QoS mechanisms rely on shallow features to identify business needs, ignore various aspects of user and device data, and cannot adjust priorities based on the real-time status of the device, resulting in poor scheduling accuracy. Third, there is a lack of dynamic adaptability. Existing technologies rely on manually preset rules or simple policies, lack intelligent decision-making mechanisms, and are difficult to respond quickly to business changes, resulting in a mismatch between resource allocation and actual needs. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a resource urgency assessment and dynamic scheduling method and system based on behavioral characteristics, which improves the efficiency of resource allocation by constructing a complete technical chain of "behavior analysis-urgency assessment-dynamic scheduling".

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for resource urgency assessment and dynamic scheduling based on behavioral characteristics, comprising: Collect device protocol feature data, device resource status data, and user historical behavior data, and pre-process them to obtain a normalized data set; By integrating a convolutional neural network, a bidirectional gated recurrent unit, and an attention mechanism network model, the spatiotemporal features and long-term and short-term dependencies of the normalized dataset are extracted to generate deep behavioral features; Based on deep behavioral characteristics, the fuzzy hierarchical analysis method is used to dynamically calculate the task urgency weight from three dimensions: business priority, equipment performance baseline, and environmental health, and generate a comprehensive urgency score; generating a dynamic scheduling strategy and executing resource allocation based on the comprehensive urgency score; Monitor resource utilization and service level agreement compliance, and provide feedback to optimize subsequent scheduling strategies.

[0007] In a second aspect, the present invention provides a resource urgency assessment and dynamic scheduling system based on behavioral characteristics, comprising: The data collection and processing module is used to collect device protocol feature data, device resource status data, and user historical behavior data, and perform preprocessing to obtain a normalized data set; A feature extraction module is used to extract the spatiotemporal features and long-term and short-term dependencies of the normalized dataset by fusing a network model with a convolutional neural network, a bidirectional gated recurrent unit, and an attention mechanism, thereby generating deep behavioral features; The scoring module is used to dynamically calculate the task urgency weight from three dimensions: business priority, equipment performance baseline, and environmental health, based on deep behavioral characteristics and using fuzzy hierarchical analysis method to generate a comprehensive urgency score; a scheduling module, configured to generate a dynamic scheduling strategy and perform resource allocation based on the comprehensive urgency score; The optimization module is used to monitor resource utilization and service level agreement compliance, and provide feedback for optimizing subsequent scheduling strategies.

[0008] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a resource urgency assessment and dynamic scheduling method based on behavioral characteristics described in the first aspect.

[0009] In a fourth aspect, the present invention provides a computer device comprising 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 steps of the method for resource urgency assessment and dynamic scheduling based on behavioral characteristics described in the first aspect are implemented.

[0010] Compared with the prior art, the present invention has the following beneficial effects: (1) Improved accuracy of dynamic resource allocation and significantly enhanced key business assurance capabilities: In response to the problem of "static strategies leading to key business stalls / non-key business resource redundancy" in the background technology, the dynamic priority evaluation model of the present invention achieves accurate quantification of resource requirements through a three-dimensional indicator system (business / equipment / environment) and fuzzy hierarchical analysis (FAHP). In a mixed business scenario (video conferencing + file transfer + background tasks), the latency jitter rate of key business (video conferencing) is reduced from 45ms in the traditional solution to 28ms (a reduction of 37.8%), and the packet loss rate is reduced from 0.8% to 0.3% (a reduction of 62.5%), and the business fluency is significantly improved (according to the ITU-T Y.1541 standard evaluation, the QoE index is improved by 40%). Through a dynamic weight adjustment mechanism (such as the three-level weight reconstruction strategy for emergency traffic), bandwidth resource utilization is increased from 65% to 89%, and the redundant resource usage of non-key business is reduced by 55%, achieving refined scheduling of "on-demand allocation".

[0011] (2) With the help of multi-dimensional data fusion technology, the scientific nature of scheduling decisions can break through traditional limitations: In response to the limitation of "one-sided use of behavioral data", the behavioral analysis engine adopted by this invention attempts to integrate 18+ dimensions of data, including real-time status of equipment (load rate, hardware configuration), historical habits (high-frequency time period), business characteristics (delay sensitivity), etc. This approach is different from the traditional shallow recognition mode that relies only on protocol ports, and is committed to processing behavioral data from a more comprehensive perspective. Through the CNN-BiGRU fusion network and KPCA data dimensionality reduction technology, the recognition accuracy of complex business scenarios has been improved from 72% of the traditional solution to 96.3% (the test data set contains 20 heterogeneous business types). For example, it can accurately distinguish between video conferencing services on high-load PCs and low-load mobile phones. The former is dynamically allocated 150Mbps bandwidth + 4-core CPU, and the latter is allocated 80Mbps + 2 cores. The resource allocation error rate is reduced from 38% to 12%.

[0012] (3) Real-time data facilitates intelligent decision-making, significantly improving dynamic response speed: To address the problem of "lack of dynamic adaptability", the hierarchical scheduling architecture (policy control center + traffic scheduling engine + device adaptation platform) constructed by this invention achieves an end-to-end closed loop from data collection to policy execution, responding to changes in business load in milliseconds. In a simulated remote collaboration emergency scenario (concurrent video conferencing instant access), traditional static policies require manual intervention and adjustment (taking an average of 5 minutes). This solution uses a sliding window to calculate network health in real time (100ms update frequency) and dynamic weight reconstruction, completing resource reallocation within 280ms. The service access success rate increased from 65% to 98%, and the response speed increased by 6 times.

[0013] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention but do not constitute a limitation of the present invention.

[0015] Figure 1 A main flow chart of a method for resource urgency assessment and dynamic scheduling based on behavioral characteristics provided by an embodiment of the present invention; Figure 2 A detailed flow chart of a method for resource urgency assessment and dynamic scheduling based on behavioral characteristics provided by an embodiment of the present invention; Figure 3 A flow chart of the data acquisition module method provided by an embodiment of the present invention; Figure 4 A flow chart of the data preprocessing module method provided by an embodiment of the present invention; Figure 5 A data processing flow chart of the depth analysis unit provided in an embodiment of the present invention; Figure 6 A flow chart of the urgency assessment module method provided in an embodiment of the present invention; Figure 7 A flowchart of a hierarchical scheduling architecture module method provided by an embodiment of the present invention; Figure 8 This is a flow chart of the real-time monitoring and feedback module method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] Currently, mainstream resource allocation technologies rely on static rules or protocol features. However, these traditional technologies have significant limitations when faced with dynamically changing business loads and complex and diverse application scenarios: (1) Extensive allocation dominated by static policies: Existing solutions generally use fixed bandwidth quotas (e.g., 50% bandwidth preset for video conferencing) or protocol priority queues (e.g., UDP traffic takes precedence over TCP) for resource scheduling. Such static policies can only handle known business scenarios and cannot cope with real-time changes in user behavior (e.g., sudden multi-way video conferencing) and device status (e.g., performance fluctuations caused by high terminal load). This often results in critical services being stalled due to insufficient resource preemption or non-critical services occupying redundant resources.

[0018] (2) Insufficient service scheduling accuracy based on shallow network features: Traditional QoS (Quality of Service) mechanisms rely solely on protocol types, such as HTTP, VoIP, and shallow features such as port numbers, to identify service needs. They do not integrate real-time user / device behavior data, such as traffic peaks, delay sensitivity, historical habit data, such as high-frequency service usage periods, and device attribute data, such as load status and device role. This leads to one-sided utilization of behavior data. For example, the actual resource requirements of the same video conferencing service on a high-load PC and a low-load mobile phone vary greatly, but existing technologies are unable to dynamically adjust priorities based on the real-time status of the device, resulting in insufficient scheduling accuracy.

[0019] (3) Lack of dynamic adaptability: With the deepening of enterprise digital transformation, business forms have become increasingly diversified, and different types of business collaboration such as real-time interaction, batch processing, and critical tasks coexist. The complexity of IoT devices has also increased significantly. Various types of devices, such as sensors, cameras, and servers, have highly differentiated resource requirements. However, most existing technical systems still rely on manually pre-set rules or adopt simple queue management strategies such as WRR weighted round-robin scheduling, and generally lack intelligent decision-making mechanisms driven by real-time data. When user behavior shows significant time-based characteristics, such as peak office software usage, or sudden scenarios, such as temporary remote collaboration needs, static resource allocation strategies are difficult to respond quickly and adaptively, resulting in a serious mismatch between resource allocation and actual business needs.

[0020] Therefore, the following embodiments provide a method, system, medium and equipment for resource urgency assessment and dynamic scheduling based on behavioral characteristics, which comprehensively collects multi-protocol, device status and historical behavior data, identifies and eliminates abnormal data, and uses an improved algorithm to process and normalize it. With the help of a specific network architecture, the spatiotemporal characteristics of the pre-processed data are extracted, and the extended multi-head attention mechanism is used to capture long-distance dependencies. The gated attention fusion mechanism is weighted to achieve deep behavioral feature mining; based on the said characteristics, the urgency is quantitatively assessed from the three dimensions of business, equipment and environment. Based on the evaluation results, a scheduling strategy is dynamically generated, and the traffic scheduling engine is quickly sent to the execution unit to adapt the platform to manage and execute heterogeneous devices. Finally, the resource utilization rate and SLA compliance rate are monitored in real time. By adjusting the pre-processing parameters, a closed-loop optimization is formed to ensure the efficient and stable operation of the system and achieve precise and adaptive resource allocation.

[0021] Example 1 like Figure 1 As shown, this embodiment discloses a resource urgency assessment and dynamic scheduling method based on behavioral characteristics, including the following steps: S1: Collect device protocol feature data, device resource status data, and user historical behavior data, and preprocess them to obtain a normalized data set; S2: By integrating a convolutional neural network, a bidirectional gated recurrent unit, and an attention mechanism, the spatiotemporal features and long-term and short-term dependencies of the normalized dataset are extracted to generate deep behavioral features. S3: Based on deep behavioral characteristics, the fuzzy hierarchical analysis method is used to dynamically calculate the task urgency weight from three dimensions: business priority, equipment performance baseline, and environmental health, and generate a comprehensive urgency score; S4: generating a dynamic scheduling strategy and executing resource allocation based on the comprehensive urgency score; S5: Monitor resource utilization and service level agreement compliance, and provide feedback for optimizing subsequent scheduling strategies.

[0022] Next, combine Figure 2 , a resource urgency assessment and dynamic scheduling method based on behavioral characteristics disclosed in this embodiment is described in detail.

[0023] (1) Data collection like Figure 3 As shown, unlike the traditional simple packet capture solution that only collects OSI layer 2-4 protocol features, this embodiment innovatively constructs a three-dimensional collection system of "deep protocol analysis + device status perception + behavior pattern recording", covering OSI layer 2-7 protocol features, device physical / virtual resource status, user historical behavior trajectory and other multimodal data sources.

[0024] 1. In-depth analysis of the protocol (1) For the network layer (OSI layer 3), in addition to parsing common IP network configuration information such as IP addresses and subnet masks, this embodiment also deeply mines the Type of Service (ToS) field of IP packets through binary field parsing technology driven by protocol specifications.

[0025] For example, according to the IP protocol standard (RFC 791), the ToS field is located in the second byte of the IP header and occupies 8 bits. The last 4 bits contain flags for delay, throughput, and reliability. By extracting and analyzing these flags, we can accurately determine the service's network quality of service requirements: if the delay flag (bit 5, corresponding to binary value 0x10) is set to 1, it is determined to be a voice or video service with high real-time requirements; if all flags are at the default value, it is determined to be a standard data transmission service.

[0026] (2) At the transport layer (OSI layer 4), in addition to identifying the TCP and UDP protocols and their corresponding port numbers, this embodiment can also analyze in detail the congestion control algorithm used by TCP connections and explore the dynamic changes in the window size based on this, thereby providing rich information for network traffic analysis.

[0027] (3) At the session layer (OSI layer 5), this embodiment extracts the characteristics of the entire life cycle of a session connection, including the key elements of the establishment, maintenance, and termination stages.

[0028] Specifically, by building an address mapping parsing model and combining static analysis with dynamic tracking, we can deconstruct the mapping mechanism between session addresses and transport addresses: First, an initial mapping table is established based on the five-tuple (source IP, destination IP, source port, destination port, protocol), and the flow identifier (Flow ID) is used to dynamically associate the session with the transport layer connection. Secondly, the Bayesian network algorithm is introduced to predict the dynamic change trend of address mapping, effectively reducing the risk of service interruption caused by address remapping.

[0029] In terms of transport quality of service (QoS) parameter configuration, we establish a correlation matrix between QoS parameters and address mappings, employ the analytic hierarchy process (AHP) to prioritize core parameters such as bandwidth, latency, and packet loss rate, and incorporate a reinforcement learning algorithm to achieve adaptive parameter adjustment, ultimately developing a configuration solution that balances reliability and efficiency. We also precisely master the technical details of session parameter negotiation.

[0030] In the data transmission link, the focus is on monitoring the deployment strategy of synchronization points and their dynamic usage. This mechanism can achieve fault location and accurate retransmission when transmission anomalies occur, providing a key basis for ensuring business continuity and reliable data transmission.

[0031] In addition, for the session layer token management system, the system will sort out the token holder identity, transmission protocol and permission control rules to build a priority determination and timing management model for business operations.

[0032] (4) For the presentation layer (OSI layer 6), use a protocol analyzer (Wireshark) to capture network data packets and parse the protocol header information, identifying the general format type that the data is converted into before transmission through specific fields. Use a log parsing engine and regular expression matching technology to deeply mine application logs: First, regular expressions are used to identify log entries containing data format conversion keywords. A log parsing engine then performs structured processing on matching entries, extracting metadata such as timestamps, source and target formats, and process IDs involved in the conversion. Finally, the system combines business process sequence diagrams to analyze the position and relationships of different format conversion operations within the business chain, accurately determining the data types and processing requirements involved (for example, conversion from ASCII to EBCDIC or between image file formats). Traffic monitoring devices are deployed along network transmission paths to analyze data traffic characteristics and identify the currently used data compression algorithm by combining signatures from common compression algorithms (e.g., specific byte sequences in gzip and ZIP). Cryptographic analysis tools are used to perform in-depth scans of transmitted data, detecting encryption protocol identifiers (e.g., SSL / TLS handshake information) to determine the encryption algorithm in use. Statistical analysis of encrypted data, combined with publicly available encryption algorithm feature libraries, infers key information such as key length to determine the security level of the data during transmission. This allows for the extraction of features such as data formatting, encryption and decryption, and compression and decompression.

[0033] (5) Application layer (OSI layer 7), which can perform in-depth analysis of various protocols such as HTTP, FTP, and SMTP. For example, it can analyze the request information in the HTTP protocol, including the request method (GET, POST, etc.), URL path, user agent information, etc., to accurately identify the business type and user operation behavior. In the process of deep protocol analysis, facing the increasingly common encryption protocols, this embodiment uses the entropy analysis method of Deep Packet Inspection (DPI) for accurate identification. The core principle of the entropy analysis method is based on the information entropy theory, and the information entropy can be calculated by the following formula: ; in, represents the random variable of the packet content, It's an event The probability of occurrence, is the number of all possible events.

[0034] For encryption protocols, the entropy characteristics of their data packets are significantly different from those of ordinary protocols. Since encrypted data packets are processed by complex encryption algorithms, the data presents a high degree of randomness, which is reflected in the relatively high entropy value. Ordinary protocol data packets have certain regularities in terms of format, content, etc., and the entropy value is relatively low. When performing deep packet inspection, this embodiment extracts the payload portion of each captured data packet as the analysis object. The entropy value of this portion of data is calculated and compared with the pre-set entropy value thresholds of different protocols. For example, when the entropy value of a data packet is detected to be higher than the upper limit of the entropy value of a conventional HTTP protocol data packet and is within the entropy value range of common encryption protocols (such as TLS), combined with other auxiliary features (such as port number, connection handshake information, etc.), it can be preliminarily determined that the data packet may use an encryption protocol.

[0035] 2. Device status awareness In the device status perception dimension, the physical resources of the device are monitored using built-in sensors and system monitoring tools to collect CPU operating status parameters in real time, such as CPU usage, temperature, core frequency, and other parameters, to comprehensively evaluate the CPU load and operating health status.

[0036] For memory resources, monitor memory resource usage indicators, such as total memory, used memory, memory swap frequency, etc., to timely understand memory usage efficiency and potential bottlenecks.

[0037] In terms of virtual resources, if the device runs a virtualized environment, the module can perceive the virtualization environment configuration and status information, such as the number of virtual machines, the resources allocated to each virtual machine (number of CPU cores, memory size), and the operating status of the virtual machine (started, paused, running).

[0038] For network devices, you can obtain network device connection and transmission indicators, such as port speed, link status, number of data packets sent and received, and other information to fully understand the network connection and transmission status of the device.

[0039] 3. Behavioral pattern recording Behavioral pattern recording focuses on collecting historical user behavior trajectories. By deploying a lightweight behavior recording agent on the device side, the user's operational behavior is continuously tracked, such as the time and frequency of users launching applications on the device, the creation, reading, modification, and deletion of files, as well as browsing history and data download volume in network applications. These behavioral data are stored in time series to form a long-term user behavior pattern database. At the same time, for device use in multi-person collaboration scenarios, it can also record interactive behaviors between users, such as file sharing operations and collaborative editing records, to provide data support for analyzing team collaboration patterns and resource requirements.

[0040] In this embodiment, data acquisition serves as the core of the system's perception layer, responsible for capturing multi-dimensional, heterogeneous data in real time. This innovative three-dimensional acquisition system enables deep protocol analysis to capture rich protocol features, device status perception to capture the physical and virtual resource status of devices, and behavioral pattern recording to track user historical behavior. This collected data can reflect changes in device and user status in real time, breaking the limitations of static policies. It overcomes the drawbacks of relying on shallow network features, improves scheduling accuracy, and provides a comprehensive, accurate, and rich raw data foundation for subsequent data processing, feature extraction, and resource scheduling decisions, effectively ensuring that the entire system can make intelligent and efficient resource allocation strategies based on multi-dimensional information.

[0041] (2) Data preprocessing like Figure 4 As shown in the figure, the collected data enters the data cleaning phase, and performs outlier detection and processing and duplicate data deletion operations in sequence to remove noise and redundant information in the data.

[0042] To handle outliers, we use the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to identify outliers in the data. The DBSCAN algorithm divides the data space into core points, boundary points, and noise points. We determine outliers by calculating the neighborhood density of a data point using the following formula: ; in, Indicates the neighborhood radius, which is set according to data distribution (for example, in network traffic scenarios, it can be set to 3 times the standard deviation of historical data). 、 represents adjacent data points, Represents a distance function. For example, for network traffic data, if the density of data points in a particular neighborhood is significantly lower than the normal density of data points, the data point is considered an outlier. For example, if network traffic typically fluctuates between 10 and 100 Mbps over a certain period of time, and a data point with a speed of 1000 Mbps appears, the DBSCAN algorithm will likely identify it as an outlier. Detected outliers are then processed using robust statistical methods, such as using median substitution.

[0043] A hash algorithm is used to quickly compare duplicate records within the large amount of collected data. Key attributes of each piece of data (such as the source IP, destination IP, source port, destination port, and protocol type combination in device protocol feature data; the device ID and timestamp combination in device resource status data; and the user operation type, operation object, and operation time combination in behavior pattern record data) are hashed to generate a unique hash value. A hash table is then established to identify and remove duplicate data.

[0044] For example, in the behavior pattern record data, if there are two user file operation records, both of which are "User A performed a read operation on the file 'report.xlsx' at 14:00 on a certain day of a certain month in 2025", and their key attributes (operation type = read, operation object = file 'report.xlsx', operation time = 2025-XX-XX 14:00:00) are exactly the same, and the hash values ​​calculated by the hash algorithm are the same, it can be determined that they are duplicate data and one of them can be deleted.

[0045] After completing the data cleaning, we enter the data conversion stage. First, we use the Z-score normalization method to normalize the data so that the data with different characteristics have the same scale and eliminate the dimensionality effect. The formula is: ,in is the original data, is the mean of the data, The standard deviation of the data. For example, network bandwidth data may range from a few Mbps to several hundred Mbps, while device memory usage data may range from 0-100%. After Z-score normalization, both values ​​are in the same scale. Categorical data is converted to numerical form using one-hot encoding or embedded encoding techniques.

[0046] Since convolutional neural networks (CNNs) will be used for feature mining later, and CNNs require fixed-dimensional tensor input, the final stage of data preprocessing requires the unified conversion of structured, semi-structured, and unstructured data into multidimensional tensors. For structured data, numerical features are mapped to the [0, 1] range using Min-Max or Z-Score normalization, and categorical features are encoded to generate fixed-length one-dimensional vectors. After parsing the fields of semi-structured data (such as network protocols), the numerical features are normalized and categorically encoded, while variable-length fields are flattened after truncation / padding and word vector encoding. Unstructured data (such as user behavior text) is converted to fixed-dimensional vectors through word segmentation, padding / truncation, and word vector models. Finally, the feature vectors of the three types of data are sequentially concatenated and reshaped into a tensor format of [batch size, total feature dimensions, 1]. Feature concatenation or channel fusion is used to construct a single input matrix to meet the CNN input requirements.

[0047] (3) Feature mining After the above data preprocessing, a regular data set that meets the CNN input requirements is formed. Figure 5 As shown, the data is connected to the deep analysis unit in a unified tensor format. The high-dimensional features are first nonlinearly reduced in dimension using KPCA (kernel principal component analysis) technology. The original data is mapped to the high-dimensional Hilbert space using the radial basis kernel function (RBF). The key feature variance is retained through kernel matrix feature decomposition (such as retaining 92% of the device status features and business behavior correlation information). The reduced features (such as from 18 dimensions to 8 dimensions) are then reconstructed into a tensor format suitable for CNN input.

[0048] On this basis, the data enters the convolutional neural network (CNN) for feature mining. CNN uses multiple convolution kernels of different sizes to slide convolution on the matrix and extract local features from the data after KPCA dimensionality reduction. For the input two-dimensional data matrix ( is the number of time steps, is the number of features), convolution kernel , the convolution operation formula is: ; in, is the position in the output feature map ( ), is the value at the corresponding position in the input matrix, is the value at the corresponding position in the convolution kernel. When the features after KPCA dimensionality reduction are processed by CNN, they can not only preserve the complex relationship between device status and business requirements (such as the power-law relationship between CPU load and bandwidth requirements) through nonlinear mapping, but also reduce the dimensional complexity of the convolution operation (reducing the computational effort by approximately 60%).

[0049] The feature map output from CNN enters the convolutional block attention module (CBAM). CBAM first performs attention calculation on the channel dimension. ( is the number of channels, is the height, is width) for global average pooling and global max pooling Operation, we get two different channel description vectors: The formula for global average pooling (GAP) is This formula represents the total spatial position of each channel (a total of The arithmetic mean operation is performed on the eigenvalues ​​of points) to obtain a length of A vector of . Among them, :,i,j represents the Row, No. The channel eigenvalues ​​at the column position, summed and divided by the total number of spatial dimensions , and finally each element corresponds to the global average response value of a channel.

[0050] The formula for global maximum pooling (GMP) is This formula means taking the maximum value of all spatial positions of each channel, and obtaining a length of The max operation traverses each channel in For all eigenvalues ​​in the spatial dimension, the strongest response value of each channel is extracted as a vector element.

[0051] These two vectors are passed through a multi-layer perceptron (MLP) respectively, and then added and activated by the Sigmoid function to obtain the attention weight of each channel. Let the MLP be , then the channel attention weight for: ,in is the Sigmoid activation function.

[0052] In the spatial dimension, CBAM also performs average pooling and maximum pooling on the feature map, concatenates the two spatial feature maps, passes them through a convolution layer, and then passes through the Sigmoid activation function to generate a spatial attention weight map.

[0053] The CBAM-optimized feature maps are reshaped into a sequence suitable for Bidirectional Gated Recurrent Unit (BiGRU) input. Assuming the feature map size is [batch_size, channels, height, width], it is reshaped to [batch_size, sequence_length, features], where sequence_length can be height * width (expanding the feature map by rows or columns), and features is the number of channels. The BiGRU can process both forward and backward information in the sequence.

[0054] In BiGRU, for the input sequence , the update gate of the forward GRU and reset gate The calculation is as follows: ; ; in, 、 、 、 represents the weight matrix, 、 represents the bias term, is the activation function, is the hidden state of the forward GRU at time t-1.

[0055] Candidate hidden states for: ; Forward hidden state for: ; The backward GRU calculation process is similar. The hidden state of the final output is the concatenation of the forward and backward hidden states: ; When processing a sequence of user actions, such as a series of steps a user takes within a software application, the forward GRU learns information from the start of the action to the current step, while the backward GRU learns information from the end of the action to the current step. The update gate and reset gate within the GRU dynamically control the transfer and forgetting of information.

[0056] After BiGRU processing, the hidden state vector enters the extended Self-Attention layer. Building on the standard Self-Attention mechanism, this unit increases the number of layers and employs various variations. Taking multi-layer Self-Attention as an example, the first Self-Attention layer calculates the degree of correlation between each vector in the input vector sequence and other vectors, generating preliminary attention features.

[0057] For the input sequence ( is the sequence length, is the vector dimension), attention score Calculated as: ; Attention Output for: ; Features output from the Self-Attention layer With the features output from BiGRU Let’s enter the gated attention fusion mechanism. The gate unit calculates a gate value through a linear transformation and Sigmoid activation function. Let the linear transformation matrix be , the bias is , then the gate value for: ; This gate value determines the fusion ratio of BiGRU features and Self-Attention features. for: ; The final fusion feature is input into the classifier or regression model for behavior analysis. In the behavior pattern recognition task, such as determining whether the network traffic behavior is an attack behavior, the classifier (such as Softmax classifier) ​​classifies it into a predefined behavior category based on the pattern of the fusion feature. , the Softmax classifier calculates the probability of each category for: ; in, It is The class weight vector, is the bias, is the total number of categories.

[0058] In behavior prediction tasks, such as predicting a user's next action, the regression model outputs predictions of future behavior based on historical behavioral patterns learned from fused features. These analysis results serve as the output of the deep analysis unit, providing data support for the subsequent urgency assessment module and helping the system make further decisions.

[0059] (IV) Urgency Assessment like Figure 6 As shown, the data from the in-depth analysis unit (the third module) is deeply analyzed. Starting from the three key dimensions of business, equipment, and environment, the fuzzy analytic hierarchy process (FAHP) is used to perform dynamic weight calculation to achieve accurate quantification of the urgency of tasks or events, providing a scientific and reliable basis for subsequent resource scheduling and decision-making.

[0060] 1. Business Dimension: This study constructs a task urgency assessment framework, considering three dimensions: business priority, timeliness, and disruption impact. Business priority is graded based on its strategic importance and alignment with organizational goals; timeliness is quantified by calculating the time difference between the task deadline and the current time; and disruption impact is quantified in terms of economic losses, reputational damage, and other factors. By comprehensively considering these three dimensions, the urgency of business-related tasks can be more accurately determined.

[0061] 2. Equipment dimension: This focuses on equipment performance, failure risk, and business criticality. Performance indicators such as CPU utilization and memory usage are scored based on the degree to which they deviate from normal ranges. Failure risk is assessed by combining equipment failure history with real-time monitoring. The criticality of the equipment is determined based on its necessity to business operations, thereby assessing task urgency.

[0062] 3. Environmental Dimension: This analysis is based on network stability, external security threats, and regulatory requirements. Stability is assessed based on fluctuations in metrics like bandwidth and latency. External threats are measured by the severity and frequency of security incidents. The impact of environmental factors on mission urgency is determined based on the urgency of business compliance with regulations and policies.

[0063] Based on the evaluation scores of each dimension and the corresponding weights, the comprehensive urgency score of the task or event is calculated. The calculation formula is: ; in In this formula, Represents the urgency weight obtained from the final comprehensive evaluation. They are the weight coefficients of business dimension, equipment dimension, and environment dimension respectively. Their sum is 1, indicating that the three dimensions together constitute a complete urgency assessment system. It is the weight obtained by evaluating the six QoE indicators mapped to the business dimension according to the ITU-TY.1541 standard. The device dimension is the weight obtained by classifying it based on RFC7228 and combining it with the hardware performance baseline modeling. The environmental dimension calculates the weights of 12 network health indicators in real time. This improved dynamic weighting formula can flexibly adjust the proportion of each dimension in the urgency assessment based on actual conditions, comprehensively considering multiple factors such as business, equipment, and environment, thereby more accurately assessing urgency.

[0064] (V) Hierarchical Scheduling Architecture like Figure 7 As shown, it plays a core role in scheduling and resource allocation throughout the system. Based on the task or event urgency scores output by the urgency assessment module, it achieves dynamic and efficient scheduling of system resources through the collaborative work of three key components: the policy control center, the traffic scheduling engine, and the device adaptation platform. This ensures that services can operate stably with appropriate resource support, improving overall system performance and responsiveness.

[0065] Policy Control Center: The Policy Control Center receives task urgency scores from the Urgency Assessment Module and, through monitoring tools such as Prometheus, collects real-time status information on various system resources, including device CPU usage, remaining memory, disk I / O queue depth, and network bandwidth utilization. To more accurately analyze the matching of resources and tasks, a reinforcement learning algorithm is introduced to build a resource scheduling policy model with the optimization goals of maximizing resource utilization and minimizing task completion time. This model is trained and optimized based on historical scheduling data to adapt to different business scenarios and resource conditions.

[0066] When formulating scheduling policies, a hierarchical scheduling strategy is adopted. For high-urgency, compute-intensive tasks, a genetic algorithm is prioritized for task allocation, finding the optimal allocation solution within a high-performance server cluster and assigning them to CPU cores and cache memory supported by Hyper-Threading technology. For I / O-intensive tasks, a simulated annealing algorithm is used to allocate them to devices with superior disk read / write performance. Furthermore, the Policy Control Center implements a distributed locking mechanism through Zookeeper to ensure the consistency and correctness of scheduling policies in a multi-node environment.

[0067] The Policy Control Center also continuously monitors the system's operational status and task execution. When new tasks emerge, a dynamic programming algorithm is used to update the scheduling strategy in real time based on the task's urgency, resource requirements, and current system load. When resource status changes, such as a device failure, a failover strategy is immediately activated, using a heartbeat detection mechanism to quickly identify the faulty node and migrating the affected tasks to healthy devices using a pre-established resource redundancy pool. If the urgency of a task changes, resource allocation is reassessed based on the new urgency, ensuring that high-priority tasks are processed promptly.

[0068] Traffic Scheduling: The traffic scheduling engine receives resource scheduling policies generated by the policy control center. It analyzes the policies in detail to understand the task and resource allocation information contained in them. Using efficient communication protocols and data transmission methods, it quickly distributes the parsed policies to the device adaptation platform. It monitors the status of policy distribution in real time to ensure that each device accurately receives the corresponding scheduling policies. If a device fails to receive the policy, it retransmits the policy and records the information related to the distribution failure for subsequent analysis and processing.

[0069] Device Adaptation Platform: The device adaptation platform receives scheduling policies issued by the traffic scheduling engine. It verifies the policies, checking their integrity and legitimacy to ensure they comply with the device's capabilities and constraints. Based on the verified policies, it dynamically configures device resources, for example, by adjusting the device's CPU scheduling parameters and allocating memory space. It then initiates the corresponding tasks on the device and monitors their execution status in real time. It then reports task execution results (such as task completion status and resource usage) to the policy control center. Based on this feedback, the policy control center further optimizes subsequent scheduling policies.

[0070] (6) Real-time monitoring and feedback like Figure 8As shown, real-time monitoring of CPU and memory usage for servers, virtual machines, and other devices is performed. Data is collected at regular intervals (e.g., 5 minutes). Using the system's built-in performance monitoring tools or installed monitoring agent software, the system obtains information such as the number of CPU cores occupied and the amount of used memory. The utilization rate is then calculated. If a server's CPU utilization consistently exceeds 80%, it indicates that the server's computing resources are limited.

[0071] Monitor the available space, used space, and I / O read and write rates of storage devices (such as hard disks and solid-state drives). Obtain storage device capacity information through the file system's interface. Use I / O monitoring tools to record the number of read and write operations and the amount of data, and calculate the I / O rate. When the available space on a storage device falls below a certain threshold (such as 10%), issue a warning indicating insufficient storage resources.

[0072] Monitor network bandwidth utilization, latency, and packet loss. Use network monitoring equipment or software to collect real-time network traffic data and calculate bandwidth usage. Measure network latency by sending test packets and recording the round-trip time. Count lost packets to calculate the packet loss rate. If network bandwidth utilization exceeds 70% for a long period of time, or if network latency suddenly increases significantly, normal business operations may be impacted.

[0073] Monitor the response time of various services provided by the system to ensure they meet the SLA standards. Measure the volume of business processed by the system per unit time to determine whether the throughput requirements specified in the SLA are met. Integrate the collected resource utilization data and SLA compliance data to form a comprehensive data set on system operation status.

[0074] Statistical analysis, machine learning, and other methods are used to detect anomalies in the integrated data. Reasonable thresholds are set to determine whether the data is abnormal. For example, if CPU utilization exceeds 90% for 10 consecutive minutes, or if service response time suddenly increases by more than 50%, these are marked as abnormal. Machine learning algorithms (such as the Isolation Forest algorithm) are also used to identify outliers in the data and detect potential abnormal behavior.

[0075] Based on the monitoring and analysis results, a system operation status assessment report is generated.

[0076] The evaluation report is promptly fed back to the policy control center in the hierarchical scheduling architecture module. Data is transmitted through message queues and API interfaces to ensure the accuracy and timeliness of feedback information.

[0077] The policy control center adjusts existing scheduling policies based on this feedback. If resource utilization is too high, it considers reallocating tasks and migrating some to devices with idle resources. If SLAs are not met, it optimizes service processes, increases resource input, or adjusts resource allocation ratios.

[0078] In terms of data processing, this embodiment collects multiple types of data and performs comprehensive preprocessing to convert heterogeneous data into a form that can be processed by the model, laying a solid foundation for subsequent analysis. In terms of feature extraction, the integration of multiple network structures can accurately extract spatiotemporal features and long-term and short-term dependencies, and deeply explore behavioral characteristics. During the evaluation process, the fuzzy hierarchical analysis method is used to dynamically calculate the task urgency weights from multiple dimensions to ensure the scientific nature of the scoring. During dynamic scheduling, resource allocation is optimized through reinforcement learning algorithms, high-urgency tasks are given priority, and a distributed lock mechanism is used to ensure scheduling consistency. In the feedback optimization link, resource utilization and service level agreement compliance rate are continuously improved through task migration, anomaly identification and process adjustment. The method realizes intelligent, dynamic and efficient scheduling of resources, which can effectively improve system performance and service quality, and adapt to complex and changing business scenario requirements.

[0079] Example 2 This embodiment provides a resource urgency assessment and dynamic scheduling system based on behavioral characteristics, including: The data collection and processing module is used to collect device protocol feature data, device resource status data, and user historical behavior data, and perform preprocessing to obtain a normalized data set; A feature extraction module is used to extract the spatiotemporal features and long-term and short-term dependencies of the normalized dataset by fusing a network model with a convolutional neural network, a bidirectional gated recurrent unit, and an attention mechanism, thereby generating deep behavioral features; The scoring module is used to dynamically calculate the task urgency weight from three dimensions: business priority, equipment performance baseline, and environmental health, based on deep behavioral characteristics and using fuzzy hierarchical analysis method to generate a comprehensive urgency score; a scheduling module, configured to generate a dynamic scheduling strategy and perform resource allocation based on the comprehensive urgency score; The optimization module is used to monitor resource utilization and service level agreement compliance, and provide feedback for optimizing subsequent scheduling strategies.

[0080] Example 3 This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the method for resource urgency assessment and dynamic scheduling based on behavioral characteristics as described in the first embodiment above are implemented.

[0081] Example 4 This embodiment provides a computer 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 steps of the method for resource urgency assessment and dynamic scheduling based on behavioral characteristics as described in the first embodiment above are implemented.

[0082] The steps or modules involved in Examples 2 to 4 above correspond to those in Example 1. For detailed implementations, please refer to the relevant description of Example 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media that includes one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to perform any method of the present invention.

[0083] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A resource urgency assessment and dynamic scheduling method based on behavioral characteristics, characterized in that: include: Collect device protocol feature data, device resource status data, and user historical behavior data, and pre-process them to obtain a normalized data set; By integrating a convolutional neural network, a bidirectional gated recurrent unit, and an attention mechanism network model, the spatiotemporal features and long-term and short-term dependencies of the normalized dataset are extracted to generate deep behavioral features; Based on deep behavioral characteristics, the fuzzy hierarchical analysis method is used to dynamically calculate the task urgency weight from three dimensions: business priority, equipment performance baseline, and environmental health, and generate a comprehensive urgency score; generating a dynamic scheduling strategy and executing resource allocation based on the comprehensive urgency score; Monitor resource utilization and service level agreement compliance, and provide feedback to optimize subsequent scheduling strategies.

2. The method for resource urgency assessment and dynamic scheduling based on behavioral characteristics according to claim 1, characterized in that: The device protocol feature data includes OSI layer 2-7 protocol parsing results, including IP network configuration information, data packet service type field, TCP window dynamic changes, HTTP request information, and encryption protocol identification results, where the encryption protocol is identified using entropy analysis based on deep packet inspection; The device resource status data includes CPU operating status parameters, memory resource usage indicators, virtualization environment configuration and status information, and network device connection and transmission indicators; The user historical behavior data includes application startup time and frequency, file operation records, network browsing history, data download volume and multi-person collaborative interaction behavior stored in time series.

3. The method for resource urgency assessment and dynamic scheduling based on behavioral characteristics according to claim 1, characterized in that: The preprocessing includes outlier replacement, duplicate data removal, standardization, format conversion and heterogeneous data input preprocessing; The heterogeneous data input preprocessing includes: For semi-structured device protocol feature data, after parsing and encoding, it is integrated into the input data structure according to preset rules; For structured device resource status data, map it to the corresponding input node; For unstructured user historical behavior data, it is converted into numerical vector form through natural language processing and then adapted to the input dimension of the model.

4. The method for resource urgency assessment and dynamic scheduling based on behavioral characteristics according to claim 1, characterized in that: The convolutional neural network is used to perform channel and spatial dimension weighting on the feature maps extracted by CNN; The attention mechanism is used to capture the global dependencies of feature sequences; The bidirectional gated recurrent unit is used to dynamically adjust the fusion ratio of the attention mechanism output and the self-attention feature.

5. The method for resource urgency assessment and dynamic scheduling based on behavioral characteristics according to claim 1, characterized in that: The dynamic calculation of task urgency weight includes: Classify service delay sensitivity and packet loss tolerance levels based on preset standards; Set equipment performance baseline weights based on preset equipment classification standards; The network bandwidth utilization and CPU load rate are calculated in real time through the sliding window algorithm to generate environmental health indicators.

6. The method for resource urgency assessment and dynamic scheduling based on behavioral characteristics according to claim 1, characterized in that: The generation of the dynamic scheduling strategy includes: Use reinforcement learning algorithms to optimize resource allocation goals, maximize resource utilization and minimize task delays; Prioritize the allocation of hyperthreaded CPU cores and cache resources to tasks with urgency higher than a preset value; The consistency of multi-node scheduling strategies is ensured through a distributed locking mechanism.

7. The method for resource urgency assessment and dynamic scheduling based on behavioral characteristics according to claim 1, characterized in that: The feedback optimization includes: When the CPU usage continuously exceeds the threshold, task migration is triggered; Identify resource allocation anomalies based on the isolation forest algorithm; Dynamically adjust service processes to improve SLA compliance.

8. A resource urgency assessment and dynamic scheduling system based on behavioral characteristics, characterized in that: include: The data collection and processing module is used to collect device protocol feature data, device resource status data, and user historical behavior data, and perform preprocessing to obtain a normalized data set; A feature extraction module is used to extract the spatiotemporal features and long-term and short-term dependencies of the normalized dataset by fusing a network model with a convolutional neural network, a bidirectional gated recurrent unit, and an attention mechanism, thereby generating deep behavioral features; The scoring module is used to dynamically calculate the task urgency weight from three dimensions: business priority, equipment performance baseline, and environmental health, based on deep behavioral characteristics and using fuzzy hierarchical analysis method to generate a comprehensive urgency score; a scheduling module, configured to generate a dynamic scheduling strategy and perform resource allocation based on the comprehensive urgency score; The optimization module is used to monitor resource utilization and service level agreement compliance, and provide feedback for optimizing subsequent scheduling strategies.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for resource urgency assessment and dynamic scheduling based on behavioral characteristics as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising 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 steps of the resource urgency assessment and dynamic scheduling method based on behavioral characteristics as described in any one of claims 1 to 7 are implemented.

Images