Intelligent machine room operation and maintenance management automatic inspection system

Abstract

The application discloses a machine room operation and maintenance management automatic inspection system based on intelligence, and belongs to the technical field of intelligence, comprising a multi-source data fusion processing module, a hidden fault probability distribution module, a fault root cause analysis module, a dynamic self-adaptive baseline generation module, a multi-element strategy autonomous healing decision module, a collaborative control execution module and a feedback optimization module. The application constructs a full-link intelligent system of multi-source data fusion, quantum neural network prediction, graph neural network tracing, dynamic baseline generation, reinforcement learning decision, transactional scheduling execution and closed-loop optimization. The hidden fault recognition capability is improved through quantum-classical hybrid calculation. The root cause is accurately positioned in combination with a causal graph. An optimal repair scheme is generated based on a dynamic baseline and a reinforcement learning strategy library. Finally, the reliability of cross-system operation is ensured through transactional scheduling, and the fault prediction accuracy is significantly improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligentization, and particularly relates to an intelligentized machine room operation and maintenance management automated inspection system. BACKGROUND

[0002] With the development of cloud computing, big data and artificial intelligence, the scale of modern machine room equipment continues to expand, and the system complexity significantly improves; traditional machine room inspection mainly relies on periodic manual inspection or rule-based monitoring alarm systems, which have problems such as response lag, weak fault identification capability, high false alarm and missed alarm rates, etc.

[0003] However, the existing machine room operation and maintenance management automated inspection still has certain defects. The existing technology mainly relies on classical neural networks, which are difficult to capture the implicit correlation between device operating states, sensitive to noise interference, and lack modeling capability for quantum entanglement characteristics, resulting in insufficient identification rate for hidden faults, easy to miss or misjudge, and mostly using five-question method or fishbone diagram for manual experience-driven causal analysis, lacking automated causal path tracing capability; the root cause positioning process relies on historical log retrieval and expert knowledge base, which is low in analysis efficiency and easy to be affected by subjective judgment, and it is difficult to quickly locate the key nodes in the complex fault chain. Therefore, an intelligentized machine room operation and maintenance management automated inspection system is proposed. SUMMARY

[0004] The purpose of the present application is to provide an intelligentized machine room operation and maintenance management automated inspection system to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides the following technical solution: an intelligentized machine room operation and maintenance management automated inspection system, comprising a multi-source data fusion processing module, a hidden fault probability distribution module, a fault root cause analysis module, a dynamic adaptive baseline generation module, a multi-element strategy autonomous healing decision module, a collaborative control execution module and a feedback optimization module.

[0006] The multi-source data fusion processing module realizes cleaning and fusion of multi-source heterogeneous data through asymmetric data alignment, synchronously completes time series feature extraction, and outputs real-time data stream containing time series features;

[0007] The hidden fault probability distribution module identifies hidden faults through quantum neural networks according to real-time data stream, and outputs fault probability distribution;

[0008] The fault root cause analysis module traces the causal path based on graph neural networks according to the fault probability distribution, locates and outputs a root cause analysis report;

[0009] The dynamic adaptive baseline generation module dynamically calculates and adjusts the performance index health threshold according to the root cause analysis report, and outputs updated dynamic baseline parameters;

[0010] The multi-element strategy autonomous healing decision module performs repair scheme optimization based on the reinforcement learning strategy library according to the root cause analysis report and the updated dynamic baseline parameter, and generates optimal autonomous healing decision instructions;

[0011] The collaborative control execution module performs transactional scheduling and execution of cross-system atomic operations according to the optimal autonomous healing decision instructions, and feeds back an execution result report;

[0012] The feedback optimization module performs optimization according to the execution result of the whole link and the system state data.

[0013] Preferably, the multi-source data fusion processing module is wirelessly connected to the hidden fault probability distribution module, the hidden fault probability distribution module is wirelessly connected to the fault root cause analysis module, the fault root cause analysis module is wirelessly connected to the multi-element strategy autonomous healing decision module and the dynamic adaptive baseline generation module, the dynamic adaptive baseline generation module is wirelessly connected to the multi-element strategy autonomous healing decision module and the hidden fault probability distribution module, the multi-element strategy autonomous healing decision module is wirelessly connected to the collaborative control execution module, the collaborative control execution module is wirelessly connected to the feedback optimization module, and the feedback optimization module is wirelessly connected to the hidden fault probability distribution module.

[0014] Preferably, the multi-source data fusion processing module accesses multi-source heterogeneous data streams, including sensor real-time readings, system log events and external API data, and parses the timestamps and key fields of each data source.

[0015] The time offset between different data sources is calculated by the dynamic time warping algorithm, and linear interpolation is applied to align the time axis to ensure that all data points are aligned at a unified time point. Then, data cleaning is performed: missing values are removed, abnormal values are corrected, and high-frequency noise is filtered. Then, data fusion is performed: the cleaned multi-source data is integrated into a unified view with device ID as the key, and the conflict values are processed by weighted average. Time series feature extraction is performed synchronously: sliding window features are calculated in real-time streams, including mean, standard deviation, volatility, periodicity coefficient and baseline deviation rate. Finally, the extracted features are attached to the original data to generate structured real-time data streams.

[0016] Preferably, the hidden fault probability distribution module maps time series features to quantum states: obtains real-time data streams, and maps time series features to quantum states through encoding. The quantum state is realized as:

[0017] ,

[0018] In the formula, represents the quantum state,​ This represents the tensor product operator. Represents the X-axis rotating quantum gate. Represents a Z-axis rotating quantum gate. Indicates the rotation angle along the X-axis. Indicates the rotation angle along the Z-axis. The initial quantum state is represented by m qubits, the zero state is represented by m qubits, and n represents the number of input timing features.

[0019] Preferably, the hidden fault probability distribution module, in the quantum-classical hybrid parameter update step, calculates the gradient of the loss function through the classical backend of the parameters, then calculates the gradient of the quantum parameters through the parameter offset rule, and performs quantum-classical hybrid parameter update through alternating optimization, as follows:

[0020] ,

[0021] In the formula, This represents the quantum parameter update amount. Indicates the learning rate. Represents the gradient of the loss function. This represents the weight coefficient of the parameter offset rule. This represents the parameter offset rule, used to calculate the gradient of quantum parameters. This represents the expected value of quantum measurement.

[0022] Preferably, the hidden fault probability distribution module includes the following steps for generating the hidden fault probability distribution: performing multiple measurements on the quantum state, statistically analyzing the frequency distribution of each measurement result, and mapping it to a fault probability distribution, which is implemented as follows:

[0023] ,

[0024] In the formula, Let B represent the failure probability distribution, and let B represent the normalization factor. Representing quantum circuits Output The probability amplitude, Indicates the calibration intensity coefficient. The reference quantum state is represented by the historical fault data, and k represents the basis vector of the quantum state.

[0025] Preferably, the fault root cause analysis module receives the fault probability distribution. Define a node set: each node represents a device component, and define a directed edge set E: edges This indicates that a failure in component u causes a failure in component v; edges are defined and weighted. The cause-and-effect diagram is as follows Failure probability distribution , , representing the probability of fault type j, initializes the feature vector for fault type nodes. Initialize feature vectors for non-fault type component nodes The node features are embedded into the graph to form an initial feature matrix. .

[0026] Preferably, the root cause analysis module aggregates neighbor node information through a graph neural network message passing mechanism, as follows:

[0027] ,

[0028] In the formula, Indicates that node v is at the th Feature representation after layer propagation This represents the activation function. Let v represent the set of neighbors of node v, and let d represent nodes that have a direct causal relationship with v. This represents the degree of node v. Indicates the first The learnable weight matrix of the layer, Indicates that node u is at the th The feature representation of the layer represents the current probability confidence of node u. During propagation, the layer... Dynamic adjustments are made to the node characteristics after propagation. Based on the propagated node features H, for each fault type j, starting from the fault node j, traverse backwards along the edges. Filter feature values The nodes are selected as candidate root cause nodes. For each candidate root cause node, a depth-first search is used to extract the shortest path from the root cause to the fault, forming a causal chain. The path probability is then selected. The highest path is taken as the primary causal path. The path represents the causal path from the root cause to the failure. A root cause analysis report is generated based on the root cause component, the primary causal path, and the path probability.

[0029] Preferably, the dynamic adaptive baseline generation module obtains the root cause analysis report, parses the root cause components, causal paths and fault type information in the report, extracts performance indicators directly related to the fault, and identifies the fluctuation characteristics of key performance indicators when the fault occurs.

[0030] Preferably, the dynamic self-adaptive baseline generation module extracts a corresponding time window from historical data according to a fault occurrence time point, screens performance index time series data related to the fault, identifies similar events in the historical data as the current fault mode, performs smoothing processing on the screened historical data, dynamically adjusts a threshold boundary according to a fault impact level in a root cause analysis report, compares a new baseline with a historical baseline, verifies whether the adjusted threshold can effectively distinguish between normal and abnormal states, and verifies the performance of the adjusted baseline in similar historical fault events through simulation testing; if the verification is passed, the adjusted baseline parameters are taken as the new baseline; and if the verification fails, the historical baseline is rolled back.

[0031] Preferably, the multi-element strategy autonomous recovery decision module determines whether the current index exceeds the baseline boundary according to the root cause analysis report and the updated dynamic baseline parameters, the root cause component name, the fault type, the impact level and the key performance indicators, records the deviation degree, constructs a system health state description in combination with the causal path and the impact level, screens candidate repair strategies matched with the fault type from a pre-constructed strategy library, filters the strategies according to the impact level: high-impact faults are preferentially matched with high-reliability strategies, and low-impact faults are matched with low-cost strategies, associates preset effect indicators with each candidate strategy, evaluates the comprehensive value of each candidate strategy based on a pre-trained reinforcement learning model, selects an optimal candidate strategy and a strategy with a comprehensive score, and maps the strategy to specific autonomous recovery decision instructions.

[0032] Preferably, the collaborative control execution module receives the autonomous recovery decision instructions, decomposes them into atomic operation tasks across systems, establishes a task dependency graph, executes the operations in a priority order through a transaction scheduling engine, configures a timeout threshold and a rollback strategy for each operation, monitors the system interface state and performance indicators in real time during the execution process, triggers a compensation mechanism immediately if an operation anomaly is detected, and finally generates a structured execution report by summarizing all operation results.

[0033] Compared with the prior art, the application has the following beneficial effects:

[0034] 1. To solve the problems of multi-source data island, hidden fault difficult to identify, low root cause positioning efficiency, single repair strategy and lagging execution response in traditional machine room operation and maintenance, the present application builds a full-link intelligent system of multi-source data fusion, quantum neural network prediction, graph neural network tracing, dynamic baseline generation, reinforcement learning decision, transactional scheduling execution and closed-loop optimization, realizes the automation, precision and self-adaptation of operation and maintenance management, breaks down the data barrier with asymmetric data alignment, improves the hidden fault identification ability through quantum-classical hybrid computing, realizes accurate root cause positioning combined with causal graph, generates the optimal repair scheme based on dynamic baseline and reinforcement learning strategy library, and finally ensures the reliability of cross-system operation through transactional scheduling, significantly improves the fault prediction accuracy, root cause positioning efficiency, repair decision success rate and system self-adaptation ability;

[0035] 2. The present application maps the time sequence characteristics into quantum states through quantum encoding, models the implicit correlation between device operating states using quantum entanglement characteristics, alternately optimizes quantum gate parameters through the classical quantum hybrid parameter update mechanism combined with gradient descent and parameter offset rules, significantly improves the sensitivity of the model to weak abnormal patterns, and generates fault probability distribution through multiple quantum state measurements to identify hidden faults that are difficult to capture;

[0036] 3. The present application builds a causal graph through a graph neural network, defines nodes and directed edges, assigns weights through historical conditional probabilities, forms a dynamic causal network, aggregates neighbor node information through the message passing mechanism of the graph neural network, extracts the main causal path combined with the depth-first search algorithm, accurately locates the fault root cause, and outputs a complete report containing fault summary, root cause component, causal path, impact assessment and action suggestion, significantly shortening the manual troubleshooting time;

[0037] 4. The present application extracts fault-related historical data through a time window, generates a prediction curve combined with smoothing processing and time series decomposition; calculates the dynamic alarm threshold based on the residual distribution, differentially adjusts the coefficient to adapt to business fluctuations, and verifies the baseline effectiveness through historical similar event simulation test in the verification stage to ensure that the new threshold can effectively distinguish between normal and abnormal states; the baseline parameters evolve adaptively with the system state to avoid failure of static threshold when the load changes or the device ages, thereby improving the abnormal detection sensitivity. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 It is a structural schematic diagram of the intelligent machine room operation and maintenance management automatic inspection system of the present application;

[0039] Figure 2 It is the running process of the intelligent machine room operation and maintenance management automatic inspection system of the present application Figure One ;

[0040] Figure 3This invention describes the operation flow of an intelligent automated inspection system for data center operation and maintenance management. Figure Two ;

[0041] Figure 4 This invention describes the operation flow of an intelligent automated inspection system for data center operation and maintenance management. Figure Three . Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example:

[0044] Please see Figures 1-4 As shown, the present invention provides a technical solution including a multi-source data fusion processing module, a hidden fault probability distribution module, a fault root cause analysis module, a dynamic adaptive baseline generation module, a multi-strategy self-healing decision-making module, a collaborative control execution module, and a feedback optimization module;

[0045] The multi-source data fusion processing module achieves the cleaning and fusion of multi-source heterogeneous data through asymmetric data alignment, simultaneously completes the extraction of time-series features, and outputs a real-time data stream in a unified format containing time-series features.

[0046] The hidden fault probability distribution module identifies hidden faults through a quantum neural network based on real-time data streams and outputs a fault probability distribution.

[0047] The fault root cause analysis module performs causal path tracing based on graph neural networks according to the fault probability distribution, locates and outputs a root cause analysis report;

[0048] The dynamic adaptive baseline generation module dynamically calculates and adjusts the health thresholds of relevant performance indicators based on the root cause analysis report, and outputs updated dynamic baseline parameters.

[0049] The multi-strategy self-healing decision module optimizes the repair scheme based on the root cause analysis report and the updated dynamic baseline parameters, and generates the optimal self-healing decision instruction.

[0050] The collaborative control execution module performs transactional scheduling and execution of cross-system atomic operations based on the optimal self-healing decision instructions, and reports the execution results.

[0051] The feedback optimization module iteratively optimizes the prediction model, analysis graph, and strategy library based on the execution results of the entire chain and the system status data.

[0052] In this embodiment, the multi-source data fusion processing module is wirelessly connected to the hidden fault probability distribution module, the hidden fault probability distribution module is wirelessly connected to the fault root cause analysis module, the fault root cause analysis module is wirelessly connected to the multi-strategy self-healing decision module and the dynamic adaptive baseline generation module, the dynamic adaptive baseline generation module is wirelessly connected to the multi-strategy self-healing decision module and the hidden fault probability distribution module, the multi-strategy self-healing decision module is wirelessly connected to the collaborative control execution module, the collaborative control execution module is wirelessly connected to the feedback optimization module, and the feedback optimization module is wirelessly connected to the hidden fault probability distribution module.

[0053] In this embodiment, the multi-source data fusion processing module accesses multi-source heterogeneous data streams, including real-time sensor readings, system log events, and external API data, and parses the timestamps and key fields of each data source.

[0054] The time offset between different data sources is calculated using a dynamic time warping algorithm, and linear interpolation is applied to align the time axis, ensuring that all data points are aligned at a unified time point. Next, data cleaning is performed: missing values ​​are removed, outliers are corrected, and high-frequency noise is filtered out. Then, data fusion is performed: using the device ID as the key, the cleaned multi-source data is integrated into a unified view, and conflicting values ​​are processed using a weighted average. Simultaneously, time-series feature extraction is performed: sliding window features, including mean, standard deviation, volatility, periodicity coefficient, and baseline deviation rate, are calculated in the real-time stream. Finally, the extracted features are appended to the original data to generate a structured real-time data stream.

[0055] In this embodiment, the hidden fault probability distribution module maps temporal features to quantum states by: acquiring real-time data streams and encoding the temporal features. Mapping to quantum state The quantum state is realized as follows:

[0056] ,

[0057] In the formula, Representing quantum states, indicating the health status of a device. This represents the tensor product operator, which combines multiple quantum gates sequentially. This represents the X-axis rotation quantum gate, used to control the amplitude of the quantum state. This represents a Z-axis rotating quantum gate used to control the phase of a quantum state. Indicates the rotation angle along the X-axis. express , represents the Z-axis rotation angle, represents , represents the initial quantum state, the zero state of m qubits, tensor product of, represents the i-th sensor index, and represents the real-time data of the device, represents the maximum value of the input time sequence feature, and n represents the number of input time sequence features;

[0058] In this embodiment, the quantum neural network architecture based on quantum gate operation takes the quantum encoded features as input in the input layer, realizes feature transformation through multi-layer quantum gate operation, and models the implicit association between device operating states by using quantum entanglement characteristics, wherein the key quantum gate parameters are realized by differentiable quantum circuits, so that the network can capture the hidden fault feature mode that is difficult to identify by traditional neural networks.

[0059] In this embodiment, the quantum-classical hybrid parameter updating step of the hidden fault probability distribution module is: calculating the gradient of the loss function through the parameter classical backend, then calculating the gradient of the quantum parameter through the parameter offset rule, and updating the quantum-classical hybrid parameter through alternating optimization, which is realized as:

[0060] ,

[0061] In the formula, represents the quantum parameter update amount, and represents the parameter the update step of the parameter, represents the learning rate, represents the loss function gradient, and represents the loss function gradient calculated by the classical backend, represents the parameter offset rule weight coefficient, which controls the influence degree of quantum noise on parameter updating, represents the parameter offset rule, which is used to calculate the gradient of the quantum parameter, represents , represents the perturbation step, represents the quantum measurement expectation value.

[0062] In this embodiment, the hidden fault probability distribution module, the hidden fault probability distribution generation step: measuring the quantum state multiple times, and counting the frequency distribution of each measurement result, mapping it as a fault probability distribution, which is realized as:

[0063] ,

[0064] In the formula, represents the fault probability distribution, the probability of the jth fault, represents the possibility of the occurrence of the jth fault of the device, and B represents the normalization factor, Quantum circuit Output Probability amplitude of Denotes the calibration intensity coefficient, the degree of influence of the control history similarity on the probability distribution, Denotes the reference quantum state generated by the historical failure data, k denotes the basis vector of the quantum state.

[0065] In this embodiment, the failure root cause analysis module receives the failure probability distribution , defines a node set: each node represents a device component or a failure type j, defines a directed edge set E: an edge represents that the failure or state change of component u may cause the failure of component v, defines the edge assignment weight : the causal strength is calculated based on the historical failure data, such as the conditional probability of v failure after u failure, and the causal relationship graph is , the node feature is initially empty, and the failure probability distribution , , represents the probability of failure type j, and the feature vector of the failure type node is initialized , the feature vector of the non-failure type component node is initialized , the node feature is embedded in the graph to form an initial feature matrix .

[0066] In this embodiment, the failure root cause analysis module aggregates neighbor node information through a message passing mechanism of a graph neural network, and is realized as:

[0067] ,

[0068] In the formula, denotes the feature representation of node v after propagation in the m-th layer, denotes the probability confidence of node v as a potential root cause, denotes an activation function, denotes a neighbor set of node v, and denotes a node directly causally related to v, denotes the degree of node v, denotes a learnable weight matrix of the m-th layer, used to learn the contribution weight of different neighbors, denotes the feature representation of node u in the m-th layer, and denotes the current probability confidence of node u, which is dynamically adjusted during the propagation process, and the node feature after propagation, denotes the probability confidence of node v as a root cause, according to the node feature H after propagation, starting from the failure node j, along the edge, the feature value is filtered , for each failure type j. ​​​ a node of represents a threshold value, as a candidate root cause node, for which the shortest path from the root cause to the failure is extracted by depth-first search to form a causal chain, and the path probability is the highest path as the main causal path, path represents the causal path from the root cause to the failure, such as root cause component → intermediate component → failure component, and the root cause analysis report is generated according to the root cause component, the main causal path and the path probability, and the root cause analysis report content includes failure summary, root cause positioning, causal path, impact assessment and action suggestion.

[0069] In this embodiment, the dynamic adaptive baseline generation module obtains the root cause analysis report, analyzes the root cause component, the causal path and the failure type information in the report, extracts the performance indicators directly related to the failure, such as temperature, voltage, response time, etc., and identifies the key performance indicator fluctuation characteristics at the time of failure.

[0070] In this embodiment, the dynamic adaptive baseline generation module extracts the corresponding time window from the historical data according to the failure time point, filters the performance indicator time series data related to the failure, such as temperature, voltage, etc., identifies events similar to the current failure mode in the historical data, and performs smoothing processing on the filtered historical data.

[0071] Specifically, the horizontal component, the trend component and the seasonal component are calculated by the time series analysis algorithm to generate a prediction curve, the dynamic alarm threshold range is determined by calculating the residual distribution of the predicted value and the actual value, and the differentiated threshold adjustment coefficient is set for failures of different impact levels.

[0072] It should be understood that according to the failure impact level in the root cause analysis report, the threshold boundary is dynamically adjusted, the new baseline is compared with the historical baseline, and it is verified whether the adjusted threshold can effectively distinguish between normal and abnormal states, the performance of the adjusted baseline in the historical similar failure events is verified by simulation test; if the verification is passed, the adjusted baseline parameter is taken as the new baseline; if the verification fails, the historical baseline is rolled back.

[0073] In this embodiment, the multi-element strategy autonomous healing decision module, root cause analysis report and updated dynamic baseline parameters, root cause component name, fault type, impact level, key performance indicators, determine whether the current indicators exceed the baseline boundary, and record the deviation degree, combined with the causal path and impact level, construct the system health status description, filter the candidate repair strategies matching the fault type from the pre-constructed strategy library, filter the strategies according to the impact level: high-impact faults preferentially match high-reliability strategies, and low-impact faults match low-cost strategies, associate each candidate strategy with a preset effect indicator, based on the pre-trained reinforcement learning model, evaluate the comprehensive value of each candidate strategy, select the optimal candidate strategy identifier and comprehensive score strategy, and map the strategy to specific autonomous healing decision instructions.

[0074] In this embodiment, the collaborative control execution module, after receiving the autonomous healing decision instructions, decomposes them into cross-system atomic operation tasks and establishes a task dependency graph, executes the operations in priority order through a transactional scheduling engine, each operation is configured with a timeout threshold and a rollback strategy, and the system interface state and performance indicators are monitored in real time during the execution process. If an operation anomaly is detected, the compensation mechanism is triggered immediately, and finally all operation results are summarized to generate a structured execution report, including task link tracking ID, operation log snapshot and baseline comparison data.

[0075] Working principle: Through the asymmetric data alignment technology, the sensor real-time data, system log events and external API data are integrated, the multi-source data timestamp offset and format difference problems are solved. First, the timestamps and key fields of each data source are parsed, the time offset is calculated using the dynamic time warping algorithm and the time axis is linearly interpolated to ensure that all data points have a unified time reference; then data cleaning is performed, and the cleaned data is fused into a unified view using device ID as the key, and conflicting values are processed by weighted average; synchronous extraction of sliding window time series features is performed, which are attached to the original data to generate structured real-time data streams;

[0076] By encoding the time sequence features into quantum states, the quantum entanglement characteristics are used to model the implicit correlation between the device operating states; the classical quantum hybrid parameter updating mechanism combines gradient descent and parameter offset rules to alternately optimize quantum gate parameters, improving the model's sensitivity to weak abnormal patterns, generating a fault probability distribution through multiple quantum state measurements, and combining historical similarity to calibrate the intensity coefficient, outputting the probability values of various faults occurring in the device; a causal graph is constructed through a graph neural network, defining device components or fault types as nodes, calculating the causal strength based on historical fault data to assign weights, and forming a directed edge causal relationship graph; the message passing mechanism aggregates neighbor node information and traverses fault nodes in reverse through a depth-first search to filter high-confidence candidate root cause nodes and extract primary causal paths; finally, an analysis report containing fault summary, root cause localization, causal path, and repair suggestions is generated; the root cause components, causal paths, and fault types in the report are analyzed to extract relevant performance indicators and their fluctuation characteristics; similar events to the current fault pattern are filtered from historical data, and a prediction curve is generated through smoothing and time series decomposition, with a residual distribution determining the dynamic alarm threshold range; different threshold adjustment coefficients are set for different impact levels to verify the effectiveness of the new baseline in historical similar events, ensuring that it can accurately distinguish between normal and abnormal states; combined with the root cause analysis results and dynamic baseline parameters, matching repair strategies are selected from a pre-defined strategy library; high-reliability or low-cost strategies are matched according to the fault impact level, and the comprehensive value of the candidate strategies is evaluated based on a pre-trained reinforcement learning model; the optimal strategy is selected as specific self-healing decision instructions; after receiving the self-healing instructions, they are decomposed into atomic operation tasks across systems, executed by a transactional scheduling engine according to dependency relationships and priorities; each operation is configured with a timeout threshold and rollback strategy, and the interface state and performance indicators are monitored in real time, triggering a compensation mechanism in case of abnormality; the operation results are summarized to generate a structured execution report containing link trace ID, log snapshot, and baseline comparison data, and finally the full-link execution results and system state data are integrated to iteratively optimize the prediction model, causal graph, and strategy library; through a closed-loop feedback mechanism, inefficient strategies are marked and efficient strategies are enhanced, continuously improving the system's adaptive ability.

[0077] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

[0078] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired thereby, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution are not creative, and should belong to the protection scope of the present application.

Claims

1. An intelligent automated inspection system for data center operation and maintenance management, characterized by: The system comprises a multi-source data fusion processing module, a hidden fault probability distribution module, a fault root cause analysis module, a dynamic adaptive baseline generation module, a multi-element strategy autonomous healing decision module, a collaborative control execution module and a feedback optimization module. The multi-source data fusion processing module realizes cleaning and fusion of multi-source heterogeneous data through asymmetric data alignment, synchronously completes time sequence feature extraction, and outputs real-time data stream containing time sequence features. The hidden fault probability distribution module identifies hidden faults through a quantum neural network according to the real-time data stream, and outputs fault probability distribution. The fault root cause analysis module traces a cause path based on a graph neural network according to the fault probability distribution, locates and outputs a root cause analysis report. The dynamic adaptive baseline generation module dynamically calculates and adjusts a performance index health threshold according to the root cause analysis report, and outputs updated dynamic baseline parameters. The multi-element strategy autonomous healing decision module performs repair scheme optimization based on a reinforcement learning strategy library according to the root cause analysis report and the updated dynamic baseline parameters, and generates optimal autonomous healing decision instructions. The collaborative control execution module performs transactional scheduling and execution of cross-system atomic operations according to the optimal autonomous healing decision instructions, and feeds back an execution result report. The feedback optimization module optimizes according to the execution result and system state data of the whole link. The hidden fault probability distribution module comprises a hidden fault probability distribution generation step of performing multiple measurements on a quantum state, counting the frequency distribution of each measurement result, mapping the frequency distribution into a fault probability distribution, and realizing the following: ， In the formula, represents a failure probability distribution, B represents a normalization factor, represents a quantum circuit output probability amplitude, represents a calibration intensity coefficient, represents a reference quantum state generated by historical failure data, k represents a base vector of a quantum state; The fault root cause analysis module receives a fault probability distribution , define a set of nodes: each node represents a device component, define a set of directed edges E: edges represent that the failure of component u leads to the failure of component v, define edge weights , the causal graph is , the fault probability distribution , , represent the probability of fault type j, initialize the feature vector of the fault type node , initialize the feature vector of the non-fault type component node , embed the node features into the graph to form an initial feature matrix ; The fault root cause analysis module aggregates neighbor node information through a message passing mechanism of a graph neural network, and realizes the following: ， In the formula, represents the feature representation of node v after the propagation of the layer, represents the activation function, represents the neighbor set of node v, which represents the nodes directly causally related to v, represents the degree of node v, represents the learned weight matrix of the layer, represents the feature representation of node u in the layer, which represents the current probability confidence of node u, which is dynamically adjusted during the propagation process, and the node feature after the propagation, According to the node feature H after the propagation, for each fault type j, starting from the fault node j, traversing the edges in reverse, Screening nodes with feature values as candidate root cause nodes, for candidate root cause nodes, extract the shortest path from the root cause to the fault through depth-first search, form a causal chain, select the path with the highest path probability as the main causal path, path represents the causal path from the root cause to the fault, and the root cause analysis report is generated according to the root cause component, the main causal path and the path probability. 2.The intelligent-based machine room operation and management automated inspection system of claim 1, wherein: The hidden failure probability distribution module, the time sequence characteristic mapping is a quantum state step: obtaining real-time data stream, mapping the time sequence characteristic into quantum state through coding Mapping into quantum state The quantum state is realized as: ， In the formula, represents a quantum state, represents a tensor product operator, represents an X-axis rotation quantum gate, represents a Z-axis rotation quantum gate, represents an X-axis rotation angle, represents a Z-axis rotation angle, represents an initial quantum state, a zero state of m quantum bits, and n represents the number of input temporal features. 3.The intelligent-based machine room operation and management automated inspection system of claim 2, characterized in that: The hidden fault probability distribution module comprises a quantum-classical hybrid parameter updating step of calculating the gradient of a loss function through a parameter classical backend, calculating the gradient of a quantum parameter through a parameter offset rule, and updating the quantum-classical hybrid parameter through alternating optimization, and realizes the following: ， In the formula, represents the quantum parameter update amount, represents the learning rate, represents the loss function gradient, represents the parameter shift rule weight coefficient, represents the parameter shift rule, used for calculating the gradient of the quantum parameter, represents the perturbation step length, represents the quantum measurement expectation value.

4. The intelligent-based machine room operation and management automated inspection system of claim 1, wherein: The dynamic adaptive baseline generation module acquires a root cause analysis report, analyzes root cause component, cause path and fault type information in the report, extracts performance indexes directly related to faults, and identifies key performance index fluctuation characteristics when faults occur.

5. The intelligent-based machine room operation and management automated inspection system of claim 4, characterized in that: The dynamic adaptive baseline generation module extracts a corresponding time window from historical data according to a fault occurrence time point, filters performance index time series data related to faults, identifies events similar to a current fault mode in historical data, performs smoothing processing on the filtered historical data, dynamically adjusts a threshold boundary according to a fault impact level in the root cause analysis report, compares a new baseline with a historical baseline, verifies whether the adjusted threshold can effectively distinguish normal and abnormal states, and verifies the performance of the adjusted baseline in historical similar fault events through simulation testing. If the verification is passed, the adjusted baseline parameters are taken as a new baseline. If the verification fails, the historical baseline is rolled back. 6.The intelligent-based machine room operation and management automated inspection system of claim 1, wherein: The multi-element strategy autonomous healing decision module, root cause analysis report and updated dynamic baseline parameters, root cause component name, fault type, impact level, key performance indicators, determines whether the current indicators exceed the baseline boundary, and records the deviation degree, constructs the system health state description combined with the causal path and the impact level, filters the candidate repair strategies from the pre-constructed strategy library matched with the fault type: high-impact faults match high-reliability strategies first, and low-impact faults match low-cost strategies, associates preset effect indicators for each candidate strategy, evaluates the comprehensive value of each candidate strategy based on the pre-trained reinforcement learning model, selects the optimal candidate strategy identifier and the comprehensive score of the strategy, and maps the strategy to specific autonomous healing decision instructions. 7.The intelligent-based machine room operation and management automated inspection system of claim 1, wherein: The collaborative control execution module receives autonomous healing decision instructions, decomposes them into cross-system atomic operation tasks, and establishes a task dependency graph. The operation is executed in priority order through a transactional scheduling engine, each operation is configured with a timeout threshold and a rollback strategy, and the system interface state and performance indicators are monitored in real time during the execution process. If an operation anomaly is detected, a compensation mechanism is triggered immediately, and finally all operation results are summarized to generate a structured execution report.

Images