Dynamic environment control system based on large model driving intelligent body and operation method of dynamic environment control system

Abstract

The invention provides a dynamic environment control system based on a large model driving agent and an operation method thereof, and the system comprises a dynamic environment system which is used for collecting sensor data; the intelligent agent module comprises a plurality of intelligent agents and a relation network among the intelligent agents; for each agent, the agent is used for obtaining a target risk index according to the target sensor data and carrying out risk prediction according to the target risk index to obtain target index risk prediction data; the target sensor data belongs to at least one item in the sensor data; and the comprehensive risk analysis module is used for obtaining agent relation characteristics according to the relation network and the target index risk prediction data, and performing comprehensive processing of root cause positioning and comprehensive risk prediction according to the agent relation characteristics to obtain a comprehensive processing result so as to adjust the operation state of the dynamic environment system. According to the system, the power environment control efficiency of the data center is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent body control technology, and in particular to a power environment control system based on a large model driving intelligent body and an operation method thereof. Background Art

[0002] The power environment management of a data center involves power supply, temperature and humidity control, fire protection and security, and environmental monitoring. For example, the power environment monitoring system can ensure a stable supply of electricity by preventing power outages through UPS and backup generators, and maintain appropriate temperature and humidity through a precision air-conditioning system to protect the stable operation of equipment; the fire protection system and security monitoring are used to ensure the safety of the data center, and the environmental monitoring system tracks air quality and potential threats in real time, thereby ensuring the efficient and safe operation of the data center.

[0003] In the related technologies, the data center power environment monitoring system integrates multiple subsystems to work together. Affected by factors such as network quality, data security, interface standards between different systems, interoperability and management difficulty, the response time of each subsystem working together is long, resulting in low subsystem collaboration efficiency, thereby reducing the operational efficiency of the data center power environment. Summary of the invention

[0004] The present invention provides a power environment control system based on a large model-driven intelligent body and an operation method thereof, so as to solve the defect that when a plurality of subsystems are integrated in the prior art for collaborative work, the response time of each subsystem for collaborative work is long, resulting in low subsystem collaboration efficiency, thereby reducing the operational efficiency of the power environment of a data center, and improving the control efficiency of the power environment of a data center.

[0005] The present invention provides a power environment control system based on a large model driving intelligent agent, comprising: A power environment system, the power environment system is used to collect sensor data, the sensor data includes at least two of power data, environment data, fire data, security data and network equipment data; An agent module, the agent module includes a plurality of agents and a relationship network between the agents; for each agent, the agent is used to obtain a target risk indicator according to target sensor data, and perform risk prediction according to the target risk indicator to obtain target indicator risk prediction data; the target sensor data belongs to at least one item of the sensor data; A comprehensive risk analysis module is used to obtain the agent relationship characteristics based on the relationship network and the target indicator risk prediction data, and perform comprehensive processing of root cause location and comprehensive risk prediction based on the agent relationship characteristics to obtain a comprehensive processing result to adjust the operating state of the power environment system.

[0006] According to a large model driven agent-based power environment control system provided by the present invention, the multi-agent-based power environment control system further comprises: A pre-trained large language model is used to drive the intelligent agent module to construct the multiple intelligent agents and conduct autonomous analysis of each intelligent agent; the pre-trained large language model is also used to drive the comprehensive risk analysis module to locate the root cause and predict the comprehensive risk.

[0007] According to a large model driven agent-based power environment control system provided by the present invention, the multi-agent-based power environment control system further comprises: An agent attribute configuration unit, the agent attribute configuration unit is used to configure the basic attributes of each agent according to professional knowledge; An agent relationship rule configuration unit, wherein the agent relationship rule configuration unit is used to configure the dependency relationship between the agents according to professional knowledge.

[0008] According to a large model driven agent-based power environment control system provided by the present invention, the multi-agent-based power environment control system further comprises: A result output module is used to encapsulate the target indicator risk prediction data and the comprehensive processing results in a target format to obtain a packaged result; the target format includes at least one of XML format, JSON format, BSON format and HTML format.

[0009] According to a power environment control system based on a large model driven intelligent agent provided by the present invention, the intelligent agent module includes: An agent management module, the agent management module is used to obtain the relationship network according to the dependency relationship between the agents, and determine the first prompt language based on the basic attributes of the agents, so that the pre-trained large language model can generate the multiple agents according to the first prompt language; An intelligent agent autonomous analysis module, wherein the intelligent agent autonomous analysis module is used to perform abnormal state analysis on the target intelligent agent according to the target sensor data to obtain the target risk index, and to process the target risk index through an index risk prediction model to obtain the target risk prediction data; the index risk prediction model is determined based on a Transformer network.

[0010] According to a power environment control system based on a large model driving intelligent agent provided by the present invention, the comprehensive processing result includes a risk root cause location result and a comprehensive risk prediction result; The comprehensive risk analysis module includes: A multi-agent relationship network analysis model, wherein the multi-agent relationship network analysis model is used to obtain agent relationship characteristics according to the relationship network and the target indicator risk prediction data; the multi-agent relationship network analysis model is determined based on a graph neural network; A risk root cause location model, wherein the risk root cause location model is used to locate the risk root cause according to the agent relationship characteristics to obtain the risk root cause location result; wherein the risk root cause location model is determined based on an LSTM network and a graph convolutional network; A comprehensive risk prediction model, wherein the comprehensive risk prediction model is used to generate a second prompt based on the agent relationship characteristics, so that the pre-trained large language model can generate the comprehensive risk prediction result based on the second prompt; the comprehensive risk prediction model is determined based on a Transformer network.

[0011] The present invention also provides an operation method of a power environment control system based on a large model driving intelligent agent, comprising: The power environment system is used to collect sensor data; the sensor data includes at least two of power data, environment data, fire data, security data and network equipment data; Based on the intelligent agent in the intelligent agent module, a target risk indicator is obtained according to the target sensor data, and risk prediction is performed according to the target risk indicator to obtain target indicator risk prediction data; the target sensor data belongs to at least one item of the sensor data; the intelligent agent module includes a plurality of intelligent agents and a relationship network between the intelligent agents; Based on the comprehensive risk analysis module, the intelligent agent relationship characteristics are obtained according to the relationship network and the target indicator risk prediction data, and the root cause location and comprehensive risk prediction are comprehensively processed according to the intelligent agent relationship characteristics to obtain a comprehensive processing result to adjust the operating state of the power environment system.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements an operating method of a power environment control system based on a large model-driven intelligent agent as described in any one of the above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for operating a power environment control system based on a large model-driven intelligent agent as described in any of the above.

[0014] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for operating a power environment control system based on a large model-driven intelligent agent.

[0015] The power environment control system based on large model driven intelligent agent and its operation method provided by the present invention collects sensor data through the power environment system, obtains target risk indicators according to target sensor data through the intelligent agent in the intelligent agent module, and performs risk prediction based on the target risk indicators to obtain target indicator risk prediction data, and finally obtains the intelligent agent relationship characteristics according to the relationship network and the target indicator risk prediction data through the comprehensive risk analysis module, and performs comprehensive processing of root cause location and comprehensive risk prediction based on the intelligent agent relationship characteristics to adjust the operating state of the power environment system, thereby improving the power environment control efficiency of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is one of the structural schematic diagrams of the power environment control system based on the large model driving intelligent agent provided by the present invention.

[0018] Figure 2 This is the second structural schematic diagram of the power environment control system based on the large model driving intelligent agent provided by the present invention.

[0019] Figure 3 This is the third structural schematic diagram of the power environment control system based on the large model driving intelligent agent provided by the present invention.

[0020] Figure 4 It is a flow chart of the operation method of the power environment control system based on the large model driving intelligent body provided by the present invention.

[0021] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention.

[0022] Reference numerals: 110: power environment system; 120: intelligent agent module; 121: intelligent agent management module; 122: Agent autonomous analysis module; 130: Comprehensive risk analysis module; 131: Multi-agent relationship network analysis model; 132: Risk root cause location model; 133: Comprehensive risk prediction model; 140: Pre-trained large language model; 150: agent attribute configuration unit; 160: agent relationship rule configuration unit; 170: Result output module. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Combine the following Figure 1-Figure 4 The invention describes a dynamic environment control system based on a large model driven intelligent agent and its operation method.

[0025] Figure 1 This is one of the structural schematic diagrams of the power environment control system based on the large model driving intelligent agent provided by the present invention, such as Figure 1 As shown, the system includes a dynamic environment system 110 , an intelligent agent module 120 and a comprehensive risk analysis module 130 .

[0026] The power environment system 110 is used to collect sensor data, and the sensor data includes at least two of power data, environment data, fire data, security data and network equipment data.

[0027] In this embodiment, the power environment system 110 includes one or more of power system sensors, environment system sensors, fire protection system sensors, security system sensors and network equipment sensors, which are used to collect and correspond power data, environment data, fire protection data, security data and network equipment data respectively.

[0028] In this embodiment, the power environment system 110 transmits the data collected by the sensors to the intelligent agent for prediction based on the deployed relevant sensors; the data center can give priority to deploying power system sensors and environmental system sensors.

[0029] The intelligent agent module 120 includes multiple intelligent agents and a relationship network between the intelligent agents; for each intelligent agent, the intelligent agent is used to obtain a target risk indicator based on target sensor data, and perform risk prediction based on the target risk indicator to obtain target indicator risk prediction data; the target sensor data belongs to at least one item of the sensor data.

[0030] In this embodiment, a type of sensor data can be sent to one or more for data prediction or association prediction. For example, the power environment system includes multiple subsystems, each subsystem may include multiple different functional modules, different functional modules correspond to corresponding functional indicators, that is, a group of functional modules corresponds to a group of functional indicators; in this embodiment, one functional indicator corresponds to one intelligent agent, that is, there are multiple intelligent agents corresponding to a group of functional indicators in the power environment system. In this embodiment, the state information predicted by each of the two intelligent agents can be combined for comparative analysis, and the intelligent agents can be optimized according to their deviations.

[0031] Specifically, the intelligent agent module 120 includes an intelligent agent management module 121 and an intelligent agent autonomous analysis module 122 .

[0032] The agent management module 121 is used to obtain a relationship network based on the dependency relationship between the agents, and determine a first prompt based on the basic attributes of each agent, so that the pre-trained large language model 140 can generate multiple agents based on the first prompt.

[0033] In this embodiment, the basic attributes include the name, type and knowledge of the agent (the agent itself stores a basic set of knowledge so that an external script can allow it to act as an agent to store certain data. For example, target location, target radius, etc. This information is used to guide the actions and decisions of the agent).

[0034] In this embodiment, the agent management module 121 is used to construct and manage agents; specifically, the agent management module 121 includes an agent construction model and an agent relationship construction model.

[0035] In this embodiment, the agent construction model generates a large first prompt (Prompt) guided by the basic attributes of the agent; the prompt is injected into the pre-trained large language model 140, and the agent is generated by the large model.

[0036] In this embodiment, the intelligent agent construction model is obtained by training a deep learning model based on the Transformer architecture; for example, the training method of Prompt Learning is used, and the labeled data is used for training, combined with the pre-trained large model for training.

[0037] The specific training steps of the agent construction model include: (1) generating a first prompt by the agent construction model based on the training data; (2) injecting the prompt into the pre-trained large language model 140, and generating an agent by the large model; the user checks whether the agent is accurate enough to the expected agent to decide how to optimize the parameters of the agent construction model in the next round.

[0038] In this embodiment, the agent relationship construction model is used to construct the relationship between multiple agents through agent relationship rules.

[0039] The agent autonomous analysis module 122 is used to analyze the abnormal state of the target agent according to the target sensor data, obtain the target risk index, and process the target risk index through the index risk prediction model to obtain the target risk prediction data; the index risk prediction model is determined based on the Transformer network In this embodiment, the agent autonomous analysis module 122 performs independent analysis on each agent through the pre-trained large language model 140, without considering the dependencies with other agents; the agent autonomous analysis module 122 includes an indicator risk detection model and an indicator risk prediction model.

[0040] In this embodiment, the indicator risk detection model analyzes the time series data sent by the sensor through a time series algorithm (such as the Anomaly Transformer algorithm), finds abnormal patterns, and stores the found abnormal conditions in an abnormal time series database. In this embodiment, the indicator risk prediction model is obtained by training a Transformer-based deep learning model, and can generate prompts for the pre-trained large language model 140, thereby stimulating the large model to predict the risk of the indicator.

[0041] This embodiment uses the Prompt Learning training method to train the indicator risk prediction model; the specific training steps are: (1) obtaining training data from the abnormal time series database; (2) generating corresponding prompts through the initial indicator risk prediction model using the training data; (3) injecting the prompts into the pre-trained large language model 140, and the large model predicts the risk; (4) checking the degree of fit between the prediction results and the data labels to determine how to optimize the parameters of the indicator risk prediction model in the next round.

[0042] The comprehensive risk analysis module 130 is used to obtain the agent relationship characteristics based on the relationship network and target indicator risk prediction data, and perform comprehensive processing of root cause location and comprehensive risk prediction based on the agent relationship characteristics to obtain a comprehensive processing result to adjust the operating status of the power environment system 110.

[0043] In this embodiment, the operating status of the environmental system includes electrical parameters (such as voltage, current, power and power factor, etc.), environmental parameters (such as temperature and humidity, etc.), and device status parameters, etc.

[0044] In this embodiment, the comprehensive risk analysis module 130 performs a comprehensive risk analysis on the collaboration process of each intelligent agent through the indicator risk prediction data obtained by the autonomous analysis of each intelligent agent and the relationship between each intelligent agent, so as to obtain the root cause location and risk type of the collaboration of each intelligent agent.

[0045] In this embodiment, the comprehensive processing result includes a risk root cause location result and a comprehensive risk prediction result.

[0046] In this embodiment, the comprehensive risk analysis module 130 includes: a multi-agent relationship network analysis model 131, a risk root cause location model 132 and a comprehensive risk prediction model 133.

[0047] The multi-agent relationship network analysis model 131 is used to obtain agent relationship characteristics based on the relationship network and target indicator risk prediction data; the multi-agent relationship network analysis model 131 is determined based on a graph neural network.

[0048] In this embodiment, the graph neural network includes graph convolutional networks (GCN), graph auto-encoders (Graph Auto-encoders), graph generative networks (GGN) or graph spatial-temporal networks (GSTN).

[0049] In this embodiment, the multi-agent relationship network analysis model 131 extracts agent dependency features with fusion relationships, namely, agent relationship features, based on the relationship network between agents.

[0050] The risk root cause location model 132 is used to locate the risk root cause according to the agent relationship characteristics to obtain the risk root cause location result; wherein, the risk root cause location model 132 is determined based on the LSTM network and the graph convolution network.

[0051] In this embodiment, the risk root cause location model 132 is trained through the following steps: (1) marking root causes according to risk types in training data; (2) generating root causes using the initial risk root cause location model 132; and (3) determining how to optimize the parameters of the risk root cause location model 132 in the next round based on an evaluation of the results.

[0052] In this embodiment, the risk root cause location model 132 processes the agent relationship features to obtain the risk root cause location results.

[0053] The comprehensive risk prediction model 133 is used to generate a second prompt based on the agent relationship characteristics, so that the pre-trained large language model 140 can generate a comprehensive risk prediction result based on the second prompt; the comprehensive risk prediction model 133 is determined based on the Transformer network.

[0054] In this embodiment, the comprehensive risk prediction model 133 is trained by the training method of Prompt Learning in combination with the pre-trained large language model 140; specifically, the above-mentioned agent relationship features are input into the comprehensive risk prediction model 133 as initial values, which can generate prompts for the pre-trained large language model 140, thereby stimulating the large model to predict the comprehensive risk.

[0055] In this embodiment, the comprehensive risk prediction model 133 is trained by the following steps: (1) obtaining training data from an abnormal time series database; (2) inputting the training data into the comprehensive risk prediction model 133 to generate a second prompt; (3) injecting the second prompt into the pre-trained large language model 140, and the pre-trained large language model 140 predicts the comprehensive risk of the overall dynamic environment system; (4) checking the degree of fit between the prediction result and the data label to determine how to optimize the parameters of the comprehensive risk prediction model 133 in the next round.

[0056] In this embodiment, the comprehensive risk prediction model 133 outputs risk root cause location results and comprehensive risk prediction results, so as to adjust the operating parameters of the power system and environmental system of the data center, and further adjust the operating status of the power environment monitoring system.

[0057] The present invention provides a power environment control system based on a large model driving intelligent agent. The power environment system collects sensor data, obtains target risk indicators according to target sensor data through the intelligent agent in the intelligent agent module, and performs risk prediction based on the target risk indicators to obtain target indicator risk prediction data. Finally, the comprehensive risk analysis module obtains the intelligent agent relationship characteristics according to the relationship network and the target indicator risk prediction data, and performs comprehensive processing of root cause location and comprehensive risk prediction based on the intelligent agent relationship characteristics to adjust the operating state of the power environment system, thereby improving the power environment control efficiency of the data center.

[0058] In some embodiments, the multi-agent based dynamic environment control system further includes: a pre-trained large language model 140 .

[0059] The pre-trained large language model 140 is used to drive the agent module 120 to construct multiple agents and conduct autonomous analysis of each agent; the pre-trained large language model 140 is also used to drive the comprehensive risk analysis module 130 to locate the root cause and conduct comprehensive risk prediction.

[0060] In this embodiment, by injecting prompts into the pre-trained large language model 140 for driving, key functions such as agent creation, agent interaction, and agent monitoring and analysis are provided.

[0061] In this embodiment, the pre-trained large language model 140 includes but is not limited to common large language models such as ChatGPT (Chat Generative Pre-trained Transformer, built based on the GPT system large model) and GLM (General Language Model).

[0062] Specifically, the pre-trained large language model 140 is used as the core engine, and the initial knowledge base, behavior rules and decision logic of the agent are automatically generated by inputting specific task descriptions or role settings; this embodiment realizes natural language communication between agents by injecting specific interactive prompts into the pre-trained large language model 140. The model can understand and generate responses that conform to the context logic, promoting effective collaboration and information sharing between agents; finally, using the generalization ability of the pre-trained large language model 140, the agent can self-learn and optimize the decision-making strategy based on historical data and environmental feedback. Through the understanding and analysis of new situations, the agent can autonomously adjust the behavior mode to better adapt to environmental changes.

[0063] The present invention provides a power environment control system based on a large model driving an intelligent agent. By setting a pre-trained large language model to drive a multi-agent system to build monitoring and prediction of the power environment detection system, it can not only ensure the monitoring and risk analysis of a single perception indicator (i.e., an indicator monitored by a sensor, such as temperature, humidity, etc.), but also perform comprehensive risk analysis and prediction while considering subsystem dependencies.

[0064] In some embodiments, the multi-agent based dynamic environment control system further includes: an agent attribute configuration unit 150 and an agent relationship rule configuration unit 160 .

[0065] The agent attribute configuration unit 150 is used to configure the basic attributes of each agent according to professional knowledge.

[0066] In this embodiment, the agent attribute configuration unit 150 can customize the basic attributes of multiple agents, such as computing power, storage space, movement speed, perception range and communication capability, and classify the roles and task requirements of the agents through the above basic attributes for targeted configuration.

[0067] In this embodiment, the agent attribute configuration unit 150 uses the knowledge and experience of domain experts, as well as historical data and successful cases, to set reasonable attribute value ranges for different types of agents, thereby ensuring that each agent has the basic capabilities required to perform its tasks upon initialization.

[0068] In this embodiment, the agent attribute configuration unit 150 also has the ability to dynamically adjust attributes. When the agent encounters performance bottlenecks or environmental changes during task execution, the unit can automatically adjust the agent's attribute values ​​based on real-time monitored data to optimize its performance.

[0069] The agent relationship rule configuration unit 160 is used to configure the dependency relationship between agents based on professional knowledge.

[0070] In this embodiment, the agent relationship rule configuration unit 160 can customize various dependency relationships that may exist between multiple agents, such as collaborative relationships, competitive relationships, and leader-follower relationships, etc., so that the agents can interact and collaborate.

[0071] In this embodiment, the agent relationship rule configuration unit 160 configures reasonable rules for the dependency relationships between agents based on the knowledge and experience of domain experts and the overall design goals of the agent system, thereby ensuring that multiple agents can effectively collaborate in a complex environment and complete tasks together.

[0072] In this embodiment, the agent relationship rule configuration unit 160 also has a conflict detection capability. When there are potential conflicts or inconsistencies between the configured rules, the unit can promptly discover and provide solutions to ensure the stability and reliability of the agent system.

[0073] The present invention provides a power environment control system based on a large model driving intelligent agent. The basic attributes of each intelligent agent are configured according to professional knowledge through an intelligent agent attribute configuration unit, and the dependency relationship between each intelligent agent is configured according to professional knowledge through an intelligent agent relationship rule configuration unit, so as to provide data support for the subsequent generation of multiple intelligent agents and the acquisition of the relationship network of each intelligent agent, thereby improving the flexibility of the configuration of multiple intelligent agents and their relationships.

[0074] In some embodiments, the multi-agent based dynamic environment control system further includes: a result output module 170 .

[0075] The result output module 170 is used to encapsulate the target indicator risk prediction data and the comprehensive processing results in a target format to obtain a packaged result; the target format includes at least one of XML format, JSON format, BSON format and HTML format.

[0076] In this embodiment, the result output module 170 first receives the target indicator risk prediction data and the comprehensive processing results, and then maps the result data to the corresponding data structure according to the selected target format; for example, for the JSON format, the result data will be converted into a key-value pair form; for the XML format, it will be converted into tags and attributes with a hierarchical structure.

[0077] In this embodiment, after the result data is mapped to the corresponding data structure, the result output module 170 performs formatting processing on the data, including but not limited to data type verification, special character escape, data sorting and grouping, etc., to ensure that the output data complies with the specifications of the selected format.

[0078] In this embodiment, the formatted data will be output to a user-specified location or storage system through a standardized interface. The user can choose to save the output data to a local file, database, or send it to a remote server through a network as needed.

[0079] The present invention provides a power environment control system based on a large model driving intelligent agent, which uses a result output module to encapsulate target indicator risk prediction data and comprehensive processing results in a target format, thereby improving the data management and transmission capabilities of the system.

[0080] Figure 2 This is the second structural diagram of the power environment control system based on the large model driving intelligent agent provided by the present invention. Figure 2 In the illustrated embodiment, system sensors, environmental system sensors, fire protection system sensors, security system sensors and network equipment sensors are deployed on the power environment system (corresponding to the dynamic environment system); wherein, sensor data output by one or more sensors are input into different intelligent agents, and each intelligent agent is driven by a pre-trained large language model to read and process the data collected by the sensor, thereby realizing simulation of the sensor detection target. For example, the temperature intelligent agent continuously reads and monitors the temperature data, and conducts a comprehensive risk analysis based on the monitoring and analysis results to obtain risk information to adjust the operating parameters of the power environment system.

[0081] Figure 3 This is the third structural diagram of the power environment control system based on the large model driving intelligent agent provided by the present invention. Figure 3In the illustrated embodiment, the basic attributes of the agents are configured according to professional knowledge by the agent basic attribute configuration unit in the agent configuration module to guide the pre-trained large language model to drive the agent construction model in the agent management module to generate multiple agents; the dependency relationship between the agents is configured according to professional knowledge by the agent relationship rule configuration unit to guide the pre-trained large language model to drive the agent relationship construction model to generate a relationship network between the agents; the pre-trained large language model is used to drive the agent autonomous analysis module to obtain the target risk index according to the sensor data collected by the power environment system, and the risk prediction is performed according to the target risk index through the indicator risk prediction model to obtain the target indicator risk prediction data; the target indicator risk prediction data and the relationship network are used as the input of the multi-agent relationship network analysis model in the comprehensive risk analysis module, the agent relationship features are output, and the agent relationship features are input into the risk root cause location model to obtain the risk root cause location result, and the agent relationship features are input into the comprehensive risk prediction model to obtain the comprehensive risk prediction result; finally, the risk root cause location result and the comprehensive risk prediction result are formatted and packaged by the result output module to obtain the packaged result and output.

[0082] The following describes the operating method of the data center power environment control system based on multiple agents provided by the present invention. The operating method of the data center power environment control system based on multiple agents described below and the power environment control system based on large model driven agents described above can be referenced to each other.

[0083] Figure 4 It is a flow chart of the operation method of the power environment control system based on the large model driving intelligent agent provided by the present invention, such as Figure 4 As shown, the method comprises the following steps: Step 410: The power environment system is used to collect sensor data; the sensor data includes at least two of power data, environment data, fire data, security data and network equipment data.

[0084] In this step, the power environment system includes one or more of power system sensors, environment system sensors, fire protection system sensors, security system sensors and network equipment sensors, which are used to collect and correspond power data, environment data, fire protection data, security data and network equipment data respectively.

[0085] In this embodiment, the power environment system transmits the data collected by the sensors to the intelligent agent for prediction based on the deployed relevant sensors; the data center can give priority to deploying power system sensors and environmental system sensors.

[0086] The intelligent agent module includes multiple intelligent agents and the relationship network between each intelligent agent; for each intelligent agent, the intelligent agent is used to obtain the target risk indicator based on the target sensor data, and perform risk prediction based on the target risk indicator to obtain the target indicator risk prediction data; the target sensor data belongs to at least one item in the sensor data.

[0087] In this embodiment, a type of sensor data can be sent to one or more agents for data prediction or association prediction, and can also be combined with the predicted state information of multiple agents for comparative analysis, and the agents can be optimized according to the deviations between the agents.

[0088] Step 420, based on the intelligent agent in the intelligent agent module, the target risk indicator is obtained according to the target sensor data, and the risk prediction is performed according to the target risk indicator to obtain the target indicator risk prediction data; the target sensor data belongs to at least one item in the sensor data; the intelligent agent module includes multiple intelligent agents and a relationship network between the intelligent agents.

[0089] In this step, the intelligent agent module includes an intelligent agent management module and an intelligent agent autonomous analysis module.

[0090] In this embodiment, the agent management module is used to obtain a relationship network based on the dependency relationship between the agents, and determine the first prompt based on the basic attributes of each agent, so that the pre-trained large language model can generate multiple agents based on the first prompt.

[0091] In this embodiment, basic attributes include the name, type, and knowledge of the agent.

[0092] In this embodiment, the agent management module is used to construct and manage the agent; specifically, the agent management module includes an agent construction model and an agent relationship construction model.

[0093] In this embodiment, the agent construction model generates a large first prompt (Prompt prompt) guided by the basic attributes of the agent; the prompt is injected into the pre-trained large language model, and the agent is generated by the large model.

[0094] In this embodiment, the intelligent agent construction model is obtained by training a deep learning model based on the Transformer architecture; for example, the training method of Prompt Learning is used, labeled data is used for training, and a pre-trained large model is used for training; the specific training steps of the intelligent agent construction model are as described above and will not be repeated in this embodiment.

[0095] In this embodiment, the agent relationship construction model is used to construct the relationship between multiple agents through agent relationship rules.

[0096] The agent autonomous analysis module is used to analyze the abnormal state of the target agent according to the target sensor data, obtain the target risk index, and process it according to the target risk index through the indicator risk prediction model to obtain the target risk prediction data; the indicator risk prediction model is determined based on the Transformer network In this embodiment, the agent autonomous analysis module independently analyzes each agent through a pre-trained large language model without considering the dependency with other agents; the agent autonomous analysis module includes an indicator risk detection model and an indicator risk prediction model.

[0097] In this embodiment, the indicator risk detection model analyzes the time series data sent by the sensor through a time series algorithm (such as the Anomaly Transformer algorithm), finds abnormal patterns, and stores the found abnormal conditions in an abnormal time series database. In this embodiment, the indicator risk prediction model is obtained by training a Transformer-based deep learning model, which can generate prompts for the pre-trained large language model to stimulate the large model to predict the risk of the indicator; this embodiment uses the Prompt Learning training method to train the indicator risk prediction model; the specific training steps are as described above and will not be repeated in this embodiment.

[0098] Step 430, based on the comprehensive risk analysis module, the intelligent agent relationship characteristics are obtained according to the relationship network and the target indicator risk prediction data, and the root cause location and comprehensive risk prediction are comprehensively processed according to the intelligent agent relationship characteristics to obtain a comprehensive processing result to adjust the operating state of the power environment system.

[0099] In this step, the comprehensive risk analysis module is used to obtain the agent relationship characteristics based on the relationship network and target indicator risk prediction data, and perform comprehensive processing of root cause location and comprehensive risk prediction based on the agent relationship characteristics to obtain a comprehensive processing result to adjust the operating status of the power environment system.

[0100] In this embodiment, the operating status of the environmental system includes electrical parameters (such as voltage, current, power and power factor, etc.), environmental parameters (such as temperature and humidity, etc.), and device status parameters, etc.

[0101] In this embodiment, the comprehensive risk analysis module conducts a comprehensive risk analysis on the collaboration process of each intelligent agent through the indicator risk prediction data obtained by the autonomous analysis of each intelligent agent and the relationship between each intelligent agent, so as to obtain the root cause location and risk type of the collaboration of each intelligent agent.

[0102] In this embodiment, the comprehensive processing result includes a risk root cause location result and a comprehensive risk prediction result.

[0103] In this embodiment, the comprehensive risk analysis module includes: a multi-agent relationship network analysis model, a risk root cause location model and a comprehensive risk prediction model.

[0104] The multi-agent relationship network analysis model is used to obtain the agent relationship characteristics based on the relationship network and target indicator risk prediction data; the multi-agent relationship network analysis model is determined based on the graph neural network.

[0105] In this embodiment, the graph neural network includes a graph convolutional network, a graph autoencoder, a graph generation network, or a graph spatiotemporal network.

[0106] In this embodiment, the multi-agent relationship network analysis model extracts agent dependency features with fusion relationships based on the relationship network between agents, namely, agent relationship features.

[0107] The risk root cause location model is used to locate the risk root cause according to the agent relationship characteristics to obtain the risk root cause location result; wherein, the risk root cause location model is determined based on the LSTM network and the graph convolutional network; the training steps of the risk root cause location model are as described above, and will not be repeated in this embodiment.

[0108] In this embodiment, the risk root cause location model processes the agent relationship characteristics to obtain the risk root cause location results; specifically, the comprehensive risk prediction model is used to generate a second prompt based on the agent relationship characteristics, so that the pre-trained large language model can generate a comprehensive risk prediction result based on the second prompt; the comprehensive risk prediction model is determined based on the Transformer network.

[0109] In this embodiment, the comprehensive risk prediction model is trained by the training method of Prompt Learning in combination with the pre-trained large language model; specifically, the above-mentioned agent relationship features are input into the comprehensive risk prediction model as initial values, which can generate prompts for the pre-trained large language model, thereby stimulating the large model to predict the comprehensive risk; the training steps of the comprehensive risk prediction model are as described above and will not be repeated in this embodiment.

[0110] In this embodiment, the comprehensive risk prediction model outputs risk root cause location results and comprehensive risk prediction results, so as to adjust the operating parameters of the power system and environmental system of the data center, and further adjust the operating status of the power environment monitoring system.

[0111] The present invention provides a control method for a power environment control system based on a large model driving an intelligent agent. The power environment system collects sensor data, obtains a target risk indicator based on target sensor data through an intelligent agent in an intelligent agent module, and performs risk prediction based on the target risk indicator to obtain target indicator risk prediction data. Finally, a comprehensive risk analysis module obtains the intelligent agent relationship characteristics based on the relationship network and the target indicator risk prediction data, and performs comprehensive processing of root cause location and comprehensive risk prediction based on the intelligent agent relationship characteristics to adjust the operating state of the power environment system, thereby improving the power environment control efficiency of the data center.

[0112] Figure 5 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the operation method of the data center power environment control system of multiple agents, the method comprising: based on the power environment system for collecting sensor data; the sensor data includes at least two of power data, environment data, fire data, security data and network equipment data; based on the agent in the agent module, the target risk index is obtained according to the target sensor data, and the risk prediction is performed according to the target risk index to obtain the target index risk prediction data; the target sensor data belongs to at least one item in the sensor data; the agent module includes multiple agents and the relationship network between each agent; based on the comprehensive risk analysis module, the agent relationship characteristics are obtained according to the relationship network and the target indicator risk prediction data, and the root cause location and comprehensive risk prediction are comprehensively processed according to the agent relationship characteristics to obtain the comprehensive processing result, so as to adjust the operation state of the power environment system.

[0113] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0114] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the operation method of the multi-agent data center power environment control system provided by the above methods, the method including: based on the power environment system for collecting sensor data; the sensor data includes at least two of power data, environmental data, fire data, security data and network equipment data; based on the agent in the agent module, obtaining the target risk index according to the target sensor data, and performing risk prediction based on the target risk index to obtain target indicator risk prediction data; the target sensor data belongs to at least one item of the sensor data; the agent module includes multiple agents and the relationship network between each agent; based on the comprehensive risk analysis module, the agent relationship characteristics are obtained according to the relationship network and the target indicator risk prediction data, and the root cause location and comprehensive risk prediction are comprehensively processed according to the agent relationship characteristics to obtain a comprehensive processing result to adjust the operation state of the power environment system.

[0115] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the operation method of the multi-agent data center power environment control system provided by the above-mentioned methods, the method comprising: based on the power environment system, used to collect sensor data; the sensor data includes at least two of power data, environmental data, fire data, security data and network equipment data; based on the agent in the agent module, obtaining the target risk index according to the target sensor data, and performing risk prediction according to the target risk index to obtain target indicator risk prediction data; the target sensor data belongs to at least one item of the sensor data; the agent module includes multiple agents and a relationship network between each agent; based on the comprehensive risk analysis module, obtaining the agent relationship characteristics according to the relationship network and the target indicator risk prediction data, and performing comprehensive processing of root cause location and comprehensive risk prediction according to the agent relationship characteristics to obtain a comprehensive processing result to adjust the operation state of the power environment system.

[0116] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0117] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power environment control system based on a large model driving intelligent agent, characterized in that: include: A power environment system, the power environment system is used to collect sensor data, the sensor data includes at least two of power data, environment data, fire data, security data and network equipment data; An agent module, the agent module includes a plurality of agents and a relationship network between the agents; for each agent, the agent is used to obtain a target risk indicator according to target sensor data, and perform risk prediction according to the target risk indicator to obtain target indicator risk prediction data; the target sensor data belongs to at least one item of the sensor data; A comprehensive risk analysis module is used to obtain the agent relationship characteristics based on the relationship network and the target indicator risk prediction data, and perform comprehensive processing of root cause location and comprehensive risk prediction based on the agent relationship characteristics to obtain a comprehensive processing result to adjust the operating state of the power environment system.

2. The power environment control system based on large model driven intelligent agent according to claim 1 is characterized in that: The multi-agent-based power environment control system also includes: A pre-trained large language model is used to drive the intelligent agent module to construct the multiple intelligent agents and conduct autonomous analysis of each intelligent agent; the pre-trained large language model is also used to drive the comprehensive risk analysis module to locate the root cause and predict the comprehensive risk.

3. The power environment control system based on large model driven intelligent agent according to claim 1, characterized in that: The multi-agent-based power environment control system also includes: An agent attribute configuration unit, the agent attribute configuration unit is used to configure the basic attributes of each agent according to professional knowledge; An agent relationship rule configuration unit, wherein the agent relationship rule configuration unit is used to configure the dependency relationship between the agents according to professional knowledge.

4. The power environment control system based on large model driven intelligent agent according to claim 1, characterized in that: The multi-agent-based power environment control system also includes: A result output module is used to encapsulate the target indicator risk prediction data and the comprehensive processing results in a target format to obtain a packaged result; the target format includes at least one of XML format, JSON format, BSON format and HTML format.

5. The power environment control system based on large model driven intelligent agent according to claim 1, characterized in that: The intelligent agent module comprises: An agent management module, the agent management module is used to obtain the relationship network according to the dependency relationship between the agents, and determine the first prompt language based on the basic attributes of the agents, so that the pre-trained large language model can generate the multiple agents according to the first prompt language; An intelligent agent autonomous analysis module, wherein the intelligent agent autonomous analysis module is used to perform abnormal state analysis on the target intelligent agent according to the target sensor data to obtain the target risk index, and to process the target risk index through an index risk prediction model to obtain the target risk prediction data; the index risk prediction model is determined based on a Transformer network.

6. The power environment control system based on large model driven intelligent agent according to claim 1, characterized in that: The comprehensive processing results include risk root cause location results and comprehensive risk prediction results; The comprehensive risk analysis module includes: A multi-agent relationship network analysis model, wherein the multi-agent relationship network analysis model is used to obtain agent relationship characteristics according to the relationship network and the target indicator risk prediction data; the multi-agent relationship network analysis model is determined based on a graph neural network; A risk root cause location model, wherein the risk root cause location model is used to locate the risk root cause according to the agent relationship characteristics to obtain the risk root cause location result; wherein the risk root cause location model is determined based on an LSTM network and a graph convolutional network; A comprehensive risk prediction model, wherein the comprehensive risk prediction model is used to generate a second prompt based on the agent relationship characteristics, so that the pre-trained large language model can generate the comprehensive risk prediction result based on the second prompt; the comprehensive risk prediction model is determined based on a Transformer network.

7. A method for operating a power environment control system based on a large model driving an intelligent agent, characterized in that: include: Power-based environmental systems are used to collect sensor data; The sensor data includes at least two of power data, environmental data, fire data, security data and network equipment data; Based on the intelligent agent in the intelligent agent module, a target risk indicator is obtained according to the target sensor data, and risk prediction is performed according to the target risk indicator to obtain target indicator risk prediction data; the target sensor data belongs to at least one item of the sensor data; the intelligent agent module includes a plurality of intelligent agents and a relationship network between the intelligent agents; Based on the comprehensive risk analysis module, the intelligent agent relationship characteristics are obtained according to the relationship network and the target indicator risk prediction data, and the root cause location and comprehensive risk prediction are comprehensively processed according to the intelligent agent relationship characteristics to obtain a comprehensive processing result to adjust the operating state of the power environment system.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the operating method of the power environment control system based on the large model driven intelligent agent as described in claim 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the operating method of the power environment control system based on a large model-driven intelligent agent as described in claim 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the operating method of the power environment control system based on a large model-driven intelligent agent as described in claim 7 is implemented.

Images