Cross analysis digital twinning intelligent operation and maintenance method and system based on artificial intelligence

Abstract

The invention relates to the technical field of host operation and maintenance, in particular to a cross analysis digital twin intelligent operation and maintenance method and system based on artificial intelligence. The method comprises the following steps that multi-mode sensing data and central air conditioner structure data are obtained, multi-mode air conditioner operation sensing is conducted, and multi-mode operation sensing data are obtained; carrying out feature weight dynamic assignment on the multi-modal operation sensing data to obtain a space-time interaction feature matrix; a real-time updating digital twinborn model is constructed based on the central air conditioner structure data; quantitatively evaluating the equipment degradation degree based on the real-time updated digital twin model to obtain a fault critical point prediction matrix; and performing double-layer game fault risk assessment on the fault critical point prediction matrix, and executing dynamic collaborative optimization of host operation parameters to obtain an intelligent operation and maintenance scheme of the central air conditioner. According to the invention, the operation and maintenance efficiency and the fault detection accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of host operation and maintenance technology, and in particular to an artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance method and system. Background Art

[0002] Traditional maintenance of central air conditioning systems relies primarily on regular manual inspections, which are inefficient and result in delayed fault detection. Due to long inspection cycles, maintenance personnel often struggle to detect potential equipment failures in a timely manner. This can lead to small problems accumulating into major failures, impacting system operation and equipment lifespan. Furthermore, simple sensor-based monitoring methods only capture limited operating parameters and fail to fully reflect equipment health, resulting in delayed and inaccurate fault detection.

[0003] Monitoring equipment operating data alone also has limitations. For example, relying solely on abnormal changes in operating parameters like temperature and pressure makes it difficult to accurately identify the type and severity of a fault. The same temperature anomaly could be caused by a malfunction in the cooling system, fan, or other components. Traditional empirical methods often lack intelligent analysis, making them prone to misjudgments or omissions, impacting the accuracy and timeliness of fault resolution.

[0004] Furthermore, central air conditioning systems involve multiple subsystems, including cooling, ventilation, and control. Data from each subsystem is often independent, lacking effective integration and collaborative analysis. This isolated data makes it difficult for operations personnel to comprehensively understand the system's operating status and quickly locate the root causes of cross-subsystem faults. This decentralized approach to operations and maintenance results in lengthy troubleshooting and repair times, further increasing the complexity and cost of system operations. Summary of the Invention

[0005] Based on this, it is necessary for the present invention to provide an artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance method and system to solve at least one of the above technical problems.

[0006] To achieve the above objectives, a cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence includes the following steps: Step S1: Acquire multimodal sensing data and central air-conditioning structure data, and perform multimodal air-conditioning operation perception based on the multimodal sensing data to obtain multimodal operation perception data; Step S2: Perform dynamic dimension slicing on the multimodal operation perception data to obtain perception dimension slice data; construct a spatiotemporal interaction map based on the perception dimension slice data, dynamically assign feature weights, and obtain a spatiotemporal interaction feature matrix; Step S3: constructing a central air-conditioning digital twin model based on the central air-conditioning structure data; inputting the spatiotemporal interaction feature matrix into the central air-conditioning digital twin model to perform cross-model update and obtain a real-time updated digital twin model; Step S4: Based on the real-time updated digital twin model, cross-abnormal pattern traceability is captured to obtain abnormal feature traceability vectors; time series fault trend modeling is performed on the abnormal feature traceability vectors to quantitatively evaluate the degree of equipment degradation and obtain a failure critical point prediction matrix; Step S5: Perform a double-layer game fault risk assessment on the fault critical point prediction matrix, and perform dynamic collaborative optimization of the host operating parameters to obtain an intelligent operation and maintenance plan for the central air conditioner.

[0007] Optionally, step S1 specifically includes: Step S11: acquiring multimodal sensor data and central air conditioning structure data, and performing data preprocessing on the multimodal sensor data to obtain multimodal sensor data to be analyzed; Step S12: dividing the multimodal sensor data to be analyzed into sensor dimensions to obtain environmental sensor data and central air conditioning sensor data; Step S13: performing central air-conditioning operating environment perception according to the environmental sensing data to obtain central air-conditioning operating environment perception data; Step S14: sensing the central air-conditioning operating state based on the central air-conditioning sensor data to obtain central air-conditioning state sensing data; Step S15: performing perception feature fusion on the central air-conditioning operating environment perception data and the central air-conditioning state perception data to obtain multimodal operation perception data.

[0008] Optionally, step S14 is specifically as follows: Step S141: performing hierarchical decoupling of operating parameters on the central air conditioning sensor data to obtain an operating parameter feature vector and an electrical parameter feature vector; Step S142: performing thermophysical parameter inversion on the operating parameter eigenvector based on a preset thermodynamic cycle physical model to obtain a thermal efficiency characteristic matrix; performing motor parameter inversion on the electrical parameter eigenvector based on a preset energy consumption load mapping model to obtain a motor power characteristic matrix; Step S143: performing graph convolutional network state association modeling based on the thermal efficiency feature matrix and the motor power feature matrix to obtain a central air conditioning state time series feature map; Step S144: performing central air conditioning abnormal state detection on the central air conditioning state time series characteristic map to obtain an abnormal state perception vector and a normal state perception vector; Step S145: performing central air-conditioning state time series merging based on the abnormal state perception vector and the normal state perception vector to obtain central air-conditioning state perception data.

[0009] Optionally, step S2 is specifically: Step S21: Dynamically slice the multimodal operation perception data, set the perception time series window length to [10 minutes, 30 minutes], and perform hierarchical time series slicing within a time step of [1 minute, 5 minutes] based on the preset sliding window [10 minutes, 30 minutes] to obtain perception dimension slice data; Step S22: construct a multi-dimensional time series feature map of the central air conditioner based on the perception dimension slice data, and perform temperature-current-wind speed-refrigerant pressure cross-correlation modeling to generate a spatiotemporal interaction map; Step S23: Perform adaptive attention dynamic weighting on the interaction strength between nodes in the spatiotemporal interaction graph to generate a dynamic spatiotemporal interaction weight matrix; Step S24: Perform thermodynamic cycle coupling modeling based on the central air conditioning structure data, set the thermodynamic cycle parameters [0°C, 100°C] and the airflow disturbance coefficient [0.01, 0.1], and perform physical coupling correction on the dynamic spatiotemporal interaction weight matrix based on the coupling modeling results to obtain a corrected spatiotemporal interaction characteristic matrix; Step S25: performing feature normalization processing on the modified spatiotemporal interaction feature matrix, and performing weight suppression of abnormal interaction regions between nodes to generate a spatiotemporal interaction feature matrix.

[0010] Optionally, step S25 is specifically as follows: Step S251: performing node normalization processing on the modified spatiotemporal interaction feature matrix, setting the normalization interval [0, 1], and performing normalization conversion based on the preset node timing window length [5min, 10min] and dynamic interaction frequency threshold [0.1, 0.5] to generate a standardized spatiotemporal interaction feature matrix; Step S252: Based on the abnormal state perception vector and the normal state perception vector, combined with the coupling modeling results, the nodes in the standardized spatiotemporal interaction feature matrix are detected for abnormal regions in a time series. If the interaction strength of a node exceeds the preset abnormal threshold [1.5σ, 3σ], the spatiotemporal trajectory of the node is marked to obtain a set of abnormally marked interaction region nodes; Step S253: Dynamically suppress the weight of the node set in the abnormally labeled interaction area. Combined with the preset time series smoothing coefficient [0.1, 0.3] and the node activity weight factor [0.5, 1.5], perform dimensionality reduction mapping of the abnormal node weight within the node time series window length [5min, 10min] to generate a suppressed spatiotemporal interaction feature matrix. Step S254: Based on the thermodynamic cycle physical model and the energy consumption load mapping model, cross-modal spatiotemporal feature fusion is performed on the spatiotemporal interaction suppression feature matrix to obtain the spatiotemporal interaction feature matrix.

[0011] Optionally, step S3 specifically includes: Step S31: extracting central air-conditioning operation features based on the multi-modal operation perception data, and identifying the air-conditioning operation mode to obtain central air-conditioning operation mode data; Step S32: extracting structural connection features based on the central air-conditioning structure data, and constructing a central air-conditioning digital twin model based on the structural connection features and the central air-conditioning operation mode data; Step S33: Input the spatiotemporal interaction feature matrix into the central air-conditioning digital twin initial model, and perform cross-mapping of thermal-electrical-environmental modal features to generate a central air-conditioning state mapping feature map; Step S34: Based on the central air conditioning state mapping feature map, dynamically align the physical state parameters with the environmental perception parameters to obtain a state parameter correction matrix; Step S35: Dynamically update the parameters of the central air-conditioning digital twin model according to the state parameter correction matrix, and perform adaptive timing correction of the twin model to obtain a real-time updated digital twin model.

[0012] Optionally, the cross-abnormal pattern tracing and capturing described in step S4 is specifically as follows: Based on the real-time updating of the digital twin model, the real-time thermal-electrical-environmental three-modal feature data of the central air conditioner is obtained, and the multi-modal feature time series hierarchical decoupling is performed to obtain the real-time thermal-electrical-environmental three-modal feature vector; Based on the real-time thermal-electrical-environmental trimodal feature vectors and combined with the abnormal state perception vector, a central air conditioning abnormal mode association graph is constructed. The cross-abnormal intensity of modal feature nodes is modeled to generate an abnormal mode cross-feature matrix. The abnormal factor traceability path is traced on the abnormal pattern cross-feature matrix. Based on the digital twin model's thermodynamic cycle parameters [20°C, 70°C], motor power parameters [0kW, 5kW], and environmental disturbance parameters [0, 0.5], the abnormal triggering factors are located and an abnormal traceability path diagram is generated. Based on the anomaly tracing path diagram, perform dynamic clustering of abnormal nodes, extract the anomaly trigger factor vector, and generate the anomaly trigger factor perception vector; The abnormal trigger factor perception vector and multimodal operation perception data are combined to perform layered traceability of abnormal patterns and obtain the abnormal feature traceability vector.

[0013] Optionally, the quantitative evaluation of the device degradation degree in step S4 is specifically: Based on the abnormal feature tracing vector, the abnormal pattern intensity time series response model is performed to obtain the abnormal pattern time series response curve; A central air conditioning operation degradation time series model is constructed based on the abnormal mode time series response curve, and an adaptive dynamic sliding window decomposition is performed on the central air conditioning operation degradation time series model to extract the degradation factor time series feature vector; Based on the degradation factor time series eigenvector, combined with the real-time update of the thermal, electrical, and environmental trimodal parameters in the digital twin model, the degradation critical threshold is set; The critical degradation threshold is used to predict the critical failure point by updating the digital twin model in real time, and the critical failure point prediction matrix is ​​obtained.

[0014] Optionally, the two-layer game failure risk assessment in step S5 is specifically as follows: Obtain historical maintenance record data and perform time series alignment on the historical maintenance record data according to the spatiotemporal interaction feature matrix to obtain the spatiotemporal maintenance interaction matrix; Combining the failure critical point prediction matrix and the spatiotemporal maintenance interaction matrix, a two-layer game model is constructed. The outer layer game subject is set as the equipment status, and the inner layer game subject is set as the maintenance record, thus obtaining the equipment failure risk assessment game model. Based on the equipment failure risk assessment game model, a two-level game is solved to determine the failure probability and the optimal solution of the maintenance strategy, and a dual game solution set is obtained; Time series tracking is performed based on the double game solution set, and combined with the real-time update of the digital twin model to predict the changing trend of equipment failure risk, the equipment failure risk change prediction data is obtained.

[0015] By integrating multimodal sensor data with central air conditioning structural data, this invention enables comprehensive perception of the air conditioning system, providing a more comprehensive and accurate understanding of the equipment's operating status. By processing multimodal operational perception data, richer perception data can be obtained, avoiding the limitations of traditional monitoring methods that only collect single or limited parameters, ensuring real-time reflection of equipment health. Furthermore, through dynamic dimensional slicing and the construction of spatiotemporal interaction maps, abnormal changes in various dimensions of the system can be more meticulously identified, avoiding the lag and risk of misjudgment associated with traditional fault diagnosis methods that are limited to single parameters or manual judgment. Furthermore, by inputting the spatiotemporal interaction feature matrix into the central air conditioning digital twin model and performing cross-model updates, the model can be dynamically adjusted as system operations change, ensuring real-time reflection of equipment operating status and constantly updating equipment health. The real-time updated digital twin model not only effectively captures various abnormal patterns in the system but also enables joint analysis across multiple subsystems, significantly improving the accuracy and response speed of fault detection. Furthermore, by tracing the source of abnormal patterns, the root cause of faults can be accurately traced, avoiding the limitation of traditional single monitoring methods that cannot identify cross-subsystem faults. By modeling the time-series failure trends of abnormal feature traceability vectors and quantifying the degree of equipment degradation, fault prediction and early warning can be achieved, identifying potential equipment issues in advance and mitigating the impact of sudden failures. This process helps develop a more accurate critical failure prediction matrix, enabling operations and maintenance personnel to understand the equipment's operating status early on and perform preventive maintenance, reducing repair costs and improving equipment reliability. A two-layer game analysis of the critical failure prediction matrix provides a more scientific basis for operation and maintenance decision-making. By dynamically and collaboratively optimizing the main unit operating parameters, an intelligent central air conditioning operation and maintenance plan is generated. This plan automatically adjusts based on actual operating conditions, improving operation and maintenance efficiency, avoiding the blindness and lag of manual inspections, ensuring optimal equipment operation, extending equipment life, and significantly reducing operation and maintenance costs.

[0016] Optionally, this specification also provides an artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance system for executing the artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance method described above. The artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance system includes: A data acquisition module is used to obtain multimodal sensor data and central air-conditioning structure data, and perform multimodal air-conditioning operation perception based on the multimodal sensor data to obtain multimodal operation perception data; The interactive feature cross-module is used to perform dynamic dimension slicing on the multimodal operation perception data to obtain perception dimension slice data; based on the perception dimension slice data, a spatiotemporal interaction graph is constructed, and feature weights are dynamically assigned to obtain a spatiotemporal interaction feature matrix; The digital twin model construction module is used to build a central air-conditioning digital twin model based on the central air-conditioning structure data; the spatiotemporal interaction feature matrix is ​​input into the central air-conditioning digital twin model to perform cross-model updates and obtain a real-time updated digital twin model; The abnormal pattern tracing module is used to trace and capture cross-abnormal patterns based on the real-time updated digital twin model to obtain abnormal feature traceability vectors. The module also models the time series fault trend of the abnormal feature traceability vectors, quantitatively assesses the degree of equipment degradation, and obtains a critical point fault prediction matrix. The fault risk assessment module is used to perform a two-layer game fault risk assessment on the fault critical point prediction matrix and perform dynamic collaborative optimization of the host operating parameters to obtain an intelligent operation and maintenance plan for the central air conditioner.

[0017] The cross-analysis digital twin intelligent operation and maintenance system based on artificial intelligence of the present invention can realize any one of the cross-analysis digital twin intelligent operation and maintenance methods based on artificial intelligence of the present invention, and is used to combine the operation and signal transmission medium between each module to complete the cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence. The internal modules of the system cooperate with each other, thereby improving the operation and maintenance efficiency, avoiding the blindness and lag in manual inspections, ensuring that the equipment operates in the optimal state, extending the service life of the equipment, and significantly reducing the operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings: Figure 1 This is a flowchart of the steps of the cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence of the present invention; Figure 2 Detailed step flow diagram of step S1 in the present invention; The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0019] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0020] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different digital twin intelligent operations and / or processor methods and / or microcontroller methods.

[0021] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0022] To achieve this, please refer to Figures 1 to 2 The present invention provides an artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance method, which includes the following steps: Step S1: Acquire multimodal sensing data and central air-conditioning structure data, and perform multimodal air-conditioning operation perception based on the multimodal sensing data to obtain multimodal operation perception data; In this embodiment, various sensors, such as temperature sensors, pressure sensors, flow sensors, and vibration sensors, are deployed on the central air conditioner main unit and key subsystem components to collect key parameters such as temperature, pressure, flow rate, and vibration in real time. A unique identifier is assigned to each data collection point. Temperature and humidity sensors, as well as air quality sensors, are placed within the air conditioner's operating environment to collect data such as ambient temperature, humidity, and air quality, which is used to comprehensively analyze the correlation between the equipment's operating status and environmental factors. Environmental data collection points are also identified. Smart meters monitor the equipment's electrical parameters, such as current, voltage, and power, and the electrical parameter data is timestamped. The sensor data obtained above is then merged to generate multimodal sensor data. Central air conditioner structural data (such as component models, parameters, and connection methods) is obtained through the central air conditioner management platform. Based on this data, a machine learning model (such as K-means clustering or a random forest algorithm) is used to detect the air conditioner's operating status and determine the system's current operating mode. For example, temperature, humidity, and wind speed data are used to determine whether the system is in cooling or heating mode, and refrigerant pressure is combined to determine whether there are inefficiencies or abnormal operating conditions. The perception data obtained can include the operating status of each component and the energy efficiency of the air conditioner, and monitor the performance of the air conditioning system in real time.

[0023] Step S2: Perform dynamic dimension slicing on the multimodal operation perception data to obtain perception dimension slice data; construct a spatiotemporal interaction map based on the perception dimension slice data, dynamically assign feature weights, and obtain a spatiotemporal interaction feature matrix; In this embodiment, the multimodal operation perception data is dynamically sliced ​​in dimensions, the perception time series window is set to 10 minutes, and the time series slice is performed with a sliding step of 5 minutes. For example, the perception data is set to include multiple dimensions such as temperature, humidity, compressor power, condenser pressure, and fan speed. The data of each time segment will be standardized to eliminate the dimension effect. Then, a spatiotemporal interaction graph is constructed based on the spatiotemporal relationship, where: the nodes represent the key components of the central air conditioner (such as the compressor , condenser , evaporator , fan ), and environmental parameters ; Edges represent the temporal and spatial relationships between nodes. The graph structure can be represented as: Node set: ; Edge set: For the edges in the spatiotemporal interaction graph, the influence of each node interaction is represented by the time decay weight. Assume that the decay factor is ,but The weight at time t can be expressed as: Then, based on the node interaction relationship within the time window, the time decay weight distribution method is adopted, that is, the closer the interaction data is to the current time, the greater the weight is, to generate the spatiotemporal interaction feature matrix. For example, at time t, the impact of the compressor power change on the condenser pressure will not appear until 5 minutes later, so it is set (in is the attenuation coefficient) to adjust the data contribution at different time steps.

[0024] Step S3: constructing a central air-conditioning digital twin model based on the central air-conditioning structure data; inputting the spatiotemporal interaction feature matrix into the central air-conditioning digital twin model to perform cross-model update and obtain a real-time updated digital twin model; In this embodiment, a digital twin model is constructed based on the structural data of the central air conditioner (e.g., physical topology, component functions, and connection relationships). This model uses a physical data fusion method based on state-parameter mapping, combining thermal system equations, electrical system models, and environmental characteristic response models to construct a multimodal fusion digital twin model. For example, thermodynamic equations are used to calculate the heat exchange efficiency of the condenser and evaporator, and the motor operating status is estimated in combination with a power loss model. Next, the spatiotemporal interaction feature matrix obtained in step S2 is input into the digital twin model, and real-time updates are achieved through cross-model updating technology. For example, data such as the power of the air conditioner compressor and the cooling load of the condenser are input to adjust the model to ensure that the digital twin model is consistent with the operating status of the actual system.

[0025] Step S4: Based on the real-time updated digital twin model, cross-abnormal pattern traceability is captured to obtain abnormal feature traceability vectors; time series fault trend modeling is performed on the abnormal feature traceability vectors to quantitatively evaluate the degree of equipment degradation and obtain a failure critical point prediction matrix; In this embodiment, a real-time updated digital twin model is used to trace the abnormal mode of the system state. For example, a temperature deviation exceeding 5°C and a compressor power fluctuation greater than 10% are set as abnormal mode triggering conditions. During the tracing process, the cross-mapping of thermal, electrical, and environmental parameters is combined to capture the abnormal mode and generate an abnormal feature tracing vector. First, a thermal-electrical-environmental three-modal abnormal causal graph is constructed to represent the abnormal impact path. The weight of each impact path is calculated through the abnormal factor tracing method based on graph convolution to find the most important abnormal triggering factor. Time series fault trend modeling is performed based on the abnormal feature tracing vector to evaluate the degree of equipment degradation. Using the abnormal feature tracing vector, the equipment degradation trend is evaluated through the time series modeling method. At this time, it is necessary to quantify the degree of equipment degradation and predict the critical point of failure. The weighted moving average method and exponential smoothing method are used to model the time series fault trend. Assuming that the equipment degradation trend changes over time, the equipment degradation is regarded as a continuous state variable D(t), which is affected by the operating state of the equipment and can be described by the following time series model: ; For the device at all times deteriorated state; is the weighting factor (for example, 0.2 means the most recent time window has a greater impact); is the influence weight of the i-th feature on the equipment degradation state; For the moment The i-th feature in the abnormal feature tracing vector; is the total characteristic quantity. Through iterative calculation, the degradation state of the equipment at each moment is obtained. The critical point of equipment failure can be represented by setting a threshold, for example, when the degradation state of the equipment Exceeding a certain critical value indicates that the device is about to fail. ;in is the predicted moment of critical failure point; The critical degradation threshold for failure is set. For example, when the degradation state of a device exceeds 90%, it is considered to be about to fail. Using the above model, a critical failure prediction matrix can be generated for different devices. Each element in the matrix represents the predicted failure time of different devices. This matrix can be expressed as: ;in, Indicates the Predict the critical failure point of each device. Through quantitative evaluation, a degradation threshold, such as a 5% change in temperature or device power, is set to identify the critical state of impending device failure.

[0026] Step S5: Perform a double-layer game fault risk assessment on the fault critical point prediction matrix, and perform dynamic collaborative optimization of the host operating parameters to obtain an intelligent operation and maintenance plan for the central air conditioner.

[0027] In this embodiment, maintenance data related to equipment failures is obtained from historical maintenance records acquired from the central air conditioning management platform. This data is then time-series aligned with the spatiotemporal interaction feature matrix to generate a spatiotemporal maintenance interaction matrix. Assuming a certain piece of equipment has a history of frequent failures, the spatiotemporal maintenance interaction matrix can capture nodes with high maintenance frequency. For example, a specific component, such as a compressor, may appear repeatedly in past maintenance records. Based on this, a two-layer game model is constructed by combining the critical failure point prediction matrix and the spatiotemporal maintenance interaction matrix. The outer layer of the game is the equipment state, and the inner layer is the maintenance record. This game model analyzes the probability of equipment failure under different operating states and the optimal maintenance strategy. For example, if the probability of compressor failure is high, a corresponding maintenance strategy (such as more frequent inspections or replacements) should be set. During the two-layer game model solution, the probability of equipment failure (e.g., 0.05) and the effectiveness of the maintenance strategy are set to optimize the maintenance plan. A Markov decision process (MDP) is used to calculate the probability of equipment state transitions. By optimizing the expected payoff of each decision node, the impact of different maintenance strategies on equipment state transitions is evaluated. After solving, the resulting dual game solution set provides a comprehensive assessment of equipment failure risk. When performing dynamic collaborative optimization of host operating parameters, the system combines real-time updated digital twin model parameters (such as compressor power, temperature, and humidity) with predicted data on equipment failure risk changes to optimize these parameters. The goal is to improve the air conditioning system's energy efficiency and reduce the occurrence of failures. Inputting this failure risk prediction data helps the system more accurately assess the health of the equipment. For example, if the failure risk data indicates a high probability of compressor failure, the system can automatically adjust the compressor's workload based on this information to avoid damage caused by excessive operation. Specifically, when the equipment's failure risk exceeds a certain threshold, preventive measures can be taken, such as reducing operating intensity, adjusting the condenser's operating status, or performing emergency maintenance on components with a high failure risk. Conversely, if the failure risk is low, the system optimizes operating efficiency and improves energy utilization. A weighted algorithm is used to collaboratively optimize various operating parameters, for example, dynamically adjusting the compressor's workload and the condenser's cooling water flow rate to ensure optimal system operation under varying load conditions. Furthermore, these optimization results are used to generate intelligent operation and maintenance plans, for example, recommending regular checks on condenser cleanliness or automatically adjusting fan speed and cooling capacity based on ambient temperature.

[0028] Optionally, step S1 specifically includes: Step S11: acquiring multimodal sensor data and central air conditioning structure data, and performing data preprocessing on the multimodal sensor data to obtain multimodal sensor data to be analyzed; In this embodiment, various sensors, such as temperature sensors, pressure sensors, flow sensors, and vibration sensors, are deployed on the central air conditioner main unit and key subsystem components to collect key parameters such as temperature, pressure, flow rate, and vibration in real time. A unique identifier is assigned to each data collection point. Temperature and humidity sensors, as well as air quality sensors, are placed within the air conditioner's operating environment to collect data such as ambient temperature, humidity, and air quality, which is used to comprehensively analyze the correlation between the equipment's operating status and environmental factors. Environmental data collection points are also identified. Smart meters monitor the equipment's electrical parameters, such as current, voltage, and power, and the electrical parameter data is timestamped. The sensor data obtained above is then merged to generate multimodal sensor data. Central air conditioner structural data (such as component models, parameters, and connection methods) is obtained through the central air conditioner management platform. During the data preprocessing stage, a data cleaning algorithm is used to remove anomalous data with missing values ​​exceeding 10%, and data with out-of-sync timestamps is interpolated to generate time-aligned multimodal sensor data for analysis.

[0029] Step S12: dividing the multimodal sensor data to be analyzed into sensor dimensions to obtain environmental sensor data and central air conditioning sensor data; In this embodiment, the multimodal sensor data to be analyzed is dimensionalized, first splitting it into environmental sensor data and central air conditioning sensor data based on the data source. Environmental sensor data includes temperature and humidity sensors, air quality sensors, and other sensors, capturing data such as ambient temperature, humidity, and air quality, while central air conditioning sensor data includes temperature, pressure, flow, and vibration sensor data. During the data segmentation process, a time window partitioning strategy based on the signal sampling frequency is employed to unify data of varying sampling rates into a 1-second time step, ensuring the timeliness and accuracy of subsequent analysis.

[0030] Step S13: performing central air-conditioning operating environment perception according to the environmental sensing data to obtain central air-conditioning operating environment perception data; In this embodiment, environmental sensor data is used to perceive the central air conditioning operating environment. A multivariate regression model is used to analyze the impact of parameters such as temperature, humidity, and wind speed on the central air conditioning heat load. For example, if the outdoor temperature exceeds 35°C and the relative humidity exceeds 80%, a high temperature and high humidity environment is identified, predicting the risk of excessive condensing pressure in the air conditioner. Furthermore, if the Air Quality Index (AQI) exceeds 150, the air conditioner is judged to be operating in a polluted environment, affecting the efficiency of the filter. Ultimately, based on the combination of different environmental parameters, central air conditioning operating environment perception data is generated and assigned different environmental weighting coefficients.

[0031] Step S14: sensing the central air-conditioning operating state based on the central air-conditioning sensor data to obtain central air-conditioning state sensing data; In this embodiment, the operating status is sensed based on the central air-conditioning sensor data. First, a data smoothing filter algorithm is used to eliminate high-frequency noise to ensure the stability of the sensor data. Then, an air-conditioning operating status perception model is constructed based on key parameters such as compressor power, evaporator and condenser temperatures, and fan speed. For example, when the compressor power is stable at around 30kW, the condenser temperature exceeds 60°C, and the evaporator temperature is lower than 5°C, the high-load operating state in the cooling mode is identified. In addition, when the fan speed is lower than 500RPM and the outlet temperature fluctuates by more than ±3°C, the system determines that the fan is operating abnormally and generates corresponding central air-conditioning status perception data.

[0032] Step S15: performing perception feature fusion on the central air-conditioning operating environment perception data and the central air-conditioning state perception data to obtain multimodal operation perception data.

[0033] In this embodiment, a spatiotemporal feature fusion method is used to construct multimodal operational perception data for central air conditioners based on their operating environment and operating status data. During the fusion process, a time series alignment method based on dynamic time warping (DTW) is employed to match environmental data with operational status data at a consistent time step. Simultaneously, a convolutional neural network (CNN) is used to extract spatiotemporal features, combining information such as ambient temperature and compressor load trends to calculate environmental adaptability under different operating conditions. For example, when the ambient temperature is above 35°C and the humidity is greater than 80%, the compressor load increases by an average of more than 15%. This analysis derives the environmental impact factor on the energy efficiency of central air conditioners, ultimately outputting multimodal operational perception data.

[0034] Optionally, step S14 is specifically as follows: Step S141: performing hierarchical decoupling of operating parameters on the central air conditioning sensor data to obtain an operating parameter feature vector and an electrical parameter feature vector; In this embodiment, the central air conditioning sensor data is subjected to hierarchical decoupling of operating parameters. First, the data source is divided according to the sensor type, including temperature and humidity sensors (detecting evaporator temperature and condenser temperature), pressure sensors (detecting evaporator pressure and condenser pressure), current sensors (detecting compressor current), etc. During the decoupling process, a signal decomposition method based on empirical mode decomposition (EMD) is used to decompose the complex sensor signal into multiple intrinsic mode functions (IMFs), and the operating parameter feature vectors (such as the evaporation temperature change rate and the condensing pressure change rate) and electrical parameter feature vectors (such as motor load fluctuations and current harmonic components) are extracted respectively. Among them, the electrical parameter feature vectors are mainly used to analyze the operating stability of the compressor, while the operating parameter feature vectors are used to evaluate the system's thermal cycle status. Finally, a set of parameter vectors after hierarchical decoupling is obtained, which provides input data for subsequent physical model calculations.

[0035] Step S142: performing thermophysical parameter inversion on the operating parameter eigenvector based on a preset thermodynamic cycle physical model to obtain a thermal efficiency characteristic matrix; performing motor parameter inversion on the electrical parameter eigenvector based on a preset energy consumption load mapping model to obtain a motor power characteristic matrix; In this embodiment, based on the preset thermodynamic cycle physical model, the operating parameter characteristic vector is subjected to thermophysical parameter inversion, and the thermal balance equation and enthalpy calculation formula are used to calculate the thermal efficiency of each stage of the central air conditioner. For example, the heat exchange capacity of the condenser is calculated based on the evaporation temperature (5°C) and the condensation temperature (60°C), and the thermal efficiency characteristic matrix of the system is inverted in combination with the refrigerant flow rate (0.1kg / s). At the same time, based on the energy consumption load mapping model, the motor parameter inversion is performed on the electrical parameter characteristic vector, and the compressor motor power (1kW~50kW) is calculated by inputting the click current (50A) and voltage (220V) to generate the motor power characteristic matrix. In order to improve the accuracy of the model, the least squares regression method is used to optimize the parameter fitting process so that the calculation error is controlled within ±2%, ensuring the accuracy of the thermophysical inversion and motor parameter inversion. Based on the thermodynamic cycle physical model, the operating parameter characteristic vector is subjected to thermophysical parameter inversion. In the specific calculation process, the enthalpy balance equation is used to calculate the enthalpy change in the refrigerant cycle, for example: ; Among them, Q is the system heat exchange capacity, is the refrigerant flow rate (unit: kg / s), , are the specific enthalpy values ​​at the refrigerant outlet and the evaporator inlet (unit: kJ / kg). Combined with the compressor power data, the real-time thermal efficiency of the system is calculated and the thermal efficiency characteristic matrix is ​​constructed. This matrix can be used to analyze the changes in heat exchange efficiency of the air-conditioning system under different working conditions, and perform adaptive corrections based on dynamic heat loads to ensure the accuracy of thermal efficiency calculations. At the same time, based on the energy consumption load mapping model, the motor torque equation is used: ;in is the motor power (unit: kW), is the input voltage (unit: V), is the current (unit: A), The power factor is calculated and the motor power characteristic matrix is ​​obtained. It is corrected in combination with the dynamic load characteristics to control the error within ±1.5%.

[0036] Step S143: performing graph convolutional network state association modeling based on the thermal efficiency feature matrix and the motor power feature matrix to obtain a central air conditioning state time series feature map; In this embodiment, a time-series feature map of the central air conditioning state is constructed based on the thermal efficiency feature matrix and the motor power feature matrix. A graph convolutional network (GCN) is used for state association modeling, where each node represents a thermal parameter or electrical parameter at a specific time step, and the weight of each edge is calculated based on the correlation between the parameters. For example, if an increase in condensing temperature leads to an increase in motor power, the weight of the associated edge increases to reflect the influence of the two. During the modeling process, a time window size of 10 minutes is selected, and a sliding window update strategy is adopted to ensure the temporal dynamics of the state features. Ultimately, a time-series feature map of the central air conditioning state is obtained, which can be used for subsequent anomaly detection and operating status analysis.

[0037] Step S144: performing central air conditioning abnormal state detection on the central air conditioning state time series characteristic map to obtain an abnormal state perception vector and a normal state perception vector; In this embodiment, abnormal state detection is performed on the central air conditioning state time series feature map. A prediction error detection method based on a long short-term memory (LSTM) network is used to calculate the prediction error of state features at different time steps. If the prediction error at a particular time step exceeds a set threshold (e.g., exceeding ±5%), the state is determined to be abnormal, and an abnormal state perception vector is generated. For example, if the condensing temperature abnormally rises to 70°C and the motor power fluctuation amplitude exceeds ±10%, the risk of condenser failure is identified and the abnormal state perception vector is recorded. Simultaneously, a normal state perception vector is generated for normal data that conforms to operating rules and stored for subsequent trend analysis and health assessment.

[0038] Step S145: performing central air-conditioning state time series merging based on the abnormal state perception vector and the normal state perception vector to obtain central air-conditioning state perception data.

[0039] In this embodiment, the central air conditioner's state time series are merged based on abnormal state perception vectors and normal state perception vectors. A weighted fusion method based on a temporal attention mechanism is used to calculate the proportion of normal and abnormal states. For example, if the abnormal state proportion exceeds 30% over the past 24 hours, the system determines that the central air conditioner has a long-term abnormal operating trend. During the state merging process, different weightings are applied to short-term anomalies (such as a brief increase in condensing temperature) and long-term anomalies (such as long-term high-load operation of the compressor) to ensure the accuracy of the state assessment. Ultimately, the merged state perception data can be used for subsequent equipment health assessment and intelligent operation and maintenance strategy formulation.

[0040] Optionally, step S2 is specifically: Step S21: Dynamically slice the multimodal operation perception data, set the perception time series window length to [10 minutes, 30 minutes], and perform hierarchical time series slicing within a time step of [1 minute, 5 minutes] based on the preset sliding window [10 minutes, 30 minutes] to obtain perception dimension slice data; In this embodiment, when dynamically slicing multimodal operational perception data, the perception time series window length is first set to [10 minutes, 30 minutes] to accommodate the varying operating states of the central air conditioner under varying load conditions. Furthermore, the sliding window size is set to [10 minutes, 30 minutes], and a time step size of [1 minute, 5 minutes] is used for stratified time series slicing to ensure temporal relevance of the data. In practice, if a 30-minute window length and a 5-minute step size are set, each 30-minute data window can generate six time slices, each containing key variables such as temperature (°C), current (A), wind speed (m / s), and refrigerant pressure (MPa) at different time points. For example, at t=10:00, the temperature is recorded as 22.5°C, the current is 8.2A, the wind speed is 3.5m / s, and the refrigerant pressure is 1.6MPa. Dynamic slicing is then performed at subsequent time steps from t=10:01 to t=10:30 to generate continuous perception dimension slices for subsequent feature modeling.

[0041] Step S22: construct a multi-dimensional time series feature map of the central air conditioner based on the perception dimension slice data, and perform temperature-current-wind speed-refrigerant pressure cross-correlation modeling to generate a spatiotemporal interaction map; In this embodiment, based on the generated perception dimension slice data, a multi-dimensional time series feature graph of the central air conditioner is constructed to describe the dynamic interaction relationship between different physical quantities. Specifically, a graph structure is established in which temperature, current, wind speed, and refrigerant pressure are respectively used as nodes of the graph, and edge connections are constructed based on the correlation of the time series. In order to enhance the accuracy of feature modeling, cross-physical modeling is also introduced to calculate the correlation between temperature and current (Pearson correlation coefficient r), as well as the effect of wind speed on refrigerant pressure (information gain calculation). For example, if the temperature-current correlation is higher than 0.8, a weight connection is added to the feature graph to increase its influence on the system state. Ultimately, the constructed spatiotemporal interaction map can fully reflect the state evolution trend of the central air conditioner.

[0042] Step S23: Perform adaptive attention dynamic weighting on the interaction strength between nodes in the spatiotemporal interaction graph to generate a dynamic spatiotemporal interaction weight matrix; In this embodiment, on the basis of the constructed spatiotemporal interaction map, an adaptive attention mechanism is introduced to adjust the weight value according to the degree of dynamic change between the physical variables to highlight the key influencing factors. The specific operation is: first, the interaction intensity between the physical variables is calculated, such as the Pearson correlation coefficient between temperature and current. If |r|>0.7, it is considered that there is a strong linear relationship between the two, and its weight is increased in the attention assignment process. Secondly, a dynamic weighting rule is set. For example, if the temperature-current interaction intensity is >0.8, the weight α=1.5 is increased. If the wind speed-refrigerant pressure interaction intensity is <0.5, the weight α=0.7 is reduced. After adaptive weighting, a dynamic spatiotemporal interaction weight matrix is ​​generated to optimize the dependency relationship between variables.

[0043] Step S24: Perform thermodynamic cycle coupling modeling based on the central air conditioning structure data, set the thermodynamic cycle parameters [0°C, 100°C] and the airflow disturbance coefficient [0.01, 0.1], and perform physical coupling correction on the dynamic spatiotemporal interaction weight matrix based on the coupling modeling results to obtain a corrected spatiotemporal interaction characteristic matrix; In this embodiment, based on the generated dynamic spatiotemporal interaction weight matrix, correction is performed in combination with the thermodynamic cycle physical model to ensure physical consistency between the data. First, the thermodynamic cycle parameter range [0°C, 100°C] is set to constrain abnormal values ​​that may appear during the calculation process, and the airflow disturbance coefficient [0.01, 0.1] is set to evaluate the impact of wind speed on the thermal efficiency of the system. Subsequently, based on the physical modeling of the system, the heat exchange of the condenser is calculated. For example, under the conditions of evaporation temperature 5°C, condensation temperature 60°C, and refrigerant flow rate 0.1kg / s, the thermal efficiency characteristic matrix is ​​inverted and fused and corrected with the dynamic spatiotemporal interaction weight matrix. For example, if the calculated thermal efficiency exceeds 100%, it means that there are unreasonable parameters in the model, and it is necessary to adjust the refrigerant flow or correct the sensor data deviation, and finally obtain the corrected spatiotemporal interaction characteristic matrix optimized by physical constraints.

[0044] Step S25: performing feature normalization processing on the modified spatiotemporal interaction feature matrix, and performing weight suppression of abnormal interaction regions between nodes to generate a spatiotemporal interaction feature matrix.

[0045] In this embodiment, based on the generated modified spatiotemporal interaction feature matrix, the data needs to be normalized and the influence of existing abnormal areas needs to be removed to improve the stability of the modeling. Specifically, the minimum-maximum normalization method is used to normalize the temperature (°C) to [0,1], the current (A) to [-1,1], the wind speed (m / s) to [0,1], and the refrigerant pressure (MPa) to [0,1] to ensure that different physical quantities are calculated on the same numerical scale. At the same time, the abnormal area is identified. If the refrigerant pressure fluctuates by more than 20% in a short period of time, it is considered that there may be abnormal interaction in the area, and its corresponding attention weight is reduced (α1=0.5). Finally, the optimized spatiotemporal interaction feature matrix is ​​obtained to provide more robust input data to support subsequent state prediction and fault detection analysis.

[0046] Optionally, step S25 is specifically as follows: Step S251: performing node normalization processing on the modified spatiotemporal interaction feature matrix, setting the normalization interval [0, 1], and performing normalization conversion based on the preset node timing window length [5min, 10min] and dynamic interaction frequency threshold [0.1, 0.5] to generate a standardized spatiotemporal interaction feature matrix; In this embodiment, when processing the modified spatiotemporal interaction feature matrix, the node data is first normalized to ensure that the numerical ranges of different feature quantities are consistent, thereby improving the stability of subsequent calculations. Specifically, a normalization interval of [0, 1] is set, and a minimum-maximum normalization method is used to map all node feature values, such as temperature (°C), current (A), wind speed (m / s), and refrigerant pressure (MPa), to a unified interval. For example, if the initial temperature data for a node ranges from [15°C to 35°C], a conventional normalization formula is used to calculate the normalized data. Furthermore, to ensure the smoothness of the time series data, a node time series window length of [5 minutes, 10 minutes] and a dynamic interaction frequency threshold of [0.1, 0.5] are introduced to perform data normalization. For example, within a 10-minute time window, if the temperature interaction frequency exceeds 0.5, it is normalized and smoothed to prevent local data mutations from interfering with system status judgment. Finally, the construction of the standardized spatiotemporal interaction feature matrix is ​​completed, providing stable input for subsequent anomaly detection.

[0047] Step S252: Based on the abnormal state perception vector and the normal state perception vector, combined with the coupling modeling results, the nodes in the standardized spatiotemporal interaction feature matrix are detected for abnormal regions in a time series. If the interaction strength of a node exceeds the preset abnormal threshold [1.5σ, 3σ], the spatiotemporal trajectory of the node is marked to obtain a set of abnormally marked interaction region nodes; In this embodiment, based on the generated standardized spatiotemporal interaction feature matrix, an abnormal state perception vector and a normal state perception vector are introduced, and combined with the coupling modeling results, the node data in different time windows are detected for abnormal areas in a time series. Combined with the coupling modeling results of the thermodynamic cycle physical model and the energy consumption load mapping model, the node data in different time windows are detected for abnormal areas in a time series to improve the physical consistency and accuracy of the detection. Based on the thermodynamic cycle physical model, the normal temperature, pressure, and flow ranges under different operating conditions are calculated. For example, under the conditions of evaporation temperature 5°C, condensation temperature 60°C, and refrigerant flow rate 0.1kg / s, the condenser heat exchange capacity is calculated according to the thermodynamic balance equation, and the theoretical energy efficiency ratio is calculated based on the compressor power. Assuming that under standard operating conditions, the COP normal range is [4.0, 6.0]. If the COP in a certain time window is lower than 3.0 or higher than 6.5, it means that there is an abnormal operating state in the time window. At the same time, combined with the energy consumption load mapping model, the theoretical current characteristic matrix of the air conditioning system under different load conditions is calculated. For example, at a set load of 50%, the theoretical current range is [8A, 12A], while at an 80% load, the current range changes to [12A, 16A]. These coupled modeling results provide a reference benchmark for anomaly detection, enabling detection to be based not only on statistical distribution (such as standard deviation σ) but also on actual physical laws, improving the accuracy of anomaly determination. After obtaining the theoretical operating range, time series anomaly detection is performed on the data of each node in the standardized spatiotemporal interaction characteristic matrix, with an anomaly threshold range of [1.5σ, 3σ] (σ is the standard deviation). If the interaction strength of a node exceeds the normal distribution range by 1.5σ, it is considered a mild anomaly; if it exceeds 3σ, it is considered a severe anomaly. In practice, all node data within a 10-minute window is traversed, and its mean μ and standard deviation σ are calculated. For example, if the mean refrigerant pressure at a certain moment is 1.6 MPa and the standard deviation σ is 0.2 MPa, then the abnormal range corresponding to a 1.5σ threshold is [1.3 MPa, 1.9 MPa], and the range corresponding to a 3σ threshold is [1.0 MPa, 2.2 MPa]. If the refrigerant pressure at a node suddenly rises to 2.3 MPa at t = 10:05, the node is considered abnormal and its spatiotemporal trajectory is marked, forming a node set in the abnormal annotated interactive region. This marking process can be used to identify potential equipment failure points and improve the system's fault warning capabilities.

[0048] Step S253: Dynamically suppress the weight of the node set in the abnormally labeled interaction area. Combined with the preset time series smoothing coefficient [0.1, 0.3] and the node activity weight factor [0.5, 1.5], perform dimensionality reduction mapping of the abnormal node weight within the node time series window length [5min, 10min] to generate a suppressed spatiotemporal interaction feature matrix. In this embodiment, a dynamic weight suppression mechanism is used for the set of nodes in the identified abnormally labeled interaction area to reduce the interference of abnormal nodes on the overall system state prediction. First, a time series smoothing coefficient of [0.1, 0.3] is set to dynamically adjust the weight of abnormal nodes so that they gradually decay within different time windows. For example, if a node is marked as abnormal at t = 10:05, its weight is decayed by 0.1 times at t = 10:10, that is, w′ = w×(1−α); where w is the initial weight, α is the smoothing coefficient (ranging from 0.1 to 0.3), and w′ is the adjusted weight. In addition, considering the impact of different node activity on the system state, a node activity weight factor of [0.5, 1.5] is introduced to perform dynamic dimensionality reduction mapping within the node time series window length of [5 minutes, 10 minutes]. For example, if a node has a high interaction frequency over the past 10 minutes (activity weight factor = 1.5), only 0.5 times of attenuation is applied when an abnormal state occurs to retain some of its information. Conversely, if the node is less active (activity weight factor = 0.5), 1.5 times of attenuation is applied to reduce its interference. Ultimately, after this processing, a matrix of suppressed spatiotemporal interaction features is generated, improving the robustness of the system's anomaly detection.

[0049] Step S254: Based on the thermodynamic cycle physical model and the energy consumption load mapping model, cross-modal spatiotemporal feature fusion is performed on the spatiotemporal interaction suppression feature matrix to obtain the spatiotemporal interaction feature matrix.

[0050] In this embodiment, based on the generated suppression spatiotemporal interaction feature matrix, the thermodynamic cycle physical model and the energy consumption load mapping model are further combined to perform cross-modal spatiotemporal feature fusion to improve the physical consistency of the data and the prediction accuracy. First, based on the thermodynamic cycle physical model, the system operating efficiency in different time windows is calculated. For example, under the conditions of evaporation temperature 5°C, condensation temperature 60°C, and refrigerant flow rate 0.1kg / s, the heat exchange capacity of the condenser is calculated, and the energy consumption load mapping model is used to invert the energy efficiency ratio of the air-conditioning system: ;in Exchange heat for the condenser, is the compressor input power. If the calculated COP is lower than 3.0 (normal range 4.0-6.0), it means that the system is in an abnormal state, and further analysis of the causes of failure such as condenser scaling and refrigerant leakage is required. Secondly, in the cross-modal fusion process, a node-level data alignment strategy is introduced to ensure that the energy consumption data output by the physical model is consistent with the data structure of the suppressed spatiotemporal interaction feature matrix. For example, if there is a time deviation (>5min) between the evaporation temperature, condensing pressure and other parameters calculated by the thermodynamic cycle and the suppressed spatiotemporal interaction data, time series alignment processing is performed to ensure that the fused data can be used for unified modeling. Finally, an optimized spatiotemporal interaction feature matrix is ​​generated to provide high-quality input for the subsequent central air-conditioning health state evolution modeling.

[0051] Optionally, step S3 specifically includes: Step S31: extracting central air-conditioning operation features based on the multi-modal operation perception data, and identifying the air-conditioning operation mode to obtain central air-conditioning operation mode data; In this embodiment, air conditioner operating features are extracted from multimodal operating perception data to identify the air conditioner's operating mode. For example, the multimodal operating perception data indicates that when the air conditioner is operating, the temperature sensor monitors a condensing temperature of 60°C, an evaporating temperature of 5°C, a current of 12A, and a wind speed of 1200 RPM. This collected data is input into a clustering algorithm, which uses a K-means clustering algorithm to classify different data patterns. Load ranges are set (e.g., low load: 10%-30%, standard load: 30%-70%, high load: 70%-100%) to identify the air conditioner's operating mode. In this process, the operating mode is determined based on dynamic threshold settings. For example, under standard load, the evaporating temperature is approximately 5°C, the condensing temperature is 60°C, and the current is stable at 12A. This system can dynamically identify and classify different air conditioner operating states, providing accurate data support for subsequent operations and optimization.

[0052] Step S32: extracting structural connection features based on the central air-conditioning structure data, and constructing a central air-conditioning digital twin model based on the structural connection features and the central air-conditioning operation mode data; In this embodiment, the structural data of the central air conditioner is analyzed in detail to extract the physical connection characteristics between each component, including the relative positions of the condenser, evaporator, compressor, and expansion valve, the piping connection method, the refrigerant flow path, and its convective heat transfer characteristics. By analyzing these structural connection characteristics, a physical connection relationship matrix between the various modules of the air conditioning system is constructed to ensure that each component is properly represented in the digital model. The connection method between each component, such as pipe connections and refrigerant flow path, can be represented in matrix form, with nodes representing components and edges representing physical connections. For example, if the central air conditioning system uses a parallel piping connection method with higher heat exchange efficiency, the refrigerant flow path under this connection method may affect the efficiency of the heat exchanger. During this process, a structural connection diagram is established based on the piping structure and refrigerant flow path. By comparing the relationship between the actual system and the theoretical structure, the connection relationship between each component is further optimized. Matrix C represents the connections between the condenser, evaporator, and compressor. A value of 1 indicates a connection, and a value of 0 indicates no connection. Next, based on the aforementioned central air conditioning operating mode data (e.g., low-load and high-load modes), a digital twin initial model is constructed through multiphysics coupling modeling, incorporating operating load conditions and equipment parameters. This model includes the dynamic states of components such as the compressor, condenser, and evaporator, as well as their interactions. This ensures that the system's state under different operating modes is accurately reflected through real-time data, creating a virtual model that closely matches the actual equipment state. The model considers the air conditioning system's operating load conditions and equipment parameters, such as sensor data such as the operating state of the condenser and compressor, as well as current and pressure. The model must not only describe the dynamic states of the equipment components but also the interactions between them. For example, it can be used to determine how the condenser's heat exchange efficiency changes with changes in compressor speed, or whether the evaporator's operating state is affected by the expansion valve's flow rate adjustment.

[0053] Step S33: Input the spatiotemporal interaction feature matrix into the central air-conditioning digital twin initial model, and perform cross-mapping of thermal-electrical-environmental modal features to generate a central air-conditioning state mapping feature map; In this embodiment, the spatiotemporal interaction feature matrix acquired from various sensors (such as temperature, current, and humidity sensors) is input into the initial digital twin model of the central air conditioner to perform cross-mapping of thermal, electrical, and environmental modalities. Specifically, the temperature sensor captures an evaporating temperature of 5°C and a condensing temperature of 60°C, the current sensor measures a current of 12A, and the humidity sensor detects an indoor humidity of 60%. This multidimensional data is cross-mapped using a graph convolutional network (GCN) to analyze and generate a spatiotemporal interaction feature map of the central air conditioner. This further reveals the complex relationship between environmental factors such as temperature, current, and humidity and the air conditioner's operating status. The cross-mapping process accounts for the coupling effects between different modalities, such as the impact of temperature changes on current during high load and the impact of humidity changes on condenser performance. The resulting feature map provides accurate data support for subsequent state correction and optimization.

[0054] Step S34: Based on the central air conditioning state mapping feature map, dynamically align the physical state parameters with the environmental perception parameters to obtain a state parameter correction matrix; In this embodiment, the theoretical evaporation temperature and condensation temperature are calculated based on the thermodynamic model and operating data, and compared with the actual measured values. For example, when the ambient temperature is 30°C, the theoretical evaporation temperature should be 5°C and the condensation temperature should be 60°C. If the actual measured evaporation temperature is 6°C and the condensation temperature is 61°C, the state parameters are corrected. By applying the adaptive Kalman filter algorithm, key state parameters such as evaporation temperature, condensation temperature, and current are dynamically corrected in real time. During the correction process, system parameters such as the output power of the compressor and the cooling water flow rate of the condenser are adjusted in combination with environmental sensors (such as outdoor temperature and humidity) and system load data to ensure that the state reflected by the model is consistent with the actual equipment. In this way, the system can correct and calibrate its state parameters in real time to ensure the consistency between the digital twin model and the actual equipment, thereby improving the operating accuracy and reliability of the system.

[0055] Step S35: Dynamically update the parameters of the central air-conditioning digital twin model according to the state parameter correction matrix, and perform adaptive timing correction of the twin model to obtain a real-time updated digital twin model.

[0056] In this embodiment, the central air conditioner digital twin model is dynamically updated based on a state parameter correction matrix, and adaptive time series correction is performed. Specifically, by first monitoring state parameters such as temperature and current in real time, combined with a previously generated central air conditioner state mapping feature map, deviations from expected operating states are identified and corrected. For example, if the current detected by the current sensor exceeds 12A and the condensing temperature exceeds the standard range of 60°C, the operating parameters of the condenser and compressor in the central air conditioner digital twin model are automatically adjusted. Then, a time series prediction model based on an LSTM (long short-term memory) network is applied to analyze load changes over the next 10 minutes. Based on the predicted results, the central air conditioner digital twin model's operating strategy (such as adjusting wind speed or refrigerant flow) is adjusted to ensure consistent operation under varying load and environmental conditions. Through this dynamic update and correction mechanism, the central air conditioner digital twin model can adaptively respond to environmental and load changes, optimizing system operating efficiency.

[0057] Optionally, the cross-abnormal pattern tracing and capturing described in step S4 is specifically as follows: Based on the real-time updating of the digital twin model, the real-time thermal-electrical-environmental three-modal feature data of the central air conditioner is obtained, and the multi-modal feature time series hierarchical decoupling is performed to obtain the real-time thermal-electrical-environmental three-modal feature vector; In this embodiment, a multimodal sensor network deployed in a central air conditioning system collects thermal, electrical, and environmental data in real time. For example, a temperature sensor records a condensing temperature of 60°C and an evaporating temperature of 5°C, a current sensor measures a current of 10A, and a wind speed sensor records a wind speed of 1100 RPM. Environmental sensor data includes, for example, an outdoor temperature of 25°C and a humidity of 65%. This sensor data is synchronously transmitted to a computing system using a real-time, updated digital twin model. Next, a time-series hierarchical decoupling method is used to perform hierarchical processing on the thermal, electrical, and environmental modal feature data based on a set time window (e.g., 10 to 30 minutes) and a dynamic feature extraction method. Specifically, for thermal features, evaporating and condensing temperature data are extracted in layers according to the time window. Electrical feature data is time-series segmented based on current and power. Environmental features are decoupled based on multiple environmental parameters such as humidity and temperature. Ultimately, this process yields real-time, updated feature vectors for the thermal, electrical, and environmental modalities, providing a foundation for subsequent abnormal pattern detection and system optimization.

[0058] Based on the real-time thermal-electrical-environmental trimodal feature vectors and combined with the abnormal state perception vector, a central air conditioning abnormal mode association graph is constructed. The cross-abnormal intensity of modal feature nodes is modeled to generate an abnormal mode cross-feature matrix. In this embodiment, a central air conditioning abnormality pattern association graph is constructed based on the obtained thermal, electrical, and environmental trimodal feature vectors, combined with abnormal state perception vectors (e.g., abnormal indicators such as condensing temperature exceeding 60°C, current exceeding 12A, and ambient temperature exceeding 30°C). First, each modal feature is represented as a node in the graph—for example, condensing temperature, current, and wind speed—and connected based on the interrelationships and dynamic changes between sensors. A graph convolutional network (GCN)-based approach is used to model the cross-anomaly intensity of modal feature nodes, calculating the interaction and anomaly intensity between nodes. For example, when the condensing temperature is too high, the current abnormally increases, and wind speed changes are also affected. These nodes are weighted in the anomaly intensity graph. During the modeling process, anomaly detection is performed by setting thresholds (e.g., when temperature exceeds a set value by 1.5σ, or when current exceeds a set value by 2σ), resulting in a matrix containing cross-features of abnormal nodes. This matrix demonstrates the interactive relationships between different modal features and can reflect changes in abnormal patterns in real time.

[0059] The abnormal factor traceability path is traced on the abnormal pattern cross-feature matrix. Based on the digital twin model's thermodynamic cycle parameters [20°C, 70°C], motor power parameters [0kW, 5kW], and environmental disturbance parameters [0, 0.5], the abnormal triggering factors are located and an abnormal traceability path diagram is generated. In this embodiment, the abnormal factor source path is traced based on the generated abnormal pattern cross-feature matrix. First, based on the obtained abnormal pattern cross-feature matrix, the abnormal factor source path is traced. This process relies on the thermodynamic cycle parameters, motor power parameters, and environmental disturbance parameters set in the digital twin model. Taking the thermodynamic cycle of a central air conditioner as an example, assume that under normal conditions, the evaporation temperature is 20°C and the condensation temperature is 60°C. However, under abnormal conditions, the condensation temperature exceeds 70°C, while the evaporation temperature is below 20°C. Through the simulation of the thermodynamic cycle in the digital twin model, these parameters are dynamically tracked and monitored. In this case, any deviation between the thermodynamic cycle parameters in the digital twin model (for example, the condensation temperature is set within the 70°C range) and the actual measured condensation temperature is marked as an anomaly. Next, based on the motor power parameter range [0kW, 5kW], it is detected that the motor power should be maintained within a specific range (for example, 3kW) under normal operating load, but under abnormal conditions, the motor power may exceed 5kW due to overload. This parameter range is applied to the model to identify abnormal increases in motor power and determine their impact on the overall operating status of the air conditioning system. Taking into account the influence of environmental disturbance parameters (such as wind speed and external temperature), simulations are performed based on these disturbance factors (for example, a wind speed disturbance coefficient within the range of 0 to 0.5). Extreme changes in the external environment (such as wind speed exceeding the set range of 0.5) can cause abnormal operation of the air conditioning system. Therefore, by analyzing the interaction between the environmental disturbance coefficient and other system parameters, the system can track the environmental factors associated with abnormal events. Through dynamic simulation and tracking of the digital twin model, changes in various parameters such as the thermal cycle, motor power, and environmental disturbances can be collaboratively modeled to identify the triggering factors of the abnormality. The digital twin model simulates how these parameters interact by establishing association rules between them (for example, the operating status of the thermal cycle, the impact of environmental changes on air conditioning load, and the relationship between motor power and system load). For example, excessively high condensing temperature can affect evaporation pressure and compressor efficiency, leading to abnormal changes in motor power. This modeling enables real-time monitoring of these parameter changes and their interactions. Digital twin models can identify the cross-influences between different modal features. For example, excessive motor power may not be directly caused by a problem with the motor itself, but rather the result of a failure in the thermal cycle system (such as unstable refrigerant flow) or changes in the external environment (such as high temperature and high humidity). By analyzing the interactions between these different modes (thermal, electrical, and environmental), the system can trace the triggering factors of the anomaly. Ultimately, an anomaly tracing path diagram is generated, which shows the relationship between the anomaly factors and their impact on the overall performance of the system. For example, if the condensing temperature is too high, the motor power may be overloaded, and environmental disturbances (such as high wind speed) exacerbate this situation.The anomaly tracing path diagram clearly depicts how these factors work together to ultimately lead to system anomalies.

[0060] Based on the anomaly tracing path diagram, perform dynamic clustering of abnormal nodes, extract the anomaly trigger factor vector, and generate the anomaly trigger factor perception vector; In this embodiment, based on the generated abnormal tracing path diagram, a dynamic clustering analysis of abnormal nodes will be performed. For example, by setting a threshold (such as the abnormality between nodes exceeds 3σ), nodes that show obvious abnormal characteristics will be clustered to form abnormal groups. In this process, the K-means clustering algorithm is used to group the nodes, and nodes belonging to the same type of abnormality are classified into one category. The central node of each cluster represents the typical trigger factor of this type of abnormality, such as excessively high condensing temperature, current overload, etc. In this way, the clustering results can be dynamically adjusted, and the perception vector of the abnormal trigger factor can be extracted in real time, for example, the joint abnormal trigger factor vector of excessively high condensing temperature and current. Through this process, possible abnormal changes can be monitored and adapted in real-time operation, providing accurate input data for subsequent abnormal pattern recognition.

[0061] The abnormal trigger factor perception vector and multimodal operation perception data are combined to perform layered traceability of abnormal patterns and obtain the abnormal feature traceability vector.

[0062] In this embodiment, the obtained abnormal trigger factor perception vector and multimodal operation perception data are combined to perform hierarchical tracing of abnormal patterns. The multimodal operation perception data (for example, temperature, humidity, current, etc.) are matched with the abnormal trigger factor perception vector, and based on the set tracing strategy (such as the occurrence time of abnormal data exceeding the set threshold for more than 5 minutes), the abnormal features are traced layer by layer. For example, when the condensing temperature is too high, other modal features such as current and wind speed are first analyzed to see if they are also abnormal, and the source of the abnormality is traced layer by layer. Finally, through the abnormal pattern hierarchical tracing process, a tracing vector containing all abnormal features is obtained, such as abnormal condensing temperature, compressor current overload, etc. These tracing vectors provide a reliable basis for subsequent abnormality detection, prediction and decision-making.

[0063] Optionally, the quantitative evaluation of the device degradation degree in step S4 is specifically: Based on the abnormal feature tracing vector, the abnormal pattern intensity time series response model is performed to obtain the abnormal pattern time series response curve; In this embodiment, a time series analysis is performed on the abnormal feature traceability vector to capture the change trend and fluctuation amplitude. Assuming that at a certain moment, the thermal cycle and electrical parameters of the air-conditioning system show abnormal changes, a response model will be established based on these data. To this end, dynamic regression analysis and time series smoothing technology are used, and the time window is set to [10min, 30min]. Each time the window slides, the response pattern is modeled. Through this modeling method, the abnormal response intensity of the system at each moment can be extracted, so that the abnormal pattern time series response curve can be drawn to capture the pattern and intensity evolution process of the abnormal occurrence. For example, if a fault causes a sudden increase in the compressor load, the time series response curve will reflect the changes in the load fluctuation.

[0064] A central air conditioning operation degradation time series model is constructed based on the abnormal mode time series response curve, and an adaptive dynamic sliding window decomposition is performed on the central air conditioning operation degradation time series model to extract the degradation factor time series feature vector; In this embodiment, based on the generated abnormal mode time series response curve, an exponential smoothing method is used to construct a central air conditioning operation degradation time series model. The window size is set to [5 minutes, 15 minutes], and each sliding window time step is 1 minute. The model identifies the performance degradation trend caused by the abnormality and extracts the main factors related to the degradation, such as decreased cooling efficiency and increased motor load. Through adaptive dynamic sliding window decomposition, the degradation time series model is decomposed to obtain a series of degradation factor time series feature vectors. For example, if the heat exchange efficiency gradually decreases, the decomposed time series feature vector will highlight the gradual decline in thermodynamic cycle performance.

[0065] Based on the degradation factor time series eigenvector, combined with the real-time update of the thermal, electrical, and environmental trimodal parameters in the digital twin model, the degradation critical threshold is set; In this embodiment, the extracted time-series feature vectors of degradation factors are combined with the three modal parameters of thermal, electrical, and environmental (e.g., compressor load, cooling water temperature, and ambient temperature) in a real-time digital twin model. The impact of each modal data change on air conditioning system performance is calculated based on the time-series variation. First, a statistical analysis method is employed to analyze the real-time digital twin model, such as a moving average (MA) or adaptive filtering method (e.g., Kalman filtering), to adjust thresholds in real time. Based on the historical fluctuation range of each parameter (thermal, electrical, and environmental), a standard deviation is dynamically calculated, and the critical degradation threshold is set as the parameter's mean plus or minus two standard deviations. For example, if the historical standard deviation of cooling water temperature fluctuations is 0.5°C, the critical degradation threshold for cooling water temperature is set to the current mean + 2 × standard deviation (i.e., a 2°C deviation range). Similarly, the thresholds for compressor power and cooling efficiency are dynamically adjusted based on their historical fluctuations using similar statistical calculation methods. For each factor, when its value exceeds the set threshold, it is considered to be causing gradual degradation of air conditioning system performance. Based on the matching of these data with the thresholds and time series analysis, we identify which factors have a significant degradation effect on the system within a specific time window. Specifically, when the cooling water temperature exceeds 2°C continuously and remains for more than 10 minutes, the weight of this parameter is automatically increased, and the degradation factor time series feature vector is updated to reflect the impact of this changing trend on the system. By comprehensively considering all relevant factors, the critical degradation threshold is finally set as follows: when the compressor load increases by 10%, the air conditioning system will enter a high-risk area; when the cooling efficiency drops by 5%, it means that the system has reached the critical point of potential failure and further measures need to be taken.

[0066] The critical degradation threshold is used to predict the critical failure point by updating the digital twin model in real time, and the critical failure point prediction matrix is ​​obtained.

[0067] In this embodiment, the digital twin model performs fault prediction based on real-time air conditioning system data (such as compressor power, cooling water temperature, and ambient temperature). Combined with the previously acquired time-series feature vectors of degradation factors, it calculates the critical point of equipment failure within the next few hours. This critical point prediction is based on a weighted average of historical data and real-time parameters. Time series analysis and regression algorithms, such as the ARIMA model, are used to infer the performance trends of the air conditioning system over the next few hours. For example, if the cooling water temperature rises by more than 2°C and the compressor power fluctuation exceeds a set threshold, the regression model predicts that the air conditioning equipment will fail within the next two hours. A critical point prediction matrix indicates the probability of failure at a specific time point (e.g., within the next two or four hours) and marks this time point as a critical point. These critical points and specific time points are then output in matrix form, resulting in a critical point prediction matrix. If the predicted critical point value approaches a set threshold (e.g., a 50% probability of failure), the system issues an alarm, prompting maintenance personnel to perform preventive maintenance or inspection at that time.

[0068] Optionally, the two-layer game failure risk assessment in step S5 is specifically as follows: Obtain historical maintenance record data and perform time series alignment on the historical maintenance record data according to the spatiotemporal interaction feature matrix to obtain the spatiotemporal maintenance interaction matrix; In this embodiment, historical maintenance records for air conditioning equipment are obtained through a central air conditioning management cloud platform. Data sources include equipment failure logs, maintenance records, parts replacement records, and maintenance reports. These maintenance records are assumed to contain information such as the time of failure, failure type, repair response time, repair duration, and repair costs during equipment operation. To ensure that these historical records can be compared with real-time data, they need to be time-series aligned using a spatiotemporal interaction feature matrix. Specifically, this spatiotemporal interaction feature matrix is ​​constructed based on the thermal, electrical, and environmental trimodal characteristics of the equipment and provides information on the state changes of the equipment at each point in time. By setting an appropriate time window (e.g., [1 hour, 24 hours]), the maintenance records for each equipment failure are time-series aligned to ensure that the maintenance data is synchronized with the real-time equipment status data. This generates a spatiotemporal maintenance interaction matrix. This matrix clearly defines the equipment's failure history, repair response, and state changes within each time period.

[0069] Combining the failure critical point prediction matrix and the spatiotemporal maintenance interaction matrix, a two-layer game model is constructed. The outer layer game subject is set as the equipment status, and the inner layer game subject is set as the maintenance record, thus obtaining the equipment failure risk assessment game model. In this embodiment, when constructing a two-layer game model, the outer game subject represents the current state of the equipment, including the thermal, electrical and environmental three-modal characteristic data of the equipment, and the inner game subject represents the maintenance record. The goal of this model is to analyze the interaction between the maintenance record strategy and the current state of the equipment through game theory, so as to assess the risk of equipment failure and formulate a maintenance strategy. The decision variables of the outer game subject are the current failure risk values ​​of the equipment. These values ​​are derived from the failure critical point prediction matrix, which predicts future failure risks through historical failure modes and equipment operation data. The decision variables of the inner game subject are maintenance strategies, including maintenance timing, maintenance resource allocation, spare parts replacement, etc. The goal of the game model is to optimize the allocation of maintenance resources and find the optimal solution between the probability of equipment failure and maintenance costs, thereby reducing the frequency of equipment failures and reducing maintenance costs. The goal of the outer game model is to maximize the operational safety and stability of the equipment. The objective function can be set to minimize the risk of equipment failure. The two-layer game objective function can be written as follows: );in: Indicates the goal of the device state game; ) represents the goal of the maintenance record game. Specifically, is the number of devices, For device index, is the number of maintenance operations, For maintenance operation index, Indicates at time Time Device The failure risk assessment value of Representation device The current state decision variable, Indicates that the device requires maintenance. 0 means the device does not require maintenance. Indicates maintenance operation On the device The spatiotemporal interaction coefficient on For maintenance operations Assuming that the equipment failure risk is related to thermal, electrical, and environmental parameters, a failure risk assessment function can be formed by weighting and summing these parameters: ;in, It is temperature, Is the electrical state, is the environmental condition, and w1, w2, and w3 are the weights of each feature. The goal of the inner-level game model is to minimize the frequency of equipment failures while controlling maintenance costs through a reasonable maintenance strategy. Maintenance strategies include determining whether to perform emergency repairs, scheduling regular inspections, and selecting maintenance personnel. The objective function of the inner-level game is a trade-off between maintenance costs and failure probability: ; and are the weights of cost and downtime, For maintenance costs, The two-layer game model is solved using the Nash equilibrium method from game theory. The outer and inner game entities interact with each other, ultimately reaching a stable equilibrium. At this equilibrium, the equipment's failure risk and maintenance strategy are optimized.

[0070] Based on the equipment failure risk assessment game model, a two-level game is solved to determine the failure probability and the optimal solution of the maintenance strategy, and a dual game solution set is obtained; In this embodiment, when solving a two-layer game, the outer-layer device state and inner-layer maintenance strategy are first optimized using a game model-solving algorithm (e.g., Nash equilibrium, evolutionary game algorithm, etc.). Assume that, given a given device state, there is a trade-off between the probability of device failure and the effectiveness of the maintenance strategy. For example, if the device failure probability is high (e.g., exceeding 70%), premature maintenance may be necessary, but this also increases maintenance costs. Through game solving, an optimal solution set is obtained, which includes the optimal control value for the failure probability and the optimal execution plan for the maintenance strategy. For example, when the failure probability is 30%, the equipment requires an inspection and preventive maintenance within 90 days; when the failure probability exceeds 50%, emergency maintenance procedures should be initiated. This optimal solution set provides the maintenance team with a basis for decision-making, ensuring that the risk of equipment failure is minimized while staying within budget.

[0071] Time series tracking is performed based on the double game solution set, and combined with the real-time update of the digital twin model to predict the changing trend of equipment failure risk, the equipment failure risk change prediction data is obtained.

[0072] In this embodiment, time-series tracking is performed using the optimal maintenance strategy determined by the dual game solution set. This tracking process requires real-time updates of the digital twin model, which monitors various equipment operating parameters (such as motor power, temperature, and humidity) in real time and adjusts maintenance plans based on the strategy derived from the game solution set. For example, when the risk of equipment failure is high, the model automatically adjusts the maintenance frequency or repair content. Time-series tracking compares real-time equipment data with failure risk prediction data to observe the changing trends of failure risk over time. Through continuous iteration and optimization, predictions of changes in equipment failure risk can be obtained, helping to predict the possible time and probability of future equipment failures. This prediction data can be used to develop more accurate maintenance plans or issue risk warnings. Dynamic updates of the digital twin model can also be used to continuously improve equipment operation and maintenance strategies.

[0073] Optionally, this specification also provides an artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance system for executing the artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance method described above. The artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance system includes: A data acquisition module is used to obtain multimodal sensor data and central air-conditioning structure data, and perform multimodal air-conditioning operation perception based on the multimodal sensor data to obtain multimodal operation perception data; The interactive feature cross-module is used to perform dynamic dimension slicing on the multimodal operation perception data to obtain perception dimension slice data; based on the perception dimension slice data, a spatiotemporal interaction graph is constructed, and feature weights are dynamically assigned to obtain a spatiotemporal interaction feature matrix; The digital twin model construction module is used to build a central air-conditioning digital twin model based on the central air-conditioning structure data; the spatiotemporal interaction feature matrix is ​​input into the central air-conditioning digital twin model to perform cross-model updates and obtain a real-time updated digital twin model; The abnormal pattern tracing module is used to trace and capture cross-abnormal patterns based on the real-time updated digital twin model to obtain abnormal feature traceability vectors. The module also models the time series fault trend of the abnormal feature traceability vectors, quantitatively assesses the degree of equipment degradation, and obtains a critical point fault prediction matrix. The fault risk assessment module is used to perform a two-layer game fault risk assessment on the fault critical point prediction matrix and perform dynamic collaborative optimization of the host operating parameters to obtain an intelligent operation and maintenance plan for the central air conditioner.

[0074] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0075] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence, characterized by: The following steps are involved: Step S1: Acquire multimodal sensing data and central air-conditioning structure data, and perform multimodal air-conditioning operation perception based on the multimodal sensing data to obtain multimodal operation perception data; Step S2: Perform dynamic dimension slicing on the multimodal operation perception data to obtain perception dimension slice data; construct a spatiotemporal interaction map based on the perception dimension slice data, dynamically assign feature weights, and obtain a spatiotemporal interaction feature matrix; Step S3: constructing a central air-conditioning digital twin model based on the central air-conditioning structure data; inputting the spatiotemporal interaction feature matrix into the central air-conditioning digital twin model to perform cross-model update and obtain a real-time updated digital twin model; Step S4: Based on the real-time updated digital twin model, cross-abnormal pattern traceability is captured to obtain abnormal feature traceability vectors; time series fault trend modeling is performed on the abnormal feature traceability vectors to quantitatively evaluate the degree of equipment degradation and obtain a failure critical point prediction matrix; Step S5: Perform a fault risk assessment on the fault critical point prediction matrix, and perform dynamic collaborative optimization of host operating parameters to obtain an intelligent operation and maintenance plan for the central air conditioner.

2. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 1 is characterized in that: Step S1 is specifically as follows: Step S11: acquiring multimodal sensor data and central air conditioning structure data, and performing data preprocessing on the multimodal sensor data to obtain multimodal sensor data to be analyzed; Step S12: dividing the multimodal sensor data to be analyzed into sensor dimensions to obtain environmental sensor data and central air conditioning sensor data; Step S13: performing central air-conditioning operating environment perception according to the environmental sensing data to obtain central air-conditioning operating environment perception data; Step S14: sensing the central air-conditioning operating state based on the central air-conditioning sensor data to obtain central air-conditioning state sensing data; Step S15: performing perception feature fusion on the central air-conditioning operating environment perception data and the central air-conditioning state perception data to obtain multimodal operation perception data.

3. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 2 is characterized in that: Step S14 is specifically as follows: Step S141: performing hierarchical decoupling of operating parameters on the central air conditioning sensor data to obtain an operating parameter feature vector and an electrical parameter feature vector; Step S142: performing thermophysical parameter inversion on the operating parameter characteristic vector based on a preset thermodynamic cycle physical model to obtain a thermal efficiency characteristic matrix; Based on the preset energy consumption load mapping model, the motor parameters are inverted for the electrical parameter characteristic vector to obtain the motor power characteristic matrix; Step S143: performing graph convolutional network state association modeling based on the thermal efficiency feature matrix and the motor power feature matrix to obtain a central air conditioning state time series feature map; Step S144: performing central air conditioning abnormal state detection on the central air conditioning state time series characteristic map to obtain an abnormal state perception vector and a normal state perception vector; Step S145: performing central air-conditioning state time series merging based on the abnormal state perception vector and the normal state perception vector to obtain central air-conditioning state perception data.

4. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S21: Dynamically slice the multimodal operation perception data, set the perception time series window length to [10 minutes, 30 minutes], and perform hierarchical time series slicing within a time step of [1 minute, 5 minutes] based on the preset sliding window [10 minutes, 30 minutes] to obtain perception dimension slice data; Step S22: construct a multi-dimensional time series feature map of the central air conditioner based on the perception dimension slice data, and perform temperature-current-wind speed-refrigerant pressure cross-correlation modeling to generate a spatiotemporal interaction map; Step S23: Perform adaptive attention dynamic weighting on the interaction strength between nodes in the spatiotemporal interaction graph to generate a dynamic spatiotemporal interaction weight matrix; Step S24: Perform thermodynamic cycle coupling modeling based on the central air conditioning structure data, set the thermodynamic cycle parameters [0°C, 100°C] and the airflow disturbance coefficient [0.01, 0.1], and perform physical coupling correction on the dynamic spatiotemporal interaction weight matrix based on the coupling modeling results to obtain a corrected spatiotemporal interaction characteristic matrix; Step S25: performing feature normalization processing on the modified spatiotemporal interaction feature matrix, and performing weight suppression of abnormal interaction regions between nodes to generate a spatiotemporal interaction feature matrix.

5. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 4 is characterized in that: Step S25 is specifically as follows: Step S251: performing node normalization processing on the modified spatiotemporal interaction feature matrix, setting the normalization interval [0, 1], and performing normalization conversion based on the preset node timing window length [5min, 10min] and dynamic interaction frequency threshold [0.1, 0.5] to generate a standardized spatiotemporal interaction feature matrix; Step S252: Based on the abnormal state perception vector and the normal state perception vector, combined with the coupling modeling results, the nodes in the standardized spatiotemporal interaction feature matrix are detected for abnormal regions in a time series. If the interaction strength of a node exceeds the preset abnormal threshold [1.5σ, 3σ], the spatiotemporal trajectory of the node is marked to obtain a set of abnormally marked interaction region nodes; Step S253: Dynamically suppress the weight of the node set in the abnormally labeled interaction area. Combined with the preset time series smoothing coefficient [0.1, 0.3] and the node activity weight factor [0.5, 1.5], perform dimensionality reduction mapping of the abnormal node weight within the node time series window length [5min, 10min] to generate a suppressed spatiotemporal interaction feature matrix. Step S254: Based on the thermodynamic cycle physical model and the energy consumption load mapping model, cross-modal spatiotemporal feature fusion is performed on the spatiotemporal interaction suppression feature matrix to obtain the spatiotemporal interaction feature matrix.

6. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 1 is characterized in that: Step S3 is specifically as follows: Step S31: extracting central air-conditioning operation features based on the multi-modal operation perception data, and identifying the air-conditioning operation mode to obtain central air-conditioning operation mode data; Step S32: extracting structural connection features based on the central air-conditioning structure data, and constructing a central air-conditioning digital twin model based on the structural connection features and the central air-conditioning operation mode data; Step S33: Input the spatiotemporal interaction feature matrix into the central air-conditioning digital twin initial model, and perform cross-mapping of thermal-electrical-environmental modal features to generate a central air-conditioning state mapping feature map; Step S34: Based on the central air conditioning state mapping feature map, dynamically align the physical state parameters with the environmental perception parameters to obtain a state parameter correction matrix; Step S35: Dynamically update the parameters of the central air-conditioning digital twin model according to the state parameter correction matrix, and perform adaptive timing correction of the twin model to obtain a real-time updated digital twin model.

7. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 1 is characterized in that: The cross-abnormal pattern tracing and capturing described in step S4 is specifically as follows: Based on the real-time updating of the digital twin model, the real-time thermal-electrical-environmental three-modal feature data of the central air conditioner is obtained, and the multi-modal feature time series hierarchical decoupling is performed to obtain the real-time thermal-electrical-environmental three-modal feature vector; Based on the real-time thermal-electrical-environmental trimodal feature vectors and combined with the abnormal state perception vector, a central air conditioning abnormal mode association graph is constructed. The cross-abnormal intensity of modal feature nodes is modeled to generate an abnormal mode cross-feature matrix. The abnormal factor traceability path is traced on the abnormal pattern cross-feature matrix. Based on the digital twin model's thermodynamic cycle parameters [20°C, 70°C], motor power parameters [0kW, 5kW], and environmental disturbance parameters [0, 0.5], the abnormal triggering factors are located and an abnormal traceability path diagram is generated. Based on the anomaly tracing path diagram, perform dynamic clustering of abnormal nodes, extract the anomaly trigger factor vector, and generate the anomaly trigger factor perception vector; The abnormal trigger factor perception vector and multimodal operation perception data are combined to perform abnormal pattern layered tracing and obtain the abnormal feature tracing vector.

8. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 1 is characterized in that: The quantitative evaluation of the equipment degradation degree described in step S4 is specifically as follows: Based on the abnormal feature tracing vector, the abnormal pattern intensity time series response model is performed to obtain the abnormal pattern time series response curve; A central air conditioning operation degradation time series model is constructed based on the abnormal mode time series response curve, and an adaptive dynamic sliding window decomposition is performed on the central air conditioning operation degradation time series model to extract the degradation factor time series feature vector; Based on the degradation factor time series eigenvector, combined with the real-time update of the thermal, electrical, and environmental trimodal parameters in the digital twin model, the degradation critical threshold is set; The critical degradation threshold is used to predict the critical failure point by updating the digital twin model in real time, and the critical failure point prediction matrix is ​​obtained.

9. The cross-analysis digital twin intelligent operation and maintenance method based on artificial intelligence according to claim 1 is characterized in that: The two-layer game failure risk assessment described in step S5 is specifically as follows: Obtain historical maintenance record data and perform time series alignment on the historical maintenance record data according to the spatiotemporal interaction feature matrix to obtain the spatiotemporal maintenance interaction matrix; Combining the failure critical point prediction matrix and the spatiotemporal maintenance interaction matrix, a two-layer game model is constructed. The outer layer game subject is set as the equipment status, and the inner layer game subject is set as the maintenance record, thus obtaining the equipment failure risk assessment game model. Based on the equipment failure risk assessment game model, a two-level game is solved to determine the failure probability and the optimal solution of the maintenance strategy, and a dual game solution set is obtained; Time series tracking is performed based on the double game solution set, and combined with the real-time update of the digital twin model to predict the changing trend of equipment failure risk, the equipment failure risk change prediction data is obtained.

10. An artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance system, characterized in that: Used to execute the artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance method according to any one of claims 1 to 9, the artificial intelligence-based cross-analysis digital twin intelligent operation and maintenance system includes: A data acquisition module is used to obtain multimodal sensor data and central air-conditioning structure data, and perform multimodal air-conditioning operation perception based on the multimodal sensor data to obtain multimodal operation perception data; The interactive feature cross-module is used to perform dynamic dimension slicing on the multimodal operation perception data to obtain perception dimension slice data; based on the perception dimension slice data, a spatiotemporal interaction graph is constructed, and feature weights are dynamically assigned to obtain a spatiotemporal interaction feature matrix; The digital twin model construction module is used to build a central air-conditioning digital twin model based on the central air-conditioning structure data; the spatiotemporal interaction feature matrix is ​​input into the central air-conditioning digital twin model to perform cross-model updates and obtain a real-time updated digital twin model; The abnormal pattern tracing module is used to trace and capture cross-abnormal patterns based on the real-time updated digital twin model to obtain abnormal feature traceability vectors. The module also models the time series fault trend of the abnormal feature traceability vectors, quantitatively assesses the degree of equipment degradation, and obtains a critical point fault prediction matrix. The fault risk assessment module is used to perform a two-layer game fault risk assessment on the fault critical point prediction matrix and perform dynamic collaborative optimization of the host operating parameters to obtain an intelligent operation and maintenance plan for the central air conditioner.

Images