A full-cycle matrix data tracing method for air conditioning in computer room energy efficiency optimization

Abstract

The present invention relates to the field of energy management and operation and maintenance. Disclosed is a full-cycle matrix data tracing method for air conditioning for energy efficiency optimization of computer rooms, the method comprising: integrating the asset configuration data of the computer room from the four stages of design, construction, commissioning and operation, as well as the operational energy efficiency data of the four dimensions of equipment, system, control and energy efficiency, to construct a full-cycle data management matrix. A grey correlation analysis model is used to calculate the impact weight of the preset energy efficiency target and generate a corresponding energy efficiency correlation matrix. When an energy efficiency anomaly is monitored, the corresponding energy efficiency correlation matrix is ​​matched according to the anomaly type, and reverse tracing is performed along the correlation path with the highest impact weight, the anomaly is accurately attributed, the key influencing factors are identified, and the optimization strategy is generated and executed based on the key influencing factors. The present invention achieves accurate positioning and closed-loop optimization of energy efficiency problems by establishing full-cycle data correlation, thereby providing an effective technical means for improving the energy efficiency of computer rooms.

Description

Technical Field

[0001] The present invention relates to the field of energy management and operation and maintenance, and in particular to an air conditioning full-cycle matrix data tracing method for optimizing the energy efficiency of a computer room. Background Art

[0002] Currently, computer room administrators save the information collected by each sub-device in different databases or time series log files, and use artificial intelligence and machine learning algorithms to predict energy efficiency, anticipate the content of operation and maintenance management work, and preliminarily optimize the energy efficiency operation and maintenance work of the computer room. These technologies have improved operational management efficiency and energy utilization to a certain extent.

[0003] In current data center and air-conditioning room operation and maintenance practices, data center infrastructure management systems are commonly used for energy efficiency monitoring. Their core function is to collect and display the operating data of the computer room in real time. During the construction of the computer room, a large amount of static, offline asset data is generated. In current technology, real-time operational data and static asset data are severely fragmented and stored in different systems, forming typical data silos. When an energy efficiency anomaly occurs, the traditional approach of operations and maintenance personnel is to observe surface phenomena and manually troubleshoot and verify. When optimizing computer room energy efficiency, existing technologies face a core technical problem throughout: because the data throughout the computer room's life cycle is fragmented and unrelated, existing monitoring systems can only monitor surface phenomena and are unable to identify decisive influencing factors. Energy efficiency anomaly attribution analysis remains at a superficial level, making it difficult to formulate accurate and fundamental energy efficiency optimization strategies.

[0004] Therefore, an air conditioning full-cycle matrix data tracing method for computer room energy efficiency optimization is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide an air conditioning full-cycle matrix data tracing method for optimizing the energy efficiency of a computer room. To achieve the above purpose, the present invention provides the following technical solutions:

[0006] An air conditioning full-cycle matrix data tracing method for optimizing computer room energy efficiency includes:

[0007] Integrate the asset configuration data of each stage of the computer room and the operational efficiency data of each dimension to obtain a full-cycle data management matrix; the stages include the design stage, construction stage, commissioning stage and operation stage; the dimensions include the equipment dimension, system dimension, control dimension and energy efficiency dimension;

[0008] Based on the grey correlation analysis model, for a plurality of preset energy efficiency targets, correlation calculation is performed on the data elements in the full-cycle data management matrix, the influence weight of the data elements on the energy efficiency targets is quantified, and energy efficiency correlation matrices corresponding to the energy efficiency targets are generated to form an energy efficiency correlation matrix set;

[0009] When a preset energy efficiency anomaly is detected in the energy efficiency dimension of the operation phase, a matching energy efficiency association matrix is ​​selected from the energy efficiency association matrix set according to the type of the energy efficiency anomaly; and a reverse tracing is performed based on the selected energy efficiency association matrix to retrieve the association path with the highest influence weight, perform attribution analysis on the energy efficiency anomaly, and identify at least one key influencing factor causing the energy efficiency anomaly;

[0010] An energy efficiency optimization strategy is generated according to the key influencing factors, and energy efficiency optimization control is performed on the computer room based on the optimization strategy.

[0011] Preferably, the step of integrating the asset configuration data of each stage of the computer room and the operational efficiency data of each dimension to construct a full-cycle data management matrix includes:

[0012] Assign a unique identity code to the air-conditioning equipment in the computer room; perform feature extraction on the asset configuration data of each stage and the operational efficiency data of each dimension to obtain a full-cycle feature vector of the computer room; based on the unique identity code, perform feature fusion on the full-cycle feature vectors of the computer room corresponding to the same equipment obtained from each stage to obtain a fused feature vector, attach a timestamp identifier to the fused feature vector, and obtain a full-cycle data management matrix.

[0013] Preferably, the correlation calculation process of the grey correlation analysis model includes:

[0014] The data of preset energy efficiency targets at continuous time points are used as a reference sequence, and the energy efficiency targets include electricity utilization efficiency, energy efficiency ratio, chiller energy efficiency ratio, system water temperature difference and overall refrigeration performance coefficient; data elements at the same time points are extracted from the full-cycle data management matrix as a comparison sequence, and the comparison sequence is dimensionlessly processed; the difference sequence between each comparison sequence and the reference sequence at each time point is calculated, and the grey correlation coefficient is calculated in combination with the preset resolution coefficient; the grey correlation coefficients of all time points are arithmetic averaged to obtain a grey correlation degree, and the grey correlation degree is the influence weight of the data element on the energy efficiency target.

[0015] Preferably, the step of generating the energy efficiency correlation matrix includes:

[0016] The selected energy efficiency target is used as the dependent variable, the data elements in the full-cycle data management matrix are used as independent variables, the grey correlation degree between the dependent variable and the independent variable is calculated based on the grey correlation analysis model, and the energy efficiency correlation matrix is ​​constructed using the grey correlation degree.

[0017] Preferably, the step of reverse tracing includes:

[0018] S1. The data element corresponding to the abnormal energy efficiency indicator is used as the target node of the critical path; starting from the target node, the previous node with the greatest influence weight on the target node is found in the energy efficiency association matrix, and the previous node is added to the association path;

[0019] S2. Using the preceding node as the new current node, repeat the S1 process and iteratively backtrack until the marginal influence weight of the nodes on the path is less than the preset tracing threshold, thereby obtaining the highest weight associated path.

[0020] Preferably, there is a mapping relationship between the type of the energy efficiency anomaly and the energy efficiency target, and the energy efficiency anomaly monitoring step also includes classifying the energy efficiency anomaly; when it is monitored that the energy efficiency indicator deviates from the dynamic benchmark, the energy efficiency target level to which the energy efficiency indicator belongs is determined, and the corresponding energy efficiency association matrix is ​​matched.

[0021] Preferably, the attribution analysis step of the key influencing factors includes:

[0022] Using a pre-trained energy efficiency prediction model, the associated path data of the historical abnormal values ​​of the key influencing factors are input to obtain the energy efficiency of the first computer room; the energy efficiency prediction model predicts the energy efficiency of the computer room by learning the historical data in the full-cycle data management matrix; the energy efficiency of the computer room is the electric energy utilization efficiency value of the computer room; the historical abnormal values ​​of the key influencing factors are corrected, and the corrected historical abnormal values ​​are input into the energy efficiency prediction model to obtain the energy efficiency of the second computer room; based on the difference between the energy efficiency of the first computer room and the energy efficiency of the second computer room, the attribution confidence of the key influencing factors is calculated and obtained.

[0023] Preferably, the process of generating an optimization strategy based on the key influencing factors includes:

[0024] When the key influencing factors belong to the control dimension, a set of optimal control parameters is issued to the manager; when the key influencing factors belong to the equipment dimension, a predictive maintenance work order is generated for the manager; when the key influencing factors belong to the system dimension, an instruction is issued to the manager to perform the migration of IT load and balance the resource distribution; when the key influencing factors belong to the energy efficiency dimension, an instruction is issued to the manager to recalibrate the energy efficiency target and correct the energy efficiency accounting model; the strategies of each dimension are integrated to obtain an optimization strategy for energy efficiency. Through the above technical solutions, the present invention realizes the full-process energy efficiency optimization and data traceability of the university computer room air-conditioning system in the four stages of design, construction, commissioning and operation, effectively improving the system's energy-saving performance and reducing operation and maintenance costs.

[0025] Compared with the prior art, the present invention also has the following beneficial effects:

[0026] At the data processing level, a causal relationship spanning the entire computer room lifecycle is achieved. Unlike most current studies that only consider data from the operational phase, this method establishes a data matrix for the entire lifecycle and can correlate the energy consumption level of the system during the operational phase with the equipment selection during the design phase and the construction method during the construction phase. This addresses the critical bottleneck of being unable to trace problems due to missing historical data.

[0027] During the energy efficiency correlation matrix generation process, the complex relationship structure is formally described and visualized, forming a knowledge graph that can be easily parsed by machines. The results of the correlation calculation are organized neatly and orderly into a matrix. This matrix represents a quantitative causal network that clearly demonstrates the degree of association between any two or more variables within the system, providing the prerequisite for subsequent reverse tracing.

[0028] The accuracy and efficiency of attribution analysis are improved during the cause tracing process. Traditional energy efficiency troubleshooting relies heavily on expert experience, resulting in low efficiency and high subjectivity. This solution quantifies impact weights using a grey correlation model and uses this to automate reverse tracing, shifting from an experience-driven approach to a data-driven one, making the attribution process faster, more objective, and more accurate. The setting of tracking thresholds allows the tracking process to focus on primary issues while ignoring secondary influencing factors. This ensures algorithm convergence and concise results, effectively avoiding analytical stagnation.

[0029] In terms of energy efficiency optimization decision-making, it provides a basis for precise optimization decision-making: by identifying and locating key influencing factors, the optimization strategy is no longer a superficial adjustment, but an analysis and solution to the root cause of the problem, thereby significantly improving the effectiveness of energy efficiency optimization control and the return on investment. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1A flow chart of a method for tracing air conditioning full-cycle matrix data for optimizing energy efficiency in a computer room, proposed in an embodiment of the present invention;

[0031] Figure 2 This is a diagram of the energy efficiency anomaly attribution and optimization interaction process proposed in the embodiment of the application of the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0033] Example 1:

[0034] See also Figures 1 to 2 The present invention provides an air conditioning full-cycle matrix data tracing method for optimizing the energy efficiency of a computer room. The technical solution is as follows:

[0035] In order to make the purpose, technical solutions and advantages of the present invention more clear, a description will be given below with reference to specific embodiments.

[0036] This embodiment is applied to a data center of a large cloud computing company. The data center has a design capacity of 3,000 cabinets and uses a chilled water air conditioning system. The operation and maintenance team monitored that the average power utilization efficiency (PUE) value of the data center during the summer peak period was as high as 1.45, significantly exceeding the design target of 1.30. The operating costs were high, and multiple manual inspections failed to find the root cause. The full-cycle matrix data tracing method for optimizing the energy efficiency of the computer room proposed by the present invention is shown in the following figure. Figure 1 As follows, including:

[0037] All energy efficiency-related assets in the computer room, especially the components of the air conditioning system, are assigned a globally unique identification code (UUID). For example, the UUID of chiller unit 4 is HNSZC-01-CH-004.

[0038] Further feature extraction of multi-source heterogeneous data: For HNSZC-01-CH-004, extract its full-cycle feature vector of the computer room.

[0039] Specifically, in the design phase: extract key parameters from the PDF format design manual, such as the designed cooling capacity: 3200kW and the rated energy efficiency ratio COP: 6.2.

[0040] Construction phase: By analyzing the as-built BIM model drawings, it was determined that the actual length of the chilled water pipeline connected to HNSZC-01-CH-004 was 180 meters, exceeding the 165 meters in the design drawings.

[0041] Debugging phase: From the debugging report in Excel format, the full-load test current of HNSZC-01-CH-004 is extracted as 580A.

[0042] Operation phase: Real-time sensor data of HNSZC-01-CH-004 is collected from the air conditioning equipment in the computer room, such as real-time power and outlet water temperature.

[0043] Reducing PUE doesn't always equate to lower costs. For example, during off-peak hours with time-of-use electricity prices, pursuing extreme PUE may not be as cost-effective as appropriately relaxing controls to extend equipment life. In addition to the existing four dimensions, a financial dimension is introduced. Data elements in this dimension include: equipment procurement costs, installation and construction costs, spare parts prices, maintenance contract costs, real-time electricity prices, water unit prices, and labor costs. By upgrading the optimization goal from a single optimal energy efficiency to a comprehensive optimal cost-effectiveness, technical decisions are directly linked to business goals, achieving deeper global optimization.

[0044] Furthermore, based on the UUID: HNSZC-01-CH-004, the feature vectors describing the same device, obtained from each stage, are fused to form a comprehensive fused feature vector. A unified timestamp is attached to the dynamic data in this vector. After all asset data is aggregated, a multi-dimensional, full-cycle data management matrix M is formed (see Table 1).

[0045] Table 1 Multi-dimensional full-cycle data management matrix M

[0046] Design-Equipment Construction-Equipment Debug-Control Operation-Equipment Run-Control Operation-Energy Efficiency Run-System Chiller Chiller model, rated cooling capacity Installation date, serial number Chilled water outlet temperature setting, pressure protection threshold Chiller operating status, load rate Real-time water outlet temperature, real-time power Real-time COP Chiller circuit Precision air conditioning in computer rooms Air conditioner model, rated air volume Delivery date, asset number Return air temperature setting, linkage strategy Air conditioning operating status, fan speed Real-time return air temperature, valve opening PUE contribution value of a single machine Service Area Chilled water pump Pump model, rated flow Installation team, pipeline connection Start-stop pressure difference setting, frequency range Pump operating status and cumulative operating time Control mode, frequency Pump frequency Chilled water pump circuit Hot channel temperature sensor Accuracy, temperature range Physical location, terminal blocks Alarm threshold setting, data reporting frequency Communication status, battery level Real-time readings Data Validity Related assets

[0047] This solves the problems of data heterogeneity and ambiguity. The unique identification code (UUID) ensures that everything from a device model on a design drawing to a construction record to a data point in the operation monitoring system can be accurately linked to the same physical entity, effectively avoiding data confusion caused by inconsistent device names and numbers. It also establishes a standardized analytical foundation. The feature extraction and fusion process transforms unstructured and semi-structured data from various sources into a unified, fused feature vector that can be directly used by machine learning models, laying a solid data foundation for subsequent association analysis and algorithm application.

[0048] Furthermore, multiple core energy efficiency targets are preset, including: power utilization efficiency PUE, system water temperature difference ΔT, chiller energy efficiency ratio, etc.

[0049] Apply the Grey Relational Analysis (GRA) model: Taking PUE analysis as an example, generate a PUE energy efficiency correlation matrix.

[0050] Specifically, a reference sequence is set: the actual daily PUE values ​​within 30 days are selected as the reference sequence Y = {y1, y2, ..., y30}, for example, Y = {1.43, 1.45, ..., 1.46}.

[0051] Set the comparison sequence: Extract other data elements at the same time point from the full-cycle data management matrix M as the comparison sequence X. For example, X1 is the total IT load, X2 is the outdoor wet-bulb temperature, X3 is the real-time power of HNSZC-01-CH-004, and X4 is the preset COP value of HNSZC-01-CH-004. Format the comparison sequence as dimensionless.

[0052] The resolution coefficient is dynamically adjusted using a dynamic adjustment mechanism that sets default resolution coefficient values ​​for different data types. The default value 1 for the resolution coefficient ρ is set to ρ = 0.5, while the default values ​​2 and 3 are set to ρ = 0.25 and 0.75, respectively. The default value 1 is used for data with a balanced effect, the default value 2 emphasizes variability, and the default value 3 is suitable for data with noisy or volatile data. In cases of high data noise, the default value 3 automatically reduces sensitivity and avoids misattribution caused by noise. In cases where high discrimination is required, the default value 2 increases sensitivity to variability, making it easier to distinguish subtle differences between different comparison sequences and the reference sequence.

[0053] Calculate the absolute value of the difference between each comparison sequence and the reference sequence at each time point Find the maximum value among all difference sequences and minimum value The calculation formula of grey relational coefficient is:

[0054] Grey relational degree is the correlation coefficient of the comparison sequence at all time points It comprehensively reflects the overall correlation strength between the comparison sequence and the reference sequence.

[0055] Calculate the grey correlation degree between each comparison sequence and the reference sequence, i.e. the influence weight.

[0056] This method provides a robust and computationally efficient mathematical tool for quantifying data correlations. Grey correlation analysis requires a small sample size and places no strict restrictions on data distribution, making it well-suited for complex systems where data may be incomplete or irregular. This offers significant advantages over traditional statistical regression or deep learning models, which require large amounts of initial data. The method is precise and interpretable: It details the entire calculation process from raw data to final impact weights, making the results generation process clear and transparent, enhancing the credibility and interpretability of the technical solution.

[0057] Furthermore, a correlation matrix is ​​generated: the calculation results may be shown in Table 2 below, which is a partial example of a PUE energy efficiency correlation matrix:

[0058] Table 2 Partial examples of PUE energy efficiency correlation matrix

[0059] Data element / independent variable Impact weight on PUE / dependent variable Dimension Stage Total IT load 0.93 system run Outdoor wet-bulb temperature 0.89 system run Total power of chiller 0.87 equipment run System chilled water temperature difference ΔT 0.75 Energy efficiency run Actual length of HNSZC-01-CH-004 pipeline 0.68 equipment construction HNSZC-01-CH-004 Design COP 0.55 equipment design

[0060] The scattered correlation calculation results are organized into a neat matrix. This matrix is ​​essentially a quantitative "causal relationship network" that intuitively shows the degree of interaction between any two or more variables in the system and is the basis for automated reverse tracing.

[0061] Furthermore, the system detected a PUE value of 1.48, which consistently exceeded the dynamic baseline of 1.32, identifying an energy efficiency anomaly at the PUE level. Based on this anomaly type, the system automatically selected a PUE energy efficiency correlation matrix from the energy efficiency correlation matrix set for tracing.

[0062] The system monitored PUE, which reached a historical low of 0.92 over a period of time, significantly lower than the dynamic baseline of 1.32. This identified energy efficiency highlights at the PUE level. Success Path Tracing: When a highlight is triggered, reverse tracing is initiated to uncover key contributing factors. In addition to detecting anomalies where energy efficiency deviates from the dynamic baseline, it also detects highlights where energy efficiency significantly exceeds the baseline. This includes highlight detection and successful path tracing capabilities. By discovering and replicating best practices, we can drive a spiral of energy efficiency improvements in the data center, achieving exceptional performance levels unattainable by traditional optimization methods.

[0063] Different energy efficiency issues, such as excessively high overall PUE and reduced chiller COP, have distinct underlying causes and networks of contributing factors. By pre-establishing the mapping between anomalies and correlation matrices, we can ensure that the most appropriate analytical model is used for specific issues, avoiding the need to use a single traditional model to analyze all issues, significantly improving attribution accuracy.

[0064] Specifically, perform reverse tracing:

[0065] S1: Starting with PUE as the target node, find the associated node with the highest influence weight in the PUE energy efficiency association matrix E, which is the chiller total power weight of 0.87. Add the associated path. The association path is from the chiller total power to PUE.

[0066] S2: Using the total chiller power as a new node, the team screened for proactively adjustable association paths based on controllable factors (i.e., factors that managers and automation systems can manipulate and adjust to change the computer room status): system chilled water temperature difference ΔT (weight 0.80) and chiller load factor (weight 0.76). The strongest association was found to be system chilled water temperature difference ΔT (weight 0.81). The association path ran from system chilled water temperature difference ΔT to total chiller power, and then to PUE. Using controllable factors as screening criteria ensures actionable results, avoiding the focus on a highly weighted, unchangeable factor and instead directly targeting operational and maintenance tasks that can be immediately executed. This also avoids tracing back and analyzing invalid paths, allowing the system to converge more quickly on the solvable root cause of the problem, saving the operations team significant time and effort.

[0067] Continuing to trace back, it was found that the key factor affecting ΔT was the total water supply flow weight of 0.79, and one of the factors affecting the total water supply flow was the pipeline resistance weight of the loop where HNSZC-01-CH-004 was located, which was 0.72. Finally, it was traced back to the actual length weight of the HNSZC-01-CH-004 pipeline, which was 0.68.

[0068] Using a greedy algorithm, the node with the highest weight is selected each time to quickly identify the most likely primary cause chain, preventing aimless searching within a large amount of data relationships. Setting a tracing threshold focuses the tracing process on key issues and ignores non-critical factors, ensuring algorithm convergence and concise results, effectively avoiding analytical stagnation.

[0069] Furthermore, key influencing factors were identified: The traceability threshold was set at 0.65. When tracing back to the actual length of pipeline HNSZC-01-CH-004, its weight of 0.68 exceeded the threshold, but since it was source data, tracing was stopped. The resulting highest-weighted association path indicated that the actual length of pipeline HNSZC-01-CH-004 during the construction phase was a key factor contributing to the low system ΔT, which in turn increased the power consumption of the water pumps and chillers, and was a key factor contributing to the low PUE.

[0070] Furthermore, the attribution confidence is calculated:

[0071] Specifically, PUE is predicted based on the LSTM deep learning model.

[0072] Input 1: The model inputs historical data of the associated path, including the actual pipe length of 180 m. The predicted energy efficiency of the first computer room is PUE = 1.47.

[0073] Input 2: Correct the actual pipe length to the design value of 165m and input the model again. The predicted energy efficiency of the second computer room is PUE = 1.41.

[0074] Based on the significant difference between the first energy efficiency and the second energy efficiency, the system calculated that the attribution confidence of this key influencing factor is 94%.

[0075] Before implementing costly corrective measures, this simulation validation step confirms that the identified key influencing factors are indeed the core of the problem. A highly reliable result gives managers greater confidence in taking action, effectively avoiding the waste of resources caused by misattribution. This enables "what-if" analysis: essentially a hypothetical inference method, providing data support for loss assessment and optimization benefit prediction.

[0076] Furthermore, in terms of equipment and systems, regarding the key influencing factor HNSZC-01-CH-004, the actual length of the pipeline: This factor is a legacy issue from the construction phase and is difficult to change in the short term. The energy efficiency optimization problem of the computer room is modeled, and the states, executable actions, and rewards that the AI ​​can understand are defined. The deep Q network model is trained in a safe simulation environment, allowing the AI ​​to learn through massive trial and error, and master the strategies for taking the most energy-efficient actions under various conditions. The trained AI model is deployed to the actual computer room. It analyzes the computer room status in real time, automatically generates and issues the optimal control instructions to the air-conditioning room, and the system generates a predictive maintenance work order for the manager with the subject "The pipe resistance of the No. 4 chiller loop is too large. It is recommended to evaluate the feasibility of pipeline optimization and modification."

[0077] In terms of control, for the total water supply flow that is directly affected: as a short-term response, the system sends a set of optimal control parameters to the manager: it is recommended to increase the opening of the linkage valve of loop 4 by 5%, and to increase the upper limit of the operating frequency of the variable frequency water pump in this loop to compensate for the flow loss caused by higher resistance.

[0078] Integration strategy: The system combines the long-term piping modification recommendations and short-term control parameter adjustment strategies to create a comprehensive energy efficiency optimization strategy for the HVAC room. The strategy clearly demonstrates the root causes of the problems, data evidence, and specific implementation recommendations.

[0079] Unlike systems that only provide vague suggestions, this solution generates clear, actionable instructions based on the core characteristics of the problem. For example, it provides specific instructions on how to proceed, significantly improving the efficiency and execution of operations and maintenance work.

[0080] Achieved multi-faceted collaborative optimization: This method understands that computer room energy efficiency is a systematic project with multiple factors interacting with each other. The strategies it provides include equipment operation and maintenance, operational control, and IT management, demonstrating the concept of comprehensive optimization. Compared with single-dimensional optimization measures, the effect is more significant.

[0081] Through the above implementation method, the data center not only solved the immediate problem of excessively high PUE, but more importantly, found a historical problem hidden in the construction phase, and achieved cross-cycle, in-depth energy efficiency problem diagnosis and optimization from the operation phase to the construction phase.

[0082] The present invention solves the problem of integrating and applying heterogeneous and unstructured data in the entire cycle of the computer room. By giving unique identifiers to equipment and using feature extraction and feature fusion technology, a traceable full-cycle data management matrix is ​​constructed, providing a high-quality data source for subsequent quantitative analysis. The complex correlation relationships are made explicit and structured, and an energy efficiency correlation matrix corresponding to specific energy efficiency targets is generated, and the fuzzy, multi-dimensional influence relationship network is converted into a clear, queryable weight matrix. An automated and standardized root cause tracing path search algorithm is provided. This method can automatically and quickly locate the most likely causal chain in a complex correlation network, avoiding the blindness and inefficiency of manual investigation. The setting of the tracing threshold ensures the depth of tracing and the effective use of computing resources. It greatly simplifies the complexity of problem diagnosis and improves the pertinence of analysis.

[0083] Example 2:

[0084] This embodiment uses the scenario of a financial data center where a large temperature difference between cabinet rows occurs, resulting in redundant cooling capacity of the air-conditioning system and reduced overall energy efficiency.

[0085] The data center uses row-based air conditioning. The monitoring system found that while the overall PUE remained around 1.45, the overall cooling coefficient of performance was declining. Furthermore, the inlet air temperature consistency across multiple rows of cabinets was poor. In particular, a localized hotspot appeared at the top of the server cabinets in row A, reaching 32°C, far above the set temperature of 24°C. This forced the air conditioning system to increase its overall cooling output, resulting in wasteful cooling.

[0086] Construct a full-cycle data management matrix M covering the design, construction, commissioning, and operation phases. In this example, special attention will be paid to data in the system and control dimensions.

[0087] System dimension data: server asset information, cabinet location, IT load, virtual machine distribution strategy, etc.

[0088] Control dimension data: fan speed setpoints for row-level air conditioners, chilled water valve openings, dynamic benchmarks for cabinet temperature sensors, etc.

[0089] Furthermore, the core of the anomaly in this case is the cooling efficiency, so the overall cooling coefficient of performance CCSF is selected as the main analysis target.

[0090] The correlation calculation process is:

[0091] Reference sequence: Extract the daily average CCSF value of the most recent month as the reference sequence Y.

[0092] Comparison sequence: Extract the corresponding IT load, the inlet / outlet air temperature of each cabinet, the row-level air conditioning operating parameters, etc. as the comparison sequence X.

[0093] Matrix construction: Generate CCSF energy efficiency correlation matrix E.

[0094] Factual data support: After calculation, some of the correlations are as follows:

[0095] The maximum temperature difference between CCSF and cabinet row level, GRG=0.93;

[0096] CCSF and average IT load of row A cabinets, GRG=0.89;

[0097] CCSF and the top temperature of cabinet A12 in row A, GRG=0.91;

[0098] The CPU utilization of CCSF and server SVR-A12-05 is GRG=0.94;

[0099] The fan speed of CCSF and row-based air conditioner ACU-A is GRG=0.85.

[0100] Furthermore, the system detects "CCSF drop" and "local hotspot in row A". The system classifies the energy efficiency anomaly and matches the energy efficiency correlation matrix E. Figure 2 Starting from the CCSF node, follow the highest weight of 0.93 to find the maximum temperature difference at the cabinet row level. This maximum temperature difference is associated with the top temperature of cabinet A12 in row A, which has a weight of 0.91. Continuing the process, this is associated with the CPU utilization of server SVR-A12-05, which has a weight of 0.94.

[0101] The highest-weighted association path is from CCSF to the maximum temperature difference at the cabinet row level, to the top temperature of cabinet A12 in row A, and then to the CPU utilization of SVR-A12-05.

[0102] Further, by identifying the key influencing factors, the end of the path points to the CPU utilization of SVR-A12-05, which is a typical system dimension problem, namely, uneven distribution of IT load.

[0103] Furthermore, the attribution confidence is calculated:

[0104] Specifically, the energy efficiency prediction model was fed historical data from the associated path, including persistently high CPU utilization on the SVR-A12-05 server. The predicted CCSF (Consumer-Level Service Factor) was 3.8, close to the observed value of 3.82. Within the input data, a simulation was performed by migrating some of the high-load virtual machines on the SVR-A12-05 server to other, less-loaded servers, reducing its CPU utilization to 60%. This data was then fed back into the model. The resulting CCSF was 4.5. This significant CCSF difference confirms that uneven IT load is the root cause of the local hotspots and the resulting decrease in system energy efficiency, with a 98% confidence level.

[0105] Furthermore, since key influencing factors are system-level, the system automatically issues instructions to the cloud management platform or data center administrator for optimization strategy generation and execution. Instruction content: Optimization Recommendation: Continuously high load detected on server SVR-A12-05 has caused local hotspots in cabinet row A, severely impacting overall cooling efficiency. Recommendation: Immediately perform IT load migration operations. Migrate the two high-computing virtual machines (VM-Compute-01 and VM-Compute-02) running on this server to servers SVR-C08-11 and SVR-D03-07, whose current loads are less than 50%, to achieve balanced resource distribution.

[0106] Execution and Closed-Loop: Following instructions, the IT administrator performed the virtual machine migration using technologies such as vMotion. After the migration, the CPU utilization and cabinet temperature of the SVR-A12-05 dropped rapidly, eliminating the hotspot in row A. The row-based air conditioner, ACU-A, automatically reduced its fan speed and cooling output. A week later, system monitoring showed that the CCSF had returned to 4.6, eliminating the hotspot without increasing energy consumption and improving the overall energy efficiency of the computer room.

[0107] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for tracing air conditioning full-cycle matrix data for optimizing energy efficiency in a computer room, characterized by: include: Integrate the asset configuration data of each stage of the computer room and the operational efficiency data of each dimension to obtain a full-cycle data management matrix; the stages include design, construction, commissioning and operation; the dimensions include equipment, system, control and energy efficiency dimensions; Based on the grey correlation analysis model, for the preset energy efficiency target, the data elements in the full-cycle data management matrix are correlated, the influence weight of the data elements on the energy efficiency target is quantified, and an energy efficiency correlation matrix corresponding to the energy efficiency target is generated to form an energy efficiency correlation matrix set; When a preset energy efficiency anomaly is detected in the energy efficiency dimension of the operation phase, a matching energy efficiency association matrix is ​​selected from the energy efficiency association matrix set according to the type of the energy efficiency anomaly; and reverse tracing is performed based on the selected energy efficiency association matrix to retrieve the association path with the highest impact weight, and an attribution analysis is performed on the energy efficiency anomaly. The pre-trained energy efficiency prediction model is used to input the association path data of the historical anomaly values ​​of the key influencing factors to obtain the energy efficiency of the first computer room; the energy efficiency prediction model predicts the energy efficiency of the computer room by learning the historical data in the full-cycle data management matrix; the energy efficiency of the computer room is the electric energy utilization efficiency value of the computer room; Correcting the historical abnormal values ​​of the key influencing factors, and inputting the corrected historical abnormal values ​​into the energy efficiency prediction model to obtain the energy efficiency of the second computer room; Calculate and obtain the attribution confidence of the key influencing factor based on the difference between the energy efficiency of the first computer room and the energy efficiency of the second computer room; Identify the key factors that lead to energy efficiency anomalies; The reverse tracing step includes: S1. The data element corresponding to the abnormal energy efficiency indicator is used as the target node of the critical path; starting from the target node, the previous node with the greatest influence weight on the target node is found in the energy efficiency association matrix, and the previous node is added to the association path; S2. Using the previous node as the new current node, repeat the S1 process, iteratively backtracking until the marginal influence weight of the nodes on the path is less than the preset tracing threshold, and the highest weighted associated path is obtained; Based on the key influencing factors, a deep Q network model is used to generate control instructions for the computer room to perform energy efficiency optimization control.

2. The air conditioning full-cycle matrix data tracing method for optimizing the energy efficiency of a computer room according to claim 1 is characterized in that: The steps of integrating the asset configuration data of each stage of the computer room and the operational efficiency data of each dimension to build a full-cycle data management matrix include: The building information model data of the design phase, the as-built drawings of the construction phase, and the test data of the commissioning phase are acquired to form the asset configuration data. At the same time, the equipment status, environmental parameters, and energy consumption readings of the operation phase are collected in real time through IoT sensors to form the operational energy efficiency data. Global asset identifiers are assigned to the physical and logical assets in the computer room. Based on the global asset identifiers, the asset configuration data belonging to different phases are associated and integrated with the operational energy efficiency data of different dimensions to construct a full-cycle data archive exclusive to each asset, and the full-cycle data archive is structured and arranged. The physical and logical assets are used as matrix rows, the combination of each phase and each dimension is used as matrix columns, and the data values ​​and time series characteristics are used as elements to generate the full-cycle data management matrix.

3. The air conditioning full cycle matrix data tracing method for optimizing the energy efficiency of a computer room according to claim 1 is characterized in that: The correlation calculation process of the grey correlation analysis model includes: The data of preset energy efficiency targets at continuous time points are used as a reference sequence, and the energy efficiency targets include electricity utilization efficiency, energy efficiency ratio, chiller energy efficiency ratio, system water temperature difference and overall refrigeration performance coefficient; data elements at the same time points are extracted from the full-cycle data management matrix as a comparison sequence, and the comparison sequence is dimensionlessly processed; the difference sequence between each comparison sequence and the reference sequence at each time point is calculated, and the grey correlation coefficient is calculated in combination with the preset resolution coefficient; the grey correlation coefficients of all time points are arithmetic averaged to obtain a grey correlation degree, and the grey correlation degree is the influence weight of the data element on the energy efficiency target.

4. The air conditioning full cycle matrix data tracing method for optimizing the energy efficiency of a computer room according to claim 1 is characterized in that: The step of generating the energy efficiency correlation matrix includes: The selected energy efficiency target is used as the dependent variable, the data elements in the full-cycle data management matrix are used as independent variables, the grey correlation degree between the dependent variable and the independent variable is calculated based on the grey correlation analysis model, and the energy efficiency correlation matrix is ​​constructed using the grey correlation degree.

5. The air conditioning full cycle matrix data tracing method for optimizing the energy efficiency of a computer room according to claim 1 is characterized in that: There is a mapping relationship between the type of the energy efficiency anomaly and the energy efficiency target, and the energy efficiency anomaly monitoring step also includes classifying the energy efficiency anomaly; when it is monitored that the energy efficiency indicator deviates from the dynamic benchmark, the energy efficiency target level to which the energy efficiency indicator belongs is determined, and the corresponding energy efficiency association matrix is ​​matched.

6. The air conditioning full cycle matrix data tracing method for optimizing the energy efficiency of a computer room according to claim 1 is characterized in that: The process of generating an optimization strategy based on the key influencing factors includes: The energy efficiency optimization problem of the computer room is modeled, and the deep Q network model is trained. By learning from the historical operation data of the computer room, an energy efficiency optimization strategy based on key influencing factors is output. When the key influencing factors belong to the control dimension, a set of optimal control parameters are issued to the manager; when the key influencing factors belong to the equipment dimension, a predictive maintenance work order is generated for the manager; when the key influencing factors belong to the system dimension, an instruction is issued to the manager to execute the migration of IT load and balance resource distribution; when the key influencing factors belong to the energy efficiency dimension, an instruction is issued to the manager to recalibrate the energy efficiency target and correct the energy efficiency accounting model; the strategies of each dimension are integrated to obtain an energy efficiency optimization strategy.

Images