An artificial intelligence-based data center operation and maintenance management method and management system

Abstract

The application relates to the technical field of data center operation and maintenance management, in particular to a data center operation and maintenance management method based on artificial intelligence, which comprises a protocol feature dynamic learning module, a time sequence arrangement optimization module, a cascading failure protection module and a multi-modal abnormality self-healing module. Through dynamic updating of a protocol knowledge base, generation of an optimal instruction sequence, prediction of a power consumption conduction path and generation of an incremental rollback scheme, the systemic failure problem in firmware upgrading of heterogeneous devices is solved. The application can intelligently coordinate instructions, predict risks and realize abnormality self-healing, and significantly improves the stability and efficiency of data center operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of data center operation and maintenance management technology, specifically an artificial intelligence-based data center operation and maintenance management method and management system. Background Technology

[0002] With the scaling up of data centers, modern data centers generally adopt a hybrid architecture using multi-brand servers to achieve a balance between cost and performance. However, in critical operational scenarios such as firmware upgrades, the differences in the proprietary extensions of the Redfish protocol implemented by various vendors' Baseboard Management Controllers (BMCs) have become a major technical bottleneck restricting operational automation. Mainstream vendors (such as HP iLO, Dell iDRAC, and Inspur BMC) generally introduce proprietary instruction sets and response timing logic when implementing the Redfish protocol. For example, iLO v2.3 requires a 300±50ms power policy negotiation window before sending the FirmwareUpdate command, while iDRAC v4.2 mandates continuous heartbeats (interval ≤2 seconds) during the firmware transmission phase. This millisecond-level timing difference causes uncontrollable deviations in the state machines of different brands of equipment during the same batch of upgrade operations, significantly increasing operational complexity.

[0003] Furthermore, when the timing misalignment of heterogeneous device upgrades exceeds the synchronization tolerance window of the Power Distribution Unit (PDU) (typically 200-500ms), it can trigger an avalanche-like failure of the power strategy. A typical case demonstrates that during a full upgrade in a hybrid rack (containing 10 HP / Dell / Inspur devices each), the instantaneous power consumption spike of the iLO device (approximately 300W / device) and the firmware verification cycle of the iDRAC (consuming 150W / device every 5 seconds) can have a cumulative effect, causing the PDU to trigger cascading overload protection within 3.2 seconds, resulting in an unexpected power outage for the entire rack. This cascading failure propagation mechanism further exacerbates systemic risks.

[0004] Existing artificial intelligence technologies have significant limitations in addressing the aforementioned problems. While traditional rule-based automated operations and maintenance systems can identify explicit protocol errors (such as HTTP status code anomalies), they cannot dynamically learn extended protocol features and still rely on manual reverse engineering of vendor instruction set knowledge bases. For new BMC firmware (such as iLO 6 with AI acceleration), feature extraction cycles can take months, making it difficult to meet the demands of rapid iteration. Furthermore, existing machine learning models (such as LSTM networks) are mostly used for macro-level resource scheduling and lack the ability to model microsecond-level instruction interaction timing, making it difficult to predict the cascading effects of cross-vendor operations.

[0005] Mainstream industry solutions typically employ manual, batch-based operations (e.g., upgrading in batches by brand). However, this not only extends the maintenance cycle by more than three times but also fails to avoid the hidden risks in scenarios where cross-brand equipment coexists. Therefore, there is an urgent need for an AI-based data center operations and maintenance method. This method should fundamentally solve the systemic failure problem caused by firmware upgrades of heterogeneous equipment by deeply analyzing protocol extension characteristics, dynamically predicting power load conduction paths, and intelligently orchestrating microsecond-level operation timing, thereby improving the automation level and reliability of data center operations and maintenance. Summary of the Invention

[0006] The purpose of this invention is to provide an artificial intelligence-based data center operation and maintenance management method, which solves the systemic failure problem of heterogeneous devices in key operation and maintenance scenarios such as firmware upgrades through intelligent command coordination, risk prediction and anomaly self-healing mechanism.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a data center operation and maintenance management method based on artificial intelligence, the operation and maintenance management method including a protocol feature dynamic learning module, a timing orchestration optimization module, a cascading failure protection module, and a multimodal anomaly self-healing module;

[0008] The protocol feature dynamic learning module is used to collect the timing features of Redfish protocol extended response from multi-vendor board management controllers (BMCs) and build a dynamically updated protocol knowledge base.

[0009] The timing orchestration optimization module is used to generate the optimal instruction transmission sequence based on the protocol knowledge base and insert a dynamic buffer interval in combination with the load data of the real-time power distribution unit (PDU).

[0010] The cascaded failure protection module is used to predict microsecond-level power transmission paths through a spatiotemporal graph attention network (ST-GAT) and generate load shaping strategies to avoid overload risks.

[0011] The multimodal anomaly self-healing module is used to synchronously analyze BMC log streams, PDU telemetry data, and network packet characteristics, and uses reinforcement learning-driven decision forests to generate incremental rollback schemes.

[0012] The protocol feature dynamic learning module is equipped with a reinforcement learning agent, which interacts with BMCs from multiple vendors. When a new protocol extension feature is detected, the reinforcement learning agent automatically extracts its timing logic and updates the protocol knowledge base. At the same time, it generates a dynamic buffer time suggestion and transmits it to the timing orchestration optimization module.

[0013] The timing orchestration optimization module includes millisecond-level convolutional branches, second-level convolutional branches, and a quantum annealing solver. The millisecond-level convolutional branches are used to extract short-term sensitive operation features, the second-level convolutional branches are used to capture long-term dependencies, and the quantum annealing solver solves the optimal instruction sequence that minimizes the PDU load fluctuation variance by mapping the device operation timing combination optimization problem. The specific formula is as follows:

[0014]

[0015] Among them, P i (t) represents the instantaneous power consumption of the i-th device at time t. Let t represent the average power consumption of all devices at time t, N represent the total number of devices, and T represent the total duration. This formula enables precise control of PDU load fluctuations, ensuring the stability of device operation timing.

[0016] The cascaded failure protection module incorporates a spatiotemporal graph attention network (ST-GAT), with a time sliding window of 1 second. The spatial adjacency matrix is ​​constructed based on circuit topology distance, and a multi-head attention mechanism captures microsecond-level power transmission paths across racks. When an overload risk is predicted, a gradient backpropagation algorithm is used to calculate the equipment operation delay, as shown in the following formula:

[0017]

[0018] Where, Δt i This represents the operation delay of the i-th device. P represents the maximum allowable load of the power distribution unit. total (t) represents the total load at time t. This represents the rate of change in power consumption of the i-th device. This formula is used to dynamically adjust the device operation timing, preventing overload from triggering the protection mechanism.

[0019] The multimodal anomaly self-healing module employs a fusion Transformer architecture. It utilizes Transformer to synchronously analyze BMC log streams, PDU telemetry data, and network packet characteristics. When a cross-layer anomaly pattern is detected, an incremental rollback scheme is generated through a reinforcement learning-driven decision forest. The splitting threshold for each tree node in the decision forest is dynamically optimized using Q-Learning, with the specific formula as follows:

[0020]

[0021] Where Q(s, a) represents the expected reward of taking action a in the current state s, α represents the learning rate, r represents the immediate reward, γ represents the discount factor, and s′ represents the next state. This formula is used to optimize the rollback strategy with the goal of minimizing the range of affected devices.

[0022] In addition, the present invention also provides an artificial intelligence-based data center operation and maintenance management system, which is applicable to the artificial intelligence-based data center operation and maintenance management method. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the module structure of the data center operation and maintenance management method based on artificial intelligence in an embodiment of the present invention.

[0024] Figure 2 This is a flowchart of the timing orchestration optimization module in an embodiment of the present invention.

[0025] Figure 3 This is a schematic diagram of the spatiotemporal graph attention network model structure of the cascaded failure protection module in an embodiment of the present invention.

[0026] Figure 4 This is a diagram illustrating the working principle of the fusion Transformer architecture and reinforcement learning-driven decision forest of the multimodal anomaly self-healing module in this embodiment of the invention. Detailed Implementation

[0027] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0028] This invention provides an artificial intelligence-based data center operation and maintenance management method. Its core lies in the collaborative work of a protocol feature dynamic learning module, a timing orchestration optimization module, a cascading failure protection module, and a multimodal anomaly self-healing module to achieve intelligent management of heterogeneous data center equipment in critical operation and maintenance scenarios. The specific implementation methods of this invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] like Figure 1As shown, the module structure of this invention includes a protocol feature dynamic learning module, a timing orchestration optimization module, a cascading failure protection module, and a multimodal anomaly self-healing module. The protocol feature dynamic learning module transmits dynamic buffer time suggestions to the timing orchestration optimization module. After the timing orchestration optimization module generates the instruction sequence, the cascading failure protection module provides real-time feedback on the load shaping strategy, and the multimodal anomaly self-healing module generates rollback instructions based on the global state, forming a closed-loop control flow. The specific implementation begins with the protocol feature dynamic learning module. The core function of this module is to collect the Redfish protocol extension response timing features of multi-vendor Baseboard Management Controllers (BMCs) and build a dynamically updated protocol knowledge base. In practical applications, data centers typically contain equipment from different vendors. These devices' BMCs support the Redfish protocol, but each vendor may have extended the protocol to varying degrees. To address this heterogeneity, this module sets up a reinforcement learning agent, which automatically extracts new protocol extension features and updates the protocol knowledge base by interacting with multi-vendor BMCs. For example, when a vendor adds a specific firmware upgrade command sequence, the reinforcement learning agent can detect this change and incorporate it into the knowledge base by analyzing its timing logic. The protocol feature dynamic learning module uses a graph database (Neo4j) to store the Redfish protocol extended response timing features of multi-vendor baseboard management controllers (BMCs). Simultaneously, this module also generates dynamic buffer time suggestions based on the new features for use by subsequent modules. The formula for calculating the suggested value of the dynamic buffer time is as follows:

[0030]

[0031] Where, Δt i Let represent the average response latency of the i-th protocol extension feature, and M represent the number of current protocol extension features. This formula ensures the rationality of the dynamic buffer time, thus providing a stable timing basis for subsequent operations.

[0032] Next, the timing orchestration optimization module generates the optimal command transmission sequence based on the protocol knowledge base generated by the protocol feature dynamic learning module, and inserts a dynamic buffer interval by combining it with the load data of the real-time power distribution unit (PDU). For example... Figure 2 As shown, this module includes millisecond-level convolutional branches, second-level convolutional branches, and a quantum annealing solver. The millisecond-level convolutional branches are used to extract short-term sensitive operational features, such as operations requiring rapid response during device startup or firmware upgrades; the second-level convolutional branches are used to capture long-term dependencies, such as performance fluctuations that may occur during long-term device operation. The quantum annealing solver is responsible for mapping the device operation timing combination optimization problem into a mathematical model and ensuring system stability by solving for the optimal instruction sequence that minimizes the variance of PDU load fluctuations. The specific optimization formula is as follows:

[0033]

[0034] Among them, P i (t) represents the instantaneous power consumption of the i-th device at time t. Let t represent the average power consumption of all devices at time t, N represent the total number of devices, and T represent the total duration. This formula allows for precise control of PDU load fluctuations, preventing system instability caused by uneven load distribution. In practical applications, assuming a data center has 100 servers, each with a power consumption ranging from 200W to 500W, the optimal instruction sequence generated by the quantum annealing solver ensures a reasonable distribution of operation timing for each device, thereby controlling PDU load fluctuations within 5%.

[0035] The cascaded failure prevention module predicts microsecond-level power conduction paths using a spatiotemporal graph attention network (ST-GAT) and generates load shaping strategies to mitigate overload risks. For example... Figure 3 As shown, the input to this module includes device firmware features and circuit topology. The time sliding window of the spatiotemporal graph attention network is set to 1 second, and the spatial adjacency matrix is ​​constructed based on the circuit topology distance. A multi-head attention mechanism is used to capture cross-rack power transmission paths. For example, in a data center, if devices in a rack start up simultaneously, it may cause a sudden surge in PDU load. In this case, the ST-GAT model will predict this overload risk and calculate the device operation latency using the gradient backpropagation algorithm, as shown in the following formula:

[0036]

[0037] Where, Δt i This represents the operation delay of the i-th device. P represents the maximum allowable load of the PDU. total (t) represents the total load at time t. This represents the rate of change in power consumption of the i-th device. Using this formula, the module can dynamically adjust the device operation sequence to avoid overload triggering the protection mechanism. For example, when the total load of devices in a cabinet approaches the maximum allowable load of a PDU, the module will automatically delay the operation of some devices, thereby keeping the load within a safe range.

[0038] The multimodal anomaly self-healing module integrates the Transformer architecture to synchronously analyze BMC log streams, PDU telemetry data, and network packet characteristics. It utilizes a reinforcement learning-driven decision forest to generate incremental rollback schemes. This decision forest consists of 100 CART trees, and the node splitting threshold is dynamically optimized through offline pre-training (historical fault dataset) and online fine-tuning (real-time Q-Learning). The state space 's' is defined as device health (0-1), PDU load deviation rate (%), and network packet loss rate (%). The action space 'a' includes rollback priority (high / medium / low) and rollback granularity (single device / rack / cluster). For example... Figure 4 As shown, the core of this module lies in extracting abnormal patterns from multimodal data using the Transformer architecture and combining it with reinforcement learning to optimize rollback strategies. For example, when a device's BMC log shows an abnormal restart record, and PDU telemetry data shows a sudden drop in power consumption, and network packet characteristics indicate communication interruption, the module will determine that the device may have a serious fault. At this time, the reinforcement learning-driven decision forest will generate an incremental rollback scheme, with the specific formula as follows:

[0039]

[0040] Where Q(s, a) represents the expected reward of taking action a in the current state s, α represents the learning rate, r represents the immediate reward, γ represents the discount factor, and s′ represents the next state. Using this formula, the module can optimize the rollback strategy to minimize the range of affected devices. For example, during a firmware upgrade, if some devices are found to be unable to function properly due to compatibility issues, the module will prioritize rolling back the firmware versions of the affected devices while retaining the upgrade results of other devices, thereby minimizing the impact on the overall system.

[0041] Furthermore, this invention also provides an artificial intelligence-based data center operation and maintenance management system, applicable to the aforementioned operation and maintenance management methods. In practical deployment, this system can support large-scale data centers through a distributed architecture. For example, in a hyperscale data center, the system can deploy a protocol feature dynamic learning module on edge nodes close to the devices to reduce data transmission latency; deploy a timing orchestration optimization module and a cascading failure protection module on a central control node to centrally handle global optimization tasks; and deploy a multimodal anomaly self-healing module in the cloud to fully utilize the high-performance computing capabilities of cloud computing. Through this layered deployment approach, the system can achieve comprehensive intelligent management of the data center while ensuring real-time performance.

[0042] In summary, this invention solves the systemic failure problem of heterogeneous devices in critical operation and maintenance scenarios such as firmware upgrades by working collaboratively through a protocol feature dynamic learning module, a timing orchestration optimization module, a cascading failure protection module, and a multimodal anomaly self-healing module. Verification through practical application scenarios shows that this invention can significantly improve the operation and maintenance efficiency and system stability of data centers, providing reliable technical support for the intelligent management of large-scale data centers.

[0043] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A data center operation and maintenance management method based on artificial intelligence, characterized in that: The operation and maintenance management method includes a protocol feature dynamic learning module, a timing orchestration optimization module, a cascading failure protection module, and a multimodal anomaly self-healing module. The protocol feature dynamic learning module is used to collect the timing features of Redfish protocol extended response from multi-vendor board management controllers and build a dynamically updated protocol knowledge base. The timing orchestration optimization module is used to generate the optimal instruction transmission sequence based on the protocol knowledge base and insert a dynamic buffer interval in combination with the load data of the real-time power distribution unit. The cascaded failure protection module is used to predict microsecond-level power transmission paths through a spatiotemporal graph attention network and generate load shaping strategies to avoid overload risks. The multimodal anomaly self-healing module is used to synchronously analyze the log stream of the baseboard management controller, the telemetry data of the power distribution unit, and the characteristics of network packets, and to generate an incremental rollback scheme using a reinforcement learning-driven decision forest.

2. The data center operation and maintenance management method based on artificial intelligence according to claim 1, characterized in that: The protocol feature dynamic learning module is equipped with a reinforcement learning agent, which interacts with the multi-vendor baseboard management controller. When a new protocol extension feature is detected, the reinforcement learning agent automatically extracts its timing logic and updates the protocol knowledge base. At the same time, it generates a dynamic buffer time suggestion and transmits it to the timing orchestration optimization module.

3. The data center operation and maintenance management method based on artificial intelligence according to claim 1, characterized in that: The timing orchestration optimization module includes millisecond-level convolutional branches, second-level convolutional branches, and a quantum annealing solver. The millisecond-level convolutional branches are used to extract short-term sensitive operation features, the second-level convolutional branches are used to capture long-term dependencies, and the quantum annealing solver solves the optimal instruction sequence that minimizes the load fluctuation variance of the power distribution unit by mapping the device operation timing combination optimization problem.

4. The data center operation and maintenance management method based on artificial intelligence according to claim 1, characterized in that: The cascaded failure protection module is equipped with a spatiotemporal graph attention network. The spatiotemporal graph attention network is used to input device firmware features and circuit topology. The time sliding window of the spatiotemporal graph attention network is set to 1 second. The spatial adjacency matrix is ​​constructed based on the circuit topology distance. The power transmission path across racks is captured through a multi-head attention mechanism. When an overload risk is predicted, the gradient backpropagation algorithm is used to calculate the device operation delay.

5. The data center operation and maintenance management method based on artificial intelligence according to claim 4, characterized in that: The cascaded failure protection module calculates the device operation delay using the following formula: ； in, Indicates the first The operational delay of the device. Indicates the maximum allowable load of the power distribution unit. express Total load at any given time Indicates the first The rate of change in power consumption of the device.

6. The data center operation and maintenance management method based on artificial intelligence according to claim 1, characterized in that: The multimodal anomaly self-healing module is equipped with a fusion Transformer architecture. It uses Transformer to synchronously analyze the log stream of the baseboard management controller, telemetry data of the power distribution unit, and network packet characteristics. When a cross-layer anomaly mode is detected, an incremental rollback scheme is generated through a reinforcement learning-driven decision forest.

7. The data center operation and maintenance management method based on artificial intelligence according to claim 6, characterized in that: The reinforcement learning-driven decision forest optimizes the rollback strategy using the following formula: ； in, Indicates the current state Take action below Expected returns Indicates the learning rate. Indicates an immediate reward. Indicates the discount factor. Indicates the next state. Indicates the state The next action to be taken.

8. A data center operation and maintenance management system based on artificial intelligence, characterized in that: The intelligent operation and maintenance management system is applicable to the data center operation and maintenance management method based on artificial intelligence as described in any one of claims 1 to 7.

9. The data center operation and maintenance management system based on artificial intelligence according to claim 8, characterized in that: The intelligent operation and maintenance management system supports large-scale data centers through a distributed architecture. The protocol feature dynamic learning module is deployed on edge nodes close to the device, the timing orchestration optimization module and the cascading failure protection module are deployed on the central control node, and the multimodal anomaly self-healing module is deployed in the cloud.

Images