College data center green energy-saving and safety linkage control method and system based on dynamic environment system

Abstract

The invention provides a green energy-saving and safety linkage control method and system for a college data center based on a dynamic environment system, and is applied to the technical field of data processing. According to the invention, a green energy-saving and safety linkage control system of a dynamic environment system is constructed for a college and university data center. Firstly, multi-dimensional data such as machine room temperature and humidity and server power consumption are collected and processed by an intelligent sensing fusion module to generate standardized feature vectors; generating a preliminary control parameter through an energy efficiency-safety dual-objective optimization module, and converting the preliminary control parameter into a final control instruction through a self-adaptive execution adjustment module; and a closed-loop feedback iteration module optimization strategy is combined, and finally a comprehensive evaluation report is generated through a visual decision support module, so that whole-process dynamic optimization and accurate management and control are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a green energy-saving and safety linkage control method and system for a university data center based on a dynamic environment system. BACKGROUND

[0002] In today's digital age, data centers and computer rooms are the core facilities for information storage and processing, and their stable operation is crucial. Inside the computer room, there are not only a large number of servers, storage devices and network communication equipment, but also complex power systems, cooling systems and fire fighting systems. The coordinated work of these systems ensures the efficient and safe operation of the computer room. However, the complexity of the computer room environment and the density of equipment also bring many challenges. Traditionally, data center energy saving mainly relies on two relatively independent management systems: IT infrastructure management system and power and environment monitoring system. The IT infrastructure management system is mainly responsible for the energy saving of servers and other IT equipment. It relies on out-of-band management technology to collect real-time energy consumption information of servers, such as power consumption, inlet / outlet temperature, air flow, etc. It analyzes the server running state and energy use, and then adjusts the fan speed, CPU frequency, etc. to achieve optimal server energy consumption. The power and environment monitoring system mainly focuses on the energy consumption optimization of the refrigeration system and the power supply system. It uses temperature and humidity sensors, current and voltage sensors, etc. to monitor the environment, power equipment, refrigeration equipment, etc. It collects real-time operation data, predicts equipment failure and environmental trends based on data analysis results, and adjusts air conditioning, power supply, cooling tower, etc. non-IT equipment to improve energy efficiency.

[0003] However, this traditional mode has significant drawbacks. Since the IT and dynamic environment systems operate independently, data cannot be shared. The dynamic environment system can only collect data at the cabinet level, and it is difficult to obtain more detailed data at the server node level and chip level, resulting in low matching of cooling capacity and server power consumption. The operation and maintenance of the refrigeration system and the power supply system are relatively extensive, and it is difficult to accurately control the temperature, which seriously affects the overall energy consumption optimization of the data center. At the same time, the traditional extensive operation and control method cannot achieve accurate temperature control, and unstable server temperature can significantly affect device performance and lifespan.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a green energy-saving and safety linkage control method and system for a university data center based on an environmental system, which at least partially overcomes the problems existing in the prior art. The method includes collecting data center environmental system multi-dimensional real-time state data, including room temperature and humidity, server power consumption, air conditioner operating parameters, and UPS state; processing the multi-dimensional real-time state data based on an intelligent perception fusion module, evaluating and fusing different source data through a dynamic weight distribution algorithm, using a time sequence feature extraction network to mine hidden correlation rules in the data, combining an abnormal value adaptive filtering mechanism to eliminate noise interference, and generating a standardized state feature vector; processing the standardized state feature vector based on an energy efficiency-safety dual target optimization module, constructing a mapping relationship between energy efficiency indicators and safety indicators through a multi-objective reinforcement learning network, designing dynamic constraint conditions to cooperatively optimize air conditioner refrigeration strategies and server cluster scheduling schemes, and generating preliminary control parameters; processing the preliminary control parameters based on an adaptive execution adjustment module and a device response characteristic database corresponding to the adaptive execution adjustment module, dynamically correcting the control parameters according to the real-time operating state of the device through a PID parameter self-tuning algorithm, combining a load prediction model to adjust the action timing of the execution mechanism in advance, and generating final control instructions that meet the operating characteristics of the device; processing the execution effect of the final control instructions based on a closed-loop feedback iteration module and a performance evaluation system corresponding to the closed-loop feedback iteration module, calculating the control deviation by monitoring the changes in energy efficiency indicators and safety indicators in real time, continuously optimizing the control model parameters using an incremental learning algorithm, and generating iteratively updated control strategies; and inputting the standardized state feature vector, the preliminary control parameters, the final control instructions, and the iteratively updated control strategies into a visual decision support module for processing, and generating a comprehensive evaluation report including energy efficiency optimization rate, safety risk early warning, and device operating state trend.

[0006] Other characteristics and advantages of the present application will become apparent from the following detailed description, or will be learned by practice of the application.

[0007] According to one aspect of the present application, a green energy-saving and safety linkage control method for a university data center based on an environmental system is provided, which includes: collecting data center environmental system multi-dimensional real-time state data, including room temperature and humidity, server power consumption, air conditioner operating parameters, and UPS state; processing the multi-dimensional real-time state data based on an intelligent perception fusion module, evaluating and fusing different source data through a dynamic weight distribution algorithm, using a time sequence feature extraction network to mine hidden correlation rules in the data, combining an abnormal value adaptive filtering mechanism to eliminate noise interference, and generating a standardized state feature vector; processing the standardized state feature vector based on an energy efficiency-safety dual target optimization module, constructing a mapping relationship between energy efficiency indicators and safety indicators through a multi-objective reinforcement learning network, designing dynamic constraint conditions to cooperatively optimize air conditioner refrigeration strategies and server cluster scheduling schemes, and generating preliminary control parameters; processing the preliminary control parameters based on an adaptive execution adjustment module and a device response characteristic database corresponding to the adaptive execution adjustment module, dynamically correcting the control parameters according to the real-time operating state of the device through a PID parameter self-tuning algorithm, combining a load prediction model to adjust the action timing of the execution mechanism in advance, and generating final control instructions that meet the operating characteristics of the device; processing the execution effect of the final control instructions based on a closed-loop feedback iteration module and a performance evaluation system corresponding to the closed-loop feedback iteration module, calculating the control deviation by monitoring the changes in energy efficiency indicators and safety indicators in real time, continuously optimizing the control model parameters using an incremental learning algorithm, and generating iteratively updated control strategies; and inputting the standardized state feature vector, the preliminary control parameters, the final control instructions, and the iteratively updated control strategies into a visual decision support module for processing, and generating a comprehensive evaluation report including energy efficiency optimization rate, safety risk early warning, and device operating state trend.

[0008] Another aspect of this application discloses a green energy-saving and safety linkage control device for a university data center based on an environmental dynamic system. The device comprises: a data acquisition module for acquiring multi-dimensional real-time status data of the data center's environmental dynamic system, including room temperature and humidity, server power consumption, air conditioning operating parameters, and UPS status; a processing module for processing the multi-dimensional real-time status data based on an intelligent sensing fusion module to generate a standardized status feature vector; processing the standardized status feature vector based on an energy efficiency-safety dual-objective optimization module to generate preliminary control parameters; processing the preliminary control parameters based on an adaptive execution adjustment module and a corresponding equipment response characteristic database to generate a final control command conforming to the equipment's operating characteristics; processing the execution effect of the final control command based on a closed-loop feedback iteration module and a corresponding performance evaluation system to generate an iteratively updated control strategy; and inputting the standardized status feature vector, preliminary control parameters, final control command, and iteratively updated control strategy into a visualization decision support module for processing to generate a comprehensive evaluation report including energy efficiency optimization rate, safety risk warning, and equipment operating status trend.

[0009] According to another aspect of this application, an electronic device is characterized by comprising: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described green energy-saving and safety linkage control method for a university data center based on an environmental system by executing the executable instructions.

[0010] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a second processor, implements the above-described green energy-saving and safety linkage control method for a university data center based on an environmental system.

[0011] This application provides a green energy-saving and safety-linked control method and system for university data centers based on a dynamic environmental system. The system collects multi-dimensional data such as room temperature and humidity, and server power consumption, generating standardized feature vectors through an intelligent sensing fusion module. It then uses an energy efficiency-safety dual-objective optimization module, combined with multi-objective reinforcement learning, to generate preliminary control parameters. These parameters are then converted into final control commands by an adaptive execution adjustment module, and the strategy is continuously optimized through a closed-loop feedback iteration module. Finally, a comprehensive evaluation report is generated through a visualization decision support module. The system encompasses data acquisition and processing modules, achieving dynamic optimization throughout the entire process, balancing energy saving and safety, adapting to the periodic load characteristics of university data centers, and improving the accuracy and efficiency of management and control.

[0012] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0013] Figure 1 This application provides a flowchart illustrating a green energy-saving and safety linkage control method for a university data center based on a dynamic environment system, according to an embodiment of this application. Figure 2 This paper presents a schematic diagram of the structure of a green energy-saving and safety linkage control device for a university data center based on a dynamic environmental system, according to an embodiment of this application. Detailed Implementation

[0014] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0015] The following is combined with Figure 1 This application describes a green energy-saving and safety linkage control method for a university data center based on an environmental system, according to exemplary embodiments of this application. It should be noted that the following application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way. Rather, the embodiments of this application can be applied to any applicable scenario.

[0016] In one embodiment, this application also proposes a green energy-saving and safety linkage control method and system for university data centers based on dynamic environmental systems. Figure 1 A schematic flowchart illustrating a green energy-saving and safety linkage control method for a university data center based on a dynamic environment system, according to an embodiment of this application, is shown. Figure 1 As shown, this method is applied to a server and includes: S101 collects multi-dimensional real-time status data of the data center's environmental system.

[0017] In one implementation, data is collected using temperature and humidity sensors deployed in different areas of the data center. For example, a high-precision temperature and humidity sensor is placed every 2 square meters in the core area with dense server racks, collecting data every minute. For instance, the sensor records that at 10:00 AM, the temperature in the core area is 23°C and the humidity is 45%, while in the edge area, the temperature is 25°C and the humidity is 50%. Data is obtained using the server's built-in power consumption monitoring interface or an external power metering device to monitor each server in real time, recording the power consumption value every 30 seconds. For example, a core server consumes 350W at 10:00:30 AM, while an edge server consumes 180W at the same time.

[0018] Relevant parameters are collected through the air conditioning control system interface, including the air conditioner's supply air temperature, return air temperature, fan speed, and compressor operating frequency, every 2 minutes. For example, at 10:02, the air conditioner's supply air temperature is 19℃, return air temperature is 24℃, and fan speed is 1200 r / min. Data is collected using the UPS's built-in monitoring system, covering information such as the UPS's input voltage, output voltage, load rate, and battery capacity, every 5 minutes. For example, at 10:05, the UPS input voltage is 380V, output voltage is 220V, load rate is 40%, and battery capacity is 95%.

[0019] S102 processes multi-dimensional real-time state data based on the intelligent perception fusion module, evaluates and fuses the credibility of data from different sources through a dynamic weight allocation algorithm, extracts hidden correlation patterns in network mining data using time-series features, removes noise interference by combining an outlier adaptive filtering mechanism, and generates a standardized state feature vector.

[0020] In one implementation, a smart sensing fusion module processes multi-dimensional real-time status data. A dynamic weighting algorithm performs hierarchical reliability assessments on data from different sources, assigning a first basic weight to high-precision sensors in core areas and a second basic weight to ordinary sensors in edge areas. Simultaneously, the weight deviation is dynamically adjusted based on historical equipment failure rates. Furthermore, spatiotemporal alignment and fusion of multi-source data are achieved through timestamp calibration and spatial coordinate mapping, eliminating information misalignment caused by time differences and location deviations in data acquisition from different devices. Based on environmental system correlation information, color gradient marking is applied to areas with weak energy efficiency and security performance in the data center's three-dimensional structural model, visually presenting the spatial distribution and warning levels of high-risk areas. For example, based on the correlation between energy efficiency indicators (such as PUE value) and safety indicators (such as equipment over-temperature probability) in the standardized state feature vector, a three-color gradient marking of "red-yellow-green" is used in the 3D model. The core server cluster area is marked in red because of its high PUE value (1.8) and over-temperature probability of 15%. The edge storage area is marked in yellow because of its medium PUE value (1.5) and over-temperature probability of 8%. The auxiliary equipment area of ​​the computer room is marked in green because of its low PUE value (1.2) and over-temperature probability of 2%. At the same time, the energy efficiency index range of each area (such as PUE 1.7-1.9 in the red area) and the safety warning threshold (such as over-temperature probability of 12%-18% in the red area) are marked. Combined with the real-time updated execution effect of the final control command (such as the temperature change after the air conditioning cooling strategy is adjusted), the marking color depth is dynamically refreshed (such as the red area turning orange-red after the air conditioning cools down to the safe range). This clearly shows the comprehensive risk distribution of different areas in terms of energy efficiency and safety, providing an intuitive reference for managers to locate and optimize key areas.

[0021] This study utilizes a temporal feature extraction network, employing a nested structure of bidirectional LSTM and an attention mechanism. The bidirectional LSTM layer captures short-term fluctuations in data over a 5-30 minute period, while the attention mechanism layer assigns 1.2-1.5 times weight to key time nodes such as peak server load periods and air conditioning start-stop switching points to enhance feature extraction. A deep fully connected network further uncovers long-term trend correlations within a 24-hour period. The nested structure of bidirectional LSTM and the attention mechanism allows the bidirectional LSTM layer to accurately capture short-term fluctuations of 5-30 minutes through the collaborative computation of forward and backward 64-dimensional neurons. For example, it identifies peak login times on the teaching platform driven by the course schedule during 10:00-10:30 AM on weekdays, causing server power consumption to steadily increase from 300W to 350W, accompanied by a slight increase in air conditioning return air temperature from 23℃ to 24℃. On weekends, this period shows no significant fluctuations, remaining around 310W.

[0022] The attention mechanism layer uses a dynamic weight allocation matrix to enhance the extraction of features at key time points: peak server load periods (such as the peak of experimental data processing from 14:00 to 16:00 on weekdays) and air conditioner start-stop switching points (such as the compressor frequency jump phase when the air conditioner starts at 8:00 every day) are given a weight of 1.3 times, and sudden load fluctuations (such as a sudden increase in power consumption from 16:30 to 17:00 caused by a temporary online test added to a course) are given a weight of 1.5 times, so as to retain the feature information of these high-impact events.

[0023] The deep fully connected network extracts long-term trends within a 24-hour period through layer-by-layer feature compression and mapping of three layers of neurons (128-dimensional → 64-dimensional → 32-dimensional). For example, it was found that from 0:00 to 6:00 every day, due to reduced teacher and student activities, server power consumption is generally below 150W, and the air conditioning supply temperature can be automatically adjusted to 22℃ to maintain the stability of the computer room; during non-teaching hours from 18:00 on Friday to 8:00 on Sunday, the overall power consumption is 20% lower than the same period on weekdays, forming a weekly periodic pattern, which provides temporal feature support for the subsequent energy efficiency-safety dual-objective optimization module.

[0024] Combining an adaptive outlier filtering mechanism, and through collaborative verification using the 3σ criterion and the Isolation Forest algorithm, outliers deviating more than three times the standard deviation from the mean are initially screened using the 3σ criterion. Then, the Isolation Forest model performs a secondary judgment on these outliers, eliminating transient noise caused by sudden electromagnetic interference and systematic deviations caused by equipment aging. Finally, a high-fidelity standardized state feature vector containing 288-dimensional temporal features and 16-dimensional static features is generated. When initially screening outliers using the 3σ criterion, differentiated calculation logic is set for different types of parameters. For data center temperature, the mean is calculated to be 24℃ and the standard deviation is 1℃ based on nearly 72 hours of stable operation data. Values ​​below 21℃ (e.g., 20℃ caused by a sudden cold draft in the early morning) or above 27℃ (e.g., 28℃ caused by a short-term air conditioner shutdown) are marked as outliers. For server power consumption, the mean of a core server is 350W and the standard deviation is 30W. Values ​​below 260W (e.g., 250W before a sudden crash) or above 440W (e.g., 450W caused by malicious program usage) are included in the outlier candidate set.

[0025] The isolated forest model then performs a second evaluation. This model uses nearly three months of normal operation data (including more than 20,000 valid samples) as the training set and constructs multiple isolated trees to perform in-depth verification of the abnormal candidate set: For instantaneous noise, such as electromagnetic interference causing the temperature to jump from 24°C to 30°C within 0.1 seconds and then immediately return to normal, the model identifies that it does not meet the characteristics of continuous fluctuation and judges it as noise that can be removed; For systematic deviations, such as an aging sensor in a certain edge area showing a temperature 2°C higher than other sensors in the same area for three consecutive days, and the deviation is stable without jumps, the model classifies it as a systematic error caused by equipment failure and marks it as removed.

[0026] The final standardized feature vector contains 288 time-series features covering the entire 24-hour cycle at 10-minute intervals. These features include 12 types of dynamic parameters (12×24=288) such as the average temperature of the server room, total power consumption of the server cluster, operating frequency of the air conditioning compressor, and UPS load rate for each time period. For example, the data recorded for the 10:00-10:10 time period is 23.5℃, total power consumption is 8500W, compressor frequency is 50Hz, and load rate is 35%. The 16-dimensional static features cover the inherent attributes of the equipment, including server model (e.g., blade / tower), air conditioning cooling capacity (e.g., 10P / 5P), sensor installation location coordinates (e.g., 3 cabinets in column A of the core area / 1 cabinet in column C of the edge area), and basic information such as the equipment's service life. This forms structured, high-fidelity input data, providing accurate feature support for subsequent energy efficiency and safety dual-objective optimization.

[0027] S103 processes the standardized state feature vector based on the energy efficiency-safety dual-objective optimization module, constructs the mapping relationship between energy efficiency indicators and safety indicators through a multi-objective reinforcement learning network, designs dynamic constraints to coordinate the optimization of air conditioning cooling strategy and server cluster scheduling scheme, and generates preliminary control parameters.

[0028] In one implementation, a standardized state feature vector is processed based on an energy efficiency-safety dual-objective optimization module. The data center operation process is modeled as an energy efficiency-safety dynamic coupling system, and a multi-objective reinforcement learning network is used to construct a nonlinear mapping relationship between energy efficiency indicators and safety indicators. The construction of the energy efficiency-safety dynamic coupling system and the nonlinear mapping relationship involves modeling the data center operation process as an energy efficiency-safety dynamic coupling system. This system encompasses the interaction relationships of core subsystems such as server clusters, air conditioning, and UPS power supply. A multi-objective reinforcement learning network (such as an improved NSGA-III algorithm, which introduces an adaptive crossover and mutation operator to enhance solution diversity) is used to construct the nonlinear mapping relationship between the two. For example, when the room temperature rises from 22°C to 24°C, the air conditioning compressor operating frequency drops from 60% to 45%, and the power efficiency ratio (PUE) drops from 1.3 to 1.2 (energy saving effect improves by about 7.7%). However, due to the reduced heat dissipation margin, the safety index (server overheating risk) of the core server increases from 5% to 8% (safety risk increases by 60%). When the server load rate increases from 50% to 80%, the total power consumption of the cluster increases from 60kW to 95kW (power efficiency ratio PUE increases from 1.2 to 1.4), but the temperature of the core server components tends to stabilize (overheating risk decreases from 8% to 3%). The network can accurately capture this complex nonlinear relationship of "temperature-driven inverse fluctuations in energy efficiency and safety, and load-driven positive correlation between energy efficiency and safety". At the same time, it can dynamically update the weight coefficients of the mapping relationship by combining the load characteristics of different time periods each day (such as peak teaching hours 8:00-22:00 and off-peak hours 22:00-8:00) to ensure that the mapping results conform to the periodic operation pattern of the university data center.

[0029] Key parameters are extracted from the standardized state feature vector to construct dynamic constraints. These dynamic constraints characterize the balance threshold between energy efficiency optimization and security assurance, ensuring that the dual-objective optimization process conforms to the physical operation rules of the data center. Key parameters are extracted from the standardized state feature vector, covering real-time load rates of server clusters (e.g., core server CPU utilization, memory usage), hotspot temperatures in different areas of the data center (including rack inlet and outlet temperatures, and the overall average temperature of the data center), air conditioning efficiency (e.g., compressor COP, supply and return air temperature difference, and fan unit airflow energy consumption), and UPS operating parameters (e.g., load rate, conversion efficiency, and remaining battery capacity). Parameter correlation analysis identifies the top 20% of core indicators with the greatest impact on energy efficiency and security. Based on these, dynamic constraints are constructed to achieve a dynamic balance between energy efficiency optimization and security assurance.

[0030] For example, the core server area (such as the server rack cluster hosting the teaching management system and research database) is subject to a three-level temperature threshold constraint: an upper limit of 25℃ for normal operation (safety baseline), a warning threshold of 23℃ (triggering the pre-cooling mechanism), and an emergency threshold of 26℃ (forced activation of backup air conditioning); the air conditioning unit load rate is dynamically adjusted according to the outdoor ambient temperature, controlled at 60%-75% in summer (outdoor temperature ≥30℃) (to avoid efficiency degradation caused by full-load operation at high temperatures), and relaxed to 70%-85% in winter (outdoor temperature ≤10℃) (utilizing natural cooling sources to improve efficiency). Energy efficiency); server clusters are constrained by service priority levels. Core services (such as online classroom systems) maintain their rated frequency when the load is below 50%, while non-core services (such as resource storage nodes) automatically reduce their frequency by 30% when the load is below 30%, and the power consumption of a single node after frequency reduction must not be lower than 40% of the rated value (to prevent hardware damage); the UPS redundancy power ratio is dynamically adjusted according to load fluctuations, maintaining a peak ratio of 1.2 times during normal load periods and increasing to 1.3 times during peak teaching periods (such as weekdays from 8:00 to 22:00) to ensure power continuity under extreme conditions. These dynamic constraints not only conform to the physical relationship between air conditioning cooling capacity and temperature, and the nonlinear characteristics of server power consumption and load, but also allow for flexible adjustment of balancing strategies based on real-time status, providing a scientific boundary for dual-objective optimization.

[0031] The standardized state feature vector is input into this module, and an improved deep deterministic policy gradient algorithm is used to collaboratively optimize the air conditioning cooling strategy and the server cluster scheduling scheme, obtaining preliminary control parameters, including the air conditioning supply air temperature range, server load balancing threshold, dynamic migration trigger delay, and corresponding UPS redundancy power ratio. The standardized state feature vector is then input into the energy efficiency-safety dual-objective optimization module, and an improved deep deterministic policy gradient algorithm (DDPG) is used to collaboratively optimize the air conditioning cooling strategy and the server cluster scheduling scheme. This algorithm improves the optimization accuracy in complex dynamic scenarios by introducing a priority experience replay mechanism (prioritizing the storage of samples with large energy efficiency-safety deviations) and a dual-delay update strategy (separating the update frequencies of the Actor network and the Critic network).

[0032] For example, targeting the peak experimental data processing load from 14:00 to 16:00 on weekdays (total power consumption of the server cluster reaches 120kW, a 40% increase compared to normal), the algorithm combines the temporal trend (heat load increases by 5% every 10 minutes during this period) and spatial distribution (temperature in the core rack area is 2-3℃ higher than in the edge area) in the standardized feature vector to generate preliminary control parameters, specifically including: the air conditioning supply temperature range is set according to regional differences, 19-20℃ in the core server area and 20-21℃ in the edge storage area (precisely matching the heat load differences through zoned air supply). Server load balancing thresholds are refined to core service node CPU utilization ≤65% and non-core nodes ≤70% (prioritizing the stability of critical business operations); dynamic migration trigger latency is tiered by service type, with real-time teaching services ≤30ms and offline computing services ≤50ms (balancing response speed and migration overhead); UPS redundancy power is configured at 1.2 times the peak load, with an additional 5% buffer (i.e., 1.26 times) reserved for sudden load fluctuations during that period, and the battery pack is linked to enter floating charging standby state (shortening the switching response time to within 2ms).

[0033] These parameters reduce overall energy consumption through zoned air conditioning temperature control (saving 8% more energy than a uniform 20°C solution across the entire area), prevent server overload through load grading and management (reducing the risk of overheating from 12% to 3%), and ensure power supply continuity, achieving synergistic optimization of energy efficiency and safety.

[0034] S104 processes the initial control parameters based on the adaptive execution adjustment module and the corresponding equipment response characteristic database. It then uses a PID parameter self-tuning algorithm to dynamically correct the control parameters according to the real-time operating status of the equipment. Combined with the load prediction model, it adjusts the timing of the actuator's action in advance to generate the final control command that conforms to the operating characteristics of the equipment.

[0035] In one implementation, the air conditioning supply temperature setpoint, server load threshold, and UPS redundancy power parameters in the preliminary control parameters are extracted and matched to generate equipment type-related parameter groups, parameter adjustment sensitivity ranges, and actuator response delay coefficients. The air conditioning supply temperature setpoint (e.g., 19-20℃ for core server areas, 20-21℃ for edge storage areas), server load thresholds (CPU utilization ≤65%, memory utilization ≤70% for core service nodes, CPU utilization ≤70%, memory utilization ≤75% for non-core nodes), and UPS redundancy power parameters (basic 1.2 times peak load, with an additional 5% buffer reserved during peak teaching periods, i.e., 1.26 times) are extracted from the preliminary control parameters and matched in multiple dimensions with the equipment response characteristic database (containing equipment operation logs for the past 12 months and performance parameter tables provided by the manufacturer).

[0036] Generate a set of parameters associated with the device type, such as a 10P precision air conditioner with a supply air temperature adjustment accuracy of ±0.5℃ (temperature fluctuations exceeding this range will cause the compressor to start and stop frequently), and a 5P ordinary air conditioner with an adjustment accuracy of ±1℃; a blade server with a load fluctuation tolerance of ±5% (short-term fluctuations exceeding this range can easily trigger hardware protection mechanisms), and a tower server with a tolerance of ±8%; and a modular UPS with a redundant power adjustment step size of ≤0.05 times (to avoid voltage fluctuations caused by sudden power changes).

[0037] Generate parameter adjustment sensitivity ranges, such as dividing air conditioner temperature sensitivity into [19℃, 22℃] (within this range, the compressor COP value is ≥3.2, and the energy efficiency is the highest), [18℃, 19℃)∪(22℃, 23℃] (COP value 2.8-3.2, energy efficiency is moderate), and <18℃ or >23℃ (COP value <2.8, energy efficiency is low); server load sensitivity ranges are core nodes [50%, 65%] (at this time, the power consumption and performance ratio is optimal), and non-core nodes [40%, 70%] (balancing energy saving and response speed).

[0038] Generate response delay coefficients for actuators, such as 0.8 seconds for air conditioner fan start-up delay and 0.5 seconds for fan speed adjustment delay; 0.3 seconds for server frequency reduction response delay and 0.2 seconds for frequency increase response delay (frequency reduction should be done step by step to avoid data loss, while frequency increase can quickly restore performance); and 0.02 seconds for UPS power switching response delay (to ensure power supply continuity). The delay coefficients will be dynamically adjusted according to the operating time of the equipment (e.g., the response delay coefficient of an air conditioner that has been running for more than 3 years will be increased by 10%).

[0039] The device type associated parameter group, parameter adjustment sensitivity range, and actuator response delay coefficient are processed to generate the initial tuning range of PID parameters, dynamic correction step size rules, and load prediction time window parameters. Based on the parameter group associated with the device type, the sensitivity range of parameter adjustment, and the response delay coefficient of the actuator, the initial tuning range of PID parameters is generated: differentiated settings for different device types. For example, for the temperature control of a 10P precision air conditioner, the proportional coefficient Kp = 2.0-3.0 (2.5-3.0 for the core area due to high temperature sensitivity, and 2.0-2.5 for the edge area), integral coefficient Ki = 0.1-0.3 (dynamically adjusted according to the response delay coefficient; Ki = 0.1-0.2 for a fan with a delay of 0.8 seconds, and Ki = 0.2-0.3 for a fan speed adjustment with a delay of 0.5 seconds), and derivative coefficient Kd = 0.05-0.15 (to suppress temperature overshoot); for the load control of a blade server, the proportional coefficient Kp = 1.5-2.5 (2.0-2.5 for core nodes to quickly respond to fluctuations), integral coefficient Ki = 0.05-0.2 (Ki = 0.05-0.15 for a load tolerance of ±5%).

[0040] By refining the sensitivity range of parameter adjustment, in temperature regulation, when the deviation is ≤1℃ and within the high-efficiency range of [19℃, 22℃], the step size is 0.2℃ (to avoid frequent adjustments affecting energy efficiency); when the deviation is >1℃ but still within the high-efficiency range, the step size is 0.3℃; when the deviation exceeds the high-efficiency range (e.g., <19℃ or >22℃), regardless of the deviation size, a step size of 0.5℃ is used to quickly return to the optimal range. In server load regulation, the step size is 3% when the load deviation of core nodes is ≤5%, and 5% when the deviation is >5%; the corresponding step sizes for non-core nodes are 5% and 8%, respectively, matching their higher tolerance.

[0041] Based on 24-hour time-series characteristics and dynamically set according to equipment response latency coefficients, for example, server clusters with a response latency of 0.2-0.3 seconds have a prediction window set to 15 minutes (covering short-term load fluctuation cycles); air conditioning systems with a response latency of 0.5-0.8 seconds and a lag in temperature field changes have a window extended to 20 minutes; UPS systems with a power switching latency of only 0.02 seconds have a window set to 10 minutes to accurately handle sudden loads. Simultaneously, the window parameters are dynamically adjusted with time periods, shortening to 10-15 minutes during peak teaching hours (8:00-22:00) to improve real-time performance, and extending to 20-30 minutes during off-peak hours (22:00-8:00) to reduce computational overhead.

[0042] Based on the initial tuning range of PID parameters, dynamic correction step size rules, and load prediction time window parameters, instantaneous load fluctuations, temperature field distribution deviations, and equipment start-stop switching signals in the real-time operation status of the equipment are classified and calibrated to generate parameter dynamic correction records. For instantaneous load fluctuations, refined processing is combined with server type differences. If the CPU utilization of a core service node (such as a teaching platform server) jumps from 60% to 75% within 1 minute (exceeding the 65% threshold), and it is predicted that it may continue to rise (due to the approaching start of classes) based on a 15-minute load prediction window, load balancing correction is immediately triggered, migrating the 10% of the load exceeding the limit to idle nodes in the same cluster. When non-core nodes (such as backup servers) experience similar fluctuations, a 30-second buffer is allowed before migration, matching their lower response priority.

[0043] To address temperature field distribution deviations, precise adjustments are made based on region and equipment characteristics. The actual temperature in the core area is 21℃, which is higher than the set value of 20℃ and falls within the efficient range of air conditioning temperature sensitivity [19℃, 22℃]. Through PID parameter tuning (using proportional coefficient Kp=2.8 and integral coefficient Ki=0.2), the air conditioning fan speed is dynamically increased by 10% while maintaining the supply air temperature at 19.8℃, avoiding energy efficiency losses caused by large adjustments. When the temperature deviation in the edge area is 1℃, due to the lower sensitivity, a parameter combination of Kp=2.0 and Ki=0.1 is used to fine-tune the supply air temperature by only 0.3℃.

[0044] For equipment start / stop switching signals, the timing is calibrated based on response delay coefficients and time period characteristics. When the air conditioner compressor start signal is triggered during peak teaching hours (e.g., 8:00 AM), a 0.8-second delay is considered, and the command is sent 0.5 seconds in advance to ensure startup is completed before the load spikes. When starting during late-night hours (e.g., 0:00 AM), the execution is delayed by 0.3 seconds to avoid frequent start / stop cycles that could damage the equipment. Server cluster start / stop signals are calibrated based on load prediction results. For example, if it is predicted that the load will drop below 30% in 10 minutes, a non-core server frequency reduction command is sent 5 minutes in advance (matching a 0.3-second response delay).

[0045] The generated parameter dynamic correction records contain multi-dimensional information, such as "14:05 Core area A column 3 cabinet temperature 21℃, +1℃ from the set value, after PID adjustment the air conditioner supply air temperature to 19.8℃, fan speed increased from 1200r / min to 1320r / min", "14:10 Core server S-01 CPU utilization 73%, +8% from the threshold, triggering dynamic migration, migrating 12% load to S-05 node, migration delay 28ms", "14:15 Air conditioner compressor start signal triggered, combined with 0.8 second delay 0.5 seconds in advance execution, 14:15:00.5 start completed, matching the predicted load peak from 14:15-14:30", the records fully reflect the correction logic and execution details.

[0046] The dynamic parameter correction records are integrated and optimized to generate final control commands that include air conditioner fan speed adjustment commands, server cluster frequency reduction / increase commands, and UPS power allocation commands. During the integration and optimization of these dynamic parameter correction records, a conflict arbitration mechanism is constructed by combining the device response characteristic database with real-time status data: when both heating and cooling commands occur simultaneously, if the absolute value of the temperature deviation in the core area is 1.5℃ (overheating) while the edge area is 0.8℃ (low temperature), the cooling command for the core area is executed first; if the absolute values ​​of the deviations are close, the load prediction results are referenced. For example, if it is predicted that the load in the core area will increase in 5 minutes, priority is given to ensuring the temperature stability of the core area. For conflicting frequency reduction and increase commands in the server cluster, the service priority is used for adjudication. Even if the load fluctuation of the core service node meets the frequency reduction conditions, if there is a need for frequency increase (such as a sudden teaching request), the frequency increase command is executed first.

[0047] The generated final control commands include fine-grained adjustment parameters: the air conditioning fan speed adjustment command is set according to regional differences, with the core server area fan speed increased from 1200r / min to 1300r / min (an 8.3% increase), while the edge storage area maintains 1100r / min (due to stable temperature), and the duration of the adjustment is also indicated (e.g., maintaining until the peak load ends at 16:00); the server cluster frequency reduction / increase command clearly specifies the graded range, with non-core servers (such as resource storage nodes) reduced by 20% (from 2.8GHz to 2.24GHz), and the power consumption of a single node after the frequency reduction is limited to a certain level. At 40% of the rated value (to avoid hardware damage), the core server (such as an online classroom node) increases its frequency by 5% (from 3.0GHz to 3.15GHz) when the load drops to 55% to improve response speed; the UPS power allocation command, combined with real-time load rate dynamic fine-tuning, adjusts the redundant power from 1.26 times to 1.28 times (an increase of 1.6%), while triggering the battery pack activation program (checking capacity every 15 minutes) to ensure stable power supply during the peak load period from 14:00 to 16:00, and the command includes recovery conditions (such as reverting to 1.26 times when the load rate drops below 60%).

[0048] These final control commands not only eliminate logical conflicts in the parameter correction process, but also conform to the physical characteristics of the equipment (such as the linear relationship between fan speed and air volume, and the non-linear relationship between server frequency and power consumption), ensuring that the execution effect is consistent with the dual objectives of energy efficiency and safety optimization.

[0049] S105 processes the execution effect of the final control command based on the closed-loop feedback iteration module and the corresponding performance evaluation system. It calculates the control deviation by monitoring the changes in energy efficiency and safety indicators in real time, and continuously optimizes the control model parameters using an incremental learning algorithm to generate an iteratively updated control strategy.

[0050] In one implementation, the air conditioning energy consumption change rate, server temperature fluctuation value, and UPS switching response time in the final control command execution effect are extracted and quantified to generate actual values ​​of energy efficiency indicators, actual values ​​of safety indicators, and dual-target deviation threshold information. When extracting and quantifying the execution effect of the final control command, the dimensions are refined according to equipment type and regional characteristics: the change rate of air conditioning energy consumption is calculated separately for the core server area and the edge storage area. In the core area, after the execution of the fan speed adjustment command, the energy consumption decreased from 50kW to 45kW within 1 hour, with a change rate of -10%. In the edge area, due to stable load, the energy consumption decreased slightly from 30kW to 29.5kW, with a change rate of -1.7%. The server temperature fluctuation value is statistically analyzed according to service priority. The temperature standard deviation of the core teaching server node increased from ±0.5℃ to ±0.8℃ within 15 minutes after the command was executed (due to increased load fluctuation), while that of the non-core storage server node decreased from ±0.6℃ to ±0.4℃ (due to the frequency reduction command taking effect). The UPS switching response time is averaged through 10 simulated power outage tests. After the execution of the power distribution command, it increased from 2ms to 2.3ms. Among them, the switching time of the core loop is stable at 2.1ms, while the edge loop fluctuates to 2.5ms due to the adjustment of redundancy configuration.

[0051] The generated energy efficiency index values ​​cover both overall and zoned data. For example, the overall PUE value of the data center is 1.25, the core area PUE value is 1.32 (due to high cooling intensity), and the edge area PUE value is 1.18 (with better energy efficiency). The actual values ​​of the safety index include a 6% risk of server overheating (8% for core nodes and 3% for edge nodes) and a 0.3% risk of UPS power outage (a decrease of 0.1% compared to before the instruction was executed). The dual-target deviation threshold information is dynamically set according to time periods. During peak teaching hours (8:00-22:00), the energy efficiency deviation threshold is relaxed to ±5% and the safety deviation threshold is strictly set to ±2%. During off-peak hours (22:00-8:00), the energy efficiency deviation threshold is tightened to ±3% and the safety deviation threshold is relaxed to ±4%, ensuring that the threshold settings are in line with the periodic load characteristics of university data centers.

[0052] The actual values ​​of energy efficiency indicators, actual values ​​of safety indicators, and dual-objective deviation threshold information are processed to generate control deviation quantification coefficients, deviation tracing path information, and model parameter adjustment priority rules. Combining the actual values ​​and deviation thresholds of energy efficiency and safety indicators, a dynamic weighting algorithm generates control deviation quantification coefficients: when the actual value of an energy efficiency indicator exceeds the threshold by 2% (e.g., a PUE value of 1.31 exceeds the upper limit of the threshold of 1.28), the coefficient is set to 0.4, considering its impact on overall energy consumption; among safety indicators, the server over-temperature risk reaches 8% (exceeding the upper limit of the threshold by 5%), exceeding the limit by 3%, and because it is directly related to equipment stability, the coefficient is set to 0.8 (the safety weight dynamically increases with the deviation magnitude); the UPS switching response time exceeds the threshold by 0.3ms (threshold 2ms, actual 2.3ms), exceeding the limit by 15%, but no safety warning is triggered, the coefficient is set to 0.3. The three factors are weighted and summed (safety factor 0.5, energy efficiency 0.3, UPS 0.2) to form a comprehensive deviation coefficient, which is 0.8×0.5+0.4×0.3+0.3×0.2=0.58 in this calculation.

[0053] The deviation source tracing information clarifies the deviation transmission chain through causal analysis model: For example, the server temperature fluctuation is too large (actual ±0.8℃, threshold ±0.5℃). Airflow simulation verification shows that it is caused by uneven air supply from the air conditioner. Further investigation reveals that the temperature difference between the core area and the edge area in the initial control parameters is only 1℃ (19-20℃ and 20-21℃), resulting in insufficient air pressure distribution. Further investigation shows that the zone temperature difference constraint setting in the energy efficiency-safety dual-objective optimization module is too loose (original constraint ≥0.5℃, actual requirement ≥1.5℃).

[0054] The model parameter adjustment priority rules are sorted according to three dimensions: "safety impact - energy efficiency sensitivity - adjustment cost": server load balancing threshold (priority 1, directly affecting the stability of core business, adjustment can reduce overheating risk by 40% within 1 hour) > proportional coefficient Kp in air conditioning PID parameters (priority 2, adjustment can reduce temperature fluctuation by 30% and has low cost) > air conditioning zone temperature range (priority 3, requires adjustment in conjunction with fan speed, adjustment cycle of 2 hours) > UPS redundancy power ratio (priority 4, current deviation has little impact on safety, adjustment benefit is only 5%). Meanwhile, the rules will be dynamically updated according to time periods, such as temporarily raising the UPS redundancy ratio priority to 2 during peak teaching periods to ensure power supply continuity.

[0055] Based on the control deviation quantification coefficient, deviation source tracing path information, and model parameter adjustment priority rules, the periodic fluctuation components, sudden disturbance signals, and systematic deviation trends in real-time monitoring data are classified and weighted, generating parameter optimization correlation records. When classifying real-time monitoring data based on the above rules, the weight allocation is refined in conjunction with the periodic operation characteristics of the data center: periodic fluctuation components are graded according to fluctuation amplitude and duration. For example, the peak experimental data processing load from 14:00 to 16:00 daily (lasting 2 hours, load 40% higher than average) is assigned a weight of 0.3, and the air conditioning pre-cooling strategy is optimized accordingly, starting the core area air conditioning pre-cooling mode 30 minutes in advance to lower the initial temperature from 20℃ to 19℃; the peak teaching system login time from 8:00 to 9:00 every Monday (lasting 1 hour, load fluctuation 30%) is assigned a weight of 0.25, and the server cluster wake-up sequence is adjusted accordingly to ensure that 80% of nodes are ready 10 minutes in advance.

[0056] Sudden disturbance signals are weighted differently according to their impact range and urgency. For example, if a temporary additional test for a certain profession causes the power consumption of 5 core servers to suddenly increase by 60% within 10 minutes (affecting online teaching), it is given a weight of 0.5, which immediately enhances the dynamic migration response speed and compresses the trigger latency from 30ms to 20ms. If construction outside the computer room causes instantaneous voltage fluctuations (lasting 0.5 seconds, without affecting equipment operation), it is given a weight of 0.1, and only the fluctuation characteristics are recorded for subsequent UPS parameter calibration.

[0057] Systematic deviation trends are weighted in conjunction with equipment lifecycle and degradation patterns. For example, if the cooling efficiency of an air conditioner decreases by 2% per month with usage time (confirmed by monitoring for 3 consecutive months), it is assigned a weight of 0.2, and the equipment aging coefficient is updated to 1.06 (baseline value 1.0). The integral coefficient Ki in the PID parameters is adjusted from 0.2 to 0.22. If the UPS battery capacity degrades by 1.5% per quarter, it is assigned a weight of 0.15, and the base value of the redundant power ratio is increased from 1.2 times to 1.22 times.

[0058] The generated parameter optimization correlation records contain multi-dimensional optimization suggestions, such as "During the peak load period from 14:00 to 16:00, the risk deviation of server overheating in the core area is +2% (actual 8%, threshold 6%). After tracing the source, it was found that the air conditioner supply temperature range of 19-20℃ is too wide, resulting in local hot spot temperatures reaching 25℃. It is recommended to lower the upper limit of the core area by 0.5℃ to 19.5℃, and at the same time increase the fan speed by 5% to enhance airflow circulation." "During the peak teaching login period from 8:00 to 9:00 every Monday, the server cluster wake-up delay causes the response speed to decrease by 15% in the first 5 minutes. It is recommended to start the preheating program of non-core nodes 15 minutes in advance to increase the readiness rate from 70% to 90%." "After the air conditioner has been running for 36 months, the cooling efficiency has decreased by 7% compared to a new machine. It is recommended to increase the filter cleaning frequency once a quarter, and at the same time lower the lower limit of the temperature adjustment sensitivity range from 19℃ to 18.5℃ to compensate for the efficiency loss."

[0059] The parameter optimization correlation records are integrated and iteratively learned to generate an iteratively updated control strategy that includes a dynamic adjustment strategy for energy efficiency and safety weights, self-calibration rules for control parameters, and an update mechanism for equipment response characteristics. During the integration of parameter optimization correlation records, historical data from the past 30 days and real-time feedback information are combined to iteratively optimize the control model using an incremental learning algorithm (such as an online update model based on random forests), generating an iteratively updated control strategy: the dynamic adjustment strategy for energy efficiency and safety weights is finely set according to time period and load type. During peak teaching periods (e.g., weekdays 8:00-22:00), the safety weight increases from 0.6 to 0.7, with a further increase to 0.75 during core teaching system operation (e.g., 9:00-12:00, 14:00-17:00) to prioritize service continuity; during non-teaching periods (22:00-8:00), the safety weight decreases to 0.5, while the energy efficiency weight increases to 0.5, allowing for a moderate relaxation of temperature constraints to improve energy-saving effects. Meanwhile, when the server cluster load rate is detected to exceed 70%, the security weight is automatically and temporarily increased by 0.1 to avoid the accumulation of security risks under high load.

[0060] The self-calibration rules for control parameters dynamically adjust the correction frequency and step size according to the equipment type and deviation amplitude: In the air conditioning PID parameters, the proportional coefficient Kp is corrected hourly according to the absolute value of the temperature deviation. When the deviation is ≤1℃, the correction is ±0.05, and when the deviation is >1℃, the correction is ±0.1. Moreover, the correction step size in the core area is 20% higher than that in the edge area. The server load balancing threshold is self-calibrated at 2:00 AM every day based on the load fluctuation data of the previous day. If the number of times the threshold is exceeded in a single day exceeds 5 times, the threshold is lowered by 3%, and vice versa, it is raised by 2%, to ensure that the threshold matches the actual load characteristics. The equipment response characteristic update mechanism sets the update cycle according to the equipment life cycle and the differences in the operating environment: the UPS switching response time is averaged by 10 simulated switching tests per month. If the test results deviate by more than 5% for two consecutive months, the response delay coefficient is updated by ±5%, and the backup power switching lead is adjusted accordingly (if the delay coefficient is increased, the lead is increased by 0.2ms). The air conditioning equipment updates the correlation curve between fan speed and cooling capacity every quarter according to the filter cleanliness and changes in outdoor temperature and humidity. For example, in the high temperature of summer, the "cooling capacity corresponding to speed of 1200r / min" is revised from the original 50kW to 48kW to match the actual energy efficiency degradation.

[0061] For example, after iterative optimization, the air conditioning supply temperature range in the core area has been slightly adjusted from 19-20℃ to 18.5-19.5℃, and a dynamic floating mechanism has been added (automatically narrowing to 18.5-19℃ when the load rate exceeds 70%). The server dynamic migration trigger latency has been shortened from 30ms to 25ms, with the real-time teaching data migration latency further compressed to 20ms, and a new protection rule has been added to "automatically pause migration when the CPU utilization exceeds 80% during the migration process". The UPS redundancy power ratio has been increased from 1.26 times to 1.3 times during peak teaching hours and maintained at 1.2 times during off-peak hours. At the same time, the charging and discharging thresholds are adjusted monthly based on battery capacity test results (e.g., if the capacity decays by 5%, the charging threshold is increased from 90% to 92%). These updated strategies retain historical optimization experience while quickly adapting to dynamic changes such as equipment aging and load fluctuations, continuously improving control accuracy.

[0062] S106 inputs the standardized state feature vector, preliminary control parameters, final control commands, and iteratively updated control strategies into the visualization decision support module for processing, generating a comprehensive evaluation report that includes energy efficiency optimization rate, safety risk warning, and equipment operating status trend.

[0063] In one implementation, the time-series characteristic fluctuation value and static characteristic attribute parameter in the standardized state feature vector, the air conditioning temperature range and server load threshold in the preliminary control parameters, the equipment adjustment amount and execution response time in the final control instruction, and the parameter optimization magnitude and weight adjustment coefficient in the iteratively updated control strategy are extracted and structured to generate an energy efficiency basic data set, a safety status dataset, a control process parameter set, and a strategy optimization trajectory set. Extract time-series characteristic fluctuation values ​​(e.g., temperature fluctuation of ±0.5℃ in the core area and ±0.8℃ in the edge area every 10 minutes, total power consumption fluctuation of the server cluster ±300W, and air conditioner compressor frequency fluctuation ±5Hz) and static characteristic attribute parameters (e.g., server model is blade HPE ProLiantDL380, service life is 2 years, air conditioner is 10P precision air conditioner, installation location coordinates (Area A, column 3), UPS is modular 200kVA, and battery cycle count is 120 times) from the standardized state feature vector; extract the air conditioning temperature range (core server area 19-20℃, edge storage area 20-21℃), server load threshold (core node CPU utilization ≤65%, memory utilization ≤70%, non-core node CPU utilization ≤70%, memory utilization ≤75%), dynamic migration trigger latency (real-time teaching service ≤30ms, offline computing service ≤50ms), and UPS redundancy power ratio (basic 1.2 times, peak teaching 1). 26 times); extract equipment adjustment quantities from the final control commands (air conditioner fan speed increased from 1200r / min to 1300r / min, non-core server frequency reduced by 20% (from 2.8GHz to 2.24GHz), UPS redundancy power fine-tuned from 1.26 times to 1.28 times), execution response time (the core area temperature drops by 1℃ 5 minutes after air conditioner adjustment, the power consumption decreases by 15% 3 minutes after server frequency reduction, and the UPS power adjustment is completed within 2ms); extract parameter optimization magnitude from the iteratively updated control strategy (the core area air conditioner temperature range is reduced by 0.5℃ from 19-20℃ to 18.5-19.5℃, the server dynamic migration trigger latency is shortened from 30ms to 25ms), weight adjustment coefficients (safety weight 0.7 and energy efficiency weight 0.3 during peak teaching hours 8:00-22:00, safety weight 0.5 and energy efficiency weight 0.5 during non-teaching hours 22:00-8:00, and the safety weight is further increased to 0.75 when the core teaching system is running).

[0064] After structured processing, the following data sets are generated: Basic energy efficiency data set (e.g., overall data center PUE value 1.25, core area PUE value 1.32, edge area PUE value 1.18, daily power saving 120kWh, weekly power saving 850kWh, average daily air conditioning energy consumption 450kWh, average daily server cluster energy consumption 680kWh); Safety status data set (e.g., server overheating risk 6% (core node 8%, edge node 3%), UPS power outage risk 0.3%, switching response time 2.3ms (core loop 2.1ms, edge loop 2.5ms), average daily number of times server room temperature exceeds limit 1, average daily number of times server load exceeds limit 3); Control process parameter set (e.g., air conditioning temperature control PID parameters Kp=2.8, K...). i=0.2, Kd=0.1, server load control PID parameters Kp=2.2, Ki=0.15, dynamic correction step size (0.2℃ when temperature deviation ≤1℃, 0.5℃ when >1℃, 3% when load deviation ≤5%, 5% when >5%), load prediction time window (15 minutes for servers, 20 minutes for air conditioners, 10 minutes for UPS)); strategy optimization trajectory set (such as the air conditioner temperature range adjustment record in the past 7 days: 19-20℃ in the core area on October 10, reduced to 18.8-19.8℃ on October 12, reduced to 18.5-19.5℃ on October 15, and the safety-energy efficiency weight adjustment record in the past 7 days: safety weight 0.7 during weekdays 8:00-22:00, and 0.65 during the same period on non-weekdays).

[0065] The system processes the energy efficiency baseline data set, safety status dataset, control process parameter set, and strategy optimization trajectory set to generate energy efficiency-safety correlation heatmap rules, control parameter sensitivity ranking standards, strategy iteration effect quantitative indicators, and anomaly warning threshold range information. A multi-level color gradient is applied based on the combination of PUE value and over-temperature risk: dark green (excellent) for PUE values ​​of 1.2-1.3 and over-temperature risk of 3%-5%; light green (good) for PUE values ​​of 1.3-1.4 and over-temperature risk of 5%-7%; orange-red (warning) for PUE values ​​of 1.3-1.4 and over-temperature risk of 7%-9%; and dark red (high risk) for PUE > 1.4 and over-temperature risk > 9%. For example, a core area with a PUE of 1.32 and an over-temperature risk of 8% meets the combination of "PUE 1.3-1.4 + risk 7%-9%" and is defined as an orange-red warning area; an edge area with a PUE of 1.18 and an over-temperature risk of 3% is displayed as dark green.

[0066] Based on a weighted ranking of the impact of parameter adjustments on energy efficiency and safety, the server load balancing threshold (which can reduce overheating risk by 40% and improve energy efficiency by 8% within 1 hour after adjustment) ranks first; followed by the proportional coefficient Kp in the air conditioning PID parameters (which reduces temperature fluctuation by 30% and energy consumption by 5% after adjustment); then the air conditioning zone temperature range (which can improve energy efficiency in the core area by 6% after adjustment, but takes 2 hours to take effect); then the UPS redundancy power ratio (which has only a 5% impact on safety and a 1% improvement in energy efficiency after adjustment); and finally the non-core server frequency reduction threshold (whose impact is limited to the edge area and improves energy efficiency by 3%).

[0067] This includes the energy efficiency optimization rate of a single iteration (e.g., after the strategy update on October 15, the daily energy consumption in the core area decreased from 50kW to 45kW, with an optimization rate of 10%), the reduction in safety risks (the risk of overheating decreased from 8% to 6%, a reduction of 25%), and the comprehensive benefit score (the weighted sum of energy efficiency improvement and risk reduction, with weights of 0.3 and 0.7 respectively, and the score for this iteration is 10%×0.3+25%×0.7=20.5 points); it also includes long-term trend indicators, such as after four consecutive weeks of iterations, the overall PUE value decreased from 1.35 to 1.25, with a cumulative optimization rate of 7.4%.

[0068] The energy efficiency dimension is further refined into a Level 1 energy efficiency warning triggered by a PUE value > 1.4 (requiring immediate inspection of the air conditioning load), and a Level 2 warning triggered by a PUE value between 1.35 and 1.4 (recommending adjustment of the supply air temperature). The safety dimension distinguishes between temperature and power supply indicators: a temperature > 25℃ in the core area of ​​the data center triggers a Level 1 safety warning (forcibly starting the backup air conditioning), and 23-25℃ triggers a Level 2 warning (increasing fan speed). A UPS switching response time > 2.5ms triggers a power supply warning (checking battery status), and > 3ms triggers an emergency warning (switching to backup power). For example, when the temperature of a core server reaches 25.5℃, a Level 1 safety warning is triggered; when the UPS switching response time is measured at 2.6ms, a power supply warning is triggered.

[0069] Based on the energy efficiency-safety correlation heatmap rules, control parameter sensitivity ranking standards, strategy iteration effect quantitative indicators, and anomaly warning threshold range information, energy efficiency improvement nodes, safety risk points, and control strategy turning points in data center operation data are classified and visualized to generate multi-dimensional operation evaluation records. When classifying and visualizing operation data based on evaluation rules, the identification logic is refined by combining the spatial distribution and time dimension of the data center. Energy efficiency improvement nodes are classified and identified according to the degree of optimization. After the strategy adjustment, a 10% decrease in energy consumption in the core area from 14:00 to 16:00 (significant optimization) is marked with a dark blue triangle, and a 3% decrease in energy consumption in the edge area (slight optimization) is marked with a light blue triangle. The specific energy saving value is also marked (e.g., "-10% (5kW)" is marked next to the dark blue triangle). Safety risk points are presented differently according to the degree of urgency. If the temperature fluctuation of a core server reaches ±0.8℃ (exceeding the threshold ±0.5℃) and lasts for more than 5 minutes, it is marked with a solid red dot. For edge servers, similar fluctuations lasting less than 1 minute are indicated by hollow red dots, with the dot size positively correlated with the fluctuation amplitude (e.g., ±0.8℃ corresponds to a diameter of 8mm, ±0.6℃ corresponds to a diameter of 6mm). Control strategy turning points are further subdivided based on event type: the air conditioning pre-cooling mode is activated 30 minutes before the peak teaching time (8:00) using a yellow square; the dynamic migration mechanism is triggered when the server cluster reaches 60% load using a yellow diamond; and the UPS redundancy power ratio is automatically increased to 1.3 times at 14:00 using a yellow pentagram. All types of indicators are labeled with trigger conditions (e.g., "Pre-cooling start: temperature ≥22℃" is noted next to the yellow square).

[0070] The generated multi-dimensional operational evaluation record includes spatiotemporal coordinates, optimization logic, and related parameters. For example, "At 8:00 AM on October 15th, due to the implementation of a 30-minute pre-cooling strategy (air conditioning supply temperature dropped from 20℃ to 19℃), the energy consumption of cabinet 3 in column A of the core area decreased by 5% from 8:00 AM to 9:00 AM compared to the same period the previous day (from 40kW to 38kW), and the risk of overheating decreased from 8% to 6%. The corresponding strategy inflection point is marked with a yellow square, and the related PID parameter Kp was adjusted from 2.5 to 2.8." "At 3:00 PM on October 15th, the server temperature fluctuation of cabinet 1 in column C of the edge area reached ±0.8℃ (the actual measured maximum was 24.5℃ and the minimum was 22.9℃), holding..." For 6 minutes, the hollow red dot indicator was triggered. The cause was traced to uneven airflow from the air conditioner. The record also shows the effect of the correction after the fan speed in this area was increased from 1100 r / min to 1200 r / min (the fluctuation narrowed to ±0.4℃ at 15:06). "At 20:00 on October 15, the dynamic migration mechanism (yellow diamond indicator) was triggered because the server cluster load was detected to reach 60%. 15% of the load was migrated from node S-01 to node S-08. After the migration, the CPU utilization of node S-01 dropped from 72% to 58%. The record includes a migration latency of 28ms and the basis for the load balancing threshold adjustment (temporarily reduced from 65% to 60%)."

[0071] The system integrates and performs in-depth analysis of multi-dimensional operational assessment records to generate a comprehensive assessment report that includes dynamic energy efficiency optimization rate curves, real-time early warning of safety risk levels, trend prediction of equipment operating status, and strategy optimization suggestions. When integrating and analyzing multi-dimensional assessment records, multi-dimensional outputs are generated by combining temporal trends and spatial distribution characteristics: the dynamic energy efficiency optimization rate curve is subdivided by time period and region. The overall daily average optimization rate over the past 30 days is 8%, with the optimization rate reaching 7% on weekdays due to stable load and rising to 12% on weekends due to extended non-teaching hours. The optimization rate in the core area (PUE value decreased from 1.32 to 1.28) is 6%, and the optimization rate in the peripheral area (PUE value decreased from 1.18 to 1.10) is 10%. The curve marks key optimization nodes (e.g., the daily optimization rate increased to 9% after the strategy iteration on October 10). The real-time warning of safety risk level adopts a five-color classification system. The core area is currently judged as "low" risk (green mark) because the temperature is stable at 23-24℃ and the risk of overheating is 6%. The peripheral area is judged as "extremely low" risk (dark green mark) because the temperature fluctuation is ≤±0.4℃ and the risk is 3%. The risk change trajectory over the past 24 hours is displayed simultaneously (e.g., the risk in the core area briefly rose to "medium" level from 14:00 to 16:00 and then fell back).

[0072] Based on the equipment operation status trend prediction combined with aging models and maintenance records, the air conditioning system's efficiency is decreasing by 2% per month due to dust accumulation on the filter. The current cooling capacity is 7% lower than that of a new machine, and it is expected to decrease by 13% in 3 months, requiring filter cleaning to be arranged in advance. In the server cluster, nodes that have been running for more than 3 years have increased power consumption by 5% compared to a new machine, and it is predicted that fan maintenance will be required in 6 months. The UPS battery capacity decreases by 1.5% per quarter, and the current remaining capacity is 92%, and it is expected that the spare battery will need to be replaced in 12 months.

[0073] The strategy optimization recommendations are generated based on a dual-objective balance algorithm. In the short term, it is recommended to increase the safety weight during peak teaching hours (8:00-22:00) from 0.7 to 0.75, and further increase it to 0.8 during the core teaching system's operating hours (9:00-12:00, 14:00-17:00). Simultaneously, the air conditioning supply temperature range should be narrowed by 0.5℃, which can further reduce the risk of overheating by 1.5%. In the long term, it is recommended to update the PID parameter tuning range quarterly according to the equipment aging coefficient. For example, the Kp value of the air conditioner should be increased by 0.1 annually as efficiency decreases to compensate for the reduction in cooling capacity.

[0074] The report includes a 3D visualization interface modeled at a 1:1 scale according to the physical layout of the data center. It displays the status of each area through a red-yellow-green color gradient overlaid with numerical labels: the core server area is marked with light green "PUE 1.28, risk 6%", the edge storage area is marked with dark green "PUE 1.10, risk 3%", and the air-conditioned data center is marked with yellow "efficiency 93%, maintenance recommended in 3 months". It also supports timeline backtracking (such as viewing the status changes of a certain area in the past 7 days) and parameter linkage query (clicking on a node can display its real-time data on load, temperature, and energy consumption), providing managers with intuitive decision support.

[0075] In one implementation, such as Figure 2 As shown, this application also provides a green energy-saving and safety linkage control device for university data centers based on a dynamic environment system, comprising: The data acquisition module 201 is used to collect multi-dimensional real-time status data of the data center's environmental system, including room temperature and humidity, server power consumption, air conditioning operating parameters, and UPS status. The processing module 202 is used to process multi-dimensional real-time state data based on the intelligent sensing fusion module to generate standardized state feature vectors; to process the standardized state feature vectors based on the energy efficiency-safety dual-objective optimization module to generate preliminary control parameters; to process the preliminary control parameters based on the adaptive execution adjustment module and the corresponding equipment response characteristic database to generate final control commands that conform to the equipment operating characteristics; to process the execution effect of the final control commands based on the closed-loop feedback iteration module and the corresponding performance evaluation system to generate an iteratively updated control strategy; and to input the standardized state feature vectors, preliminary control parameters, final control commands, and iteratively updated control strategies into the visualization decision support module for processing to generate a comprehensive evaluation report that includes energy efficiency optimization rate, safety risk warning, and equipment operating status trend.

[0076] The computer-readable storage medium provided in the above embodiments of this application and the green energy-saving and safety linkage control method for university data centers based on dynamic environmental systems provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

Claims

1. A green energy-saving and safety linkage control method for university data centers based on dynamic environmental systems, characterized in that, include: Collect multi-dimensional real-time status data of the data center's environmental system, including room temperature and humidity, server power consumption, air conditioning operating parameters, and UPS status; The intelligent perception fusion module processes multi-dimensional real-time state data, uses a dynamic weight allocation algorithm to evaluate and fuse data from different sources, extracts hidden correlation patterns in network mining data using time-series features, and removes noise interference by combining an outlier adaptive filtering mechanism to generate standardized state feature vectors. Based on the energy efficiency-safety dual-objective optimization module, the standardized state feature vector is processed. The mapping relationship between energy efficiency indicators and safety indicators is constructed through a multi-objective reinforcement learning network. Dynamic constraints are designed to coordinate the optimization of air conditioning cooling strategy and server cluster scheduling scheme, and preliminary control parameters are generated. The initial control parameters are processed based on the adaptive execution adjustment module and the corresponding equipment response characteristic database. The control parameters are dynamically corrected according to the real-time operating status of the equipment through the PID parameter self-tuning algorithm. Combined with the load prediction model, the timing of the actuator action is adjusted in advance to generate the final control command that conforms to the operating characteristics of the equipment. The execution effect of the final control command is processed based on the closed-loop feedback iteration module and the corresponding performance evaluation system. The control deviation is calculated by monitoring the changes in energy efficiency and safety indicators in real time, and the control model parameters are continuously optimized by using incremental learning algorithms to generate iteratively updated control strategies. The standardized state feature vector, preliminary control parameters, final control commands, and iteratively updated control strategies are input into the visualization decision support module for processing, generating a comprehensive evaluation report that includes energy efficiency optimization rate, safety risk warning, and equipment operating status trend.

2. The method as described in claim 1, characterized in that, The system processes multidimensional real-time state data based on an intelligent sensing fusion module. It uses a dynamic weight allocation algorithm to assess and fuse data from different sources, extracts hidden correlations from network mining data using time-series features, and removes noise interference using an adaptive outlier filtering mechanism. This generates a standardized state feature vector, including: The intelligent sensing fusion module processes multi-dimensional real-time status data. The dynamic weight allocation algorithm performs hierarchical credibility assessment on data from different sources. The high-precision sensors in the core area are assigned the first basic weight, and the ordinary sensors in the edge area are assigned the second basic weight. At the same time, the weight deviation value is dynamically adjusted in combination with the historical failure rate of the equipment. The spatiotemporal alignment fusion of multi-source data is completed through timestamp calibration and spatial coordinate mapping to eliminate information misalignment caused by the time difference and position deviation of different devices. The temporal feature extraction network utilizes a nested structure of bidirectional LSTM and attention mechanism. The bidirectional LSTM layer captures the short-term fluctuation patterns of data within a 5-30 minute period, while the attention mechanism layer assigns 1.2-1.5 times weight to the features of key time nodes such as peak server load periods and air conditioner start-stop switching points to enhance extraction. The deep fully connected network further explores the long-term trend correlations of data within a 24-hour period. By combining an adaptive outlier filtering mechanism and co-verifying the 3σ criterion with the isolated forest algorithm, outliers that deviate from the mean by more than three times the standard deviation are initially screened out using the 3σ criterion. Then, the isolated forest model is used to make a second judgment on the outliers, eliminating instantaneous noise caused by sudden electromagnetic interference and systematic deviations caused by equipment aging. Finally, a high-fidelity standardized state feature vector containing 288-dimensional temporal features and 16-dimensional static features is generated.

3. The method as described in claim 1, characterized in that, The standardized state feature vector is processed based on an energy efficiency-safety dual-objective optimization module. A mapping relationship between energy efficiency and safety indicators is constructed using a multi-objective reinforcement learning network. Dynamic constraints are designed to collaboratively optimize air conditioning cooling strategies and server cluster scheduling schemes, generating preliminary control parameters, including: Based on the dual-objective optimization module of energy efficiency and safety, the standardized state feature vector is processed, and the data center operation process is modeled as a dynamic coupling system of energy efficiency and safety. A multi-objective reinforcement learning network is used to construct a nonlinear mapping relationship between energy efficiency indicators and safety indicators. Key parameters are extracted from the standardized state feature vector to construct dynamic constraints. These dynamic constraints are used to characterize the balance threshold between energy efficiency optimization and security assurance, ensuring that the dual-objective optimization process complies with the physical operation rules of the data center. The standardized state feature vector is input into this module, and the improved deep deterministic strategy gradient algorithm is used to coordinate the optimization of the air conditioning cooling strategy and the server cluster scheduling scheme to obtain preliminary control parameters, including the air conditioning supply temperature range, the server load balancing threshold, the dynamic migration trigger delay, and the corresponding UPS redundancy power ratio.

4. The method as described in claim 1, characterized in that, The initial control parameters are processed based on the adaptive execution adjustment module and the corresponding equipment response characteristic database. A PID parameter self-tuning algorithm dynamically corrects the control parameters according to the real-time operating status of the equipment. Combined with a load prediction model, the timing of actuator actions is adjusted in advance to generate final control commands that conform to the equipment's operating characteristics, including: The air conditioning supply temperature setpoint, server load threshold, and UPS redundant power parameters in the preliminary control parameters are extracted and matched to generate equipment type associated parameter groups, parameter adjustment sensitivity ranges, and actuator response delay coefficients. The device type associated parameter group, parameter adjustment sensitivity range, and actuator response delay coefficient are processed to generate the initial tuning range of PID parameters, dynamic correction step size rules, and load prediction time window parameters. Based on the initial tuning range of PID parameters, dynamic correction step size rules, and load prediction time window parameters, instantaneous load fluctuations, temperature field distribution deviations, and equipment start-stop switching signals in the real-time operating status of the equipment are classified and calibrated to generate parameter dynamic correction records. The dynamic parameter correction records are merged and optimized to generate final control commands that include air conditioner fan speed adjustment commands, server cluster frequency reduction / increase commands, and UPS power allocation commands.

5. The method as described in claim 4, characterized in that, The execution effect of the final control command is processed based on the closed-loop feedback iterative module and the corresponding performance evaluation system. Control deviation is calculated by real-time monitoring of changes in energy efficiency and safety indicators. Incremental learning algorithms are used to continuously optimize the control model parameters, generating iteratively updated control strategies, including: The air conditioning energy consumption change rate, server temperature fluctuation value, and UPS switching response time in the final control command execution effect are extracted and quantified to generate actual values ​​of energy efficiency indicators, actual values ​​of safety indicators, and dual-target deviation threshold information. The actual values ​​of energy efficiency indicators, actual values ​​of safety indicators, and dual-objective deviation threshold information are processed to generate control deviation quantification coefficients, deviation source tracing path information, and model parameter adjustment priority rules. Based on the control deviation quantification coefficient, deviation source tracing path information, and model parameter adjustment priority rules, the periodic fluctuation components, sudden disturbance signals, and systematic deviation trends in real-time monitoring data are classified and weighted, generating parameter optimization correlation records. The parameter optimization correlation records are integrated and iteratively learned to generate an iteratively updated control strategy that includes a dynamic adjustment strategy for energy efficiency and safety weights, self-calibration rules for control parameters, and an update mechanism for equipment response characteristics.

6. The method as described in claim 1, characterized in that, The standardized state feature vector, preliminary control parameters, final control commands, and iteratively updated control strategies are input into the visualization decision support module for processing, generating a comprehensive evaluation report that includes energy efficiency optimization rate, safety risk warnings, and equipment operating status trends, including: The time-series characteristic fluctuation values ​​and static characteristic attribute parameters in the standardized state feature vector, the air conditioning temperature range and server load threshold in the preliminary control parameters, the equipment adjustment amount and execution response time in the final control command, and the parameter optimization magnitude and weight adjustment coefficient in the control strategy after iterative update are extracted and structured to generate energy efficiency basic data set, safety status dataset, control process parameter set, and strategy optimization trajectory set; The energy efficiency basic data set, safety status dataset, control process parameter set, and strategy optimization trajectory set are processed to generate energy efficiency-safety correlation heat map rules, control parameter sensitivity ranking criteria, strategy iteration effect quantitative indicators, and anomaly warning threshold range information. Based on the energy efficiency-safety correlation heatmap rules, control parameter sensitivity ranking standards, strategy iteration effect quantitative indicators, and anomaly warning threshold range information, energy efficiency improvement nodes, safety risk points, and control strategy turning points in data center operation data are classified and visualized to generate multi-dimensional operation evaluation records. The system integrates and performs in-depth analysis of multi-dimensional operational assessment records to generate a comprehensive assessment report that includes dynamic energy efficiency optimization rate curves, real-time early warning of safety risk levels, trend prediction of equipment operating status, and strategy optimization suggestions.

7. A green energy-saving and safety linkage control device for a university data center based on a dynamic environment system, characterized in that, The device includes: The data acquisition module is used to collect multi-dimensional real-time status data of the data center's environmental system, including room temperature and humidity, server power consumption, air conditioning operating parameters, and UPS status. The processing module is used to process multi-dimensional real-time state data based on the intelligent sensing fusion module to generate standardized state feature vectors; to process the standardized state feature vectors based on the energy efficiency-safety dual-objective optimization module to generate preliminary control parameters; to process the preliminary control parameters based on the adaptive execution adjustment module and the corresponding equipment response characteristic database to generate final control commands that conform to the equipment operating characteristics; to process the execution effect of the final control commands based on the closed-loop feedback iteration module and the corresponding performance evaluation system to generate iteratively updated control strategies; and to input the standardized state feature vectors, preliminary control parameters, final control commands, and iteratively updated control strategies into the visualization decision support module for processing to generate a comprehensive evaluation report that includes energy efficiency optimization rate, safety risk warning, and equipment operating status trends.

8. An electronic device, characterized in that, include: First processor; and memory for storing executable instructions of the first processor; The first processor is configured to execute the green energy-saving and safety linkage control method for a university data center based on an environmental system, as described in any one of claims 1 to 6, by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the green energy-saving and safety linkage control method for university data centers based on dynamic environmental systems as described in any one of claims 1 to 6.

Images