An air conditioning system full life cycle operation and maintenance diagnosis method based on a large model

Abstract

The application discloses a kind of based on big model's air conditioning system full life cycle operation and maintenance diagnosis method, it is relevant, the steps of this method include: collecting the operation and maintenance data of air conditioning system full life cycle and pre-processing;The internal correlation feature of operation and maintenance data is analyzed to change, and correlation structure skewness is constructed;Based on the correlation structure skewness, the time sequence characteristics of air conditioning system fault evolution are analyzed, and state evolution accumulation degree is constructed;Based on the state evolution accumulation degree, the information value that current data sample has for constructing decision tree is analyzed, and decision value index is constructed;Based on the decision value index, the probability of extracting current data sample when constructing decision tree in random forest algorithm is improved, and the operation and maintenance diagnosis of air conditioning system full life cycle is realized.The application solves the problem that the sample of normal operating state in the operation and maintenance diagnosis process of air conditioning system full life cycle occupies the vast majority, and forms typical data imbalance problem.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method for full lifecycle operation and maintenance diagnosis of air conditioning systems based on a large model. Background Technology

[0002] Full lifecycle operation and maintenance diagnosis of air conditioning systems refers to the use of advanced sensing technology, data analysis, and intelligent algorithms to monitor the system's status in real time, diagnose faults, evaluate performance, and perform predictive maintenance throughout the entire lifecycle of an air conditioning system, from design and manufacturing to installation, commissioning, operation, and disposal. Its main function is to proactively identify potential fault risks, optimize system operating parameters, reduce energy consumption and maintenance costs, and extend equipment lifespan. Performing full lifecycle operation and maintenance diagnosis of air conditioning systems is a key technological means to ensure the comfort, reliability, and energy efficiency of modern buildings, and is of great significance for promoting green building and sustainable development.

[0003] In existing technologies, the random forest algorithm, as an ensemble learning method, constructs multiple decision trees and integrates their prediction results, offering advantages such as fast training speed, high accuracy, good tolerance to noise and outliers, and low overfitting. However, the entire lifecycle operation and maintenance diagnosis process of air conditioning systems is characterized by a large volume of operational data and sparse fault sample data, meaning that samples from normal operating states constitute the vast majority, forming a typical data imbalance problem. If the original random forest algorithm is directly used to train on this imbalanced data, the decision trees tend to learn the features of the majority class, i.e., the features of the normal state, resulting in lower accuracy in identifying the minority class, i.e., the features of various faults, leading to a high false negative rate and failing to meet the high accuracy requirements of operation and maintenance diagnosis for fault detection.

[0004] Therefore, this invention provides a method for full lifecycle operation and maintenance diagnosis of air conditioning systems based on a large model. Summary of the Invention

[0005] The purpose of this invention is to provide a method for operation and maintenance diagnosis of air conditioning systems throughout their entire lifecycle based on a large model, so as to solve the existing problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for full lifecycle operation and maintenance diagnosis of air conditioning systems based on a large model, comprising the following steps:

[0007] Step S1: Collect and preprocess the operation and maintenance data of the air conditioning system throughout its entire life cycle;

[0008] Step S2: Analyze the changes in the inherent correlation characteristics of the operation and maintenance data, and construct the correlation structure skewness;

[0009] Step S3: Based on the correlation structure skewness analysis, analyze the temporal characteristics of the air conditioning system fault evolution and construct the state evolution cumulative degree;

[0010] Step S4: Based on the state evolution cumulative analysis, analyze the information value of the current data sample used to construct the decision tree, and construct the decision value index;

[0011] Step S5: Based on the decision value index, improve the probability of extracting the current data sample when constructing the decision tree in the random forest algorithm to realize the operation and maintenance diagnosis of the entire life cycle of the air conditioning system.

[0012] A further improvement of the present invention is that step S2 specifically includes the following steps:

[0013] S21. The time window consisting of the n times preceding time t is denoted as the operation and maintenance monitoring window.

[0014] S22. Calculate the Pearson correlation coefficients between any two dimensions within the operation and maintenance monitoring window, and denote the matrix formed by these coefficients as the operation and maintenance correlation matrix. Then the dimension of the matrix is ​​N×N, where N represents the dimension of the operation and maintenance data;

[0015] S23. Perform eigenvalue decomposition on the operation and maintenance related matrix to obtain N eigenvalues, and record the sequence of eigenvalues ​​after descending order as the current eigenvalue sequence.

[0016] S24. Using the same processing method as step S1, collect the full life cycle operation and maintenance data of the air conditioner in a healthy state as determined by technical personnel, and obtain the health operation and maintenance correlation matrix and health feature value sequence of the full life cycle operation and maintenance data in a healthy state within the same time window through steps S21-S23.

[0017] S25. Combine the output data from steps S23 and S24 to obtain the associated structural skewness.

[0018] A further improvement of this invention is that the method for calculating the skewness of the associated structure is as follows:

[0019] ;

[0020] in The value at time t represents the correlation skewness, and N represents the dimension of the air conditioning system's full lifecycle operation and maintenance data. Represents the operation and maintenance related matrix The value in the i-th row and j-th column is the Pearson correlation coefficient between the i-th dimension of operation and maintenance data and the j-th dimension of operation and maintenance data in the operation and maintenance monitoring window. This represents the value in the i-th row and j-th column of the health operation and maintenance correlation matrix, which is the Pearson correlation coefficient between the i-th dimension and the j-th dimension of the air conditioning system's full lifecycle operation and maintenance data under healthy conditions.

[0021] Represents the correlation structure divergence at time t. Represented by the operation and maintenance related matrix The k-th feature value obtained in the current feature value sequence, Represented by the operation and maintenance related matrix The sum of all eigenvalues ​​in the current eigenvalue sequence is obtained. This represents the k-th feature value in the sequence of health feature values. This represents the sum of all characteristic values ​​in the health characteristic value sequence;

[0022] The product of the correlation numerical skewness and the correlation structure divergence at time t is taken as the correlation structure skewness.

[0023] A further improvement of this invention is that step S3 constructs a state evolution cumulative degree based on the correlation structure skewness combined with time persistence to quantify the cumulative impact of fault characteristics in the air conditioning system. The specific process includes: based on the implementation process of step S2, calculating the correlation structure skewness at each time point, and recording the sequence of correlation structure skewness within the operation and maintenance monitoring window as the nearest neighbor correlation structure skewness sequence; calculating the mean of the correlation structure skewness of the full life cycle operation and maintenance data running in a healthy state, and using it as a benchmark threshold; thus obtaining the state evolution cumulative degree.

[0024] A further improvement of this invention is that the method for calculating the cumulative degree of state evolution is as follows:

[0025] ;

[0026] in, Represents the cumulative degree of state evolution at time t. This represents the attenuation coefficient, used to control the rate at which historical operation and maintenance data decays. This represents the cumulative degree of state evolution at time (t-1), and max() represents the function to find the maximum value. This represents the correlation structure skewness at time t. This represents the baseline threshold.

[0027] A further improvement of this invention lies in constructing a decision value index based on the state evolution cumulative degree, which is used to evaluate the amount of information contained in the operation and maintenance data when used for fault detection in the random forest model. The construction process of the decision value index is as follows:

[0028] Extract the cumulative state evolution at each historical moment, and use the cumulative state evolution at each historical moment and the cumulative state evolution at time t as inputs to the KNN algorithm. The output is the cumulative state evolution of the K nearest neighbors at time t. The set formed by these cumulative states is denoted as the K nearest neighbor set, and the decision value index is obtained.

[0029] A further improvement of this invention is that the calculation method for the decision value index is as follows:

[0030] ;

[0031] in Let represent the decision value index at time t, and K represent the number of nearest neighbors selected in the KNN classification algorithm. Represents the cumulative degree of state evolution at time t. Let represent the value of the b-th element in the K-nearest neighbor set of the cumulative degree of state evolution at time t, where 1 indicates a hyperparameter.

[0032] A further improvement of this invention lies in extracting the decision value index at each time point, improving the sampling probability when constructing the decision tree in the random forest algorithm based on the decision value index, thereby improving the random forest algorithm, and realizing the operation and maintenance diagnosis of the same entire life cycle of air conditioners based on the improved random forest algorithm. The specific process is as follows: sum the decision value indices at all times to obtain the summed feature value, and use the ratio of the decision value index at time t to the summed feature value as the sampling probability at that time. In the original random forest algorithm, uniform random sampling with replacement is adopted.

[0033] A further improvement of the present invention is that step S5 further includes using real-time operation and maintenance data as input to the random forest algorithm after training, and outputting the diagnostic results of the current system.

[0034] A further improvement of this invention is that the operation and maintenance data specifically includes: the temperature of the chiller unit, the compressor current, and the inlet and outlet temperatures of the cooling water; the operating frequency, current, and differential pressure of the distribution system; and environmental parameters, including indoor temperature and humidity.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] 1. This invention first constructs a correlation structure skewness by analyzing the changes in the inherent correlation characteristics of operation and maintenance data of different dimensions during the operation of the air conditioning system, which can more sensitively identify minor anomalies or early faults in the system status.

[0037] 2. Secondly, based on the correlation structure skewness analysis, the cumulative effect of the air conditioning system failure is analyzed, and then the state evolution cumulative degree is constructed to effectively distinguish between persistent failures and transient disturbances, thereby enhancing the accuracy of operation and maintenance diagnosis results;

[0038] 3. Based on the information value of the samples used in constructing the decision tree, the cumulative degree of state evolution analysis is used to construct a decision value index. Based on the decision value index, samples are probabilistically selected as the samples needed to construct the decision tree. This solves the problem of the algorithm's accuracy in identifying a few fault samples due to data imbalance. It enables the algorithm to significantly improve the detection rate of various faults while maintaining a high distinction rate for the normal state of the air conditioner. This provides more reliable decision support for the predictive maintenance of the air conditioning system and ensures the safe and stable operation of the system. Attached Figure Description

[0039] Figure 1 This is a flowchart of a full lifecycle operation and maintenance diagnosis method for air conditioning systems based on a large model, according to the present invention. Detailed Implementation

[0040] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0041] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0042] Example 1

[0043] Figure 1 This embodiment presents a flowchart of a full lifecycle operation and maintenance diagnosis method for an air conditioning system based on a large model. The steps are as follows:

[0044] Step S1: Collect and preprocess the operation and maintenance data of the air conditioning system throughout its entire life cycle.

[0045] During the entire lifecycle operation and maintenance diagnosis of the air conditioning system, various sensors and devices deployed at key system nodes comprehensively collect multi-dimensional data reflecting the system's operating status. Specifically, this includes: chiller unit temperature, compressor current, and cooling water inlet and outlet temperatures; operating frequency, current, and differential pressure of the distribution system (such as water pumps and fans); and environmental parameters, including indoor temperature and humidity. The data collection interval for the above data is set to 1 second in this application, but can be selected according to the specific circumstances.

[0046] During the data collection process, missing values ​​may be present due to environmental interference, noise, or other factors. Therefore, appropriate preprocessing is required, including: filling in missing values ​​using the nearest neighbor method to ensure data integrity; and using the maximum-minimum normalization method to map all dimensions of data to the [0,1] interval to standardize the data, in order to eliminate the influence of different physical dimensions on the algorithm's calculation results. The nearest neighbor method and the maximum-minimum normalization method are well-known techniques and will not be elaborated upon here.

[0047] As described above, we obtain the preprocessed operation and maintenance data of the entire lifecycle of the air conditioning system that needs to be monitored.

[0048] Step S2: Analyze the changes in the inherent correlation characteristics of the operation and maintenance data and construct the correlation structure skewness.

[0049] The operation of an air conditioning system is a multivariable, strongly coupled thermodynamic process. During normal operation, the various physical parameters follow specific laws of energy conservation and thermodynamics, exhibiting a stable and highly correlated intrinsic relationship. For example, an increase in compressor current should be accompanied by a corresponding change in refrigerant evaporation pressure. When an early potential fault occurs in the system, such as a minor refrigerant leak or heat exchanger scaling, the change in a single parameter may not be significant and may still remain within the normal threshold range. However, the intrinsic correlation between parameters will be disrupted first, resulting in decoupling phenomena that deviate from the normal pattern.

[0050] Based on the above analysis, this application constructs a correlation structure skewness to quantify the degree of deviation between operation and maintenance data and normal correlation patterns, thereby improving the sensitivity to potential fault identification. The construction process of the correlation structure skewness is as follows:

[0051] Taking time t as an example, the time window consisting of the n times preceding time t is denoted as the operation and maintenance monitoring window. The value of n is 120 in this application, but can be selected according to the situation. The Pearson correlation coefficients between each pair of dimensions within the operation and maintenance monitoring window are calculated, and the matrix formed by these coefficients is denoted as the operation and maintenance correlation matrix. The dimension of the matrix is ​​N×N, where N represents the dimension of the operation and maintenance data. The matrix is ​​decomposed into eigenvalues ​​to obtain N eigenvalues. The sequence of eigenvalues ​​after descending order is denoted as the eigenvalue sequence.

[0052] Following the same processing method as step S1, collect full lifecycle operation and maintenance data of the air conditioner operating in a healthy state, as confirmed by technical personnel. Then, following the same processing method as the above steps, obtain the health operation and maintenance correlation matrix and health feature value sequence of the full lifecycle operation and maintenance data in a healthy state within the same time window.

[0053] Based on the above processing steps, the method for calculating the skewness of the associated structure in this application is as follows:

[0054] ;

[0055] ;

[0056] in The value at time t represents the correlation skewness, and N represents the dimension of the air conditioning system's full lifecycle operation and maintenance data. Represents the operation and maintenance related matrix The value in the i-th row and j-th column is the Pearson correlation coefficient between the i-th dimension of operation and maintenance data and the j-th dimension of operation and maintenance data in the operation and maintenance monitoring window. This represents the value in the i-th row and j-th column of the health operation and maintenance correlation matrix, which is the Pearson correlation coefficient between the i-th dimension and the j-th dimension of the air conditioning system's full lifecycle operation and maintenance data under healthy conditions.

[0057] Represents the correlation structure divergence at time t. Represented by the operation and maintenance related matrix The k-th feature value obtained in the current feature value sequence, Represented by the operation and maintenance related matrix The sum of all eigenvalues ​​in the current eigenvalue sequence is obtained. This represents the k-th feature value in the sequence of health feature values. This represents the sum of all characteristic values ​​in the health characteristic value sequence;

[0058] In the correlation numerical skewness, the operation and maintenance correlation matrix is ​​calculated. The sum of Manhattan distances between elements in the health and operations related matrix measures the overall numerical difference between the two matrices. When the correlation of operations data in individual dimensions changes, the calculated correlation values ​​will be more skewed.

[0059] In the correlation structure divergence, by comparing the operation and maintenance correlation matrix The differences in eigenvalues ​​of the health and maintenance correlation matrix quantify the similarity of the association structure of the entire lifecycle maintenance data of the air conditioning system. When the system operating mode undergoes a fundamental change, such as switching from cooling mode to heating mode, or when a serious fault occurs, the association structure of different dimensions will change significantly, causing the maintenance correlation matrix to... The eigenvalues ​​of the correlation matrix with health operation and maintenance show significant differences, resulting in a large divergence in the calculated correlation structure.

[0060] The product of the correlation numerical skewness and the correlation structural divergence at time t is taken as the correlation structural skewness, so that the value of the correlation structural skewness can comprehensively reflect the differences in structure and data between the full life cycle operation and maintenance data of the air conditioning system that needs to be monitored and the full life cycle operation and maintenance data running in a healthy state.

[0061] Step S3: Construct state evolution cumulative degree based on the temporal characteristics of air conditioning system fault evolution through correlation structure skewness analysis.

[0062] When diagnosing faults in air conditioning systems, the evolution of faults is temporal. A momentary high deviation in the associated structure may be caused by temporary external disturbances or noise, and does not necessarily represent a true, persistent equipment fault. A true, gradually worsening fault, however, will manifest as a sustained, cumulative high level of multidimensional correlation deviation over time, such as duct blockage or loose fan belts. This temporal persistence and cumulative effect of the fault state can effectively distinguish between a true fault and a momentary disturbance.

[0063] Therefore, this application constructs a state evolution cumulative degree based on the correlation structure skewness combined with time persistence to quantify the cumulative impact of fault characteristics in the air conditioning system, thereby improving the robustness of operation and maintenance diagnosis results. The construction process of the state evolution cumulative degree is as follows:

[0064] Following the same steps as above, calculate the correlation structure skewness at each time point, and record the sequence of correlation structure skewness within the operation and maintenance monitoring window as the nearest neighbor correlation structure skewness sequence. Calculate the mean of the correlation structure skewness of the entire lifecycle operation and maintenance data running in a healthy state, and use it as the benchmark threshold.

[0065] Based on the above processing steps, the method for calculating the state evolution cumulative degree in this application is as follows:

[0066] ;

[0067] in Represents the cumulative degree of state evolution at time t. This represents the attenuation coefficient, used to control the attenuation rate of historical operation and maintenance data. In this application, it is set to 0.2, but can be selected according to the specific circumstances. This represents the cumulative degree of state evolution at time (t-1), and max() represents the function to find the maximum value. This represents the correlation structure skewness at time t. This represents the baseline threshold.

[0068] In the state evolution cumulative degree, by combining the exponential moving weighted average and the anomaly activation mechanism, if the deviation of the associated structure skewness at time t exceeds the mean value of the associated structure skewness of the air conditioning system's full life cycle operation and maintenance data under healthy conditions, i.e., the benchmark threshold, only the excess part will participate in the calculation of the state evolution cumulative degree, i.e., be included in the cumulative value. This allows the state evolution cumulative degree to effectively accumulate significant fault characteristics. At the same time, due to the existence of the attenuation coefficient, the state evolution cumulative degree can ignore the occasional normal fluctuations. When the operating state of the air conditioning system is continuously abnormal, the associated structure skewness at time t will continue to be greater than the mean value of the associated structure skewness under healthy conditions. The more obvious the cumulative effect of the fault, the larger the calculated state evolution cumulative degree.

[0069] Step S4: Analyze the information value of the sample used to construct the decision tree based on the state evolution cumulative analysis and construct the decision value index.

[0070] In an imbalanced data space composed of massive normal samples and sparse fault samples, the information value of fault samples is not equal. Some fault samples represent the nascent or early stages of a fault; their feature vectors are very close to the boundary of the normal sample cluster in the multidimensional space, making them easily confused with noise or disturbances under normal operating conditions. These are the boundary samples that define the classification boundary. Other samples represent the typical stage where the fault has fully developed; their feature vectors are far from the normal sample cluster in the space, and they are the core samples that define the core features of the fault category. When training the data using the random forest algorithm, both types of samples have extremely high learning value. However, in the original random forest algorithm, when dealing with imbalanced samples, in order to improve the classification accuracy of massive normal samples, it ignores the few but information-rich boundary samples and core samples, resulting in insufficient generalization ability of the random forest model when detecting air conditioning faults.

[0071] Based on the above analysis, this application constructs a decision value index based on state evolution cumulancy to evaluate the amount of information contained in operation and maintenance data when used for fault detection in a random forest model. The construction process of the decision value index is as follows:

[0072] Following the same steps as described above, the cumulative state evolution at each historical time point is obtained. The cumulative state evolution at each historical time point, along with the cumulative state evolution at time t, are used as inputs to the KNN algorithm. The output is the cumulative state evolution of the K nearest neighbors at time t. This set is denoted as the K nearest neighbor set. The KNN algorithm is a well-known technique and will not be elaborated upon here.

[0073] Based on the above processing steps, the calculation method of the decision value index in this application is as follows:

[0074] ;

[0075] in The decision value index at time t is represented by K, which represents the number of nearest neighbors selected in the KNN classification algorithm. In this application, the value is 10, but it can be selected according to the situation. Represents the cumulative degree of state evolution at time t. This represents the value of the b-th element in the K-nearest neighbor set of the cumulative degree of state evolution at time t. 1 indicates a hyperparameter used to avoid the denominator being zero, which would prevent calculation. It can be selected as needed.

[0076] When the cumulative state evolution at time t is high, or the cumulative state evolution of the surrounding neighborhood differs significantly from it, the corresponding or A larger value indicates that the moment is located at the boundary between the normal operation time and the fault time of the air conditioner, or at the fault time. That is, it belongs to the boundary sample or core sample. In the random forest algorithm, these are important samples that need to be learned. They have higher value when constructing the decision tree, resulting in a larger calculated decision value index.

[0077] Step S5: Improve the probability of sampling when constructing the decision tree in the random forest algorithm based on the decision value index to achieve operation and maintenance diagnosis of the entire life cycle of the air conditioning system.

[0078] The decision value index obtained through the above steps reflects the information value of the samples at each time point when constructing the decision tree. That is, for moments with a higher decision value index, the samples should be used more frequently in constructing the decision tree; for moments with a lower decision value index, the samples should be used with less probability. Therefore, this application improves the probability of sampling when constructing the decision tree in the random forest algorithm based on the decision value index, thereby improving the random forest algorithm. Based on the improved random forest algorithm, it implements operation and maintenance diagnosis for the same entire lifecycle of air conditioners. The specific process is as follows:

[0079] The decision value indices at all times are summed to obtain the summed feature value. The ratio of the decision value index at time t to the summed feature value is used as the sampling probability at that time. In the original random forest algorithm, uniform random sampling with replacement is used, that is, each sample has an equal probability of being selected. This application uses probability sampling with replacement, and uses the sampling probability as the probability of selecting the sample to build the decision tree, so that samples with higher classification value are more likely to be selected, thereby making the constructed decision tree more valuable for decision-making.

[0080] The above steps are used to improve the random forest algorithm and train it. Real-time operation and maintenance data are used as input to the trained random forest algorithm, and the output is the diagnostic results of the current system, such as the specific fault type, such as normal operation, compressor failure, refrigerant leakage, etc., thereby realizing high-precision and intelligent real-time operation and maintenance diagnosis of the air conditioning system.

[0081] The threshold and weight settings can be set by default according to the present invention, or they can be set by those skilled in the art.

[0082] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for full lifecycle operation and maintenance diagnosis of air conditioning systems based on a large model, characterized in that: Includes the following steps: Step S1: Collect and preprocess the operation and maintenance data of the air conditioning system throughout its entire life cycle; Step S2: Analyze the changes in the inherent correlation characteristics of the operation and maintenance data, and construct the correlation structure skewness; Step S3: Based on the correlation structure skewness analysis, analyze the temporal characteristics of the air conditioning system fault evolution and construct the state evolution cumulative degree; Step S4: Based on the state evolution cumulative analysis, analyze the information value of the current data sample used to construct the decision tree, and construct the decision value index; Step S5: Based on the decision value index, improve the probability of extracting the current data sample when constructing the decision tree in the random forest algorithm to realize the operation and maintenance diagnosis of the entire life cycle of the air conditioning system; The specific steps of step S2 include: S21. The time window consisting of the n times preceding time t is denoted as the operation and maintenance monitoring window. S22. Calculate the Pearson correlation coefficients between any two dimensions within the operation and maintenance monitoring window, and denote the matrix formed by these coefficients as the operation and maintenance correlation matrix. Then the dimension of the matrix is ​​N×N, where N represents the dimension of the operation and maintenance data; S23. Perform eigenvalue decomposition on the operation and maintenance related matrix to obtain N eigenvalues, and record the sequence of eigenvalues ​​after descending order as the current eigenvalue sequence. S24. Using the same processing method as step S1, collect the full life cycle operation and maintenance data of the air conditioner in a healthy state as determined by technical personnel, and obtain the health operation and maintenance correlation matrix and health feature value sequence of the full life cycle operation and maintenance data in a healthy state within the same time window through steps S21-S23. S25. Combine the output data from steps S23 and S24 to obtain the associated structural skewness; the calculation method for the associated structural skewness is as follows: ； in The value at time t represents the correlation skewness, and N represents the dimension of the air conditioning system's full lifecycle operation and maintenance data. Represents the operation and maintenance related matrix The value in the i-th row and j-th column is the Pearson correlation coefficient between the i-th dimension of operation and maintenance data and the j-th dimension of operation and maintenance data in the operation and maintenance monitoring window. This represents the value in the i-th row and j-th column of the health operation and maintenance correlation matrix, which is the Pearson correlation coefficient between the i-th dimension and the j-th dimension of the air conditioning system's full lifecycle operation and maintenance data under healthy conditions. Represents the correlation structure divergence at time t. Represented by the operation and maintenance related matrix The k-th feature value obtained in the current feature value sequence, Represented by the operation and maintenance related matrix The sum of all eigenvalues ​​in the current eigenvalue sequence is obtained. This represents the k-th feature value in the sequence of health feature values. This represents the sum of all characteristic values ​​in the health characteristic value sequence; The product of the correlation numerical skewness and the correlation structure divergence at time t is taken as the correlation structure skewness. Step S3 constructs the state evolution cumulative degree based on the correlation structure skewness combined with time persistence to quantify the cumulative impact of fault characteristics in the air conditioning system. The specific process includes: based on the implementation process of step S2, calculating the correlation structure skewness at each time point, and recording the sequence of correlation structure skewness within the operation and maintenance monitoring window as the nearest neighbor correlation structure skewness sequence; calculating the mean of the correlation structure skewness of the full life cycle operation and maintenance data running in a healthy state, and using it as the benchmark threshold; thus obtaining the state evolution cumulative degree. A decision value index is constructed based on the state evolution cumulative degree to evaluate the amount of information contained in the operation and maintenance data when used for fault detection in the random forest model. The construction process of the decision value index is as follows: Extract the cumulative state evolution at each historical moment, and use the cumulative state evolution at each historical moment and the cumulative state evolution at time t as inputs to the KNN algorithm. The output is the cumulative state evolution of the K nearest neighbors at time t. The set formed by these cumulative states is denoted as the K nearest neighbor set, and the decision value index is obtained.

2. The method for full lifecycle operation and maintenance diagnosis of an air conditioning system based on a large model as described in claim 1, characterized in that: The method for calculating the cumulative degree of state evolution is as follows: ； in, Represents the cumulative degree of state evolution at time t. This represents the attenuation coefficient, used to control the rate at which historical operation and maintenance data decays. This represents the cumulative degree of state evolution at time (t-1), and max() represents the function to find the maximum value. This represents the correlation structure skewness at time t. This represents the baseline threshold.

3. The method for full lifecycle operation and maintenance diagnosis of an air conditioning system based on a large model according to claim 2, characterized in that: The decision value index is calculated as follows: ； in Let represent the decision value index at time t, and K represent the number of nearest neighbors selected in the KNN classification algorithm. Represents the cumulative degree of state evolution at time t. Let represent the value of the b-th element in the K-nearest neighbor set of the cumulative degree of state evolution at time t, where 1 indicates a hyperparameter.

4. The method for full lifecycle operation and maintenance diagnosis of an air conditioning system based on a large model as described in claim 3, characterized in that: The decision value index at each time point is extracted. Based on the decision value index, the probability of sampling when constructing the decision tree in the random forest algorithm is improved, thereby improving the random forest algorithm. Based on the improved random forest algorithm, the operation and maintenance diagnosis of the same whole life cycle of the air conditioner is realized. The specific process is as follows: the decision value index at all times is summed to obtain the summed feature value. The ratio of the decision value index at time t to the summed feature value is used as the sampling probability at that time. In the original random forest algorithm, uniform random sampling with replacement is adopted.

5. The method for full lifecycle operation and maintenance diagnosis of an air conditioning system based on a large model according to claim 4, characterized in that: Step S5 further includes using real-time operation and maintenance data as input to the trained random forest algorithm, and outputting the diagnostic results of the current system.

6. The method for full lifecycle operation and maintenance diagnosis of an air conditioning system based on a large model according to claim 1, characterized in that: The specific maintenance data includes: chiller unit temperature, compressor current, cooling water inlet and outlet temperatures; operating frequency, current, and differential pressure of the distribution system; and environmental parameters, including indoor temperature and humidity.

Images