Data center dynamic environment decision-making method and system

Abstract

The invention provides a data center dynamic environment decision-making method and system, which can be applied to the technical field of data center intelligent dynamic environment operation and maintenance. The data center dynamic environment decision-making method comprises the following steps: collecting dynamic environment multi-modal data of a data center in real time; performing time sequence feature extraction processing on the moving ring multi-modal data to obtain moving ring time sequence features, and constructing unified physical state representation by using a clustering analysis result of the historical fault information and the moving ring time sequence features; executing a physical rule embedding operation on the unified physical state representation to obtain a physical constraint result, and executing a simulation deduction operation based on a world simulation engine on the unified physical state representation to obtain a simulation deduction result; executing a decision generation operation based on a large language model by using the physical constraint result and the simulation deduction result to obtain an initial decision scheme, and executing a multi-dimensional evaluation operation and a scheme optimization operation based on physical verification on the initial decision scheme to obtain a decision scheme; and interactively displaying the decision scheme to the user.

Description

Technical Field

[0001] This invention relates to the field of intelligent power environment operation and maintenance technology for data centers, and specifically to a data center power environment decision-making method and system. Background Technology

[0002] Data center environmental monitoring (ORM) systems are core systems for ensuring the stable operation of data centers. Through real-time monitoring, intelligent early warning, and centralized management, they improve operational efficiency and reduce risks. Currently, data center ORM systems suffer from several problems in fault handling: they can only provide a single ORM decision-making solution, lacking the ability to generate multiple feasible ORM decisions; they lack an evaluation mechanism for the generated ORM decisions; the generated ORM decisions ignore physical constraints; and some purely data-driven ORM decisions lack forward-looking analysis capabilities. Furthermore, some existing data center ORM systems typically use a simple chain approach of "physical model diagnosis + AI (Artificial Intelligence) suggestions" in the ORM decision-making process, failing to achieve deep two-way integration of physical rules and large-scale models, resulting in a lack of closed-loop feedback in the ORM decisions. Summary of the Invention

[0003] In view of the above problems, the present invention provides an improved method and system for making environmental decisions in data centers, which at least solves one of the above-mentioned technical problems.

[0004] According to a first aspect of the present invention, a data center dynamic environment decision-making method is provided, comprising: real-time acquisition of dynamic environment multimodal data of the data center; performing time-series feature extraction processing on the dynamic environment multimodal data to obtain dynamic environment time-series features, and constructing a unified physical state representation using the clustering analysis results of historical fault information and the dynamic environment time-series features; performing a physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain physical constraint results, and performing a simulation inference operation based on a world simulation engine on the unified physical state representation to obtain simulation inference results; performing a decision generation operation based on a large language model using the physical constraint results and simulation inference results to obtain an initial decision scheme, and performing a multi-dimensional evaluation operation and a scheme optimization operation based on physical verification on the initial decision scheme to obtain a decision scheme; and interactively displaying the decision scheme to the user.

[0005] According to an embodiment of the present invention, the above-mentioned time series feature extraction processing of dynamic environment multimodal data to obtain dynamic environment time series features includes: performing data standardization operation on dynamic environment multimodal data to obtain dynamic environment multimodal standard data; extracting time series features from dynamic environment multimodal standard data using sliding window technology and physical constraint regularization technology to obtain dynamic environment initial time series features; and performing data verification on the dynamic environment initial time series features to obtain dynamic environment time series features.

[0006] According to an embodiment of the present invention, the above-mentioned construction of a unified physical state representation using the clustering analysis results of historical fault information and dynamic environment time-series features includes: performing physical dimension analysis on the dynamic environment time-series features to obtain dynamic environment time-series features that conform to physical dimension rules; constructing a physical state vector and scene feature indicators for the data center, and fusing the physical state vector and scene feature indicators into the dynamic environment time-series features that conform to physical dimension rules to obtain the dynamic environment target time-series features; performing similarity matching and clustering analysis operations on historical fault information to obtain the clustering analysis results of historical fault information; and constructing a unified physical state representation using the dynamic environment target time-series features and the clustering analysis results of historical fault information.

[0007] According to an embodiment of the present invention, the above-mentioned physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain physical constraint results includes: performing a dynamic transformation operation on the unified physical state representation to obtain initial natural language prompt information and initial structured constraint information; performing a constraint injection operation on the initial natural language prompt information and initial structured constraint information using a preset large model to obtain natural language prompt information and structured constraint information; embedding the physical rules contained in the unified physical state representation as an inherent constraint mechanism into the decision generation operation based on the large language model using the natural language prompt information and structured constraint information; and, during the execution of the dynamic transformation operation and the constraint injection operation, performing a scene-adaptive rule strength adjustment operation on the physical rules based on a fault classification system using a dynamic strength adjustment algorithm to obtain physical constraint results.

[0008] According to embodiments of the present invention, the above-mentioned dynamic transformation operation on the unified physical state representation to obtain initial natural language prompt information and initial structured constraint information includes: performing a dynamic transformation operation on the unified physical state representation using rule template technology to obtain original natural language prompt information and original structured constraint information; performing a dynamic adjustment operation on the original natural language prompt information and original structured constraint information using a scene-aware rule generation mechanism based on the current fault scenario information of the data center to obtain intermediate natural language prompt information and intermediate structured constraint information; during the execution of the dynamic transformation operation and the dynamic adjustment operation, the automatic identification and processing of rule conflicts is achieved by introducing a rule conflict detection mechanism; and performing multi-granularity expression processing on the intermediate natural language prompt information and intermediate structured constraint information to obtain initial natural language prompt information and initial structured constraint information.

[0009] According to an embodiment of the present invention, the process of embedding the physical rules implied by the unified physical state representation as an intrinsic constraint mechanism into the decision generation operation based on a large language model by utilizing natural language prompts and structured constraint information includes: adding natural language prompts and structured constraint information before the input data using prefix injection technology to obtain input data with physical constraint prompts; using the input data with physical constraint prompts to guide the decision generation operation based on the large language model to conform to physical laws; performing a filtering operation based on a constraint decoding mechanism on tokens that violate physical rules in the token generation stage of the decision generation operation based on the large language model, and dynamically adjusting the token generation probability distribution based on a physical consistency sampling strategy; and dynamically comparing and correcting the result of the decision generation operation based on the large language model with the physical rules based on the introduction of a rule-decision closed-loop mechanism.

[0010] According to an embodiment of the present invention, the above-mentioned simulation and deduction operation based on the world simulation engine is performed on the unified physical state representation to obtain the simulation and deduction results, including: performing multi-physics field coupling simulation operation on the unified physical state representation to obtain a multimodal physical simulator of the world simulation engine; performing simulation and fault propagation risk assessment operation on the fault propagation of the data center based on the fault propagation rules, using the critical path analysis algorithm and the fault propagation graph model of the world simulation engine; and performing time-based hybrid simulation and deduction operation on the fault of the data center using the multimodal physical simulator and the large language model to obtain the simulation and deduction results.

[0011] According to an embodiment of the present invention, the above-mentioned multi-physics field coupling simulation operation performed on the unified physical state representation to obtain a multi-modal physical simulator of the world simulation engine includes: performing multi-physics field coupling simulation operation on the unified physical state representation using the order reduction model technique to obtain a multi-modal physical simulator, wherein the multi-modal physical simulator includes a thermal field model based on computational fluid dynamics and an electric field model based on a directed graph.

[0012] According to an embodiment of the present invention, the above-described hybrid simulation and deduction operation based on the time dimension of a data center fault, performed using a multimodal physical simulator and a large language model, to obtain simulation and deduction results includes: performing a detection operation on key points of the data center using the key point prediction model of the world simulation engine to obtain key point prediction results; constructing prompt word instructions using the key point prediction operation; using the prompt word instructions to drive the large language model to perform a simulation and deduction operation based on the time dimension to obtain the large language model simulation and deduction results; performing a simulation and deduction operation based on the time dimension of the physical state of the data center using a multimodal physical simulator to obtain physical simulation and deduction results; and performing a weighted fusion processing of the large language model simulation results and the physical simulation results to obtain the simulation and deduction results.

[0013] According to a second aspect of the present invention, a data center environmental decision-making system is provided, comprising: a data acquisition module for real-time acquisition of environmental multimodal data from the data center; a data representation and fusion module for performing time-series feature extraction processing on the environmental multimodal data to obtain environmental time-series features, and constructing a unified physical state representation using the clustering analysis results of historical fault information and the environmental time-series features; a rule embedding and engine simulation module for performing a physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain physical constraint results, and performing a simulation deduction operation based on a world simulation engine on the unified physical state representation to obtain simulation deduction results; a decision generation and evaluation module for performing a decision generation operation based on a large language model using the physical constraint results and simulation deduction results to obtain an initial decision scheme, and performing a multi-dimensional evaluation operation and a scheme optimization operation based on physical verification on the initial decision scheme to obtain a decision scheme; and a user interaction module for interactively displaying the decision scheme to the user.

[0014] A third aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0015] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0016] The data center environmental impact assessment (EIA) decision-making method provided by this invention focuses on optimizing decisions for known faults in data centers. It provides a comprehensive data foundation for EIA decision generation through multimodal data fusion, ensures the physical feasibility of generated decisions through simulation based on physical rule embedding and a world simulation engine, and rapidly generates decision solutions by combining the reasoning capabilities of a large language model. Furthermore, it optimizes decisions through multi-dimensional evaluation and physical verification mechanisms, enhancing the scientific rigor and reliability of the decisions. This invention addresses the shortcomings of traditional data center EIA decision-making methods, which often only provide single solutions or lack systematic evaluation mechanisms. It offers scientific, objective, and actionable decision support for data center EIA management, significantly improving the efficiency and quality of data center fault handling. Attached Figure Description

[0017] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0018] Figure 1 The diagram illustrates an application scenario of the data center environmental decision-making method according to an embodiment of the present invention.

[0019] Figure 2 A flowchart illustrating a data center environmental decision-making method according to an embodiment of the present invention is shown.

[0020] Figure 3 The diagram illustrates a multi-level intelligent decision-making process for the data center's dynamic environment according to an embodiment of the present invention.

[0021] Figure 4 A schematic diagram illustrating the data processing flow of a world simulation engine according to an embodiment of the present invention is shown.

[0022] Figure 5 A flowchart illustrating decision generation and evaluation according to an embodiment of the present invention is shown schematically.

[0023] Figure 6 A schematic diagram of a data center environmental decision system according to an embodiment of the present invention is shown.

[0024] Figure 7 A block diagram of an electronic device suitable for implementing a data center environmental decision-making method according to an embodiment of the present invention is shown schematically. Detailed Implementation

[0025] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0028] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0029] The current data center environmental monitoring system has the following problems in the fault handling stage (i.e., operation and maintenance decision stage): (1) When the fault has been detected, the existing data center environmental monitoring system can only provide a single solution (i.e., operation and maintenance decision scheme) or simple operation suggestions, and lacks the ability to generate multiple feasible solutions. When faced with a confirmed fault, operation and maintenance personnel can only perform fixed operations according to the preset process and cannot flexibly choose the optimal path according to the actual situation. (2) The traditional data center environmental monitoring system lacks a systematic evaluation mechanism for different operation and maintenance decision schemes. The unique multi-system coupling characteristics of data centers (the mutual influence of power, cooling, IT (Information Technology) load, etc.) make the scheme evaluation more complex, and a single-dimensional evaluation cannot reflect the comprehensive effect of the scheme. Even if experienced operation and maintenance personnel can propose multiple handling schemes, there is a lack of objective and quantitative multi-dimensional evaluation standards. The decision-making process is highly dependent on personal experience, resulting in unstable decision quality. (3) Existing data center dynamic environment decision-making technologies often neglect physical constraints. For example, in a cooling system failure scenario, it may be suggested to "increase the air conditioner setting temperature to save energy," without considering that this may cause local hotspots to exceed the safe operating temperature of the equipment, resulting in secondary failures. (4) Some other operation and maintenance decision-making methods based on pure data lack forward-looking analysis capabilities and cannot predict the evolution of the system state after the implementation of different decision-making schemes. This leads to decisions that only focus on short-term effects while ignoring long-term impacts, which may cause problems such as shortened equipment lifespan. (5) Existing technical solutions mostly adopt a simple chain approach of "physical model diagnosis + AI-provided suggestions," which fails to achieve deep bidirectional integration of physical rules and large models. Physical rules are only used as a post-verification tool, rather than an inherent constraint mechanism for decision generation, resulting in a lack of closed-loop feedback in the decision-making process.

[0030] The aforementioned issues often prevent data centers from choosing the truly optimal solution when faced with known faults, resulting in unnecessary business interruptions or resource waste.

[0031] To address at least one of the problems in existing technologies, this invention provides a data center environmental dynamics (EM) decision-making method, applied in the field of data center intelligent power environment (EM) operation and maintenance technology. This method generates multiple feasible decision-making schemes for identified and confirmed faults, and systematically evaluates each scheme using a world model that integrates physical rules and a large language model to select the optimal implementation path. This system can be widely applied to data centers of various sizes, providing scientific, objective, and actionable fault handling decision support for operators of data center EEM monitoring systems.

[0032] Figure 1 The diagram illustrates an application scenario of the data center environmental decision-making method according to an embodiment of the present invention.

[0033] like Figure 1 As shown, application scenario 100 according to this embodiment may include the field of intelligent power environment operation and maintenance technology for data centers. Network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0034] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0035] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0036] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0037] It should be noted that the data center environmental decision-making method provided in this embodiment of the invention can generally be executed by server 105. Correspondingly, the data center environmental decision-making system provided in this embodiment of the invention can generally be set up in server 105. The data center environmental decision-making method provided in this embodiment of the invention can also be executed by a server or server cluster that is different from server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the data center environmental decision-making system provided in this embodiment of the invention can also be set up in a server or server cluster that is different from server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0038] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0039] The following will be based on Figure 1 The described scene, through Figures 2-5 The data center environmental decision-making method according to the disclosed embodiments is described in detail.

[0040] Figure 2 A flowchart illustrating a data center environmental decision-making method according to an embodiment of the present invention is shown.

[0041] like Figure 2 As shown, the data center environmental decision-making method of this embodiment includes operations S210 to S250.

[0042] The S210 is used to collect dynamic environmental multimodal data from the data center in real time.

[0043] Since a data center environmental monitoring system includes multiple subsystems, such as power systems, cooling systems, and IT systems, and each subsystem has different data content and formats, in order to generate more accurate and scientific decision-making solutions, it is necessary to integrate the data from each subsystem, comprehensively capture the operational status of the data center, and improve the breadth and depth of data utilization.

[0044] In operation S220, time-series feature extraction processing is performed on the dynamic environment multimodal data to obtain dynamic environment time-series features, and a unified physical state representation is constructed using the clustering analysis results of historical fault information and dynamic environment time-series features.

[0045] By utilizing temporal feature extraction and historical fault clustering analysis, a unified physical state representation is constructed, which enhances the accuracy of state description and the inheritance of historical experience.

[0046] In operation S230, a physical rule embedding operation based on a two-way fusion mechanism is performed on the unified physical state representation to obtain physical constraint results, and a simulation deduction operation based on a world simulation engine is performed on the unified physical state representation to obtain simulation deduction results.

[0047] By embedding physical rules and a world simulation engine in the simulation, the physical feasibility of the decision-making is ensured, and the effects of different options are predicted through simulation.

[0048] In operation S240, a decision generation operation based on a large language model is performed using physical constraint results and simulation results to obtain an initial decision scheme. Then, a multi-dimensional evaluation operation and a scheme optimization operation based on physical verification are performed on the initial decision scheme to obtain a final decision scheme.

[0049] By combining the reasoning capabilities of large language models, preliminary decision-making schemes can be generated quickly, and optimized through multi-dimensional evaluation and physical verification, thereby improving the scientific nature and reliability of the decisions.

[0050] When operating S250, the decision-making options are interactively displayed to the user.

[0051] The decision-making solution is displayed to the user, along with fault alarm information, operation guides, and risk warning information.

[0052] It should be noted that the data center environmental decision-making method provided by this invention generates multiple feasible decision schemes for identified and confirmed faults, rather than for the detection, location or root cause tracing of faults. It is a technical solution for decision generation and optimization.

[0053] The data center environmental impact assessment (EIA) decision-making method provided by this invention focuses on optimizing decisions for known faults in data centers. It provides a comprehensive data foundation for EIA decision generation through multimodal data fusion, ensures the physical feasibility of generated decisions through simulation based on physical rule embedding and a world simulation engine, and rapidly generates decision solutions by combining the reasoning capabilities of a large language model. Furthermore, it optimizes decisions through multi-dimensional evaluation and physical verification mechanisms, enhancing the scientific rigor and reliability of the decisions. This invention addresses the shortcomings of traditional data center EIA decision-making methods, which often only provide single solutions or lack systematic evaluation mechanisms. It offers scientific, objective, and actionable decision support for data center EIA management, significantly improving the efficiency and quality of data center fault handling.

[0054] The following describes specific implementation methods in conjunction with appendices. Figure 3 The data center environmental decision-making method provided by this invention will be described in further detail.

[0055] Figure 3The diagram illustrates a multi-level intelligent decision-making process for the data center's dynamic environment according to an embodiment of the present invention.

[0056] like Figure 3 As shown, the data center dynamic environment decision-making method provided by this invention is a multi-layered fusion decision intelligence generation and optimization method. It deeply embeds the unique physical rules of data centers, such as thermodynamics and electrical systems, into the decision-making process of a large language model, and constructs a fault decision-making world model that can accurately simulate the physical environment of the data center.

[0057] like Figure 3 As shown, before decision generation and optimization, dynamic environment multimodal data is acquired through an external system interface. This dynamic environment multimodal data mainly refers to confirmed fault information and associated state data, such as... Figure 3 The DCIM (Data Center Infrastructure management) system, BMS (Battery Management System), IT (Information Technology) monitoring system, and digital twin platform provide multi-source data.

[0058] like Figure 3 As shown, in the data representation and fusion layer, the dynamic environment multimodal data obtained from data acquisition is processed, and a physical state space is constructed based on the processing results of the dynamic environment multimodal data. By extracting historical fault modes, the fault state is transformed into a unified physical state representation in the physical state space, thereby using the unified physical state representation as data support for physical rule embedding and simulation inference.

[0059] The data representation and fusion layer, serving as the foundational layer for multi-level intelligent decision-making in the data center's environmental dynamics (Environment and Environment) system, is responsible for transforming the state information of detected faults into a unified physical representation. This is the basis for the entire decision generation and optimization process. The data representation and fusion layer employs a multi-stage data processing workflow to ensure that the state information input to the decision generation and evaluation layer is both comprehensive and conforms to physical laws. This layer focuses on the state representation of detected faults and does not involve the fault detection process, providing an accurate input basis for subsequent decision generation. Through physical consistency processing, it solves the problems of multi-source heterogeneity and implicit physical relationships in data center environmental dynamics data, providing upper layers with a state representation that conforms to physical laws. This is the foundation for achieving the fusion of physical rules and large-scale models.

[0060] like Figure 3As shown, physical constraints (i.e., physical constraint results) and simulation results (i.e., simulation results) are obtained by utilizing a physical rule embedding layer and a world model simulation engine (i.e., the world simulation engine). This achieves deep bidirectional fusion of data center-specific physical rules and large-scale models, and employs a dynamic rule encoding mechanism and dynamic constraint strength adjustment technology during the physical constraint and simulation process. Specifically, the physical rule embedding layer mainly includes a physical knowledge base, rule encoder, constraint injection mechanism, and rule strength adjustment; the world model simulation engine mainly includes multiphysics coupling simulation, fault propagation simulation, and time-dimensional simulation.

[0061] The physical rule embedding layer, as the core innovation layer of the architecture, achieves deep bidirectional fusion of data center-specific physical rules and the large model (i.e., the large language model), which is crucial to ensuring the physical feasibility of decision-making solutions. This physical rule embedding layer employs an innovative bidirectional fusion mechanism, making physical rules not only post-validation tools but also intrinsic constraints on decision generation. This layer breaks through the limitation of traditional methods where physical rules are merely post-validation tools, making physical laws an intrinsic constraint mechanism for decision generation, ensuring that all recommendations meet physical feasibility requirements.

[0062] like Figure 3 As shown, in the decision generation and evaluation layer, based on physical constraints and simulation results, a large model (i.e., a large language model) is used to drive the generation of decision schemes, resulting in multiple feasible initial decision schemes for the same environmental fault. The optimal decision scheme is then selected. Finally, the optimal decision scheme is displayed to the user through the user interaction layer, along with fault alarm information, operation instructions, and risk warning information.

[0063] This invention addresses the limitations of data center environmental monitoring systems (or data center environmental management systems) that can only provide a single solution or lack a system evaluation mechanism. It achieves a precise transformation from "known faults" to "optimal decisions," solving the core pain point in data center environmental management of "knowing the problem but not knowing how to solve it optimally." It provides maintenance personnel with scientific, objective, and actionable fault handling decision support. At the same time, the decision-making method provided by this invention can significantly reduce the mean time to repair faults in data centers, improve energy efficiency, and ensure business continuity.

[0064] Through the above specific implementation methods and in combination Figure 3 It can be seen that:

[0065] This invention provides a fault handling decision generation architecture, which focuses on the fault handling decision-making process by designing a decision optimization architecture for known faults.

[0066] The physical rule embedding mechanism provided by this invention realizes the dynamic encoding of dynamic and environmental rules. It dynamically transforms data center-specific physical rules, such as those involving thermodynamic models, power system models, and environmental parameter coupling models, into natural language prompts and structured constraints that can be understood by the large model based on the characteristics of the detected faults.

[0067] This invention enables dynamic adjustment of the constraint strength of physical rules. It automatically adjusts the constraint strength of physical rules according to the severity of the detected fault. For example, hard constraints are used for level 1 faults (emergency), soft constraints are used for level 2 faults (early warning), and suggestive constraints are used for level 3 faults (potential), thus achieving scenario-adaptive decision-making strategies.

[0068] This invention provides a world model simulation engine (i.e., world simulation engine), which, by constructing a lightweight multiphysics coupled simulator, can deduce the long-term impact of data center fault development processes and decision-making schemes, providing forward-looking analysis in the time dimension, and taking into account data center-specific factors such as IT load distribution and airflow organization.

[0069] This invention provides a multi-dimensional evaluation matrix for data centers. For example, by designing a multi-dimensional evaluation system that considers business continuity, power usage effectiveness (PUE), physical feasibility, operational complexity, and compliance, different decision-making schemes for the same fault can be quantitatively scored to select the globally optimal solution.

[0070] According to an embodiment of the present invention, the above-mentioned time series feature extraction processing of dynamic environment multimodal data to obtain dynamic environment time series features includes: performing data standardization operation on dynamic environment multimodal data to obtain dynamic environment multimodal standard data; extracting time series features from dynamic environment multimodal standard data using sliding window technology and physical constraint regularization technology to obtain dynamic environment initial time series features; and performing data verification on the dynamic environment initial time series features to obtain dynamic environment time series features.

[0071] According to an embodiment of the present invention, the above-mentioned construction of a unified physical state representation using the clustering analysis results of historical fault information and dynamic environment time-series features includes: performing physical dimension analysis on the dynamic environment time-series features to obtain dynamic environment time-series features that conform to physical dimension rules; constructing a physical state vector and scene feature indicators for the data center, and fusing the physical state vector and scene feature indicators into the dynamic environment time-series features that conform to physical dimension rules to obtain the dynamic environment target time-series features; performing similarity matching and clustering analysis operations on historical fault information to obtain the clustering analysis results of historical fault information; and constructing a unified physical state representation using the dynamic environment target time-series features and the clustering analysis results of historical fault information.

[0072] The following detailed explanation of the acquisition process of the above-mentioned dynamic environmental timing characteristics and the construction process of the unified physical state representation are provided through specific implementation methods.

[0073] The collected environmental multimodal data undergoes multi-dimensional data preprocessing. For example, during data integration, standardized interface protocols are used to support multiple data formats such as JSON (JavaScript Object Notation) and XML (Extensible Markup Language), thereby ensuring compatibility with mainstream data center environmental management systems.

[0074] For time series data, a sliding window technique is used to extract key features, such as the rate of temperature change and the amplitude of current fluctuations, to capture the dynamic characteristics of fault development. Simultaneously, a data quality inspection mechanism is constructed, for example, to automatically identify and handle outliers and missing values, ensuring the reliability of the input data.

[0075] The construction of the physical state space ensures that the extracted features conform to the rules of physical dimensions and conservation laws, which is crucial for achieving physically consistent characterization. Dimensional analysis techniques are employed to verify the unit consistency between heat load and cooling capacity, ensuring the relationship Q=mcΔT holds true, where Q represents heat, m represents mass, c represents the specific heat capacity of the substance, and ΔT represents the temperature change.

[0076] Construct a physical state vector unique to the data center, with dimensions covering key physical relationships such as temperature field distribution, power load balance, and airflow organization.

[0077] Physical constraint regularization technology is introduced to incorporate physical conservation law constraints into the extraction of dynamic cycle time series features, ensuring that the state representation conforms to basic physical laws.

[0078] For special scenarios where some data centers have high-density server rooms, this invention designs a local hotspot risk index to quantitatively assess the risk of local temperature exceeding the standard.

[0079] In the process of extracting historical failure modes (i.e., historical failure information), typical failure modes and handling experience are extracted from historical data to provide a reference for decision-making.

[0080] In the process of processing historical fault information, cluster analysis technology is used to divide the historical fault information (or historical fault data) into several typical patterns, such as "cooling system capacity decline" and "power fluctuations cause equipment restart".

[0081] Construct a fault handling case library to record typical handling solutions and their effectiveness evaluations for each fault mode. Design a similarity matching algorithm to automatically match historical similar cases based on current fault characteristics, providing a reference for decision-making. Implement a case learning mechanism to automatically incorporate newly handled fault cases into the knowledge base, enabling the continuous evolution of the data center environmental management system.

[0082] According to an embodiment of the present invention, the above-mentioned physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain physical constraint results includes: performing a dynamic transformation operation on the unified physical state representation to obtain initial natural language prompt information and initial structured constraint information; performing a constraint injection operation on the initial natural language prompt information and initial structured constraint information using a preset large model to obtain natural language prompt information and structured constraint information; embedding the physical rules contained in the unified physical state representation as an inherent constraint mechanism into the decision generation operation based on the large language model using the natural language prompt information and structured constraint information; and, during the execution of the dynamic transformation operation and the constraint injection operation, performing a scene-adaptive rule strength adjustment operation on the physical rules based on a fault classification system using a dynamic strength adjustment algorithm to obtain physical constraint results.

[0083] The above embodiments achieve a dual guarantee of interpretable expression and machine-readable description of physical rules by dynamically transforming the unified physical state representation into natural language prompts and structured constraint information; in addition, the use of a pre-set large model for constraint injection ensures the semantic accuracy and logical consistency of physical rules.

[0084] Based on the fault classification system, the above embodiments adopt a dynamic intensity adjustment algorithm, which can automatically adjust the constraint intensity of the rules according to different fault scenarios, enhance the constraint force of the rules in critical fault scenarios, maintain appropriate flexibility in normal scenarios, and achieve precise constraint control.

[0085] The above embodiments directly embed physical rules as an intrinsic constraint mechanism into the decision generation process, ensuring that the decision schemes generated by the large language model always conform to physical laws, and avoiding the decision deviation problem caused by the separation of physical rules and AI decisions in traditional data center dynamic and environmental decision generation methods.

[0086] According to embodiments of the present invention, the above-mentioned dynamic transformation operation on the unified physical state representation to obtain initial natural language prompt information and initial structured constraint information includes: performing a dynamic transformation operation on the unified physical state representation using rule template technology to obtain original natural language prompt information and original structured constraint information; performing a dynamic adjustment operation on the original natural language prompt information and original structured constraint information using a scene-aware rule generation mechanism based on the current fault scenario information of the data center to obtain intermediate natural language prompt information and intermediate structured constraint information; during the execution of the dynamic transformation operation and the dynamic adjustment operation, the automatic identification and processing of rule conflicts is achieved by introducing a rule conflict detection mechanism; and performing multi-granularity expression processing on the intermediate natural language prompt information and intermediate structured constraint information to obtain initial natural language prompt information and initial structured constraint information.

[0087] The above embodiments, through a scene-aware rule generation mechanism, can dynamically adjust natural language prompts and structured constraint information according to the current fault scenario of the data center, enabling the decision support system to accurately respond to the needs of different fault types and severity. This flexibility ensures a high degree of matching between the rule template and the real-time scenario.

[0088] The rule conflict detection mechanism introduced in the above embodiments can automatically identify and handle contradictions between multiple rules. For example, when temperature alarms and load alarms are triggered at the same time, the system can prioritize handling the more urgent conflict to avoid confusion in decision-making logic, which significantly improves the reliability of decision-making in complex environments.

[0089] The multi-granularity expression optimization mechanism used in the above embodiments can meet the data needs of different decision-making levels.

[0090] According to an embodiment of the present invention, the process of embedding the physical rules implied by the unified physical state representation as an intrinsic constraint mechanism into the decision generation operation based on a large language model by utilizing natural language prompts and structured constraint information includes: adding natural language prompts and structured constraint information before the input data using prefix injection technology to obtain input data with physical constraint prompts; using the input data with physical constraint prompts to guide the decision generation operation based on the large language model to conform to physical laws; performing a filtering operation based on a constraint decoding mechanism on tokens that violate physical rules in the token generation stage of the decision generation operation based on the large language model, and dynamically adjusting the token generation probability distribution based on a physical consistency sampling strategy; and dynamically comparing and correcting the result of the decision generation operation based on the large language model with the physical rules based on the introduction of a rule-decision closed-loop mechanism.

[0091] The above embodiments incorporate natural language prompts and structured constraint information into the input data through prefix injection technology, so that the large language model always follows physical laws during the decision generation stage and avoids generating solutions that violate the dynamic and environmental operation rules of the data center.

[0092] The above embodiments filter out outputs that violate physical rules in real time during the token generation phase, and adjust the probability distribution by combining a physical consistency sampling strategy to ensure that the generated decision scheme (such as a cooling system scheduling strategy) meets both business requirements and equipment safety thresholds. Furthermore, a rule-decision closed-loop mechanism dynamically compares the generated results with physical rules to achieve real-time correction of the decision scheme.

[0093] The physical rule embedding operation provided by the present invention will be further described in detail below through specific embodiments.

[0094] The thermodynamic equations, power balance constraints, heating, cooling and air conditioning standards and other rules specific to data centers are dynamically transformed into natural language prompts and structured constraints that the large model can understand, based on the characteristics of the detected faults.

[0095] This system employs rule template technology to extract calculation methods, rules (e.g., Q=m+c) or decision rules (e.g., if m>0.5 then c=1) from a physical knowledge base. The extracted methods, rules, or decision rules are then used to calculate a unified physical state representation (e.g., m=0.1; c=0.2). The calculation results are then converted into scenario-adapted natural language expressions (i.e., natural language prompts), such as "Cooling capacity must meet: at least 0.86kW cooling capacity is required for every 1kW IT monitoring system load." A scenario-aware rule generation mechanism is designed (e.g., using a knowledge extraction algorithm based on a graph neural network (GNN). Inputting "scenario information," the algorithm extracts scenario-related calculation or decision rules from the physical knowledge base). The system dynamically adjusts the detail and constraint strength of the rules based on the current fault scenario, using strict expressions for emergency fault scenarios and guiding expressions for early warning states. This achieves multi-granularity rule expression, providing both natural language prompts and precise mathematical constraints (a type of structured constraint information) to meet the needs of different decision-making environments. A rule conflict detection mechanism is introduced to automatically identify and resolve potential conflicts between rules, ensuring the consistency of the rule set.

[0096] Constraint Injection Mechanism: Physical rules are embedded as an intrinsic constraint mechanism into the decision-making process of the large-scale decision model, rather than merely serving as a post-validation tool. Prefix injection technology is employed to add physical constraint hints before input to the large-scale decision model, guiding it to generate solutions that conform to physical laws. Constraint decoding technology is implemented to filter options that violate physical rules during token generation, ensuring the physical feasibility of the generated solutions. A physical consistency sampling strategy is designed to adjust the token generation probability distribution, making the model more inclined to generate solutions that conform to physical laws. A rule-decision closed-loop mechanism is introduced to compare the solutions generated by the large-scale model with the physical rules in real time, promptly correcting decisions that deviate from physical laws. For example, the constraint injection mechanism is used to optimize rule information expression: the generated physical constraints (i.e., calculation results) are transcribed into formatted text according to the prompt word injection template of the decision-level large-scale model, rather than merely serving as a post-validation tool. During the rule information expression optimization process, a built-in large language model is used. By inputting rule templates and using rule templateization technology to calculate physical constraints, the system guides the generation of physical constraint information that conforms to the needs of the decision-generation layer large-scale model, ensuring the feasibility of the generated physical constraints.

[0097] Rule strength adjustment: The constraint strength of physical rules is automatically adjusted based on the severity of the detected fault, enabling scenario-adaptive decision-making strategies. A three-level fault classification system is designed: Level 1 fault (urgent), Level 2 fault (early warning), and Level 3 fault (potential), each corresponding to a different constraint strength. Simultaneously, a dynamic strength adjustment algorithm is employed to adjust the constraint strength in real time according to the fault development trend, ensuring that the decision-making strategy is synchronized with the fault evolution.

[0098] Among them, the first-level fault (emergency) adopts hard constraints: strictly filtering out schemes that violate physical rules to ensure safety as a priority.

[0099] Among them, the second-level fault (early warning) adopts soft constraints: the scheme that violates the physical rules is subject to weighted penalties, allowing a certain degree of flexibility.

[0100] Among them, Level 3 faults (potential) adopt suggestive constraints: providing low-weight suggestions, focusing on the innovativeness and optimization potential of the solution.

[0101] Rule strength adjustment can be achieved through a specific mathematical mapping model: Based on a pre-defined fault classification system (Level 1 - Emergency, Level 2 - Warning, Level 3 - Potential), a basic constraint strength coefficient k_base is assigned to each level (e.g., 1.0 for Level 1, 0.6 for Level 2, and 0.3 for Level 3). Simultaneously, a real-time situation factor α is introduced, determined by the rate of deterioration of the fault indicator (e.g., the rate of temperature increase per minute). The final rule strength K_final = k_base * (1 + α). In LLM decision generation, this strength coefficient is used to adjust the penalty weight for violating physical rules; for example, a term λ * K_final * Violation_Score is added to the loss function, where Violation_Score is the quantified value of the degree to which the scheme violates physical rules.

[0102] According to an embodiment of the present invention, the above-mentioned simulation and deduction operation based on the world simulation engine is performed on the unified physical state representation to obtain the simulation and deduction results, including: performing multi-physics field coupling simulation operation on the unified physical state representation to obtain a multimodal physical simulator of the world simulation engine; performing simulation and fault propagation risk assessment operation on the fault propagation of the data center based on the fault propagation rules, using the critical path analysis algorithm and the fault propagation graph model of the world simulation engine; and performing time-based hybrid simulation and deduction operation on the fault of the data center using the multimodal physical simulator and the large language model to obtain the simulation and deduction results.

[0103] The above embodiments, by inputting a unified physical state representation into the world simulation engine, can construct a multimodal physical simulator to achieve coupled simulation of multiple physical fields such as mechanics, thermodynamics, and electromagnetics in data centers. This capability enables the system to accurately simulate complex physical processes such as equipment stress, heat transfer, and electromagnetic interference, providing simulation basis that conforms to physical laws for decision-making. Simultaneously, based on fault propagation graph models and critical path analysis algorithms, the system can simulate and assess data center faults. Furthermore, by combining the multimodal physical simulator with a large language model, it supports time-based hybrid simulation and deduction.

[0104] According to an embodiment of the present invention, the above-mentioned multi-physics field coupling simulation operation performed on the unified physical state representation to obtain a multi-modal physical simulator of the world simulation engine includes: performing multi-physics field coupling simulation operation on the unified physical state representation using the order reduction model technique to obtain a multi-modal physical simulator, wherein the multi-modal physical simulator includes a thermal field model based on computational fluid dynamics and an electric field model based on a directed graph.

[0105] According to an embodiment of the present invention, the above-described hybrid simulation and deduction operation based on the time dimension of a data center fault, performed using a multimodal physical simulator and a large language model, to obtain simulation and deduction results includes: performing a detection operation on key points of the data center using the key point prediction model of the world simulation engine to obtain key point prediction results; constructing prompt word instructions using the key point prediction operation; using the prompt word instructions to drive the large language model to perform a simulation and deduction operation based on the time dimension to obtain the large language model simulation and deduction results; performing a simulation and deduction operation based on the time dimension of the physical state of the data center using a multimodal physical simulator to obtain physical simulation and deduction results; and performing a weighted fusion processing of the large language model simulation results and the physical simulation results to obtain the simulation and deduction results.

[0106] The hybrid simulation provided in the above embodiments can capture both the physical characteristics of equipment operation (e.g., temperature and voltage fluctuations) and simulate the complex logical chain of fault propagation, improving the comprehensiveness of decision-making. Simultaneously, utilizing the key point prediction model of the world simulation engine, core risk nodes in the data center can be dynamically identified; the time-dimensional hybrid simulation can simulate the short-term outbreaks (e.g., voltage drops) and long-term evolutions (e.g., equipment aging) of faults, providing a time-series basis for preventative maintenance decisions. Furthermore, by weighting and fusing the logical deduction results of the large language model with the precise data from the physical simulator, the generalization ability of AI is preserved while ensuring the constraint of physical rules.

[0107] The following describes specific implementation methods in conjunction with appendices. Figure 4 The construction and simulation process of the world simulation engine provided by this invention will be explained in further detail.

[0108] Figure 4 A schematic diagram illustrating the data processing flow of a world simulation engine according to an embodiment of the present invention is shown.

[0109] The world simulation engine achieves accurate verification and forward-looking analysis of decision-making schemes through deep integration with large-scale models. This integration employs a hybrid simulation architecture combining physical and large-scale models, achieving complementary advantages. For example, the physical model is prioritized to ensure accuracy under normal conditions, while the large-scale model is invoked to enhance predictive capabilities in complex nonlinear regions. Through a two-way knowledge flow mechanism, the physical model provides physical constraints to the large-scale model, while the large-scale model provides insights into complex relationships to the physical model, improving simulation accuracy and applicability.

[0110] like Figure 4As shown, lightweight coupled simulations of thermodynamics, electricity, and other physical fields are achieved in the multiphysics coupled simulation process to verify the physical feasibility of the decision-making scheme. A lightweight physical simulator is constructed using a reduced-order model technique to meet real-time decision-making requirements. Specifically, the thermal field model is based on a simplified CFD (Computational Fluid Dynamics) model using the thermal resistance network method, retaining key temperature nodes and airflow paths. The calculation formula is: R... th =1 / (hA)+L / (kA), R th This is the total equivalent thermal resistance, where h represents the convective heat transfer coefficient, A represents the area, L represents the material thickness, and k represents the thermal conductivity. The power model is based on graph theory-based distribution network analysis, modeling the power distribution system as a directed graph and using Kirchhoff's laws to calculate load distribution.

[0111] like Figure 4 As shown, during the fault propagation simulation, the propagation path and impact range of faults in the data center system are predicted, providing forward-looking analysis for decision-making. First, a fault propagation graph model is constructed, where nodes represent system components and edges represent fault propagation paths. A probabilistic graphical model is used to calculate the propagation probability. Second, a data center-specific propagation rule base is designed, such as "cooling system failure → temperature rise → IT equipment frequency reduction / shutdown" and "power fluctuation → UPS (Uninterruptible Power Supply) switching → cooling capacity reduction".

[0112] Next, a critical path analysis algorithm is implemented to identify key nodes and paths in fault development and solution implementation, guiding decision-making priorities. Finally, a propagation risk assessment mechanism is introduced to quantitatively assess the continuous impact of fault propagation on business operations.

[0113] like Figure 4 As shown, in the time-based simulation process, the fault evolves over time: First, key point prediction (e.g., key point prediction based on the Transformer framework) identifies critical turning points in the system state, guiding the simulation focus and avoiding wasting computational resources in stable regions. Second, large model prediction prompts are built based on the key point prediction model to drive the large model to make predictions. Finally, a hybrid simulation strategy is employed. For example, deterministic simulation based on physical equations is used for normal states, while hybrid simulation combining large model predictions is used for complex states. The simulation results from both are then weighted and fused, with the weights set empirically.

[0114] The following describes specific implementation methods in conjunction with appendices. Figure 5 The decision evaluation and optimization process provided by this invention will be described in further detail.

[0115] Figure 5 A flowchart illustrating decision generation and evaluation according to an embodiment of the present invention is shown schematically.

[0116] As the application layer of the architecture, the decision generation and evaluation layer provides complete support for environmental fault decision-making. This layer achieves a precise transformation from "known faults" to "optimal decisions," providing maintenance personnel with scientific, objective, and actionable fault handling decisions.

[0117] like Figure 5 As shown, in the large model-driven solution generation process: based on the large model with embedded physical rules, multiple feasible solutions are generated for the same known environmental fault. First, a physical rule embedded prompting project is designed to deeply integrate physical rules into the large model input, guiding the model to generate solutions that conform to physical laws. Second, a solution diversity guarantee mechanism is implemented, ensuring that the generated solutions have significant differences by evaluating the feature distance and operational path differences between solutions. Finally, a large model-assisted solution correction technique is used to check the physical rule compliance of the initially generated solutions and intelligently correct solutions that violate physical rules.

[0118] like Figure 5 As shown, the multi-dimensional evaluation process achieves a comprehensive assessment of decision-making solutions from different dimensions: quantifying the solutions from five dimensions—business continuity, energy efficiency, physical feasibility, operational complexity, and compliance. A dynamic weight adjustment mechanism is also designed to automatically adjust the weights of each dimension based on data center operation strategies and fault characteristics, ensuring that the evaluation results meet actual needs.

[0119] The business continuity dimension assesses the impact of decision-making on IT operations, including metrics such as expected downtime and the percentage of IT load affected.

[0120] Among them, the energy efficiency dimension assesses the impact of decision-making schemes on energy efficiency and calculates changes in energy consumption and energy efficiency losses.

[0121] Among them, the physical feasibility dimension assesses whether the decision-making scheme conforms to physical laws, including key indicators such as temperature field distribution and power load balance.

[0122] The operational complexity dimension assesses the ease of implementing the decision-making plan, including the required manual intervention steps, spare parts needs, and operational skill requirements.

[0123] Among them, the compliance dimension: assess whether the decision-making plan complies with industry standards and norms.

[0124] like Figure 5As shown, in the process of optimizing decision-making schemes based on physical verification: candidate schemes are optimized and screened by combining the verification results of the world simulation engine. A closed-loop mechanism for decision-making scheme correction is implemented, and the physical verification results are fed back to guide the optimization of decision-making schemes. A decision-making scheme selection algorithm is designed, which comprehensively considers multi-dimensional evaluation results and physical verification data to calculate the comprehensive score of each decision-making scheme. Detailed decision-making basis is provided, including key indicator comparison, risk point analysis, and expected effect prediction.

[0125] The advantages and effectiveness of the data center environmental decision-making method provided by this invention will be illustrated and verified through specific experiments below.

[0126] The data center environmental decision-making method provided by this invention focuses on decision optimization for known faults. Through operations such as known fault input, multi-decision scheme generation, physical feasibility verification, multi-dimensional evaluation, and optimal scheme selection, it realizes a complete closed loop from "known fault" to "optimal decision". It solves the problem that traditional data center environmental monitoring systems can only provide a single solution or lack a systematic evaluation mechanism. It provides scientific, objective, and executable decision support for data center environmental management, and significantly improves the efficiency and quality of data center fault handling.

[0127] During the fault input phase, ensure the accuracy and completeness of the received fault information to provide a reliable data foundation for subsequent decision-making: receive confirmed fault information from external systems, including fault type, severity, and scope of impact; verify the completeness and accuracy of the fault information to ensure the reliability of the input data; convert the fault information into a standardized format for easy system processing; and identify key characteristics of the fault, such as the fault source, affected systems, and development trends.

[0128] In the multi-decision generation stage, it breaks through the limitation of traditional data center environmental monitoring systems that can only provide a single solution, and provides diverse choices for decision-making: based on fault status and physical rules, it generates 3-5 feasible decision-making solutions; ensures that each solution has significant differences and solves the problem from different perspectives; adopts large model-driven solution generation technology to ensure that the solutions conform to physical laws; implements a diversity guarantee mechanism to avoid solutions being too similar; and generates detailed solution descriptions, including operation steps, expected effects, and potential risks.

[0129] During the physical feasibility verification phase, ensure that all candidate decision-making schemes conform to physical laws and avoid proposing suggestions that are not feasible for engineering purposes: verify the physical feasibility of each scheme through a world simulation engine; eliminate infeasible decision-making schemes that violate physical laws; identify key risk points in the implementation process of decision-making schemes; predict the evolution of the system state after the implementation of decision-making schemes; and generate detailed verification reports, including key indicator change curves and risk warnings.

[0130] During the multi-dimensional evaluation phase, objective and quantitative evaluation criteria are provided to avoid over-reliance on personal experience in decision-making: five-dimensional quantitative scoring is performed on the retained decision-making options; the weights of each dimension are adjusted according to the data center operation strategy; a decision-making option comparison matrix is ​​generated to intuitively display the advantages and disadvantages of each option; the key advantages and potential risks of each decision-making option are identified; and the comprehensive score of each decision-making option is calculated to provide a basis for selecting the optimal decision-making option.

[0131] During the optimal decision selection phase, actionable decision results are provided: the comprehensive score of each decision option is calculated, and the decision option with the highest score is selected as the recommended decision option; detailed decision basis is generated, explaining the reasons for selecting the decision option; detailed operation guidelines are provided, including key operation steps and precautions; risk warnings and backup decision options are generated to ensure the robustness of the decision; the decision process and results are recorded for continuous optimization of the large model (large language model).

[0132] Based on the above-mentioned data center environmental decision-making method, this invention also provides a data center environmental decision-making system. The following will combine... Figure 6 The system is described in detail.

[0133] Figure 6 A schematic diagram of a data center environmental decision system according to an embodiment of the present invention is shown.

[0134] like Figure 6 As shown, the data center environmental decision system 600 includes a data acquisition module 610, a data representation and fusion module 620, a rule embedding and engine simulation module 630, a decision generation and evaluation module 640, and a user interaction module 650.

[0135] The data acquisition module 610 is used to acquire dynamic and environmental multimodal data from the data center in real time. In one embodiment, the data acquisition module 610 can be used to perform the operation S210 described above, which will not be repeated here.

[0136] The data representation and fusion module 620 is used to extract time-series features from the dynamic environment multimodal data to obtain dynamic environment time-series features, and to construct a unified physical state representation using the clustering analysis results of historical fault information and the dynamic environment time-series features. In one embodiment, the data representation and fusion module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0137] The rule embedding and engine simulation module 630 is used to perform a physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain physical constraint results, and to perform a simulation deduction operation based on a world simulation engine on the unified physical state representation to obtain simulation deduction results. In one embodiment, the rule embedding and engine simulation module 630 can be used to perform the operation S230 described above, which will not be repeated here.

[0138] The decision generation and evaluation module 640 is used to perform a decision generation operation based on a large language model using physical constraint results and simulation results to obtain an initial decision scheme, and then perform a multi-dimensional evaluation operation and a scheme optimization operation based on physical verification on the initial decision scheme to obtain a final decision scheme. In one embodiment, the decision generation and evaluation module 640 can be used to perform the operation S240 described above, which will not be repeated here.

[0139] User interaction module 650 is used to interactively display the decision-making scheme to the user. In one embodiment, user interaction module 650 can be used to perform the operation S250 described above, which will not be repeated here.

[0140] According to embodiments of the present invention, any plurality of modules among the data acquisition module 610, data representation and fusion module 620, rule embedding and engine simulation module 630, decision generation and evaluation module 640, and user interaction module 650 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of the present invention, at least one of the data acquisition module 610, data representation and fusion module 620, rule embedding and engine simulation module 630, decision generation and evaluation module 640, and user interaction module 650 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations or in a suitable combination of any of these. Alternatively, at least one of the data acquisition module 610, data representation and fusion module 620, rule embedding and engine simulation module 630, decision generation and evaluation module 640, and user interaction module 650 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0141] Figure 7A block diagram of an electronic device suitable for implementing a data center environmental decision-making method according to an embodiment of the present invention is shown schematically.

[0142] like Figure 7 As shown, an electronic device 700 according to an embodiment of the present invention includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0143] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0144] According to an embodiment of the present invention, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the input / output (I / O) interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0145] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0146] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 702 and / or RAM 703 and / or one or more memories other than ROM 702 and RAM 703 described above.

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0148] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0149] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A data center dynamic loop decision method, characterized in that, The method comprises: Real-time acquisition of dynamic environment multi-modal data of a data center; Performing time series feature extraction processing on the dynamic environment multi-modal data to obtain dynamic environment time series features, and constructing a unified physical state representation using clustering analysis results of historical fault information and the dynamic environment time series features; Performing physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain physical constraint results, and performing simulation deduction operation based on a world simulation engine on the unified physical state representation to obtain simulation deduction results; Performing decision generation operation based on a large language model using the physical constraint results and the simulation deduction results to obtain an initial decision scheme, and performing multi-dimensional evaluation operation and scheme optimization operation based on physical verification on the initial decision scheme to obtain a decision scheme; Interactively displaying the decision scheme to a user.

2. The method of claim 1, wherein, Performing time series feature extraction processing on the dynamic environment multi-modal data to obtain dynamic environment time series features comprises: Performing data standardization operation on the dynamic environment multi-modal data to obtain dynamic environment multi-modal standard data; Extracting time series features from the dynamic environment multi-modal standard data using sliding window technology and physical constraint regularization technology to obtain dynamic environment initial time series features, and performing data verification on the dynamic environment initial time series features to obtain the dynamic environment time series features.

3. The method of claim 2, wherein, Constructing a unified physical state representation using clustering analysis results of historical fault information and the dynamic environment time series features comprises: Performing physical dimension analysis on the dynamic environment time series features to obtain dynamic environment time series features conforming to physical dimension rules; Constructing physical state vectors and scene feature indicators of the data center, and fusing the physical state vectors and the scene feature indicators into the dynamic environment time series features conforming to physical dimension rules to obtain dynamic environment target time series features; Performing similarity matching operation and clustering analysis operation on the historical fault information to obtain clustering analysis results of the historical fault information; Constructing a unified physical state representation using the dynamic environment target time series features and the clustering analysis results of the historical fault information.

4. The method of claim 1, wherein, Performing physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain physical constraint results comprises: Performing dynamic conversion operation on the unified physical state representation to obtain initial natural language prompt information and initial structured constraint information; Performing constraint injection operation on the initial natural language prompt information and the initial structured constraint information using a preset large model to obtain natural language prompt information and structured constraint information; Embedding physical rules contained in the unified physical state representation as an internal constraint mechanism into the process of the decision generation operation based on a large language model using the natural language prompt information and the structured constraint information; During the execution of the dynamic conversion operation and the constraint injection operation, performing scene-adaptive rule strength adjustment operation based on a fault classification system on the physical rules using a dynamic strength adjustment algorithm to obtain the physical constraint results.

5. The method of claim 4, wherein, Performing dynamic conversion operation on the unified physical state representation to obtain initial natural language prompt information and initial structured constraint information comprises: Performing a dynamic conversion operation on the unified physical state representation by using a rule templating technique to obtain original natural language prompt information and original structured constraint information; Performing a dynamic adjustment operation on the original natural language prompt information and the original structured constraint information by using a scene-aware rule generation mechanism according to current fault scenario information of the data center to obtain intermediate natural language prompt information and intermediate structured constraint information; In the execution process of the dynamic conversion operation and the dynamic adjustment operation, a rule conflict detection mechanism is introduced to realize automatic identification and processing of rule conflicts; Performing multi-granularity expression processing on the intermediate natural language prompt information and the intermediate structured constraint information to obtain the initial natural language prompt information and the initial structured constraint information.

6. The method of claim 4, wherein, Embedding the physical rules contained in the unified physical state representation as an internal constraint mechanism into the process of the large language model-based decision generation operation by using the natural language prompt information and the structured constraint information includes: Adding the natural language prompt information and the structured constraint information in front of the input data by using a prefix injection technique to obtain input data with physical constraint prompt information; Guiding the large language model-based decision generation operation to conform to physical laws by using the input data with physical constraint prompt information; Performing a filtering operation based on a constraint decoding mechanism on tokens that violate the physical rules in the token generation stage of the large language model-based decision generation operation, and performing a dynamic adjustment on the token generation probability distribution based on a physical consistency sampling strategy; Performing a dynamic comparison and correction based on an introduced rule-decision closed loop mechanism on the result of the large language model-based decision generation operation and the physical rules.

7. The method of claim 1, wherein, Performing a simulation and deduction operation based on a world simulation engine on the unified physical state representation to obtain a simulation and deduction result includes: Performing a multi-physical field coupling simulation operation on the unified physical state representation to obtain a multi-modal physical simulator of the world simulation engine; Performing a simulation and deduction operation based on a world simulation engine on the unified physical state representation to obtain a simulation and deduction result includes: Performing a simulation and deduction operation based on a world simulation engine on the unified physical state representation to obtain a simulation and deduction result includes:

8. The method of claim 7, wherein, Performing a multi-physical field coupling simulation operation on the unified physical state representation to obtain a multi-modal physical simulator of the world simulation engine includes: Performing a multi-physical field coupling simulation operation on the unified physical state representation to obtain a multi-modal physical simulator of the world simulation engine includes:

9. The method of claim 7, wherein, Performing a multi-physical field coupling simulation operation on the unified physical state representation to obtain a multi-modal physical simulator of the world simulation engine includes: Performing a multi-physical field coupling simulation operation on the unified physical state representation to obtain a multi-modal physical simulator of the world simulation engine includes: Perform a detection operation on the key points of the data center by using a key point prediction model of the world simulation engine to obtain a key point prediction result; Construct a prompt word instruction by using the key point prediction operation; Drive the large language model to perform a time dimension-based simulation reasoning operation by using the prompt word instruction to obtain a large language model simulation reasoning result; Perform a time dimension-based simulation reasoning operation on the physical state of the data center by using the multi-modal physical simulator to obtain a physical simulation reasoning result; Weight and fuse the large language model simulation result and the physical simulation result to obtain the simulation reasoning result.

10. A data center dynamic loop decision system, comprising: The system comprises: A data acquisition module configured to acquire dynamic environment multi-modal data of a data center in real time; A data representation and fusion module configured to perform time series feature extraction processing on the dynamic environment multi-modal data to obtain dynamic environment time series features, and construct a unified physical state representation by using clustering analysis results of historical fault information and the dynamic environment time series features; A rule embedding and engine simulation module configured to perform a physical rule embedding operation based on a bidirectional fusion mechanism on the unified physical state representation to obtain a physical constraint result, and perform a simulation reasoning operation based on a world simulation engine on the unified physical state representation to obtain a simulation reasoning result; A decision generation and evaluation module configured to perform a decision generation operation based on a large language model on the physical constraint result and the simulation reasoning result to obtain an initial decision scheme, and perform a multi-dimensional evaluation operation and a scheme optimization operation based on physical verification on the initial decision scheme to obtain a decision scheme; A user interaction module configured to interactively display the decision scheme to a user.

Images