Central air conditioning operation abnormality monitoring method based on digital twinning

Abstract

The application discloses a central air conditioner operation abnormality monitoring method based on digital twinning, relates to the technical field of air conditioner monitoring, and realizes abnormality perception and false judgment elimination by constructing a digital twinning body, establishing a dynamic efficiency evaluation baseline based on working condition clustering and transient process identification, establishing a gradual factor parameter library, screening to-be-evaluated factors by matching induced characteristics, utilizing the digital twinning body to perform orthogonal experiment simulation, constructing a response surface model and solving an inverse problem, calculating the specific contribution weight of each gradual factor to current energy efficiency abnormality, and realizing accurate decoupling and quantitative root cause positioning of the coupling influence of multiple implicit and cumulative deterioration factors. A long-period natural degradation benchmark curve is established, and statistical significance testing of a short-term performance trend is combined, so that early differentiation and early warning of a sudden abnormality drop and a natural aging process of equipment are realized. The defects that a traditional method cannot trace back to cross-domain faults, cannot quantify gradual coupling influence, and cannot misjudge dynamic processes are overcome.

Description

Technical Field

[0001] This invention relates to the field of air conditioning monitoring technology, and in particular to a method for monitoring abnormal operation of central air conditioning systems based on digital twins. Background Technology

[0002] With the expansion of modern building scale and the increasing demands for indoor environmental quality, central air conditioning systems have become core energy-consuming equipment in large public buildings. Traditional methods for monitoring anomalies in central air conditioning systems mainly rely on fixed threshold alarms, periodic manual inspections, or simple model analysis based on a single physical domain (such as purely electrical or purely thermal), which have significant limitations when dealing with complex systems. Digital twins, on the other hand, are virtual models of physical entities created digitally. Through data connections and interactions, they enable real-time mapping, monitoring, prediction, and optimization of physical entities, representing a technology that provides a two-way dynamic association between physical entities and virtual models.

[0003] In practical applications, a single physical domain anomaly can trigger a chain reaction across domains, ultimately presenting as a complex set of symptoms. Traditional methods can only observe the final symptoms but cannot trace the cross-physical domain propagation chain of the fault or the original root cause, resulting in a lack of targeted maintenance.

[0004] Central air conditioning systems contain numerous static, slowly accumulating factors that cause performance degradation, such as heat exchanger scaling, minor refrigerant leaks, and sensor drift. While these factors may appear unhealthy individually and within acceptable thresholds, their combined effects can lead to significant energy efficiency decline. Current technologies lack the ability to effectively isolate and quantitatively attribute these latent, interconnected faults. Existing methods based on fixed operating conditions or static thresholds are prone to misinterpreting normal dynamic adjustments as abnormalities or masking early signs of faults during transient processes, resulting in false alarms and missed alarms. Summary of the Invention

[0005] This application provides a digital twin-based method for monitoring abnormal operation of central air conditioning systems. It solves the problems in existing technologies, such as the inability to effectively trace the fault propagation path across physical domains and the difficulty in quantifying the coupling effects of multiple implicit factors. This method enables multi-physical coupling analysis of cross-domain implicit gradual change factors, thereby achieving the technical effect of accurate anomaly location.

[0006] This application provides a method for monitoring abnormal operation of central air conditioning systems based on digital twins, including:

[0007] S1: Collect the physical parameter set of the central air conditioning system and construct a digital twin, extract the operating condition feature vector and perform clustering to automatically divide it into several operating condition clusters; calculate the actual energy efficiency ratio under different time health conditions in each operating condition cluster to form a normal efficiency baseline, identify extreme transient events, and establish a transient process efficiency line.

[0008] S2: Construct a gradient factor parameter library; obtain the operating energy efficiency ratio, operating condition feature vector and physical parameter set during anomalies; perform feature matching based on the parameter features of the physical parameter set and the induced data features, and mark the gradient factors corresponding to the induced data features with a matching degree greater than 85% as factors to be evaluated;

[0009] S3: Generate the baseline simulation energy efficiency ratio and the actual simulation energy efficiency ratio for the factors to be evaluated, and output the combination of parameters to be evaluated and the corresponding changes in energy efficiency ratio;

[0010] S4: For each gradient factor in the parameter combination to be evaluated, calculate the local sensitivity and the optimal estimation coefficient; use the product of the local sensitivity and the optimal estimation coefficient as the absolute contribution; calculate the contribution weight of each gradient factor based on the absolute contribution; generate the fault propagation path based on the contribution weight.

[0011] The gradual change factor refers to a performance parameter that is a static physical process and can continuously accumulate degradation.

[0012] Further, step S1 includes: S11: In real-time monitoring, determine whether the current situation is within the defined transient process event window; if yes, use the transient process performance line for judgment; if no, enter the working condition discrimination process, perform clustering, and judge with the normal performance baseline; if the judgment is abnormal, trigger an abnormal flag.

[0013] S12: While triggering the anomaly flag, quantify and record the severity of the anomaly, calculate the difference between the current measured real-time operating energy efficiency ratio and the median value of the energy efficiency ratio of the normal efficiency baseline of the corresponding operating condition cluster, and record it as the degree of energy efficiency deviation.

[0014] Furthermore, the operating condition feature vector includes the total load rate of the central air conditioning system, the comprehensive outdoor ambient temperature, and the coded equipment operating combination mode; the coded equipment operating combination mode is a discrete classification feature, using a multi-dimensional vector to represent the actual operating equipment in the current central air conditioning system, used to characterize the equipment operating status; the normal efficiency baseline refers to the actual energy efficiency level and fluctuation range exhibited when it is under the current operating condition during its historical healthy period.

[0015] Furthermore, the method also includes: S5: setting a long period, filtering out steady-state operating condition cluster data based on the physical parameter set, and calculating the corresponding standard energy efficiency ratio; arranging the standard energy efficiency ratios calculated in different periods in chronological order to form a historical trend sequence, and then fitting it into a natural degradation baseline curve;

[0016] S6: Obtain the new standard energy efficiency ratio sample value under the operating condition and compare it with the predicted value of the natural degradation baseline curve at the current moment to calculate the standardized offset value; set a short period, analyze the local change trend of the standard energy efficiency ratio within the short period, and calculate the short-term slope; count the average number of times the standard offset value exceeds the deviation threshold in the historical abnormal performance inflection point data within the long period. If the number of times the currently monitored standardized offset value exceeds the preset deviation threshold is greater than the average number, and the deviation between the short-term slope and the long-term natural degradation slope is greater than the slope threshold, then mark the abnormal performance inflection point; determine the final cause of the failure based on the natural degradation baseline curve and the abnormal performance inflection point.

[0017] Furthermore, establishing a transient process performance line includes: identifying the start and end times of transient events from historical data; aligning all identified historical data segments of similar events with the start time as the reference point; on the aligned time axis, for each relative time point t, collecting the actual operating energy efficiency ratios of all historical events at the relative time point, and calculating the 5th and 95th percentiles of the actual operating energy efficiency ratio set as upper and lower limits; connecting the upper and lower limits of all relative time points t to form a transient process performance line; the relative time point is the offset time with the start of the transient event as the zero point of time.

[0018] Furthermore, for the factors to be evaluated, a baseline simulation energy efficiency ratio and an actual simulation energy efficiency ratio are generated, including: setting the influence coefficients of all gradual factors to 0 in the digital twin and running the simulation to obtain the baseline simulation energy efficiency ratio; for the factors to be evaluated, several discrete disturbance levels are randomly defined within a predetermined value range, and orthogonal experiments are performed to determine several sets of parameter combinations to be evaluated, and simulation is performed on each set of parameter combinations to obtain the corresponding actual simulation energy efficiency ratio.

[0019] Furthermore, a gradient factor parameter library is constructed, including: extracting gradient factors based on historical data, and associating each gradient factor with a set of induced data features, wherein the induced data features refer to the data change features of the theoretically associated physical quantities caused by changes in the gradient factor parameters.

[0020] Further, the calculation of local sensitivity and optimal estimation coefficients includes: constructing a response surface function to calculate local sensitivity; establishing an objective function, setting an optimization algorithm to solve for the optimal combination of gradient factor parameters, and obtaining the optimal estimation coefficients for each gradient factor; the local sensitivity refers to the rate of change of the digital twin output value relative to a certain input parameter when a unit change occurs at a specific parameter point.

[0021] Furthermore, the benchmark simulation energy efficiency ratio is obtained by running the simulation, including: loading the simulation with boundary conditions completely consistent with the current central air conditioning system in the digital twin, wherein the boundary conditions include environmental boundary conditions, load boundary conditions, system operating setpoint and topology and state initialization.

[0022] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0023] By constructing a digital twin and integrating geometric models, multi-physics domain simulation models, and data fusion models, real-time dynamic mapping of the central air conditioning physical system is achieved. The multi-physics domain simulation covers fluid networks, heat transfer, mechanical rotation, and circuit models. Through bidirectional coupling simulation of cross-domain interaction, it solves the problem that traditional methods only focus on a single parameter and ignore coupling effects. The operating condition feature vector clustering technology is introduced to automatically divide the operating data into typical operating condition clusters and establish targeted normal performance baselines and transient process performance lines, effectively distinguishing steady-state and transient events and improving monitoring adaptability.

[0024] By using a gradient factor parameter library and feature matching mechanism, latent fault parameters are identified, and contribution weights are calculated using digital twin simulation to generate fault propagation paths, accurately locating the root cause of multi-factor coupled faults and overcoming the one-sidedness of traditional diagnosis.

[0025] Long-term trend analysis distinguishes between natural aging and abnormal deterioration of equipment by comparing the natural degradation baseline curve and the short-term slope, enabling early warning. Through multi-dimensional data fusion and simulation optimization, it significantly reduces the false alarm rate and improves energy efficiency. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of a method for monitoring abnormal operation of a central air conditioning system based on digital twins, as described in an embodiment of the present invention. Detailed Implementation

[0027] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] Example 1: As Figure 1 As shown, a method for monitoring abnormal operation of a central air conditioning system based on digital twins is provided, the method comprising:

[0030] S1: Collect the physical parameter set of the central air conditioning system and construct a digital twin, extract the operating condition feature vector and perform clustering to automatically divide it into several operating condition clusters; calculate the actual energy efficiency ratio under different time health conditions in each operating condition cluster to form a normal efficiency baseline, identify extreme transient events, and establish a transient process efficiency line.

[0031] In some embodiments, high-precision, high-frequency sensors are pre-deployed in the refrigeration unit, water pump, cooling tower, air handling unit, and key pipeline nodes of the central air conditioning system to collect raw physical parameters across physical domains in real time, such as temperature, pressure, flow rate, vibration, current harmonics, and valve opening, forming a raw data stream. The raw data stream is then cleaned, aligned, and timestamped to generate a time-space aligned normalized state dataset as a set of physical parameters.

[0032] A digital twin is constructed based on a set of physical parameters. This digital twin is a complete digital entity consisting of three layers: a geometric model, a multiphysics simulation model, and a data fusion model. The geometric model is a 3D visualization model built based on CAD / BIM, providing a static visual representation of spatial structure and component relationships. The multiphysics simulation model layer is a computable model composed of mathematical equations, physical laws, and algorithms, used to simulate the dynamic interactive behavior of the air conditioning system in multiple physical domains such as thermodynamics, fluid mechanics, mechanical dynamics, and electrical control. The data fusion continuously acquires physical data through a sensor network, driving the simulation model's operation, and continuously calibrates model parameters through machine learning algorithms to ensure synchronization and predictive consistency between the virtual simulation and the physical entity's state.

[0033] Specifically, the multiphysics simulation model includes a fluid network model, a heat transfer model, a mechanical rotation model, and a circuit model; the fluid network model is a reduced-order model of computational fluid dynamics and pipe network hydraulics. The air conditioning water system (chilled water, cooling water) and refrigerant circulation system are abstracted as a topological network composed of nodes (equipment interfaces, branch points) and branches (pipes, valves); based on the mass conservation equation and momentum conservation equation; a flow balance equation is established for each node, and a pressure drop balance equation is established for each loop; the hydraulic impedance coefficient is used in the model to characterize the parameters of fluid resistance of pipes, valves, filters, etc., calibrated using historical data; the performance curves of pumps / fans are used to describe the flow-head (air volume-air pressure) relationship in polynomial form; a real-time query function for temperature-pressure-density-enthalpy is established; the flow distribution, pressure distribution, and velocity changes of each branch of the central air conditioning system are calculated, and problems such as insufficient flow, hydraulic imbalance, and abnormal refrigerant charge are identified.

[0034] The heat transfer model is a distributed parameter model based on the three fundamental laws of heat transfer (conduction, convection, and radiation). For heat exchangers (evaporators and condensers), the efficiency-number of heat transfer units method is used to establish the heat transfer calculation equation. For air ducts and rooms, a two-dimensional unsteady-state heat balance equation is established, considering the convective heat transfer between air and walls, the heat conduction of the building envelope, and the heat generation of indoor heat sources. The heat transfer coefficient and air property coefficient are determined, and the heat transfer of each heat exchanger, the refrigerant phase change process, and the room temperature field distribution are calculated. Anomalies such as decreased heat transfer efficiency, insulation failure, and load mismatch are identified.

[0035] The mechanical rotation model focuses on the dynamic characteristics of rotating machinery (compressors, fans, pumps); it establishes the torque balance equation of rotating components based on Newton's second law; it simplifies the rotating machinery into a mass-spring-damped system and establishes a multi-degree-of-freedom vibration equation to analyze the characteristic frequency response of faults such as bearing wear and impeller imbalance; it simulates the volumetric efficiency changes of compressors, the performance curve deviation of fans / pumps, the generation and propagation of vibration signals, and identifies faults such as mechanical wear, dynamic balance failure, and misalignment.

[0036] The circuit model, based on circuit theory and electrical machinery, describes the behavior of electrical drive and control systems, calculates the real-time input power, efficiency, and power factor of the motor, analyzes harmonic distortion rate, simulates the impact of inverter failures on central air conditioning, and identifies problems such as decreased electrical insulation, deteriorated power quality, and abnormal control signals.

[0037] In practical applications, overall simulation is achieved through strong bidirectional coupling. Examples include: fluid-heat transfer coupling: the fluid model provides velocity and flow rate boundary conditions for the heat transfer model; the heat transfer model provides temperature distribution for the fluid model, thus affecting fluid properties (viscosity, density) and buoyancy effects; and mechanical-electrical coupling: the electromagnetic torque of the motor drives mechanical rotation; changes in mechanical load affect the motor's current and power factor, thereby altering the operating state of the circuit model. Using co-simulation technology, each physical domain model solves within its own time step, exchanging data and iterating at coupling boundaries (such as power, flow rate, and temperature) until the entire model converges.

[0038] In existing technologies, anomalies in a single physical domain of a central air conditioning system can trigger chain reactions in other domains, ultimately presenting as complex symptoms. Traditional monitoring only observes the final symptoms (such as a decrease in energy efficiency ratio) and cannot trace the cross-domain propagation path and the original root cause. This embodiment constructs a digital twin by using a multi-physical parameter set to achieve multi-physical coupling, accurately determine the cross-domain propagation path and the original root cause, and accurately trace the source.

[0039] In some embodiments, operating condition feature vectors are extracted based on a set of physical parameters and automatically divided into several operating condition clusters. Physical parameters include, but are not limited to, outdoor dry-bulb temperature, outdoor wet-bulb temperature, indoor temperature in each area, chilled water supply and return water temperature, cooling water supply and return water temperature, chilled water flow rate, cooling water flow rate, chilled water pump frequency, cooling water pump frequency, cooling tower fan frequency, number of operating compressors and their load percentage, and total electrical power of the central air conditioning system.

[0040] The operating condition feature vector includes the total load rate of the central air conditioning system, the comprehensive outdoor ambient temperature, and the coded equipment operating combination mode. The total load rate of the central air conditioning system is a dimensionless percentage value, reflecting the current cooling output of the system relative to its maximum capacity. The comprehensive outdoor ambient temperature is generated by weighted summation of outdoor dry-bulb temperature and outdoor wet-bulb temperature, maintaining a weight sum of 1. The specific weighting needs to be set according to the building type and system characteristics. Initially, the weight of outdoor dry-bulb temperature can be set to 0.3, and the weight of outdoor wet-bulb temperature to 0.7 to highlight the importance of wet-bulb temperature. The weights can be adjusted based on actual operating statistics. The coded equipment operating combination mode is a discrete classification feature, using a multi-dimensional vector to represent the actual operating equipment in the system. For example, a multi-dimensional vector can be used to represent: [number of operating chilled water pumps, number of operating cooling towers, compressor load rate], used to characterize the equipment operating status and represent the topology and control strategy state of the central air conditioning system. For example, the efficiency of a single large pump operating is different from that of two small pumps operating in parallel when delivering the same flow rate; the heat dissipation power consumption and effect are also different between one cooling tower operating at full speed and two cooling towers operating at half speed. The optimal energy efficiency baseline for a system will inevitably differ depending on the combination of operating equipment. By coding the equipment operating combination mode, the normal efficiency baseline can be adjusted in real time according to the current specific equipment operating status.

[0041] Based on the operating condition feature vectors and combined with unsupervised clustering algorithms, historical operating data is automatically divided into several typical operating condition clusters. Each typical operating condition cluster is a set of historical data points with highly similar operating condition feature vectors. For example, cluster C1 represents a high-temperature, high-load single-pump operation mode, containing all data points from historical data where the overall outdoor ambient temperature is greater than 32 degrees Celsius, the total load rate is greater than 80%, and a single main pump is used. Through automatic clustering analysis, an independent and accurate performance baseline is established for each scenario, further improving the accuracy and efficiency of anomaly monitoring. Similarly, comparing the operating energy efficiency ratio at different times is only meaningful under highly similar operating conditions (within the same cluster). Comparing operating energy efficiency ratios under different weather conditions and loads introduces significant noise, making it impossible to detect true performance degradation.

[0042] In some embodiments, the DBSCAN (Density-Based Spatial Clustering with Noise) algorithm is used for clustering, automatically discovering clusters of arbitrary shapes and identifying noise points that do not belong to any cluster (i.e., rare, atypical transitional operating conditions). The preprocessed feature vector set is input into the DBSCAN algorithm. The algorithm outputs: each data point is assigned a cluster label (C1, C2, ..., Cn); some points are marked as noise, and for noise points, they are associated with the nearest cluster center or temporarily excluded from baseline establishment. Finally, several operating condition clusters are obtained. The statistical characteristics of the original physical parameters of the data points within each cluster are analyzed to give each cluster a clear meaning, for example: Cluster C3: Spring medium-load dual-pump operation condition.

[0043] In some embodiments, the actual operating energy efficiency ratio under different time health conditions is calculated within each operating condition cluster to form a normal efficiency baseline. The normal efficiency baseline refers to the actual energy efficiency level and fluctuation range exhibited when the data is under the current operating condition during a historical health period. Specifically, points whose timestamps are within the health state (assessed based on actual operating data and expert experience) are selected from all historical data points belonging to cluster Ci.

[0044] Calculate the instantaneous actual operating energy efficiency ratio for each data point based on the actual measurement data corresponding to the timestamp: ,

[0045] in, The actual cooling capacity refers to the amount of heat actually removed from the building by the central air conditioning system per unit time, measured in kilowatts. Total input power refers to the total active power consumed by all power-consuming equipment in the central air conditioning system (including the chiller, chilled water pump, cooling water pump, cooling tower fan, etc.) at the same time. The unit is kilowatt, and it is obtained based on real-time measurement.

[0046] Actual cooling capacity The calculation formula is as follows: ,

[0047] in, It is the density of water; It is the specific heat capacity of water; It is the cold water flow rate; It is the chilled water return temperature, the chilled water temperature that returns from the building's end units (such as fan coil units and air conditioning units) to the central air conditioning unit; This refers to the chilled water supply temperature, the temperature of the chilled water supplied from the central air conditioning unit to the building terminals. In practical applications, the return water temperature should be higher than the supply water temperature. If a negative value appears, it is considered abnormal, and there is no need to calculate the actual cooling capacity.

[0048] The actual operating energy efficiency ratios of all healthy data points within cluster Ci are treated as a sample set. The median, 10th percentile (EERp10, where 10% of the values ​​in the dataset are less than or equal to this value), and 90th percentile (EERp90, where 90% of the values ​​in the dataset are less than or equal to this value) of this set are calculated. This constitutes the normal efficiency range [EERp10, EERp90]. This range covers 80% of healthy operating energy efficiency performance under this condition, defining the normal boundary. The median represents the most typical energy efficiency level under this condition and is insensitive to extreme values. Percentiles are a statistical method describing data distribution, used to avoid the mean being affected by extreme values ​​and more robustly reflecting the actual fluctuation range of most healthy data.

[0049] In some embodiments, extreme transient events are identified. These extreme transient events refer to events in which the system's operating state undergoes drastic and non-steady-state changes. A change threshold is pre-set for each air conditioner operating parameter, and the rate of change of each parameter within a sampling period is identified. If the absolute value of the rate of change is greater than the threshold, it is determined that a drastic change has occurred; if the air conditioner operating parameter does not enter a predefined steady-state judgment interval within several consecutive sampling periods, it is determined to be a non-steady-state change. The change threshold is set based on the parameter's historical steady-state statistical characteristics and physical response boundaries, or in conjunction with the equipment manufacturer's recommended values, maintenance manuals, or historical fault cases. These are all well-known technologies, and this application does not impose specific limitations on them. The sampling period is set according to the different physical characteristics and operating parameters of the equipment, and is less than the minimum dynamic response time of the system to ensure the observability of the dynamic process. For example, the adjustment response time of equipment such as compressors and water pumps is generally between 15 seconds and 2 minutes, and the sampling period is less than the minimum response time, such as 10 seconds. Several consecutive sampling periods should be greater than the maximum expected number of periods of random disturbance to filter out accidental fluctuations. The interval is set to [3, 10] sampling periods. In this embodiment, it is set to 5 sampling periods. The steady state judgment interval is defined as the range of 3 times the standard deviation of the parameter mean under healthy steady state. Extreme transient events include: start-up process: from the state of all equipment stopped, the cooling tower, water pump and host are started in sequence until the water temperature reaches the set value and the room temperature begins to drop, which lasts for 20-45 minutes; sudden change in load: such as the sudden end of a meeting in a large conference room and the rapid departure of people, which causes the load to drop by more than 50% within 10 minutes; or the sudden start of large equipment, which causes a surge in indoor heat load; equipment switching process: such as switching from one pump to another, or the transition process of adding or removing load from the host.

[0050] In some embodiments, establishing a transient process performance line includes: identifying the start and end times of the transient event from historical data; aligning all identified historical data segments of similar events with the start time as the reference point; on the aligned time axis, for each relative time point t, collecting the actual operating energy efficiency ratios of all historical events at the relative time point, and calculating the 5th percentile and 95th percentile of the actual operating energy efficiency ratio set as upper and lower limits; connecting the upper and lower limits of all relative time points t to form a transient process performance line; wherein the relative time point is the offset time with the start of the transient event as the zero point of time. Specifically, based on historical data, including equipment start / stop signals and load rate changes, the start and end times of transient events are automatically identified (defined as when key parameters such as supply and return water temperature difference re-enter a stable fluctuation range). Historical data segments of all identified similar events (such as all chiller start-up processes) are aligned using the start time as a reference point. On the aligned timeline, for each relative time point t (e.g., the 5th minute after start-up), the actual operating energy efficiency ratio (EERP) of all historical events at that relative time point is collected, and the 5th percentile (EERp5) and 95th percentile (EERp95) of the actual operating EERP set are used as upper and lower limits. At relative time point t, the reasonable upper and lower limits of the actual operating EERP fluctuation are [EERp5, EERp95]. Connecting the upper and lower limits of all relative time points t forms the transient process efficiency line for this type of event, which is a dynamic range with time as the independent variable. The relative time point is the offset time from the start of the transient event as the zero point (e.g., the time when the start-up command is issued), in minutes (min).

[0051] Based on the content of this embodiment, step S1 includes: S11: In real-time monitoring, determine whether the current state is within the defined transient process event window; if yes, use the transient process performance line for judgment; if no, enter the operating condition discrimination process, perform clustering, and judge with the normal performance baseline; if the judgment is abnormal, trigger an abnormal flag. This solves the problem of mistakenly judging normal dynamic processes as abnormal.

[0052] When a central air conditioning system is running in real time, it first determines its operating condition cluster Ci based on the current set of physical parameters. It then checks whether the current time point falls within a defined time window of an extreme transient process (such as a start-up process). If a device start-up / stop signal or load change signal is detected, and the current time is within a preset duration after the signal is triggered (set based on statistical analysis of the duration of transient events in historical data, such as the average duration of events), it is determined to be a transient process. The system then compares the currently measured real-time operating energy efficiency ratio (EER). If the current state falls into a typical operating condition cluster, it is compared with the corresponding normal efficiency baseline to determine whether the currently measured real-time EER is within the normal EER range defined by that baseline.

[0053] If the current state is determined to be in an extreme transient process, it is compared with the transient process efficiency line corresponding to this type of transient process to determine whether the measured real-time operating energy efficiency ratio falls between the reasonable upper and lower limits of fluctuation defined by the transient process efficiency line at the current moment (the time point counted from the start of the transient process).

[0054] An anomaly flag is triggered based on the comparison results. If the currently measured real-time operating energy efficiency ratio continuously (e.g., for more than 5 consecutive sampling periods or stably exceeds the range for 10 minutes) falls outside the corresponding normal energy efficiency ratio range or the reasonable fluctuation range of the transient process, it is determined to be a valid anomaly, and the efficiency deviation anomaly flag is triggered and its status is set to true.

[0055] S12: Upon triggering the anomaly flag, quantify and record the severity of the anomaly, calculate the difference between the currently measured real-time operating energy efficiency ratio and the median energy efficiency ratio of the corresponding operating condition cluster's normal efficiency baseline, and record this as the degree of energy efficiency deviation. The degree of energy efficiency deviation is a positive or negative value; its absolute value reflects the extent to which the current efficiency deviates from the historical healthy normal, and its sign indicates the direction of deviation (e.g., a positive value indicates energy efficiency better than the historical median, and a negative value indicates energy efficiency worse than the historical median).

[0056] Preferably, after the transient process ends, a smooth transition to a certain steady-state operating condition cluster Ci is recorded, and the result is verified using a normal performance baseline. If the real-time operating efficiency ratio remains abnormal in the steady state after the transient process ends, it indicates that the transient process caused a persistent problem, requiring continued monitoring and reporting of the anomaly.

[0057] In this embodiment, a high-precision sensor network is deployed to collect a set of physical parameters, constructing a digital twin composed of a geometric model, a multi-physics domain simulation model, and a data fusion model. Operating condition feature vectors are extracted, and a clustering algorithm is used to divide historical data into operating condition clusters. Within each cluster, a normal performance baseline is established based on healthy period data. Extreme transient events are identified, and the performance lines of the transient process are aligned with historical data to form a dynamic fluctuation range. In real-time monitoring, it is first determined whether the system is within a transient event window, and the corresponding performance line is selected for comparison; otherwise, the normal baseline is matched according to the operating condition cluster, and the degree of energy efficiency deviation is quantified when an anomaly flag is triggered.

[0058] This technology addresses the problems of misjudgment caused by fixed thresholds, inability to adapt to dynamic events, and noise introduced by changes in operating conditions in existing technologies. It achieves accurate simulation across multiple physical domains through digital twins, avoiding missed detection of cross-domain coupled faults; operating condition clustering and transient line design improve monitoring adaptability and reduce false alarms; real-time comparison and quantification of deviations lay the foundation for root cause analysis, thus improving the overall accuracy of anomaly identification.

[0059] Example 2: Example 1 established a precise foundation for anomaly monitoring through digital twins and operational condition clustering. However, for complex systems with multiple intertwined factors, such as latent faults caused by gradual changes like scaling and leakage, the accuracy in identifying the degree of impact is low. Therefore, this example makes further improvements based on the above.

[0060] The method further includes: S2: constructing a gradient factor parameter library; obtaining the operating energy efficiency ratio, operating condition feature vector and physical parameter set during anomalies; performing feature matching based on the parameter features of the physical parameter set and the induced data features, and marking the gradient factors corresponding to the induced data features with a matching degree greater than 85% as factors to be evaluated.

[0061] In some embodiments, a gradient factor parameter library is constructed, including: extracting gradient factors based on historical data, and associating each gradient factor with a set of induced data features. The induced data features refer to the data change characteristics of the theoretically associated physical quantities caused by changes in the gradient factor parameters. The gradient factor refers to a performance parameter whose physical state is a static process and can continuously accumulate degradation. Its relationship with the fault cannot be accurately assessed relative to the initial state. Each gradient factor includes the following elements: Baseline state (0 point): clearly defining the physical state corresponding to when the factor value is 0, representing the ideal design state of the equipment or system being brand new, clean, leak-free, wear-free, and without characteristic drift; Influence scale, used to linearly or non-linearly characterize the severity of degradation; Example of a linear scale: the refrigerant leakage factor ranges from [0, -0.3]. A refrigerant leakage factor of -0.15 indicates that 15% of the total refrigerant has been lost, with the negative sign indicating the direction of loss. Example of nonlinear scaling: The scaling factor ranges from [0,1]. A scaling factor of 0.25 does not directly represent that 25% of the area is blocked, but rather that the heat transfer coefficient caused by scaling has decreased to (1-0.25)=75% of the clean state.

[0062] Different gradual change factors correspond to their respective physical mapping models. In the digital twin, the corresponding physical parameters are modified based on the current value of the factor. For the fouling factor x, its effect is to modify the overall heat transfer coefficient U of the heat exchanger. The mapping model is: U current = U design * (1 - ks * x), where ks is a known constant reflecting the sensitivity of this type of heat exchanger to fouling (for example, ks = 0.8, which means that when x = 1, U can drop to a maximum of 20% of the design value).

[0063] Based on the changes in the value of the scaling factor, the changing trends of theoretically related physical quantities (such as temperature, pressure, current, and vibration) can be identified. For example, an increase in the scaling factor on the condenser will result in a systematic upward shift of the curve corresponding to the condensing temperature and condensing pressure, and a gradual increase in the condenser terminal temperature difference (condensing temperature - cooling water outlet temperature).

[0064] In practical applications, the corresponding gradient factor can be selected according to the characteristics of the above-mentioned gradient factors and actual needs. In this embodiment, the gradient factors include heat exchanger fouling factor, refrigerant micro-leakage factor, resistance gain factor, sensor drift factor, and regulating valve characteristic offset factor. The heat exchanger fouling factor represents the heat transfer efficiency reduction rate caused by fouling (0 for clean, 1 for complete blockage). The refrigerant micro-leakage factor represents the proportion of refrigerant charge loss relative to the rated value (e.g., 0 for no leakage, -0.1 for a 10% loss). The resistance gain factor includes two types of resistance gain factors for water systems and air systems, representing the percentage of additional flow resistance caused by filter blockage, pipe corrosion, etc. (0 for design resistance, >0 for increased resistance). The sensor drift factor represents the measurement value offset of a specific sensor (e.g., return air temperature sensor). The regulating valve characteristic offset factor represents the degree of nonlinear deviation between the actual valve opening and the commanded opening (e.g., 0 for normal characteristics, 0.2 for an actual opening of only 40% under a 50% opening command). In the gradient factor parameter library, each gradient factor is associated with a set of induced data features. The induced data features refer to the data change characteristics of the theoretically associated physical quantities caused by the change of the gradient factor parameters. For example, the heat exchanger fouling factor is associated with the slow shift in the correspondence between condensing temperature and condensing pressure.

[0065] S3: Generate the baseline simulation energy efficiency ratio and the actual simulation energy efficiency ratio for the factors to be evaluated, and output the combination of parameters to be evaluated and the corresponding changes in energy efficiency ratio.

[0066] Specifically, the process of obtaining the baseline simulation energy efficiency ratio (EER) includes: running a simulation in a digital twin with boundary conditions identical to the current central air conditioning system, including environmental boundary conditions, load boundary conditions, system operating setpoint, and topology and state initialization. For the factors to be evaluated, baseline and actual simulation EERs are generated, including: setting the influence coefficients of all gradual change factors to 0 in the digital twin and running the simulation to obtain the baseline simulation EER; for the factors to be evaluated, randomly defining several discrete disturbance levels within a predetermined value range and performing orthogonal experiments to determine several sets of parameter combinations to be evaluated; simulating each set of parameter combinations to obtain the corresponding actual simulation EER; and outputting the parameter combinations to be evaluated and the corresponding changes in the EER.

[0067] In some embodiments, boundary conditions identical to those of the current central air conditioning system are loaded into the digital twin, and the influence coefficients of all gradient factors are set to 0 (i.e., ideal state). The simulation yields a baseline simulation energy efficiency ratio. Under ideal conditions without gradient factors, the theoretical energy efficiency ratio that the system should have under the current operating conditions is obtained, serving as a benchmark for subsequent comparative analysis. The real-time operating state of the current physical system is completely mapped to the digital twin. Boundary conditions include environmental boundary conditions, load boundary conditions, system operating setpoints, and topology and state initialization. The environmental boundary conditions include at least outdoor dry-bulb temperature, outdoor wet-bulb temperature, and solar radiation intensity. The load boundary conditions are the total building cooling load calculated from the building energy consumption model or obtained through actual measurement, or the real-time total cooling capacity demand summarized from terminal data. The system operating setpoints include at least chilled water supply temperature setpoints, cooling water return temperature setpoints, and control commands for each major device. The topology and state initialization confirms the start / stop status of all pumps, fans, valves, compressors, and other equipment, and sets them consistently in the digital twin. Setting the influence coefficients of all gradient factors to 0 indicates simulation calculation under completely ideal conditions. The simulation process is set up according to the actual experimental requirements and is a standard simulation method, so it will not be described in detail.

[0068] In some embodiments, for the factor to be evaluated, several discrete perturbation levels are randomly defined within a predetermined value range to perform orthogonal experiments to determine several combinations of parameters to be evaluated. The value range is set based on historical engineering experience and absolute physical limits (theoretical upper and lower limits). For example, a fouling factor of 0 indicates no fouling, and the heat transfer coefficient is the design value. The upper limit is determined based on the complete blockage of the heat exchange tube or the fouling thermal resistance reaching a critical value that causes heat transfer to stagnate. According to engineering experience, for water-cooled condensers, the fouling thermal resistance will not exceed 0.0005 m²·K / W, which can be converted into a heat transfer coefficient reduction rate, such as a theoretical upper limit of 0.5 (indicating a 50% reduction in heat transfer capacity). Several discrete perturbation levels are randomly defined within a reasonable value range for each factor to be evaluated.

[0069] Discrete perturbation level refers to uniformly selecting a finite number of specific numerical points within the range of the gradual factor. For example, [0, 0.4] can be divided into 5 levels: {0.0, 0.1, 0.2, 0.3, 0.4}. The number of levels can range from 3 to 5. The more levels there are, the more refined the characterization of the factor's influence, but the number of combinations will increase exponentially.

[0070] Orthogonal experimental design is a method for scientifically arranging multi-factor, multi-level experiments based on orthogonal arrays. It can efficiently evaluate the main effects of each factor on the index and the interaction effects between factors with the fewest number of trials (parameter combinations), significantly reducing simulation computation while ensuring the validity of the analysis. Specific selection criteria are as follows: Balance: In the generated parameter combination set, each level of each factor appears exactly the same number of times. Orthogonality: Any combination of different levels of any two factors appears the same number of times in all combinations. Computational efficiency: The number of trials is much smaller than the total number of combinations. Information completeness: Based on the data obtained from the orthogonal array, the significance and contribution rate of each factor to the energy efficiency ratio are quantified using analysis of variance. Orthogonal experiments are a well-known experimental method, and will not be elaborated upon further in this application.

[0071] In some embodiments, several sets of parameter combinations to be evaluated are determined based on orthogonal experimental results and selection criteria. Simulations are performed on each set of parameter combinations to obtain the corresponding actual simulated energy efficiency ratio (EER). The baseline simulated EER and the actual simulated EER are compared to obtain the EER change, and finally, the parameter combinations to be evaluated and their corresponding EER changes are output.

[0072] S4: For each gradient factor in the parameter combination to be evaluated, calculate the local sensitivity and the optimal estimation coefficient; use the product of the local sensitivity and the optimal estimation coefficient as the absolute contribution; calculate the contribution weight of each gradient factor based on the absolute contribution; generate the fault propagation path based on the contribution weight.

[0073] The calculation of local sensitivity and optimal estimation coefficients includes: constructing a response surface function to calculate local sensitivity; establishing an objective function, setting an optimization algorithm to solve for the optimal combination of gradient factor parameters, and obtaining the optimal estimation coefficients for each gradient factor; the local sensitivity refers to the rate of change of the digital twin output value relative to a certain input parameter at a specific parameter point when a unit change occurs. Specifically, for each gradient factor in the parameter combination to be evaluated, a response surface function is constructed based on multinomial regression to calculate local sensitivity; an objective function and constraints are established, and an optimization algorithm is set to solve for the optimal combination of gradient factor parameters to obtain the optimal estimation coefficients for each gradient factor; the product of the local sensitivity and the optimal estimation coefficients is used as the absolute contribution; the absolute value of the absolute contribution is normalized to obtain the contribution weight of each gradient factor; and a fault propagation path is generated based on the contribution weights.

[0074] In some embodiments, the response surface function uses a polynomial containing first-order and second-order terms as its basic structure, and each gradient factor in the combination of parameters to be evaluated is denoted as a vector. The response surface function represents the predicted change in energy efficiency ratio, and the formula is as follows: ,

[0075] in, This is a constant term. When all factors are 0 (ideal state), the predicted change in energy efficiency ratio should theoretically be 0. It is the coefficient of the first-order term, representing the factor. The intensity of the linear effect on changes in energy efficiency ratio; It is the coefficient of the quadratic term, representing the factor. The intensity of the nonlinear influence of the change in energy efficiency ratio. Based on the combination of parameters to be evaluated output in step S3 and the corresponding change in energy efficiency ratio, the coefficients of the first term and the coefficients of the second term are solved by the least squares method.

[0076] The local sensitivity refers to the rate of change of the digital twin output value relative to a certain input parameter when a small unit change occurs at a specific parameter point (reference point, such as F=0). It reflects the severity of the impact of a single input parameter on the output near that point. In the early stages of a fault, the influence of each factor is relatively small. Calculating the local sensitivity at the reference point can quickly assess the initial order of strength of the impact of a unit change in each factor on energy efficiency.

[0077] Calculate based on response surface function Partial derivative at F=0: Therefore, the coefficient of the first-order term represents the local sensitivity of the i-th gradient factor.

[0078] An objective function and constraints are established, and an optimization algorithm is set up to solve for the optimal combination of gradient factor parameters, obtaining the optimal estimated coefficients for each gradient factor. Using the known gradient factors and energy efficiency changes, the most likely input combination leading to a specific output result is inferred, with the least squares method used as the matching criterion. A set of specific values ​​for the gradient factors is found such that the difference between the energy efficiency ratio change predicted by the response surface function based on these values ​​and the actual observed abnormal change in energy efficiency ratio is minimized. These specific values ​​of the gradient factors are the estimates of the severity of each factor that best explain the current anomaly.

[0079] Using the physically reasonable upper limit of each gradient factor as a constraint, the objective function is set as follows: ,

[0080] in, It is the objective function, which is the square of the difference between the predicted value and the measured value, and measures the degree of matching. This refers to the degree of energy efficiency deviation. The estimated severity of the i-th gradient factor is obtained based on the objective function. This is an estimate, taking values ​​(0, 1], representing the optimal estimation coefficient, indicating the minimum severity of the gradient factor under the current fault condition. The objective function is solved iteratively. In each iteration, the objective function value and its gradient (the change in the objective function compared to the previous iteration, such as an increase or decrease) are calculated, and the search direction is determined according to the algorithm rules. A line search is performed to determine the step size, the value of the gradient factor is updated, and it is determined whether the convergence condition is met (such as a sufficiently small gradient or the upper limit of the number of iterations). After convergence, the optimal solution is output.

[0081] The product of local sensitivity and optimal estimation coefficient is taken as the absolute contribution, representing the estimated contribution of the gradual factor to energy efficiency change. It is usually negative, indicating a decrease in energy efficiency. The contribution weight of each gradual factor is obtained by normalizing the absolute value of the contribution.

[0082]

[0083] in, It is the contribution weight of the i-th gradient factor; It is the absolute value of the absolute contribution of the i-th gradual change factor, representing its influence magnitude; It is the optimal estimated coefficient of the i-th gradient factor.

[0084] Fault propagation paths are generated based on contribution weights. The analysis outputs triplet information, including the name of the gradient factor, the optimal estimation coefficient, and the contribution weight. For example, for the condenser scaling factor: the optimal estimation coefficient (severity) is 0.18, and the contribution weight is 65%. Combined with the simulation state of the digital twin under the corresponding gradient factor parameters, fault propagation paths across physical domains are generated synchronously.

[0085] In this embodiment, a gradual factor parameter library is constructed based on historical data. Each factor is associated with a set of induced data features, characterizing its theoretical physical quantity change trend. When an anomaly occurs, the real-time operating energy efficiency ratio, operating condition characteristics, and physical parameter set are acquired. Factors with a matching degree higher than 85% are marked as factors to be evaluated through feature matching. The current boundary conditions are loaded into the digital twin, and the influence coefficients of all factors are set to zero to simulate the baseline energy efficiency ratio. For the factors to be evaluated, discrete perturbations are performed within the value range, and orthogonal experiments are conducted to obtain parameter combinations. The actual energy efficiency ratio change is simulated. The local sensitivity of each factor is calculated through response surface functions, and the optimal estimated coefficients are solved by combining optimization algorithms. The product of the sensitivity and the estimated coefficient is taken as the absolute contribution, normalized to obtain the contribution weight, and a fault propagation path is generated.

[0086] It solves the problems of existing technologies being unable to quantify the impact of multiple failure factors and identify coupled root causes. Through digital twin simulation, it accurately recreates failure scenarios, avoiding reliance on experience-based judgment; contribution weights clearly define the responsibility ratio of each factor, supporting targeted maintenance; and it visualizes the cross-domain propagation chain of failure paths, enhancing diagnostic depth.

[0087] Example 3: Over several years of operation, the performance degradation of a central air conditioning system may be due to a combination of normal aging and sudden malfunctions. If only instantaneous anomalies are considered, progressive degradation signals may be easily overlooked. This example further improves upon the above.

[0088] The method further includes: S5: setting a long period, filtering out steady-state operating condition cluster data based on the physical parameter set, and calculating the corresponding standard energy efficiency ratio; arranging the standard energy efficiency ratios calculated in different periods in chronological order to form a historical trend sequence, and then fitting it into a natural degradation baseline curve.

[0089] In some embodiments, the long-term period is set based on the theoretical usage time of the actual equipment, collecting historical operating data for no less than 24 months as the initial long-term data for modeling; in the initial stage of operation, no less than 12 months. The performance of central air conditioning is significantly affected by the seasons. 24 months of data ensures that at least two complete cooling / heating season cycles are included, thereby eliminating the interference of seasonal fluctuations on long-term trend analysis and extracting performance degradation signals purely driven by equipment aging.

[0090] The natural degradation baseline curve is a slowly declining linear or exponential curve, and its downward slope represents the average annual rate of energy efficiency ratio degradation caused by the natural aging of the equipment under normal maintenance conditions.

[0091] S6: Obtain the new standard energy efficiency ratio sample value under the operating condition and compare it with the predicted value of the natural degradation baseline curve at the current moment to calculate the standardized offset value; set a short period, analyze the local change trend of the standard energy efficiency ratio within the short period, and calculate the short-term slope; count the average number of times the standard offset value exceeds the deviation threshold in the historical abnormal performance inflection point data within the long period. If the number of times the currently monitored standardized offset value exceeds the preset deviation threshold is greater than the average number, and the deviation between the short-term slope and the long-term natural degradation slope is greater than the slope threshold, then mark the abnormal performance inflection point; determine the final cause of the failure based on the natural degradation baseline curve and the abnormal performance inflection point.

[0092] In some embodiments, the short cycle is set to 7 to 30 days, with an initial value of 14 days. The short cycle duration is dynamically adjusted based on the accuracy of the trend analysis results; if the accuracy is low, the short cycle duration is increased. The short cycle needs to be sensitive to detect abrupt changes that deviate from the long-term natural aging trend. The occurrence and development of faults or imbalances will be reflected in performance data within days to weeks. A cycle that is too short (e.g., <7 days) is susceptible to random noise and daily fluctuations; a cycle that is too long (e.g., >30 days) will lead to delayed response and miss early warning windows.

[0093] The analysis of local trends in Standard Energy Efficiency Ratio (CEER) over a short period includes: analyzing a series of CEER data points collected within the short period using linear regression. The CEER samples arranged chronologically within the short period are considered the dependent variable, and the corresponding sampling time points (which can be converted into the number of days from the start of the period) are considered the independent variables. The least squares method is applied to fit a straight line to this dataset, resulting in a fitted straight line. The slope of this fitted line is the desired short-term slope, used to quantify the average daily (or per sampling interval) rate of change of the CEER within a short time window.

[0094] In some embodiments, the deviation threshold is used to determine whether a single sample deviates significantly from expectations. It is preset based on the control limit principle in statistical process control and the historical statistical characteristics of equipment performance fluctuations. For example, the standard deviation of the standard energy efficiency ratio in historical data is calculated. According to the 3σ criterion in statistics, for a normal distribution, approximately 99.7% of the data points fall within the range of the mean ± 3 times the standard deviation; therefore, the deviation threshold is set to 3.0. The slope threshold is used to determine whether the rate of performance degradation is abnormal, distinguishing between the natural aging rate and the abnormal accelerated degradation rate. Based on statistics estimated from limited historical data, statistical hypothesis testing is performed to estimate the uncertainty of the natural aging rate, thereby determining with high confidence whether the recently observed accelerated performance degradation is real and significant, rather than an illusion caused by estimation errors in the natural aging rate. Neither the deviation threshold nor the slope threshold is a fixed value; they need to be preset and dynamically adjusted based on real-time monitored equipment information and historical data. This application does not impose specific limitations on these values.

[0095] In some embodiments, the abnormal performance inflection point includes the following information: time information: the specific date and time when the inflection point is confirmed; equipment identification: the specific central air conditioning system number, service area, and associated main subsystem identification where the inflection point occurs; performance quantification data: the actual sampled value of the standard energy efficiency ratio that triggered the inflection point, the predicted value of the natural degradation baseline curve at the current moment, the standardized offset value, the short-term slope, and the long-term natural degradation slope.

[0096] If a clear contribution weight is calculated in the above embodiment, the diagnosis conclusion is: there is a hidden coupling fault, the main cause is micro-fouling of the condenser (contribution 60%), the secondary cause is slightly insufficient cooling water flow (contribution 40%), and a fault propagation path across physical domains is generated.

[0097] If no clear contribution weight is calculated, it indicates that the gradual factors are all in a normal state, but an anomaly of unknown cause still occurs. This indicates that the system performance has deviated from the natural aging trajectory and has an abnormal decline. Check the components that are loose, worn, or have drifted control parameters, and combine the natural degradation baseline curve and the abnormal performance inflection point to point out the parameter that has changed most significantly before and after the performance mutation (such as an increase in the harmonics of a certain pump current) as clues, and then determine the final cause of the failure.

[0098] The method described in the above embodiments ultimately generates a structured diagnostic report, which clearly lists: the type of anomaly (coupled fault / abnormal aging / normal transient); the root cause and contribution; the location of the anomaly; the quantified impact (percentage of energy efficiency loss); and maintenance priority recommendations.

[0099] In this embodiment, long-term steady-state operating condition data is collected, filtered, and the standard energy efficiency ratio is calculated. This data is then arranged in a time series and fitted to a natural degradation baseline curve to reflect the natural aging trend of the equipment. New sampled values ​​are acquired and compared with the curve's predicted values ​​to calculate the standardized offset value. A short-term period is set, and local slopes are analyzed using linear regression. If the offset value continuously exceeds the deviation threshold and the deviation between the short-term and long-term slopes exceeds the slope difference threshold, an abnormal performance inflection point is marked. Based on the inflection point and the baseline curve, natural aging and abnormal degradation are distinguished, and the final cause of failure is determined.

[0100] This technology addresses the shortcomings of existing technologies, such as the inability to distinguish between normal aging and sudden failures in long-term performance evolution, and the lack of trend warning capabilities. The natural degradation baseline curve provides an aging reference baseline, avoiding misjudging normal degradation as abnormality; short-term slope analysis sensitively captures performance abrupt changes, enabling early warning; inflection point markers combined with historical data support root cause tracing, such as identifying potential problems like loosening and wear.

[0101] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring abnormal operation of central air conditioning systems based on digital twins, characterized in that, include: S1: Collect the physical parameter set of the central air conditioning system and construct a digital twin, extract the operating condition feature vector and perform clustering to automatically divide it into several operating condition clusters; calculate the actual energy efficiency ratio under different time health conditions in each operating condition cluster to form a normal efficiency baseline, identify extreme transient events, and establish a transient process efficiency line. S2: Construct a gradient factor parameter library; obtain the operating energy efficiency ratio, operating condition feature vector and physical parameter set during anomalies; perform feature matching based on the parameter features of the physical parameter set and the induced data features, and mark the gradient factors corresponding to the induced data features with a matching degree greater than 85% as factors to be evaluated; S3: Generate the baseline simulation energy efficiency ratio and the actual simulation energy efficiency ratio for the factors to be evaluated, and output the combination of parameters to be evaluated and the corresponding changes in energy efficiency ratio; S4: For each gradient factor in the parameter combination to be evaluated, calculate the local sensitivity and the optimal estimation coefficient; use the product of the local sensitivity and the optimal estimation coefficient as the absolute contribution; calculate the contribution weight of each gradient factor based on the absolute contribution; generate the fault propagation path based on the contribution weight; the gradient factor refers to the performance parameter whose physical state is a static process and can continuously accumulate degradation. S5: Set a long period, filter out steady-state operating condition cluster data based on the physical parameter set, and calculate the corresponding standard energy efficiency ratio; arrange the standard energy efficiency ratios calculated in different periods in chronological order to form a historical trend sequence, and then fit it into a natural degradation baseline curve. S6: Obtain the new standard energy efficiency ratio sample value under the operating condition and compare it with the predicted value of the natural degradation baseline curve at the current moment to calculate the standardized offset value; set a short period, analyze the local change trend of the standard energy efficiency ratio within the short period, and calculate the short-term slope; count the average number of times the standardized offset value exceeds the deviation threshold in the historical abnormal performance inflection point data within the long period. If the number of times the currently monitored standardized offset value exceeds the preset deviation threshold is greater than the average number, and the deviation between the short-term slope and the long-term natural degradation slope is greater than the slope threshold, then mark the abnormal performance inflection point. The final cause of failure was determined based on the natural degradation baseline curve and abnormal performance inflection points. Step S1 includes: S11: In real-time monitoring, determine whether the current situation is within the defined transient process event window; if yes, use the transient process performance line for judgment; if no, enter the working condition discrimination process, perform clustering, and judge with the normal performance baseline; if the judgment is abnormal, trigger the abnormal flag. S12: While triggering the anomaly flag, quantify and record the severity of the anomaly, calculate the difference between the current measured real-time operating energy efficiency ratio and the median value of the energy efficiency ratio of the normal efficiency baseline of the corresponding operating condition cluster, and record it as the degree of energy efficiency deviation.

2. The method as described in claim 1, characterized in that, Establishing a transient process performance line includes: identifying the start and end times of transient events from historical data; aligning all identified historical data segments of similar events with the start time as the reference point; on the aligned time axis, for each relative time point t, collecting the actual operating energy efficiency ratios of all historical events at the relative time point, and calculating the 5th and 95th percentiles of the actual operating energy efficiency ratio set as upper and lower limits; connecting the upper and lower limits of all relative time points t to form a transient process performance line; the relative time points are offset times from the start of the transient event as the zero point.

3. The method as described in claim 1, characterized in that, For the factors to be evaluated, a baseline simulation energy efficiency ratio and an actual simulation energy efficiency ratio are generated, including: setting the influence coefficients of all gradual factors to 0 in the digital twin and running the simulation to obtain the baseline simulation energy efficiency ratio; for the factors to be evaluated, several discrete disturbance levels are randomly defined within a predetermined value range, and orthogonal experiments are performed to determine several sets of parameter combinations to be evaluated, and simulation is performed on each set of parameter combinations to obtain the corresponding actual simulation energy efficiency ratio.

4. The method as described in claim 1, characterized in that, Constructing a gradient factor parameter library includes: extracting gradient factors based on historical data, with each gradient factor associated with a set of induced data features, wherein the induced data features refer to the data change characteristics of the theoretically associated physical quantities caused by changes in the gradient factor parameters.

5. The method as described in claim 1, characterized in that, The calculation of local sensitivity and optimal estimation coefficients includes: constructing a response surface function to calculate local sensitivity; establishing an objective function, setting an optimization algorithm to solve for the optimal combination of gradient factor parameters, and obtaining the optimal estimation coefficients for each gradient factor; the local sensitivity refers to the rate of change of the digital twin output value relative to a certain input parameter when a unit change occurs at a specific parameter point.

6. The method as described in claim 5, characterized in that, The response surface function uses a polynomial containing first-order and second-order terms as its basic structure. Each gradient factor in the combination of parameters to be evaluated is denoted as a vector. The response surface function represents the predicted change in energy efficiency ratio, and the formula is as follows: ,in, It is a constant term; It is the coefficient of the first-order term, representing the factor. The intensity of the linear effect on changes in energy efficiency ratio; It is the coefficient of the quadratic term, representing the factor. The intensity of the nonlinear influence of changes in energy efficiency ratio.

7. The method as described in claim 1, characterized in that, The operating condition feature vector includes the total load rate of the central air conditioning system, the comprehensive outdoor ambient temperature, and the coded equipment operating combination mode; the coded equipment operating combination mode is a discrete classification feature, using a multi-dimensional vector to represent the actual operating equipment in the current central air conditioning system, used to characterize the equipment operating status; the normal efficiency baseline refers to the actual energy efficiency level and fluctuation range exhibited when it is under the current operating condition during its historical healthy period.

8. The method as described in claim 3, characterized in that, The benchmark simulation energy efficiency ratio is obtained by running the simulation, including loading boundary conditions completely consistent with the current central air conditioning system in the digital twin. The boundary conditions include environmental boundary conditions, load boundary conditions, system operating setpoint, and topology and state initialization.

Images