Financial operation and maintenance management method and system based on deep learning
By constructing a financial operation and maintenance health model and combining deep learning and data analysis, the shortcomings of traditional methods in processor selection and fault identification in large-scale complex systems are solved. This enables intelligent matching of tasks and servers and resource optimization, thereby improving system stability and fault diagnosis accuracy.
Patent Information
- Application Number
- CN202511487852.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-17
- Publication Date
- 2026-02-27
AI Technical Summary
Traditional root cause analysis methods based on predefined rules or experience-based judgments are difficult to handle new or unknown faults in large-scale complex systems. Deep learning models are limited by data quality and computing resources in financial data operation and maintenance. Processor selection and anomaly identification lack automation, and dynamic allocation and priority processing cannot handle a large number of tasks in a timely manner.
A financial operations and maintenance health model is constructed. By analyzing financial operations and maintenance data through deep learning, the operation and maintenance capability value and health value are calculated. The priority of tasks is dynamically calculated by combining time pressure and security level to achieve cross-gradient elastic scheduling. Kafka buffer and Flink backpressure mechanism are used to deal with data floods. A three-dimensional evaluation model is constructed for resource scheduling.
It achieves intelligent matching of tasks and server processing capabilities, improves system resource utilization and fault diagnosis accuracy, reduces server overload risk, ensures data security and system stability, and supports cross-gradient elastic scheduling and automated resource scheduling.
Smart Images
Figure CN121582013A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of financial operation and maintenance management, and particularly relates to a financial operation and maintenance management method and system based on deep learning. BACKGROUND
[0002] With the development of information technology, the data environment is increasingly complex, there are numerous servers, data centers, various applications and network devices, and the number of hidden faults also increases; the traditional fault root cause analysis method based on predefined rules or experience judgment will have an explosive increase in the number of rules when facing large-scale complex systems, which is difficult to maintain and difficult to cope with new or unknown fault modes; a large amount of logs and monitoring indicators are accumulated in the process of financial data operation and maintenance; the development of big data technology makes it possible to store, manage and analyze these massive data, providing a rich source of data for deep learning models; deep learning automatically learns patterns and features in data and has achieved success in image recognition and natural language processing; In the prior art, the performance of a deep learning model is highly dependent on high-quality training data; if the data has noise, missing values or inconsistency, the accuracy and generalization ability of the model will be affected; in the actual financial data operation and maintenance environment, due to the wide range of data sources, the data quality is difficult to guarantee; the training of a deep learning model requires a large amount of computing resources and time, and the model needs to be optimized to obtain the best performance; different processors have different data operation and maintenance performance, and the prior art still has defects in selecting a processor with appropriate data operation and maintenance performance for different processing tasks and automatically identifying potential abnormalities; it is unable to timely process the dynamic allocation of processors and adjust the priority processing order when there are a large number of processing tasks; Therefore, it is necessary to provide a financial operation and maintenance management method and system based on deep learning. SUMMARY
[0003] The present application aims to provide a financial operation and maintenance management method and system based on deep learning, and to solve the above-mentioned problems in the prior art, the present application realizes the following technical solutions: In a first aspect, the present application provides a financial operation and maintenance management method based on deep learning, which specifically includes the following steps: Step one: build a financial operation and maintenance data collection system, unify the representation of financial operation and maintenance data and perform fusion, and construct a financial operation and maintenance health model; Step two: calculate the operation and maintenance capability value and the operation and maintenance health value based on the financial operation and maintenance health model, and comprehensively analyze to obtain the operation and maintenance performance value to judge the server operation and maintenance performance; Step three: calculate the task difficulty value based on the analysis of the server to-be-processed tasks, analyze and compare the obtained operation and maintenance performance value and task difficulty value, and select a processor to process the to-be-processed tasks; Step four: based on the overload of processing tasks, the time efficiency pressure value and the safety level weight of the overload processing tasks are combined with the performance gradient analysis to calculate the task priority, and the server is allocated according to the task priority.
[0004] In a second aspect, the embodiment of the present application provides a financial operation and maintenance management system based on deep learning, which specifically comprises the following modules: The data preprocessing module: a financial operation and maintenance data collection system is built, financial operation and maintenance data are unified and fused, and a financial operation and maintenance health model is constructed. The data analysis module: based on the financial operation and maintenance health model, the operation and maintenance capability value and the operation and maintenance health value are calculated, the operation and maintenance performance value is obtained by comprehensive analysis to judge the server operation and maintenance performance, and the task difficulty value is calculated based on the server to-be-processed task analysis; The operation and maintenance management module: the operation and maintenance performance value and the task difficulty value are analyzed and compared, and the processor is selected to process the to-be-processed task; The optimization decision module: based on the overload of processing tasks, the time efficiency pressure value and the safety level weight of the overload processing tasks are combined with the performance gradient analysis to calculate the task priority, and the server is allocated according to the task priority.
[0005] The beneficial effects of the present application are as follows: 1. The fine financial operation and maintenance health model is constructed, the limitations of traditional single index monitoring are broken through, the bidirectional matching mechanism is proposed, the time efficiency pressure and the operation and maintenance health value are combined for secondary sorting, the intelligent matching of tasks and server processing capacity is realized, the priority is dynamically calculated based on the time efficiency pressure and the safety level for the overload task, the cross-gradient elastic scheduling is supported, the system resource utilization rate and the intelligent task processing are improved; 2. The three-dimensional evaluation model including operation and maintenance capability, health state and safety constraint is constructed, the index weight is dynamically allocated by the entropy method to avoid subjective assignment deviation; the operation and maintenance full-link data are covered, the machine learning and graph model technology are combined to accurately capture system abnormalities and improve the accuracy of fault diagnosis and health evaluation; based on the gradient matching and dynamic scheduling strategy, the task load balancing is realized, and the server overload risk is reduced; the Kafka buffer and Flink back pressure mechanism effectively cope with data flood peak to guarantee the stability of the collection and processing system; the safety level weight and compliance risk rate evaluation are used to strengthen data security and operation audit; and resource automatic scheduling is realized. BRIEF DESCRIPTION OF DRAWINGS
[0006] In order to make the technical solutions in the embodiments of the present application clearer, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.
[0007] Figure 1 is a step flow chart of a financial operation and maintenance management method based on deep learning provided by Embodiment 1 of the present application; Figure 2 is a structural schematic diagram of a financial operation and maintenance management system based on deep learning provided by Embodiment 2 of the present application. DETAILED DESCRIPTION
[0008] In order to make those skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the present application.
[0009] Embodiment 1 As shown in the figure, the financial operation and maintenance management method based on deep learning provided by the embodiments of the present application specifically includes the following steps: Figure 1 Step one: build a financial operation and maintenance data collection system, unify the representation of financial operation and maintenance data and perform fusion, and construct a financial operation and maintenance health model; In specific embodiments, the method for building the financial operation and maintenance data collection system is: Based on the OpenTelemetry standardized collection system, integrate and fuse operation and maintenance data, which includes but is not limited to log files, performance indicators, network traffic data, configuration information and user behavior data; It should be noted that OpenTelemetry is an open-source, vendor-neutral observability framework for collecting, processing and exporting telemetry data to help developers and operation and maintenance personnel better understand and monitor modern distributed systems; For log files, system logs and application logs are real-time captured through a distributed log collector and a lightweight log collector, and are collected according to log levels to reduce the amount of invalid data; For performance indicators, the Prometheus pull mode is used to collect monitoring server CPU utilization and memory utilization, database slow queries, and Grafana is used for index visualization. In the container environment, the Kubernetes API is used to obtain Pod resource consumption. It should be noted that Prometheus is an open-source system monitoring and alerting toolkit. For network traffic data, NetFlow protocol is used to collect network traffic data, and Elasticsearch is used to store traffic logs. High-frequency packet data is cleaned locally using edge computing nodes to filter irrelevant port traffic, and aggregated statistical indicators are uploaded to the central server. It should be noted that NetFlow is a network traffic monitoring technology used to collect and analyze network traffic data, and Elasticsearch is an open-source distributed search and analysis engine used to quickly store, search, and analyze massive amounts of data. For configuration information, server parameters including server model and IP address are obtained through the configuration management database API, network topology is parsed using the OSPF protocol, and a server dependency graph is constructed, which is stored in the form of an adjacency matrix. It should be noted that the Open Shortest Path First (OSPF) protocol is a link-state routing protocol that can be used to discover server connection relationships in a network. Specifically, a connection is established with the configuration management database, and appropriate API interfaces are used to ensure the stability and security of communication. Send a request to obtain server parameters to obtain server model and IP address. Use the server's IP address as the node identifier and the server model as the node's attribute. In the graph, each node represents a server. According to the connection relationship between servers in the network topology, add the corresponding edges in the graph. For example, if there is a physical or logical connection between two servers, an edge is added between the nodes representing the two servers. Set the weight of the edge according to the link bandwidth between servers. Log in to the core device in the network and view the interface configuration information through the command line interface. Traverse all interfaces to obtain the minimum bandwidth value, and mark the minimum bandwidth value as the approximate value of the minimum bandwidth in the network. The minimum bandwidth approximation value obtained is processed by ratio to obtain the weight of the edge. The dependency relationship between servers is stored in the form of an adjacency matrix. It should be noted that the adjacency matrix is a two-dimensional array, in which the rows and columns of the matrix correspond to the IP addresses of the servers; if there is a connection between server i and server j, the value at the corresponding position in the adjacency matrix is 1, otherwise it is 0, and the adjacency matrix stores the weight of the edge; For user behavior data, the software development kit (SDK) is embedded in the application layer to collect user access frequency, page load time, and error request URL, and the user IP and access path are associated with the Nginx access log; A Kafka cluster is deployed between the collection end and the processing layer, and the number of partitions is set equal to the number of collection nodes. When a data flood occurs, the Kafka buffer temporarily stores data to prevent the collection service from being overloaded. In combination with the back pressure mechanism of Flink, the data processing rate is dynamically adjusted. It should be noted that Kafka is an open source stream processing platform for processing real-time data streams and message passing in distributed systems. Flink is an open source distributed stream processing framework that performs stateful computation on unbounded and bounded data streams. When the collection node produces a data flood due to sudden traffic, the Kafka partition acts as a buffer to temporarily store excess data. Each collection node corresponds to a partition, and Kafka maintains a message queue for each partition. If the Flink processing speed cannot keep up with the production speed, messages will accumulate in the queue to prevent the collection node from being overloaded due to data backlog. The collection Agent periodically sends a heartbeat packet to the central server, and the server scans the timeout devices through a timing task to trigger reconnection; It should be noted that the collection Agent is a software component that typically runs on the server where the data source is located, and is used to collect data from various data sources and send the collected data to a data processing platform or storage system; Based on CPU load and network bandwidth, time series samples are generated based on a sliding window, and Z-Score normalization is used to eliminate dimensions, and time features are added to enhance periodicity recognition; Based on BERT, pre-training log corpus for semantic encoding converts unstructured logs into multi-dimensional vectors, extracts context-dependent features through BiLSTM, and finally generates event type labels; The network topology and microservice call chain are modeled as a directed graph, with nodes representing servers and edges representing dependency relationships, and weights representing call frequencies. GraphSAGE algorithm is used to generate node embedding multi-dimensional vectors to represent the structural importance of nodes in the graph. It should be noted that GraphSAGE is an inductive representation learning algorithm for graph data; Using timestamps as anchors, time-series features, log features, and graph features from the same time period are input into the Transformer, and weights are calculated through a self-attention mechanism to generate fused features. It should be noted that Transformer is a deep learning model architecture based on attention mechanisms, primarily used for natural language processing tasks; Based on the obtained fusion features and operation and maintenance data, a financial operation and maintenance health model is constructed using Transformer. Step 2: Calculate the operation and maintenance capability value and operation and maintenance health value based on the financial operation and maintenance health model, and comprehensively analyze the operation and maintenance performance value to judge the server operation and maintenance performance; In a specific embodiment, the method for calculating the operation and maintenance capability value and the operation and maintenance health value is as follows: Deploy a traffic monitoring agent on the server's network interface card; Collect real-time financial data according to the financial data collection cycle, and record the number of bytes sent and received within the financial data collection cycle; The number of bytes transmitted during the financial data collection period is obtained by summing the number of bytes sent and received during the financial data collection period. The local standard deviation is calculated using a preset sliding window. Peaks that significantly deviate from the normal range within the window are identified and removed to obtain the number of bytes C transmitted. zj ; Through formula The average throughput ΔTu within the financial data collection period is calculated, where N represents the number of financial data collection periods. Obtain the container expansion / shrinkage success rate, and multiply the obtained expansion / shrinkage success rate by the time consumption coefficient to obtain the processing elasticity efficiency Ee. It should be noted that the time consumption coefficient is inversely proportional to the time consumed by the container to successfully expand or shrink its capacity; the longer the container takes to successfully expand or shrink its capacity, the smaller the time consumption coefficient. Obtain the server's fault recovery time, set a standard fault recovery time, and calculate the fault recovery speed Rf by comparing the standard fault recovery time with the obtained fault recovery time. Get the combined load rate of the server's CPU and GPU, set the combined load rate range to the ideal range of [60%, 80%], and mark the computing power utilization rate Uc as 1. If the combined load rate deviates from the ideal range, the computing power utilization rate will decrease linearly. The processing capacity value Cn is obtained by weighted summing of the processing elasticity efficiency, fault recovery speed and computing power utilization. It should be noted that the multiple sample data of the elastic efficiency Ee, the fault recovery speed Rf and the computing power utilization rate Uc are normalized by using the minimum-maximum normalization method to map the data to the [0, 1] interval; the entropy value of each index is calculated, and the weight is calculated according to the entropy value; Based on the obtained average throughput and service capacity value, the formula The operation and maintenance capability value SCI is calculated, wherein Tu max Indicates the theoretical peak value of traffic transmission; The response time variance in the preset sliding window is obtained, and the reciprocal of the response time variance is calculated to obtain the performance stability value Wd, wherein the maximum value of the performance stability value is 1, and if the response time variance is less than 0.1, the performance stability value is the maximum value, that is, 1; According to the ERROR level log and the WARNING level log output by the BERT-IT model; The number of ERROR level logs and the number of WARNING level logs are obtained; The number of ERROR level logs is uniformly converted into the number of WARNING level logs by a preset ratio; The number of WARNING level logs obtained by converting the number of ERROR level logs is summed with the number of WARNING level logs output to obtain the abnormal log weight Yc; According to the log file, the number of server error requests Qw and the total number of requests Qa are obtained; Based on the calculated performance stability value and abnormal log weight, the formula The operation and maintenance health value SHI is calculated; Based on the preset sliding window, the number of high-risk operations is detected by the OPA policy engine, the ratio of the obtained number of high-risk operations to the total number of operations is processed to obtain the compliance risk rate, and the complement of the compliance risk rate with respect to 1 is obtained to obtain the safety constraint value As; It should be noted that the OPA policy engine is an open source general policy engine, which manages and decides the access and operation of various systems; Based on the obtained operation and maintenance capability value SCI, the operation and maintenance health value SHI and the safety constraint value As, the weighted sum is performed to obtain the operation and maintenance performance value SPI of the server; It should be noted that in the operation and maintenance work, the processing capacity of the server, the response speed of the system and the effective use of resources are directly related to the normal operation of the business and the experience of the customers; the health status of the system includes the failure rate of the server, the stability of the software and the connectivity of the network, which is the basis for stable operation of the system, and the system failure or instability leads to business interruption and serious loss; security is an important part of operation and maintenance work that cannot be ignored, including data security and network security, and under normal circumstances, the probability of security incidents is relatively low, and a series of security measures are usually taken to prevent and protect; according to the degree of influence on the operation and maintenance performance, the weights of the operation and maintenance capability value, the operation and maintenance health value and the security constraint value are distributed in order from large to small, and the weight size is adjusted according to the actual situation; Step three: based on the server pending task analysis and calculation task difficulty value, according to the obtained operation and maintenance performance value and task difficulty value, analysis and comparison are carried out and the processor is selected to process the pending task; In specific embodiments, the method for analyzing and calculating the task difficulty value is: For the task to be processed by the server, the number of floating point operations per second of the server is obtained; The number of floating point operations per second required by the task is obtained; Specifically, for the experience data of the number of floating point operations per second of similar tasks, the similarity between the current task and the known task is calculated; If the new task is similar to the known task in terms of algorithm structure and data size, the data dimension or parameter changes, and the known number of floating point operations per second is adjusted according to these changes to obtain the number of floating point operations per second of the new task; For complex tasks, if it is difficult to directly calculate the number of floating point operations per second, the task is decomposed into multiple subtasks with known number of floating point operations per second; The number of floating point operations per second of the whole task is calculated by accumulation; The obtained number of floating point operations per second required by the task is processed by ratio with the number of floating point operations per second of the server to obtain the operation demand ratio; The memory requirement of the task is obtained, and the memory requirement of the task is processed by ratio with the average idle memory of the node to obtain the memory demand ratio; The obtained operation demand ratio and memory demand ratio are multiplied to obtain the calculation complexity Dc; The deadline of the task is obtained, and the ratio processing is performed between the benchmark processing time of the task and the deadline of the task to obtain the time efficiency pressure value Dt; According to the transmission encryption mode of the task, the task is divided into public level, sensitive level and confidential level, and the security level of the confidential level is greater than that of the sensitive level, and the security level of the sensitive level is greater than that of the public level; The computational complexity Dc is proportionally summed with the time pressure value Dt, and a task difficulty quantification model is established by combining the task's security level to obtain the task difficulty value. It should be noted that, according to the order of security level from low to high, corresponding numerical weights are assigned to obtain the security level weights. The computational complexity Dc, the time pressure value Dt, and the security level weights are proportionally summed to obtain the task difficulty value. Obtain the operational performance values of all servers and sort the servers from largest to smallest based on their operational performance values; Obtain the task difficulty value of the tasks to be processed, and sort the tasks to be processed from largest to smallest according to the task difficulty value; Servers are assigned to handle pending tasks based on their operational performance values from highest to lowest, and the operational performance value of each server is greater than the sum of the computational complexity of the assigned pending tasks. Based on operational performance values, servers are divided into three performance tiers: the first performance tier, the second performance tier, and the third performance tier. Based on the servers within each performance tier, a secondary sorting is performed from largest to smallest according to their operational health values; Based on the task difficulty value, the tasks to be processed are divided into the first difficulty level, the second difficulty level, and the third difficulty level; Based on the tasks to be processed within each difficulty level, they are sorted in descending order of time pressure value; Match the performance gradient obtained from the partitioning with the difficulty gradient obtained from the partitioning; Specifically, tasks with the first difficulty level are matched to servers with the first performance level, tasks with the second difficulty level are matched to servers with the second performance level, and tasks with the third difficulty level are matched to servers with the third performance level. For any task to be processed within any gradient, it is matched with servers sorted from largest to smallest based on time pressure value; Step 4: Based on the fact that the processing tasks are overloaded and suitable processors cannot be allocated in time, the priority of the tasks is calculated by combining the time pressure value and security level weight of the overloaded processing tasks with performance gradient analysis, and servers are allocated according to the task priority. In a specific embodiment, the method for calculating the priority of the task to be processed is as follows: Obtain the time pressure value and security level weight of all overloaded processing tasks; For any overloaded processing task, the time pressure value is compared with the gradient value to obtain the task priority; Based on task priority, the overloaded processing tasks within the gradient are sorted from largest to smallest, and a priority list is generated. The servers are matched based on the prioritized list and the servers sorted from large to small according to the operation and maintenance health value, and meanwhile, if all the servers in the gradient are overloaded, the servers from the low-order performance gradient to the high-order performance gradient are traversed to match the servers that are not overloaded; For example, if the servers of the third performance gradient are all overloaded, the servers from the third performance gradient to the second performance gradient are traversed to match the servers that are not overloaded, and if the servers of the second performance gradient are all overloaded, the servers from the second performance gradient to the first performance gradient are traversed to match the servers that are not overloaded; The technical scheme of the embodiment of the application is as follows: real-time collection and stream processing, modeling of log semantics and network topology dependency relationship by combining algorithms, generation of fused features and construction of financial operation and maintenance health model; operation and maintenance capacity value is calculated by entropy method weighting based on traffic throughput, container expansion and contraction efficiency, fault recovery speed and computing power utilization; operation and maintenance health value is obtained based on preset proportion coefficient quantization by combining response time stability, abnormal log weight and error request rate; high-risk operation is detected by OPA policy engine, compliance risk rate is calculated and converted into safety constraint value, and finally server operation and maintenance performance value is obtained by weighting; task difficulty value is defined, and task difficulty quantification model is constructed by comprehensively considering operation demand ratio, memory demand ratio, time efficiency pressure value and safety level; servers are divided into three gradients according to operation and maintenance performance, tasks are classified according to difficulty, gradient matching is performed, and secondary sorting and matching are performed based on time efficiency pressure and operation and maintenance health value; when tasks cannot be allocated in time, task priority is calculated according to time efficiency pressure value and safety level weight, and a prioritized list is generated; if the current gradient server is overloaded, gradient traversal is performed from low to high, and the server that is not overloaded is matched to ensure efficient task processing; a refined financial operation and maintenance health model is constructed, the limitations of traditional single index monitoring are broken through, a two-way matching mechanism is proposed, secondary sorting is performed based on time efficiency pressure and operation and maintenance health value, and intelligent matching of task and server processing capacity is realized; for overloaded tasks, priority is dynamically calculated based on time efficiency pressure and safety level, cross-gradient elastic scheduling is supported, and the intelligentization of system resource utilization and task processing is improved; a three-dimensional evaluation model including operation and maintenance capacity, health state and safety constraint is constructed, index weight is dynamically allocated by entropy method to avoid subjective assignment bias; full-link operation and maintenance data are covered, machine learning and graph model technology are combined, system abnormalities are accurately captured, and the accuracy of fault diagnosis and health evaluation is improved; task load balancing is realized based on gradient matching and dynamic scheduling strategy, and server overload risk is reduced; Kafka buffer and Flink back pressure mechanism effectively cope with data flood peak, and guarantee the stability of collection and processing system; data security and operation audit are strengthened through safety level weight and compliance risk rate evaluation; resource automatic scheduling is realized.
[0010] Embodiment 2 As Figure 2As shown, the financial operation and maintenance management system based on deep learning provided by the embodiment of the present application specifically comprises the following modules: The data preprocessing module: build a financial operation and maintenance data collection system, unify the representation of financial operation and maintenance data and perform fusion, and construct a financial operation and maintenance health model; The data analysis module: calculate the operation and maintenance capability value and the operation and maintenance health value based on the financial operation and maintenance health model, comprehensively analyze to obtain the operation and maintenance performance value to judge the server operation and maintenance performance, and analyze and calculate the task difficulty value based on the server pending task; The operation and maintenance management module: analyze and compare the obtained operation and maintenance performance value and task difficulty value and select a processor to process the pending task; The optimization decision module: based on the overload processing task that cannot be timely matched with a suitable processor, according to the time efficiency pressure value and the safety level weight of the overload processing task, combined with the performance gradient analysis, the task priority is calculated, and the server is allocated according to the task priority.
[0011] The above describes one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application; the above formulas are all de-dimensioned to calculate the numerical value, the formula is obtained by collecting a large amount of data to simulate the formula of the nearest real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation and historical experience, which can be adjusted according to the actual situation; the above is only the preferred embodiment of the present application, and cannot be used to limit the present application, and all equivalent changes and improvements made according to the scope of the present application should belong to the patent coverage range of the present application.
Claims
1. A deep learning-based financial operations and maintenance management method, characterized in that, Includes the following steps: Step 1: Establish a financial operations and maintenance data collection system, uniformly represent and integrate financial operations and maintenance data, and construct a financial operations and maintenance health model; Step 2: Calculate the operation and maintenance capability value and operation and maintenance health value based on the financial operation and maintenance health model, and comprehensively analyze the operation and maintenance performance value to judge the server operation and maintenance performance; Step 3: Calculate the task difficulty value based on the server's pending tasks, analyze and compare the obtained operation and maintenance performance value with the task difficulty value, and select a processor to process the pending tasks. Step 4: If the processing tasks are overloaded and suitable processors cannot be allocated in time, calculate the task priority based on the time pressure value and security level weight of the overloaded processing tasks, combined with performance gradient analysis, and allocate servers according to the task priority.
2. The deep learning-based financial operations and maintenance management method according to claim 1, characterized in that, The method for unifying and integrating financial operations and maintenance data is as follows: Based on the OpenTelemetry standardized data acquisition system, operational data is integrated and fused. The network topology and microservice call chain are modeled as a directed graph, with nodes representing servers, edges representing dependencies, and weights representing call frequencies. The GraphSAGE algorithm is used to generate multi-dimensional vectors for node embeddings, which represent the structural importance of nodes in the graph. Using timestamps as anchors, time-series features, log features, and graph features from the same time period are input into the Transformer, and weights are calculated through a self-attention mechanism to generate fused features.
3. The deep learning-based financial operations and maintenance management method according to claim 1, characterized in that, The method for obtaining the operation and maintenance capability value is as follows: Collect real-time financial data according to the financial data collection cycle, and record the number of bytes sent and received within the financial data collection cycle; The number of bytes transmitted during the financial data collection period is obtained by summing the number of bytes sent and received during the financial data collection period. The local standard deviation is calculated using a preset sliding window. Peaks that significantly deviate from the normal range within the window are identified and removed to obtain the number of bytes C transmitted. zj ; The average throughput ΔTu during the financial data collection period is calculated using a formula. The processing capacity value Cn is obtained by weighted summing of the processing elasticity efficiency, fault recovery speed and computing power utilization. Based on the obtained average throughput and service capacity values, using the formula The operational capability value SCI is calculated, where Tu max This represents the theoretical peak value of financial transmission.
4. The deep learning-based financial operations and maintenance management method according to claim 1, characterized in that, The method for obtaining the operational health value is as follows: The number of ERROR-level logs is converted to the number of WARNING-level logs, and the sum of the number of WARNING-level logs output is obtained to get the abnormal log weight Yc. Obtain the number of erroneous requests (Qw) and the total number of requests (Qa) from the log file. Based on the calculated stable performance value and the weight of the abnormal logs, the formula is used. The operational health value SHI is calculated, where Wd represents the performance stability value.
5. The deep learning-based financial operations and maintenance management method according to claim 1, characterized in that, The method for obtaining the operation and maintenance performance value is as follows: Based on a preset sliding window, the number of high-risk operations is detected by the OPA strategy engine. The ratio of the number of high-risk operations to the total number of operations is calculated to obtain the compliance risk rate. The complement of the compliance risk rate with respect to 1 is obtained to obtain the security constraint value As. Based on the obtained operation and maintenance capability value SCI, operation and maintenance health value SHI, and security constraint value As, a weighted sum is performed to obtain the server's operation and maintenance performance value SPI.
6. The deep learning-based financial operations and maintenance management method according to claim 1, characterized in that, The method for obtaining the task difficulty value is as follows: Obtain the deadline of the task, and then calculate the ratio between the baseline processing time of the task and the deadline of the task to obtain the time pressure value Dt. The computational complexity Dc is proportionally summed with the time pressure value Dt, and a task difficulty quantification model is established by combining the task's security level to obtain the task difficulty value.
7. The deep learning-based financial operations and maintenance management method according to claim 6, characterized in that, The method for obtaining the computational complexity is as follows: Empirical data on the number of floating-point operations per second for similar tasks are used to calculate the current task based on its similarity to known tasks. The total number of floating-point operations per second for the entire task is calculated by summing them up. The computational requirement ratio is obtained by comparing the number of floating-point operations per second required for the task with the server's baseline number of floating-point operations per second. Obtain the memory requirements of the task, and then calculate the memory requirement ratio by comparing the task's memory requirements with the average free memory of the nodes. Multiply the ratio of computational requirements by the ratio of memory requirements to obtain the computational complexity.
8. The deep learning-based financial operations and maintenance management method according to claim 1, characterized in that, The method for performing analysis, comparison, and processor selection is as follows: Servers are assigned to handle pending tasks based on their operational performance values from highest to lowest, and the operational performance value of each server is greater than the sum of the computational complexity of the assigned pending tasks. Based on operational performance values, servers are divided into three performance tiers: the first performance tier, the second performance tier, and the third performance tier. Match the performance gradient obtained from the partitioning with the difficulty gradient obtained from the partitioning; For any task to be processed within any gradient, it is matched with servers sorted from largest to smallest based on timeliness pressure value.
9. The deep learning-based financial operations and maintenance management method according to claim 1, characterized in that, The method for calculating task priority is as follows: Obtain the time pressure value and security level weight of all overloaded processing tasks; For any overloaded processing task, the time pressure value is compared with the gradient value to obtain the task priority; Based on task priority, the overloaded processing tasks within the gradient are sorted from largest to smallest, and a priority sorting list is generated. The system matches servers based on a priority list and servers sorted from largest to smallest based on their operational health values. If all servers within a gradient are overloaded, the system iterates from lower-order performance gradients to higher-order performance gradients to match servers that are not overloaded.
10. A deep learning-based financial operations and maintenance management system, the system being used to execute the management method described in any one of claims 1-9, characterized in that, include: Data preprocessing module: Establish a financial operations and maintenance data collection system, uniformly represent and integrate financial operations and maintenance data, and construct a financial operations and maintenance health model; Data analysis module: Calculates operation and maintenance capability value and operation and maintenance health value based on financial operation and maintenance health model, comprehensively analyzes to obtain operation and maintenance performance value to judge server operation and maintenance performance, and calculates task difficulty value based on server pending tasks analysis; Operations and maintenance management module: Analyzes and compares the obtained operations and maintenance performance values with the task difficulty values, and selects processors to process the tasks to be processed; Optimization Decision Module: Based on the overload of processing tasks and the inability to allocate suitable processors in a timely manner, the priority of tasks is calculated by combining the time pressure value and security level weight of the tasks to be processed with performance gradient analysis, and servers are allocated according to the task priority.