Fault self-healing and resource optimization method and system for medical image SaaS platform

Abstract

The invention discloses a fault self-healing and resource optimization method and system for a medical image SaaS platform, and the method comprises the steps: obtaining multi-source observation data in the operation process of the medical image SaaS platform, and constructing a digital twinborn model of the platform; based on the digital twinborn model and multi-source observation data, performing intelligent identification and root cause positioning on an abnormal state of the platform so as to execute cognitive analysis on a platform operation problem; and executing a preset corresponding fault repair operation and a resource optimization strategy according to the result of the cognitive analysis. The full-link transparency of the medical image SaaS platform is realized by constructing the digital twinborn model, the fault recovery time is greatly shortened, and the capabilities of the platform for automatically sensing, diagnosing and repairing system abnormity and adaptively optimizing resources are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image platform service operation and maintenance, in particular to a medical image SaaS platform fault self-healing and resource optimization method and system. BACKGROUND

[0002] With the rapid development of medical informatization, the medical image SaaS platform has become a key infrastructure supporting clinical diagnosis and treatment, carrying a large number of high-concurrency image review and AI analysis requests, and its stability, performance and security directly affect the quality of medical services and patient safety. However, the current medical image SaaS platform faces severe challenges in operation and maintenance management. Due to the complexity of micro-service architecture, the internal running state of the platform is like a black box, and it is difficult for developers and operators to fully grasp the real-time running status of the system. Often, performance degradation or hidden faults are only discovered passively after user complaints, resulting in time-consuming and laborious fault location. Existing monitoring systems mostly rely on static thresholds for alarm, and such simple rules will produce a large number of invalid alarms to cause alarm storms during business peak periods, while remaining silent in the face of slow performance degradation, leaving operators in a dilemma. At the same time, the configuration of cloud resources mostly depends on the experience estimation of operators, making it difficult to accurately match the dynamically changing business load, resulting in either insufficient resources affecting service quality or resource redundancy causing cost waste. More seriously, traditional firewalls and web application firewalls are difficult to resist complex attacks on the business logic of medical SaaS platforms, such as malicious high-frequency calls to exhaust AI resources, batch fake requests for data crawling, etc. The update speed of security rules is far behind the evolution speed of attack methods, making the platform face increasingly severe security threats. These problems not only affect the availability and user experience of medical image services, but also greatly increase the operation and maintenance costs and security risks, and an integrated solution is urgently needed to achieve comprehensive system state perception, accurate abnormality identification, rapid fault location and self-healing, and intelligent resource optimization, in order to improve the reliability, efficiency and security of the medical image SaaS platform. SUMMARY

[0003] Therefore, the present application provides a medical image SaaS platform fault self-healing and resource optimization method and system, which can realize comprehensive system state perception, accurate abnormality identification, fault location and resource optimization. The present application provides the following technical solutions:

[0004] A medical image SaaS platform fault self-healing and resource optimization method, the method comprising:

[0005] Obtaining multi-source observation data in the running process of the medical image SaaS platform, and constructing a digital twin model of the platform;

[0006] Intelligently identify an abnormal state of the platform and locate a root cause based on the digital twin model and multi-source observation data to perform cognitive analysis on a platform operation problem;

[0007] Perform a preset corresponding fault repair operation and resource optimization strategy according to a result of the cognitive analysis.

[0008] Optionally, the multi-source observation data includes log data, trace data, and index data.

[0009] The digital twin model of the construction platform specifically includes:

[0010] The dispersed log data, trace data, and index data are associated by injecting a unified context identifier;

[0011] A service call relationship topology graph is constructed and updated in real time based on the associated multi-source observation data, a node of the topology graph represents a microservice, an edge represents a service call relationship, and a weight of the edge contains at least one of real-time delay, request volume, and error rate.

[0012] Optionally, the intelligently identifying the abnormal state of the platform includes:

[0013] Performance indicators KPIs in a platform operation process are obtained based on the multi-source observation data, and a high-dimensional vector is formed;

[0014] Normal operation modes in historical data of the platform are learned by an unsupervised clustering algorithm, and a centroid of the normal modes is determined;

[0015] When a distance between a high-dimensional vector corresponding to a current KPI and all known normal mode centroids exceeds a preset threshold, it is determined that the system is in an abnormal state.

[0016] Optionally, the root cause locating specifically includes:

[0017] A causal inference algorithm is applied to the service call relationship topology graph to analyze a propagation path of the abnormal state on the topology graph;

[0018] A causal influence degree of each node as a fault source on a global abnormal phenomenon is calculated, and a node with the largest causal influence degree is determined as a fault root cause.

[0019] Optionally, the performing the preset corresponding fault repair operation and resource optimization strategy according to the result of the cognitive analysis includes:

[0020] When the cognitive analysis result determines that the fault root cause is a specific microservice or component, a repair operation matched with the fault type is performed;

[0021] Based on the system state information obtained through cognitive analysis, a reinforcement learning agent is constructed through a deep Q network, and the reinforcement learning agent determines an optimal resource adjustment action according to a current system state vector;

[0022] The system state vector includes at least one of resource configuration, key performance indicators, service load and cognitive analysis results, and the resource adjustment action includes at least one of increasing or decreasing a computing instance, adjusting CPU / GPU resource configuration, and modifying a load balancing strategy.

[0023] The reinforcement learning agent adopts a joint reward function of cost and performance, gives a performance reward when an actual service level agreement (SLA) is better than a target SLA, and at the same time, imposes a penalty on the cost of the resource adjustment action, so as to maximize resource utilization efficiency.

[0024] Optionally, the method further includes:

[0025] A normal operation behavior sequence baseline based on a Markov model or a recurrent neural network is established for a platform user;

[0026] The occurrence probability of a current behavior sequence of the user is calculated in real time, and when a negative log-likelihood value of the occurrence probability exceeds a safety threshold, an adaptive security defense mechanism is triggered, the adaptive security defense mechanism including at least one of flow limiting, forced multi-factor authentication or session termination.

[0027] The application further discloses a fault self-healing and resource optimization system of a medical image SaaS platform, including:

[0028] A data acquisition and model construction module is configured to acquire multi-source observation data in a running process of the medical image SaaS platform and construct a digital twin model of the platform.

[0029] An abnormality analysis module is configured to intelligently identify and locate a root cause of an abnormal state of the platform based on the digital twin model and the multi-source observation data, so as to perform cognitive analysis on a running problem of the platform.

[0030] An execution optimization module is configured to perform a preset corresponding fault repair operation and resource optimization strategy according to a result of the cognitive analysis.

[0031] The application further discloses a computer readable storage medium, the storage medium storing a computer program, and the computer program is executed by a processor to implement the method.

[0032] The application further discloses an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the program to implement the method.

[0033] The application further discloses a computer program product comprising a computer program which, when executed by a processor, implements the method described above.

[0034] According to the technical scheme of the application, by acquiring multi-source observation data in the running process of the medical image SaaS platform and constructing a digital twin model of the platform, comprehensive perception of the system state and full-link transparency are realized, effectively solving the system black box problem under the traditional micro-service architecture; on this basis, intelligent identification and root cause positioning of abnormal states are performed based on the digital twin model and multi-source observation data, significantly improving the accuracy of abnormal detection and greatly shortening the root cause positioning time, and reducing the fault recovery time; further, according to the result of cognitive analysis, a preset corresponding fault repair operation and resource optimization strategy are performed, not only realizing automatic perception, diagnosis and repair of system abnormalities, but also reducing the cost of cloud infrastructure through intelligent resource dynamic allocation, while significantly improving service availability and resource utilization, thereby significantly reducing operation and maintenance costs and security risks while ensuring the quality of medical image services. BRIEF DESCRIPTION OF DRAWINGS

[0035] For the purpose of illustration and not limitation, the present application will now be described in conjunction with embodiments thereof and the accompanying drawings, in which:

[0036] Figure 1 is a flowchart of a fault self-healing and resource optimization method of a medical image SaaS platform in an embodiment of the application;

[0037] Figure 2 is a structural schematic diagram of a fault self-healing and resource optimization system of a medical image SaaS platform in an embodiment of the application;

[0038] Figure 3 is a structural schematic diagram of an electronic device in an embodiment of the application. DETAILED DESCRIPTION

[0039] In order to enable persons skilled in the art to 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 accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative labor should fall within the scope of protection of the present application.

[0040] It should be noted that, in the case of no conflict, the features in the embodiments of the present application and the embodiments can be combined with each other. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0041] ReferenceFigure 1 The embodiment discloses a fault self-healing and resource optimization method of a medical image SaaS platform, and the method comprises the following steps:

[0042] S100: acquiring multi-source observation data in the running process of the medical image SaaS platform, and constructing a digital twin model of the platform. The multi-source observation data is acquired by using a non-invasive full-link data acquisition technology.

[0043] Specifically, an OpenTelemetry Agent is deployed in the Kubernetes cluster, which can automatically identify and access various microservices (including services developed in different languages such as Java, Python and Go) running in the platform, and can collect three types of key information, i.e. log data, trace data and index data, without modifying the business code. In the data collection process, a unique Trace ID is generated for each complete user request (such as an image review operation), and a Span ID is assigned to the processing unit of the request in each service, according to the formula: wherein, all events are represented by event, event∈Traceij represents that the time belongs to a certain Trace, Context(eventi) represents the context information of event i, ← represents the assignment operation, TraceID is a unique identifier of a complete user request, and SpanID is an identifier of the processing unit of the request in a certain service. The context injection mechanism described above injects these identifiers into all related events, thereby associating the originally dispersed log, trace and index data, and restoring the complete life cycle of the user request from initiation to response. Based on the associated multi-source observation data, a service calling relationship topology graph is constructed and continuously updated in real time by using a graph database, and the process complies with the formula

[0044] wherein, Gtopo(t) is a service topology graph at time t, BuildGraph is a graph construction function, Spancaller is a Span of a caller, Spancallee is a Span of a callee, and Latencyt is a calling delay at time t. For all the traces. The nodes of the topology graph represent microservice instances, the edges represent the calling relationship between services, and the weight of the edge dynamically reflects real-time delay, request volume, and error rate, and other key indicators. An example is given in this embodiment: in the actual application of a certain medical image SaaS platform, the digital twin model enables the operation and maintenance team to obtain a dynamic and global system map for the first time. When a user complains that the image loading is slow, the team can locate the bottleneck to a slow query of the image archiving service to the metadata database within 1 minute through traceability, rather than the previously suspected network problem. This effectively solves the system black box problem under the traditional microservice architecture and realizes the full-link transparency from user clicks to database queries, providing a data foundation for subsequent intelligent anomaly detection and root cause positioning.

[0045] S200: Based on the digital twin model and multi-source observation data, intelligent identification and root cause positioning of abnormal states of the platform are performed to carry out cognitive analysis of the operation problems of the platform.

[0046] After the digital twin model of the medical image SaaS platform is constructed in step S100, intelligent identification and root cause positioning of abnormal states are implemented based on the model and multi-source observation data. First, hundreds of key performance indicators (KPIs) are extracted from the digital twin model, including QPS, P99 delay, CPU / memory usage, database connection number, etc. of each microservice, and these indicators are combined into a high-dimensional vector KPIVector t , where t represents the current time point. Subsequently, the DBSCAN clustering algorithm is applied to analyze the historical data, and multiple normal operation modes are automatically learned and divided, each mode corresponding to a centroid Centroid k , k represents the kth normal mode. When the current KPIVector t is detected to be more than a preset threshold ∈ k from all known normal mode centroids Centroid k , it is determined to be in an abnormal state. This determination logic follows the formula: , where Anomaly represents whether the system is in an abnormal state, and dist is the distance function. An example of an actual application is given in this embodiment: when the system automatically alarms on Tuesday afternoon that the error rate and P99 delay combination vector of the AI reasoning service deviates from all known normal modes, the system immediately starts the root cause positioning process. This process applies a causal inference algorithm on the service calling relationship topology graph Gtopo, and through the formula: The analysis is performed, where v represents a node (microservice) in the topology graph, P(Anomaly|do(v=fault)) represents the probability of observing a global anomaly when node v fails, do(v=fault) represents the intervention operation performed on node v, RootCause represents the root cause, and argmax represents finding the parameter that maximizes the function. The causal influence of each node as a source of failure on the global anomaly is calculated, and the node with the highest probability of the global anomaly is finally determined as the root cause. In the example above, within 90 seconds, a memory leak was accurately identified in the newly launched AI model version v2.1's handling of a rare CT scan protocol, causing the GPU instance to OOM (Out of Memory), thus triggering a service anomaly. This allowed the operations team to immediately roll back the model version, significantly reducing the mean time to repair (MTTR) from over two hours in the traditional solution to less than five minutes, while also improving the alarm accuracy to over 95%, effectively solving the alarm storm and silent problem caused by traditional static threshold monitoring.

[0047] S300: Based on the results of cognitive analysis, execute the preset corresponding fault repair operations and resource optimization strategies.

[0048] After intelligently identifying and locating the root cause of abnormal states, based on the results of cognitive analysis, pre-set corresponding fault repair operations and resource optimization strategies are executed. Specifically, this is achieved through the formula: Once the root cause of the fault is identified, an automated remediation process matching the fault type is immediately triggered. For example, if a memory leak is detected due to AI model version v2.1 (as in the previous example), a version rollback operation is automatically performed to restore the service to the previous stable version, and a remediation report is sent to the operations team. For slow database query issues, preset index optimization strategies or query rewriting rules are automatically applied. Regarding resource optimization, based on system state information obtained from cognitive analysis, a reinforcement learning agent is constructed using a deep Q-network, which follows the formula: R... t =w1·(SLA) target -SLA actual )-w2·Cost(Action t ), where R t This represents the immediate reward obtained at time t, where w1 and w2 represent the weighting coefficients of performance reward and cost penalty, respectively, and SLA. target It is the target value of the Service Level Agreement (SLA). actual It refers to the actual level of service achieved, Cost (Action). t ) is the action to be performed. t The resulting resource costs, Action t This represents the action performed by the agent. The agent operates based on the current system state vector, State.t determining the optimal resource adjustment action, which follows the formula: where Action* represents the optimal action, Q(State t , a; θ) is a deep Q-network used to approximate the long-term expected return of performing action a in state State t , and θ represents the parameters of the neural network.

[0049] S400: Establish a normal operation behavior sequence baseline for platform users based on a Markov model or a recurrent neural network, and calculate the occurrence probability of the current behavior sequence of the user in real time. When the negative log-likelihood value of the occurrence probability exceeds a safety threshold, trigger an adaptive security defense mechanism, which includes at least one of flow limiting, forced multi-factor authentication, or session termination.

[0050] First, the operation history of all users on the platform is subjected to deep learning, and a personalized normal operation behavior sequence baseline is established for each user or user role (such as doctors, technicians, administrators). For a doctor user, the operation type is defined as a discrete state space, including "login", "query patient", "review CT", "review MRI", "call AI_A", "call AI_B", "print report", and more than 20 basic operations. Based on these operation sequences, a state transition probability matrix is constructed using a Markov model, and the formula P(op t+1 |op t ,...,op1) = MarkovModel(UserID) is followed, where op t represents the operation performed by the user at time t, UserID identifies a specific user or user role, and P(op t+1 |op t ,..., op1) represents the probability of predicting the occurrence of the next operation op t under the condition of known historical operation sequences op1 to op t+1 . For advanced users with more complex behavior patterns, a recurrent neural network (RNN) is used instead of a Markov model, which can capture longer operation dependencies.

[0051] When the system monitors the real-time operation of the user, the occurrence probability P(op1,op2,...,op n ) of the current behavior sequence is immediately calculated, and abnormality is determined according to the formula: , where -log P(op1,...,op n ) represents the negative log-likelihood value of the user behavior sequence, and τ sec is a dynamically adjusted safety threshold. The safety threshold τ secNot a fixed value, but dynamically adjusted according to user role, current time period, historical behavior stability, etc. For example, set a relatively loose threshold for the chief physician role during the daytime on weekdays, and set a more stringent threshold for non-working hours or for newly hired doctors. When abnormal behavior is detected, the system automatically triggers the corresponding adaptive security defense mechanism according to the risk level: for low-risk behavior, only record logs and send monitoring alerts; for medium-risk behavior, force the user to perform secondary multi-factor authentication (MFA); for high-risk behavior, immediately terminate the session and notify the security team. The risk level is determined by comparing the value of -log P with the threshold. An example is given in this embodiment: the system once monitored a chief physician account in a hospital, which accessed the electronic film list page of 500 different patients under his department with an accurate interval of 1.2 seconds within 3 minutes, but did not open any specific image. The probability calculation result of this behavior sequence is P = 1.2 × 10-287, and the corresponding negative log likelihood value -log P = 661.3, which is much higher than the threshold τ sec = 200 for this role in this time period, the system immediately determines that it is a data crawling type of tapping behavior, triggers the medium-risk security policy, and forces the account to perform secondary MFA authentication. Since the attacker cannot provide the hospital intranet dynamic token required for secondary authentication, the authentication fails, and the system automatically terminates the session, successfully preventing a potential medical data leakage event, effectively solving the problem of business logic level attacks that traditional WAFs cannot handle.

[0052] Reference Figure 2 The embodiment further discloses a fault self-healing and resource optimization system of a medical image SaaS platform, comprising:

[0053] A data acquisition and model construction module 21 is configured to acquire multi-source observation data in the running process of the medical image SaaS platform, and construct a digital twin model of the platform;

[0054] An abnormality analysis module 22 is configured to intelligently identify and locate the root cause of the abnormal state of the platform based on the digital twin model and the multi-source observation data, to perform cognitive analysis on the running problems of the platform;

[0055] An execution optimization module 23 is configured to execute a preset corresponding fault repair operation and resource optimization strategy according to the result of the cognitive analysis;

[0056] A security defense module 24 is configured to establish a normal operation behavior sequence baseline for platform users based on a Markov model or a recurrent neural network, and to calculate the occurrence probability of the current behavior sequence of the user in real time. When the negative log likelihood value of the occurrence probability exceeds a security threshold, an adaptive security defense mechanism is triggered, which includes at least one of flow limiting, forced multi-factor authentication, or session termination.

[0057] Figure 3 An electronic device entity structure schematic diagram provided by the embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the electronic device 50 comprises a processor 501, a memory 502 and a bus 503. Figure 3

[0058] The processor 501 and the memory 502 can communicate with each other through the bus 503. The processor 501 is configured to invoke program instructions in the memory 502 to execute the method provided by each method embodiment.

[0059] The embodiment provides a non-transitory computer readable storage medium storing computer instructions, and the computer instructions cause a computer to execute the method provided by each method embodiment.

[0060] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, and the foregoing program can be stored in a computer readable storage medium, and the program executes the steps of the above-mentioned method embodiments when executed; and the foregoing storage medium includes ROM, RAM, magnetic disk or optical disk and various storage media that can store program codes.

[0061] The device embodiments described above are only schematic, and units described as separate components can or can not be physically separate, and components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Some or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0062] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk or optical disk, and include a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute the method of each embodiment or some part of the embodiment.

[0063] ​The foregoing detailed description has set forth various embodiments of the devices and / or processes via the use of specific terminology. However, embodiments thereof can be practiced with the exact description not being presented in detail. The term "device" should be understood to encompass devices operating in various modes, such as active mode, sleep mode, hibernate mode, and the like. The terms "coupled" and "connected," along with derivatives thereof, can be used. It should be understood that these terms are not intended as synonyms for each other. Rather, particular circuitry that can be said to be coupled to, or connected with, other circuitry can be some of the other circuitry that can communicate in some way, while other circuitry can be some of the other circuitry that can not be in communication with that particular circuitry.

Claims

1. A fault self-healing and resource optimization method of a medical image SaaS platform, characterized in that, The method comprises: acquiring multi-source observation data in the running process of a medical image SaaS platform, and constructing a digital twin model of the platform; based on the digital twin model and multi-source observation data, intelligently identifying and locating the root cause of the abnormal state of the platform to perform cognitive analysis of the platform running problem; according to the results of cognitive analysis, performing pre-set corresponding fault repair operations and resource optimization strategies.

2. The method of claim 1, wherein, The multi-source observation data includes log data, tracking data and index data; The construction of the digital twin model of the platform specifically includes: correlate the dispersed log data, tracking data and index data by injecting a unified context identifier; based on the correlated multi-source observation data, real-time construction and update of service call relationship topology graph, the nodes of the topology graph represent microservices, the edges represent service call relationship, and the weight of the edge contains at least one of real-time delay, request volume and error rate.

3. The method of claim 2, wherein, The intelligent identification of the abnormal state of the platform includes: based on the multi-source observation data, acquiring performance indicators KPI in the running process of the platform, and forming a high-dimensional vector; learn the normal operation mode in the historical data of the platform by unsupervised clustering algorithm and determine the centroid of the normal mode; when the distance between the current KPI corresponding high-dimensional vector and all known normal mode centroids exceeds the pre-set threshold, it is determined that the system is in an abnormal state.

4. The method of claim 3, wherein, The root cause positioning specifically includes: apply causal inference algorithm to the service call relationship topology graph to analyze the propagation path of the abnormal state on the topology graph; calculate the causal influence degree of each node as a fault source on the global abnormal phenomenon, and determine the node with the maximum causal influence degree as the fault root cause.

5. The method for failure self-healing and resource optimization of claim 1, wherein, According to the results of cognitive analysis, performing pre-set corresponding fault repair operations and resource optimization strategies includes: when the cognitive analysis result determines that the fault root cause is a specific microservice or component, perform the repair operation matching the fault type; based on the system state information obtained by cognitive analysis, build a reinforcement learning agent through a deep Q network, which determines the optimal resource adjustment action according to the current system state vector; wherein the system state vector includes at least one of resource configuration, key performance indicator, business load and cognitive analysis result, and the resource adjustment action includes at least one of increasing or decreasing computing instances, adjusting CPU / GPU resource configuration, and modifying load balancing strategy; The reinforcement learning agent uses a joint reward function of cost and performance. When the actual service level agreement SLA is better than the target SLA, a performance reward is given, and a penalty is imposed on the cost of resource adjustment action to maximize resource utilization efficiency.

6. The method of claim 1, wherein, Further comprising: establish a normal operation behavior sequence baseline for platform users based on a Markov model or a recurrent neural network; real-time calculation of the occurrence probability of the current behavior sequence of the user, when the negative log likelihood value of the occurrence probability exceeds the safety threshold, trigger the adaptive security defense mechanism, which includes at least one of flow limiting, forced multi-factor authentication or session termination.

7. A fault self-healing and resource optimization system of a medical image SaaS platform, characterized in that, It includes: A data acquisition and model construction module is configured to acquire multi-source observation data in the running process of the medical image SaaS platform and construct a digital twin model of the platform; An anomaly analysis module is configured to intelligently identify and locate the root cause of an abnormal state of the platform based on the digital twin model and the multi-source observation data, so as to perform cognitive analysis on the running problem of the platform; An execution optimization module is configured to execute a preset corresponding fault repair operation and resource optimization strategy according to the result of the cognitive analysis.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1-6.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the method in any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the method in any one of claims 1-6.

Images