Artificial intelligence-based system maintenance task intelligent scheduling system

Abstract

The application is specifically an artificial intelligence-based system maintenance task intelligent scheduling system, and relates to the technical field of artificial intelligence, nonlinear dynamics and computer system operation and maintenance, comprising: a univariate time series data acquisition and denoising module; a phase space reconstruction module; a chaotic feature quantitative calculation module; a dynamics state discrimination and early warning module; and an entropy-driven task blocking and scheduling execution module.In the application, Takens embedding theorem is used to realize univariate micro time series signal acquisition, and the global dynamics state of the system can be completely characterized by only CPU interrupt response delay and memory bus bandwidth occupancy, thereby reducing data transmission, storage and calculation overhead; and the maximum Lyapunov exponent is calculated to realize accurate quantification of steady state, bifurcation and chaotic state.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, nonlinear dynamics and computer system operation and maintenance technology, and in particular to an intelligent scheduling system for system maintenance tasks based on artificial intelligence. Background Technology

[0002] Current computer system maintenance and task scheduling technologies generally rely on multi-dimensional collection and threshold comparison of macro performance indicators, mainly monitoring routine operating parameters such as CPU utilization, memory usage, disk I / O, and network throughput. This approach suffers from redundant data collection dimensions and high data transmission and computational overhead.

[0003] Such technologies can only reflect the surface-level operating state of the system and cannot collect key signals that characterize the micro-dynamic characteristics of the system, such as CPU interrupt response delay and memory bus bandwidth fluctuation. Furthermore, they do not model the internal evolution law of the system based on nonlinear dynamics theory, making it difficult to capture early minor anomalies such as hidden hardware faults, software logic conflicts, and memory leak precursors. They generally suffer from defects such as delayed fault perception, low early warning accuracy, and passive response rather than proactive prediction.

[0004] Existing system scheduling and fault handling schemes lack quantitative analysis of chaotic characteristics and entropy-driven optimization mechanisms. They fail to accurately characterize the dynamic state of the system through phase space reconstruction and maximum Lyapunov exponent calculation, and cannot effectively distinguish between the three evolutionary stages of steady state, bifurcation, and chaos.

[0005] Therefore, an intelligent scheduling system for system maintenance tasks based on artificial intelligence is proposed to address the aforementioned problems. Summary of the Invention

[0006] The purpose of this invention is to propose an intelligent scheduling system for system maintenance tasks based on artificial intelligence in order to solve the above-mentioned problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: An AI-based intelligent scheduling system for system maintenance tasks includes: The single-variable time series data acquisition and denoising module is configured to acquire physical indicators that reflect the micro-dynamic state of the system, filter out irrelevant random noise, and retain the core signals of the system's dynamic evolution. The phase space reconstruction module is configured to determine the optimal delay time. and embedding dimension The system dynamic trajectory is reconstructed by mapping a one-dimensional pure time-series signal to a high-dimensional phase space. The chaotic feature quantification calculation module is configured to quantify the chaotic evolution state of the system through real-time calculation of the maximum Lyapunov exponent in subsystem resource scenarios. The dynamic state discrimination and early warning module is configured to divide the system into three dynamic regions—steady state, bifurcation, and chaos—based on the maximum Lyapunov exponent and its changing trend. The entropy-driven task blocking and scheduling execution module is configured to execute the low-priority task circuit breaker mechanism in the bifurcation zone, the micro-restart in the chaotic zone, and the process isolation active reset strategy according to the system's dynamic state and warning level.

[0008] Preferably, the univariate time-series data acquisition and denoising module specifically includes: The data collected includes CPU interrupt response latency or memory bus bandwidth utilization. The sampling frequency setting must simultaneously meet two requirements: first, it must satisfy the Nyquist sampling theorem; second, it must meet the requirement of capturing the dynamic characteristics of the system. Nyquist sampling theorem: Let the highest frequency of the sampled signal be... Then the sampling frequency Must meet ; Capture system dynamics characteristics: The sampling frequency is higher than the preset minimum sampling frequency; The acquired continuous numerical signals contain high-frequency random noise, which is denoised using wavelet transform, including: Signal decomposition: The acquired one-dimensional time-series signal is decomposed into 5 layers; Thresholding: For values ​​whose absolute value is less than a threshold... The detail factor is set to 0, and the absolute value is greater than 0. Detail factor minus ; Signal reconstruction: Perform inverse wavelet transform on the processed detail coefficients and approximation coefficients to obtain the denoised clean time series signal.

[0009] Preferably, the phase space reconstruction module specifically includes: The collected one-dimensional time series is , , The number of sampling points, and the reconstructed phase space vector. ,for: ; in, For delay time; For the embedding dimension; A phase space vector; The optimal delay time is calculated using the mutual information method. The specific calculation steps and parameters are as follows: For the original time series ,calculate and Mutual information between ; Drawing mutual information Follow The changing curve, when the mutual information first reaches the preset minimum value, corresponds to... This is the optimal delay time.

[0010] Preferably, the method further includes an optimal embedding dimension. Determination: Embedding dimension The initial value is 1, and the maximum value is 10; For each embedding dimension Calculate all vectors in phase space nearest neighbor Then increase the embedding dimension to ,calculate exist nearest neighbor in phase space ,like and If the rate of change of distance exceeds a preset threshold, then it is determined that... For false neighbors, calculate the proportion of false neighbors to the total number of neighbors; Gradually increase the embedding dimension The calculation is performed on the proportion of false neighbors to the total number of neighbors. The calculation stops when the proportion of false neighbors first falls below a preset threshold. This is the optimal embedding dimension.

[0011] Preferably, the chaotic feature quantization calculation module specifically includes: Select the trajectory vector after phase space reconstruction Data is extracted using a sliding window. For each vector within the window Select the k nearest neighbors of the trajectories in phase space and record the initial distance between the adjacent trajectories. ; Track the evolution of adjacent trajectories over time and calculate the distance at each time step. Calculate the divergence rate of a single trajectory. Repeatedly calculate the divergence rate of all adjacent trajectories; The average divergence rate of all trajectories is used to obtain the current window's [divergence rate]. Values ​​are updated via a sliding window and output in real time. result.

[0012] Preferably, the method further includes: Using a sliding window, the trajectory vector after phase space reconstruction is extracted. ; Calculate the autocorrelation function of the trajectory vector. ; For autocorrelation function Taking the natural logarithm, we get ,draw Follow The slope of the changing curve is... value.

[0013] Preferably, the dynamic state discrimination and early warning module specifically includes: Combination Based on the numerical characteristics and operational laws of the computer system, the system dynamics are divided into three distinct regions: the steady-state region, the bifurcation region, and the chaotic region. Each region is quantified. The judgment criteria, corresponding system status, and failure risks specifically include: Steady-state region: The value remains consistently negative and is stable at [value range]. ≤-0.1; bifurcation zone: It oscillates around 0, i.e., -0.1 < <0.1; Chaos Zone: It remains positive and the value gradually increases, that is... ≥0.1.

[0014] Preferably, the method further includes: To avoid due to Random fluctuations can lead to misjudgments. To address this, a dynamic discrimination approach combined with historical trajectory backtracking is employed. The specific steps are as follows: Real-time acquisition of the output of the chaotic feature quantization module Value, recording the most recent h samples Trajectory; Based on the judgment criteria for the three regions, the current system status is preliminarily determined; If it is initially determined to be a bifurcation region / chaotic region, backtrack to the nearest... sampling points If the trend of change is consistent with the current preliminary judgment, then the state judgment result is confirmed; if the trend is inconsistent, then the judgment is changed to... Occasional fluctuations do not trigger warnings; only the fluctuation details are recorded. Based on the dynamic state, a three-level early warning mechanism is set up, with each level corresponding to different early warning signals, early warning methods, and processing priorities.

[0015] Preferably, the entropy-driven task blocking and scheduling execution module specifically includes: Two core scheduling strategies are built-in, corresponding to the bifurcation region and the chaotic region, respectively: Strategy A, corresponding to the bifurcation region: Terminate all currently executing low-priority tasks and record their termination status; Pause all low-priority tasks pending execution and add them to the task queue; Allocate CPU, memory, and bus resources used by low-priority tasks to high-priority core business tasks. After the pause time ends, or when the system recovers from the bifurcation region to the steady state region, the execution of low-priority tasks will automatically resume, and unfinished tasks will be resumed first according to the order of the task queue.

[0016] Preferably, the method further includes: Strategy B, corresponding to the chaotic region: Identify the core system processes, record their running status and data, then terminate them, restart them immediately, and restore their running status and data after restarting. By tracing the anomaly logs, the cause can be located. For abnormal processes that have increased in frequency, migrate them to a separate container and limit their CPU and memory usage; at the same time, restart the core processes that the abnormal processes depend on to restore the normal system logic. After process isolation, the system automatically monitors the status of abnormal processes. If the abnormal process recovers, it is removed from the container and normal resource allocation is restored. If the abnormal process continues to be abnormal, the process is automatically terminated and a fault notification is sent to the administrator.

[0017] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention achieves single-variable micro-time series signal acquisition through Takens embedding theorem, and can fully characterize the global dynamic state of the system by only using CPU interrupt response delay and memory bus bandwidth utilization, thereby reducing data transmission, storage and computation overhead; and achieves accurate quantification of steady state, bifurcation and chaotic state by calculating the maximum Lyapunov exponent.

[0018] 2. This invention captures early, minute anomalies such as hidden hardware faults and software conflict precursors that are imperceptible by traditional technologies. It characterizes the instability evolution law of the system from the essence of nonlinear dynamics, upgrades fault prediction from passive threshold alarm to active quantitative prediction, significantly improves the sensitivity of system anomaly perception and the timeliness of early warning, and provides core theoretical and technical support for high-reliability computing scenarios. Attached Figure Description

[0019] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1This is a system structure diagram of the present invention. Detailed Implementation

[0020] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0021] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0022] Example 1

[0023] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0024] Appendix Figure 1 The block diagram of the intelligent scheduling system for system maintenance tasks based on artificial intelligence provided in the embodiments of the present invention shows the connection relationship between the single-variable time-series data acquisition and denoising module and the entropy-driven task blocking and scheduling execution module, and marks the main functional interaction flow of each module.

[0025] In this embodiment, it includes: The single-variable time-series data acquisition and denoising module is configured to acquire single continuous numerical physical indicators that reflect the micro-dynamic state of the system, such as CPU interrupt response delay, in real time at high frequency. It filters out irrelevant random noise through the soft threshold denoising method of db4 wavelet transform, retains the core signals of the system's dynamic evolution, and achieves the dual goals of low transmission cost and high effectiveness of monitoring data. Specifically, it includes: The core principle for selecting data collection objects is that a single variable can reflect the overall system. Based on the characteristics of Takens' theorem, priority is given to physical indicators that best reflect the system's microscopic dynamic state and are most sensitive to system faults. Specifically, this includes two types of core indicators, which can be flexibly selected according to the actual application scenario (servers, industrial control computers, embedded systems, etc.), while clearly defining the selection criteria: The primary metric is CPU interrupt latency. The selection criteria are as follows: As the core computing unit of a computer system, the CPU's interrupt latency directly reflects the coordinated state of the system's hardware (CPU, bus) and software (interrupt controller, process scheduler). Any minor hardware malfunction (such as corrupted CPU cache or abnormal bus transmission) or software logic conflict (such as precursors to process deadlock or disordered interrupt priorities) will cause slight fluctuations in interrupt latency. These fluctuations are the core characteristics of chaotic system evolution and can accurately capture changes in the system's internal state. The specific sampling range is 0-100μs (for a normal stable system), and can exceed 100μs under abnormal conditions.

[0026] Alternative indicator: Memory bus bandwidth utilization. Selection criteria: The memory bus is the core channel connecting the CPU and memory. Continuous changes in its bandwidth utilization directly reflect core anomalies such as memory read / write efficiency, process memory access frequency, and memory leaks. In particular, system crashes caused by memory leaks will show irregular fluctuations (chaotic characteristics) in memory bus bandwidth utilization before the crash, and such fluctuations can be captured by continuous sampling; the specific sampling range is 0-100%, and the sampling accuracy is 0.01%.

[0027] Methods for determining the sampling frequency: The sampling frequency setting must simultaneously meet two core requirements: first, it must satisfy the Nyquist sampling theorem (to avoid signal aliasing and ensure that the acquired signal can accurately reproduce the original signal); second, it must meet the requirement of capturing system dynamic characteristics (ensuring that it can capture small fluctuations in interrupt response delay and memory bus bandwidth utilization, the period of which is usually between 10 and 100 ms). The specific setting method is as follows: Nyquist sampling theorem: Let the highest frequency of the sampled signal be... (The highest fluctuation frequency of CPU interrupt response latency is 10Hz, i.e., a fluctuation period of 100ms), then the sampling frequency... Must meet That is, the minimum sampling frequency is 20Hz (50ms / time); Capturing system dynamics: To ensure complete capture of the minute fluctuations in the chaotic evolution of the system, the sampling frequency should be higher than the preset minimum sampling frequency. It is recommended to set it to 100ms / time (10Hz). This can avoid data redundancy and increased transmission pressure caused by excessively high sampling frequency, and also avoid loss of dynamics due to excessively low sampling frequency. Adjustable design: The module supports flexible adjustment of the sampling frequency (50ms / time to 500ms / time) according to the system type. During adjustment, it automatically adapts to the Nyquist sampling theorem without manual intervention, enhancing the versatility of the solution.

[0028] The specific implementation of noise reduction processing: The acquired continuous numerical signals contain a large amount of high-frequency random noise (such as small deviations in CPU interrupt delay caused by power grid fluctuations and fluctuations in memory bus bandwidth caused by external electromagnetic interference). This noise will interfere with the capture of the system dynamic trend. Therefore, it is necessary to filter out irrelevant noise and retain the core trend signal through noise reduction processing. This module uses wavelet transform as the core denoising algorithm. Compared to traditional low-pass filtering and mean filtering, wavelet transform can remove noise while preserving the signal's abrupt change characteristics (key features of chaotic system evolution), including: Signal decomposition: The db4 wavelet is used (the selection criteria for the wavelet basis: the db4 wavelet has good time-frequency locality and can accurately distinguish noise from signal change features) to decompose the acquired one-dimensional time series signal into 5 layers (the decomposition layer criteria: 5 layers of decomposition can fully separate noise from the core signal; too many decomposition layers will lead to signal distortion, and too few layers will not be able to completely remove noise). Thresholding: Soft thresholding is applied to the detail coefficients (high-frequency components, mainly noise) of each decomposed layer. The threshold is set to... ,in The standard deviation of noise ( , (for the first level of detail coefficients) The number of sampling points (e.g., 100ms / sample, 600 points sampled per minute). =600); The core logic of soft threshold processing is: to handle values ​​whose absolute value is less than the threshold. The detail factor is set to 0, and the absolute value is greater than 0. Detail factor minus To avoid signal abrupt distortion caused by hard thresholding; Signal reconstruction: The processed detail coefficients and approximation coefficients (low-frequency components, core dynamic trend signals) are subjected to inverse wavelet transform to obtain a denoised and clean time-series signal, which is used in the subsequent phase space reconstruction module.

[0029] The phase space reconstruction module is configured to use Takens' embedding theorem as its core theoretical support and automatically determine the optimal delay time through mutual information and spurious neighbor methods. and embedding dimension By mapping a one-dimensional pure time series signal to a high-dimensional phase space, the dynamic trajectory of the system can be reconstructed, and the effectiveness of the reconstruction can be verified by the correlation dimension, providing a geometric basis for the quantification of chaotic features. Specifically, it includes: One-dimensional pure timing signals (CPU interrupt response delay or memory bus bandwidth utilization timing signals) are mapped to a high-dimensional phase space through mathematical transformations to reconstruct a trajectory (i.e., an attractor) that can truly reflect the internal dynamics of the computer system. The essence of phase space reconstruction is to transform a one-dimensional time domain signal into a geometric structure in a high-dimensional space. Through changes in the geometric structure, the three states of the system—stable, periodic, and chaotic—are intuitively presented, providing a geometric basis for subsequent quantification of chaotic features.

[0030] The core theoretical support is the Takens embedding theorem (proposed by mathematician Floris Takens, a core theorem in the field of nonlinear dynamics), the core content of which is: For an n-dimensional chaotic dynamical system, it is only necessary to collect a continuous and measurable univariate time series of the system. By using the delayed embedding method, it is possible to... In 3D phase space ( The reconstructed dynamic trajectory is equivalent to the original system's topology; that is, the time-series signal of a single variable can completely reflect the entire system. The evolutionary law of the dimensional system is also the core theoretical basis of the single-variable data collection in this scheme, which completely distinguishes it from the technical logic of traditional multi-indicator data collection.

[0031] The specific mathematical expression is as follows: The collected one-dimensional time series is , , The number of sampling points, and the reconstructed phase space vector. , ,for: ; The physical meaning and setting basis of each parameter are as follows: The delay time represents the time interval between two adjacent elements in the phase space vector in the original time series. Its core function is to ensure the independence of each element in the phase space vector, avoid linear correlation between elements, and ensure that the reconstructed trajectory can truly reflect the dynamic characteristics of the system. The embedding dimension represents the dimension of the phase space. Its core function is to ensure that the reconstructed trajectory can untangle the dynamic trajectory knot of the original system, avoid the overlap of trajectories in the phase space, and ensure that the reconstructed trajectory is topologically equivalent to the original system. Let each vector be a point in the phase space. The set of all phase space vectors is the reconstructed system dynamic trajectory (attractor).

[0032] Determining the optimal delay time τ: The optimal delay time is calculated using the mutual information method. Compared to the traditional autocorrelation function method (which can only capture linear correlations), the mutual information method can capture nonlinear correlations in time series, making it more suitable for the characteristics of chaotic systems. The specific calculation steps and parameters are as follows: Define mutual information: for the original time series ,calculate and Mutual information between The physical meaning of mutual information is Includes about The amount of information is calculated using the following formula: ,in for The marginal probability, for and The joint probability; Parameter settings: The value range is 1~50 (corresponding to a sampling frequency of 100ms / time). =1, which means 100ms. =50, which is 5s), the probability calculation uses the histogram method, which divides the range of values ​​of the time series into 20 intervals (the number of intervals is based on the fact that 20 intervals can balance calculation accuracy and efficiency). optimal Determine: Draw mutual information Follow The changing curve, when the mutual information first reaches the preset minimum value, corresponds to... This is the optimal delay time; the core logic is: at this point... and It exhibits the lowest correlation and strongest independence, avoiding redundancy in elements of the phase space vector and ensuring the accuracy of the reconstructed trajectory. Experiments have verified that for CPU interrupt response delay timing signals, the optimal [method / mechanism] is [suitable / effective]. Typically 5~10 (500ms~1s).

[0033] Also includes optimal embedding dimension Determination: The optimal embedding dimension is calculated using the spurious neighbor method. The core logic of this method is that as the embedding dimension *m* increases, the originally overlapping trajectories in the phase space gradually unravel, and the proportion of spurious neighbors (false neighboring points caused by trajectory overlap) gradually decreases. When the proportion of spurious neighbors drops below a preset threshold, the corresponding... This is the optimal embedding dimension. The specific calculation steps and parameters are as follows: Embedding dimension The initial value is 1, and the maximum value is 10 (according to Takens' theorem). , The chaos dimension of a system is typically 2 to 3. A maximum value of 10 is sufficient to meet the requirements. The threshold for the proportion of false neighbor points is set to 5% (a 5% threshold can balance reconstruction accuracy and computational efficiency; below 5%, the reconstructed trajectory has no significant optimization, and above 5%, the trajectory overlaps). False neighbor computation: for each embedding dimension Calculate all vectors in phase space nearest neighbor Then increase the embedding dimension to ,calculate exist nearest neighbor in phase space ,like and If the rate of change of distance exceeds a preset threshold (usually 10), then it is determined that... For false neighbors, calculate the proportion of false neighbors to the total number of neighbors; optimal Determine: Gradually increase the embedding dimension The calculation is performed on the proportion of false neighbors to the total number of neighbors. The calculation stops when the proportion of false neighbors first drops below a preset threshold (5%). This is the optimal embedding dimension; for timing signals related to CPU interrupt response latency and memory bus bandwidth utilization, the optimal... It is usually 3 to 5.

[0034] To ensure the accuracy of the reconstructed trajectory, the module adds a reconstruction verification step: calculate the correlation dimension of the reconstructed trajectory. If the correlation dimension is a fixed value (usually 2~3, which conforms to the chaotic characteristics of computer systems), the reconstruction is deemed valid; if the correlation dimension is not a fixed value, denoising processing or sampling parameter adjustment is performed again to ensure the input quality of subsequent modules.

[0035] The chaotic feature quantification calculation module is configured to perform real-time and accurate calculation of the maximum Lyapunov exponent in subsystem resource scenarios for the reconstructed high-dimensional dynamic trajectory by improving the Wolf algorithm and Rosenstein algorithm. It quantifies the chaotic evolution state of the system and provides core quantitative basis for system fault prediction. Specifically, it includes: Quantitative analysis is performed on the high-dimensional dynamic trajectories (attractors) output by the phase space reconstruction module to extract core mathematical indicators that characterize the chaotic state of the system. The core objective is to calculate the maximum Lyapunov exponent. ,pass The magnitude and trend of the values ​​can be used to accurately determine the evolution of the system's reconstruction trajectory, including convergence (system stability), periodicity (system in a cyclical state with no risk of failure), or divergence (system entering a chaotic state, indicating an impending failure), providing a quantitative basis for subsequent state identification and early warning.

[0036] Core quantitative indicator: Maximum Lyapunov index: The maximum Lyapunov exponent is a core quantitative indicator in the field of chaotic dynamics used to determine the chaotic state of a system. Its physical meaning is the average exponential divergence rate of two adjacent trajectories in phase space over time. It accurately characterizes the system's sensitivity to initial conditions. Computer system failures (crashes, freezes, shutdowns) are essentially amplifications of the system's sensitivity to small initial anomalies (such as minor interruption delay fluctuations or memory leaks), ultimately leading to trajectory divergence and entry into a chaotic state. The ability to accurately capture this evolutionary process is based on the following specific rules and their physical significance: <0: The system is in a stable state. Adjacent trajectories in the phase space will gradually converge to a fixed point (fixed point), indicating that the internal dynamics of the system are stable and there is no risk of failure. This corresponds to the normal operating state of a computer system. =0: The system is in a stable boundary or periodic motion state. The adjacent trajectories in the phase space neither converge nor diverge, but change periodically. This indicates that there is a minor anomaly in the system (such as a slight process scheduling conflict), but it has not yet entered a chaotic state. This is a warning precursor. >0: The system enters a chaotic state. Adjacent trajectories in the phase space will diverge at an exponential rate, indicating that the system is highly sensitive to initial conditions (butterfly effect). Small initial anomalies will be rapidly amplified, indicating that the system is about to fail (crash, freeze, collapse), and immediate intervention is required. Additional notes: The magnitude of this value is positively correlated with the risk of failure, that is... The larger the value, the faster the trajectory diverges and the shorter the time to failure (e.g., When the value is 0.1, the fault may occur in 10 minutes. When the value is 0.5, the fault may occur in 1 minute, providing a basis for setting subsequent early warning priorities.

[0037] Small data volume method: This algorithm is an optimization of the traditional Wolf algorithm. The core improvement is the use of a sliding window for computation, which reduces reliance on historical data and increases computation speed. The specific steps and parameters are as follows: Data preparation: Selecting trajectory vectors after phase space reconstruction Data is extracted using a sliding window, with the window size set to 100-500 sampling points (100-500 sampling points balances computational accuracy and real-time performance; a window that is too small will result in insufficient accuracy, while a window that is too large will lead to excessive computational latency). The window sliding step size is set to 10 sampling points (1 second, to ensure real-time updates). result); Adjacent trajectory selection: For each vector within the window Select its k nearest neighbors in phase space (k=5~10, balancing computational accuracy and efficiency), and record the initial distance between adjacent trajectories. ; Divergence rate calculation: Track the evolution of adjacent trajectories over time and calculate the distance at each time step. According to the formula Calculate the divergence rate of a single trajectory Repeatedly calculate the divergence rate of all adjacent trajectories; Calculation: Take the average of the divergence rates of all trajectories to obtain the current window's... Values ​​are updated via a sliding window and output in real time. result; This includes adding an outlier filtering step, which treats trajectories with a divergence rate exceeding a reasonable range (-0.5 to 0.5) as abnormal trajectories and excludes them from the evaluation process. Calculations should be performed to avoid extreme values ​​affecting calculation accuracy.

[0038] Rosenstein algorithm: The core advantages are low computational complexity and low resource consumption; it does not require selecting adjacent trajectories and directly calculates the autocorrelation characteristics of the time series. The specific steps and parameters are as follows: Data preparation: Using the same sliding window (100-500 sampling points) as the small data volume method, the trajectory vector after phase space reconstruction is extracted. ; Autocorrelation function calculation: Calculate the autocorrelation function of the trajectory vector. The formula is ,in The average value of the trajectory vector; Calculate: For the autocorrelation function Taking the natural logarithm, we get ,draw Follow The slope of the changing curve is... value; The slope of the curve is calculated using linear fitting, and the fitting interval is selected. =10~50 sampling points to avoid excessive fitting deviation and ensure calculation accuracy.

[0039] Algorithm selection logic: The module has built-in algorithm selection logic that can automatically select an algorithm based on the system's CPU utilization and memory size: when the system has ≥30% idle resources, it automatically selects the small data volume method (high accuracy); when the system has <30% idle resources, it automatically switches to the Rosenstein algorithm (low resource consumption); it also supports manual algorithm switching to adapt to different application scenarios.

[0040] The Wolf algorithm is improved by using a sliding window combined with outlier filtering to achieve... Real-time computation solves the technical shortcomings of the traditional Wolf algorithm, such as high computational complexity and inability to output in real time; It also provides two algorithms adapted to different resource scenarios, enhancing the versatility of the solution and giving it a significant advantage over existing technologies that only support a single algorithm.

[0041] The dynamic state discrimination and early warning module is configured to divide the system into three dynamic regions—steady state, bifurcation, and chaos—based on the maximum Lyapunov exponent and its changing trend. It accurately determines the system state through dynamic discrimination and historical trajectory backtracking, sets up a three-level hierarchical early warning mechanism corresponding to the state, and realizes anomaly tracing to provide a clear control basis for subsequent intervention operations. Specifically, it includes: Output based on chaotic feature quantization module The numerical values ​​and trends, combined with preset state judgment rules, accurately determine the current health state of the computer system (steady state, bifurcation state, chaotic state), and output corresponding early warning signals (tiered early warning) according to different states. At the same time, the state change trajectory is recorded to provide clear control basis for subsequent task scheduling and execution modules. The core objective is to predict faults in advance, provide tiered early warnings, and accurately locate anomalies to achieve early detection, early warning, and early intervention, and avoid serious faults such as system crashes and shutdowns.

[0042] Combination Based on the numerical characteristics and operational laws of the computer system, the system dynamics are divided into three distinct regions: the steady-state region, the bifurcation region, and the chaotic region. Each region is quantified. The judgment criteria, corresponding system status, and failure risks specifically include: Steady-state region: The value remains consistently negative and is stable at [value range]. ≤-0.1 (quantitative standard); Corresponding system state: The internal dynamics of the system are stable, there are no hidden anomalies, the hardware and software operate normally in coordination, and the risk of failure is 0; Judgment conditions: All 30 consecutive sampling points (3s) meet the requirements If the value is ≤-0.1, it is determined that the system has entered the steady-state region.

[0043] bifurcation zone: It oscillates around 0, i.e., -0.1 < <0.1 (quantitative standard), or a sudden mutation occurs (such as...) The value instantly increased from -0.2 to 0.05; corresponding system state: the system is on the verge of a kinetic phase transition, similar to the critical point where water is about to freeze. Minor latent anomalies have appeared internally (such as precursors to memory leaks, minor CPU interrupt controller anomalies, and process scheduling conflicts), but these have not yet affected normal system operation. The failure risk is 30%~70% (depending on the situation). (Approaching 0.1 and then increasing); Judgment criteria: Ten consecutive sampling points (1s) satisfy -0.1 < <0.1, or a single sample If the mutation magnitude exceeds 0.1 (e.g., from -0.15 to 0.05), it is determined to have entered the bifurcation region.

[0044] Chaos Zone: It remains positive and the value gradually increases, that is... ≥0.1 (quantitative standard); corresponding system state: the system enters a chaotic state, is highly sensitive to initial conditions, internal hidden anomalies have been amplified, the trajectory diverges rapidly, indicating that the system is about to experience serious failures such as crashes, freezes, and breakdowns, with a failure risk ≥80%. The higher the value, the faster the fault occurs; judgment criteria: Five consecutive sampling points (0.5s) satisfy the condition. If the value is ≥0.1, it is determined that the region has entered the chaotic region.

[0045] Also includes: State determination logic: To avoid due to Random fluctuations can lead to misjudgments. To address this, a dynamic discrimination approach combined with historical trajectory backtracking is employed. The specific steps are as follows: Real-time acquisition of the output of the chaotic feature quantization module The value records the values ​​of the most recent h (h = 60) sampling points (6 seconds). Trajectory; Based on the judgment criteria for the three regions, the current system status is preliminarily determined; Historical trajectory backtracking verification: If initially determined to be a bifurcation region / chaotic region, backtrack to the most recent... 30 sampling points (3s) The trend of change, if the trend is consistent with the current preliminary judgment (e.g.) If the trend continues to rise or oscillates around 0, then the status determination result is confirmed; if the trend is inconsistent (e.g., ... If it rises to 0.1 by chance and then quickly falls back to below -0.1, it is judged as... Occasional fluctuations do not trigger warnings; only the fluctuation details are recorded. State transition determination: When the system transitions from the steady state region to the bifurcation region, from the bifurcation region to the chaotic region, or from the chaotic region to the bifurcation region, or from the bifurcation region to the steady state region, the state must be consistent for 5 consecutive sampling points (0.5s) to avoid misjudgment caused by instantaneous switching and to ensure the stability of state determination.

[0046] Based on the system's dynamic state, a three-level early warning mechanism is set up, with each level corresponding to different early warning signals, early warning methods, and processing priorities to ensure the accuracy and practicality of the early warning system, as detailed below: Level 1 Warning (Low Risk, corresponding to the initial stage of the bifurcation zone): At −0.1 Oscillations between <0 indicate a 30%~50% risk of failure; Warning signal: System background logs record warning information (including...). (Value values, status change time, preliminary anomaly analysis), does not pop up a front-end warning window, and does not affect the normal operation of the system; processing priority: low, only requires background monitoring, no manual intervention required; Level 2 warning (medium risk, corresponding to the later stage of the bifurcation zone): In 0≤ The system oscillates between <0.1, indicating a 50%~70% risk of failure. Warning signals include: background log recording and a pop-up light warning window (indicating a hidden anomaly in the system, indicating an impending high-risk state, please pay attention), and an SMS / email warning sent to the system administrator. Priority: Medium; it is recommended that the administrator monitor the system status, but no immediate intervention is required. Level 3 Warning (High Risk, Corresponding to Chaos Zone): If the value is ≥0.1 and continues to rise, the risk of failure is ≥80%; Warning signals: background log recording plus a front-end pop-up emergency warning window (red warning, indicating that the system is about to crash, please handle immediately), administrator SMS / email warning, system beeping prompt; Handling priority: high, requires immediate intervention from the subsequent task scheduling and execution module to avoid system failure.

[0047] The entropy-driven task blocking and scheduling execution module is configured with entropy-driven core logic. Based on the system's dynamic state and warning level, it executes the low-priority task circuit breaker mechanism in the bifurcation zone, the micro-restart in the chaotic zone, and the process isolation active reset strategy, respectively. It can also adaptively adjust the scheduling parameters according to the system type to achieve proactive intervention and precise recovery of the system, minimizing the impact on core business. Specifically, it includes: As the execution terminal of this solution, the module performs corresponding physical and software-level intervention operations based on the system state (steady-state region, bifurcation region, chaotic region) and warning level output by the dynamic state discrimination and early warning module. The core logic is entropy-driven. By blocking low-priority tasks and actively resetting the system state, the entropy value (disorder) inside the system is reduced, and the dynamic trajectory of the system is forcibly pulled back to the steady-state attractor region, so as to prevent the system from entering the chaotic region or recovering from the chaotic region to the stable state. The core objective is to intervene quickly, minimize impact, and proactively recover, ensuring stable system operation and reducing losses caused by failures.

[0048] Additional explanation: Here, entropy refers to the degree of disorder within a system. When a system enters a bifurcation or chaotic region, the entropy value increases (disorder increases). By performing operations such as task blocking and active reset, resource competition and logical conflicts within the system can be reduced, the entropy value can be lowered, and the system can be restored to order (steady state).

[0049] The module has two built-in core scheduling strategies, corresponding to the bifurcation region and the chaotic region respectively, and also supports adaptive adjustment of the strategies. The specific implementation details are as follows: Strategy A (circuit breaker mechanism), corresponding to the bifurcation zone (level 1 and level 2 warnings): Core objective: To proactively block low-priority tasks, reduce system resource contention, lower system entropy, and prevent the system from further slipping into a chaotic state, while ensuring the normal operation of core system functions; specific execution steps and parameters are as follows: Task Prioritization: All tasks in the system are pre-classified into three priority levels to clearly define the scope of blocking and avoid mistakenly blocking core tasks. High-priority tasks: Core business tasks (such as database queries, core process operations, and user interaction tasks) must not be blocked to ensure the normal operation of core businesses; Medium priority tasks: Regular system tasks (such as system updates, process scheduling, and core log recording) can be temporarily paused (maximum pause time 30 minutes) without interruption; Low-priority tasks: Non-core maintenance tasks (such as log archiving, system backup, defragmentation, and memory reclamation of non-core processes) can be directly blocked to free up resources for core tasks.

[0050] Blocking trigger condition: When the system enters the bifurcation zone and triggers a level 1 or level 2 warning, the circuit breaker mechanism will be triggered immediately; Specific blocking procedures: Immediately terminate all low-priority tasks in progress and record their termination status (such as log archiving progress and backup completion status) to facilitate subsequent recovery. Pause all low-priority tasks pending execution and add them to the task queue. Set the pause time to 10-30 minutes (adjustable according to the warning level: 10 minutes for a level 1 warning and 30 minutes for a level 2 warning). Resource release: CPU, memory, and bus resources occupied by low-priority tasks are allocated to high-priority core business tasks to ensure the stability of core business operations. Task recovery logic: After the pause time ends, or when the system recovers from the bifurcation zone to the steady state zone, the execution of low-priority tasks will be automatically resumed. In the order of the task queue, unfinished tasks (such as continued log archiving and backup) will be resumed first, without manual intervention.

[0051] 10. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 9, characterized in that it further comprises: Strategy B (Active Reset), corresponding to the chaos zone (Level 3 Warning): Core objective: To quickly pull the system back from the chaotic region to the steady state region, preventing system crashes and shutdowns, minimizing downtime, and protecting core data from loss; specific execution steps and parameters are as follows: Active reset methods (two methods, automatic selection): Preferred approach: High-priority micro-reboot, which only restarts the system's core processes (such as interrupt controller processes, memory management processes, and process scheduler processes), without restarting the entire machine. The restart time is controlled within 100ms-500ms (ensuring that core business operations are not interrupted and users are unaware of the process). Applicable scenario: When the system's core business operations are running and shutdown is not an option. Specific steps: Identify the core system processes (which can be configured in advance, such as the systemd process in Linux and the winlogon process in Windows), record the running status and data of the core processes, then terminate the core processes, restart them immediately, and restore the running status and data of the core processes after restarting. Parameter settings: The micro-reboot timeout is set to 500ms. If the core process fails to restart successfully within 500ms, it will automatically switch to process isolation mode.

[0052] Alternative approach: Process isolation, which will cause abnormal processes to enter the chaotic region of the system (through...). The process (including tracking changes and tracing logs) is isolated to an independent container environment (such as a Docker container), limiting its resource consumption, and restarting the core dependent processes corresponding to the abnormal process. Applicable scenarios: micro-restart failure, or the abnormal process is clearly identified (such as a third-party application process). Specific steps: Locate the cause by tracing the exception logs. For abnormal processes with elevated CPU usage, migrate them to an independent container and limit their CPU usage to ≤10% and memory usage to ≤10% to prevent them from interfering with core system processes; at the same time, restart the core processes that the abnormal process depends on (such as database processes and network processes) to restore normal system logic. Follow-up processing: After process isolation, the system automatically monitors the status of the abnormal process. If the abnormal process recovers to normal ( If the process drops to a stable state, it will be removed from the container and normal resource allocation will be restored; if the abnormal process continues to be abnormal, the process will be automatically terminated and a fault notification will be sent to the administrator.

[0053] Data protection mechanism: During the active reset process, core data (such as user interaction data, temporary database data, and core process running data) is automatically backed up to an independent memory buffer to avoid data loss during restart and isolation; if the reset fails, the system state can be restored through the backup data.

[0054] Post-reset verification: After the active reset is completed, monitor the subsequent 10 sampling points (1s). Value, if Reduced to steady state region ( If ≤−0.1), then the reset is considered successful; if If the system is still in the chaotic zone, the active reset operation will be repeated (up to 3 times). If all 3 attempts fail, an emergency fault notification will be sent to the administrator, suggesting manual intervention.

[0055] Adaptive adjustment of scheduling strategy: The module has built-in adaptive adjustment logic that can automatically adjust scheduling parameters based on system type and core business requirements. Server scenario: Micro-rebooting is preferred, and the pause time for low-priority tasks in the circuit breaker mechanism is extended to 30 minutes to ensure the continuity of core business. Embedded system scenarios: Process isolation should be prioritized, and low-priority tasks triggered by the circuit breaker mechanism should be terminated directly (without recovery) to reduce resource consumption; Industrial control computer scenario: Micro-reboot time is reduced to 100ms to ensure uninterrupted industrial control tasks, while increasing the frequency of data backup (backing up core data every 100ms).

[0056] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0057] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0058] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0059] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0060] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0061] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0064] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0065] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An intelligent scheduling system for system maintenance tasks based on artificial intelligence, characterized in that, include: The single-variable time series data acquisition and denoising module is configured to acquire physical indicators that reflect the micro-dynamic state of the system, filter out irrelevant random noise, and retain the core signals of the system's dynamic evolution. The phase space reconstruction module is configured to determine the optimal delay time. and embedding dimension The system dynamic trajectory is reconstructed by mapping a one-dimensional pure time-series signal to a high-dimensional phase space. The chaotic feature quantification calculation module is configured to quantify the chaotic evolution state of the system through real-time calculation of the maximum Lyapunov exponent in subsystem resource scenarios. The dynamic state discrimination and early warning module is configured to divide the system into three dynamic regions—steady state, bifurcation, and chaos—based on the maximum Lyapunov exponent and its changing trend. The entropy-driven task blocking and scheduling execution module is configured to execute the low-priority task circuit breaker mechanism in the bifurcation zone, the micro-restart in the chaotic zone, and the process isolation active reset strategy according to the system's dynamic state and warning level.

2. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 1, characterized in that, The univariate time series data acquisition and denoising module specifically includes: The data collected includes CPU interrupt response latency or memory bus bandwidth utilization. The sampling frequency setting must simultaneously meet two requirements: first, it must satisfy the Nyquist sampling theorem; second, it must meet the requirement of capturing the dynamic characteristics of the system. Nyquist sampling theorem: Let the highest frequency of the sampled signal be... Then the sampling frequency Must meet ; Capture system dynamics characteristics: The sampling frequency is higher than the preset minimum sampling frequency; The acquired continuous numerical signals contain high-frequency random noise, which is denoised using wavelet transform, including: Signal decomposition: The acquired one-dimensional time-series signal is decomposed into 5 layers; Thresholding: For values ​​whose absolute value is less than a threshold... The detail factor is set to 0, and the absolute value is greater than 0. Detail factor minus ; Signal reconstruction: Perform inverse wavelet transform on the processed detail coefficients and approximation coefficients to obtain the denoised clean time series signal.

3. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 1, characterized in that, The phase space reconstruction module specifically includes: The collected one-dimensional time series is , , The number of sampling points, and the reconstructed phase space vector. ,for: ； in, For delay time; For the embedding dimension; A phase space vector; The optimal delay time is calculated using the mutual information method. The specific calculation steps and parameters are as follows: For the original time series ,calculate and Mutual information between ; Drawing mutual information Follow The changing curve, when the mutual information first reaches the preset minimum value, corresponds to... This is the optimal delay time.

4. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 3, characterized in that, Also includes optimal embedding dimension Determination: Embedding dimension The initial value is 1, and the maximum value is 10; For each embedding dimension Calculate all vectors in phase space nearest neighbor Then increase the embedding dimension to ,calculate exist nearest neighbor in phase space ,like and If the rate of change of distance exceeds a preset threshold, then it is determined that... For false neighbors, calculate the proportion of false neighbors to the total number of neighbors; Gradually increase the embedding dimension The calculation is performed on the proportion of false neighbors to the total number of neighbors. The calculation stops when the proportion of false neighbors first falls below a preset threshold. This is the optimal embedding dimension.

5. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 1, characterized in that, The chaotic feature quantization calculation module specifically includes: Select the trajectory vector after phase space reconstruction Data is extracted using a sliding window. For each vector within the window Select the k nearest neighbors of the trajectories in phase space and record the initial distance between the adjacent trajectories. ; Track the evolution of adjacent trajectories over time and calculate the distance at each time step. Calculate the divergence rate of a single trajectory. Repeatedly calculate the divergence rate of all adjacent trajectories; The average divergence rate of all trajectories is used to obtain the current window's [divergence rate]. Values ​​are updated via a sliding window and output in real time. result.

6. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 5, characterized in that, Also includes: Using a sliding window, the trajectory vector after phase space reconstruction is extracted. ; Calculate the autocorrelation function of the trajectory vector. ; For autocorrelation function Taking the natural logarithm, we get ,draw Follow The slope of the changing curve is... value.

7. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 1, characterized in that, The dynamic state discrimination and early warning module specifically includes: Combination Based on the numerical characteristics and operational laws of the computer system, the system dynamics are divided into three distinct regions: the steady-state region, the bifurcation region, and the chaotic region. Each region is quantified. The judgment criteria, corresponding system status, and failure risks specifically include: Steady-state region: The value remains consistently negative and is stable at [value range]. ≤-0.1; bifurcation zone: It oscillates around 0, i.e., -0.1 < <0.1; Chaos Zone: It remains positive and the value gradually increases, that is... ≥0.

1.

8. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 7, characterized in that, Also includes: To avoid due to Random fluctuations can lead to misjudgments. To address this, a dynamic discrimination approach combined with historical trajectory backtracking is employed. The specific steps are as follows: Real-time acquisition of the output of the chaotic feature quantization module Value, recording the most recent h samples Trajectory; Based on the judgment criteria for the three regions, the current system status is preliminarily determined; If it is initially determined to be a bifurcation region / chaotic region, backtrack to the nearest... sampling points If the trend is consistent with the current preliminary judgment, then the state judgment result is confirmed. If the trends are inconsistent, then it is determined as Occasional fluctuations do not trigger warnings; only the fluctuation details are recorded. Based on the dynamic state, a three-level early warning mechanism is set up, with each level corresponding to different early warning signals, early warning methods, and processing priorities.

9. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 1, characterized in that, The entropy-driven task blocking and scheduling execution module specifically includes: Two core scheduling strategies are built-in, corresponding to the bifurcation region and the chaotic region, respectively: Strategy A, corresponding to the bifurcation region: Terminate all currently executing low-priority tasks and record their termination status; Pause all low-priority tasks pending execution and add them to the task queue; Allocate CPU, memory, and bus resources used by low-priority tasks to high-priority core business tasks. After the pause time ends, or when the system recovers from the bifurcation region to the steady state region, the execution of low-priority tasks will automatically resume, and unfinished tasks will be resumed first according to the order of the task queue.

10. The intelligent scheduling system for system maintenance tasks based on artificial intelligence according to claim 9, characterized in that, Also includes: Strategy B, corresponding to the chaotic region: Identify the core system processes, record their running status and data, then terminate them, restart them immediately, and restore their running status and data after restarting. By tracing the anomaly logs, the cause can be located. For abnormal processes that have increased in frequency, migrate them to a separate container and limit their CPU and memory usage; at the same time, restart the core processes that the abnormal processes depend on to restore the normal system logic. After process isolation, the system automatically monitors the status of abnormal processes. If the abnormal process recovers, it is removed from the container and normal resource allocation is restored. If the abnormal process continues to be abnormal, the process is automatically terminated and a fault notification is sent to the administrator.

Images