Intelligent terminal test data mode clustering analysis method and system

Abstract

This invention discloses a method and system for clustering analysis of test data patterns in smart terminals, belonging to the field of data analysis technology. The method includes: acquiring operational data of smart terminal components; identifying micro-deviation events based on the operational data and preset characteristic fluctuation patterns, wherein the preset characteristic fluctuation patterns reflect the fluctuation of system load under historical fault interference; constructing a directed acyclic graph (DAG) based on the micro-deviation events; extracting structural features from the DAG, wherein the structural features reflect the internal logical relationships of the DAG; and clustering the structural features using a preset clustering algorithm to identify fault patterns, wherein the fault patterns reflect the complete event chain of fault occurrence. This invention can combine micro-deviation events and structural features to identify fault patterns, thereby achieving data pattern clustering analysis and improving accuracy and reliability.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method and system for clustering analysis of smart terminal test data patterns. Background Technology

[0002] In scenarios involving large-scale automated testing of smart terminals, systems generate high-dimensional, high-frequency, multi-source heterogeneous time-series data. Some key performance issues, such as momentary stuttering, are not caused by drastic changes in a single parameter, but rather by a series of weak, ordered fluctuations in multiple system parameters over a short period. Existing analysis methods typically identify these issues based on data snapshots within a specific time window, making it difficult to accurately identify fault modes. Specifically, the signal characteristics of performance problems are weak and easily masked by background fluctuations generated during normal system operation, leading to missed detections. For example, a momentary stutter might manifest as a tiny spike in the load of a processor core, while a processing stage in the graphics processing unit experiences a delay of a few milliseconds. These minute fluctuations are extremely similar to the background noise during normal system operation, making them difficult for existing analysis methods to effectively identify, resulting in low data analysis accuracy and reliability.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this invention is to propose a method and system for clustering analysis of smart terminal test data patterns. This method can identify fault modes by combining micro-deviation events and structural features to achieve data pattern clustering analysis, thereby improving accuracy and reliability.

[0005] On one hand, embodiments of the present invention provide a method for clustering analysis of smart terminal test data patterns, including the following steps: The system acquires operational data from intelligent terminal components, including a central processing unit, a graphics processing unit, memory, and a storage device. Based on the operating data and preset characteristic fluctuation patterns, micro-deviation events are identified. The preset characteristic fluctuation patterns are used to reflect the fluctuation of system load under historical fault interference. The micro-deviation events include CPU load surge events, graphics processing unit rendering delay fluctuation events, memory read / write rate decrease events, and memory input / output wait time increase events. Based on the aforementioned micro-deviation events, construct a directed acyclic graph of events; Based on the event directed acyclic graph, structural features are extracted, including node features, edge features, and topological features, which are used to reflect the internal logical relationships of the event directed acyclic graph; The structural features are clustered using a preset clustering algorithm to identify fault modes. The preset clustering algorithm includes spectral clustering, hierarchical clustering, and association graph clustering. The fault modes are used to reflect the complete event chain of the fault occurrence.

[0006] On the other hand, embodiments of the present invention provide a smart terminal test data pattern clustering analysis system, including: The data acquisition module is used to acquire the operating data of the smart terminal component, which includes a central processing unit, a graphics processing unit, a memory, and a storage device. The micro-deviation event identification module is used to identify micro-deviation events based on the operating data and preset characteristic fluctuation patterns. The preset characteristic fluctuation patterns are used to reflect the fluctuation of system load under historical fault interference. The micro-deviation events include CPU load surge events, graphics processing unit rendering delay fluctuation events, memory read / write rate decrease events, and memory input / output wait time increase events. The event directed acyclic graph construction module is used to construct an event directed acyclic graph based on the micro-deviation events; The structural feature extraction module is used to extract structural features based on the event directed acyclic graph. The structural features include node features, edge features, and topological features, which are used to reflect the internal logical relationships of the event directed acyclic graph. The fault mode identification module is used to cluster the structural features using a preset clustering algorithm to identify fault modes. The preset clustering algorithm includes spectral clustering, hierarchical clustering, and association graph clustering. The fault mode is used to reflect the complete event chain of the fault occurrence.

[0007] The embodiments of this application include at least the following beneficial effects: First, the embodiments of this application acquire the operating data of the intelligent terminal components. Based on the operating data and preset characteristic fluctuation patterns, micro-deviation events are identified. Then, based on the micro-deviation events, a directed acyclic graph of events is constructed. Next, based on the directed acyclic graph of events, structural features are extracted. Finally, a preset clustering algorithm is used to cluster the structural features to identify fault modes. Thus, it is possible to combine micro-deviation events and structural features to identify fault modes, thereby realizing data pattern clustering analysis and improving accuracy and reliability.

[0008] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description and the drawings. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0010] Figure 1 This is a flowchart of a smart terminal test data pattern clustering analysis method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a smart terminal test data pattern clustering analysis system according to an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0012] In related technologies, scenarios involving large-scale automated testing of smart terminals generate high-dimensional, high-frequency, multi-source heterogeneous time-series data. Existing analysis methods typically rely on data snapshots within a specific time window for identification, making it difficult to accurately identify fault modes. Specifically, existing analysis methods usually perform preliminary classifications based on some preset rules, which are insufficient to handle complex situations that occur occasionally or are intertwined with multiple problems. Furthermore, existing analysis methods often fail to clarify the chronological order of events, thus failing to fully reconstruct how the fault occurred. Especially in large-scale automated testing, equipment operates continuously for extended periods, generating massive amounts of data. Many critical performance issues, such as momentary stuttering, are not caused by drastic changes in a single parameter, but rather by a series of weak, orderly fluctuations in multiple system parameters over a short period. Existing analysis methods typically analyze only data snapshots at a specific point in time, making it difficult to detect these weak signals, which are easily masked by various fluctuations during normal system operation. Moreover, the higher the data dimension, the more complex the normal operating state becomes; many normal fluctuations resemble fault characteristics, leading to numerous errors in the analysis results. Existing analysis methods do not consider the temporal sequence of data points, making it impossible to identify fault modes composed of a series of specific events or to extract the complete process from the occurrence to the manifestation of a fault from the data. Performance problems often exhibit weak signal characteristics, easily masked by background fluctuations during normal system operation, leading to missed detections. For example, a momentary lag might manifest as a tiny spike in the load of a single processor core, while a processing stage in the graphics processing unit might be delayed by a few milliseconds. These minute fluctuations are extremely similar to the background noise during normal system operation, making them difficult for existing analysis methods to effectively identify, resulting in low accuracy and reliability in data analysis.

[0013] For example, during the intensive testing phase before the market launch of a new smart terminal model, thousands of prototypes are deployed on an automated testing platform for 24 / 7 stress testing and functional verification. The testing system continuously records the operating data of dozens of key components on each device, such as the load of each processor core, memory read / write speed and page replacement frequency, graphics processing unit rendering pipeline status, memory I / O latency, battery temperature, and signal strength of specific communication modules. This data is aggregated in time-series format to a central data analysis server. In the early stages of testing, the analysis system primarily relies on a clustering method based on feature vector distance. This method takes the data from all monitoring points within a certain time window as a snapshot, extracts its statistical features, and then groups snapshots with similar features into one category. This method is effective in locating serious faults with obvious characteristics. For example, if an application experiences a memory leak, it will cause the memory usage to continuously and unidirectionally increase, forming a very unique trajectory that can be identified and classified into the "memory anomaly" category.

[0014] However, as testing progressed, testers discovered that some devices, when running graphics-intensive applications, occasionally experienced momentary stutters that were almost imperceptible to users but could be detected by high-precision frame rate monitoring tools. These stutters were extremely short, typically lasting only tens to hundreds of milliseconds; the device did not crash or report errors, and no direct error records were left in the system logs. Analysts attempted to find patterns in these momentary stutters from massive amounts of data, but found the characteristics to be very subtle. At the data level, a stutter might only manifest as a tiny spike in the load of a processor core, or a delay of a few milliseconds in a processing stage of the graphics processing unit. These minute fluctuations are extremely similar to the massive "background noise" fluctuations generated during normal system operation due to background task scheduling, data cache refresh, etc. Therefore, existing analysis methods could not effectively isolate these momentary stutter events; data snapshots were incorrectly categorized into the vast "normal operation" category, causing the problem to be overwhelmed and unable to be accurately located.

[0015] The embodiments of this application will be explained in detail below with reference to the accompanying drawings: Figure 1 This is an optional flowchart of a smart terminal test data pattern clustering analysis method provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0016] Step S101: Obtain the operating data of the smart terminal components, which include a central processing unit, a graphics processing unit, memory, and a storage device; Step S102: Identify micro-deviation events based on the operating data and preset characteristic fluctuation patterns. The preset characteristic fluctuation patterns are used to reflect the fluctuation of system load under historical fault interference. Micro-deviation events include CPU load surge events, graphics processing unit rendering delay fluctuation events, memory read / write rate decrease events, and memory input / output wait time increase events. Step S103: Construct a directed acyclic graph of events based on the micro-deviation events; Step S104: Extract structural features from the event directed acyclic graph. The structural features include node features, edge features, and topological features, which are used to reflect the internal logical relationships of the event directed acyclic graph. Step S105: Cluster the structural features using a preset clustering algorithm to identify fault modes. The preset clustering algorithms include spectral clustering, hierarchical clustering, and association graph clustering. Fault modes are used to reflect the complete event chain of the fault occurrence.

[0017] Steps S101 to S105 as shown in the embodiments of this application can combine microscopic deviation events and structural features to identify failure modes, thereby achieving data pattern clustering analysis and improving accuracy and reliability.

[0018] In some embodiments, steps S101-S105 may first acquire operational data of the smart terminal components, including a central processing unit (CPU), a graphics processing unit (GPU), memory, and storage. Lightweight data acquisition agents can be deployed on the smart terminal to monitor performance counters and system logs of core components such as the CPU, GPU, memory, and storage in real time. These agents can transmit the collected data to a data processing center via a network. Alternatively, the API interfaces provided by the smart terminal operating system can be used to periodically query the operational status and resource usage of each component, and this data can be stored in a structured manner. For example, data such as CPU utilization, GPU frame rate, available memory space, and storage read / write bandwidth can be collected once per second. It is understood that operational data refers to performance indicators and status information collected in real time from various core hardware components of the smart terminal. This data is typically in the form of time-series data, reflecting the operational status and load of the components at a specific point in time.

[0019] Then, based on the operating data and preset characteristic fluctuation patterns, micro-deviation events are identified. These preset characteristic fluctuation patterns reflect the fluctuations in system load under historical fault interference. Micro-deviation events include CPU load surges, graphics processing unit rendering latency fluctuations, memory read / write rate decreases, and memory I / O latency increases. The collected operating data can be preprocessed, such as through smoothing filtering or trend decomposition, to remove noise and highlight potential fluctuations. The processed data is then compared with the preset characteristic fluctuation patterns. These patterns can be stored in a pattern library, with each pattern containing a series of time-series features, such as mean, variance, peak, and trough values ​​within a specific time window. During the comparison, the similarity between the operating data and each preset pattern can be calculated, for example, using a Dynamic Time Warping (DTW) algorithm. When the similarity exceeds a preset similarity threshold, a micro-deviation event is considered identified. For example, if the CPU load experiences a very short but small spike within a short period, and this spike highly matches the preset pattern of CPU load surges in historical faults, it can be identified as a CPU load surge event. A preset similarity threshold can be calibrated. It can be understood that a preset characteristic fluctuation pattern refers to a series of system load fluctuation patterns predefined based on historical fault data. These patterns aim to capture specific trends or abnormal behaviors in the system's performance parameters when subjected to fault disturbances, such as periodic load fluctuations, instantaneous spikes, or continuous declines. Micro-deviation events refer to subtle events in the operating data of intelligent terminal components that match the preset characteristic fluctuation pattern and indicate potential anomalies. These events themselves may not be sufficient to trigger traditional alarms, but their combination and sequence are crucial for identifying complex faults.

[0020] Next, a directed acyclic graph (DAG) is constructed based on the micro-deviation events. A node can be created for each identified micro-deviation event. Directed edges are added between nodes based on the chronological order of events and potential causal relationships. For example, if event A occurs before event B, and event A may cause event B, a directed edge can be added from event A to event B. Causal relationships can be determined based on the co-occurrence frequency of events within a time window or statistical causal inference methods. For example, if a decrease in memory read / write speed always occurs before an increase in memory I / O latency, and the two are closely linked in time, then the decrease in memory read / write speed can be considered a potential cause of the increase in memory I / O latency, and a corresponding directed edge can be constructed. In essence, a DAG is a graph structure where nodes represent identified micro-deviation events, and directed edges represent causal or chronological relationships between events. The "acyclic" nature of this graph ensures the logicality and traceability of the event sequence, avoiding circular dependencies.

[0021] Based on the event-directed acyclic graph (DAG), structural features are extracted. These features include node features, edge features, and topological features, reflecting the internal logical relationships within the DAG. Quantitative analysis of the event-directed acyclic graph can be performed from multiple dimensions. Node features can include statistics on the number of different types of micro-deviation events in the graph, such as the number of CPU load spikes or graphics processing unit (GPU) rendering latency fluctuations. Edge features can reflect the strength of associations between events or the weight of propagation paths, for example, by calculating time intervals, co-occurrence frequencies, or based on component dependencies. Topological features can include graph connectivity, diameter, average path length, etc., which describe the complexity and propagation range of event chains. For example, the longest path from the initial event to the final event can be calculated to reflect the longest chain of fault propagation. In essence, structural features refer to quantitative indicators extracted from the event-directed acyclic graph to describe the graph structure and event relationships. These features collectively reflect the internal logical relationships of the event sequence and the fault propagation paths.

[0022] Finally, pre-defined clustering algorithms are used to cluster the structural features and identify fault modes. These algorithms include spectral clustering, hierarchical clustering, and association graph clustering. Fault modes reflect the complete event chain of a fault occurrence. Different clustering algorithms can be selected based on actual needs. For example, when using spectral clustering, a similarity matrix can be constructed, reflecting the degree of similarity of structural features between directed acyclic graphs (DAGs) of different events. Feature decomposition is performed on the similarity matrix, mapping the graph to a low-dimensional space, and traditional clustering algorithms such as K-means are used for clustering. When using hierarchical clustering, the distance between structural features of DAGs of different events can be calculated, and cluster trees can be constructed through aggregation or splitting, with appropriate levels for partitioning to identify fault modes. When using association graph clustering, the software module call relationships and concurrent task scheduling information of the smart terminal can be combined to construct an association graph, thereby enabling more refined aggregation of structural features. Each cluster in the clustering results represents a specific fault mode, reflecting faults with similar event chains and propagation paths. Understandably, a failure mode refers to a pattern that reflects the complete event chain of a failure, consisting of a series of specific micro-deviation events arranged in a specific order and with strong correlation. Identifying these patterns helps to gain a deeper understanding of the root causes and evolution of failures.

[0023] This embodiment first acquires real-time operational data from intelligent terminal components. Based on the acquired operational data and preset characteristic fluctuation patterns, potential micro-deviation events are identified. By identifying micro-deviation events, key signals related to faults can be filtered from massive amounts of data, avoiding the missed detection problem caused by weak signals in traditional methods. Next, the identified micro-deviation events are organized into a directed acyclic graph (DAG). This graph not only records the occurrence of events but also reveals the causal or temporal relationships between events through directed edges. This graph structure can connect discrete micro-events into a complete event chain, providing a structured representation for subsequent fault mode analysis. Subsequently, node features, edge features, and topological features are extracted from the constructed DAG. These structural features are a quantitative description of the internal logical relationships of the DAG, comprehensively reflecting the composition of the event chain, the strength of the correlation between events, and the path and complexity of event propagation. By extracting these high-level structural features, different types of event sequences can be effectively distinguished, avoiding the false alarm problem caused by the presence of a large number of benign fluctuations in high-dimensional data in traditional methods. Finally, a pre-defined clustering algorithm is used to cluster the extracted structural features, thereby identifying fault modes. This algorithm groups faults with similar event chains together based on the similarity of structural features. Each clustering result represents a specific fault mode, reflecting the complete event chain of the fault occurrence. In this way, this embodiment can reconstruct the complete path from cause to effect from the data, accurately uncovering hidden fault modes caused by specific event sequences. This embodiment overcomes the limitations of existing technologies in processing complex time-series data and identifying event-sequence faults. It can not only identify weak fault signals but also effectively distinguish between benign fluctuations and fault modes. It can understand and analyze the temporal sequence and inherent logical relationships between data points, thereby accurately identifying fault modes composed of specific event sequences, significantly improving the efficiency and accuracy of fault diagnosis.

[0024] Through the above technical solution, this embodiment can capture subtle abnormal signals that are easily overlooked by traditional methods by identifying micro-deviation events. By constructing a directed acyclic graph (DAG), discrete micro-events can be linked into event chains with causal or temporal relationships, thereby reconstructing the complete path from "cause" to "effect" from the data. This graph-structured representation allows this embodiment to extract structural features that reflect the internal logical relationships of the DAG. These structural features are high-level abstractions of fault propagation paths and event correlations, effectively distinguishing different types of fault modes. Finally, by clustering these structural features using a preset clustering algorithm, fault modes reflecting the complete event chain of fault occurrence can be identified. For example, when dealing with momentary lag issues, this embodiment can identify micro-deviation events and construct an event DAG, such as the complete event chain from "inappropriate data prefetching operations by background applications" to "suboptimal decisions by the system scheduler," then to "sudden increases in processor load," ultimately leading to "fluctuations in graphics processing unit rendering latency." By clustering the structural features of this directed acyclic graph of events, this embodiment can identify the specific fault mode of "instantaneous pause," thereby providing deeper insights for fault diagnosis. Thus, this embodiment not only effectively solves the problems of weak performance signal features and false alarms caused by numerous benign fluctuations in high-dimensional data, but also overcomes the limitations of existing methods that ignore the temporal dependencies of data points. This allows for accurate identification of fault modes composed of specific event sequences, significantly improving the efficiency and accuracy of fault diagnosis.

[0025] In some embodiments, step S102, identifying micro-deviation events based on operating data and preset characteristic fluctuation patterns, may include, but is not limited to, the following steps: Step S201: Perform statistical analysis on the operating data and extract statistical features, including load change trend, load change rate, local peak value, local valley value, peak value duration and valley value duration. Step S202: Compare the statistical features with the preset feature fluctuation pattern and calculate the feature matching degree; Step S203: Identify micro-deviation events based on feature matching degree.

[0026] In some embodiments, statistical analysis can be performed on the operational data to extract statistical features, aiming to extract key indicators reflecting the system state and behavior from the raw, continuous operational data. Statistical features include load change trends, load change rates, local peaks, local troughs, peak duration, and trough duration. Load change trends reflect the overall direction of the system load, such as whether it is continuously rising, falling, or remaining stable; the load change rate quantifies the speed of load changes, helping to capture sudden system behaviors; local peaks and local troughs identify high and low points in the data sequence, which often correspond to abnormal fluctuations in system performance; peak duration and trough duration further describe the duration of these abnormal states. The extraction of these statistical features provides a quantitative basis for subsequent pattern comparison.

[0027] The statistical characteristics are then compared with preset characteristic fluctuation patterns to calculate the feature matching degree, which aims to assess the similarity between the current system operating state and known fault modes. The preset characteristic fluctuation patterns are constructed based on historical fault data and are used to describe typical fluctuation patterns that may occur in system load under specific fault disturbances. The feature matching degree is calculated through comparison, and this matching measure quantifies the degree of agreement between the current statistical characteristics and the preset pattern.

[0028] Then, based on the feature matching degree, micro-deviation events are identified. When the feature matching degree reaches a preset matching degree threshold, it indicates that the current system operating state is highly similar to a certain preset fault fluctuation pattern, thus allowing the determination of a corresponding micro-deviation event, such as a sudden increase in CPU load, fluctuation in graphics processing unit rendering latency, a decrease in memory read / write speed, or an increase in memory input / output latency. The preset matching degree threshold can be calibrated.

[0029] This embodiment performs refined statistical analysis on the operational data of intelligent terminal components to extract multi-dimensional statistical features. These features comprehensively and quantitatively reflect the dynamic changes in system load. These statistical features can capture load trends, rates, and local anomalies and their durations, enabling the system to more sensitively detect potential abnormal behaviors. Subsequently, by comparing these quantified statistical features with preset feature patterns reflecting system load fluctuations under historical fault interference, and calculating the feature matching degree, an objective quantitative indicator is provided to assess the similarity between the current system state and known fault modes. This method based on statistical features and pattern comparison allows the identification of micro-deviation events to move beyond simple threshold judgments and instead identify more complex and representative system behavior patterns, thereby improving the accuracy and robustness of identification.

[0030] Through the above technical solution, this embodiment enables a deeper and more detailed analysis of the operating data of intelligent terminals, thereby improving the accuracy and sensitivity of micro-deviation event identification. Specifically, by extracting statistical features such as load change trends, load change rates, local peaks, local troughs, peak duration, and trough duration, the dynamic characteristics of the system load can be comprehensively captured, avoiding false alarms or missed alarms caused by judgment based on a single indicator. Simultaneously, comparing these statistical features with preset feature fluctuation patterns and calculating feature matching degrees makes the identification process of micro-deviation events more quantitative and objective, effectively distinguishing between normal fluctuations and fault precursors. This provides more reliable input for subsequent fault mode clustering analysis, improving the overall efficiency and accuracy of fault diagnosis.

[0031] In some embodiments, step S202, comparing statistical features with a preset feature fluctuation pattern and calculating the feature matching degree, may include, but is not limited to, the following steps: Acquire environmental parameters and operating status information of smart terminals. Environmental parameters include temperature and humidity. Operating status information includes application type, resource usage and system service status. System service status includes network service status, file system activity status and scheduler status. Calculate the rate of change of environmental parameters based on the environmental parameters; Determine the initial load fluctuation range based on the operating condition information; Calculate the fluctuation range adjustment amount based on environmental parameters and their rate of change; The initial load fluctuation range is adjusted according to the fluctuation range adjustment amount to obtain the target load fluctuation range in order to compensate for environmental interference. The preset characteristic fluctuation pattern is modified according to the target load fluctuation range; The statistical features are compared with the corrected preset feature fluctuation patterns to calculate the feature matching degree.

[0032] In some embodiments, due to the complex and variable operating environment of smart terminals, environmental parameters such as temperature and humidity may have a significant impact on system load, causing the preset characteristic fluctuation mode to fail to accurately reflect the system behavior under different environmental conditions, thereby affecting the accuracy of micro-deviation event identification.

[0033] To this end, we can first obtain environmental parameters and the operating status information of the smart terminal. Environmental parameters refer to the physical quantities of the external environment in which the smart terminal is located, such as temperature and humidity. These parameters can directly or indirectly affect the performance of the internal components of the smart terminal and the system load. Operating status information reflects the current working status of the smart terminal, including the types of applications currently running, the usage of system resources, and the status of various system services. System service status includes network service status, file system activity status, and scheduler status. This information helps to understand the underlying driving factors of system load.

[0034] Then, based on environmental parameters, the rate of change of these parameters is calculated to capture dynamic changes in environmental factors, such as rapid increases or decreases in temperature, which may indicate abnormalities in system heat dissipation or power consumption. Simultaneously, based on operating condition information, the initial load fluctuation range is determined. This initial load fluctuation range is determined based on historical data from the smart terminal under different operating conditions and represents the normal fluctuation range of system load under ideal or standard environments. The corresponding initial load fluctuation range can be determined based on the current operating condition information.

[0035] Next, based on environmental parameters and their rate of change, the fluctuation range adjustment is calculated to quantify the impact of environmental disturbances on system load fluctuations. For example, in high-temperature and high-humidity environments, the normal fluctuation range of the system load may expand, or resource consumption may increase significantly during certain application runs. The fluctuation range adjustment can be determined by consulting a mapping table between environmental parameters, their rate of change, and the fluctuation range adjustment. This mapping table can be set according to actual needs and is not specifically limited. Based on the fluctuation range adjustment, the initial load fluctuation range is adjusted to obtain the target load fluctuation range to compensate for environmental disturbances. By applying the fluctuation range adjustment to the initial load fluctuation range, a target load fluctuation range that better reflects the current environmental conditions can be obtained, effectively compensating for the impact of environmental disturbances on system load performance.

[0036] Finally, based on the target load fluctuation range, the preset characteristic fluctuation pattern is modified, meaning that environmental factors are incorporated into the definition of the system's normal behavior pattern. The modified preset characteristic fluctuation pattern can more accurately reflect the expected fluctuation characteristics of the system load under fault-free conditions in the current environment and operating conditions. Statistical characteristics are then compared with the modified preset characteristic fluctuation pattern to calculate the feature matching degree, thereby improving the accuracy of identifying micro-deviation events.

[0037] This embodiment addresses the potential accuracy issues in identifying micro-deviation events under complex and ever-changing environments by introducing environmental parameters and the operating conditions of the smart terminal, and dynamically adjusting a preset characteristic fluctuation pattern based on this information. Specifically, the normal system load fluctuation range changes when the smart terminal operates under different temperatures, humidity levels, or uses different applications. By acquiring environmental parameters and operating condition information, and calculating the rate of change of environmental parameters, the impact of these external and internal factors on the system load can be quantified. Based on this, an initial load fluctuation range is determined, and the adjustment amount is calculated by combining environmental parameters and their rate of change. This adjustment yields a target load fluctuation range that reflects the current actual operating conditions. Thus, the preset characteristic fluctuation pattern is corrected to adapt to the current environment, allowing subsequent comparisons of statistical features with the corrected pattern to more accurately reflect whether the system has truly deviated, rather than merely exhibiting normal fluctuations caused by environmental changes. This dynamic correction mechanism ensures that the identification of micro-deviation events is no longer limited to a static preset pattern but can adapt to the actual operating conditions of the smart terminal in various complex environments.

[0038] To illustrate this technical solution more clearly, a specific example is used below. Assume that when a smart terminal performs a high-intensity graphics rendering task, the load on its central processing unit (CPU) and graphics processing unit (GPU) will increase significantly. If the ambient temperature is also high, for example, reaching 40 degrees Celsius, then the normal fluctuation range of the system load will be wider than at 25 degrees Celsius. This embodiment first obtains the current ambient temperature as 40 degrees Celsius and the operating condition as high-intensity graphics rendering. According to historical data, the initial load fluctuation range for high-intensity graphics rendering at 25 degrees Celsius is (x, y). Since the current temperature has increased by 15 degrees Celsius, the system can determine the fluctuation range adjustment amount z by querying a mapping table of environmental parameters, the rate of change of environmental parameters, and the fluctuation range adjustment amount. Applying the fluctuation range adjustment amount z to the initial load fluctuation range (x, y) yields a wider target load fluctuation range of (x, y + z). Subsequently, the preset characteristic fluctuation mode will be corrected based on this new target load fluctuation range. When the actual statistical characteristics are compared with the corrected pattern, only when the load fluctuation exceeds the range of (x, y+z) will it be judged as a micro deviation event, thus avoiding the misjudgment of normal load increases caused by high temperature as a precursor to a fault and ensuring the accuracy of identification.

[0039] Through the above technical solution, this embodiment can effectively compensate for the impact of environmental interference on the load fluctuations of the intelligent terminal system, significantly improving the accuracy and robustness of micro-deviation event identification. This embodiment, by dynamically adjusting preset characteristic fluctuation patterns, enables the system to distinguish between normal fluctuations caused by environmental changes and genuine micro-deviation events, thereby avoiding unnecessary alarms and ensuring timely detection of real fault precursors. This not only enhances the practicality of the intelligent terminal test data pattern clustering analysis method but also provides more reliable basic data for subsequent fault mode identification.

[0040] In some embodiments, in step S103, constructing a directed acyclic graph of events based on the micro-deviation events may include, but is not limited to, the following steps: Obtain the event parameters of the micro-deviation event, including the occurrence time, event type, event causal relationship, and involved components; The event time, event type, and involved components are structured to obtain structured event information; Based on the occurrence time and causal relationship of different micro-deviation events, directed connections are made to the structured information of multiple events to construct a directed acyclic graph of events.

[0041] In some embodiments, event parameters for micro-deviation events can be obtained first. Event parameters are key attributes describing each micro-deviation event, encompassing the event's occurrence time, type, causal relationship between events, and the smart terminal components involved. The occurrence time can be accurate to milliseconds or microseconds to ensure the accuracy of the event sequence. The event type specifies the particular micro-deviation event, such as a CPU load surge or a graphics processing unit rendering latency fluctuation. The event causal relationship characterizes whether there is a direct or indirect triggering, influencing, or dependent relationship between different events, which can be inferred through a predefined rule base. The involved components specify the hardware or software components affected when the event occurs.

[0042] Then, the occurrence time, event type, and involved components are structured to obtain structured event information, aiming to transform this raw data into unified and standardized structured event information. This structuring process can take the form of key-value pairs, JSON objects, or specific data structures to facilitate subsequent storage, querying, and analysis. Through structuring, it is possible to ensure that micro-deviation event data from different sources or of different types have a consistent representation, thereby improving the efficiency and accuracy of data processing.

[0043] Based on the occurrence time and causal relationships of different micro-deviation events, directed connections are made to the structured information of multiple events, constructing a directed acyclic graph (DAG). The establishment of directed connections is based on the order of event occurrence and the identified causal relationships. For example, if event C occurs before event D, and event C is the cause of event D, then a directed edge will be established from event C to event D in the DAG. Nodes in the DAG represent structured micro-deviation events, and directed edges represent causal or temporal dependencies between events. The acyclic nature of this graph ensures the logicality and traceability of the event chain, avoiding circular dependencies.

[0044] This embodiment obtains detailed event parameters of micro-deviation events, including occurrence time, event type, causal relationship, and involved components, providing comprehensive foundational data for subsequent graph construction. By structuring the occurrence time, event type, and involved components, the standardization and consistency of event information are ensured, which is crucial for accurately identifying and representing the relationships between events. Subsequently, based on the occurrence time and explicit causal relationships, directed connections are made to this structured information, accurately depicting the temporal and logical dependencies between micro-deviation events. Thus, the constructed directed acyclic graph clearly shows the complete propagation path of a fault from its initial cause to its final manifestation, providing an intuitive and logically rigorous view for in-depth analysis of the root causes and evolution of the fault.

[0045] Through the above technical solution, this embodiment can organize and represent the micro-deviation events identified in smart terminals in a systematic and structured manner. Specifically, by clarifying event parameters and performing structured processing, the integrity and operability of event information are ensured. Furthermore, directed connections based on occurrence time and causal relationships enable the constructed directed acyclic graph of events to accurately reflect the actual propagation path and mutual influence mechanisms of events, avoiding potential problems of ambiguous causal relationships or disordered timing. This not only improves the accuracy and reliability of fault mode analysis but also provides a solid data foundation and logical support for subsequent fault diagnosis and prediction.

[0046] In some embodiments, step S104, extracting structural features based on the directed acyclic graph of events, may include, but is not limited to, the following steps: Step S301: Based on the directed acyclic graph of events, count the number of micro-deviation events and the proportion of each event type as node features; Step S302: Calculate the longest path and average path length in the directed acyclic graph of events as topological features; Step S303: Count the total number of edges in the directed acyclic graph. Step S304: Construct edge features based on the total number of edges.

[0047] In some embodiments, the number of micro-deviation events and the proportion of each event type can be statistically analyzed based on the directed acyclic graph of events (DAG) as node features. Node features reflect the attributes of each node in the DAG. Specifically, by statistically analyzing the number of micro-deviation events, the total number of events in the graph can be quantified; by statistically analyzing the proportion of each event type, the distribution and relative importance of different types of events (e.g., CPU load spikes, graphics processing unit rendering latency fluctuations, memory read / write speed drops, and memory I / O latency increases) in the entire event chain can be revealed. These statistical information collectively constitute a description of the node level of the DAG.

[0048] Then, the longest path and average path length in the directed acyclic graph (DAG) of events are calculated as topological features. Topological features describe the overall structure and connectivity of the DAG. Specifically, the longest path can reflect the longest chain of fault propagation or the most complex causal relationship path, while the average path length measures the tightness of the association between events and the efficiency of information transmission. These path calculations help to understand the depth and breadth of the evolution of fault events.

[0049] Next, count the total number of edges in the directed acyclic graph (DAG). The total number of edges reflects the strength and complexity of the associations between micro-deviation events in the DAG, where edges represent causal or temporal dependencies between events. Based on the total number of edges, construct edge features. For example, consider edge weights, edge types, etc., to more finely characterize the strength and properties of the associations between events.

[0050] This embodiment extracts structural features that comprehensively reflect the internal logical relationships of a directed acyclic graph (DAG) by performing multi-dimensional quantitative analysis on the event DAG. Specifically, the extraction of node features quantifies the composition and distribution of micro-deviation events in the DAG, providing foundational data for subsequent clustering analysis. The calculation of topological features reveals the depth and breadth of event propagation and the overall structural complexity of the graph, aiding in understanding the path of fault evolution. Simultaneously, by counting the total number of edges and constructing edge features, the strength and properties of causal relationships between events are effectively captured. This comprehensive extraction of structural features allows the complex event DAG to be transformed into a numerical representation that can be processed by clustering algorithms, thus laying the foundation for accurate fault mode identification.

[0051] Through the above technical solution, this embodiment ensures that the internal logical relationships and causal chains of the directed acyclic graph of events can be fully captured and represented by meticulously defining node features, topological features, and edge features. This provides high-quality, high-dimensional input data for subsequent fault mode clustering analysis, significantly improving the accuracy and interpretability of fault mode recognition, thereby effectively supporting fault diagnosis and performance optimization of smart terminals.

[0052] In some embodiments, in step S304, constructing edge features based on the total number of edges may include, but is not limited to, the following steps: Obtain current operational lifespan information and historical performance degradation trends of smart terminal components; Based on current operational lifespan information and historical performance degradation trends, the weights of edges in the directed acyclic graph of events are determined. The edge weights are used to reflect the impact of component performance degradation on the strength of event correlation. Calculate the average edge weight based on the edge weights in the directed acyclic graph of events; Edge features are constructed based on the average edge weight and the total number of edges.

[0053] In some embodiments, the current operational lifespan information and historical performance degradation trends of the smart terminal components can be obtained first. The current operational lifespan information of the smart terminal components refers to the cumulative working time, cumulative number of working cycles, or predicted remaining lifespan of the components since they were put into use. This information can quantify the actual usage of the components. Historical performance degradation trends refer to the patterns or regularities in the changes of the components' key performance indicators (such as the stability of the CPU's clock speed, the rendering frame rate of the graphics processing unit, the read / write latency of memory, and the input / output throughput of the memory) over a past period. For example, a performance degradation curve can be fitted using historical monitoring data, or a probability distribution of performance degradation can be obtained through statistical analysis.

[0054] Then, based on current operational lifetime information and historical performance degradation trends, the weights of the edges in the directed acyclic graph (DAG) are determined. These edge weights reflect the impact of component performance degradation on the strength of event correlation. This means that when a component's performance significantly degrades, the correlation strength between it and related micro-deviation events may strengthen or weaken; this change needs to be quantified through edge weights. For example, if the performance degradation trend of a memory component indicates that its input / output latency is continuously increasing, then the weight of the edge related to the "increased memory input / output latency event" might be increased to highlight its importance in the failure chain.

[0055] Next, based on the edge weights in the event-driven acyclic graph (DAG), the average edge weight is calculated. The average edge weight can be obtained by summing the weights of all edges in the DAG and dividing by the total number of edges. Its purpose is to provide a holistic indicator reflecting the average level of event association strength. Furthermore, based on the average edge weight and the total number of edges, more informative edge features can be constructed. For example, the average edge weight can be combined with the total number of edges to form a vector, thus comprehensively characterizing the edge structure of the event-driven acyclic graph.

[0056] This embodiment incorporates current operational lifespan information and historical performance degradation trends of intelligent terminal components, enabling the construction of edge features in the directed acyclic graph (DAG) of events to fully consider the health status and potential degradation risks of the components. Specifically, after obtaining the current operational lifespan information and historical performance degradation trends of the components, the system can assess the degree of performance degradation of the components. Based on this assessment, the weights of edges in the DAG of events can be dynamically adjusted. For example, for events related to components with significantly degraded performance, the weights of their connecting edges will be increased accordingly to highlight the key role of these events in the fault propagation chain; conversely, for components in good health, the weights of their related edges may remain at a lower level. In this way, the weights of edges are no longer simply a matter of presence or absence, but contain in-depth information about the health status of the components, thus more accurately reflecting the actual correlation strength between micro-deviation events. Subsequently, by calculating the average value of these weighted edges and combining it with the total number of edges, edge features that better reflect the complex correlations and potential risks within the system can be constructed.

[0057] To illustrate this technical solution more clearly, a specific example is used below. Suppose that the CPU of a smart terminal has shown a slow upward trend in average load over the past year under specific application scenarios, and its historical performance degradation trend model predicts that the CPU's performance will decrease by 10% within the next three months. When the system detects a "CPU load surge event," since both the CPU's current operating life information (e.g., it has exceeded 80% of its design life) and historical performance degradation trend indicate that it is in a degraded state, the weight of the edge related to this "CPU load surge event" will be significantly increased. For example, a weighting factor can be set that is proportional to the degree of performance degradation of the component. If a memory module's read / write rate shows a periodic downward trend in historical data, and its current operating life is close to the warning threshold, then when a "memory read / write rate decrease event" occurs, the weight of the edge connecting to this event will also increase accordingly. In this way, the system can distinguish between events caused by occasional anomalies in healthy components and events caused by the continuous deterioration of performance-degraded components, thus assigning higher importance to the latter when constructing edge features. Finally, these weighted edges are averaged and combined with the total number of edges in the directed acyclic graph of events to form a comprehensive edge feature vector, which is used for subsequent failure mode clustering analysis.

[0058] Through the above technical solution, this embodiment overcomes the limitations of constructing edge features solely based on the total number of edges. By considering the current operational lifespan information and historical performance degradation trends of intelligent terminal components, the determined edge weights can more realistically and meticulously reflect the impact of component performance degradation on the strength of event correlations. This allows the constructed edge features to not only include the topological relationships between events but also incorporate the key dimension of component health status, thereby significantly improving the ability of the directed acyclic graph of events to characterize the actual operating state of the system. Consequently, in subsequent cluster analysis, more accurate and diagnostically valuable fault modes can be identified. Especially for those hidden faults caused by long-term component performance degradation, this solution can provide stronger identification capabilities and more accurate location, thereby improving the accuracy and efficiency of intelligent terminal fault diagnosis.

[0059] In some embodiments, in step S105, when the preset clustering algorithm is spectral clustering, the preset clustering algorithm is used to cluster structural features and identify fault modes, which may include, but is not limited to, the following steps: A similarity matrix is ​​constructed by performing structural similarity analysis on structural features using graph kernel functions. Based on the similarity matrix, the structural features are mapped to a low-dimensional space to obtain multiple structural feature points; The K-means clustering algorithm is used to cluster multiple structural feature points to identify fault modes.

[0060] In some embodiments, a similarity matrix can be constructed by first performing structural similarity analysis on structural features using graph kernel functions. Graph kernel functions, specifically designed for comparing graph structures or graph-related data, can be used to measure the similarity of structural features such as node features, edge features, and topological features extracted from event-directed acyclic graphs (DAGs). Graph kernel functions can effectively capture the topological structure information of a graph, such as subgraph isomorphism and path similarity, thereby quantifying the degree of structural similarity between DAGs of different events. The purpose is to transform non-vectorized graph structure data into a similarity metric that can be numerically calculated, laying the foundation for subsequent clustering analysis. The similarity matrix is ​​a symmetric matrix where each element represents the similarity value between any two structural features; a higher similarity value indicates greater similarity between the two structural features.

[0061] Then, based on the similarity matrix, the structural features are mapped to a low-dimensional space, resulting in multiple structural feature points. This can be achieved through spectral decomposition (e.g., eigenvalue decomposition of the Laplacian matrix), utilizing the information from the similarity matrix to project the original high-dimensional structural feature data into a lower-dimensional space. In this lower-dimensional space, structural feature points that were difficult to distinguish in the original space become easier to differentiate and cluster due to their enhanced inherent structural similarity. The aim is to reduce data complexity, remove redundant information, and highlight key structural patterns in the data, enabling subsequent clustering algorithms to work more efficiently and accurately. The resulting structural feature points are vector representations of each event's directed acyclic graph in the lower-dimensional space.

[0062] Then, the K-means clustering algorithm is used to cluster multiple structural feature points to identify failure modes. After mapping the structural features to a low-dimensional space and obtaining structural feature points, the K-means algorithm can be used to group these low-dimensional feature points. The K-means algorithm iteratively calculates the mean of each cluster center and assigns each feature point to the nearest cluster center until the cluster centers no longer change significantly. Its purpose is to group directed acyclic graphs of events with similar structural features into one class, with each class representing a specific failure mode. In this way, patterns that reflect the complete event chain of failure occurrence can be effectively identified.

[0063] To illustrate this technical solution more clearly, a specific example is used below. Suppose that during the operation of a smart terminal, a series of micro-deviation events are collected, and multiple event directed acyclic graphs (DAGs) are constructed based on these events. These graphs may represent different fault precursors or actual fault chains. To identify these fault modes, firstly, for each event DAG, node features (such as event type distribution), edge features (such as event association strength), and topological features (such as longest path) are extracted. Subsequently, these structural features are compared using graph kernel functions based on random walks (such as the Gärtner kernel or the Weisfeiler-Lehman kernel), calculating the structural similarity between features, and constructing an N×N similarity matrix, where N is the number of event DAGs. This similarity matrix reflects the degree of structural similarity between any two event DAGs. Next, based on this similarity matrix, the structural features of each event DAG are mapped to a lower-dimensional space using spectral decomposition techniques (e.g., calculating the eigenvectors of the Laplacian matrix), for example, mapping from high-dimensional feature vectors to 2D or 3D feature points. In this low-dimensional space, feature points corresponding to directed acyclic graphs (DAGs) of events with similar structures will move closer to each other, while those with large structural differences will move further apart. Finally, the K-means clustering algorithm is used to cluster these low-dimensional structural feature points. For example, with a preset K value of 3, the K-means algorithm divides these feature points into 3 clusters, each representing a unique fault mode. By analyzing the common features of the DAGs of events within each cluster, specific fault modes such as "system lag caused by memory leaks," "application crashes caused by CPU overload," or "slow response caused by memory I / O bottlenecks" can be identified. In this way, this embodiment can effectively summarize complex event sequences into understandable fault modes, providing clear guidance for fault diagnosis of smart terminals.

[0064] Through the above technical solution, this embodiment overcomes the limitations in processing complex graph structure data, significantly improving the accuracy and robustness of fault mode recognition for smart terminals. Specifically, by performing structural similarity analysis on structural features using graph kernel functions, the nonlinear structural relationships between directed acyclic graphs of events can be captured more precisely, avoiding the risk of losing key topological information due to simple feature vectorization. Furthermore, mapping structural features to a low-dimensional space not only effectively reduces computational complexity but also enables clearer separation of different fault modes during K-means clustering in the low-dimensional space, thereby improving the clustering effect. Therefore, this embodiment can more effectively identify patterns reflecting the complete event chain of fault occurrence, providing a more accurate basis for fault diagnosis and prediction of smart terminals, thereby improving system reliability and maintenance efficiency.

[0065] In some embodiments, in step S105, when the preset clustering algorithm is hierarchical clustering, the preset clustering algorithm is used to cluster structural features and identify fault modes, which may include, but is not limited to, the following steps: A preset distance algorithm is used to calculate the distance between structural features to obtain the feature distance. The preset distance algorithm includes the Euclidean distance algorithm and the cosine similarity algorithm. Based on feature distance, structural features are aggregated to obtain a clustering tree; The clustering tree is hierarchically divided to identify fault modes.

[0066] In some embodiments, a preset distance algorithm can be used to calculate the distance between structural features to obtain the feature distance. The preset distance algorithm refers to a mathematical method used to quantify the similarity or difference between different structural features, and may include Euclidean distance and cosine similarity algorithms. In practical applications, an appropriate distance metric can be selected based on the type of structural feature and the data distribution. For example, the Euclidean distance algorithm is suitable for measuring the straight-line distance between two points in a multidimensional space. When the dimensions of structural features have similar dimensions and distributions, it can effectively reflect the overall differences in features. The cosine similarity algorithm measures the similarity between two vectors by calculating the cosine value of the angle between them, focusing more on direction. It is suitable for processing high-dimensional data or scenarios where the pattern or trend of the feature is more critical. Its purpose is to provide a reliable basis for similarity or difference in subsequent clustering and aggregation. The feature distance is a value calculated according to the selected preset distance algorithm, which can intuitively reflect the proximity between different structural features. For example, a smaller Euclidean distance or a larger cosine similarity usually indicates that two structural features are more similar.

[0067] Then, based on feature distances, structural features are aggregated to obtain a clustering tree. The most similar structural features or existing clusters can be merged progressively based on the calculated feature distances to construct the clustering tree. This process can employ bottom-up (agglomerative) or top-down (split-like) methods. For example, in agglomerative hierarchical clustering, each structural feature is initially treated as an independent cluster, and then the nearest clusters are iteratively merged until all features are merged into one large cluster. A clustering tree is a graphical representation of the hierarchical clustering process, showing the aggregation process of structural features from individuals to the final cluster in a tree structure. Each node in the tree represents a cluster, while edges represent merging relationships between clusters. The height of the tree is typically related to the distance or similarity at the time of merging.

[0068] The clustering tree is then hierarchically divided to identify fault modes. Based on a preset pruning strategy, "cutting" can be performed at different levels of the clustering tree to obtain the final clustering result. For example, a distance threshold can be set, and all edges in the clustering tree with a merge distance exceeding that threshold can be cut; or a desired number of clusters can be set, and appropriate levels can be selected in the clustering tree to obtain the specified number of clusters. The aim is to extract discrete, practically meaningful fault modes from the continuous aggregation process.

[0069] To illustrate this technical solution more clearly, a specific example is used below. Assume a series of structural features have been extracted from the operational data of the smart terminal, each representing a potential chain of fault events. To identify fault modes within these event chains, hierarchical clustering can be employed. Specifically, first, the Euclidean distance algorithm is selected as the preset distance algorithm, and the Euclidean distance between any two structural features is calculated, resulting in a feature distance matrix. This matrix reflects the similarity of each event chain in the multidimensional feature space. Next, an agglomerative hierarchical clustering method is used, treating each structural feature as an initial cluster, and then iteratively merging the two closest clusters until all structural features are merged into a single cluster, thus constructing a complete clustering tree. After the clustering tree is constructed, it can be hierarchically divided according to actual fault diagnosis needs, such as setting a distance threshold or specifying the number of fault modes to be identified (e.g., 5 main fault modes). For example, if five clusters are defined, the system will cut the cluster tree at the corresponding levels to identify five different fault modes. Each mode represents a chain of fault events with similar structural characteristics. In this way, various fault modes existing in smart terminals can be clearly identified, and their characteristics and occurrence patterns can be further analyzed.

[0070] Through the above technical solution, the hierarchical clustering method in this embodiment does not require pre-specifying the number of clusters. Instead, by constructing a clustering tree, it allows users or systems to classify fault modes at different granularities according to actual needs, thus adapting to the diversity and complexity of fault modes in smart terminals. Furthermore, the structure of the clustering tree intuitively displays the hierarchical relationships and evolution processes between fault modes, helping to deeply understand the internal mechanisms and propagation paths of faults, providing valuable insights for root cause analysis and prevention. This method not only improves the accuracy and robustness of fault mode identification but also enhances the interpretability and practicality of fault diagnosis results.

[0071] In some embodiments, in step S105, when the preset clustering algorithm is the association graph clustering method, the preset clustering algorithm is used to cluster the structural features and identify the fault modes, which may include, but is not limited to, the following steps: Obtain software module call relationship information and concurrent task scheduling information of the smart terminal; Based on the software module call relationship information, a software module association graph is constructed. The software module association graph is used to reflect the call relationship and dependency relationship between software modules in the directed acyclic graph of events. Based on the concurrent task scheduling information, a concurrent task association graph is constructed. The concurrent task association graph is used to reflect the scheduling relationship and resource competition relationship between concurrent tasks in the directed acyclic graph of events. Based on the software module association diagram and the concurrent task association diagram, structural features are aggregated to identify fault modes.

[0072] In some embodiments, software module call relationship information and concurrent task scheduling information of the smart terminal can be obtained first. Software module call relationship information of the smart terminal refers to data on mutual calls and dependencies between various software modules within the smart terminal during execution. This can include function call stacks, application programming interface call logs, inter-module communication records, etc., and its purpose is to reveal the logical connections and data flow between software components. Concurrent task scheduling information refers to data on how the smart terminal's operating system or runtime environment manages and allocates resources for multiple concurrent tasks, such as task priority, execution status, resource usage, scheduling queue information, etc., and its purpose is to reflect the parallel execution and resource competition among tasks.

[0073] Then, based on the software module call relationship information, a software module association graph is constructed. This graph reflects the call and dependency relationships between software modules in a directed acyclic graph (DAG). Each software module in the smart terminal can be considered a node in the association graph, and the calls or dependencies between nodes can be considered edges. The direction of the edge can represent the initiator and receiver of the call, and the weight of the edge can represent the call frequency or dependency strength. The purpose of constructing this software module association graph is to intuitively demonstrate the internal connections of the software architecture, thereby better understanding the propagation path of micro-deviation events at the software level.

[0074] Based on the concurrent task scheduling information, a concurrent task association graph is constructed. This graph reflects the scheduling relationships and resource contention relationships between concurrent tasks in a directed acyclic graph (DAG). Concurrent tasks running on the smart terminal can be used as nodes in the graph, and the scheduling relationships (e.g., task synchronization, asynchronicity, mutual exclusion) or resource contention relationships (e.g., competition for shared memory, CPU time slices, and input / output devices) between nodes can be used as edges. The purpose of constructing this concurrent task association graph is to reveal potential bottlenecks and conflicts during task execution, thereby helping to identify failure modes caused by concurrency issues.

[0075] Finally, based on the software module association graph and the concurrent task association graph, structural features are aggregated to identify failure modes. Node features, edge features, and topological features extracted from the directed acyclic graph of events can be fused with the additional information provided by the newly constructed software module association graph and concurrent task association graph. This aggregation can be achieved in various ways. For example, features of the association graph (such as the average call depth between modules, the average competition intensity between tasks, and the number of modules or tasks on the critical path) can be added as new feature dimensions to the original structural feature set. Alternatively, graph embedding techniques can be used to encode the information of the association graph into vectors, which are then concatenated or fused with the structural features. The goal is to form a more comprehensive and insightful feature representation so that subsequent clustering algorithms can more accurately identify failure modes.

[0076] Through the above technical solution, this embodiment can significantly improve the accuracy and depth of fault mode identification in smart terminals. By integrating software module call relationships and concurrent task scheduling information, and constructing a corresponding association graph, cluster analysis is no longer limited to the surface structure of events, but can delve into the software architecture and runtime behavior of smart terminals. Therefore, it can identify complex fault modes caused by defects in specific software modules, improper calls between modules, abnormal concurrent task competition, or scheduling problems. These fault modes are difficult to accurately capture based solely on the structural features of directed acyclic graphs of events. This more comprehensive feature aggregation makes the classification of fault modes more refined, helps to more accurately locate the root cause of faults, and provides more valuable guidance for fault diagnosis and performance optimization of smart terminals, thereby effectively addressing the limitations in handling complex software-related faults.

[0077] The beneficial effects of implementing the embodiments of the present invention include: First, the embodiments of this application acquire the operating data of the intelligent terminal components. Based on the operating data and preset characteristic fluctuation patterns, micro-deviation events are identified. Then, based on the micro-deviation events, a directed acyclic graph of events is constructed. Based on the directed acyclic graph of events, structural features are extracted. Finally, a preset clustering algorithm is used to cluster the structural features to identify fault modes. Thus, it is possible to combine micro-deviation events and structural features to identify fault modes, thereby realizing data pattern clustering analysis and improving accuracy and reliability.

[0078] like Figure 2 As shown in the figure, this embodiment of the invention also provides a smart terminal test data pattern clustering analysis system, including: The data acquisition module 401 is used to acquire the operating data of the smart terminal component, which includes a central processing unit, a graphics processing unit, memory, and a storage device. The micro-deviation event identification module 402 is used to identify micro-deviation events based on operating data and preset characteristic fluctuation patterns. The preset characteristic fluctuation patterns are used to reflect the fluctuation of system load under historical fault interference. Micro-deviation events include CPU load surge events, graphics processing unit rendering delay fluctuation events, memory read / write rate decrease events, and memory input / output wait time increase events. The event directed acyclic graph construction module 403 is used to construct an event directed acyclic graph based on micro-deviation events; The structural feature extraction module 404 is used to extract structural features from the event directed acyclic graph. The structural features include node features, edge features and topological features, which are used to reflect the internal logical relationships of the event directed acyclic graph. The fault mode identification module 405 is used to cluster structural features using a preset clustering algorithm to identify fault modes. The preset clustering algorithms include spectral clustering, hierarchical clustering, and association graph clustering. The fault mode is used to reflect the complete event chain of the fault occurrence.

[0079] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0080] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

Claims

1. A clustering analysis method for smart terminal test data patterns, characterized in that, Includes the following steps: The system acquires operational data from intelligent terminal components, including a central processing unit, a graphics processing unit, memory, and a storage device. Based on the operating data and preset characteristic fluctuation patterns, micro-deviation events are identified. The preset characteristic fluctuation patterns are used to reflect the fluctuation of system load under historical fault interference. The micro-deviation events include CPU load surge events, graphics processing unit rendering delay fluctuation events, memory read / write rate decrease events, and memory input / output wait time increase events. Based on the aforementioned micro-deviation events, construct a directed acyclic graph of events; Based on the event directed acyclic graph, structural features are extracted, including node features, edge features, and topological features, which are used to reflect the internal logical relationships of the event directed acyclic graph; The structural features are clustered using a preset clustering algorithm to identify fault modes. The preset clustering algorithm includes spectral clustering, hierarchical clustering, and association graph clustering. The fault modes are used to reflect the complete event chain of the fault occurrence.

2. The method according to claim 1, characterized in that, The step of identifying micro-deviation events based on the operational data and preset characteristic fluctuation patterns includes: The operational data is statistically analyzed to extract statistical features, including load change trend, load change rate, local peaks, local valleys, peak duration, and valley duration. The statistical features are compared with the preset feature fluctuation pattern to calculate the feature matching degree; The micro-deviation event is identified based on the feature matching degree.

3. The method according to claim 2, characterized in that, The step of comparing the statistical features with the preset feature fluctuation pattern and calculating the feature matching degree includes: The system acquires environmental parameters and operating status information of the smart terminal. The environmental parameters include temperature and humidity, and the operating status information includes application type, resource usage, and system service status. The system service status includes network service status, file system activity status, and scheduler status. Calculate the rate of change of the environmental parameters based on the environmental parameters; Based on the aforementioned operating condition information, determine the initial load fluctuation range; Calculate the fluctuation range adjustment amount based on the environmental parameters and the rate of change of the environmental parameters; The initial load fluctuation range is adjusted according to the fluctuation range adjustment amount to obtain the target load fluctuation range in order to compensate for environmental interference. The preset characteristic fluctuation pattern is corrected based on the target load fluctuation range; The statistical features are compared with the corrected preset feature fluctuation pattern to calculate the feature matching degree.

4. The method according to claim 1, characterized in that, The construction of a directed acyclic graph of events based on the micro-deviation events includes: Obtain the event parameters of the micro-deviation event, including the occurrence time, event type, event causal relationship, and involved components; The occurrence time, event type, and involved components are structured to obtain structured event information; Based on the occurrence time and causal relationship of different micro-deviation events, directed connections are made to the structured information of multiple events to construct the directed acyclic graph of the events.

5. The method according to claim 1, characterized in that, The step of extracting structural features from the directed acyclic graph of events includes: Based on the directed acyclic graph of events, the number of micro-deviation events and the proportion of each event type are statistically analyzed and used as the node features; Calculate the longest path and average path length in the directed acyclic graph of the events, and use them as the topological features; Count the total number of edges in the directed acyclic graph containing the events described. The edge features are constructed based on the total number of edges.

6. The method according to claim 5, characterized in that, The step of constructing the edge features based on the total number of edges includes: Obtain current operational lifespan information and historical performance degradation trends of smart terminal components; Based on the current operating life information and the historical performance degradation trend, the weights of the edges in the directed acyclic graph of the event are determined, and the weights of the edges are used to reflect the impact of component performance degradation on the event correlation strength. Calculate the average edge weight based on the edge weights in the directed acyclic graph of the events. The edge features are constructed based on the average edge weight and the total number of edges.

7. The method according to claim 1, characterized in that, When the preset clustering algorithm is spectral clustering, the step of clustering the structural features using the preset clustering algorithm to identify fault modes includes: A similarity matrix is ​​constructed by performing structural similarity analysis on the structural features using a graph kernel function. Based on the similarity matrix, the structural features are mapped to a low-dimensional space to obtain multiple structural feature points; The K-means clustering algorithm is used to cluster the multiple structural feature points to identify the fault modes.

8. The method according to claim 1, characterized in that, When the preset clustering algorithm is hierarchical clustering, the step of using the preset clustering algorithm to cluster the structural features and identify fault modes includes: The structural features are calculated using a preset distance algorithm to obtain the feature distance. The preset distance algorithm includes the Euclidean distance algorithm and the cosine similarity algorithm. Based on the feature distance, the structural features are aggregated to obtain a clustering tree; The clustering tree is hierarchically divided to identify the fault modes.

9. The method according to claim 1, characterized in that, When the preset clustering algorithm is an association graph clustering method, the step of using the preset clustering algorithm to cluster the structural features and identify fault modes includes: Obtain software module call relationship information and concurrent task scheduling information of the smart terminal; Based on the software module call relationship information, a software module association graph is constructed. The software module association graph is used to reflect the call relationship and dependency relationship between software modules in the directed acyclic graph of events. Based on the concurrent task scheduling information, a concurrent task association graph is constructed. The concurrent task association graph is used to reflect the scheduling relationship and resource competition relationship between concurrent tasks in the directed acyclic graph of events. Based on the software module association diagram and the concurrent task association diagram, the structural features are aggregated to identify the fault modes.

10. A clustering analysis system for smart terminal test data, characterized in that, include: The data acquisition module is used to acquire the operating data of the smart terminal component, which includes a central processing unit, a graphics processing unit, a memory, and a storage device. The micro-deviation event identification module is used to identify micro-deviation events based on the operating data and preset characteristic fluctuation patterns. The preset characteristic fluctuation patterns are used to reflect the fluctuation of system load under historical fault interference. The micro-deviation events include CPU load surge events, graphics processing unit rendering delay fluctuation events, memory read / write rate decrease events, and memory input / output wait time increase events. The event directed acyclic graph construction module is used to construct an event directed acyclic graph based on the micro-deviation events; The structural feature extraction module is used to extract structural features based on the event directed acyclic graph. The structural features include node features, edge features, and topological features, which are used to reflect the internal logical relationships of the event directed acyclic graph. The fault mode identification module is used to cluster the structural features using a preset clustering algorithm to identify fault modes. The preset clustering algorithm includes spectral clustering, hierarchical clustering, and association graph clustering. The fault mode is used to reflect the complete event chain of the fault occurrence.

Images