A method for self-recovery processing of hardware device failure

Abstract

The application relates to the technical field of hardware device fault self-recovery, and discloses a hardware device fault self-recovery processing method, which comprises the following steps: if the type of a fault is a first-level fault, a feature vector of the fault is extracted, a fault case library C is set, a classification model prediction strategy is executed according to the feature vector X and the fault case library C, the probability that a fault solution of each fault case successfully solves the current fault is predicted, a sorting model optimization strategy is executed according to the feature vector X and a candidate category set, the priority of each fault solution in the candidate category set is optimized, a target solution is applied to repair the current fault, and a feedback result is recorded as When a hardware device with a first-level fault is repaired, a maintenance plan is dynamically generated according to multidimensional factors, the self-adaptive capability is used to accurately process hardware device faults, the fault processing time is reduced, and therefore the overall operation and maintenance efficiency is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of hardware device fault self-healing technology, specifically to a hardware device fault self-healing method. Background Technology

[0002] Current Status of Related Technologies in this Field: With the acceleration of information technology construction, large data centers are increasingly demanding higher stability and responsiveness from their IT infrastructure. While existing operation and maintenance mechanisms can monitor equipment status in real time, they often lead to service interruptions due to delays in manual intervention during sudden emergency failures, impacting user experience and business continuity. Corresponding Existing Techniques: Traditional IT operation and maintenance strategies mainly include setting threshold alarms and assigning personnel to monitor equipment. Once an anomaly is detected, an alert is immediately issued, awaiting on-site inspection and intervention by technical personnel. Limitations of Existing Technologies: Current fault handling heavily relies on manual decision-making and intervention. In emergency situations at night or on holidays, response times are long, failing to promptly prevent potential system crashes, thus affecting the overall system performance and reliability.

[0003] Therefore, this solution proposes a self-healing method for hardness testing equipment faults, which solves the problems mentioned in the background technology. Summary of the Invention

[0004] This invention provides a hardware device fault self-healing method, which helps to solve the problems mentioned in the background art.

[0005] This invention provides the following technical solution: a hardware device fault self-healing method, comprising:

[0006] Hardware device failures are categorized into Level 1 and Level 2 failures.

[0007] Classifying faults into primary and secondary faults helps to accurately pinpoint the severity of problems and ensure that critical issues are addressed first.

[0008] When a hardware device malfunctions, determine whether the malfunction is classified as a Level 1 malfunction.

[0009] If the fault type is a level one fault, then extract the fault feature vector X = {x1, x2, ..., x...} n}, where x i This indicates the frequency and performance metrics of keywords in the device logs;

[0010] Define a fault case library C, which contains all historical fault cases, and represent each fault case as C. i =(A i P i R i Ti ), where A i P is the fault feature vector. i For troubleshooting solutions, R i To solve the problem, T i This serves as a timestamp for failure cases; by introducing a failure case library and matching feature vectors with historical data, we can ensure that the solution is supported by data and improve accuracy.

[0011] Based on the feature vector X and the fault case library C, a classification model prediction strategy is executed to predict the probability that the fault solution for each fault case will successfully resolve the current fault.

[0012] Obtain the probability of all failure solutions, calculate the mean of the probabilities, and record the result as the average probability.

[0013] Collect all types of fault solutions with a probability greater than the average probability, and denote them as the candidate type set;

[0014] Based on the feature vector X and the candidate category set, a ranking model optimization strategy is executed to optimize the priority of each fault solution in the candidate category set;

[0015] The probability screening and ranking mechanism of the candidate solution set avoids blind attempts and reduces the risk of processing failure.

[0016] Identify the highest priority solution to the fault, denoted as the target solution P. u ;

[0017] Application Target Solution P u Repair the current fault and record the feedback result as R. u ;

[0018] Based on the feedback results, the model for recommending fault solutions is dynamically optimized;

[0019] Determine whether the faulty equipment has been successfully repaired. If it has been successfully repaired, the hardware equipment that has been repaired for a Level 1 fault is considered to have experienced a Level 2 fault.

[0020] After repairing hardware devices with a Level 1 fault, a maintenance plan is dynamically generated based on multi-dimensional factors.

[0021] If the repair fails, each element of the target solution removed from the candidate set will be used sequentially to repair the hardware device.

[0022] Preferably, the step of executing a classification model prediction strategy based on the feature vector X and the fault case library C to predict the probability that the fault solution for each fault case will successfully resolve the current fault includes:

[0023] Obtain all distinct fault solutions from the fault case library and form a solution category set;

[0024] For any fault solution in the set of solution categories, assign a score Z to the fault solution.

[0025] Obtain the feature vector X = {x1, x2, ..., x3} n};

[0026] Calculate Z = W·X + b, where W is the weight matrix, X is the model parameters, and b is the bias term;

[0027] The following formula converts the scores of fault solutions into probabilities:

[0028] Obtain the number of elements in the solution category set, denoted as the total number of categories K;

[0029] Number the fault solutions in the solution category set from 1 to K;

[0030] calculate Where j = 1, 2, ..., K, P j Let j be the probability of the faulty solution.

[0031] Calculate the cross-entropy loss:

[0032] Where n is the number of samples, K is the number of classes, and y ij For the one-hot encoding of the real label, P ij For probability;

[0033] The result of calculating the cross-entropy loss is recorded as the loss value;

[0034] A loss threshold is set, which is used to determine the accuracy of the prediction result;

[0035] When the loss value is less than or equal to the loss threshold, the classification model prediction strategy is terminated.

[0036] When the loss value exceeds the loss threshold, gradient descent is used to update parameters W and b, specifically:

[0037] calculate As the new W;

[0038] calculate As the new b, where η is the learning rate;

[0039] The classification model prediction strategy is executed again using the new W and b as parameters.

[0040] Preferably, the step of executing a ranking model optimization strategy based on the feature vector X and the candidate category set to optimize the priority of each fault solution in the candidate category set includes:

[0041] The number of all elements in the candidate category set is obtained, denoted as F;

[0042] Sort each fault solution in the candidate category set from 1 to F;

[0043] For any fault solution in the candidate set, set a score sco for the fault solution;

[0044] Obtain the feature vector X = {x1, x2, ..., x3} n};

[0045] in, Used to extract feature vector X and candidate schemes s j The features are W, the parameters of the model, and b, the bias term.

[0046] Obtain the score for each fault solution in the candidate category set, and sort the fault solutions according to the scores;

[0047] Let s be any two fault solutions obtained from the candidate set. p and s q , where s p The score is higher than s q ;

[0048] Calculate ∑ (i，j) max(0, 1-(sco(X, s)) p )-sco(X, s q Record the result as the sort value;

[0049] Compare the sorted value with 1: if the sorted value > 1, stop executing the sorting model optimization strategy;

[0050] If the sort value is ≤1, then the gradient descent method is used to update the parameters W and b, specifically:

[0051] calculate As the new W;

[0052] calculate As the new b, where η is the learning rate;

[0053] The sorting model optimization strategy is executed again using the new W and b as parameters.

[0054] By optimizing the priority of candidate solutions using a ranking model, the most likely solution to solve the problem can be quickly identified, improving efficiency. Dynamically adjusting parameters W and b makes the model more adaptable to new problems, ensuring the accuracy of fault handling decisions.

[0055] Preferably, the step of dynamically optimizing the model for recommending fault solutions based on feedback results includes:

[0056] The feedback results in, -1 indicates the fault is resolved, -1 indicates the fault is not resolved, T max Maximum allowed processing time;

[0057] Define strategy π(P) u |X u ;θ), indicating that the fault characteristic is X u Time-based fault solution P u The probability distribution of u, where 0 ≤ u ≤ n;

[0058] Set a discount factor γ, 0≤γ≤1, to balance immediate rewards and long-term rewards;

[0059] Calculate cumulative rewards

[0060] Set a reward threshold;

[0061] When the cumulative reward is less than or equal to the reward threshold, the policy gradient method is used to optimize θ.

[0062]

[0063] The model is optimized through feedback learning, avoiding the repeated execution of inefficient or erroneous solutions. Rapid fault repair and self-healing reduce equipment downtime and maximize equipment availability. Actual fault resolution results (success or failure) are fed back to the model, continuously evolving classification and ranking strategies. A cumulative reward mechanism is introduced to balance short-term and long-term effects, enhancing the intelligence of decision-making.

[0064] Preferably, after repairing a hardware device that has experienced a Level 1 fault, a maintenance plan is dynamically generated based on multi-dimensional factors, including:

[0065] Establish equipment state model: M t =M0-α.V t -β.Y t +γ.L t ;

[0066] Among them, M t This represents the health status of the hardware device at time t, where 0 indicates a fault and 1 indicates normal operation.

[0067] M0 represents the initial healthy state;

[0068] V t The workload intensity of the hardware device at time t;

[0069] Y t The hardware device accumulates its running time over time t;

[0070] L t The strength of maintenance measures for hardware equipment at time t;

[0071] α, β, and γ are the weighting factors for the impact of workload, service life, and maintenance measures on health status, respectively.

[0072] Preferably, after repairing a hardware device that has experienced a Level 1 fault, a maintenance plan is dynamically generated based on multi-dimensional factors.

[0073] Extend equipment lifespan and reduce replacement costs due to malfunctions by dynamically generating maintenance plans. Maintenance plans are dynamically generated based on equipment workload, service life, and health status to precisely meet equipment maintenance needs. Health status models and priority scoring are used to identify maintenance strategies for high-priority equipment, reducing potential risks. After repairing a primary fault, preventative maintenance measures are developed through multi-dimensional factor analysis to reduce the probability of secondary faults. Critical equipment is prioritized for maintenance to reduce the risk of large-scale equipment damage caused by cascading failures.

[0074] The present invention has the following beneficial effects:

[0075] 1. This hardware device fault self-healing method, by adopting an intelligent emergency fault diagnosis and handling process, automates the entire process from fault monitoring and judgment to handling, greatly reducing the time cost required for human intervention, thereby significantly improving the overall stability and service quality of the system.

[0076] 2. This hardware device fault self-healing method, by incorporating intelligent decision support based on historical data, can accurately select the most suitable emergency measures for the current situation, greatly reducing trial and error costs and improving the success rate of the first repair.

[0077] 3. This hardware equipment fault self-healing method, combined with a regular preventive maintenance mechanism, forms a forward-looking operation and maintenance management model, effectively avoiding potential crises that may occur during long-term operation, ensuring that the equipment is in good condition for a long time, and indirectly saving a lot of maintenance expenses.

[0078] 4. This hardware device fault self-healing method, through classification model prediction and ranking optimization, divides the fault handling process into precise steps, ensuring that primary faults are addressed first and avoiding resource waste due to inappropriate prioritization. The classification model quickly matches historical cases based on feature vectors to select possible solutions; the ranking model further optimizes the priority of solutions, allowing the highest priority solutions to be applied rapidly. Gradient descent dynamically adjusts model parameters to ensure real-time adaptation to actual operating conditions. These methods effectively reduce fault handling time, especially in complex systems, and can significantly improve overall operational efficiency.

[0079] 5. This hardware device fault self-healing method, through the establishment of a fault case library and its matching mechanism with feature vectors, bases fault handling on historical data and scientific models. By calculating the success probability of solutions, the system selects only candidate solutions with a probability greater than the average, reducing interference from inefficient or even erroneous solutions. Ranking optimization further ensures the priority use of the best solution through feature extraction and scoring of candidate solutions. The introduction of a feedback mechanism guarantees continuous model optimization during use; its adaptive capability enables it to accurately handle new problems, thereby greatly improving the accuracy and reliability of the solution.

[0080] 6. This hardware fault self-healing method addresses the traditional reliance on the experience and skills of specialized personnel for fault handling, which typically requires significant time and effort in troubleshooting and repair. This solution, through automated classification, sorting, and prediction, significantly reduces the workload of the operations and maintenance team, especially providing intelligent assistance when dealing with complex problems. Furthermore, the optimized recommended solutions lower the technical barrier, allowing even less experienced maintenance personnel to quickly find efficient solutions, reducing reliance on specialized knowledge and improving the overall efficiency of the team. Attached Figure Description

[0081] Figure 1 This is a schematic diagram of the process of the present invention.

[0082] Figure 2 This is a schematic diagram of the method of the present invention.

[0083] Figure 3 This is a schematic diagram of the prediction strategy process of the classification model of the present invention.

[0084] Figure 4 This is a schematic diagram of the optimization strategy process for the sorting model of the present invention. Detailed Implementation

[0085] 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.

[0086] Example 1, refer to Figure 2 A method for self-healing hardware device faults, comprising:

[0087] Hardware device failures are categorized into Level 1 and Level 2 failures.

[0088] When a hardware device malfunctions, determine whether the malfunction is classified as a Level 1 malfunction.

[0089] If the fault type is a level one fault, then extract the fault feature vector X = {x1, x2, ..., x...} n}, where x i This indicates the frequency and performance metrics of keywords in the device logs;

[0090] Define a fault case library C, which contains all historical fault cases, and represent each fault case as C. i =(A i P i R i T i ), where A i P is the fault feature vector. i For troubleshooting solutions, R i To solve the problem, T i Timestamps for failure cases;

[0091] By introducing a fault case library and feature vector analysis, the system achieves structured storage and processing of fault information. Each fault has a detailed classification, solution, result feedback, and timestamp, forming a traceable historical record. This transparent management approach makes the fault handling process clearer and facilitates review and analysis by relevant personnel. Furthermore, standardized processing procedures such as classification prediction and sorting optimization avoid inconsistencies in processing quality caused by differences in human experience, laying the foundation for standardized management of equipment operation and maintenance.

[0092] Based on the feature vector X and the fault case library C, a classification model prediction strategy is executed to predict the probability that the fault solution for each fault case will successfully resolve the current fault.

[0093] By using classification models to predict the success rate of failure solutions, we can quickly screen potentially effective solutions and reduce trial and error time.

[0094] Obtain the probability of all failure solutions, calculate the mean of the probabilities, and record the result as the average probability.

[0095] Collect all types of fault solutions with a probability greater than the average probability, and denote them as the candidate type set;

[0096] Based on the feature vector X and the candidate category set, a ranking model optimization strategy is executed to optimize the priority of each fault solution in the candidate category set;

[0097] Identify the highest priority solution to the fault, denoted as the target solution P. u ;

[0098] Application Target Solution R u Repair the current fault and record the feedback result as R. u ;

[0099] Based on the feedback results, the model for recommending fault solutions is dynamically optimized;

[0100] Determine whether the faulty equipment has been successfully repaired. If it has been successfully repaired, the hardware equipment that has been repaired for a Level 1 fault is considered to have experienced a Level 2 fault.

[0101] After repairing hardware devices with a Level 1 fault, a maintenance plan is dynamically generated based on multi-dimensional factors.

[0102] If the repair fails, each element of the target solution removed from the candidate set will be used sequentially to repair the hardware device.

[0103] In this embodiment, refer to Figure 1 The key technologies of the present invention are shown in detail.

[0104] The implementation of this technological solution aligns deeply with the Internet of Things (IoT) and big data analytics in modern industry, providing a technological foundation for enterprise digital transformation. By collecting and analyzing equipment data in real time, the system achieves intelligent management across the entire process, from fault handling to preventative maintenance. Combining historical data analysis with machine learning, the system can gradually achieve higher levels of adaptability and automation, providing strong support for enterprises to introduce AI-driven management models, thereby propelling enterprises into a new stage of intelligent upgrading.

[0105] The step of executing a classification model prediction strategy based on feature vector X and fault case library C to predict the probability that the fault solution for each fault case will successfully resolve the current fault includes:

[0106] Obtain all distinct fault solutions from the fault case library and form a solution category set;

[0107] For any fault solution in the set of solution categories, assign a score Z to the fault solution.

[0108] Obtain the feature vector X = {x1, x2, ..., x3} n};

[0109] Calculate Z = W·X + b, where W is the weight matrix, X is the model parameters, and b is the bias term;

[0110] The following formula converts the scores of fault solutions into probabilities:

[0111] Obtain the number of elements in the solution category set, denoted as the total number of categories K;

[0112] Number the fault solutions in the solution category set from 1 to K;

[0113] calculate Where j = 1, 2, ..., K, P j Let j be the probability of the faulty solution.

[0114] Calculate the cross-entropy loss:

[0115] Where n is the number of samples, K is the number of classes, and y ij For the one-hot encoding of the real label, P ij For probability;

[0116] The result of calculating the cross-entropy loss is recorded as the loss value;

[0117] A loss threshold is set, which is used to determine the accuracy of the prediction result;

[0118] When the loss value is less than or equal to the loss threshold, the classification model prediction strategy is terminated.

[0119] When the loss value exceeds the loss threshold, gradient descent is used to update parameters W and b, specifically:

[0120] calculate As the new W;

[0121] calculate As the new b, where η is the learning rate;

[0122] The classification model prediction strategy is executed again using the new W and b as parameters.

[0123] In this embodiment, refer to Figure 3 This is the prediction strategy process for the classification model.

[0124] The step of executing a ranking model optimization strategy based on the feature vector X and the candidate category set to optimize the priority of each fault solution in the candidate category set includes:

[0125] The number of all elements in the candidate category set is obtained, denoted as F;

[0126] Sort each fault solution in the candidate category set from 1 to F;

[0127] For any fault solution in the candidate set, set a score sco for the fault solution;

[0128] Obtain the feature vector X = {x1, x2, ..., x} n};

[0129] in, Used to extract feature vector X and candidate schemes s j The features are W, the parameters of the model, and b, the bias term.

[0130] Obtain the score for each fault solution in the candidate category set, and sort the fault solutions according to the scores;

[0131] Let s be any two fault solutions obtained from the candidate set. p and s q , where s p The score is higher than s q ;

[0132] Calculate ∑(i,j)max(0,1-(sco(X,s)) p )-sco(X, s q Record the result as the sort value;

[0133] Compare the sorted value with 1: if the sorted value > 1, stop executing the sorting model optimization strategy;

[0134] If the sort value is ≤1, then the gradient descent method is used to update the parameters W and b, specifically:

[0135] calculate As the new W;

[0136] calculate As the new b, where η is the learning rate;

[0137] The sorting model optimization strategy is executed again using the new W and b as parameters.

[0138] In this embodiment, refer to Figure 4 This optimizes the strategy process for the ranking model.

[0139] Through an intelligent classification and sorting model, this solution avoids the trial-and-error costs of traditional fault handling methods. Accurate prediction and optimization reduce the time and resource consumption of repeatedly trying incorrect solutions, lowering operation and maintenance costs. Furthermore, rapid repair shortens equipment downtime, improves equipment utilization, and thus reduces indirect economic losses caused by production interruptions. The system also reduces the costs of equipment replacement or major repairs due to faults through preventative maintenance, fundamentally achieving cost optimization and efficient resource utilization.

[0140] The model for dynamically optimizing and recommending fault solutions based on feedback results includes:

[0141] The feedback results

[0142] in, -1 indicates the problem is resolved, -1 indicates the problem is not resolved.

[0143] T max Maximum allowed processing time;

[0144] Define strategy π(P) u |X u ;θ), indicating that the fault characteristic is X u Time-based fault solution P u The probability distribution of u, where 0 ≤ u ≤ n;

[0145] Set a discount factor γ, 0≤γ≤1, to balance immediate rewards and long-term rewards;

[0146] Calculate cumulative rewards

[0147] Set a reward threshold;

[0148] When the cumulative reward is less than or equal to the reward threshold, the policy gradient method is used to optimize θ.

[0149]

[0150] The system continuously optimizes its classification and ranking models through a feedback learning mechanism, achieving real-time adaptive adjustment. Every repair result during the fault resolution process is incorporated into the system, guiding subsequent fault handling strategy updates. A cumulative reward mechanism further balances short-term and long-term benefits, enabling the system not only to handle current faults but also to support long-term optimization. The model dynamically adjusts parameters W and b to adapt to changes in the equipment's operating environment, ensuring effective solutions even when new problems arise, significantly improving the system's intelligence level.

[0151] When a hardware device that has experienced a Level 1 fault is repaired, a maintenance plan is dynamically generated based on multiple factors, including:

[0152] Establish equipment state model: M t =M0-α.V t -β.Y t +γ.L t ;

[0153] Among them, M t This represents the health status of the hardware device at time t, where 0 indicates a fault and 1 indicates normal operation.

[0154] Initial healthy state;

[0155] V t The workload intensity of the hardware device at time t;

[0156] Y t The hardware device accumulates its running time over time t;

[0157] L t The strength of maintenance measures for hardware equipment at time t;

[0158] α, β, and γ are the weighting factors for the impact of workload, service life, and maintenance measures on health status, respectively.

[0159] After repairing a hardware device that has experienced a Level 1 fault, a maintenance plan is dynamically generated based on multi-dimensional factors.

[0160] After a fault is repaired, the system automatically generates a personalized maintenance plan, accurately predicting and formulating maintenance strategies based on multiple dimensions such as equipment workload, service life, and health status. Prioritization scoring identifies high-risk equipment and ensures resources are focused on critical issues. Preventative maintenance measures effectively reduce equipment failure rates, lower the risk of sudden downtime, and extend equipment lifespan. This approach not only protects the health of the equipment but also achieves scientific management throughout its entire lifecycle.

[0161] By optimizing equipment health and operational efficiency, the system reduces energy waste caused by malfunctions or performance degradation. For example, high-efficiency equipment typically consumes less energy, and a lower failure rate means a significant reduction in unnecessary resource consumption. Furthermore, timely replacement of critical components through personalized maintenance plans effectively prevents the release of potentially harmful emissions during equipment failures. This green operation and maintenance approach not only improves the economic efficiency of equipment but also aligns with the company's environmental protection and sustainable development goals.

[0162] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do 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 process, method, article, or apparatus.

[0163] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for self-healing hardware device faults, characterized in that, include: Hardware device failures are categorized into Level 1 and Level 2 failures. When a hardware device malfunctions, determine whether the malfunction is classified as a Level 1 malfunction. If the type of the fault is a first-class fault, a feature vector X = {x1, x2,..., x n} of the fault is extracted, wherein x i represents a device log keyword frequency and a performance index; A fault case library C is set, which contains all the historical fault cases, and each fault case is represented as C i = (A i , P i , R i , T i ), wherein A i is a fault feature vector, P i is a fault solution, R i is a solution result, and T i is a timestamp of the fault case; Based on the feature vector X and the fault case library C, a classification model prediction strategy is executed to predict the probability that the fault solution for each fault case will successfully resolve the current fault. Obtain the probability of all failure solutions, calculate the mean of the probabilities, and record the result as the average probability. Collect all types of fault solutions with a probability greater than the average probability, and denote them as the candidate type set; Based on the feature vector X and the candidate category set, a ranking model optimization strategy is executed to optimize the priority of each fault solution in the candidate category set; obtaining a fault solution with the highest priority, denoted as target solution P u ; Apply target solution P u Repair current fault, record feedback result as R u ; Based on the feedback results, the model for recommending fault solutions is dynamically optimized; Determine whether the faulty equipment has been successfully repaired. If it has been successfully repaired, the hardware equipment that has been repaired for a Level 1 fault is considered to have experienced a Level 2 fault. After repairing hardware devices with a Level 1 fault, a maintenance plan is dynamically generated based on multi-dimensional factors. If the repair fails, each element of the target solution removed from the candidate set will be used sequentially to repair the hardware device.

2. The hardware device fault self-healing method according to claim 1, characterized in that, The step of executing a classification model prediction strategy based on feature vector X and fault case library C to predict the probability that the fault solution for each fault case will successfully resolve the current fault includes: Obtain all distinct fault solutions from the fault case library and form a solution category set; For any fault solution in the set of solution categories, assign a score Z to the fault solution. obtaining a feature vector X = {x1, x2,..., x n}; Calculate Z = W·X + b, where W is the weight matrix, X is the model parameters, and b is the bias term; The following formula converts the scores of fault solutions into probabilities: Obtain the number of elements in the solution category set, denoted as the total number of categories K; Number the fault solutions in the solution category set from 1 to K; calculate Where j = 1, 2, ..., K, P j Let j be the probability of the faulty solution. Calculate the cross-entropy loss: Where n is the number of samples, K is the number of classes, and y ij For the one-hot encoding of the real label, P ij For probability; The result of calculating the cross-entropy loss is recorded as the loss value; A loss threshold is set, which is used to determine the accuracy of the prediction result; When the loss value is less than or equal to the loss threshold, the classification model prediction strategy is terminated. When the loss value exceeds the loss threshold, gradient descent is used to update parameters W and b, specifically: calculate As the new W, Cross-entropy loss; calculate As the new b, where η is the learning rate; The classification model prediction strategy is executed again using the new W and b as parameters.

3. The hardware device fault self-healing method according to claim 1, characterized in that, The step of executing a ranking model optimization strategy based on the feature vector X and the candidate category set to optimize the priority of each fault solution in the candidate category set includes: The number of all elements in the candidate category set is obtained, denoted as F; Sort each fault solution in the candidate category set from 1 to F; For any fault solution in the candidate set, set a score sco for the fault solution; obtaining a feature vector X = {x1, x2,..., x n}; Among them, s j For the fault solution j, 1≤j≤F, Used to extract feature vector X and candidate schemes s j The features are: W is the weight matrix, represents the parameters of the model, and b is the bias term. Obtain the score for each fault solution in the candidate category set, and sort the fault solutions according to the scores; Let the two failure resolution solutions in the arbitrary acquired candidate category set be denoted as s p and s q , respectively, where the score of s p is higher than that of s q ; Calculate ∑ (p，q) max(0, 1-(sco(X, s)) p )-sco(X, s q Record the result as the sort value; Compare the sorted value with 1: if the sorted value > 1, stop executing the sorting model optimization strategy; If the sort value is ≤1, then the gradient descent method is used to update the parameters W and b, specifically: calculate As the new W; calculate As the new b, where η is the learning rate; The sorting model optimization strategy is executed again using the new W and b as parameters.

4. The hardware device fault self-healing method according to claim 1, characterized in that, The model for dynamically optimizing and recommending fault solutions based on feedback results includes: The feedback results in, -1 indicates the fault is resolved, -1 indicates the fault is not resolved, T max Maximum allowed processing time; Define strategy π(P) u |X u ;θ), indicating that the fault characteristic is X u Time-based fault solution P u The probability distribution of u, where 0 ≤ u ≤ n; Set a discount factor γ, 0≤γ≤1, to balance immediate rewards and long-term rewards; Calculate cumulative rewards Set a reward threshold; When the cumulative reward is less than or equal to the reward threshold, the policy gradient method is used to optimize θ.

5. The hardware device fault self-healing method according to claim 1, characterized in that, When a hardware device that has experienced a Level 1 fault is repaired, a maintenance plan is dynamically generated based on multiple factors, including: Establish equipment state model: M t =M0-α·V t -β·Y t +γ·L t ; Among them, M t This represents the health status of the hardware device at time t, where 0 indicates a fault and 1 indicates normal operation. M0 represents the initial healthy state; V t The workload intensity of the hardware device at time t; Y t The hardware device accumulates its running time over time t; L t The strength of maintenance measures for hardware equipment at time t; α, β, and γ are the weighting factors for the impact of workload, service life, and maintenance measures on health status, respectively.

6. The hardware device fault self-healing method according to claim 1, characterized in that, After repairing a hardware device that has experienced a Level 1 fault, a maintenance plan is dynamically generated based on multi-dimensional factors.

Images